net/ipv6/ip6_fib.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      Forwarding Information Database
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  *
   9  *      Changes:
  10  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  11  *                              remove ip6_null_entry from the top of
  12  *                              routing table.
  13  *      Ville Nuorvala:         Fixed routing subtrees.
  14  */
  15
  16 #define pr_fmt(fmt) "IPv6: " fmt
  17
  18 #include <linux/errno.h>
  19 #include <linux/types.h>
  20 #include <linux/net.h>
  21 #include <linux/route.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/in6.h>
  24 #include <linux/init.h>
  25 #include <linux/list.h>
  26 #include <linux/slab.h>
  27
  28 #include <net/ip.h>
  29 #include <net/ipv6.h>
  30 #include <net/ndisc.h>
  31 #include <net/addrconf.h>
  32 #include <net/lwtunnel.h>
  33 #include <net/fib_notifier.h>
  34
  35 #include <net/ip6_fib.h>
  36 #include <net/ip6_route.h>
  37
  38 static struct kmem_cache *fib6_node_kmem __read_mostly;
  39
  40 struct fib6_cleaner {
  41         struct fib6_walker w;
  42         struct net *net;
  43         int (*func)(struct fib6_info *, void *arg);
  44         int sernum;
  45         void *arg;
  46         bool skip_notify;
  47 };
  48
  49 #ifdef CONFIG_IPV6_SUBTREES
  50 #define FWS_INIT FWS_S
  51 #else
  52 #define FWS_INIT FWS_L
  53 #endif
  54
  55 static struct fib6_info *fib6_find_prefix(struct net *net,
  56                                          struct fib6_table *table,
  57                                          struct fib6_node *fn);
  58 static struct fib6_node *fib6_repair_tree(struct net *net,
  59                                           struct fib6_table *table,
  60                                           struct fib6_node *fn);
  61 static int fib6_walk(struct net *net, struct fib6_walker *w);
  62 static int fib6_walk_continue(struct fib6_walker *w);
  63
  64 /*
  65  *      A routing update causes an increase of the serial number on the
  66  *      affected subtree. This allows for cached routes to be asynchronously
  67  *      tested when modifications are made to the destination cache as a
  68  *      result of redirects, path MTU changes, etc.
  69  */
  70
  71 static void fib6_gc_timer_cb(struct timer_list *t);
  72
  73 #define FOR_WALKERS(net, w) \
  74         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  75
  76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  77 {
  78         write_lock_bh(&net->ipv6.fib6_walker_lock);
  79         list_add(&w->lh, &net->ipv6.fib6_walkers);
  80         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  81 }
  82
  83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  84 {
  85         write_lock_bh(&net->ipv6.fib6_walker_lock);
  86         list_del(&w->lh);
  87         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  88 }
  89
  90 static int fib6_new_sernum(struct net *net)
  91 {
  92         int new, old;
  93
  94         do {
  95                 old = atomic_read(&net->ipv6.fib6_sernum);
  96                 new = old < INT_MAX ? old + 1 : 1;
  97         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
  98                                 old, new) != old);
  99         return new;
 100 }
 101
 102 enum {
 103         FIB6_NO_SERNUM_CHANGE = 0,
 104 };
 105
 106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 107 {
 108         struct fib6_node *fn;
 109
 110         fn = rcu_dereference_protected(f6i->fib6_node,
 111                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 112         if (fn)
 113                 fn->fn_sernum = fib6_new_sernum(net);
 114 }
 115
 116 /*
 117  *      Auxiliary address test functions for the radix tree.
 118  *
 119  *      These assume a 32bit processor (although it will work on
 120  *      64bit processors)
 121  */
 122
 123 /*
 124  *      test bit
 125  */
 126 #if defined(__LITTLE_ENDIAN)
 127 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 128 #else
 129 # define BITOP_BE32_SWIZZLE     0
 130 #endif
 131
 132 static __be32 addr_bit_set(const void *token, int fn_bit)
 133 {
 134         const __be32 *addr = token;
 135         /*
 136          * Here,
 137          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 138          * is optimized version of
 139          *      htonl(1 << ((~fn_bit)&0x1F))
 140          * See include/asm-generic/bitops/le.h.
 141          */
 142         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 143                addr[fn_bit >> 5];
 144 }
 145
 146 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
 147 {
 148         struct fib6_info *f6i;
 149         size_t sz = sizeof(*f6i);
 150
 151         if (with_fib6_nh)
 152                 sz += sizeof(struct fib6_nh);
 153
 154         f6i = kzalloc(sz, gfp_flags);
 155         if (!f6i)
 156                 return NULL;
 157
 158         /* fib6_siblings is a union with nh_list, so this initializes both */
 159         INIT_LIST_HEAD(&f6i->fib6_siblings);
 160         refcount_set(&f6i->fib6_ref, 1);
 161
 162         return f6i;
 163 }
 164
 165 void fib6_info_destroy_rcu(struct rcu_head *head)
 166 {
 167         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 168
 169         WARN_ON(f6i->fib6_node);
 170
 171         if (f6i->nh)
 172                 nexthop_put(f6i->nh);
 173         else
 174                 fib6_nh_release(f6i->fib6_nh);
 175
 176         ip_fib_metrics_put(f6i->fib6_metrics);
 177         kfree(f6i);
 178 }
 179 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 180
 181 static struct fib6_node *node_alloc(struct net *net)
 182 {
 183         struct fib6_node *fn;
 184
 185         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 186         if (fn)
 187                 net->ipv6.rt6_stats->fib_nodes++;
 188
 189         return fn;
 190 }
 191
 192 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 193 {
 194         kmem_cache_free(fib6_node_kmem, fn);
 195         net->ipv6.rt6_stats->fib_nodes--;
 196 }
 197
 198 static void node_free_rcu(struct rcu_head *head)
 199 {
 200         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 201
 202         kmem_cache_free(fib6_node_kmem, fn);
 203 }
 204
 205 static void node_free(struct net *net, struct fib6_node *fn)
 206 {
 207         call_rcu(&fn->rcu, node_free_rcu);
 208         net->ipv6.rt6_stats->fib_nodes--;
 209 }
 210
 211 static void fib6_free_table(struct fib6_table *table)
 212 {
 213         inetpeer_invalidate_tree(&table->tb6_peers);
 214         kfree(table);
 215 }
 216
 217 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 218 {
 219         unsigned int h;
 220
 221         /*
 222          * Initialize table lock at a single place to give lockdep a key,
 223          * tables aren't visible prior to being linked to the list.
 224          */
 225         spin_lock_init(&tb->tb6_lock);
 226         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 227
 228         /*
 229          * No protection necessary, this is the only list mutatation
 230          * operation, tables never disappear once they exist.
 231          */
 232         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 233 }
 234
 235 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 236
 237 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 238 {
 239         struct fib6_table *table;
 240
 241         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 242         if (table) {
 243                 table->tb6_id = id;
 244                 rcu_assign_pointer(table->tb6_root.leaf,
 245                                    net->ipv6.fib6_null_entry);
 246                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 247                 inet_peer_base_init(&table->tb6_peers);
 248         }
 249
 250         return table;
 251 }
 252
 253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 254 {
 255         struct fib6_table *tb;
 256
 257         if (id == 0)
 258                 id = RT6_TABLE_MAIN;
 259         tb = fib6_get_table(net, id);
 260         if (tb)
 261                 return tb;
 262
 263         tb = fib6_alloc_table(net, id);
 264         if (tb)
 265                 fib6_link_table(net, tb);
 266
 267         return tb;
 268 }
 269 EXPORT_SYMBOL_GPL(fib6_new_table);
 270
 271 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 272 {
 273         struct fib6_table *tb;
 274         struct hlist_head *head;
 275         unsigned int h;
 276
 277         if (id == 0)
 278                 id = RT6_TABLE_MAIN;
 279         h = id & (FIB6_TABLE_HASHSZ - 1);
 280         rcu_read_lock();
 281         head = &net->ipv6.fib_table_hash[h];
 282         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 283                 if (tb->tb6_id == id) {
 284                         rcu_read_unlock();
 285                         return tb;
 286                 }
 287         }
 288         rcu_read_unlock();
 289
 290         return NULL;
 291 }
 292 EXPORT_SYMBOL_GPL(fib6_get_table);
 293
 294 static void __net_init fib6_tables_init(struct net *net)
 295 {
 296         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 297         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 298 }
 299 #else
 300
 301 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 302 {
 303         return fib6_get_table(net, id);
 304 }
 305
 306 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 307 {
 308           return net->ipv6.fib6_main_tbl;
 309 }
 310
 311 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 312                                    const struct sk_buff *skb,
 313                                    int flags, pol_lookup_t lookup)
 314 {
 315         struct rt6_info *rt;
 316
 317         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 318         if (rt->dst.error == -EAGAIN) {
 319                 ip6_rt_put_flags(rt, flags);
 320                 rt = net->ipv6.ip6_null_entry;
 321                 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
 322                         dst_hold(&rt->dst);
 323         }
 324
 325         return &rt->dst;
 326 }
 327
 328 /* called with rcu lock held; no reference taken on fib6_info */
 329 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 330                 struct fib6_result *res, int flags)
 331 {
 332         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
 333                                  res, flags);
 334 }
 335
 336 static void __net_init fib6_tables_init(struct net *net)
 337 {
 338         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 339 }
 340
 341 #endif
 342
 343 unsigned int fib6_tables_seq_read(struct net *net)
 344 {
 345         unsigned int h, fib_seq = 0;
 346
 347         rcu_read_lock();
 348         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 349                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 350                 struct fib6_table *tb;
 351
 352                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 353                         fib_seq += tb->fib_seq;
 354         }
 355         rcu_read_unlock();
 356
 357         return fib_seq;
