root/net/ipv6/ip6_fib.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. fib6_walker_link
  2. fib6_walker_unlink
  3. fib6_new_sernum
  4. fib6_update_sernum
  5. addr_bit_set
  6. fib6_info_alloc
  7. fib6_info_destroy_rcu
  8. node_alloc
  9. node_free_immediate
  10. node_free_rcu
  11. node_free
  12. fib6_free_table
  13. fib6_link_table
  14. fib6_alloc_table
  15. fib6_new_table
  16. fib6_get_table
  17. fib6_tables_init
  18. fib6_new_table
  19. fib6_get_table
  20. fib6_rule_lookup
  21. fib6_lookup
  22. fib6_tables_init
  23. fib6_tables_seq_read
  24. call_fib6_entry_notifier
  25. call_fib6_entry_notifiers
  26. call_fib6_multipath_entry_notifiers
  27. fib6_rt_dump
  28. fib6_node_dump
  29. fib6_table_dump
  30. fib6_tables_dump
  31. fib6_dump_node
  32. fib6_dump_end
  33. fib6_dump_done
  34. fib6_dump_table
  35. inet6_dump_fib
  36. fib6_metric_set
  37. fib6_add_1
  38. __fib6_drop_pcpu_from
  39. fib6_nh_drop_pcpu_from
  40. fib6_drop_pcpu_from
  41. fib6_purge_rt
  42. fib6_add_rt2node
  43. fib6_start_gc
  44. fib6_force_start_gc
  45. __fib6_update_sernum_upto_root
  46. fib6_update_sernum_upto_root
  47. fib6_update_sernum_stub
  48. fib6_add
  49. fib6_node_lookup_1
  50. fib6_node_lookup
  51. fib6_locate_1
  52. fib6_locate
  53. fib6_find_prefix
  54. fib6_repair_tree
  55. fib6_del_route
  56. fib6_del
  57. fib6_walk_continue
  58. fib6_walk
  59. fib6_clean_node
  60. fib6_clean_tree
  61. __fib6_clean_all
  62. fib6_clean_all
  63. fib6_clean_all_skip_notify
  64. fib6_flush_trees
  65. fib6_age
  66. fib6_run_gc
  67. fib6_gc_timer_cb
  68. fib6_net_init
  69. fib6_net_exit
  70. fib6_init
  71. fib6_gc_cleanup
  72. ipv6_route_seq_show
  73. ipv6_route_yield
  74. ipv6_route_seq_setup_walk
  75. ipv6_route_seq_next_table
  76. ipv6_route_check_sernum
  77. ipv6_route_seq_next
  78. ipv6_route_seq_start
  79. ipv6_route_iter_active
  80. ipv6_route_seq_stop

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      Forwarding Information Database
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  *
   9  *      Changes:
  10  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  11  *                              remove ip6_null_entry from the top of
  12  *                              routing table.
  13  *      Ville Nuorvala:         Fixed routing subtrees.
  14  */
  15 
  16 #define pr_fmt(fmt) "IPv6: " fmt
  17 
  18 #include <linux/errno.h>
  19 #include <linux/types.h>
  20 #include <linux/net.h>
  21 #include <linux/route.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/in6.h>
  24 #include <linux/init.h>
  25 #include <linux/list.h>
  26 #include <linux/slab.h>
  27 
  28 #include <net/ip.h>
  29 #include <net/ipv6.h>
  30 #include <net/ndisc.h>
  31 #include <net/addrconf.h>
  32 #include <net/lwtunnel.h>
  33 #include <net/fib_notifier.h>
  34 
  35 #include <net/ip6_fib.h>
  36 #include <net/ip6_route.h>
  37 
  38 static struct kmem_cache *fib6_node_kmem __read_mostly;
  39 
  40 struct fib6_cleaner {
  41         struct fib6_walker w;
  42         struct net *net;
  43         int (*func)(struct fib6_info *, void *arg);
  44         int sernum;
  45         void *arg;
  46         bool skip_notify;
  47 };
  48 
  49 #ifdef CONFIG_IPV6_SUBTREES
  50 #define FWS_INIT FWS_S
  51 #else
  52 #define FWS_INIT FWS_L
  53 #endif
  54 
  55 static struct fib6_info *fib6_find_prefix(struct net *net,
  56                                          struct fib6_table *table,
  57                                          struct fib6_node *fn);
  58 static struct fib6_node *fib6_repair_tree(struct net *net,
  59                                           struct fib6_table *table,
  60                                           struct fib6_node *fn);
  61 static int fib6_walk(struct net *net, struct fib6_walker *w);
  62 static int fib6_walk_continue(struct fib6_walker *w);
  63 
  64 /*
  65  *      A routing update causes an increase of the serial number on the
  66  *      affected subtree. This allows for cached routes to be asynchronously
  67  *      tested when modifications are made to the destination cache as a
  68  *      result of redirects, path MTU changes, etc.
  69  */
  70 
  71 static void fib6_gc_timer_cb(struct timer_list *t);
  72 
  73 #define FOR_WALKERS(net, w) \
  74         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  75 
  76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  77 {
  78         write_lock_bh(&net->ipv6.fib6_walker_lock);
  79         list_add(&w->lh, &net->ipv6.fib6_walkers);
  80         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  81 }
  82 
  83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  84 {
  85         write_lock_bh(&net->ipv6.fib6_walker_lock);
  86         list_del(&w->lh);
  87         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  88 }
  89 
  90 static int fib6_new_sernum(struct net *net)
  91 {
  92         int new, old;
  93 
  94         do {
  95                 old = atomic_read(&net->ipv6.fib6_sernum);
  96                 new = old < INT_MAX ? old + 1 : 1;
  97         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
  98                                 old, new) != old);
  99         return new;
 100 }
 101 
 102 enum {
 103         FIB6_NO_SERNUM_CHANGE = 0,
 104 };
 105 
 106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 107 {
 108         struct fib6_node *fn;
 109 
 110         fn = rcu_dereference_protected(f6i->fib6_node,
 111                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 112         if (fn)
 113                 fn->fn_sernum = fib6_new_sernum(net);
 114 }
 115 
 116 /*
 117  *      Auxiliary address test functions for the radix tree.
 118  *
 119  *      These assume a 32bit processor (although it will work on
 120  *      64bit processors)
 121  */
 122 
 123 /*
 124  *      test bit
 125  */
 126 #if defined(__LITTLE_ENDIAN)
 127 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 128 #else
 129 # define BITOP_BE32_SWIZZLE     0
 130 #endif
 131 
 132 static __be32 addr_bit_set(const void *token, int fn_bit)
 133 {
 134         const __be32 *addr = token;
 135         /*
 136          * Here,
 137          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 138          * is optimized version of
 139          *      htonl(1 << ((~fn_bit)&0x1F))
 140          * See include/asm-generic/bitops/le.h.
 141          */
 142         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 143                addr[fn_bit >> 5];
 144 }
 145 
 146 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
 147 {
 148         struct fib6_info *f6i;
 149         size_t sz = sizeof(*f6i);
 150 
 151         if (with_fib6_nh)
 152                 sz += sizeof(struct fib6_nh);
 153 
 154         f6i = kzalloc(sz, gfp_flags);
 155         if (!f6i)
 156                 return NULL;
 157 
 158         /* fib6_siblings is a union with nh_list, so this initializes both */
 159         INIT_LIST_HEAD(&f6i->fib6_siblings);
 160         refcount_set(&f6i->fib6_ref, 1);
 161 
 162         return f6i;
 163 }
 164 
 165 void fib6_info_destroy_rcu(struct rcu_head *head)
 166 {
 167         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 168 
 169         WARN_ON(f6i->fib6_node);
 170 
 171         if (f6i->nh)
 172                 nexthop_put(f6i->nh);
 173         else
 174                 fib6_nh_release(f6i->fib6_nh);
 175 
 176         ip_fib_metrics_put(f6i->fib6_metrics);
 177         kfree(f6i);
 178 }
 179 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 180 
 181 static struct fib6_node *node_alloc(struct net *net)
 182 {
 183         struct fib6_node *fn;
 184 
 185         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 186         if (fn)
 187                 net->ipv6.rt6_stats->fib_nodes++;
 188 
 189         return fn;
 190 }
 191 
 192 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 193 {
 194         kmem_cache_free(fib6_node_kmem, fn);
 195         net->ipv6.rt6_stats->fib_nodes--;
 196 }
 197 
 198 static void node_free_rcu(struct rcu_head *head)
 199 {
 200         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 201 
 202         kmem_cache_free(fib6_node_kmem, fn);
 203 }
 204 
 205 static void node_free(struct net *net, struct fib6_node *fn)
 206 {
 207         call_rcu(&fn->rcu, node_free_rcu);
 208         net->ipv6.rt6_stats->fib_nodes--;
 209 }
 210 
 211 static void fib6_free_table(struct fib6_table *table)
 212 {
 213         inetpeer_invalidate_tree(&table->tb6_peers);
 214         kfree(table);
 215 }
 216 
 217 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 218 {
 219         unsigned int h;
 220 
 221         /*
 222          * Initialize table lock at a single place to give lockdep a key,
 223          * tables aren't visible prior to being linked to the list.
 224          */
 225         spin_lock_init(&tb->tb6_lock);
 226         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 227 
 228         /*
 229          * No protection necessary, this is the only list mutatation
 230          * operation, tables never disappear once they exist.
