net/ipv6/ip6_fib.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      Forwarding Information Database
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  *
   9  *      Changes:
  10  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  11  *                              remove ip6_null_entry from the top of
  12  *                              routing table.
  13  *      Ville Nuorvala:         Fixed routing subtrees.
  14  */
  15
  16 #define pr_fmt(fmt) "IPv6: " fmt
  17
  18 #include <linux/errno.h>
  19 #include <linux/types.h>
  20 #include <linux/net.h>
  21 #include <linux/route.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/in6.h>
  24 #include <linux/init.h>
  25 #include <linux/list.h>
  26 #include <linux/slab.h>
  27
  28 #include <net/ip.h>
  29 #include <net/ipv6.h>
  30 #include <net/ndisc.h>
  31 #include <net/addrconf.h>
  32 #include <net/lwtunnel.h>
  33 #include <net/fib_notifier.h>
  34
  35 #include <net/ip6_fib.h>
  36 #include <net/ip6_route.h>
  37
  38 static struct kmem_cache *fib6_node_kmem __read_mostly;
  39
  40 struct fib6_cleaner {
  41         struct fib6_walker w;
  42         struct net *net;
  43         int (*func)(struct fib6_info *, void *arg);
  44         int sernum;
  45         void *arg;
  46         bool skip_notify;
  47 };
  48
  49 #ifdef CONFIG_IPV6_SUBTREES
  50 #define FWS_INIT FWS_S
  51 #else
  52 #define FWS_INIT FWS_L
  53 #endif
  54
  55 static struct fib6_info *fib6_find_prefix(struct net *net,
  56                                          struct fib6_table *table,
  57                                          struct fib6_node *fn);
  58 static struct fib6_node *fib6_repair_tree(struct net *net,
  59                                           struct fib6_table *table,
  60                                           struct fib6_node *fn);
  61 static int fib6_walk(struct net *net, struct fib6_walker *w);
  62 static int fib6_walk_continue(struct fib6_walker *w);
  63
  64 /*
  65  *      A routing update causes an increase of the serial number on the
  66  *      affected subtree. This allows for cached routes to be asynchronously
  67  *      tested when modifications are made to the destination cache as a
  68  *      result of redirects, path MTU changes, etc.
  69  */
  70
  71 static void fib6_gc_timer_cb(struct timer_list *t);
  72
  73 #define FOR_WALKERS(net, w) \
  74         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  75
  76 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  77 {
  78         write_lock_bh(&net->ipv6.fib6_walker_lock);
  79         list_add(&w->lh, &net->ipv6.fib6_walkers);
  80         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  81 }
  82
  83 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  84 {
  85         write_lock_bh(&net->ipv6.fib6_walker_lock);
  86         list_del(&w->lh);
  87         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  88 }
  89
  90 static int fib6_new_sernum(struct net *net)
  91 {
  92         int new, old;
  93
  94         do {
  95                 old = atomic_read(&net->ipv6.fib6_sernum);
  96                 new = old < INT_MAX ? old + 1 : 1;
  97         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
  98                                 old, new) != old);
  99         return new;
 100 }
 101
 102 enum {
 103         FIB6_NO_SERNUM_CHANGE = 0,
 104 };
 105
 106 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 107 {
 108         struct fib6_node *fn;
 109
 110         fn = rcu_dereference_protected(f6i->fib6_node,
 111                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 112         if (fn)
 113                 WRITE_ONCE(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 = pol_lookup_func(lookup,
 318                         net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 319         if (rt->dst.error == -EAGAIN) {
 320                 ip6_rt_put_flags(rt, flags);
 321                 rt = net->ipv6.ip6_null_entry;
 322                 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
 323                         dst_hold(&rt->dst);
 324         }
 325
 326         return &rt->dst;
 327 }
 328
 329 /* called with rcu lock held; no reference taken on fib6_info */
 330 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 331                 struct fib6_result *res, int flags)
 332 {
 333         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
 334                                  res, flags);
 335 }
 336
 337 static void __net_init fib6_tables_init(struct net *net)
 338 {
 339         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 340 }
 341
 342 #endif
 343
 344 unsigned int fib6_tables_seq_read(struct net *net)
 345 {
 346         unsigned int h, fib_seq = 0;
 347
 348         rcu_read_lock();
 349         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 350                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 351                 struct fib6_table *tb;
 352
 353                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 354                         fib_seq += tb->fib_seq;
 355         }
 356         rcu_read_unlock();
 357
 358         return fib_seq;
 359 }
 360
 361 static int call_fib6_entry_notifier(struct notifier_block *nb,
 362                                     enum fib_event_type event_type,
 363                                     struct fib6_info *rt,
 364                                     struct netlink_ext_ack *extack)
 365 {
 366         struct fib6_entry_notifier_info info = {
 367                 .info.extack = extack,
 368                 .rt = rt,
 369         };
 370
 371         return call_fib6_notifier(nb, event_type, &info.info);
 372 }
 373
 374 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
 375                                               enum fib_event_type event_type,
 376                                               struct fib6_info *rt,
 377                                               unsigned int nsiblings,
 378                                               struct netlink_ext_ack *extack)
 379 {
 380         struct fib6_entry_notifier_info info = {
 381                 .info.extack = extack,
 382                 .rt = rt,
 383                 .nsiblings = nsiblings,
 384         };
 385
 386         return call_fib6_notifier(nb, event_type, &info.info);
 387 }
 388
 389 int call_fib6_entry_notifiers(struct net *net,
 390                               enum fib_event_type event_type,
 391                               struct fib6_info *rt,
 392                               struct netlink_ext_ack *extack)
 393 {
 394         struct fib6_entry_notifier_info info = {
 395                 .info.extack = extack,
 396                 .rt = rt,
 397         };
 398
 399         rt->fib6_table->fib_seq++;
 400         return call_fib6_notifiers(net, event_type, &info.info);
 401 }
 402
 403 int call_fib6_multipath_entry_notifiers(struct net *net,
 404                                         enum fib_event_type event_type,
 405                                         struct fib6_info *rt,
 406                                         unsigned int nsiblings,
 407                                         struct netlink_ext_ack *extack)
 408 {
 409         struct fib6_entry_notifier_info info = {
 410                 .info.extack = extack,
 411                 .rt = rt,
 412                 .nsiblings = nsiblings,
 413         };
 414
 415         rt->fib6_table->fib_seq++;
 416         return call_fib6_notifiers(net, event_type, &info.info);
 417 }
 418
 419 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
 420 {
 421         struct fib6_entry_notifier_info info = {
 422                 .rt = rt,
 423                 .nsiblings = rt->fib6_nsiblings,
 424         };
 425
 426         rt->fib6_table->fib_seq++;
 427         return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
 428 }
 429
 430 struct fib6_dump_arg {
 431         struct net *net;
 432         struct notifier_block *nb;
 433         struct netlink_ext_ack *extack;
 434 };
 435
 436 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 437 {
 438         enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
 439         int err;
 440
 441         if (!rt || rt == arg->net->ipv6.fib6_null_entry)
 442                 return 0;
 443
 444         if (rt->fib6_nsiblings)
 445                 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
 446                                                          rt,
 447                                                          rt->fib6_nsiblings,
 448                                                          arg->extack);
 449         else
 450                 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
 451                                                arg->extack);
 452
 453         return err;
 454 }
 455
 456 static int fib6_node_dump(struct fib6_walker *w)
 457 {
 458         int err;
 459
 460         err = fib6_rt_dump(w->leaf, w->args);
 461         w->leaf = NULL;
 462         return err;
 463 }
 464
 465 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
 466                            struct fib6_walker *w)
 467 {
 468         int err;
 469
 470         w->root = &tb->tb6_root;
 471         spin_lock_bh(&tb->tb6_lock);
 472         err = fib6_walk(net, w);
 473         spin_unlock_bh(&tb->tb6_lock);
 474         return err;
 475 }
 476
 477 /* Called with rcu_read_lock() */
 478 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
 479                      struct netlink_ext_ack *extack)
 480 {
 481         struct fib6_dump_arg arg;
 482         struct fib6_walker *w;
 483         unsigned int h;
 484         int err = 0;
 485
 486         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 487         if (!w)
 488                 return -ENOMEM;
 489
 490         w->func = fib6_node_dump;
 491         arg.net = net;
 492         arg.nb = nb;
 493         arg.extack = extack;
 494         w->args = &arg;
 495
 496         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 497                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 498                 struct fib6_table *tb;
 499
 500                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 501                         err = fib6_table_dump(net, tb, w);
 502                         if (err < 0)
 503                                 goto out;
 504                 }
 505         }
 506
 507 out:
 508         kfree(w);
 509
 510         return err;
 511 }
 512
 513 static int fib6_dump_node(struct fib6_walker *w)
 514 {
 515         int res;
 516         struct fib6_info *rt;
 517
 518         for_each_fib6_walker_rt(w) {
 519                 res = rt6_dump_route(rt, w->args, w->skip_in_node);
 520                 if (res >= 0) {
 521                         /* Frame is full, suspend walking */
 522                         w->leaf = rt;
 523
 524                         /* We'll restart from this node, so if some routes were
 525                          * already dumped, skip them next time.