 358 }
 359
 360 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 361                                     enum fib_event_type event_type,
 362                                     struct fib6_info *rt)
 363 {
 364         struct fib6_entry_notifier_info info = {
 365                 .rt = rt,
 366         };
 367
 368         return call_fib6_notifier(nb, net, event_type, &info.info);
 369 }
 370
 371 int call_fib6_entry_notifiers(struct net *net,
 372                               enum fib_event_type event_type,
 373                               struct fib6_info *rt,
 374                               struct netlink_ext_ack *extack)
 375 {
 376         struct fib6_entry_notifier_info info = {
 377                 .info.extack = extack,
 378                 .rt = rt,
 379         };
 380
 381         rt->fib6_table->fib_seq++;
 382         return call_fib6_notifiers(net, event_type, &info.info);
 383 }
 384
 385 int call_fib6_multipath_entry_notifiers(struct net *net,
 386                                         enum fib_event_type event_type,
 387                                         struct fib6_info *rt,
 388                                         unsigned int nsiblings,
 389                                         struct netlink_ext_ack *extack)
 390 {
 391         struct fib6_entry_notifier_info info = {
 392                 .info.extack = extack,
 393                 .rt = rt,
 394                 .nsiblings = nsiblings,
 395         };
 396
 397         rt->fib6_table->fib_seq++;
 398         return call_fib6_notifiers(net, event_type, &info.info);
 399 }
 400
 401 struct fib6_dump_arg {
 402         struct net *net;
 403         struct notifier_block *nb;
 404 };
 405
 406 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 407 {
 408         if (rt == arg->net->ipv6.fib6_null_entry)
 409                 return;
 410         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 411 }
 412
 413 static int fib6_node_dump(struct fib6_walker *w)
 414 {
 415         struct fib6_info *rt;
 416
 417         for_each_fib6_walker_rt(w)
 418                 fib6_rt_dump(rt, w->args);
 419         w->leaf = NULL;
 420         return 0;
 421 }
 422
 423 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 424                             struct fib6_walker *w)
 425 {
 426         w->root = &tb->tb6_root;
 427         spin_lock_bh(&tb->tb6_lock);
 428         fib6_walk(net, w);
 429         spin_unlock_bh(&tb->tb6_lock);
 430 }
 431
 432 /* Called with rcu_read_lock() */
 433 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 434 {
 435         struct fib6_dump_arg arg;
 436         struct fib6_walker *w;
 437         unsigned int h;
 438
 439         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 440         if (!w)
 441                 return -ENOMEM;
 442
 443         w->func = fib6_node_dump;
 444         arg.net = net;
 445         arg.nb = nb;
 446         w->args = &arg;
 447
 448         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 449                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 450                 struct fib6_table *tb;
 451
 452                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 453                         fib6_table_dump(net, tb, w);
 454         }
 455
 456         kfree(w);
 457
 458         return 0;
 459 }
 460
 461 static int fib6_dump_node(struct fib6_walker *w)
 462 {
 463         int res;
 464         struct fib6_info *rt;
 465
 466         for_each_fib6_walker_rt(w) {
 467                 res = rt6_dump_route(rt, w->args, w->skip_in_node);
 468                 if (res >= 0) {
 469                         /* Frame is full, suspend walking */
 470                         w->leaf = rt;
 471
 472                         /* We'll restart from this node, so if some routes were
 473                          * already dumped, skip them next time.
 474                          */
 475                         w->skip_in_node += res;
 476
 477                         return 1;
 478                 }
 479                 w->skip_in_node = 0;
 480
 481                 /* Multipath routes are dumped in one route with the
 482                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 483                  * last sibling of this route (no need to dump the
 484                  * sibling routes again)
 485                  */
 486                 if (rt->fib6_nsiblings)
 487                         rt = list_last_entry(&rt->fib6_siblings,
 488                                              struct fib6_info,
 489                                              fib6_siblings);
 490         }
 491         w->leaf = NULL;
 492         return 0;
 493 }
 494
 495 static void fib6_dump_end(struct netlink_callback *cb)
 496 {
 497         struct net *net = sock_net(cb->skb->sk);
 498         struct fib6_walker *w = (void *)cb->args[2];
 499
 500         if (w) {
 501                 if (cb->args[4]) {
 502                         cb->args[4] = 0;
 503                         fib6_walker_unlink(net, w);
 504                 }
 505                 cb->args[2] = 0;
 506                 kfree(w);
 507         }
 508         cb->done = (void *)cb->args[3];
 509         cb->args[1] = 3;
 510 }
 511
 512 static int fib6_dump_done(struct netlink_callback *cb)
 513 {
 514         fib6_dump_end(cb);
 515         return cb->done ? cb->done(cb) : 0;
 516 }
 517
 518 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 519                            struct netlink_callback *cb)
 520 {
 521         struct net *net = sock_net(skb->sk);
 522         struct fib6_walker *w;
 523         int res;
 524
 525         w = (void *)cb->args[2];
 526         w->root = &table->tb6_root;
 527
 528         if (cb->args[4] == 0) {
 529                 w->count = 0;
 530                 w->skip = 0;
 531                 w->skip_in_node = 0;
 532
 533                 spin_lock_bh(&table->tb6_lock);
 534                 res = fib6_walk(net, w);
 535                 spin_unlock_bh(&table->tb6_lock);
 536                 if (res > 0) {
 537                         cb->args[4] = 1;
 538                         cb->args[5] = w->root->fn_sernum;
 539                 }
 540         } else {
 541                 if (cb->args[5] != w->root->fn_sernum) {
 542                         /* Begin at the root if the tree changed */
 543                         cb->args[5] = w->root->fn_sernum;
 544                         w->state = FWS_INIT;
 545                         w->node = w->root;
 546                         w->skip = w->count;
 547                         w->skip_in_node = 0;
 548                 } else
 549                         w->skip = 0;
 550
 551                 spin_lock_bh(&table->tb6_lock);
 552                 res = fib6_walk_continue(w);
 553                 spin_unlock_bh(&table->tb6_lock);
 554                 if (res <= 0) {
 555                         fib6_walker_unlink(net, w);
 556                         cb->args[4] = 0;
 557                 }
 558         }
 559
 560         return res;
 561 }
 562
 563 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 564 {
 565         struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
 566                                          .filter.dump_routes = true };
 567         const struct nlmsghdr *nlh = cb->nlh;
 568         struct net *net = sock_net(skb->sk);
 569         unsigned int h, s_h;
 570         unsigned int e = 0, s_e;
 571         struct fib6_walker *w;
 572         struct fib6_table *tb;
 573         struct hlist_head *head;
 574         int res = 0;
 575
 576         if (cb->strict_check) {
 577                 int err;
 578
 579                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 580                 if (err < 0)
 581                         return err;
 582         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 583                 struct rtmsg *rtm = nlmsg_data(nlh);
 584
 585                 if (rtm->rtm_flags & RTM_F_PREFIX)
 586                         arg.filter.flags = RTM_F_PREFIX;
 587         }
 588
 589         w = (void *)cb->args[2];
 590         if (!w) {
 591                 /* New dump:
 592                  *
 593                  * 1. hook callback destructor.
 594                  */
 595                 cb->args[3] = (long)cb->done;
 596                 cb->done = fib6_dump_done;
 597
 598                 /*
 599                  * 2. allocate and initialize walker.
 600                  */
 601                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 602                 if (!w)
 603                         return -ENOMEM;
 604                 w->func = fib6_dump_node;
 605                 cb->args[2] = (long)w;
 606         }
 607
 608         arg.skb = skb;
 609         arg.cb = cb;
 610         arg.net = net;
 611         w->args = &arg;
 612
 613         if (arg.filter.table_id) {
 614                 tb = fib6_get_table(net, arg.filter.table_id);
 615                 if (!tb) {
 616                         if (arg.filter.dump_all_families)
 617                                 goto out;
 618
 619                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 620                         return -ENOENT;
 621                 }
 622
 623                 if (!cb->args[0]) {
 624                         res = fib6_dump_table(tb, skb, cb);
 625                         if (!res)
 626                                 cb->args[0] = 1;
 627                 }
 628                 goto out;
 629         }
 630
 631         s_h = cb->args[0];
 632         s_e = cb->args[1];
 633
 634         rcu_read_lock();
 635         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 636                 e = 0;
 637                 head = &net->ipv6.fib_table_hash[h];
 638                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 639                         if (e < s_e)
 640                                 goto next;
 641                         res = fib6_dump_table(tb, skb, cb);
 642                         if (res != 0)
 643                                 goto out_unlock;
 644 next:
 645                         e++;
 646                 }
 647         }
 648 out_unlock:
 649         rcu_read_unlock();
 650         cb->args[1] = e;
 651         cb->args[0] = h;
 652 out:
 653         res = res < 0 ? res : skb->len;
 654         if (res <= 0)
 655                 fib6_dump_end(cb);
 656         return res;
 657 }
 658
 659 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 660 {
 661         if (!f6i)
 662                 return;
 663
 664         if (f6i->fib6_metrics == &dst_default_metrics) {
 665                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 666
 667                 if (!p)
 668                         return;
 669
 670                 refcount_set(&p->refcnt, 1);
 671                 f6i->fib6_metrics = p;
 672         }
 673
 674         f6i->fib6_metrics->metrics[metric - 1] = val;
 675 }
 676
 677 /*
 678  *      Routing Table
 679  *
 680  *      return the appropriate node for a routing tree "add" operation
 681  *      by either creating and inserting or by returning an existing
 682  *      node.