 231          */
 232         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 233 }
 234 
 235 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 236 
 237 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 238 {
 239         struct fib6_table *table;
 240 
 241         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 242         if (table) {
 243                 table->tb6_id = id;
 244                 rcu_assign_pointer(table->tb6_root.leaf,
 245                                    net->ipv6.fib6_null_entry);
 246                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 247                 inet_peer_base_init(&table->tb6_peers);
 248         }
 249 
 250         return table;
 251 }
 252 
 253 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 254 {
 255         struct fib6_table *tb;
 256 
 257         if (id == 0)
 258                 id = RT6_TABLE_MAIN;
 259         tb = fib6_get_table(net, id);
 260         if (tb)
 261                 return tb;
 262 
 263         tb = fib6_alloc_table(net, id);
 264         if (tb)
 265                 fib6_link_table(net, tb);
 266 
 267         return tb;
 268 }
 269 EXPORT_SYMBOL_GPL(fib6_new_table);
 270 
 271 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 272 {
 273         struct fib6_table *tb;
 274         struct hlist_head *head;
 275         unsigned int h;
 276 
 277         if (id == 0)
 278                 id = RT6_TABLE_MAIN;
 279         h = id & (FIB6_TABLE_HASHSZ - 1);
 280         rcu_read_lock();
 281         head = &net->ipv6.fib_table_hash[h];
 282         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 283                 if (tb->tb6_id == id) {
 284                         rcu_read_unlock();
 285                         return tb;
 286                 }
 287         }
 288         rcu_read_unlock();
 289 
 290         return NULL;
 291 }
 292 EXPORT_SYMBOL_GPL(fib6_get_table);
 293 
 294 static void __net_init fib6_tables_init(struct net *net)
 295 {
 296         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 297         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 298 }
 299 #else
 300 
 301 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 302 {
 303         return fib6_get_table(net, id);
 304 }
 305 
 306 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 307 {
 308           return net->ipv6.fib6_main_tbl;
 309 }
 310 
 311 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 312                                    const struct sk_buff *skb,
 313                                    int flags, pol_lookup_t lookup)
 314 {
 315         struct rt6_info *rt;
 316 
 317         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 318         if (rt->dst.error == -EAGAIN) {
 319                 ip6_rt_put_flags(rt, flags);
 320                 rt = net->ipv6.ip6_null_entry;
 321                 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
 322                         dst_hold(&rt->dst);
 323         }
 324 
 325         return &rt->dst;
 326 }
 327 
 328 /* called with rcu lock held; no reference taken on fib6_info */
 329 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 330                 struct fib6_result *res, int flags)
 331 {
 332         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
 333                                  res, flags);
 334 }
 335 
 336 static void __net_init fib6_tables_init(struct net *net)
 337 {
 338         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 339 }
 340 
 341 #endif
 342 
 343 unsigned int fib6_tables_seq_read(struct net *net)
 344 {
 345         unsigned int h, fib_seq = 0;
 346 
 347         rcu_read_lock();
 348         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 349                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 350                 struct fib6_table *tb;
 351 
 352                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 353                         fib_seq += tb->fib_seq;
 354         }
 355         rcu_read_unlock();
 356 
 357         return fib_seq;
 358 }
 359 
 360 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 361                                     enum fib_event_type event_type,
 362                                     struct fib6_info *rt)
 363 {
 364         struct fib6_entry_notifier_info info = {
 365                 .rt = rt,
 366         };
 367 
 368         return call_fib6_notifier(nb, net, event_type, &info.info);
 369 }
 370 
 371 int call_fib6_entry_notifiers(struct net *net,
 372                               enum fib_event_type event_type,
 373                               struct fib6_info *rt,
 374                               struct netlink_ext_ack *extack)
 375 {
 376         struct fib6_entry_notifier_info info = {
 377                 .info.extack = extack,
 378                 .rt = rt,
 379         };
 380 
 381         rt->fib6_table->fib_seq++;
 382         return call_fib6_notifiers(net, event_type, &info.info);
 383 }
 384 
 385 int call_fib6_multipath_entry_notifiers(struct net *net,
 386                                         enum fib_event_type event_type,
 387                                         struct fib6_info *rt,
 388                                         unsigned int nsiblings,
 389                                         struct netlink_ext_ack *extack)
 390 {
 391         struct fib6_entry_notifier_info info = {
 392                 .info.extack = extack,
 393                 .rt = rt,
 394                 .nsiblings = nsiblings,
 395         };
 396 
 397         rt->fib6_table->fib_seq++;
 398         return call_fib6_notifiers(net, event_type, &info.info);
 399 }
 400 
 401 struct fib6_dump_arg {
 402         struct net *net;
 403         struct notifier_block *nb;
 404 };
 405 
 406 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 407 {
 408         if (rt == arg->net->ipv6.fib6_null_entry)
 409                 return;
 410         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 411 }
 412 
 413 static int fib6_node_dump(struct fib6_walker *w)
 414 {
 415         struct fib6_info *rt;
 416 
 417         for_each_fib6_walker_rt(w)
 418                 fib6_rt_dump(rt, w->args);
 419         w->leaf = NULL;
 420         return 0;
 421 }
 422 
 423 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 424                             struct fib6_walker *w)
 425 {
 426         w->root = &tb->tb6_root;
 427         spin_lock_bh(&tb->tb6_lock);
 428         fib6_walk(net, w);
 429         spin_unlock_bh(&tb->tb6_lock);
 430 }
 431 
 432 /* Called with rcu_read_lock() */
 433 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 434 {
 435         struct fib6_dump_arg arg;
 436         struct fib6_walker *w;
 437         unsigned int h;
 438 
 439         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 440         if (!w)
 441                 return -ENOMEM;
 442 
 443         w->func = fib6_node_dump;
 444         arg.net = net;
 445         arg.nb = nb;
 446         w->args = &arg;
 447 
 448         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 449                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 450                 struct fib6_table *tb;
 451 
 452                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 453                         fib6_table_dump(net, tb, w);
 454         }
 455 
 456         kfree(w);
 457 
 458         return 0;
 459 }
 460 
 461 static int fib6_dump_node(struct fib6_walker *w)
 462 {
 463         int res;
 464         struct fib6_info *rt;
 465 
 466         for_each_fib6_walker_rt(w) {
 467                 res = rt6_dump_route(rt, w->args, w->skip_in_node);
 468                 if (res >= 0) {
 469                         /* Frame is full, suspend walking */
 470                         w->leaf = rt;
 471 
 472                         /* We'll restart from this node, so if some routes were
 473                          * already dumped, skip them next time.
 474                          */
 475                         w->skip_in_node += res;
 476 
 477                         return 1;
 478                 }
 479                 w->skip_in_node = 0;
 480 
 481                 /* Multipath routes are dumped in one route with the
 482                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 483                  * last sibling of this route (no need to dump the
 484                  * sibling routes again)
 485                  */
 486                 if (rt->fib6_nsiblings)
 487                         rt = list_last_entry(&rt->fib6_siblings,
 488                                              struct fib6_info,
 489                                              fib6_siblings);
 490         }
 491         w->leaf = NULL;
 492         return 0;
 493 }
 494 
 495 static void fib6_dump_end(struct netlink_callback *cb)
 496 {
 497         struct net *net = sock_net(cb->skb->sk);
 498         struct fib6_walker *w = (void *)cb->args[2];
 499 
 500         if (w) {
 501                 if (cb->args[4]) {
 502                         cb->args[4] = 0;
 503                         fib6_walker_unlink(net, w);
 504                 }
 505                 cb->args[2] = 0;
 506                 kfree(w);
 507         }
 508         cb->done = (void *)cb->args[3];
 509         cb->args[1] = 3;
 510 }
 511 
 512 static int fib6_dump_done(struct netlink_callback *cb)
 513 {
 514         fib6_dump_end(cb);
 515         return cb->done ? cb->done(cb) : 0;
 516 }
 517 
 518 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 519                            struct netlink_callback *cb)
 520 {
 521         struct net *net = sock_net(skb->sk);
 522         struct fib6_walker *w;
 523         int res;
 524 
 525         w = (void *)cb->args[2];
 526         w->root = &table->tb6_root;
 527 
 528         if (cb->args[4] == 0) {
 529                 w->count = 0;
 530                 w->skip = 0;
 531                 w->skip_in_node = 0;
 532 
 533                 spin_lock_bh(&table->tb6_lock);
 534                 res = fib6_walk(net, w);
 535                 spin_unlock_bh(&table->tb6_lock);
 536                 if (res > 0) {
 537                         cb->args[4] = 1;
 538                         cb->args[5] = w->root->fn_sernum;
 539                 }
 540         } else {
 541                 if (cb->args[5] != w->root->fn_sernum) {
 542                         /* Begin at the root if the tree changed */
 543                         cb->args[5] = w->root->fn_sernum;
 544                         w->state = FWS_INIT;
 545                         w->node = w->root;
 546                         w->skip = w->count;
 547                         w->skip_in_node = 0;
 548                 } else
 549                         w->skip = 0;
 550 
 551                 spin_lock_bh(&table->tb6_lock);
 552                 res = fib6_walk_continue(w);
 553                 spin_unlock_bh(&table->tb6_lock);
 554                 if (res <= 0) {
 555                         fib6_walker_unlink(net, w);
 556                         cb->args[4] = 0;
 557                 }
 558         }
 559 
 560         return res;
 561 }
 562 
 563 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 564 {
 565         struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
 566                                          .filter.dump_routes = true };
 567         const struct nlmsghdr *nlh = cb->nlh;
 568         struct net *net = sock_net(skb->sk);
 569         unsigned int h, s_h;
 570         unsigned int e = 0, s_e;
 571         struct fib6_walker *w;
 572         struct fib6_table *tb;
 573         struct hlist_head *head;
 574         int res = 0;
 575 
 576         if (cb->strict_check) {
 577                 int err;
 578 
 579                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 580                 if (err < 0)
 581                         return err;
 582         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 583                 struct rtmsg *rtm = nlmsg_data(nlh);
 584 
 585                 if (rtm->rtm_flags & RTM_F_PREFIX)
 586                         arg.filter.flags = RTM_F_PREFIX;
 587         }
 588 
 589         w = (void *)cb->args[2];
 590         if (!w) {
 591                 /* New dump:
 592                  *
 593                  * 1. hook callback destructor.
 594                  */
 595                 cb->args[3] = (long)cb->done;
 596                 cb->done = fib6_dump_done;
 597 
 598                 /*
 599                  * 2. allocate and initialize walker.