 526                          */
 527                         w->skip_in_node += res;
 528
 529                         return 1;
 530                 }
 531                 w->skip_in_node = 0;
 532
 533                 /* Multipath routes are dumped in one route with the
 534                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 535                  * last sibling of this route (no need to dump the
 536                  * sibling routes again)
 537                  */
 538                 if (rt->fib6_nsiblings)
 539                         rt = list_last_entry(&rt->fib6_siblings,
 540                                              struct fib6_info,
 541                                              fib6_siblings);
 542         }
 543         w->leaf = NULL;
 544         return 0;
 545 }
 546
 547 static void fib6_dump_end(struct netlink_callback *cb)
 548 {
 549         struct net *net = sock_net(cb->skb->sk);
 550         struct fib6_walker *w = (void *)cb->args[2];
 551
 552         if (w) {
 553                 if (cb->args[4]) {
 554                         cb->args[4] = 0;
 555                         fib6_walker_unlink(net, w);
 556                 }
 557                 cb->args[2] = 0;
 558                 kfree(w);
 559         }
 560         cb->done = (void *)cb->args[3];
 561         cb->args[1] = 3;
 562 }
 563
 564 static int fib6_dump_done(struct netlink_callback *cb)
 565 {
 566         fib6_dump_end(cb);
 567         return cb->done ? cb->done(cb) : 0;
 568 }
 569
 570 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 571                            struct netlink_callback *cb)
 572 {
 573         struct net *net = sock_net(skb->sk);
 574         struct fib6_walker *w;
 575         int res;
 576
 577         w = (void *)cb->args[2];
 578         w->root = &table->tb6_root;
 579
 580         if (cb->args[4] == 0) {
 581                 w->count = 0;
 582                 w->skip = 0;
 583                 w->skip_in_node = 0;
 584
 585                 spin_lock_bh(&table->tb6_lock);
 586                 res = fib6_walk(net, w);
 587                 spin_unlock_bh(&table->tb6_lock);
 588                 if (res > 0) {
 589                         cb->args[4] = 1;
 590                         cb->args[5] = READ_ONCE(w->root->fn_sernum);
 591                 }
 592         } else {
 593                 int sernum = READ_ONCE(w->root->fn_sernum);
 594                 if (cb->args[5] != sernum) {
 595                         /* Begin at the root if the tree changed */
 596                         cb->args[5] = sernum;
 597                         w->state = FWS_INIT;
 598                         w->node = w->root;
 599                         w->skip = w->count;
 600                         w->skip_in_node = 0;
 601                 } else
 602                         w->skip = 0;
 603
 604                 spin_lock_bh(&table->tb6_lock);
 605                 res = fib6_walk_continue(w);
 606                 spin_unlock_bh(&table->tb6_lock);
 607                 if (res <= 0) {
 608                         fib6_walker_unlink(net, w);
 609                         cb->args[4] = 0;
 610                 }
 611         }
 612
 613         return res;
 614 }
 615
 616 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 617 {
 618         struct rt6_rtnl_dump_arg arg = { .filter.dump_exceptions = true,
 619                                          .filter.dump_routes = true };
 620         const struct nlmsghdr *nlh = cb->nlh;
 621         struct net *net = sock_net(skb->sk);
 622         unsigned int h, s_h;
 623         unsigned int e = 0, s_e;
 624         struct fib6_walker *w;
 625         struct fib6_table *tb;
 626         struct hlist_head *head;
 627         int res = 0;
 628
 629         if (cb->strict_check) {
 630                 int err;
 631
 632                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 633                 if (err < 0)
 634                         return err;
 635         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 636                 struct rtmsg *rtm = nlmsg_data(nlh);
 637
 638                 if (rtm->rtm_flags & RTM_F_PREFIX)
 639                         arg.filter.flags = RTM_F_PREFIX;
 640         }
 641
 642         w = (void *)cb->args[2];
 643         if (!w) {
 644                 /* New dump:
 645                  *
 646                  * 1. hook callback destructor.
 647                  */
 648                 cb->args[3] = (long)cb->done;
 649                 cb->done = fib6_dump_done;
 650
 651                 /*
 652                  * 2. allocate and initialize walker.
 653                  */
 654                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 655                 if (!w)
 656                         return -ENOMEM;
 657                 w->func = fib6_dump_node;
 658                 cb->args[2] = (long)w;
 659         }
 660
 661         arg.skb = skb;
 662         arg.cb = cb;
 663         arg.net = net;
 664         w->args = &arg;
 665
 666         if (arg.filter.table_id) {
 667                 tb = fib6_get_table(net, arg.filter.table_id);
 668                 if (!tb) {
 669                         if (rtnl_msg_family(cb->nlh) != PF_INET6)
 670                                 goto out;
 671
 672                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 673                         return -ENOENT;
 674                 }
 675
 676                 if (!cb->args[0]) {
 677                         res = fib6_dump_table(tb, skb, cb);
 678                         if (!res)
 679                                 cb->args[0] = 1;
 680                 }
 681                 goto out;
 682         }
 683
 684         s_h = cb->args[0];
 685         s_e = cb->args[1];
 686
 687         rcu_read_lock();
 688         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 689                 e = 0;
 690                 head = &net->ipv6.fib_table_hash[h];
 691                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 692                         if (e < s_e)
 693                                 goto next;
 694                         res = fib6_dump_table(tb, skb, cb);
 695                         if (res != 0)
 696                                 goto out_unlock;
 697 next:
 698                         e++;
 699                 }
 700         }
 701 out_unlock:
 702         rcu_read_unlock();
 703         cb->args[1] = e;
 704         cb->args[0] = h;
 705 out:
 706         res = res < 0 ? res : skb->len;
 707         if (res <= 0)
 708                 fib6_dump_end(cb);
 709         return res;
 710 }
 711
 712 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 713 {
 714         if (!f6i)
 715                 return;
 716
 717         if (f6i->fib6_metrics == &dst_default_metrics) {
 718                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 719
 720                 if (!p)
 721                         return;
 722
 723                 refcount_set(&p->refcnt, 1);
 724                 f6i->fib6_metrics = p;
 725         }
 726
 727         f6i->fib6_metrics->metrics[metric - 1] = val;
 728 }
 729
 730 /*
 731  *      Routing Table
 732  *
 733  *      return the appropriate node for a routing tree "add" operation
 734  *      by either creating and inserting or by returning an existing
 735  *      node.
 736  */
 737
 738 static struct fib6_node *fib6_add_1(struct net *net,
 739                                     struct fib6_table *table,
 740                                     struct fib6_node *root,
 741                                     struct in6_addr *addr, int plen,
 742                                     int offset, int allow_create,
 743                                     int replace_required,
 744                                     struct netlink_ext_ack *extack)
 745 {
 746         struct fib6_node *fn, *in, *ln;
 747         struct fib6_node *pn = NULL;
 748         struct rt6key *key;
 749         int     bit;
 750         __be32  dir = 0;
 751
 752         RT6_TRACE("fib6_add_1\n");
 753
 754         /* insert node in tree */
 755
 756         fn = root;
 757
 758         do {
 759                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 760                                             lockdep_is_held(&table->tb6_lock));
 761                 key = (struct rt6key *)((u8 *)leaf + offset);
 762
 763                 /*
 764                  *      Prefix match
 765                  */
 766                 if (plen < fn->fn_bit ||
 767                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 768                         if (!allow_create) {
 769                                 if (replace_required) {
 770                                         NL_SET_ERR_MSG(extack,
 771                                                        "Can not replace route - no match found");
 772                                         pr_warn("Can't replace route, no match found\n");
 773                                         return ERR_PTR(-ENOENT);
 774                                 }
 775                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 776                         }
 777                         goto insert_above;
 778                 }
 779
 780                 /*
 781                  *      Exact match ?
 782                  */
 783
 784                 if (plen == fn->fn_bit) {
 785                         /* clean up an intermediate node */
 786                         if (!(fn->fn_flags & RTN_RTINFO)) {
 787                                 RCU_INIT_POINTER(fn->leaf, NULL);
 788                                 fib6_info_release(leaf);
 789                         /* remove null_entry in the root node */
 790                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 791                                    rcu_access_pointer(fn->leaf) ==
 792                                    net->ipv6.fib6_null_entry) {
 793                                 RCU_INIT_POINTER(fn->leaf, NULL);
 794                         }
 795
 796                         return fn;
 797                 }
 798
 799                 /*
 800                  *      We have more bits to go
 801                  */
 802
 803                 /* Try to walk down on tree. */
 804                 dir = addr_bit_set(addr, fn->fn_bit);
 805                 pn = fn;
 806                 fn = dir ?
 807                      rcu_dereference_protected(fn->right,
 808                                         lockdep_is_held(&table->tb6_lock)) :
 809                      rcu_dereference_protected(fn->left,
 810                                         lockdep_is_held(&table->tb6_lock));
 811         } while (fn);
 812
 813         if (!allow_create) {
 814                 /* We should not create new node because
 815                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 816                  * I assume it is safe to require NLM_F_CREATE when
 817                  * REPLACE flag is used! Later we may want to remove the
 818                  * check for replace_required, because according
 819                  * to netlink specification, NLM_F_CREATE
 820                  * MUST be specified if new route is created.