 683  */
 684
 685 static struct fib6_node *fib6_add_1(struct net *net,
 686                                     struct fib6_table *table,
 687                                     struct fib6_node *root,
 688                                     struct in6_addr *addr, int plen,
 689                                     int offset, int allow_create,
 690                                     int replace_required,
 691                                     struct netlink_ext_ack *extack)
 692 {
 693         struct fib6_node *fn, *in, *ln;
 694         struct fib6_node *pn = NULL;
 695         struct rt6key *key;
 696         int     bit;
 697         __be32  dir = 0;
 698
 699         RT6_TRACE("fib6_add_1\n");
 700
 701         /* insert node in tree */
 702
 703         fn = root;
 704
 705         do {
 706                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 707                                             lockdep_is_held(&table->tb6_lock));
 708                 key = (struct rt6key *)((u8 *)leaf + offset);
 709
 710                 /*
 711                  *      Prefix match
 712                  */
 713                 if (plen < fn->fn_bit ||
 714                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 715                         if (!allow_create) {
 716                                 if (replace_required) {
 717                                         NL_SET_ERR_MSG(extack,
 718                                                        "Can not replace route - no match found");
 719                                         pr_warn("Can't replace route, no match found\n");
 720                                         return ERR_PTR(-ENOENT);
 721                                 }
 722                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 723                         }
 724                         goto insert_above;
 725                 }
 726
 727                 /*
 728                  *      Exact match ?
 729                  */
 730
 731                 if (plen == fn->fn_bit) {
 732                         /* clean up an intermediate node */
 733                         if (!(fn->fn_flags & RTN_RTINFO)) {
 734                                 RCU_INIT_POINTER(fn->leaf, NULL);
 735                                 fib6_info_release(leaf);
 736                         /* remove null_entry in the root node */
 737                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 738                                    rcu_access_pointer(fn->leaf) ==
 739                                    net->ipv6.fib6_null_entry) {
 740                                 RCU_INIT_POINTER(fn->leaf, NULL);
 741                         }
 742
 743                         return fn;
 744                 }
 745
 746                 /*
 747                  *      We have more bits to go
 748                  */
 749
 750                 /* Try to walk down on tree. */
 751                 dir = addr_bit_set(addr, fn->fn_bit);
 752                 pn = fn;
 753                 fn = dir ?
 754                      rcu_dereference_protected(fn->right,
 755                                         lockdep_is_held(&table->tb6_lock)) :
 756                      rcu_dereference_protected(fn->left,
 757                                         lockdep_is_held(&table->tb6_lock));
 758         } while (fn);
 759
 760         if (!allow_create) {
 761                 /* We should not create new node because
 762                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 763                  * I assume it is safe to require NLM_F_CREATE when
 764                  * REPLACE flag is used! Later we may want to remove the
 765                  * check for replace_required, because according
 766                  * to netlink specification, NLM_F_CREATE
 767                  * MUST be specified if new route is created.
 768                  * That would keep IPv6 consistent with IPv4
 769                  */
 770                 if (replace_required) {
 771                         NL_SET_ERR_MSG(extack,
 772                                        "Can not replace route - no match found");
 773                         pr_warn("Can't replace route, no match found\n");
 774                         return ERR_PTR(-ENOENT);
 775                 }
 776                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 777         }
 778         /*
 779          *      We walked to the bottom of tree.
 780          *      Create new leaf node without children.
 781          */
 782
 783         ln = node_alloc(net);
 784
 785         if (!ln)
 786                 return ERR_PTR(-ENOMEM);
 787         ln->fn_bit = plen;
 788         RCU_INIT_POINTER(ln->parent, pn);
 789
 790         if (dir)
 791                 rcu_assign_pointer(pn->right, ln);
 792         else
 793                 rcu_assign_pointer(pn->left, ln);
 794
 795         return ln;
 796
 797
 798 insert_above:
 799         /*
 800          * split since we don't have a common prefix anymore or
 801          * we have a less significant route.
 802          * we've to insert an intermediate node on the list
 803          * this new node will point to the one we need to create
 804          * and the current
 805          */
 806
 807         pn = rcu_dereference_protected(fn->parent,
 808                                        lockdep_is_held(&table->tb6_lock));
 809
 810         /* find 1st bit in difference between the 2 addrs.
 811
 812            See comment in __ipv6_addr_diff: bit may be an invalid value,
 813            but if it is >= plen, the value is ignored in any case.
 814          */
 815
 816         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 817
 818         /*
 819          *              (intermediate)[in]
 820          *                /        \
 821          *      (new leaf node)[ln] (old node)[fn]
 822          */
 823         if (plen > bit) {
 824                 in = node_alloc(net);
 825                 ln = node_alloc(net);
 826
 827                 if (!in || !ln) {
 828                         if (in)
 829                                 node_free_immediate(net, in);
 830                         if (ln)
 831                                 node_free_immediate(net, ln);
 832                         return ERR_PTR(-ENOMEM);
 833                 }
 834
 835                 /*
 836                  * new intermediate node.
 837                  * RTN_RTINFO will
 838                  * be off since that an address that chooses one of
 839                  * the branches would not match less specific routes
 840                  * in the other branch
 841                  */
 842
 843                 in->fn_bit = bit;
 844
 845                 RCU_INIT_POINTER(in->parent, pn);
 846                 in->leaf = fn->leaf;
 847                 fib6_info_hold(rcu_dereference_protected(in->leaf,
 848                                 lockdep_is_held(&table->tb6_lock)));
 849
 850                 /* update parent pointer */
 851                 if (dir)
 852                         rcu_assign_pointer(pn->right, in);
 853                 else
 854                         rcu_assign_pointer(pn->left, in);
 855
 856                 ln->fn_bit = plen;
 857
 858                 RCU_INIT_POINTER(ln->parent, in);
 859                 rcu_assign_pointer(fn->parent, in);
 860
 861                 if (addr_bit_set(addr, bit)) {
 862                         rcu_assign_pointer(in->right, ln);
 863                         rcu_assign_pointer(in->left, fn);
 864                 } else {
 865                         rcu_assign_pointer(in->left, ln);
 866                         rcu_assign_pointer(in->right, fn);
 867                 }
 868         } else { /* plen <= bit */
 869
 870                 /*
 871                  *              (new leaf node)[ln]
 872                  *                /        \
 873                  *           (old node)[fn] NULL
 874                  */
 875
 876                 ln = node_alloc(net);
 877
 878                 if (!ln)
 879                         return ERR_PTR(-ENOMEM);
 880
 881                 ln->fn_bit = plen;
 882
 883                 RCU_INIT_POINTER(ln->parent, pn);
 884
 885                 if (addr_bit_set(&key->addr, plen))
 886                         RCU_INIT_POINTER(ln->right, fn);
 887                 else
 888                         RCU_INIT_POINTER(ln->left, fn);
 889
 890                 rcu_assign_pointer(fn->parent, ln);
 891
 892                 if (dir)
 893                         rcu_assign_pointer(pn->right, ln);
 894                 else
 895                         rcu_assign_pointer(pn->left, ln);
 896         }
 897         return ln;
 898 }
 899
 900 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
 901                                   const struct fib6_info *match,
 902                                   const struct fib6_table *table)
 903 {
 904         int cpu;
 905
 906         if (!fib6_nh->rt6i_pcpu)
 907                 return;
 908
 909         /* release the reference to this fib entry from
 910          * all of its cached pcpu routes
 911          */
 912         for_each_possible_cpu(cpu) {
 913                 struct rt6_info **ppcpu_rt;
 914                 struct rt6_info *pcpu_rt;
 915
 916                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
 917                 pcpu_rt = *ppcpu_rt;
 918
 919                 /* only dropping the 'from' reference if the cached route
 920                  * is using 'match'. The cached pcpu_rt->from only changes
 921                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
 922                  * change from one fib6_info reference to another
 923                  */
 924                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
 925                         struct fib6_info *from;
 926
 927                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 928                         fib6_info_release(from);
 929                 }
 930         }
 931 }
 932
 933 struct fib6_nh_pcpu_arg {
 934         struct fib6_info        *from;
 935         const struct fib6_table *table;
 936 };
 937
 938 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
 939 {
 940         struct fib6_nh_pcpu_arg *arg = _arg;
 941
 942         __fib6_drop_pcpu_from(nh, arg->from, arg->table);
 943         return 0;
 944 }
 945
 946 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
 947                                 const struct fib6_table *table)
 948 {
 949         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
 950          * while we are cleaning them here.
 951          */
 952         f6i->fib6_destroying = 1;
 953         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
 954
 955         if (f6i->nh) {
 956                 struct fib6_nh_pcpu_arg arg = {
 957                         .from = f6i,
 958                         .table = table
 959                 };
 960
 961                 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
 962                                          &arg);
 963         } else {
 964                 struct fib6_nh *fib6_nh;
 965
 966                 fib6_nh = f6i->fib6_nh;
 967                 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
 968         }
 969 }
 970
 971 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
 972                           struct net *net)
 973 {
 974         struct fib6_table *table = rt->fib6_table;
 975
 976         fib6_drop_pcpu_from(rt, table);
 977
 978         if (rt->nh && !list_empty(&rt->nh_list))
 979                 list_del_init(&rt->nh_list);
 980
 981         if (refcount_read(&rt->fib6_ref) != 1) {
 982                 /* This route is used as dummy address holder in some split
 983                  * nodes. It is not leaked, but it still holds other resources,
 984                  * which must be released in time. So, scan ascendant nodes
 985                  * and replace dummy references to this route with references
 986                  * to still alive ones.