 600                  */
 601                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 602                 if (!w)
 603                         return -ENOMEM;
 604                 w->func = fib6_dump_node;
 605                 cb->args[2] = (long)w;
 606         }
 607 
 608         arg.skb = skb;
 609         arg.cb = cb;
 610         arg.net = net;
 611         w->args = &arg;
 612 
 613         if (arg.filter.table_id) {
 614                 tb = fib6_get_table(net, arg.filter.table_id);
 615                 if (!tb) {
 616                         if (rtnl_msg_family(cb->nlh) != PF_INET6)
 617                                 goto out;
 618 
 619                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 620                         return -ENOENT;
 621                 }
 622 
 623                 if (!cb->args[0]) {
 624                         res = fib6_dump_table(tb, skb, cb);
 625                         if (!res)
 626                                 cb->args[0] = 1;
 627                 }
 628                 goto out;
 629         }
 630 
 631         s_h = cb->args[0];
 632         s_e = cb->args[1];
 633 
 634         rcu_read_lock();
 635         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 636                 e = 0;
 637                 head = &net->ipv6.fib_table_hash[h];
 638                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 639                         if (e < s_e)
 640                                 goto next;
 641                         res = fib6_dump_table(tb, skb, cb);
 642                         if (res != 0)
 643                                 goto out_unlock;
 644 next:
 645                         e++;
 646                 }
 647         }
 648 out_unlock:
 649         rcu_read_unlock();
 650         cb->args[1] = e;
 651         cb->args[0] = h;
 652 out:
 653         res = res < 0 ? res : skb->len;
 654         if (res <= 0)
 655                 fib6_dump_end(cb);
 656         return res;
 657 }
 658 
 659 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 660 {
 661         if (!f6i)
 662                 return;
 663 
 664         if (f6i->fib6_metrics == &dst_default_metrics) {
 665                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 666 
 667                 if (!p)
 668                         return;
 669 
 670                 refcount_set(&p->refcnt, 1);
 671                 f6i->fib6_metrics = p;
 672         }
 673 
 674         f6i->fib6_metrics->metrics[metric - 1] = val;
 675 }
 676 
 677 /*
 678  *      Routing Table
 679  *
 680  *      return the appropriate node for a routing tree "add" operation
 681  *      by either creating and inserting or by returning an existing
 682  *      node.
 683  */
 684 
 685 static struct fib6_node *fib6_add_1(struct net *net,
 686                                     struct fib6_table *table,
 687                                     struct fib6_node *root,
 688                                     struct in6_addr *addr, int plen,
 689                                     int offset, int allow_create,
 690                                     int replace_required,
 691                                     struct netlink_ext_ack *extack)
 692 {
 693         struct fib6_node *fn, *in, *ln;
 694         struct fib6_node *pn = NULL;
 695         struct rt6key *key;
 696         int     bit;
 697         __be32  dir = 0;
 698 
 699         RT6_TRACE("fib6_add_1\n");
 700 
 701         /* insert node in tree */
 702 
 703         fn = root;
 704 
 705         do {
 706                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 707                                             lockdep_is_held(&table->tb6_lock));
 708                 key = (struct rt6key *)((u8 *)leaf + offset);
 709 
 710                 /*
 711                  *      Prefix match
 712                  */
 713                 if (plen < fn->fn_bit ||
 714                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 715                         if (!allow_create) {
 716                                 if (replace_required) {
 717                                         NL_SET_ERR_MSG(extack,
 718                                                        "Can not replace route - no match found");
 719                                         pr_warn("Can't replace route, no match found\n");
 720                                         return ERR_PTR(-ENOENT);
 721                                 }
 722                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 723                         }
 724                         goto insert_above;
 725                 }
 726 
 727                 /*
 728                  *      Exact match ?
 729                  */
 730 
 731                 if (plen == fn->fn_bit) {
 732                         /* clean up an intermediate node */
 733                         if (!(fn->fn_flags & RTN_RTINFO)) {
 734                                 RCU_INIT_POINTER(fn->leaf, NULL);
 735                                 fib6_info_release(leaf);
 736                         /* remove null_entry in the root node */
 737                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 738                                    rcu_access_pointer(fn->leaf) ==
 739                                    net->ipv6.fib6_null_entry) {
 740                                 RCU_INIT_POINTER(fn->leaf, NULL);
 741                         }
 742 
 743                         return fn;
 744                 }
 745 
 746                 /*
 747                  *      We have more bits to go
 748                  */
 749 
 750                 /* Try to walk down on tree. */
 751                 dir = addr_bit_set(addr, fn->fn_bit);
 752                 pn = fn;
 753                 fn = dir ?
 754                      rcu_dereference_protected(fn->right,
 755                                         lockdep_is_held(&table->tb6_lock)) :
 756                      rcu_dereference_protected(fn->left,
 757                                         lockdep_is_held(&table->tb6_lock));
 758         } while (fn);
 759 
 760         if (!allow_create) {
 761                 /* We should not create new node because
 762                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 763                  * I assume it is safe to require NLM_F_CREATE when
 764                  * REPLACE flag is used! Later we may want to remove the
 765                  * check for replace_required, because according
 766                  * to netlink specification, NLM_F_CREATE
 767                  * MUST be specified if new route is created.
 768                  * That would keep IPv6 consistent with IPv4
 769                  */
 770                 if (replace_required) {
 771                         NL_SET_ERR_MSG(extack,
 772                                        "Can not replace route - no match found");
 773                         pr_warn("Can't replace route, no match found\n");
 774                         return ERR_PTR(-ENOENT);
 775                 }
 776                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 777         }
 778         /*
 779          *      We walked to the bottom of tree.
 780          *      Create new leaf node without children.
 781          */
 782 
 783         ln = node_alloc(net);
 784 
 785         if (!ln)
 786                 return ERR_PTR(-ENOMEM);
 787         ln->fn_bit = plen;
 788         RCU_INIT_POINTER(ln->parent, pn);
 789 
 790         if (dir)
 791                 rcu_assign_pointer(pn->right, ln);
 792         else
 793                 rcu_assign_pointer(pn->left, ln);
 794 
 795         return ln;
 796 
 797 
 798 insert_above:
 799         /*
 800          * split since we don't have a common prefix anymore or
 801          * we have a less significant route.
 802          * we've to insert an intermediate node on the list
 803          * this new node will point to the one we need to create
 804          * and the current
 805          */
 806 
 807         pn = rcu_dereference_protected(fn->parent,
 808                                        lockdep_is_held(&table->tb6_lock));
 809 
 810         /* find 1st bit in difference between the 2 addrs.
 811 
 812            See comment in __ipv6_addr_diff: bit may be an invalid value,
 813            but if it is >= plen, the value is ignored in any case.
 814          */
 815 
 816         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 817 
 818         /*
 819          *              (intermediate)[in]
 820          *                /        \
 821          *      (new leaf node)[ln] (old node)[fn]
 822          */
 823         if (plen > bit) {
 824                 in = node_alloc(net);
 825                 ln = node_alloc(net);
 826 
 827                 if (!in || !ln) {
 828                         if (in)
 829                                 node_free_immediate(net, in);
 830                         if (ln)
 831                                 node_free_immediate(net, ln);
 832                         return ERR_PTR(-ENOMEM);
 833                 }
 834 
 835                 /*
 836                  * new intermediate node.
 837                  * RTN_RTINFO will
 838                  * be off since that an address that chooses one of
 839                  * the branches would not match less specific routes
 840                  * in the other branch
 841                  */
 842 
 843                 in->fn_bit = bit;
 844 
 845                 RCU_INIT_POINTER(in->parent, pn);
 846                 in->leaf = fn->leaf;
 847                 fib6_info_hold(rcu_dereference_protected(in->leaf,
 848                                 lockdep_is_held(&table->tb6_lock)));
 849 
 850                 /* update parent pointer */
 851                 if (dir)
 852                         rcu_assign_pointer(pn->right, in);
 853                 else
 854                         rcu_assign_pointer(pn->left, in);
 855 
 856                 ln->fn_bit = plen;
 857 
 858                 RCU_INIT_POINTER(ln->parent, in);
 859                 rcu_assign_pointer(fn->parent, in);
 860 
 861                 if (addr_bit_set(addr, bit)) {
 862                         rcu_assign_pointer(in->right, ln);
 863                         rcu_assign_pointer(in->left, fn);
 864                 } else {
 865                         rcu_assign_pointer(in->left, ln);
 866                         rcu_assign_pointer(in->right, fn);
 867                 }
 868         } else { /* plen <= bit */
 869 
 870                 /*
 871                  *              (new leaf node)[ln]
 872                  *                /        \
 873                  *           (old node)[fn] NULL
 874                  */
 875 
 876                 ln = node_alloc(net);
 877 
 878                 if (!ln)
 879                         return ERR_PTR(-ENOMEM);
 880 
 881                 ln->fn_bit = plen;
 882 
 883                 RCU_INIT_POINTER(ln->parent, pn);
 884 
 885                 if (addr_bit_set(&key->addr, plen))
 886                         RCU_INIT_POINTER(ln->right, fn);
 887                 else
 888                         RCU_INIT_POINTER(ln->left, fn);
 889 
 890                 rcu_assign_pointer(fn->parent, ln);
 891 
 892                 if (dir)
 893                         rcu_assign_pointer(pn->right, ln);
 894                 else
 895                         rcu_assign_pointer(pn->left, ln);
 896         }
 897         return ln;
 898 }
 899 
 900 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
 901                                   const struct fib6_info *match,
 902                                   const struct fib6_table *table)
 903 {
 904         int cpu;
 905 
 906         if (!fib6_nh->rt6i_pcpu)
 907                 return;
 908 
 909         /* release the reference to this fib entry from
 910          * all of its cached pcpu routes
 911          */
 912         for_each_possible_cpu(cpu) {
 913                 struct rt6_info **ppcpu_rt;
 914                 struct rt6_info *pcpu_rt;
 915 
 916                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
 917                 pcpu_rt = *ppcpu_rt;
 918 
 919                 /* only dropping the 'from' reference if the cached route
 920                  * is using 'match'. The cached pcpu_rt->from only changes
 921                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
 922                  * change from one fib6_info reference to another
 923                  */
 924                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
 925                         struct fib6_info *from;
 926 
 927                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 928                         fib6_info_release(from);
 929                 }
 930         }
 931 }
 932 
 933 struct fib6_nh_pcpu_arg {
 934         struct fib6_info        *from;
 935         const struct fib6_table *table;
 936 };
 937 
 938 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
 939 {
 940         struct fib6_nh_pcpu_arg *arg = _arg;
 941 
 942         __fib6_drop_pcpu_from(nh, arg->from, arg->table);
 943         return 0;
 944 }
 945 
 946 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
 947                                 const struct fib6_table *table)
 948 {
 949         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
 950          * while we are cleaning them here.
