net/ipv6/ip6_fib.c

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