 987                  */
 988                 while (fn) {
 989                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 990                                             lockdep_is_held(&table->tb6_lock));
 991                         struct fib6_info *new_leaf;
 992                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 993                                 new_leaf = fib6_find_prefix(net, table, fn);
 994                                 fib6_info_hold(new_leaf);
 995
 996                                 rcu_assign_pointer(fn->leaf, new_leaf);
 997                                 fib6_info_release(rt);
 998                         }
 999                         fn = rcu_dereference_protected(fn->parent,
1000                                     lockdep_is_held(&table->tb6_lock));
1001                 }
1002         }
1003 }
1004
1005 /*
1006  *      Insert routing information in a node.
1007  */
1008
1009 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1010                             struct nl_info *info,
1011                             struct netlink_ext_ack *extack)
1012 {
1013         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1014                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1015         struct fib6_info *iter = NULL;
1016         struct fib6_info __rcu **ins;
1017         struct fib6_info __rcu **fallback_ins = NULL;
1018         int replace = (info->nlh &&
1019                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1020         int add = (!info->nlh ||
1021                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
1022         int found = 0;
1023         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1024         u16 nlflags = NLM_F_EXCL;
1025         int err;
1026
1027         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1028                 nlflags |= NLM_F_APPEND;
1029
1030         ins = &fn->leaf;
1031
1032         for (iter = leaf; iter;
1033              iter = rcu_dereference_protected(iter->fib6_next,
1034                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1035                 /*
1036                  *      Search for duplicates
1037                  */
1038
1039                 if (iter->fib6_metric == rt->fib6_metric) {
1040                         /*
1041                          *      Same priority level
1042                          */
1043                         if (info->nlh &&
1044                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1045                                 return -EEXIST;
1046
1047                         nlflags &= ~NLM_F_EXCL;
1048                         if (replace) {
1049                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1050                                         found++;
1051                                         break;
1052                                 }
1053                                 if (rt_can_ecmp)
1054                                         fallback_ins = fallback_ins ?: ins;
1055                                 goto next_iter;
1056                         }
1057
1058                         if (rt6_duplicate_nexthop(iter, rt)) {
1059                                 if (rt->fib6_nsiblings)
1060                                         rt->fib6_nsiblings = 0;
1061                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1062                                         return -EEXIST;
1063                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1064                                         fib6_clean_expires(iter);
1065                                 else
1066                                         fib6_set_expires(iter, rt->expires);
1067
1068                                 if (rt->fib6_pmtu)
1069                                         fib6_metric_set(iter, RTAX_MTU,
1070                                                         rt->fib6_pmtu);
1071                                 return -EEXIST;
1072                         }
1073                         /* If we have the same destination and the same metric,
1074                          * but not the same gateway, then the route we try to
1075                          * add is sibling to this route, increment our counter
1076                          * of siblings, and later we will add our route to the
1077                          * list.
1078                          * Only static routes (which don't have flag
1079                          * RTF_EXPIRES) are used for ECMPv6.
1080                          *
1081                          * To avoid long list, we only had siblings if the
1082                          * route have a gateway.
1083                          */
1084                         if (rt_can_ecmp &&
1085                             rt6_qualify_for_ecmp(iter))
1086                                 rt->fib6_nsiblings++;
1087                 }
1088
1089                 if (iter->fib6_metric > rt->fib6_metric)
1090                         break;
1091
1092 next_iter:
1093                 ins = &iter->fib6_next;
1094         }
1095
1096         if (fallback_ins && !found) {
1097                 /* No ECMP-able route found, replace first non-ECMP one */
1098                 ins = fallback_ins;
1099                 iter = rcu_dereference_protected(*ins,
1100                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1101                 found++;
1102         }
1103
1104         /* Reset round-robin state, if necessary */
1105         if (ins == &fn->leaf)
1106                 fn->rr_ptr = NULL;
1107
1108         /* Link this route to others same route. */
1109         if (rt->fib6_nsiblings) {
1110                 unsigned int fib6_nsiblings;
1111                 struct fib6_info *sibling, *temp_sibling;
1112
1113                 /* Find the first route that have the same metric */
1114                 sibling = leaf;
1115                 while (sibling) {
1116                         if (sibling->fib6_metric == rt->fib6_metric &&
1117                             rt6_qualify_for_ecmp(sibling)) {
1118                                 list_add_tail(&rt->fib6_siblings,
1119                                               &sibling->fib6_siblings);
1120                                 break;
1121                         }
1122                         sibling = rcu_dereference_protected(sibling->fib6_next,
1123                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1124                 }
1125                 /* For each sibling in the list, increment the counter of
1126                  * siblings. BUG() if counters does not match, list of siblings
1127                  * is broken!
1128                  */
1129                 fib6_nsiblings = 0;
1130                 list_for_each_entry_safe(sibling, temp_sibling,
1131                                          &rt->fib6_siblings, fib6_siblings) {
1132                         sibling->fib6_nsiblings++;
1133                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1134                         fib6_nsiblings++;
1135                 }
1136                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1137                 rt6_multipath_rebalance(temp_sibling);
1138         }
1139
1140         /*
1141          *      insert node
1142          */
1143         if (!replace) {
1144                 if (!add)
1145                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1146
1147 add:
1148                 nlflags |= NLM_F_CREATE;
1149
1150                 if (!info->skip_notify_kernel) {
1151                         err = call_fib6_entry_notifiers(info->nl_net,
1152                                                         FIB_EVENT_ENTRY_ADD,
1153                                                         rt, extack);
1154                         if (err) {
1155                                 struct fib6_info *sibling, *next_sibling;
1156
1157                                 /* If the route has siblings, then it first
1158                                  * needs to be unlinked from them.
1159                                  */
1160                                 if (!rt->fib6_nsiblings)
1161                                         return err;
1162
1163                                 list_for_each_entry_safe(sibling, next_sibling,
1164                                                          &rt->fib6_siblings,
1165                                                          fib6_siblings)
1166                                         sibling->fib6_nsiblings--;
1167                                 rt->fib6_nsiblings = 0;
1168                                 list_del_init(&rt->fib6_siblings);
1169                                 rt6_multipath_rebalance(next_sibling);
1170                                 return err;
1171                         }
1172                 }
1173
1174                 rcu_assign_pointer(rt->fib6_next, iter);
1175                 fib6_info_hold(rt);
1176                 rcu_assign_pointer(rt->fib6_node, fn);
1177                 rcu_assign_pointer(*ins, rt);
1178                 if (!info->skip_notify)
1179                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1180                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1181
1182                 if (!(fn->fn_flags & RTN_RTINFO)) {
1183                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1184                         fn->fn_flags |= RTN_RTINFO;
1185                 }
1186
1187         } else {
1188                 int nsiblings;
1189
1190                 if (!found) {
1191                         if (add)
1192                                 goto add;
1193                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1194                         return -ENOENT;
1195                 }
1196
1197                 if (!info->skip_notify_kernel) {
1198                         err = call_fib6_entry_notifiers(info->nl_net,
1199                                                         FIB_EVENT_ENTRY_REPLACE,
1200                                                         rt, extack);
1201                         if (err)
1202                                 return err;
1203                 }
1204
1205                 fib6_info_hold(rt);
1206                 rcu_assign_pointer(rt->fib6_node, fn);
1207                 rt->fib6_next = iter->fib6_next;
1208                 rcu_assign_pointer(*ins, rt);
1209                 if (!info->skip_notify)
1210                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1211                 if (!(fn->fn_flags & RTN_RTINFO)) {
1212                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1213                         fn->fn_flags |= RTN_RTINFO;
1214                 }
1215                 nsiblings = iter->fib6_nsiblings;
1216                 iter->fib6_node = NULL;
1217                 fib6_purge_rt(iter, fn, info->nl_net);
1218                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1219                         fn->rr_ptr = NULL;
1220                 fib6_info_release(iter);
1221
1222                 if (nsiblings) {
1223                         /* Replacing an ECMP route, remove all siblings */
1224                         ins = &rt->fib6_next;
1225                         iter = rcu_dereference_protected(*ins,
1226                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1227                         while (iter) {
1228                                 if (iter->fib6_metric > rt->fib6_metric)
1229                                         break;
1230                                 if (rt6_qualify_for_ecmp(iter)) {
1231                                         *ins = iter->fib6_next;
1232                                         iter->fib6_node = NULL;
1233                                         fib6_purge_rt(iter, fn, info->nl_net);
1234                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1235                                                 fn->rr_ptr = NULL;
1236                                         fib6_info_release(iter);
1237                                         nsiblings--;
1238                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1239                                 } else {
1240                                         ins = &iter->fib6_next;
1241                                 }
1242                                 iter = rcu_dereference_protected(*ins,
1243                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1244                         }
1245                         WARN_ON(nsiblings != 0);
1246                 }
1247         }
1248
1249         return 0;
1250 }
1251
1252 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1253 {
1254         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1255             (rt->fib6_flags & RTF_EXPIRES))
1256                 mod_timer(&net->ipv6.ip6_fib_timer,
1257                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1258 }
1259
1260 void fib6_force_start_gc(struct net *net)
1261 {
1262         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1263                 mod_timer(&net->ipv6.ip6_fib_timer,
1264                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1265 }
1266
1267 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1268                                            int sernum)
1269 {
1270         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1271                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1272
1273         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1274         smp_wmb();
1275         while (fn) {
1276                 fn->fn_sernum = sernum;
1277                 fn = rcu_dereference_protected(fn->parent,
1278                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1279         }
1280 }
1281
1282 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1283 {
1284         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1285 }
1286
1287 /* allow ipv4 to update sernum via ipv6_stub */
1288 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1289 {
1290         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1291         fib6_update_sernum_upto_root(net, f6i);
1292         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1293 }
1294
1295 /*
1296  *      Add routing information to the routing tree.