 821                  * That would keep IPv6 consistent with IPv4
 822                  */
 823                 if (replace_required) {
 824                         NL_SET_ERR_MSG(extack,
 825                                        "Can not replace route - no match found");
 826                         pr_warn("Can't replace route, no match found\n");
 827                         return ERR_PTR(-ENOENT);
 828                 }
 829                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 830         }
 831         /*
 832          *      We walked to the bottom of tree.
 833          *      Create new leaf node without children.
 834          */
 835
 836         ln = node_alloc(net);
 837
 838         if (!ln)
 839                 return ERR_PTR(-ENOMEM);
 840         ln->fn_bit = plen;
 841         RCU_INIT_POINTER(ln->parent, pn);
 842
 843         if (dir)
 844                 rcu_assign_pointer(pn->right, ln);
 845         else
 846                 rcu_assign_pointer(pn->left, ln);
 847
 848         return ln;
 849
 850
 851 insert_above:
 852         /*
 853          * split since we don't have a common prefix anymore or
 854          * we have a less significant route.
 855          * we've to insert an intermediate node on the list
 856          * this new node will point to the one we need to create
 857          * and the current
 858          */
 859
 860         pn = rcu_dereference_protected(fn->parent,
 861                                        lockdep_is_held(&table->tb6_lock));
 862
 863         /* find 1st bit in difference between the 2 addrs.
 864
 865            See comment in __ipv6_addr_diff: bit may be an invalid value,
 866            but if it is >= plen, the value is ignored in any case.
 867          */
 868
 869         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 870
 871         /*
 872          *              (intermediate)[in]
 873          *                /        \
 874          *      (new leaf node)[ln] (old node)[fn]
 875          */
 876         if (plen > bit) {
 877                 in = node_alloc(net);
 878                 ln = node_alloc(net);
 879
 880                 if (!in || !ln) {
 881                         if (in)
 882                                 node_free_immediate(net, in);
 883                         if (ln)
 884                                 node_free_immediate(net, ln);
 885                         return ERR_PTR(-ENOMEM);
 886                 }
 887
 888                 /*
 889                  * new intermediate node.
 890                  * RTN_RTINFO will
 891                  * be off since that an address that chooses one of
 892                  * the branches would not match less specific routes
 893                  * in the other branch
 894                  */
 895
 896                 in->fn_bit = bit;
 897
 898                 RCU_INIT_POINTER(in->parent, pn);
 899                 in->leaf = fn->leaf;
 900                 fib6_info_hold(rcu_dereference_protected(in->leaf,
 901                                 lockdep_is_held(&table->tb6_lock)));
 902
 903                 /* update parent pointer */
 904                 if (dir)
 905                         rcu_assign_pointer(pn->right, in);
 906                 else
 907                         rcu_assign_pointer(pn->left, in);
 908
 909                 ln->fn_bit = plen;
 910
 911                 RCU_INIT_POINTER(ln->parent, in);
 912                 rcu_assign_pointer(fn->parent, in);
 913
 914                 if (addr_bit_set(addr, bit)) {
 915                         rcu_assign_pointer(in->right, ln);
 916                         rcu_assign_pointer(in->left, fn);
 917                 } else {
 918                         rcu_assign_pointer(in->left, ln);
 919                         rcu_assign_pointer(in->right, fn);
 920                 }
 921         } else { /* plen <= bit */
 922
 923                 /*
 924                  *              (new leaf node)[ln]
 925                  *                /        \
 926                  *           (old node)[fn] NULL
 927                  */
 928
 929                 ln = node_alloc(net);
 930
 931                 if (!ln)
 932                         return ERR_PTR(-ENOMEM);
 933
 934                 ln->fn_bit = plen;
 935
 936                 RCU_INIT_POINTER(ln->parent, pn);
 937
 938                 if (addr_bit_set(&key->addr, plen))
 939                         RCU_INIT_POINTER(ln->right, fn);
 940                 else
 941                         RCU_INIT_POINTER(ln->left, fn);
 942
 943                 rcu_assign_pointer(fn->parent, ln);
 944
 945                 if (dir)
 946                         rcu_assign_pointer(pn->right, ln);
 947                 else
 948                         rcu_assign_pointer(pn->left, ln);
 949         }
 950         return ln;
 951 }
 952
 953 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
 954                                   const struct fib6_info *match,
 955                                   const struct fib6_table *table)
 956 {
 957         int cpu;
 958
 959         if (!fib6_nh->rt6i_pcpu)
 960                 return;
 961
 962         /* release the reference to this fib entry from
 963          * all of its cached pcpu routes
 964          */
 965         for_each_possible_cpu(cpu) {
 966                 struct rt6_info **ppcpu_rt;
 967                 struct rt6_info *pcpu_rt;
 968
 969                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
 970                 pcpu_rt = *ppcpu_rt;
 971
 972                 /* only dropping the 'from' reference if the cached route
 973                  * is using 'match'. The cached pcpu_rt->from only changes
 974                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
 975                  * change from one fib6_info reference to another
 976                  */
 977                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
 978                         struct fib6_info *from;
 979
 980                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 981                         fib6_info_release(from);
 982                 }
 983         }
 984 }
 985
 986 struct fib6_nh_pcpu_arg {
 987         struct fib6_info        *from;
 988         const struct fib6_table *table;
 989 };
 990
 991 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
 992 {
 993         struct fib6_nh_pcpu_arg *arg = _arg;
 994
 995         __fib6_drop_pcpu_from(nh, arg->from, arg->table);
 996         return 0;
 997 }
 998
 999 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1000                                 const struct fib6_table *table)
1001 {
1002         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1003          * while we are cleaning them here.
1004          */
1005         f6i->fib6_destroying = 1;
1006         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1007
1008         if (f6i->nh) {
1009                 struct fib6_nh_pcpu_arg arg = {
1010                         .from = f6i,
1011                         .table = table
1012                 };
1013
1014                 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1015                                          &arg);
1016         } else {
1017                 struct fib6_nh *fib6_nh;
1018
1019                 fib6_nh = f6i->fib6_nh;
1020                 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1021         }
1022 }
1023
1024 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1025                           struct net *net)
1026 {
1027         struct fib6_table *table = rt->fib6_table;
1028
1029         /* Flush all cached dst in exception table */
1030         rt6_flush_exceptions(rt);
1031         fib6_drop_pcpu_from(rt, table);
1032
1033         if (rt->nh && !list_empty(&rt->nh_list))
1034                 list_del_init(&rt->nh_list);
1035
1036         if (refcount_read(&rt->fib6_ref) != 1) {
1037                 /* This route is used as dummy address holder in some split
1038                  * nodes. It is not leaked, but it still holds other resources,
1039                  * which must be released in time. So, scan ascendant nodes
1040                  * and replace dummy references to this route with references
1041                  * to still alive ones.
1042                  */
1043                 while (fn) {
1044                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1045                                             lockdep_is_held(&table->tb6_lock));
1046                         struct fib6_info *new_leaf;
1047                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1048                                 new_leaf = fib6_find_prefix(net, table, fn);
1049                                 fib6_info_hold(new_leaf);
1050
1051                                 rcu_assign_pointer(fn->leaf, new_leaf);
1052                                 fib6_info_release(rt);
1053                         }
1054                         fn = rcu_dereference_protected(fn->parent,
1055                                     lockdep_is_held(&table->tb6_lock));
1056                 }
1057         }
1058 }
1059
1060 /*
1061  *      Insert routing information in a node.
1062  */
1063
1064 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1065                             struct nl_info *info,
1066                             struct netlink_ext_ack *extack)
1067 {
1068         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1069                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1070         struct fib6_info *iter = NULL;
1071         struct fib6_info __rcu **ins;
1072         struct fib6_info __rcu **fallback_ins = NULL;
1073         int replace = (info->nlh &&
1074                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1075         int add = (!info->nlh ||
1076                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
1077         int found = 0;
1078         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1079         bool notify_sibling_rt = false;
1080         u16 nlflags = NLM_F_EXCL;
1081         int err;
1082
1083         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1084                 nlflags |= NLM_F_APPEND;
1085
1086         ins = &fn->leaf;
1087
1088         for (iter = leaf; iter;
1089              iter = rcu_dereference_protected(iter->fib6_next,
1090                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1091                 /*
1092                  *      Search for duplicates
1093                  */
1094
1095                 if (iter->fib6_metric == rt->fib6_metric) {
1096                         /*
1097                          *      Same priority level
1098                          */
1099                         if (info->nlh &&
1100                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1101                                 return -EEXIST;
1102
1103                         nlflags &= ~NLM_F_EXCL;
1104                         if (replace) {
1105                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1106                                         found++;
1107                                         break;
1108                                 }
1109                                 fallback_ins = fallback_ins ?: ins;
1110                                 goto next_iter;
1111                         }
1112
1113                         if (rt6_duplicate_nexthop(iter, rt)) {
1114                                 if (rt->fib6_nsiblings)
1115                                         rt->fib6_nsiblings = 0;
1116                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1117                                         return -EEXIST;
1118                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1119                                         fib6_clean_expires(iter);
1120                                 else
1121                                         fib6_set_expires(iter, rt->expires);
1122
1123                                 if (rt->fib6_pmtu)
1124                                         fib6_metric_set(iter, RTAX_MTU,
1125                                                         rt->fib6_pmtu);
1126                                 return -EEXIST;
1127                         }
1128                         /* If we have the same destination and the same metric,
1129                          * but not the same gateway, then the route we try to
1130                          * add is sibling to this route, increment our counter
1131                          * of siblings, and later we will add our route to the
1132                          * list.