 951          */
 952         f6i->fib6_destroying = 1;
 953         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
 954 
 955         if (f6i->nh) {
 956                 struct fib6_nh_pcpu_arg arg = {
 957                         .from = f6i,
 958                         .table = table
 959                 };
 960 
 961                 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
 962                                          &arg);
 963         } else {
 964                 struct fib6_nh *fib6_nh;
 965 
 966                 fib6_nh = f6i->fib6_nh;
 967                 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
 968         }
 969 }
 970 
 971 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
 972                           struct net *net)
 973 {
 974         struct fib6_table *table = rt->fib6_table;
 975 
 976         fib6_drop_pcpu_from(rt, table);
 977 
 978         if (rt->nh && !list_empty(&rt->nh_list))
 979                 list_del_init(&rt->nh_list);
 980 
 981         if (refcount_read(&rt->fib6_ref) != 1) {
 982                 /* This route is used as dummy address holder in some split
 983                  * nodes. It is not leaked, but it still holds other resources,
 984                  * which must be released in time. So, scan ascendant nodes
 985                  * and replace dummy references to this route with references
 986                  * to still alive ones.
 987                  */
 988                 while (fn) {
 989                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 990                                             lockdep_is_held(&table->tb6_lock));
 991                         struct fib6_info *new_leaf;
 992                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 993                                 new_leaf = fib6_find_prefix(net, table, fn);
 994                                 fib6_info_hold(new_leaf);
 995 
 996                                 rcu_assign_pointer(fn->leaf, new_leaf);
 997                                 fib6_info_release(rt);
 998                         }
 999                         fn = rcu_dereference_protected(fn->parent,
1000                                     lockdep_is_held(&table->tb6_lock));
1001                 }
1002         }
1003 }
1004 
1005 /*
1006  *      Insert routing information in a node.
1007  */
1008 
1009 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1010                             struct nl_info *info,
1011                             struct netlink_ext_ack *extack)
1012 {
1013         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1014                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1015         struct fib6_info *iter = NULL;
1016         struct fib6_info __rcu **ins;
1017         struct fib6_info __rcu **fallback_ins = NULL;
1018         int replace = (info->nlh &&
1019                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1020         int add = (!info->nlh ||
1021                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
1022         int found = 0;
1023         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1024         u16 nlflags = NLM_F_EXCL;
1025         int err;
1026 
1027         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1028                 nlflags |= NLM_F_APPEND;
1029 
1030         ins = &fn->leaf;
1031 
1032         for (iter = leaf; iter;
1033              iter = rcu_dereference_protected(iter->fib6_next,
1034                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1035                 /*
1036                  *      Search for duplicates
1037                  */
1038 
1039                 if (iter->fib6_metric == rt->fib6_metric) {
1040                         /*
1041                          *      Same priority level
1042                          */
1043                         if (info->nlh &&
1044                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1045                                 return -EEXIST;
1046 
1047                         nlflags &= ~NLM_F_EXCL;
1048                         if (replace) {
1049                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1050                                         found++;
1051                                         break;
1052                                 }
1053                                 fallback_ins = fallback_ins ?: ins;
1054                                 goto next_iter;
1055                         }
1056 
1057                         if (rt6_duplicate_nexthop(iter, rt)) {
1058                                 if (rt->fib6_nsiblings)
1059                                         rt->fib6_nsiblings = 0;
1060                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1061                                         return -EEXIST;
1062                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1063                                         fib6_clean_expires(iter);
1064                                 else
1065                                         fib6_set_expires(iter, rt->expires);
1066 
1067                                 if (rt->fib6_pmtu)
1068                                         fib6_metric_set(iter, RTAX_MTU,
1069                                                         rt->fib6_pmtu);
1070                                 return -EEXIST;
1071                         }
1072                         /* If we have the same destination and the same metric,
1073                          * but not the same gateway, then the route we try to
1074                          * add is sibling to this route, increment our counter
1075                          * of siblings, and later we will add our route to the
1076                          * list.
1077                          * Only static routes (which don't have flag
1078                          * RTF_EXPIRES) are used for ECMPv6.
1079                          *
1080                          * To avoid long list, we only had siblings if the
1081                          * route have a gateway.
1082                          */
1083                         if (rt_can_ecmp &&
1084                             rt6_qualify_for_ecmp(iter))
1085                                 rt->fib6_nsiblings++;
1086                 }
1087 
1088                 if (iter->fib6_metric > rt->fib6_metric)
1089                         break;
1090 
1091 next_iter:
1092                 ins = &iter->fib6_next;
1093         }
1094 
1095         if (fallback_ins && !found) {
1096                 /* No matching route with same ecmp-able-ness found, replace
1097                  * first matching route
1098                  */
1099                 ins = fallback_ins;
1100                 iter = rcu_dereference_protected(*ins,
1101                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1102                 found++;
1103         }
1104 
1105         /* Reset round-robin state, if necessary */
1106         if (ins == &fn->leaf)
1107                 fn->rr_ptr = NULL;
1108 
1109         /* Link this route to others same route. */
1110         if (rt->fib6_nsiblings) {
1111                 unsigned int fib6_nsiblings;
1112                 struct fib6_info *sibling, *temp_sibling;
1113 
1114                 /* Find the first route that have the same metric */
1115                 sibling = leaf;
1116                 while (sibling) {
1117                         if (sibling->fib6_metric == rt->fib6_metric &&
1118                             rt6_qualify_for_ecmp(sibling)) {
1119                                 list_add_tail(&rt->fib6_siblings,
1120                                               &sibling->fib6_siblings);
1121                                 break;
1122                         }
1123                         sibling = rcu_dereference_protected(sibling->fib6_next,
1124                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1125                 }
1126                 /* For each sibling in the list, increment the counter of
1127                  * siblings. BUG() if counters does not match, list of siblings
1128                  * is broken!
1129                  */
1130                 fib6_nsiblings = 0;
1131                 list_for_each_entry_safe(sibling, temp_sibling,
1132                                          &rt->fib6_siblings, fib6_siblings) {
1133                         sibling->fib6_nsiblings++;
1134                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1135                         fib6_nsiblings++;
1136                 }
1137                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1138                 rt6_multipath_rebalance(temp_sibling);
1139         }
1140 
1141         /*
1142          *      insert node
1143          */
1144         if (!replace) {
1145                 if (!add)
1146                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1147 
1148 add:
1149                 nlflags |= NLM_F_CREATE;
1150 
1151                 if (!info->skip_notify_kernel) {
1152                         err = call_fib6_entry_notifiers(info->nl_net,
1153                                                         FIB_EVENT_ENTRY_ADD,
1154                                                         rt, extack);
1155                         if (err) {
1156                                 struct fib6_info *sibling, *next_sibling;
1157 
1158                                 /* If the route has siblings, then it first
1159                                  * needs to be unlinked from them.
1160                                  */
1161                                 if (!rt->fib6_nsiblings)
1162                                         return err;
1163 
1164                                 list_for_each_entry_safe(sibling, next_sibling,
1165                                                          &rt->fib6_siblings,
1166                                                          fib6_siblings)
1167                                         sibling->fib6_nsiblings--;
1168                                 rt->fib6_nsiblings = 0;
1169                                 list_del_init(&rt->fib6_siblings);
1170                                 rt6_multipath_rebalance(next_sibling);
1171                                 return err;
1172                         }
1173                 }
1174 
1175                 rcu_assign_pointer(rt->fib6_next, iter);
1176                 fib6_info_hold(rt);
1177                 rcu_assign_pointer(rt->fib6_node, fn);
1178                 rcu_assign_pointer(*ins, rt);
1179                 if (!info->skip_notify)
1180                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1181                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1182 
1183                 if (!(fn->fn_flags & RTN_RTINFO)) {
1184                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1185                         fn->fn_flags |= RTN_RTINFO;
1186                 }
1187 
1188         } else {
1189                 int nsiblings;
1190 
1191                 if (!found) {
1192                         if (add)
1193                                 goto add;
1194                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1195                         return -ENOENT;
1196                 }
1197 
1198                 if (!info->skip_notify_kernel) {
1199                         err = call_fib6_entry_notifiers(info->nl_net,
1200                                                         FIB_EVENT_ENTRY_REPLACE,
1201                                                         rt, extack);
1202                         if (err)
1203                                 return err;
1204                 }
1205 
1206                 fib6_info_hold(rt);
1207                 rcu_assign_pointer(rt->fib6_node, fn);
1208                 rt->fib6_next = iter->fib6_next;
1209                 rcu_assign_pointer(*ins, rt);
1210                 if (!info->skip_notify)
1211                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1212                 if (!(fn->fn_flags & RTN_RTINFO)) {
1213                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1214                         fn->fn_flags |= RTN_RTINFO;
1215                 }
1216                 nsiblings = iter->fib6_nsiblings;
1217                 iter->fib6_node = NULL;
1218                 fib6_purge_rt(iter, fn, info->nl_net);
1219                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1220                         fn->rr_ptr = NULL;
1221                 fib6_info_release(iter);
1222 
1223                 if (nsiblings) {
1224                         /* Replacing an ECMP route, remove all siblings */
1225                         ins = &rt->fib6_next;
1226                         iter = rcu_dereference_protected(*ins,
1227                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1228                         while (iter) {
1229                                 if (iter->fib6_metric > rt->fib6_metric)
1230                                         break;
1231                                 if (rt6_qualify_for_ecmp(iter)) {
1232                                         *ins = iter->fib6_next;
1233                                         iter->fib6_node = NULL;
1234                                         fib6_purge_rt(iter, fn, info->nl_net);
1235                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1236                                                 fn->rr_ptr = NULL;
1237                                         fib6_info_release(iter);
1238                                         nsiblings--;
1239                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1240                                 } else {
1241                                         ins = &iter->fib6_next;
1242                                 }
1243                                 iter = rcu_dereference_protected(*ins,
1244                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1245                         }
1246                         WARN_ON(nsiblings != 0);
1247                 }
1248         }
1249 
1250         return 0;
1251 }
1252 
1253 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1254 {
1255         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1256             (rt->fib6_flags & RTF_EXPIRES))
1257                 mod_timer(&net->ipv6.ip6_fib_timer,
1258                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1259 }
1260 
1261 void fib6_force_start_gc(struct net *net)
1262 {
1263         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1264                 mod_timer(&net->ipv6.ip6_fib_timer,
1265                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1266 }
1267 
1268 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1269                                            int sernum)
1270 {
1271         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1272                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1273 
1274         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1275         smp_wmb();
1276         while (fn) {
1277                 fn->fn_sernum = sernum;
1278                 fn = rcu_dereference_protected(fn->parent,
1279                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1280         }
1281 }
1282 
1283 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1284 {
1285         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1286 }
1287 
1288 /* allow ipv4 to update sernum via ipv6_stub */
1289 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1290 {
1291         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1292         fib6_update_sernum_upto_root(net, f6i);
1293         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1294 }
1295 
1296 /*
1297  *      Add routing information to the routing tree.