1297  *      <destination addr>/<source addr>
1298  *      with source addr info in sub-trees
1299  *      Need to own table->tb6_lock
1300  */
1301
1302 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1303              struct nl_info *info, struct netlink_ext_ack *extack)
1304 {
1305         struct fib6_table *table = rt->fib6_table;
1306         struct fib6_node *fn, *pn = NULL;
1307         int err = -ENOMEM;
1308         int allow_create = 1;
1309         int replace_required = 0;
1310         int sernum = fib6_new_sernum(info->nl_net);
1311
1312         if (info->nlh) {
1313                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1314                         allow_create = 0;
1315                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1316                         replace_required = 1;
1317         }
1318         if (!allow_create && !replace_required)
1319                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1320
1321         fn = fib6_add_1(info->nl_net, table, root,
1322                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1323                         offsetof(struct fib6_info, fib6_dst), allow_create,
1324                         replace_required, extack);
1325         if (IS_ERR(fn)) {
1326                 err = PTR_ERR(fn);
1327                 fn = NULL;
1328                 goto out;
1329         }
1330
1331         pn = fn;
1332
1333 #ifdef CONFIG_IPV6_SUBTREES
1334         if (rt->fib6_src.plen) {
1335                 struct fib6_node *sn;
1336
1337                 if (!rcu_access_pointer(fn->subtree)) {
1338                         struct fib6_node *sfn;
1339
1340                         /*
1341                          * Create subtree.
1342                          *
1343                          *              fn[main tree]
1344                          *              |
1345                          *              sfn[subtree root]
1346                          *                 \
1347                          *                  sn[new leaf node]
1348                          */
1349
1350                         /* Create subtree root node */
1351                         sfn = node_alloc(info->nl_net);
1352                         if (!sfn)
1353                                 goto failure;
1354
1355                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1356                         rcu_assign_pointer(sfn->leaf,
1357                                            info->nl_net->ipv6.fib6_null_entry);
1358                         sfn->fn_flags = RTN_ROOT;
1359
1360                         /* Now add the first leaf node to new subtree */
1361
1362                         sn = fib6_add_1(info->nl_net, table, sfn,
1363                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1364                                         offsetof(struct fib6_info, fib6_src),
1365                                         allow_create, replace_required, extack);
1366
1367                         if (IS_ERR(sn)) {
1368                                 /* If it is failed, discard just allocated
1369                                    root, and then (in failure) stale node
1370                                    in main tree.
1371                                  */
1372                                 node_free_immediate(info->nl_net, sfn);
1373                                 err = PTR_ERR(sn);
1374                                 goto failure;
1375                         }
1376
1377                         /* Now link new subtree to main tree */
1378                         rcu_assign_pointer(sfn->parent, fn);
1379                         rcu_assign_pointer(fn->subtree, sfn);
1380                 } else {
1381                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1382                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1383                                         offsetof(struct fib6_info, fib6_src),
1384                                         allow_create, replace_required, extack);
1385
1386                         if (IS_ERR(sn)) {
1387                                 err = PTR_ERR(sn);
1388                                 goto failure;
1389                         }
1390                 }
1391
1392                 if (!rcu_access_pointer(fn->leaf)) {
1393                         if (fn->fn_flags & RTN_TL_ROOT) {
1394                                 /* put back null_entry for root node */
1395                                 rcu_assign_pointer(fn->leaf,
1396                                             info->nl_net->ipv6.fib6_null_entry);
1397                         } else {
1398                                 fib6_info_hold(rt);
1399                                 rcu_assign_pointer(fn->leaf, rt);
1400                         }
1401                 }
1402                 fn = sn;
1403         }
1404 #endif
1405
1406         err = fib6_add_rt2node(fn, rt, info, extack);
1407         if (!err) {
1408                 if (rt->nh)
1409                         list_add(&rt->nh_list, &rt->nh->f6i_list);
1410                 __fib6_update_sernum_upto_root(rt, sernum);
1411                 fib6_start_gc(info->nl_net, rt);
1412         }
1413
1414 out:
1415         if (err) {
1416 #ifdef CONFIG_IPV6_SUBTREES
1417                 /*
1418                  * If fib6_add_1 has cleared the old leaf pointer in the
1419                  * super-tree leaf node we have to find a new one for it.
1420                  */
1421                 if (pn != fn) {
1422                         struct fib6_info *pn_leaf =
1423                                 rcu_dereference_protected(pn->leaf,
1424                                     lockdep_is_held(&table->tb6_lock));
1425                         if (pn_leaf == rt) {
1426                                 pn_leaf = NULL;
1427                                 RCU_INIT_POINTER(pn->leaf, NULL);
1428                                 fib6_info_release(rt);
1429                         }
1430                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1431                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1432                                                            pn);
1433 #if RT6_DEBUG >= 2
1434                                 if (!pn_leaf) {
1435                                         WARN_ON(!pn_leaf);
1436                                         pn_leaf =
1437                                             info->nl_net->ipv6.fib6_null_entry;
1438                                 }
1439 #endif
1440                                 fib6_info_hold(pn_leaf);
1441                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1442                         }
1443                 }
1444 #endif
1445                 goto failure;
1446         }
1447         return err;
1448
1449 failure:
1450         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1451          * 1. fn is an intermediate node and we failed to add the new
1452          * route to it in both subtree creation failure and fib6_add_rt2node()
1453          * failure case.
1454          * 2. fn is the root node in the table and we fail to add the first
1455          * default route to it.
1456          */
1457         if (fn &&
1458             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1459              (fn->fn_flags & RTN_TL_ROOT &&
1460               !rcu_access_pointer(fn->leaf))))
1461                 fib6_repair_tree(info->nl_net, table, fn);
1462         return err;
1463 }
1464
1465 /*
1466  *      Routing tree lookup
1467  *
1468  */
1469
1470 struct lookup_args {
1471         int                     offset;         /* key offset on fib6_info */
1472         const struct in6_addr   *addr;          /* search key                   */
1473 };
1474
1475 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1476                                             struct lookup_args *args)
1477 {
1478         struct fib6_node *fn;
1479         __be32 dir;
1480
1481         if (unlikely(args->offset == 0))
1482                 return NULL;
1483
1484         /*
1485          *      Descend on a tree
1486          */
1487
1488         fn = root;
1489
1490         for (;;) {
1491                 struct fib6_node *next;
1492
1493                 dir = addr_bit_set(args->addr, fn->fn_bit);
1494
1495                 next = dir ? rcu_dereference(fn->right) :
1496                              rcu_dereference(fn->left);
1497
1498                 if (next) {
1499                         fn = next;
1500                         continue;
1501                 }
1502                 break;
1503         }
1504
1505         while (fn) {
1506                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1507
1508                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1509                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1510                         struct rt6key *key;
1511
1512                         if (!leaf)
1513                                 goto backtrack;
1514
1515                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1516
1517                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1518 #ifdef CONFIG_IPV6_SUBTREES
1519                                 if (subtree) {
1520                                         struct fib6_node *sfn;
1521                                         sfn = fib6_node_lookup_1(subtree,
1522                                                                  args + 1);
1523                                         if (!sfn)
1524                                                 goto backtrack;
1525                                         fn = sfn;
1526                                 }
1527 #endif
1528                                 if (fn->fn_flags & RTN_RTINFO)
1529                                         return fn;
1530                         }
1531                 }
1532 backtrack:
1533                 if (fn->fn_flags & RTN_ROOT)
1534                         break;
1535
1536                 fn = rcu_dereference(fn->parent);
1537         }
1538
1539         return NULL;
1540 }
1541
1542 /* called with rcu_read_lock() held
1543  */
1544 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1545                                    const struct in6_addr *daddr,
1546                                    const struct in6_addr *saddr)
1547 {
1548         struct fib6_node *fn;
1549         struct lookup_args args[] = {
1550                 {
1551                         .offset = offsetof(struct fib6_info, fib6_dst),
1552                         .addr = daddr,
1553                 },
1554 #ifdef CONFIG_IPV6_SUBTREES
1555                 {
1556                         .offset = offsetof(struct fib6_info, fib6_src),
1557                         .addr = saddr,
1558                 },
1559 #endif
1560                 {
1561                         .offset = 0,    /* sentinel */
1562                 }
1563         };
1564
1565         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1566         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1567                 fn = root;
1568
1569         return fn;
1570 }
1571
1572 /*
1573  *      Get node with specified destination prefix (and source prefix,
1574  *      if subtrees are used)
1575  *      exact_match == true means we try to find fn with exact match of
1576  *      the passed in prefix addr
1577  *      exact_match == false means we try to find fn with longest prefix
1578  *      match of the passed in prefix addr. This is useful for finding fn
1579  *      for cached route as it will be stored in the exception table under
1580  *      the node with longest prefix length.