1133                          * Only static routes (which don't have flag
1134                          * RTF_EXPIRES) are used for ECMPv6.
1135                          *
1136                          * To avoid long list, we only had siblings if the
1137                          * route have a gateway.
1138                          */
1139                         if (rt_can_ecmp &&
1140                             rt6_qualify_for_ecmp(iter))
1141                                 rt->fib6_nsiblings++;
1142                 }
1143
1144                 if (iter->fib6_metric > rt->fib6_metric)
1145                         break;
1146
1147 next_iter:
1148                 ins = &iter->fib6_next;
1149         }
1150
1151         if (fallback_ins && !found) {
1152                 /* No matching route with same ecmp-able-ness found, replace
1153                  * first matching route
1154                  */
1155                 ins = fallback_ins;
1156                 iter = rcu_dereference_protected(*ins,
1157                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1158                 found++;
1159         }
1160
1161         /* Reset round-robin state, if necessary */
1162         if (ins == &fn->leaf)
1163                 fn->rr_ptr = NULL;
1164
1165         /* Link this route to others same route. */
1166         if (rt->fib6_nsiblings) {
1167                 unsigned int fib6_nsiblings;
1168                 struct fib6_info *sibling, *temp_sibling;
1169
1170                 /* Find the first route that have the same metric */
1171                 sibling = leaf;
1172                 notify_sibling_rt = true;
1173                 while (sibling) {
1174                         if (sibling->fib6_metric == rt->fib6_metric &&
1175                             rt6_qualify_for_ecmp(sibling)) {
1176                                 list_add_tail(&rt->fib6_siblings,
1177                                               &sibling->fib6_siblings);
1178                                 break;
1179                         }
1180                         sibling = rcu_dereference_protected(sibling->fib6_next,
1181                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1182                         notify_sibling_rt = false;
1183                 }
1184                 /* For each sibling in the list, increment the counter of
1185                  * siblings. BUG() if counters does not match, list of siblings
1186                  * is broken!
1187                  */
1188                 fib6_nsiblings = 0;
1189                 list_for_each_entry_safe(sibling, temp_sibling,
1190                                          &rt->fib6_siblings, fib6_siblings) {
1191                         sibling->fib6_nsiblings++;
1192                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1193                         fib6_nsiblings++;
1194                 }
1195                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1196                 rt6_multipath_rebalance(temp_sibling);
1197         }
1198
1199         /*
1200          *      insert node
1201          */
1202         if (!replace) {
1203                 if (!add)
1204                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1205
1206 add:
1207                 nlflags |= NLM_F_CREATE;
1208
1209                 /* The route should only be notified if it is the first
1210                  * route in the node or if it is added as a sibling
1211                  * route to the first route in the node.
1212                  */
1213                 if (!info->skip_notify_kernel &&
1214                     (notify_sibling_rt || ins == &fn->leaf)) {
1215                         enum fib_event_type fib_event;
1216
1217                         if (notify_sibling_rt)
1218                                 fib_event = FIB_EVENT_ENTRY_APPEND;
1219                         else
1220                                 fib_event = FIB_EVENT_ENTRY_REPLACE;
1221                         err = call_fib6_entry_notifiers(info->nl_net,
1222                                                         fib_event, rt,
1223                                                         extack);
1224                         if (err) {
1225                                 struct fib6_info *sibling, *next_sibling;
1226
1227                                 /* If the route has siblings, then it first
1228                                  * needs to be unlinked from them.
1229                                  */
1230                                 if (!rt->fib6_nsiblings)
1231                                         return err;
1232
1233                                 list_for_each_entry_safe(sibling, next_sibling,
1234                                                          &rt->fib6_siblings,
1235                                                          fib6_siblings)
1236                                         sibling->fib6_nsiblings--;
1237                                 rt->fib6_nsiblings = 0;
1238                                 list_del_init(&rt->fib6_siblings);
1239                                 rt6_multipath_rebalance(next_sibling);
1240                                 return err;
1241                         }
1242                 }
1243
1244                 rcu_assign_pointer(rt->fib6_next, iter);
1245                 fib6_info_hold(rt);
1246                 rcu_assign_pointer(rt->fib6_node, fn);
1247                 rcu_assign_pointer(*ins, rt);
1248                 if (!info->skip_notify)
1249                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1250                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1251
1252                 if (!(fn->fn_flags & RTN_RTINFO)) {
1253                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1254                         fn->fn_flags |= RTN_RTINFO;
1255                 }
1256
1257         } else {
1258                 int nsiblings;
1259
1260                 if (!found) {
1261                         if (add)
1262                                 goto add;
1263                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1264                         return -ENOENT;
1265                 }
1266
1267                 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1268                         err = call_fib6_entry_notifiers(info->nl_net,
1269                                                         FIB_EVENT_ENTRY_REPLACE,
1270                                                         rt, extack);
1271                         if (err)
1272                                 return err;
1273                 }
1274
1275                 fib6_info_hold(rt);
1276                 rcu_assign_pointer(rt->fib6_node, fn);
1277                 rt->fib6_next = iter->fib6_next;
1278                 rcu_assign_pointer(*ins, rt);
1279                 if (!info->skip_notify)
1280                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1281                 if (!(fn->fn_flags & RTN_RTINFO)) {
1282                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1283                         fn->fn_flags |= RTN_RTINFO;
1284                 }
1285                 nsiblings = iter->fib6_nsiblings;
1286                 iter->fib6_node = NULL;
1287                 fib6_purge_rt(iter, fn, info->nl_net);
1288                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1289                         fn->rr_ptr = NULL;
1290                 fib6_info_release(iter);
1291
1292                 if (nsiblings) {
1293                         /* Replacing an ECMP route, remove all siblings */
1294                         ins = &rt->fib6_next;
1295                         iter = rcu_dereference_protected(*ins,
1296                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1297                         while (iter) {
1298                                 if (iter->fib6_metric > rt->fib6_metric)
1299                                         break;
1300                                 if (rt6_qualify_for_ecmp(iter)) {
1301                                         *ins = iter->fib6_next;
1302                                         iter->fib6_node = NULL;
1303                                         fib6_purge_rt(iter, fn, info->nl_net);
1304                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1305                                                 fn->rr_ptr = NULL;
1306                                         fib6_info_release(iter);
1307                                         nsiblings--;
1308                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1309                                 } else {
1310                                         ins = &iter->fib6_next;
1311                                 }
1312                                 iter = rcu_dereference_protected(*ins,
1313                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1314                         }
1315                         WARN_ON(nsiblings != 0);
1316                 }
1317         }
1318
1319         return 0;
1320 }
1321
1322 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1323 {
1324         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1325             (rt->fib6_flags & RTF_EXPIRES))
1326                 mod_timer(&net->ipv6.ip6_fib_timer,
1327                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1328 }
1329
1330 void fib6_force_start_gc(struct net *net)
1331 {
1332         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1333                 mod_timer(&net->ipv6.ip6_fib_timer,
1334                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1335 }
1336
1337 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1338                                            int sernum)
1339 {
1340         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1341                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1342
1343         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1344         smp_wmb();
1345         while (fn) {
1346                 WRITE_ONCE(fn->fn_sernum, sernum);
1347                 fn = rcu_dereference_protected(fn->parent,
1348                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1349         }
1350 }
1351
1352 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1353 {
1354         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1355 }
1356
1357 /* allow ipv4 to update sernum via ipv6_stub */
1358 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1359 {
1360         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1361         fib6_update_sernum_upto_root(net, f6i);
1362         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1363 }
1364
1365 /*
1366  *      Add routing information to the routing tree.
1367  *      <destination addr>/<source addr>
1368  *      with source addr info in sub-trees
1369  *      Need to own table->tb6_lock
1370  */
1371
1372 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1373              struct nl_info *info, struct netlink_ext_ack *extack)
1374 {
1375         struct fib6_table *table = rt->fib6_table;
1376         struct fib6_node *fn, *pn = NULL;
1377         int err = -ENOMEM;
1378         int allow_create = 1;
1379         int replace_required = 0;
1380
1381         if (info->nlh) {
1382                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1383                         allow_create = 0;
1384                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1385                         replace_required = 1;
1386         }
1387         if (!allow_create && !replace_required)
1388                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1389
1390         fn = fib6_add_1(info->nl_net, table, root,
1391                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1392                         offsetof(struct fib6_info, fib6_dst), allow_create,
1393                         replace_required, extack);
1394         if (IS_ERR(fn)) {
1395                 err = PTR_ERR(fn);
1396                 fn = NULL;
1397                 goto out;
1398         }
1399
1400         pn = fn;
1401
1402 #ifdef CONFIG_IPV6_SUBTREES
1403         if (rt->fib6_src.plen) {
1404                 struct fib6_node *sn;
1405
1406                 if (!rcu_access_pointer(fn->subtree)) {
1407                         struct fib6_node *sfn;
1408
1409                         /*
1410                          * Create subtree.