1298  *      <destination addr>/<source addr>
1299  *      with source addr info in sub-trees
1300  *      Need to own table->tb6_lock
1301  */
1302 
1303 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1304              struct nl_info *info, struct netlink_ext_ack *extack)
1305 {
1306         struct fib6_table *table = rt->fib6_table;
1307         struct fib6_node *fn, *pn = NULL;
1308         int err = -ENOMEM;
1309         int allow_create = 1;
1310         int replace_required = 0;
1311         int sernum = fib6_new_sernum(info->nl_net);
1312 
1313         if (info->nlh) {
1314                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1315                         allow_create = 0;
1316                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1317                         replace_required = 1;
1318         }
1319         if (!allow_create && !replace_required)
1320                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1321 
1322         fn = fib6_add_1(info->nl_net, table, root,
1323                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1324                         offsetof(struct fib6_info, fib6_dst), allow_create,
1325                         replace_required, extack);
1326         if (IS_ERR(fn)) {
1327                 err = PTR_ERR(fn);
1328                 fn = NULL;
1329                 goto out;
1330         }
1331 
1332         pn = fn;
1333 
1334 #ifdef CONFIG_IPV6_SUBTREES
1335         if (rt->fib6_src.plen) {
1336                 struct fib6_node *sn;
1337 
1338                 if (!rcu_access_pointer(fn->subtree)) {
1339                         struct fib6_node *sfn;
1340 
1341                         /*
1342                          * Create subtree.
1343                          *
1344                          *              fn[main tree]
1345                          *              |
1346                          *              sfn[subtree root]
1347                          *                 \
1348                          *                  sn[new leaf node]
1349                          */
1350 
1351                         /* Create subtree root node */
1352                         sfn = node_alloc(info->nl_net);
1353                         if (!sfn)
1354                                 goto failure;
1355 
1356                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1357                         rcu_assign_pointer(sfn->leaf,
1358                                            info->nl_net->ipv6.fib6_null_entry);
1359                         sfn->fn_flags = RTN_ROOT;
1360 
1361                         /* Now add the first leaf node to new subtree */
1362 
1363                         sn = fib6_add_1(info->nl_net, table, sfn,
1364                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1365                                         offsetof(struct fib6_info, fib6_src),
1366                                         allow_create, replace_required, extack);
1367 
1368                         if (IS_ERR(sn)) {
1369                                 /* If it is failed, discard just allocated
1370                                    root, and then (in failure) stale node
1371                                    in main tree.
1372                                  */
1373                                 node_free_immediate(info->nl_net, sfn);
1374                                 err = PTR_ERR(sn);
1375                                 goto failure;
1376                         }
1377 
1378                         /* Now link new subtree to main tree */
1379                         rcu_assign_pointer(sfn->parent, fn);
1380                         rcu_assign_pointer(fn->subtree, sfn);
1381                 } else {
1382                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1383                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1384                                         offsetof(struct fib6_info, fib6_src),
1385                                         allow_create, replace_required, extack);
1386 
1387                         if (IS_ERR(sn)) {
1388                                 err = PTR_ERR(sn);
1389                                 goto failure;
1390                         }
1391                 }
1392 
1393                 if (!rcu_access_pointer(fn->leaf)) {
1394                         if (fn->fn_flags & RTN_TL_ROOT) {
1395                                 /* put back null_entry for root node */
1396                                 rcu_assign_pointer(fn->leaf,
1397                                             info->nl_net->ipv6.fib6_null_entry);
1398                         } else {
1399                                 fib6_info_hold(rt);
1400                                 rcu_assign_pointer(fn->leaf, rt);
1401                         }
1402                 }
1403                 fn = sn;
1404         }
1405 #endif
1406 
1407         err = fib6_add_rt2node(fn, rt, info, extack);
1408         if (!err) {
1409                 if (rt->nh)
1410                         list_add(&rt->nh_list, &rt->nh->f6i_list);
1411                 __fib6_update_sernum_upto_root(rt, sernum);
1412                 fib6_start_gc(info->nl_net, rt);
1413         }
1414 
1415 out:
1416         if (err) {
1417 #ifdef CONFIG_IPV6_SUBTREES
1418                 /*
1419                  * If fib6_add_1 has cleared the old leaf pointer in the
1420                  * super-tree leaf node we have to find a new one for it.
1421                  */
1422                 if (pn != fn) {
1423                         struct fib6_info *pn_leaf =
1424                                 rcu_dereference_protected(pn->leaf,
1425                                     lockdep_is_held(&table->tb6_lock));
1426                         if (pn_leaf == rt) {
1427                                 pn_leaf = NULL;
1428                                 RCU_INIT_POINTER(pn->leaf, NULL);
1429                                 fib6_info_release(rt);
1430                         }
1431                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1432                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1433                                                            pn);
1434 #if RT6_DEBUG >= 2
1435                                 if (!pn_leaf) {
1436                                         WARN_ON(!pn_leaf);
1437                                         pn_leaf =
1438                                             info->nl_net->ipv6.fib6_null_entry;
1439                                 }
1440 #endif
1441                                 fib6_info_hold(pn_leaf);
1442                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1443                         }
1444                 }
1445 #endif
1446                 goto failure;
1447         }
1448         return err;
1449 
1450 failure:
1451         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1452          * 1. fn is an intermediate node and we failed to add the new
1453          * route to it in both subtree creation failure and fib6_add_rt2node()
1454          * failure case.
1455          * 2. fn is the root node in the table and we fail to add the first
1456          * default route to it.
1457          */
1458         if (fn &&
1459             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1460              (fn->fn_flags & RTN_TL_ROOT &&
1461               !rcu_access_pointer(fn->leaf))))
1462                 fib6_repair_tree(info->nl_net, table, fn);
1463         return err;
1464 }
1465 
1466 /*
1467  *      Routing tree lookup
1468  *
1469  */
1470 
1471 struct lookup_args {
1472         int                     offset;         /* key offset on fib6_info */
1473         const struct in6_addr   *addr;          /* search key                   */
1474 };
1475 
1476 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1477                                             struct lookup_args *args)
1478 {
1479         struct fib6_node *fn;
1480         __be32 dir;
1481 
1482         if (unlikely(args->offset == 0))
1483                 return NULL;
1484 
1485         /*
1486          *      Descend on a tree
1487          */
1488 
1489         fn = root;
1490 
1491         for (;;) {
1492                 struct fib6_node *next;
1493 
1494                 dir = addr_bit_set(args->addr, fn->fn_bit);
1495 
1496                 next = dir ? rcu_dereference(fn->right) :
1497                              rcu_dereference(fn->left);
1498 
1499                 if (next) {
1500                         fn = next;
1501                         continue;
1502                 }
1503                 break;
1504         }
1505 
1506         while (fn) {
1507                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1508 
1509                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1510                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1511                         struct rt6key *key;
1512 
1513                         if (!leaf)
1514                                 goto backtrack;
1515 
1516                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1517 
1518                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1519 #ifdef CONFIG_IPV6_SUBTREES
1520                                 if (subtree) {
1521                                         struct fib6_node *sfn;
1522                                         sfn = fib6_node_lookup_1(subtree,
1523                                                                  args + 1);
1524                                         if (!sfn)
1525                                                 goto backtrack;
1526                                         fn = sfn;
1527                                 }
1528 #endif
1529                                 if (fn->fn_flags & RTN_RTINFO)
1530                                         return fn;
1531                         }
1532                 }
1533 backtrack:
1534                 if (fn->fn_flags & RTN_ROOT)
1535                         break;
1536 
1537                 fn = rcu_dereference(fn->parent);
1538         }
1539 
1540         return NULL;
1541 }
1542 
1543 /* called with rcu_read_lock() held
1544  */
1545 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1546                                    const struct in6_addr *daddr,
1547                                    const struct in6_addr *saddr)
1548 {
1549         struct fib6_node *fn;
1550         struct lookup_args args[] = {
1551                 {
1552                         .offset = offsetof(struct fib6_info, fib6_dst),
1553                         .addr = daddr,
1554                 },
1555 #ifdef CONFIG_IPV6_SUBTREES
1556                 {
1557                         .offset = offsetof(struct fib6_info, fib6_src),
1558                         .addr = saddr,
1559                 },
1560 #endif
1561                 {
1562                         .offset = 0,    /* sentinel */
1563                 }
1564         };
1565 
1566         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1567         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1568                 fn = root;
1569 
1570         return fn;
1571 }
1572 
1573 /*
1574  *      Get node with specified destination prefix (and source prefix,
1575  *      if subtrees are used)
1576  *      exact_match == true means we try to find fn with exact match of
1577  *      the passed in prefix addr
1578  *      exact_match == false means we try to find fn with longest prefix
1579  *      match of the passed in prefix addr. This is useful for finding fn
1580  *      for cached route as it will be stored in the exception table under
1581  *      the node with longest prefix length.