1581  */
1582
1583
1584 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1585                                        const struct in6_addr *addr,
1586                                        int plen, int offset,
1587                                        bool exact_match)
1588 {
1589         struct fib6_node *fn, *prev = NULL;
1590
1591         for (fn = root; fn ; ) {
1592                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1593                 struct rt6key *key;
1594
1595                 /* This node is being deleted */
1596                 if (!leaf) {
1597                         if (plen <= fn->fn_bit)
1598                                 goto out;
1599                         else
1600                                 goto next;
1601                 }
1602
1603                 key = (struct rt6key *)((u8 *)leaf + offset);
1604
1605                 /*
1606                  *      Prefix match
1607                  */
1608                 if (plen < fn->fn_bit ||
1609                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1610                         goto out;
1611
1612                 if (plen == fn->fn_bit)
1613                         return fn;
1614
1615                 if (fn->fn_flags & RTN_RTINFO)
1616                         prev = fn;
1617
1618 next:
1619                 /*
1620                  *      We have more bits to go
1621                  */
1622                 if (addr_bit_set(addr, fn->fn_bit))
1623                         fn = rcu_dereference(fn->right);
1624                 else
1625                         fn = rcu_dereference(fn->left);
1626         }
1627 out:
1628         if (exact_match)
1629                 return NULL;
1630         else
1631                 return prev;
1632 }
1633
1634 struct fib6_node *fib6_locate(struct fib6_node *root,
1635                               const struct in6_addr *daddr, int dst_len,
1636                               const struct in6_addr *saddr, int src_len,
1637                               bool exact_match)
1638 {
1639         struct fib6_node *fn;
1640
1641         fn = fib6_locate_1(root, daddr, dst_len,
1642                            offsetof(struct fib6_info, fib6_dst),
1643                            exact_match);
1644
1645 #ifdef CONFIG_IPV6_SUBTREES
1646         if (src_len) {
1647                 WARN_ON(saddr == NULL);
1648                 if (fn) {
1649                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1650
1651                         if (subtree) {
1652                                 fn = fib6_locate_1(subtree, saddr, src_len,
1653                                            offsetof(struct fib6_info, fib6_src),
1654                                            exact_match);
1655                         }
1656                 }
1657         }
1658 #endif
1659
1660         if (fn && fn->fn_flags & RTN_RTINFO)
1661                 return fn;
1662
1663         return NULL;
1664 }
1665
1666
1667 /*
1668  *      Deletion
1669  *
1670  */
1671
1672 static struct fib6_info *fib6_find_prefix(struct net *net,
1673                                          struct fib6_table *table,
1674                                          struct fib6_node *fn)
1675 {
1676         struct fib6_node *child_left, *child_right;
1677
1678         if (fn->fn_flags & RTN_ROOT)
1679                 return net->ipv6.fib6_null_entry;
1680
1681         while (fn) {
1682                 child_left = rcu_dereference_protected(fn->left,
1683                                     lockdep_is_held(&table->tb6_lock));
1684                 child_right = rcu_dereference_protected(fn->right,
1685                                     lockdep_is_held(&table->tb6_lock));
1686                 if (child_left)
1687                         return rcu_dereference_protected(child_left->leaf,
1688                                         lockdep_is_held(&table->tb6_lock));
1689                 if (child_right)
1690                         return rcu_dereference_protected(child_right->leaf,
1691                                         lockdep_is_held(&table->tb6_lock));
1692
1693                 fn = FIB6_SUBTREE(fn);
1694         }
1695         return NULL;
1696 }
1697
1698 /*
1699  *      Called to trim the tree of intermediate nodes when possible. "fn"
1700  *      is the node we want to try and remove.
1701  *      Need to own table->tb6_lock
1702  */
1703
1704 static struct fib6_node *fib6_repair_tree(struct net *net,
1705                                           struct fib6_table *table,
1706                                           struct fib6_node *fn)
1707 {
1708         int children;
1709         int nstate;
1710         struct fib6_node *child;
1711         struct fib6_walker *w;
1712         int iter = 0;
1713
1714         /* Set fn->leaf to null_entry for root node. */
1715         if (fn->fn_flags & RTN_TL_ROOT) {
1716                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1717                 return fn;
1718         }
1719
1720         for (;;) {
1721                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1722                                             lockdep_is_held(&table->tb6_lock));
1723                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1724                                             lockdep_is_held(&table->tb6_lock));
1725                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1726                                             lockdep_is_held(&table->tb6_lock));
1727                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1728                                             lockdep_is_held(&table->tb6_lock));
1729                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1730                                             lockdep_is_held(&table->tb6_lock));
1731                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1732                                             lockdep_is_held(&table->tb6_lock));
1733                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1734                                             lockdep_is_held(&table->tb6_lock));
1735                 struct fib6_info *new_fn_leaf;
1736
1737                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1738                 iter++;
1739
1740                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1741                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1742                 WARN_ON(fn_leaf);
1743
1744                 children = 0;
1745                 child = NULL;
1746                 if (fn_r)
1747                         child = fn_r, children |= 1;
1748                 if (fn_l)
1749                         child = fn_l, children |= 2;
1750
1751                 if (children == 3 || FIB6_SUBTREE(fn)
1752 #ifdef CONFIG_IPV6_SUBTREES
1753                     /* Subtree root (i.e. fn) may have one child */
1754                     || (children && fn->fn_flags & RTN_ROOT)
1755 #endif
1756                     ) {
1757                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1758 #if RT6_DEBUG >= 2
1759                         if (!new_fn_leaf) {
1760                                 WARN_ON(!new_fn_leaf);
1761                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1762                         }
1763 #endif
1764                         fib6_info_hold(new_fn_leaf);
1765                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1766                         return pn;
1767                 }
1768
1769 #ifdef CONFIG_IPV6_SUBTREES
1770                 if (FIB6_SUBTREE(pn) == fn) {
1771                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1772                         RCU_INIT_POINTER(pn->subtree, NULL);
1773                         nstate = FWS_L;
1774                 } else {
1775                         WARN_ON(fn->fn_flags & RTN_ROOT);
1776 #endif
1777                         if (pn_r == fn)
1778                                 rcu_assign_pointer(pn->right, child);
1779                         else if (pn_l == fn)
1780                                 rcu_assign_pointer(pn->left, child);
1781 #if RT6_DEBUG >= 2
1782                         else
1783                                 WARN_ON(1);
1784 #endif
1785                         if (child)
1786                                 rcu_assign_pointer(child->parent, pn);
1787                         nstate = FWS_R;
1788 #ifdef CONFIG_IPV6_SUBTREES
1789                 }
1790 #endif
1791
1792                 read_lock(&net->ipv6.fib6_walker_lock);
1793                 FOR_WALKERS(net, w) {
1794                         if (!child) {
1795                                 if (w->node == fn) {
1796                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1797                                         w->node = pn;
1798                                         w->state = nstate;
1799                                 }
1800                         } else {
1801                                 if (w->node == fn) {
1802                                         w->node = child;
1803                                         if (children&2) {
1804                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1805                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1806                                         } else {
1807                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1808                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1809                                         }
1810                                 }
1811                         }
1812                 }
1813                 read_unlock(&net->ipv6.fib6_walker_lock);
1814
1815                 node_free(net, fn);
1816                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1817                         return pn;
1818
1819                 RCU_INIT_POINTER(pn->leaf, NULL);
1820                 fib6_info_release(pn_leaf);
1821                 fn = pn;
1822         }
1823 }
1824
1825 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1826                            struct fib6_info __rcu **rtp, struct nl_info *info)
1827 {
1828         struct fib6_walker *w;
1829         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1830                                     lockdep_is_held(&table->tb6_lock));
1831         struct net *net = info->nl_net;
1832
1833         RT6_TRACE("fib6_del_route\n");
1834
1835         /* Unlink it */
1836         *rtp = rt->fib6_next;
1837         rt->fib6_node = NULL;
1838         net->ipv6.rt6_stats->fib_rt_entries--;
1839         net->ipv6.rt6_stats->fib_discarded_routes++;
1840
1841         /* Flush all cached dst in exception table */
1842         rt6_flush_exceptions(rt);
1843
1844         /* Reset round-robin state, if necessary */
1845         if (rcu_access_pointer(fn->rr_ptr) == rt)
1846                 fn->rr_ptr = NULL;
1847
1848         /* Remove this entry from other siblings */
1849         if (rt->fib6_nsiblings) {
1850                 struct fib6_info *sibling, *next_sibling;
1851
1852                 list_for_each_entry_safe(sibling, next_sibling,
1853                                          &rt->fib6_siblings, fib6_siblings)
1854                         sibling->fib6_nsiblings--;
1855                 rt->fib6_nsiblings = 0;
1856                 list_del_init(&rt->fib6_siblings);
1857                 rt6_multipath_rebalance(next_sibling);
1858         }
1859
1860         /* Adjust walkers */
1861         read_lock(&net->ipv6.fib6_walker_lock);
1862         FOR_WALKERS(net, w) {
1863                 if (w->state == FWS_C && w->leaf == rt) {
1864                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1865                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1866                                             lockdep_is_held(&table->tb6_lock));
1867                         if (!w->leaf)
1868                                 w->state = FWS_U;
1869                 }
1870         }
1871         read_unlock(&net->ipv6.fib6_walker_lock);
1872
1873         /* If it was last route, call fib6_repair_tree() to:
1874          * 1. For root node, put back null_entry as how the table was created.
1875          * 2. For other nodes, expunge its radix tree node.