1411                          *
1412                          *              fn[main tree]
1413                          *              |
1414                          *              sfn[subtree root]
1415                          *                 \
1416                          *                  sn[new leaf node]
1417                          */
1418
1419                         /* Create subtree root node */
1420                         sfn = node_alloc(info->nl_net);
1421                         if (!sfn)
1422                                 goto failure;
1423
1424                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1425                         rcu_assign_pointer(sfn->leaf,
1426                                            info->nl_net->ipv6.fib6_null_entry);
1427                         sfn->fn_flags = RTN_ROOT;
1428
1429                         /* Now add the first leaf node to new subtree */
1430
1431                         sn = fib6_add_1(info->nl_net, table, sfn,
1432                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1433                                         offsetof(struct fib6_info, fib6_src),
1434                                         allow_create, replace_required, extack);
1435
1436                         if (IS_ERR(sn)) {
1437                                 /* If it is failed, discard just allocated
1438                                    root, and then (in failure) stale node
1439                                    in main tree.
1440                                  */
1441                                 node_free_immediate(info->nl_net, sfn);
1442                                 err = PTR_ERR(sn);
1443                                 goto failure;
1444                         }
1445
1446                         /* Now link new subtree to main tree */
1447                         rcu_assign_pointer(sfn->parent, fn);
1448                         rcu_assign_pointer(fn->subtree, sfn);
1449                 } else {
1450                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1451                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1452                                         offsetof(struct fib6_info, fib6_src),
1453                                         allow_create, replace_required, extack);
1454
1455                         if (IS_ERR(sn)) {
1456                                 err = PTR_ERR(sn);
1457                                 goto failure;
1458                         }
1459                 }
1460
1461                 if (!rcu_access_pointer(fn->leaf)) {
1462                         if (fn->fn_flags & RTN_TL_ROOT) {
1463                                 /* put back null_entry for root node */
1464                                 rcu_assign_pointer(fn->leaf,
1465                                             info->nl_net->ipv6.fib6_null_entry);
1466                         } else {
1467                                 fib6_info_hold(rt);
1468                                 rcu_assign_pointer(fn->leaf, rt);
1469                         }
1470                 }
1471                 fn = sn;
1472         }
1473 #endif
1474
1475         err = fib6_add_rt2node(fn, rt, info, extack);
1476         if (!err) {
1477                 if (rt->nh)
1478                         list_add(&rt->nh_list, &rt->nh->f6i_list);
1479                 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1480                 fib6_start_gc(info->nl_net, rt);
1481         }
1482
1483 out:
1484         if (err) {
1485 #ifdef CONFIG_IPV6_SUBTREES
1486                 /*
1487                  * If fib6_add_1 has cleared the old leaf pointer in the
1488                  * super-tree leaf node we have to find a new one for it.
1489                  */
1490                 if (pn != fn) {
1491                         struct fib6_info *pn_leaf =
1492                                 rcu_dereference_protected(pn->leaf,
1493                                     lockdep_is_held(&table->tb6_lock));
1494                         if (pn_leaf == rt) {
1495                                 pn_leaf = NULL;
1496                                 RCU_INIT_POINTER(pn->leaf, NULL);
1497                                 fib6_info_release(rt);
1498                         }
1499                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1500                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1501                                                            pn);
1502 #if RT6_DEBUG >= 2
1503                                 if (!pn_leaf) {
1504                                         WARN_ON(!pn_leaf);
1505                                         pn_leaf =
1506                                             info->nl_net->ipv6.fib6_null_entry;
1507                                 }
1508 #endif
1509                                 fib6_info_hold(pn_leaf);
1510                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1511                         }
1512                 }
1513 #endif
1514                 goto failure;
1515         } else if (fib6_requires_src(rt)) {
1516                 fib6_routes_require_src_inc(info->nl_net);
1517         }
1518         return err;
1519
1520 failure:
1521         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1522          * 1. fn is an intermediate node and we failed to add the new
1523          * route to it in both subtree creation failure and fib6_add_rt2node()
1524          * failure case.
1525          * 2. fn is the root node in the table and we fail to add the first
1526          * default route to it.
1527          */
1528         if (fn &&
1529             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1530              (fn->fn_flags & RTN_TL_ROOT &&
1531               !rcu_access_pointer(fn->leaf))))
1532                 fib6_repair_tree(info->nl_net, table, fn);
1533         return err;
1534 }
1535
1536 /*
1537  *      Routing tree lookup
1538  *
1539  */
1540
1541 struct lookup_args {
1542         int                     offset;         /* key offset on fib6_info */
1543         const struct in6_addr   *addr;          /* search key                   */
1544 };
1545
1546 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1547                                             struct lookup_args *args)
1548 {
1549         struct fib6_node *fn;
1550         __be32 dir;
1551
1552         if (unlikely(args->offset == 0))
1553                 return NULL;
1554
1555         /*
1556          *      Descend on a tree
1557          */
1558
1559         fn = root;
1560
1561         for (;;) {
1562                 struct fib6_node *next;
1563
1564                 dir = addr_bit_set(args->addr, fn->fn_bit);
1565
1566                 next = dir ? rcu_dereference(fn->right) :
1567                              rcu_dereference(fn->left);
1568
1569                 if (next) {
1570                         fn = next;
1571                         continue;
1572                 }
1573                 break;
1574         }
1575
1576         while (fn) {
1577                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1578
1579                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1580                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1581                         struct rt6key *key;
1582
1583                         if (!leaf)
1584                                 goto backtrack;
1585
1586                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1587
1588                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1589 #ifdef CONFIG_IPV6_SUBTREES
1590                                 if (subtree) {
1591                                         struct fib6_node *sfn;
1592                                         sfn = fib6_node_lookup_1(subtree,
1593                                                                  args + 1);
1594                                         if (!sfn)
1595                                                 goto backtrack;
1596                                         fn = sfn;
1597                                 }
1598 #endif
1599                                 if (fn->fn_flags & RTN_RTINFO)
1600                                         return fn;
1601                         }
1602                 }
1603 backtrack:
1604                 if (fn->fn_flags & RTN_ROOT)
1605                         break;
1606
1607                 fn = rcu_dereference(fn->parent);
1608         }
1609
1610         return NULL;
1611 }
1612
1613 /* called with rcu_read_lock() held
1614  */
1615 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1616                                    const struct in6_addr *daddr,
1617                                    const struct in6_addr *saddr)
1618 {
1619         struct fib6_node *fn;
1620         struct lookup_args args[] = {
1621                 {
1622                         .offset = offsetof(struct fib6_info, fib6_dst),
1623                         .addr = daddr,
1624                 },
1625 #ifdef CONFIG_IPV6_SUBTREES
1626                 {
1627                         .offset = offsetof(struct fib6_info, fib6_src),
1628                         .addr = saddr,
1629                 },
1630 #endif
1631                 {
1632                         .offset = 0,    /* sentinel */
1633                 }
1634         };
1635
1636         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1637         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1638                 fn = root;
1639
1640         return fn;
1641 }
1642
1643 /*
1644  *      Get node with specified destination prefix (and source prefix,
1645  *      if subtrees are used)
1646  *      exact_match == true means we try to find fn with exact match of
1647  *      the passed in prefix addr
1648  *      exact_match == false means we try to find fn with longest prefix
1649  *      match of the passed in prefix addr. This is useful for finding fn
1650  *      for cached route as it will be stored in the exception table under
1651  *      the node with longest prefix length.