1582  */
1583 
1584 
1585 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1586                                        const struct in6_addr *addr,
1587                                        int plen, int offset,
1588                                        bool exact_match)
1589 {
1590         struct fib6_node *fn, *prev = NULL;
1591 
1592         for (fn = root; fn ; ) {
1593                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1594                 struct rt6key *key;
1595 
1596                 /* This node is being deleted */
1597                 if (!leaf) {
1598                         if (plen <= fn->fn_bit)
1599                                 goto out;
1600                         else
1601                                 goto next;
1602                 }
1603 
1604                 key = (struct rt6key *)((u8 *)leaf + offset);
1605 
1606                 /*
1607                  *      Prefix match
1608                  */
1609                 if (plen < fn->fn_bit ||
1610                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1611                         goto out;
1612 
1613                 if (plen == fn->fn_bit)
1614                         return fn;
1615 
1616                 if (fn->fn_flags & RTN_RTINFO)
1617                         prev = fn;
1618 
1619 next:
1620                 /*
1621                  *      We have more bits to go
1622                  */
1623                 if (addr_bit_set(addr, fn->fn_bit))
1624                         fn = rcu_dereference(fn->right);
1625                 else
1626                         fn = rcu_dereference(fn->left);
1627         }
1628 out:
1629         if (exact_match)
1630                 return NULL;
1631         else
1632                 return prev;
1633 }
1634 
1635 struct fib6_node *fib6_locate(struct fib6_node *root,
1636                               const struct in6_addr *daddr, int dst_len,
1637                               const struct in6_addr *saddr, int src_len,
1638                               bool exact_match)
1639 {
1640         struct fib6_node *fn;
1641 
1642         fn = fib6_locate_1(root, daddr, dst_len,
1643                            offsetof(struct fib6_info, fib6_dst),
1644                            exact_match);
1645 
1646 #ifdef CONFIG_IPV6_SUBTREES
1647         if (src_len) {
1648                 WARN_ON(saddr == NULL);
1649                 if (fn) {
1650                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1651 
1652                         if (subtree) {
1653                                 fn = fib6_locate_1(subtree, saddr, src_len,
1654                                            offsetof(struct fib6_info, fib6_src),
1655                                            exact_match);
1656                         }
1657                 }
1658         }
1659 #endif
1660 
1661         if (fn && fn->fn_flags & RTN_RTINFO)
1662                 return fn;
1663 
1664         return NULL;
1665 }
1666 
1667 
1668 /*
1669  *      Deletion
1670  *
1671  */
1672 
1673 static struct fib6_info *fib6_find_prefix(struct net *net,
1674                                          struct fib6_table *table,
1675                                          struct fib6_node *fn)
1676 {
1677         struct fib6_node *child_left, *child_right;
1678 
1679         if (fn->fn_flags & RTN_ROOT)
1680                 return net->ipv6.fib6_null_entry;
1681 
1682         while (fn) {
1683                 child_left = rcu_dereference_protected(fn->left,
1684                                     lockdep_is_held(&table->tb6_lock));
1685                 child_right = rcu_dereference_protected(fn->right,
1686                                     lockdep_is_held(&table->tb6_lock));
1687                 if (child_left)
1688                         return rcu_dereference_protected(child_left->leaf,
1689                                         lockdep_is_held(&table->tb6_lock));
1690                 if (child_right)
1691                         return rcu_dereference_protected(child_right->leaf,
1692                                         lockdep_is_held(&table->tb6_lock));
1693 
1694                 fn = FIB6_SUBTREE(fn);
1695         }
1696         return NULL;
1697 }
1698 
1699 /*
1700  *      Called to trim the tree of intermediate nodes when possible. "fn"
1701  *      is the node we want to try and remove.
1702  *      Need to own table->tb6_lock
1703  */
1704 
1705 static struct fib6_node *fib6_repair_tree(struct net *net,
1706                                           struct fib6_table *table,
1707                                           struct fib6_node *fn)
1708 {
1709         int children;
1710         int nstate;
1711         struct fib6_node *child;
1712         struct fib6_walker *w;
1713         int iter = 0;
1714 
1715         /* Set fn->leaf to null_entry for root node. */
1716         if (fn->fn_flags & RTN_TL_ROOT) {
1717                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1718                 return fn;
1719         }
1720 
1721         for (;;) {
1722                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1723                                             lockdep_is_held(&table->tb6_lock));
1724                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1725                                             lockdep_is_held(&table->tb6_lock));
1726                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1727                                             lockdep_is_held(&table->tb6_lock));
1728                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1729                                             lockdep_is_held(&table->tb6_lock));
1730                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1731                                             lockdep_is_held(&table->tb6_lock));
1732                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1733                                             lockdep_is_held(&table->tb6_lock));
1734                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1735                                             lockdep_is_held(&table->tb6_lock));
1736                 struct fib6_info *new_fn_leaf;
1737 
1738                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1739                 iter++;
1740 
1741                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1742                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1743                 WARN_ON(fn_leaf);
1744 
1745                 children = 0;
1746                 child = NULL;
1747                 if (fn_r)
1748                         child = fn_r, children |= 1;
1749                 if (fn_l)
1750                         child = fn_l, children |= 2;
1751 
1752                 if (children == 3 || FIB6_SUBTREE(fn)
1753 #ifdef CONFIG_IPV6_SUBTREES
1754                     /* Subtree root (i.e. fn) may have one child */
1755                     || (children && fn->fn_flags & RTN_ROOT)
1756 #endif
1757                     ) {
1758                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1759 #if RT6_DEBUG >= 2
1760                         if (!new_fn_leaf) {
1761                                 WARN_ON(!new_fn_leaf);
1762                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1763                         }
1764 #endif
1765                         fib6_info_hold(new_fn_leaf);
1766                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1767                         return pn;
1768                 }
1769 
1770 #ifdef CONFIG_IPV6_SUBTREES
1771                 if (FIB6_SUBTREE(pn) == fn) {
1772                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1773                         RCU_INIT_POINTER(pn->subtree, NULL);
1774                         nstate = FWS_L;
1775                 } else {
1776                         WARN_ON(fn->fn_flags & RTN_ROOT);
1777 #endif
1778                         if (pn_r == fn)
1779                                 rcu_assign_pointer(pn->right, child);
1780                         else if (pn_l == fn)
1781                                 rcu_assign_pointer(pn->left, child);
1782 #if RT6_DEBUG >= 2
1783                         else
1784                                 WARN_ON(1);
1785 #endif
1786                         if (child)
1787                                 rcu_assign_pointer(child->parent, pn);
1788                         nstate = FWS_R;
1789 #ifdef CONFIG_IPV6_SUBTREES
1790                 }
1791 #endif
1792 
1793                 read_lock(&net->ipv6.fib6_walker_lock);
1794                 FOR_WALKERS(net, w) {
1795                         if (!child) {
1796                                 if (w->node == fn) {
1797                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1798                                         w->node = pn;
1799                                         w->state = nstate;
1800                                 }
1801                         } else {
1802                                 if (w->node == fn) {
1803                                         w->node = child;
1804                                         if (children&2) {
1805                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1806                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1807                                         } else {
1808                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1809                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1810                                         }
1811                                 }
1812                         }
1813                 }
1814                 read_unlock(&net->ipv6.fib6_walker_lock);
1815 
1816                 node_free(net, fn);
1817                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1818                         return pn;
1819 
1820                 RCU_INIT_POINTER(pn->leaf, NULL);
1821                 fib6_info_release(pn_leaf);
1822                 fn = pn;
1823         }
1824 }
1825 
1826 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1827                            struct fib6_info __rcu **rtp, struct nl_info *info)
1828 {
1829         struct fib6_walker *w;
1830         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1831                                     lockdep_is_held(&table->tb6_lock));
1832         struct net *net = info->nl_net;
1833 
1834         RT6_TRACE("fib6_del_route\n");
1835 
1836         /* Unlink it */
1837         *rtp = rt->fib6_next;
1838         rt->fib6_node = NULL;
1839         net->ipv6.rt6_stats->fib_rt_entries--;
1840         net->ipv6.rt6_stats->fib_discarded_routes++;
1841 
1842         /* Flush all cached dst in exception table */
1843         rt6_flush_exceptions(rt);
1844 
1845         /* Reset round-robin state, if necessary */
1846         if (rcu_access_pointer(fn->rr_ptr) == rt)
1847                 fn->rr_ptr = NULL;
1848 
1849         /* Remove this entry from other siblings */
1850         if (rt->fib6_nsiblings) {
1851                 struct fib6_info *sibling, *next_sibling;
1852 
1853                 list_for_each_entry_safe(sibling, next_sibling,
1854                                          &rt->fib6_siblings, fib6_siblings)
1855                         sibling->fib6_nsiblings--;
1856                 rt->fib6_nsiblings = 0;
1857                 list_del_init(&rt->fib6_siblings);
1858                 rt6_multipath_rebalance(next_sibling);
1859         }
1860 
1861         /* Adjust walkers */
1862         read_lock(&net->ipv6.fib6_walker_lock);
1863         FOR_WALKERS(net, w) {
1864                 if (w->state == FWS_C && w->leaf == rt) {
1865                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1866                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1867                                             lockdep_is_held(&table->tb6_lock));
1868                         if (!w->leaf)
1869                                 w->state = FWS_U;
1870                 }
1871         }
1872         read_unlock(&net->ipv6.fib6_walker_lock);
1873 
1874         /* If it was last route, call fib6_repair_tree() to:
1875          * 1. For root node, put back null_entry as how the table was created.
1876          * 2. For other nodes, expunge its radix tree node.