1876          */
1877         if (!rcu_access_pointer(fn->leaf)) {
1878                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1879                         fn->fn_flags &= ~RTN_RTINFO;
1880                         net->ipv6.rt6_stats->fib_route_nodes--;
1881                 }
1882                 fn = fib6_repair_tree(net, table, fn);
1883         }
1884
1885         fib6_purge_rt(rt, fn, net);
1886
1887         if (!info->skip_notify_kernel)
1888                 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1889         if (!info->skip_notify)
1890                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1891
1892         fib6_info_release(rt);
1893 }
1894
1895 /* Need to own table->tb6_lock */
1896 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1897 {
1898         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1899                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1900         struct fib6_table *table = rt->fib6_table;
1901         struct net *net = info->nl_net;
1902         struct fib6_info __rcu **rtp;
1903         struct fib6_info __rcu **rtp_next;
1904
1905         if (!fn || rt == net->ipv6.fib6_null_entry)
1906                 return -ENOENT;
1907
1908         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1909
1910         /*
1911          *      Walk the leaf entries looking for ourself
1912          */
1913
1914         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1915                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1916                                         lockdep_is_held(&table->tb6_lock));
1917                 if (rt == cur) {
1918                         fib6_del_route(table, fn, rtp, info);
1919                         return 0;
1920                 }
1921                 rtp_next = &cur->fib6_next;
1922         }
1923         return -ENOENT;
1924 }
1925
1926 /*
1927  *      Tree traversal function.
1928  *
1929  *      Certainly, it is not interrupt safe.
1930  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1931  *      It means, that we can modify tree during walking
1932  *      and use this function for garbage collection, clone pruning,
1933  *      cleaning tree when a device goes down etc. etc.
1934  *
1935  *      It guarantees that every node will be traversed,
1936  *      and that it will be traversed only once.
1937  *
1938  *      Callback function w->func may return:
1939  *      0 -> continue walking.
1940  *      positive value -> walking is suspended (used by tree dumps,
1941  *      and probably by gc, if it will be split to several slices)
1942  *      negative value -> terminate walking.
1943  *
1944  *      The function itself returns:
1945  *      0   -> walk is complete.
1946  *      >0  -> walk is incomplete (i.e. suspended)
1947  *      <0  -> walk is terminated by an error.
1948  *
1949  *      This function is called with tb6_lock held.
1950  */
1951
1952 static int fib6_walk_continue(struct fib6_walker *w)
1953 {
1954         struct fib6_node *fn, *pn, *left, *right;
1955
1956         /* w->root should always be table->tb6_root */
1957         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1958
1959         for (;;) {
1960                 fn = w->node;
1961                 if (!fn)
1962                         return 0;
1963
1964                 switch (w->state) {
1965 #ifdef CONFIG_IPV6_SUBTREES
1966                 case FWS_S:
1967                         if (FIB6_SUBTREE(fn)) {
1968                                 w->node = FIB6_SUBTREE(fn);
1969                                 continue;
1970                         }
1971                         w->state = FWS_L;
1972 #endif
1973                         /* fall through */
1974                 case FWS_L:
1975                         left = rcu_dereference_protected(fn->left, 1);
1976                         if (left) {
1977                                 w->node = left;
1978                                 w->state = FWS_INIT;
1979                                 continue;
1980                         }
1981                         w->state = FWS_R;
1982                         /* fall through */
1983                 case FWS_R:
1984                         right = rcu_dereference_protected(fn->right, 1);
1985                         if (right) {
1986                                 w->node = right;
1987                                 w->state = FWS_INIT;
1988                                 continue;
1989                         }
1990                         w->state = FWS_C;
1991                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1992                         /* fall through */
1993                 case FWS_C:
1994                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1995                                 int err;
1996
1997                                 if (w->skip) {
1998                                         w->skip--;
1999                                         goto skip;
2000                                 }
2001
2002                                 err = w->func(w);
2003                                 if (err)
2004                                         return err;
2005
2006                                 w->count++;
2007                                 continue;
2008                         }
2009 skip:
2010                         w->state = FWS_U;
2011                         /* fall through */
2012                 case FWS_U:
2013                         if (fn == w->root)
2014                                 return 0;
2015                         pn = rcu_dereference_protected(fn->parent, 1);
2016                         left = rcu_dereference_protected(pn->left, 1);
2017                         right = rcu_dereference_protected(pn->right, 1);
2018                         w->node = pn;
2019 #ifdef CONFIG_IPV6_SUBTREES
2020                         if (FIB6_SUBTREE(pn) == fn) {
2021                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2022                                 w->state = FWS_L;
2023                                 continue;
2024                         }
2025 #endif
2026                         if (left == fn) {
2027                                 w->state = FWS_R;
2028                                 continue;
2029                         }
2030                         if (right == fn) {
2031                                 w->state = FWS_C;
2032                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2033                                 continue;
2034                         }
2035 #if RT6_DEBUG >= 2
2036                         WARN_ON(1);
2037 #endif
2038                 }
2039         }
2040 }
2041
2042 static int fib6_walk(struct net *net, struct fib6_walker *w)
2043 {
2044         int res;
2045
2046         w->state = FWS_INIT;
2047         w->node = w->root;
2048
2049         fib6_walker_link(net, w);
2050         res = fib6_walk_continue(w);
2051         if (res <= 0)
2052                 fib6_walker_unlink(net, w);
2053         return res;
2054 }
2055
2056 static int fib6_clean_node(struct fib6_walker *w)
2057 {
2058         int res;
2059         struct fib6_info *rt;
2060         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2061         struct nl_info info = {
2062                 .nl_net = c->net,
2063                 .skip_notify = c->skip_notify,
2064         };
2065
2066         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2067             w->node->fn_sernum != c->sernum)
2068                 w->node->fn_sernum = c->sernum;
2069
2070         if (!c->func) {
2071                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2072                 w->leaf = NULL;
2073                 return 0;
2074         }
2075
2076         for_each_fib6_walker_rt(w) {
2077                 res = c->func(rt, c->arg);
2078                 if (res == -1) {
2079                         w->leaf = rt;
2080                         res = fib6_del(rt, &info);
2081                         if (res) {
2082 #if RT6_DEBUG >= 2
2083                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2084                                          __func__, rt,
2085                                          rcu_access_pointer(rt->fib6_node),
2086                                          res);
2087 #endif
2088                                 continue;
2089                         }
2090                         return 0;
2091                 } else if (res == -2) {
2092                         if (WARN_ON(!rt->fib6_nsiblings))
2093                                 continue;
2094                         rt = list_last_entry(&rt->fib6_siblings,
2095                                              struct fib6_info, fib6_siblings);
2096                         continue;
2097                 }
2098                 WARN_ON(res != 0);
2099         }
2100         w->leaf = rt;
2101         return 0;
2102 }
2103
2104 /*
2105  *      Convenient frontend to tree walker.
2106  *
2107  *      func is called on each route.
2108  *              It may return -2 -> skip multipath route.
2109  *                            -1 -> delete this route.
2110  *                            0  -> continue walking
2111  */
2112
2113 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2114                             int (*func)(struct fib6_info *, void *arg),
2115                             int sernum, void *arg, bool skip_notify)
2116 {
2117         struct fib6_cleaner c;
2118
2119         c.w.root = root;
2120         c.w.func = fib6_clean_node;
2121         c.w.count = 0;
2122         c.w.skip = 0;
2123         c.w.skip_in_node = 0;
2124         c.func = func;
2125         c.sernum = sernum;
2126         c.arg = arg;
2127         c.net = net;
2128         c.skip_notify = skip_notify;
2129
2130         fib6_walk(net, &c.w);
2131 }
2132
2133 static void __fib6_clean_all(struct net *net,
2134                              int (*func)(struct fib6_info *, void *),
2135                              int sernum, void *arg, bool skip_notify)
2136 {
2137         struct fib6_table *table;
2138         struct hlist_head *head;
2139         unsigned int h;
2140
2141         rcu_read_lock();
2142         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2143                 head = &net->ipv6.fib_table_hash[h];
2144                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2145                         spin_lock_bh(&table->tb6_lock);
2146                         fib6_clean_tree(net, &table->tb6_root,
2147                                         func, sernum, arg, skip_notify);
2148                         spin_unlock_bh(&table->tb6_lock);
2149                 }
2150         }
2151         rcu_read_unlock();
2152 }
2153
2154 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2155                     void *arg)
2156 {
2157         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2158 }
2159
2160 void fib6_clean_all_skip_notify(struct net *net,
2161                                 int (*func)(struct fib6_info *, void *),
2162                                 void *arg)
2163 {
2164         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2165 }
2166
2167 static void fib6_flush_trees(struct net *net)
2168 {
2169         int new_sernum = fib6_new_sernum(net);
2170
2171         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2172 }
2173
2174 /*
2175  *      Garbage collection
2176  */
2177
2178 static int fib6_age(struct fib6_info *rt, void *arg)
2179 {
2180         struct fib6_gc_args *gc_args = arg;
2181         unsigned long now = jiffies;
2182
2183         /*
2184          *      check addrconf expiration here.
2185          *      Routes are expired even if they are in use.
2186          */
2187
2188         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2189                 if (time_after(now, rt->expires)) {
2190                         RT6_TRACE("expiring %p\n", rt);
2191                         return -1;
2192                 }
2193                 gc_args->more++;
2194         }
2195
2196         /*      Also age clones in the exception table.
2197          *      Note, that clones are aged out
2198          *      only if they are not in use now.