1652  */
1653
1654
1655 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1656                                        const struct in6_addr *addr,
1657                                        int plen, int offset,
1658                                        bool exact_match)
1659 {
1660         struct fib6_node *fn, *prev = NULL;
1661
1662         for (fn = root; fn ; ) {
1663                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1664                 struct rt6key *key;
1665
1666                 /* This node is being deleted */
1667                 if (!leaf) {
1668                         if (plen <= fn->fn_bit)
1669                                 goto out;
1670                         else
1671                                 goto next;
1672                 }
1673
1674                 key = (struct rt6key *)((u8 *)leaf + offset);
1675
1676                 /*
1677                  *      Prefix match
1678                  */
1679                 if (plen < fn->fn_bit ||
1680                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1681                         goto out;
1682
1683                 if (plen == fn->fn_bit)
1684                         return fn;
1685
1686                 if (fn->fn_flags & RTN_RTINFO)
1687                         prev = fn;
1688
1689 next:
1690                 /*
1691                  *      We have more bits to go
1692                  */
1693                 if (addr_bit_set(addr, fn->fn_bit))
1694                         fn = rcu_dereference(fn->right);
1695                 else
1696                         fn = rcu_dereference(fn->left);
1697         }
1698 out:
1699         if (exact_match)
1700                 return NULL;
1701         else
1702                 return prev;
1703 }
1704
1705 struct fib6_node *fib6_locate(struct fib6_node *root,
1706                               const struct in6_addr *daddr, int dst_len,
1707                               const struct in6_addr *saddr, int src_len,
1708                               bool exact_match)
1709 {
1710         struct fib6_node *fn;
1711
1712         fn = fib6_locate_1(root, daddr, dst_len,
1713                            offsetof(struct fib6_info, fib6_dst),
1714                            exact_match);
1715
1716 #ifdef CONFIG_IPV6_SUBTREES
1717         if (src_len) {
1718                 WARN_ON(saddr == NULL);
1719                 if (fn) {
1720                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1721
1722                         if (subtree) {
1723                                 fn = fib6_locate_1(subtree, saddr, src_len,
1724                                            offsetof(struct fib6_info, fib6_src),
1725                                            exact_match);
1726                         }
1727                 }
1728         }
1729 #endif
1730
1731         if (fn && fn->fn_flags & RTN_RTINFO)
1732                 return fn;
1733
1734         return NULL;
1735 }
1736
1737
1738 /*
1739  *      Deletion
1740  *
1741  */
1742
1743 static struct fib6_info *fib6_find_prefix(struct net *net,
1744                                          struct fib6_table *table,
1745                                          struct fib6_node *fn)
1746 {
1747         struct fib6_node *child_left, *child_right;
1748
1749         if (fn->fn_flags & RTN_ROOT)
1750                 return net->ipv6.fib6_null_entry;
1751
1752         while (fn) {
1753                 child_left = rcu_dereference_protected(fn->left,
1754                                     lockdep_is_held(&table->tb6_lock));
1755                 child_right = rcu_dereference_protected(fn->right,
1756                                     lockdep_is_held(&table->tb6_lock));
1757                 if (child_left)
1758                         return rcu_dereference_protected(child_left->leaf,
1759                                         lockdep_is_held(&table->tb6_lock));
1760                 if (child_right)
1761                         return rcu_dereference_protected(child_right->leaf,
1762                                         lockdep_is_held(&table->tb6_lock));
1763
1764                 fn = FIB6_SUBTREE(fn);
1765         }
1766         return NULL;
1767 }
1768
1769 /*
1770  *      Called to trim the tree of intermediate nodes when possible. "fn"
1771  *      is the node we want to try and remove.
1772  *      Need to own table->tb6_lock
1773  */
1774
1775 static struct fib6_node *fib6_repair_tree(struct net *net,
1776                                           struct fib6_table *table,
1777                                           struct fib6_node *fn)
1778 {
1779         int children;
1780         int nstate;
1781         struct fib6_node *child;
1782         struct fib6_walker *w;
1783         int iter = 0;
1784
1785         /* Set fn->leaf to null_entry for root node. */
1786         if (fn->fn_flags & RTN_TL_ROOT) {
1787                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1788                 return fn;
1789         }
1790
1791         for (;;) {
1792                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1793                                             lockdep_is_held(&table->tb6_lock));
1794                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1795                                             lockdep_is_held(&table->tb6_lock));
1796                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1797                                             lockdep_is_held(&table->tb6_lock));
1798                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1799                                             lockdep_is_held(&table->tb6_lock));
1800                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1801                                             lockdep_is_held(&table->tb6_lock));
1802                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1803                                             lockdep_is_held(&table->tb6_lock));
1804                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1805                                             lockdep_is_held(&table->tb6_lock));
1806                 struct fib6_info *new_fn_leaf;
1807
1808                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1809                 iter++;
1810
1811                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1812                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1813                 WARN_ON(fn_leaf);
1814
1815                 children = 0;
1816                 child = NULL;
1817                 if (fn_r) {
1818                         child = fn_r;
1819                         children |= 1;
1820                 }
1821                 if (fn_l) {
1822                         child = fn_l;
1823                         children |= 2;
1824                 }
1825
1826                 if (children == 3 || FIB6_SUBTREE(fn)
1827 #ifdef CONFIG_IPV6_SUBTREES
1828                     /* Subtree root (i.e. fn) may have one child */
1829                     || (children && fn->fn_flags & RTN_ROOT)
1830 #endif
1831                     ) {
1832                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1833 #if RT6_DEBUG >= 2
1834                         if (!new_fn_leaf) {
1835                                 WARN_ON(!new_fn_leaf);
1836                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1837                         }
1838 #endif
1839                         fib6_info_hold(new_fn_leaf);
1840                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1841                         return pn;
1842                 }
1843
1844 #ifdef CONFIG_IPV6_SUBTREES
1845                 if (FIB6_SUBTREE(pn) == fn) {
1846                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1847                         RCU_INIT_POINTER(pn->subtree, NULL);
1848                         nstate = FWS_L;
1849                 } else {
1850                         WARN_ON(fn->fn_flags & RTN_ROOT);
1851 #endif
1852                         if (pn_r == fn)
1853                                 rcu_assign_pointer(pn->right, child);
1854                         else if (pn_l == fn)
1855                                 rcu_assign_pointer(pn->left, child);
1856 #if RT6_DEBUG >= 2
1857                         else
1858                                 WARN_ON(1);
1859 #endif
1860                         if (child)
1861                                 rcu_assign_pointer(child->parent, pn);
1862                         nstate = FWS_R;
1863 #ifdef CONFIG_IPV6_SUBTREES
1864                 }
1865 #endif
1866
1867                 read_lock(&net->ipv6.fib6_walker_lock);
1868                 FOR_WALKERS(net, w) {
1869                         if (!child) {
1870                                 if (w->node == fn) {
1871                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1872                                         w->node = pn;
1873                                         w->state = nstate;
1874                                 }
1875                         } else {
1876                                 if (w->node == fn) {
1877                                         w->node = child;
1878                                         if (children&2) {
1879                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1880                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1881                                         } else {
1882                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1883                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1884                                         }
1885                                 }
1886                         }
1887                 }
1888                 read_unlock(&net->ipv6.fib6_walker_lock);
1889
1890                 node_free(net, fn);
1891                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1892                         return pn;
1893
1894                 RCU_INIT_POINTER(pn->leaf, NULL);
1895                 fib6_info_release(pn_leaf);
1896                 fn = pn;
1897         }
1898 }
1899
1900 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1901                            struct fib6_info __rcu **rtp, struct nl_info *info)
1902 {
1903         struct fib6_info *leaf, *replace_rt = NULL;
1904         struct fib6_walker *w;
1905         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1906                                     lockdep_is_held(&table->tb6_lock));
1907         struct net *net = info->nl_net;
1908         bool notify_del = false;
1909
1910         RT6_TRACE("fib6_del_route\n");
1911
1912         /* If the deleted route is the first in the node and it is not part of
1913          * a multipath route, then we need to replace it with the next route
1914          * in the node, if exists.
1915          */
1916         leaf = rcu_dereference_protected(fn->leaf,
1917                                          lockdep_is_held(&table->tb6_lock));
1918         if (leaf == rt && !rt->fib6_nsiblings) {
1919                 if (rcu_access_pointer(rt->fib6_next))
1920                         replace_rt = rcu_dereference_protected(rt->fib6_next,
1921                                             lockdep_is_held(&table->tb6_lock));
1922                 else
1923                         notify_del = true;
1924         }
1925
1926         /* Unlink it */
1927         *rtp = rt->fib6_next;
1928         rt->fib6_node = NULL;
1929         net->ipv6.rt6_stats->fib_rt_entries--;
1930         net->ipv6.rt6_stats->fib_discarded_routes++;
1931
1932         /* Reset round-robin state, if necessary */
1933         if (rcu_access_pointer(fn->rr_ptr) == rt)
1934                 fn->rr_ptr = NULL;
1935
1936         /* Remove this entry from other siblings */
1937         if (rt->fib6_nsiblings) {
1938                 struct fib6_info *sibling, *next_sibling;
1939
1940                 /* The route is deleted from a multipath route. If this
1941                  * multipath route is the first route in the node, then we need
1942                  * to emit a delete notification. Otherwise, we need to skip
1943                  * the notification.
1944                  */
1945                 if (rt->fib6_metric == leaf->fib6_metric &&
1946                     rt6_qualify_for_ecmp(leaf))
1947                         notify_del = true;
1948                 list_for_each_entry_safe(sibling, next_sibling,
1949                                          &rt->fib6_siblings, fib6_siblings)
1950                         sibling->fib6_nsiblings--;
1951                 rt->fib6_nsiblings = 0;
1952                 list_del_init(&rt->fib6_siblings);
1953                 rt6_multipath_rebalance(next_sibling);
1954         }
1955
1956         /* Adjust walkers */
1957         read_lock(&net->ipv6.fib6_walker_lock);
1958         FOR_WALKERS(net, w) {
1959                 if (w->state == FWS_C && w->leaf == rt) {
1960                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1961                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1962                                             lockdep_is_held(&table->tb6_lock));
1963                         if (!w->leaf)
1964                                 w->state = FWS_U;
1965                 }
1966         }
1967         read_unlock(&net->ipv6.fib6_walker_lock);
1968
1969         /* If it was last route, call fib6_repair_tree() to:
1970          * 1. For root node, put back null_entry as how the table was created.
1971          * 2. For other nodes, expunge its radix tree node.