1877          */
1878         if (!rcu_access_pointer(fn->leaf)) {
1879                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1880                         fn->fn_flags &= ~RTN_RTINFO;
1881                         net->ipv6.rt6_stats->fib_route_nodes--;
1882                 }
1883                 fn = fib6_repair_tree(net, table, fn);
1884         }
1885 
1886         fib6_purge_rt(rt, fn, net);
1887 
1888         if (!info->skip_notify_kernel)
1889                 call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1890         if (!info->skip_notify)
1891                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1892 
1893         fib6_info_release(rt);
1894 }
1895 
1896 /* Need to own table->tb6_lock */
1897 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1898 {
1899         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1900                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1901         struct fib6_table *table = rt->fib6_table;
1902         struct net *net = info->nl_net;
1903         struct fib6_info __rcu **rtp;
1904         struct fib6_info __rcu **rtp_next;
1905 
1906         if (!fn || rt == net->ipv6.fib6_null_entry)
1907                 return -ENOENT;
1908 
1909         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1910 
1911         /*
1912          *      Walk the leaf entries looking for ourself
1913          */
1914 
1915         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1916                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1917                                         lockdep_is_held(&table->tb6_lock));
1918                 if (rt == cur) {
1919                         fib6_del_route(table, fn, rtp, info);
1920                         return 0;
1921                 }
1922                 rtp_next = &cur->fib6_next;
1923         }
1924         return -ENOENT;
1925 }
1926 
1927 /*
1928  *      Tree traversal function.
1929  *
1930  *      Certainly, it is not interrupt safe.
1931  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1932  *      It means, that we can modify tree during walking
1933  *      and use this function for garbage collection, clone pruning,
1934  *      cleaning tree when a device goes down etc. etc.
1935  *
1936  *      It guarantees that every node will be traversed,
1937  *      and that it will be traversed only once.
1938  *
1939  *      Callback function w->func may return:
1940  *      0 -> continue walking.
1941  *      positive value -> walking is suspended (used by tree dumps,
1942  *      and probably by gc, if it will be split to several slices)
1943  *      negative value -> terminate walking.
1944  *
1945  *      The function itself returns:
1946  *      0   -> walk is complete.
1947  *      >0  -> walk is incomplete (i.e. suspended)
1948  *      <0  -> walk is terminated by an error.
1949  *
1950  *      This function is called with tb6_lock held.
1951  */
1952 
1953 static int fib6_walk_continue(struct fib6_walker *w)
1954 {
1955         struct fib6_node *fn, *pn, *left, *right;
1956 
1957         /* w->root should always be table->tb6_root */
1958         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1959 
1960         for (;;) {
1961                 fn = w->node;
1962                 if (!fn)
1963                         return 0;
1964 
1965                 switch (w->state) {
1966 #ifdef CONFIG_IPV6_SUBTREES
1967                 case FWS_S:
1968                         if (FIB6_SUBTREE(fn)) {
1969                                 w->node = FIB6_SUBTREE(fn);
1970                                 continue;
1971                         }
1972                         w->state = FWS_L;
1973 #endif
1974                         /* fall through */
1975                 case FWS_L:
1976                         left = rcu_dereference_protected(fn->left, 1);
1977                         if (left) {
1978                                 w->node = left;
1979                                 w->state = FWS_INIT;
1980                                 continue;
1981                         }
1982                         w->state = FWS_R;
1983                         /* fall through */
1984                 case FWS_R:
1985                         right = rcu_dereference_protected(fn->right, 1);
1986                         if (right) {
1987                                 w->node = right;
1988                                 w->state = FWS_INIT;
1989                                 continue;
1990                         }
1991                         w->state = FWS_C;
1992                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1993                         /* fall through */
1994                 case FWS_C:
1995                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1996                                 int err;
1997 
1998                                 if (w->skip) {
1999                                         w->skip--;
2000                                         goto skip;
2001                                 }
2002 
2003                                 err = w->func(w);
2004                                 if (err)
2005                                         return err;
2006 
2007                                 w->count++;
2008                                 continue;
2009                         }
2010 skip:
2011                         w->state = FWS_U;
2012                         /* fall through */
2013                 case FWS_U:
2014                         if (fn == w->root)
2015                                 return 0;
2016                         pn = rcu_dereference_protected(fn->parent, 1);
2017                         left = rcu_dereference_protected(pn->left, 1);
2018                         right = rcu_dereference_protected(pn->right, 1);
2019                         w->node = pn;
2020 #ifdef CONFIG_IPV6_SUBTREES
2021                         if (FIB6_SUBTREE(pn) == fn) {
2022                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2023                                 w->state = FWS_L;
2024                                 continue;
2025                         }
2026 #endif
2027                         if (left == fn) {
2028                                 w->state = FWS_R;
2029                                 continue;
2030                         }
2031                         if (right == fn) {
2032                                 w->state = FWS_C;
2033                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2034                                 continue;
2035                         }
2036 #if RT6_DEBUG >= 2
2037                         WARN_ON(1);
2038 #endif
2039                 }
2040         }
2041 }
2042 
2043 static int fib6_walk(struct net *net, struct fib6_walker *w)
2044 {
2045         int res;
2046 
2047         w->state = FWS_INIT;
2048         w->node = w->root;
2049 
2050         fib6_walker_link(net, w);
2051         res = fib6_walk_continue(w);
2052         if (res <= 0)
2053                 fib6_walker_unlink(net, w);
2054         return res;
2055 }
2056 
2057 static int fib6_clean_node(struct fib6_walker *w)
2058 {
2059         int res;
2060         struct fib6_info *rt;
2061         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2062         struct nl_info info = {
2063                 .nl_net = c->net,
2064                 .skip_notify = c->skip_notify,
2065         };
2066 
2067         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2068             w->node->fn_sernum != c->sernum)
2069                 w->node->fn_sernum = c->sernum;
2070 
2071         if (!c->func) {
2072                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2073                 w->leaf = NULL;
2074                 return 0;
2075         }
2076 
2077         for_each_fib6_walker_rt(w) {
2078                 res = c->func(rt, c->arg);
2079                 if (res == -1) {
2080                         w->leaf = rt;
2081                         res = fib6_del(rt, &info);
2082                         if (res) {
2083 #if RT6_DEBUG >= 2
2084                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2085                                          __func__, rt,
2086                                          rcu_access_pointer(rt->fib6_node),
2087                                          res);
2088 #endif
2089                                 continue;
2090                         }
2091                         return 0;
2092                 } else if (res == -2) {
2093                         if (WARN_ON(!rt->fib6_nsiblings))
2094                                 continue;
2095                         rt = list_last_entry(&rt->fib6_siblings,
2096                                              struct fib6_info, fib6_siblings);
2097                         continue;
2098                 }
2099                 WARN_ON(res != 0);
2100         }
2101         w->leaf = rt;
2102         return 0;
2103 }
2104 
2105 /*
2106  *      Convenient frontend to tree walker.
2107  *
2108  *      func is called on each route.
2109  *              It may return -2 -> skip multipath route.
2110  *                            -1 -> delete this route.
2111  *                            0  -> continue walking
2112  */
2113 
2114 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2115                             int (*func)(struct fib6_info *, void *arg),
2116                             int sernum, void *arg, bool skip_notify)
2117 {
2118         struct fib6_cleaner c;
2119 
2120         c.w.root = root;
2121         c.w.func = fib6_clean_node;
2122         c.w.count = 0;
2123         c.w.skip = 0;
2124         c.w.skip_in_node = 0;
2125         c.func = func;
2126         c.sernum = sernum;
2127         c.arg = arg;
2128         c.net = net;
2129         c.skip_notify = skip_notify;
2130 
2131         fib6_walk(net, &c.w);
2132 }
2133 
2134 static void __fib6_clean_all(struct net *net,
2135                              int (*func)(struct fib6_info *, void *),
2136                              int sernum, void *arg, bool skip_notify)
2137 {
2138         struct fib6_table *table;
2139         struct hlist_head *head;
2140         unsigned int h;
2141 
2142         rcu_read_lock();
2143         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2144                 head = &net->ipv6.fib_table_hash[h];
2145                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2146                         spin_lock_bh(&table->tb6_lock);
2147                         fib6_clean_tree(net, &table->tb6_root,
2148                                         func, sernum, arg, skip_notify);
2149                         spin_unlock_bh(&table->tb6_lock);
2150                 }
2151         }
2152         rcu_read_unlock();
2153 }
2154 
2155 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2156                     void *arg)
2157 {
2158         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2159 }
2160 
2161 void fib6_clean_all_skip_notify(struct net *net,
2162                                 int (*func)(struct fib6_info *, void *),
2163                                 void *arg)
2164 {
2165         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2166 }
2167 
2168 static void fib6_flush_trees(struct net *net)
2169 {
2170         int new_sernum = fib6_new_sernum(net);
2171 
2172         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2173 }
2174 
2175 /*
2176  *      Garbage collection
2177  */
2178 
2179 static int fib6_age(struct fib6_info *rt, void *arg)
2180 {
2181         struct fib6_gc_args *gc_args = arg;
2182         unsigned long now = jiffies;
2183 
2184         /*
2185          *      check addrconf expiration here.
2186          *      Routes are expired even if they are in use.
2187          */
2188 
2189         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2190                 if (time_after(now, rt->expires)) {
2191                         RT6_TRACE("expiring %p\n", rt);
2192                         return -1;
2193                 }
2194                 gc_args->more++;
2195         }
2196 
2197         /*      Also age clones in the exception table.
2198          *      Note, that clones are aged out
2199          *      only if they are not in use now.