2199          */
2200         rt6_age_exceptions(rt, gc_args, now);
2201
2202         return 0;
2203 }
2204
2205 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2206 {
2207         struct fib6_gc_args gc_args;
2208         unsigned long now;
2209
2210         if (force) {
2211                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2212         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2213                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2214                 return;
2215         }
2216         gc_args.timeout = expires ? (int)expires :
2217                           net->ipv6.sysctl.ip6_rt_gc_interval;
2218         gc_args.more = 0;
2219
2220         fib6_clean_all(net, fib6_age, &gc_args);
2221         now = jiffies;
2222         net->ipv6.ip6_rt_last_gc = now;
2223
2224         if (gc_args.more)
2225                 mod_timer(&net->ipv6.ip6_fib_timer,
2226                           round_jiffies(now
2227                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2228         else
2229                 del_timer(&net->ipv6.ip6_fib_timer);
2230         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2231 }
2232
2233 static void fib6_gc_timer_cb(struct timer_list *t)
2234 {
2235         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2236
2237         fib6_run_gc(0, arg, true);
2238 }
2239
2240 static int __net_init fib6_net_init(struct net *net)
2241 {
2242         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2243         int err;
2244
2245         err = fib6_notifier_init(net);
2246         if (err)
2247                 return err;
2248
2249         spin_lock_init(&net->ipv6.fib6_gc_lock);
2250         rwlock_init(&net->ipv6.fib6_walker_lock);
2251         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2252         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2253
2254         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2255         if (!net->ipv6.rt6_stats)
2256                 goto out_timer;
2257
2258         /* Avoid false sharing : Use at least a full cache line */
2259         size = max_t(size_t, size, L1_CACHE_BYTES);
2260
2261         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2262         if (!net->ipv6.fib_table_hash)
2263                 goto out_rt6_stats;
2264
2265         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2266                                           GFP_KERNEL);
2267         if (!net->ipv6.fib6_main_tbl)
2268                 goto out_fib_table_hash;
2269
2270         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2271         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2272                            net->ipv6.fib6_null_entry);
2273         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2274                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2275         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2276
2277 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2278         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2279                                            GFP_KERNEL);
2280         if (!net->ipv6.fib6_local_tbl)
2281                 goto out_fib6_main_tbl;
2282         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2283         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2284                            net->ipv6.fib6_null_entry);
2285         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2286                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2287         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2288 #endif
2289         fib6_tables_init(net);
2290
2291         return 0;
2292
2293 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2294 out_fib6_main_tbl:
2295         kfree(net->ipv6.fib6_main_tbl);
2296 #endif
2297 out_fib_table_hash:
2298         kfree(net->ipv6.fib_table_hash);
2299 out_rt6_stats:
2300         kfree(net->ipv6.rt6_stats);
2301 out_timer:
2302         fib6_notifier_exit(net);
2303         return -ENOMEM;
2304 }
2305
2306 static void fib6_net_exit(struct net *net)
2307 {
2308         unsigned int i;
2309
2310         del_timer_sync(&net->ipv6.ip6_fib_timer);
2311
2312         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2313                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2314                 struct hlist_node *tmp;
2315                 struct fib6_table *tb;
2316
2317                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2318                         hlist_del(&tb->tb6_hlist);
2319                         fib6_free_table(tb);
2320                 }
2321         }
2322
2323         kfree(net->ipv6.fib_table_hash);
2324         kfree(net->ipv6.rt6_stats);
2325         fib6_notifier_exit(net);
2326 }
2327
2328 static struct pernet_operations fib6_net_ops = {
2329         .init = fib6_net_init,
2330         .exit = fib6_net_exit,
2331 };
2332
2333 int __init fib6_init(void)
2334 {
2335         int ret = -ENOMEM;
2336
2337         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2338                                            sizeof(struct fib6_node),
2339                                            0, SLAB_HWCACHE_ALIGN,
2340                                            NULL);
2341         if (!fib6_node_kmem)
2342                 goto out;
2343
2344         ret = register_pernet_subsys(&fib6_net_ops);
2345         if (ret)
2346                 goto out_kmem_cache_create;
2347
2348         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2349                                    inet6_dump_fib, 0);
2350         if (ret)
2351                 goto out_unregister_subsys;
2352
2353         __fib6_flush_trees = fib6_flush_trees;
2354 out:
2355         return ret;
2356
2357 out_unregister_subsys:
2358         unregister_pernet_subsys(&fib6_net_ops);
2359 out_kmem_cache_create:
2360         kmem_cache_destroy(fib6_node_kmem);
2361         goto out;
2362 }
2363
2364 void fib6_gc_cleanup(void)
2365 {
2366         unregister_pernet_subsys(&fib6_net_ops);
2367         kmem_cache_destroy(fib6_node_kmem);
2368 }
2369
2370 #ifdef CONFIG_PROC_FS
2371 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2372 {
2373         struct fib6_info *rt = v;
2374         struct ipv6_route_iter *iter = seq->private;
2375         struct fib6_nh *fib6_nh = rt->fib6_nh;
2376         unsigned int flags = rt->fib6_flags;
2377         const struct net_device *dev;
2378
2379         if (rt->nh)
2380                 fib6_nh = nexthop_fib6_nh(rt->nh);
2381
2382         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2383
2384 #ifdef CONFIG_IPV6_SUBTREES
2385         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2386 #else
2387         seq_puts(seq, "00000000000000000000000000000000 00 ");
2388 #endif
2389         if (fib6_nh->fib_nh_gw_family) {
2390                 flags |= RTF_GATEWAY;
2391                 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2392         } else {
2393                 seq_puts(seq, "00000000000000000000000000000000");
2394         }
2395
2396         dev = fib6_nh->fib_nh_dev;
2397         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2398                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2399                    flags, dev ? dev->name : "");
2400         iter->w.leaf = NULL;
2401         return 0;
2402 }
2403
2404 static int ipv6_route_yield(struct fib6_walker *w)
2405 {
2406         struct ipv6_route_iter *iter = w->args;
2407
2408         if (!iter->skip)
2409                 return 1;
2410
2411         do {
2412                 iter->w.leaf = rcu_dereference_protected(
2413                                 iter->w.leaf->fib6_next,
2414                                 lockdep_is_held(&iter->tbl->tb6_lock));
2415                 iter->skip--;
2416                 if (!iter->skip && iter->w.leaf)
2417                         return 1;
2418         } while (iter->w.leaf);
2419
2420         return 0;
2421 }
2422
2423 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2424                                       struct net *net)
2425 {
2426         memset(&iter->w, 0, sizeof(iter->w));
2427         iter->w.func = ipv6_route_yield;
2428         iter->w.root = &iter->tbl->tb6_root;
2429         iter->w.state = FWS_INIT;
2430         iter->w.node = iter->w.root;
2431         iter->w.args = iter;
2432         iter->sernum = iter->w.root->fn_sernum;
2433         INIT_LIST_HEAD(&iter->w.lh);
2434         fib6_walker_link(net, &iter->w);
2435 }
2436
2437 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2438                                                     struct net *net)
2439 {
2440         unsigned int h;
2441         struct hlist_node *node;
2442
2443         if (tbl) {
2444                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2445                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2446         } else {
2447                 h = 0;
2448                 node = NULL;
2449         }
2450
2451         while (!node && h < FIB6_TABLE_HASHSZ) {
2452                 node = rcu_dereference_bh(
2453                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2454         }
2455         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2456 }
2457
2458 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2459 {
2460         if (iter->sernum != iter->w.root->fn_sernum) {
2461                 iter->sernum = iter->w.root->fn_sernum;
2462                 iter->w.state = FWS_INIT;
2463                 iter->w.node = iter->w.root;
2464                 WARN_ON(iter->w.skip);
2465                 iter->w.skip = iter->w.count;
2466         }
2467 }
2468
2469 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2470 {
2471         int r;
2472         struct fib6_info *n;
2473         struct net *net = seq_file_net(seq);
2474         struct ipv6_route_iter *iter = seq->private;
2475
2476         if (!v)
2477                 goto iter_table;
2478
2479         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2480         if (n) {
2481                 ++*pos;
2482                 return n;
2483         }
2484
2485 iter_table:
2486         ipv6_route_check_sernum(iter);
2487         spin_lock_bh(&iter->tbl->tb6_lock);
2488         r = fib6_walk_continue(&iter->w);
2489         spin_unlock_bh(&iter->tbl->tb6_lock);
2490         if (r > 0) {
2491                 if (v)
2492                         ++*pos;
2493                 return iter->w.leaf;
2494         } else if (r < 0) {
2495                 fib6_walker_unlink(net, &iter->w);
2496                 return NULL;
2497         }
2498         fib6_walker_unlink(net, &iter->w);
2499
2500         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2501         if (!iter->tbl)
2502                 return NULL;
2503
2504         ipv6_route_seq_setup_walk(iter, net);
2505         goto iter_table;
2506 }
2507
2508 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2509         __acquires(RCU_BH)
2510 {
2511         struct net *net = seq_file_net(seq);
2512         struct ipv6_route_iter *iter = seq->private;
2513
2514         rcu_read_lock_bh();
2515         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2516         iter->skip = *pos;
2517
2518         if (iter->tbl) {
2519                 ipv6_route_seq_setup_walk(iter, net);
2520                 return ipv6_route_seq_next(seq, NULL, pos);
2521         } else {
2522                 return NULL;
2523         }
2524 }
2525
2526 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2527 {
2528         struct fib6_walker *w = &iter->w;
2529         return w->node && !(w->state == FWS_U && w->node == w->root);
2530 }
2531
2532 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2533         __releases(RCU_BH)
2534 {
2535         struct net *net = seq_file_net(seq);
2536         struct ipv6_route_iter *iter = seq->private;
2537
2538         if (ipv6_route_iter_active(iter))
2539                 fib6_walker_unlink(net, &iter->w);
2540
2541         rcu_read_unlock_bh();
2542 }
2543
2544 const struct seq_operations ipv6_route_seq_ops = {
2545         .start  = ipv6_route_seq_start,
2546         .next   = ipv6_route_seq_next,
2547         .stop   = ipv6_route_seq_stop,
2548         .show   = ipv6_route_seq_show
2549 };
2550 #endif /* CONFIG_PROC_FS */