1972          */
1973         if (!rcu_access_pointer(fn->leaf)) {
1974                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1975                         fn->fn_flags &= ~RTN_RTINFO;
1976                         net->ipv6.rt6_stats->fib_route_nodes--;
1977                 }
1978                 fn = fib6_repair_tree(net, table, fn);
1979         }
1980
1981         fib6_purge_rt(rt, fn, net);
1982
1983         if (!info->skip_notify_kernel) {
1984                 if (notify_del)
1985                         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1986                                                   rt, NULL);
1987                 else if (replace_rt)
1988                         call_fib6_entry_notifiers_replace(net, replace_rt);
1989         }
1990         if (!info->skip_notify)
1991                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1992
1993         fib6_info_release(rt);
1994 }
1995
1996 /* Need to own table->tb6_lock */
1997 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1998 {
1999         struct net *net = info->nl_net;
2000         struct fib6_info __rcu **rtp;
2001         struct fib6_info __rcu **rtp_next;
2002         struct fib6_table *table;
2003         struct fib6_node *fn;
2004
2005         if (rt == net->ipv6.fib6_null_entry)
2006                 return -ENOENT;
2007
2008         table = rt->fib6_table;
2009         fn = rcu_dereference_protected(rt->fib6_node,
2010                                        lockdep_is_held(&table->tb6_lock));
2011         if (!fn)
2012                 return -ENOENT;
2013
2014         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2015
2016         /*
2017          *      Walk the leaf entries looking for ourself
2018          */
2019
2020         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2021                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2022                                         lockdep_is_held(&table->tb6_lock));
2023                 if (rt == cur) {
2024                         if (fib6_requires_src(cur))
2025                                 fib6_routes_require_src_dec(info->nl_net);
2026                         fib6_del_route(table, fn, rtp, info);
2027                         return 0;
2028                 }
2029                 rtp_next = &cur->fib6_next;
2030         }
2031         return -ENOENT;
2032 }
2033
2034 /*
2035  *      Tree traversal function.
2036  *
2037  *      Certainly, it is not interrupt safe.
2038  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2039  *      It means, that we can modify tree during walking
2040  *      and use this function for garbage collection, clone pruning,
2041  *      cleaning tree when a device goes down etc. etc.
2042  *
2043  *      It guarantees that every node will be traversed,
2044  *      and that it will be traversed only once.
2045  *
2046  *      Callback function w->func may return:
2047  *      0 -> continue walking.
2048  *      positive value -> walking is suspended (used by tree dumps,
2049  *      and probably by gc, if it will be split to several slices)
2050  *      negative value -> terminate walking.
2051  *
2052  *      The function itself returns:
2053  *      0   -> walk is complete.
2054  *      >0  -> walk is incomplete (i.e. suspended)
2055  *      <0  -> walk is terminated by an error.
2056  *
2057  *      This function is called with tb6_lock held.
2058  */
2059
2060 static int fib6_walk_continue(struct fib6_walker *w)
2061 {
2062         struct fib6_node *fn, *pn, *left, *right;
2063
2064         /* w->root should always be table->tb6_root */
2065         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2066
2067         for (;;) {
2068                 fn = w->node;
2069                 if (!fn)
2070                         return 0;
2071
2072                 switch (w->state) {
2073 #ifdef CONFIG_IPV6_SUBTREES
2074                 case FWS_S:
2075                         if (FIB6_SUBTREE(fn)) {
2076                                 w->node = FIB6_SUBTREE(fn);
2077                                 continue;
2078                         }
2079                         w->state = FWS_L;
2080                         fallthrough;
2081 #endif
2082                 case FWS_L:
2083                         left = rcu_dereference_protected(fn->left, 1);
2084                         if (left) {
2085                                 w->node = left;
2086                                 w->state = FWS_INIT;
2087                                 continue;
2088                         }
2089                         w->state = FWS_R;
2090                         fallthrough;
2091                 case FWS_R:
2092                         right = rcu_dereference_protected(fn->right, 1);
2093                         if (right) {
2094                                 w->node = right;
2095                                 w->state = FWS_INIT;
2096                                 continue;
2097                         }
2098                         w->state = FWS_C;
2099                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
2100                         fallthrough;
2101                 case FWS_C:
2102                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2103                                 int err;
2104
2105                                 if (w->skip) {
2106                                         w->skip--;
2107                                         goto skip;
2108                                 }
2109
2110                                 err = w->func(w);
2111                                 if (err)
2112                                         return err;
2113
2114                                 w->count++;
2115                                 continue;
2116                         }
2117 skip:
2118                         w->state = FWS_U;
2119                         fallthrough;
2120                 case FWS_U:
2121                         if (fn == w->root)
2122                                 return 0;
2123                         pn = rcu_dereference_protected(fn->parent, 1);
2124                         left = rcu_dereference_protected(pn->left, 1);
2125                         right = rcu_dereference_protected(pn->right, 1);
2126                         w->node = pn;
2127 #ifdef CONFIG_IPV6_SUBTREES
2128                         if (FIB6_SUBTREE(pn) == fn) {
2129                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2130                                 w->state = FWS_L;
2131                                 continue;
2132                         }
2133 #endif
2134                         if (left == fn) {
2135                                 w->state = FWS_R;
2136                                 continue;
2137                         }
2138                         if (right == fn) {
2139                                 w->state = FWS_C;
2140                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2141                                 continue;
2142                         }
2143 #if RT6_DEBUG >= 2
2144                         WARN_ON(1);
2145 #endif
2146                 }
2147         }
2148 }
2149
2150 static int fib6_walk(struct net *net, struct fib6_walker *w)
2151 {
2152         int res;
2153
2154         w->state = FWS_INIT;
2155         w->node = w->root;
2156
2157         fib6_walker_link(net, w);
2158         res = fib6_walk_continue(w);
2159         if (res <= 0)
2160                 fib6_walker_unlink(net, w);
2161         return res;
2162 }
2163
2164 static int fib6_clean_node(struct fib6_walker *w)
2165 {
2166         int res;
2167         struct fib6_info *rt;
2168         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2169         struct nl_info info = {
2170                 .nl_net = c->net,
2171                 .skip_notify = c->skip_notify,
2172         };
2173
2174         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2175             READ_ONCE(w->node->fn_sernum) != c->sernum)
2176                 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2177
2178         if (!c->func) {
2179                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2180                 w->leaf = NULL;
2181                 return 0;
2182         }
2183
2184         for_each_fib6_walker_rt(w) {
2185                 res = c->func(rt, c->arg);
2186                 if (res == -1) {
2187                         w->leaf = rt;
2188                         res = fib6_del(rt, &info);
2189                         if (res) {
2190 #if RT6_DEBUG >= 2
2191                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2192                                          __func__, rt,
2193                                          rcu_access_pointer(rt->fib6_node),
2194                                          res);
2195 #endif
2196                                 continue;
2197                         }
2198                         return 0;
2199                 } else if (res == -2) {
2200                         if (WARN_ON(!rt->fib6_nsiblings))
2201                                 continue;
2202                         rt = list_last_entry(&rt->fib6_siblings,
2203                                              struct fib6_info, fib6_siblings);
2204                         continue;
2205                 }
2206                 WARN_ON(res != 0);
2207         }
2208         w->leaf = rt;
2209         return 0;
2210 }
2211
2212 /*
2213  *      Convenient frontend to tree walker.
2214  *
2215  *      func is called on each route.
2216  *              It may return -2 -> skip multipath route.
2217  *                            -1 -> delete this route.
2218  *                            0  -> continue walking
2219  */
2220
2221 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2222                             int (*func)(struct fib6_info *, void *arg),
2223                             int sernum, void *arg, bool skip_notify)
2224 {
2225         struct fib6_cleaner c;
2226
2227         c.w.root = root;
2228         c.w.func = fib6_clean_node;
2229         c.w.count = 0;
2230         c.w.skip = 0;
2231         c.w.skip_in_node = 0;
2232         c.func = func;
2233         c.sernum = sernum;
2234         c.arg = arg;
2235         c.net = net;
2236         c.skip_notify = skip_notify;
2237
2238         fib6_walk(net, &c.w);
2239 }
2240
2241 static void __fib6_clean_all(struct net *net,
2242                              int (*func)(struct fib6_info *, void *),
2243                              int sernum, void *arg, bool skip_notify)
2244 {
2245         struct fib6_table *table;
2246         struct hlist_head *head;
2247         unsigned int h;
2248
2249         rcu_read_lock();
2250         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2251                 head = &net->ipv6.fib_table_hash[h];
2252                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2253                         spin_lock_bh(&table->tb6_lock);
2254                         fib6_clean_tree(net, &table->tb6_root,
2255                                         func, sernum, arg, skip_notify);
2256                         spin_unlock_bh(&table->tb6_lock);
2257                 }
2258         }
2259         rcu_read_unlock();
2260 }
2261
2262 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2263                     void *arg)
2264 {
2265         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2266 }
2267
2268 void fib6_clean_all_skip_notify(struct net *net,
2269                                 int (*func)(struct fib6_info *, void *),
2270                                 void *arg)
2271 {
2272         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2273 }
2274
2275 static void fib6_flush_trees(struct net *net)
2276 {
2277         int new_sernum = fib6_new_sernum(net);
2278
2279         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2280 }
2281
2282 /*
2283  *      Garbage collection
2284  */
2285
2286 static int fib6_age(struct fib6_info *rt, void *arg)
2287 {
2288         struct fib6_gc_args *gc_args = arg;
2289         unsigned long now = jiffies;
2290
2291         /*
2292          *      check addrconf expiration here.