2200          */
2201         rt6_age_exceptions(rt, gc_args, now);
2202 
2203         return 0;
2204 }
2205 
2206 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2207 {
2208         struct fib6_gc_args gc_args;
2209         unsigned long now;
2210 
2211         if (force) {
2212                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2213         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2214                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2215                 return;
2216         }
2217         gc_args.timeout = expires ? (int)expires :
2218                           net->ipv6.sysctl.ip6_rt_gc_interval;
2219         gc_args.more = 0;
2220 
2221         fib6_clean_all(net, fib6_age, &gc_args);
2222         now = jiffies;
2223         net->ipv6.ip6_rt_last_gc = now;
2224 
2225         if (gc_args.more)
2226                 mod_timer(&net->ipv6.ip6_fib_timer,
2227                           round_jiffies(now
2228                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2229         else
2230                 del_timer(&net->ipv6.ip6_fib_timer);
2231         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2232 }
2233 
2234 static void fib6_gc_timer_cb(struct timer_list *t)
2235 {
2236         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2237 
2238         fib6_run_gc(0, arg, true);
2239 }
2240 
2241 static int __net_init fib6_net_init(struct net *net)
2242 {
2243         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2244         int err;
2245 
2246         err = fib6_notifier_init(net);
2247         if (err)
2248                 return err;
2249 
2250         spin_lock_init(&net->ipv6.fib6_gc_lock);
2251         rwlock_init(&net->ipv6.fib6_walker_lock);
2252         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2253         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2254 
2255         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2256         if (!net->ipv6.rt6_stats)
2257                 goto out_timer;
2258 
2259         /* Avoid false sharing : Use at least a full cache line */
2260         size = max_t(size_t, size, L1_CACHE_BYTES);
2261 
2262         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2263         if (!net->ipv6.fib_table_hash)
2264                 goto out_rt6_stats;
2265 
2266         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2267                                           GFP_KERNEL);
2268         if (!net->ipv6.fib6_main_tbl)
2269                 goto out_fib_table_hash;
2270 
2271         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2272         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2273                            net->ipv6.fib6_null_entry);
2274         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2275                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2276         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2277 
2278 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2279         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2280                                            GFP_KERNEL);
2281         if (!net->ipv6.fib6_local_tbl)
2282                 goto out_fib6_main_tbl;
2283         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2284         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2285                            net->ipv6.fib6_null_entry);
2286         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2287                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2288         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2289 #endif
2290         fib6_tables_init(net);
2291 
2292         return 0;
2293 
2294 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2295 out_fib6_main_tbl:
2296         kfree(net->ipv6.fib6_main_tbl);
2297 #endif
2298 out_fib_table_hash:
2299         kfree(net->ipv6.fib_table_hash);
2300 out_rt6_stats:
2301         kfree(net->ipv6.rt6_stats);
2302 out_timer:
2303         fib6_notifier_exit(net);
2304         return -ENOMEM;
2305 }
2306 
2307 static void fib6_net_exit(struct net *net)
2308 {
2309         unsigned int i;
2310 
2311         del_timer_sync(&net->ipv6.ip6_fib_timer);
2312 
2313         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2314                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2315                 struct hlist_node *tmp;
2316                 struct fib6_table *tb;
2317 
2318                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2319                         hlist_del(&tb->tb6_hlist);
2320                         fib6_free_table(tb);
2321                 }
2322         }
2323 
2324         kfree(net->ipv6.fib_table_hash);
2325         kfree(net->ipv6.rt6_stats);
2326         fib6_notifier_exit(net);
2327 }
2328 
2329 static struct pernet_operations fib6_net_ops = {
2330         .init = fib6_net_init,
2331         .exit = fib6_net_exit,
2332 };
2333 
2334 int __init fib6_init(void)
2335 {
2336         int ret = -ENOMEM;
2337 
2338         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2339                                            sizeof(struct fib6_node),
2340                                            0, SLAB_HWCACHE_ALIGN,
2341                                            NULL);
2342         if (!fib6_node_kmem)
2343                 goto out;
2344 
2345         ret = register_pernet_subsys(&fib6_net_ops);
2346         if (ret)
2347                 goto out_kmem_cache_create;
2348 
2349         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2350                                    inet6_dump_fib, 0);
2351         if (ret)
2352                 goto out_unregister_subsys;
2353 
2354         __fib6_flush_trees = fib6_flush_trees;
2355 out:
2356         return ret;
2357 
2358 out_unregister_subsys:
2359         unregister_pernet_subsys(&fib6_net_ops);
2360 out_kmem_cache_create:
2361         kmem_cache_destroy(fib6_node_kmem);
2362         goto out;
2363 }
2364 
2365 void fib6_gc_cleanup(void)
2366 {
2367         unregister_pernet_subsys(&fib6_net_ops);
2368         kmem_cache_destroy(fib6_node_kmem);
2369 }
2370 
2371 #ifdef CONFIG_PROC_FS
2372 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2373 {
2374         struct fib6_info *rt = v;
2375         struct ipv6_route_iter *iter = seq->private;
2376         struct fib6_nh *fib6_nh = rt->fib6_nh;
2377         unsigned int flags = rt->fib6_flags;
2378         const struct net_device *dev;
2379 
2380         if (rt->nh)
2381                 fib6_nh = nexthop_fib6_nh(rt->nh);
2382 
2383         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2384 
2385 #ifdef CONFIG_IPV6_SUBTREES
2386         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2387 #else
2388         seq_puts(seq, "00000000000000000000000000000000 00 ");
2389 #endif
2390         if (fib6_nh->fib_nh_gw_family) {
2391                 flags |= RTF_GATEWAY;
2392                 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2393         } else {
2394                 seq_puts(seq, "00000000000000000000000000000000");
2395         }
2396 
2397         dev = fib6_nh->fib_nh_dev;
2398         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2399                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2400                    flags, dev ? dev->name : "");
2401         iter->w.leaf = NULL;
2402         return 0;
2403 }
2404 
2405 static int ipv6_route_yield(struct fib6_walker *w)
2406 {
2407         struct ipv6_route_iter *iter = w->args;
2408 
2409         if (!iter->skip)
2410                 return 1;
2411 
2412         do {
2413                 iter->w.leaf = rcu_dereference_protected(
2414                                 iter->w.leaf->fib6_next,
2415                                 lockdep_is_held(&iter->tbl->tb6_lock));
2416                 iter->skip--;
2417                 if (!iter->skip && iter->w.leaf)
2418                         return 1;
2419         } while (iter->w.leaf);
2420 
2421         return 0;
2422 }
2423 
2424 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2425                                       struct net *net)
2426 {
2427         memset(&iter->w, 0, sizeof(iter->w));
2428         iter->w.func = ipv6_route_yield;
2429         iter->w.root = &iter->tbl->tb6_root;
2430         iter->w.state = FWS_INIT;
2431         iter->w.node = iter->w.root;
2432         iter->w.args = iter;
2433         iter->sernum = iter->w.root->fn_sernum;
2434         INIT_LIST_HEAD(&iter->w.lh);
2435         fib6_walker_link(net, &iter->w);
2436 }
2437 
2438 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2439                                                     struct net *net)
2440 {
2441         unsigned int h;
2442         struct hlist_node *node;
2443 
2444         if (tbl) {
2445                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2446                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2447         } else {
2448                 h = 0;
2449                 node = NULL;
2450         }
2451 
2452         while (!node && h < FIB6_TABLE_HASHSZ) {
2453                 node = rcu_dereference_bh(
2454                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2455         }
2456         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2457 }
2458 
2459 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2460 {
2461         if (iter->sernum != iter->w.root->fn_sernum) {
2462                 iter->sernum = iter->w.root->fn_sernum;
2463                 iter->w.state = FWS_INIT;
2464                 iter->w.node = iter->w.root;
2465                 WARN_ON(iter->w.skip);
2466                 iter->w.skip = iter->w.count;
2467         }
2468 }
2469 
2470 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2471 {
2472         int r;
2473         struct fib6_info *n;
2474         struct net *net = seq_file_net(seq);
2475         struct ipv6_route_iter *iter = seq->private;
2476 
2477         if (!v)
2478                 goto iter_table;
2479 
2480         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2481         if (n) {
2482                 ++*pos;
2483                 return n;
2484         }
2485 
2486 iter_table:
2487         ipv6_route_check_sernum(iter);
2488         spin_lock_bh(&iter->tbl->tb6_lock);
2489         r = fib6_walk_continue(&iter->w);
2490         spin_unlock_bh(&iter->tbl->tb6_lock);
2491         if (r > 0) {
2492                 if (v)
2493                         ++*pos;
2494                 return iter->w.leaf;
2495         } else if (r < 0) {
2496                 fib6_walker_unlink(net, &iter->w);
2497                 return NULL;
2498         }
2499         fib6_walker_unlink(net, &iter->w);
2500 
2501         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2502         if (!iter->tbl)
2503                 return NULL;
2504 
2505         ipv6_route_seq_setup_walk(iter, net);
2506         goto iter_table;
2507 }
2508 
2509 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2510         __acquires(RCU_BH)
2511 {
2512         struct net *net = seq_file_net(seq);
2513         struct ipv6_route_iter *iter = seq->private;
2514 
2515         rcu_read_lock_bh();
2516         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2517         iter->skip = *pos;
2518 
2519         if (iter->tbl) {
2520                 ipv6_route_seq_setup_walk(iter, net);
2521                 return ipv6_route_seq_next(seq, NULL, pos);
2522         } else {
2523                 return NULL;
2524         }
2525 }
2526 
2527 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2528 {
2529         struct fib6_walker *w = &iter->w;
2530         return w->node && !(w->state == FWS_U && w->node == w->root);
2531 }
2532 
2533 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2534         __releases(RCU_BH)
2535 {
2536         struct net *net = seq_file_net(seq);
2537         struct ipv6_route_iter *iter = seq->private;
2538 
2539         if (ipv6_route_iter_active(iter))
2540                 fib6_walker_unlink(net, &iter->w);
2541 
2542         rcu_read_unlock_bh();
2543 }
2544 
2545 const struct seq_operations ipv6_route_seq_ops = {
2546         .start  = ipv6_route_seq_start,
2547         .next   = ipv6_route_seq_next,
2548         .stop   = ipv6_route_seq_stop,
2549         .show   = ipv6_route_seq_show
2550 };
2551 #endif /* CONFIG_PROC_FS */

/* [<][>][^][v][top][bottom][index][help] */