2293          *      Routes are expired even if they are in use.
2294          */
2295
2296         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2297                 if (time_after(now, rt->expires)) {
2298                         RT6_TRACE("expiring %p\n", rt);
2299                         return -1;
2300                 }
2301                 gc_args->more++;
2302         }
2303
2304         /*      Also age clones in the exception table.
2305          *      Note, that clones are aged out
2306          *      only if they are not in use now.
2307          */
2308         rt6_age_exceptions(rt, gc_args, now);
2309
2310         return 0;
2311 }
2312
2313 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2314 {
2315         struct fib6_gc_args gc_args;
2316         unsigned long now;
2317
2318         if (force) {
2319                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2320         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2321                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2322                 return;
2323         }
2324         gc_args.timeout = expires ? (int)expires :
2325                           net->ipv6.sysctl.ip6_rt_gc_interval;
2326         gc_args.more = 0;
2327
2328         fib6_clean_all(net, fib6_age, &gc_args);
2329         now = jiffies;
2330         net->ipv6.ip6_rt_last_gc = now;
2331
2332         if (gc_args.more)
2333                 mod_timer(&net->ipv6.ip6_fib_timer,
2334                           round_jiffies(now
2335                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2336         else
2337                 del_timer(&net->ipv6.ip6_fib_timer);
2338         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2339 }
2340
2341 static void fib6_gc_timer_cb(struct timer_list *t)
2342 {
2343         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2344
2345         fib6_run_gc(0, arg, true);
2346 }
2347
2348 static int __net_init fib6_net_init(struct net *net)
2349 {
2350         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2351         int err;
2352
2353         err = fib6_notifier_init(net);
2354         if (err)
2355                 return err;
2356
2357         spin_lock_init(&net->ipv6.fib6_gc_lock);
2358         rwlock_init(&net->ipv6.fib6_walker_lock);
2359         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2360         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2361
2362         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2363         if (!net->ipv6.rt6_stats)
2364                 goto out_timer;
2365
2366         /* Avoid false sharing : Use at least a full cache line */
2367         size = max_t(size_t, size, L1_CACHE_BYTES);
2368
2369         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2370         if (!net->ipv6.fib_table_hash)
2371                 goto out_rt6_stats;
2372
2373         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2374                                           GFP_KERNEL);
2375         if (!net->ipv6.fib6_main_tbl)
2376                 goto out_fib_table_hash;
2377
2378         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2379         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2380                            net->ipv6.fib6_null_entry);
2381         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2382                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2383         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2384
2385 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2386         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2387                                            GFP_KERNEL);
2388         if (!net->ipv6.fib6_local_tbl)
2389                 goto out_fib6_main_tbl;
2390         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2391         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2392                            net->ipv6.fib6_null_entry);
2393         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2394                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2395         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2396 #endif
2397         fib6_tables_init(net);
2398
2399         return 0;
2400
2401 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2402 out_fib6_main_tbl:
2403         kfree(net->ipv6.fib6_main_tbl);
2404 #endif
2405 out_fib_table_hash:
2406         kfree(net->ipv6.fib_table_hash);
2407 out_rt6_stats:
2408         kfree(net->ipv6.rt6_stats);
2409 out_timer:
2410         fib6_notifier_exit(net);
2411         return -ENOMEM;
2412 }
2413
2414 static void fib6_net_exit(struct net *net)
2415 {
2416         unsigned int i;
2417
2418         del_timer_sync(&net->ipv6.ip6_fib_timer);
2419
2420         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2421                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2422                 struct hlist_node *tmp;
2423                 struct fib6_table *tb;
2424
2425                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2426                         hlist_del(&tb->tb6_hlist);
2427                         fib6_free_table(tb);
2428                 }
2429         }
2430
2431         kfree(net->ipv6.fib_table_hash);
2432         kfree(net->ipv6.rt6_stats);
2433         fib6_notifier_exit(net);
2434 }
2435
2436 static struct pernet_operations fib6_net_ops = {
2437         .init = fib6_net_init,
2438         .exit = fib6_net_exit,
2439 };
2440
2441 int __init fib6_init(void)
2442 {
2443         int ret = -ENOMEM;
2444
2445         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2446                                            sizeof(struct fib6_node),
2447                                            0, SLAB_HWCACHE_ALIGN,
2448                                            NULL);
2449         if (!fib6_node_kmem)
2450                 goto out;
2451
2452         ret = register_pernet_subsys(&fib6_net_ops);
2453         if (ret)
2454                 goto out_kmem_cache_create;
2455
2456         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2457                                    inet6_dump_fib, 0);
2458         if (ret)
2459                 goto out_unregister_subsys;
2460
2461         __fib6_flush_trees = fib6_flush_trees;
2462 out:
2463         return ret;
2464
2465 out_unregister_subsys:
2466         unregister_pernet_subsys(&fib6_net_ops);
2467 out_kmem_cache_create:
2468         kmem_cache_destroy(fib6_node_kmem);
2469         goto out;
2470 }
2471
2472 void fib6_gc_cleanup(void)
2473 {
2474         unregister_pernet_subsys(&fib6_net_ops);
2475         kmem_cache_destroy(fib6_node_kmem);
2476 }
2477
2478 #ifdef CONFIG_PROC_FS
2479 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2480 {
2481         struct fib6_info *rt = v;
2482         struct ipv6_route_iter *iter = seq->private;
2483         struct fib6_nh *fib6_nh = rt->fib6_nh;
2484         unsigned int flags = rt->fib6_flags;
2485         const struct net_device *dev;
2486
2487         if (rt->nh)
2488                 fib6_nh = nexthop_fib6_nh_bh(rt->nh);
2489
2490         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2491
2492 #ifdef CONFIG_IPV6_SUBTREES
2493         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2494 #else
2495         seq_puts(seq, "00000000000000000000000000000000 00 ");
2496 #endif
2497         if (fib6_nh->fib_nh_gw_family) {
2498                 flags |= RTF_GATEWAY;
2499                 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2500         } else {
2501                 seq_puts(seq, "00000000000000000000000000000000");
2502         }
2503
2504         dev = fib6_nh->fib_nh_dev;
2505         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2506                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2507                    flags, dev ? dev->name : "");
2508         iter->w.leaf = NULL;
2509         return 0;
2510 }
2511
2512 static int ipv6_route_yield(struct fib6_walker *w)
2513 {
2514         struct ipv6_route_iter *iter = w->args;
2515
2516         if (!iter->skip)
2517                 return 1;
2518
2519         do {
2520                 iter->w.leaf = rcu_dereference_protected(
2521                                 iter->w.leaf->fib6_next,
2522                                 lockdep_is_held(&iter->tbl->tb6_lock));
2523                 iter->skip--;
2524                 if (!iter->skip && iter->w.leaf)
2525                         return 1;
2526         } while (iter->w.leaf);
2527
2528         return 0;
2529 }
2530
2531 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2532                                       struct net *net)
2533 {
2534         memset(&iter->w, 0, sizeof(iter->w));
2535         iter->w.func = ipv6_route_yield;
2536         iter->w.root = &iter->tbl->tb6_root;
2537         iter->w.state = FWS_INIT;
2538         iter->w.node = iter->w.root;
2539         iter->w.args = iter;
2540         iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2541         INIT_LIST_HEAD(&iter->w.lh);
2542         fib6_walker_link(net, &iter->w);
2543 }
2544
2545 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2546                                                     struct net *net)
2547 {
2548         unsigned int h;
2549         struct hlist_node *node;
2550
2551         if (tbl) {
2552                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2553                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2554         } else {
2555                 h = 0;
2556                 node = NULL;
2557         }
2558
2559         while (!node && h < FIB6_TABLE_HASHSZ) {
2560                 node = rcu_dereference_bh(
2561                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2562         }
2563         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2564 }
2565
2566 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2567 {
2568         int sernum = READ_ONCE(iter->w.root->fn_sernum);
2569
2570         if (iter->sernum != sernum) {
2571                 iter->sernum = sernum;
2572                 iter->w.state = FWS_INIT;
2573                 iter->w.node = iter->w.root;
2574                 WARN_ON(iter->w.skip);
2575                 iter->w.skip = iter->w.count;
2576         }
2577 }
2578
2579 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2580 {
2581         int r;
2582         struct fib6_info *n;
2583         struct net *net = seq_file_net(seq);
2584         struct ipv6_route_iter *iter = seq->private;
2585
2586         ++(*pos);
2587         if (!v)
2588                 goto iter_table;
2589
2590         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2591         if (n)
2592                 return n;
2593
2594 iter_table:
2595         ipv6_route_check_sernum(iter);
2596         spin_lock_bh(&iter->tbl->tb6_lock);
2597         r = fib6_walk_continue(&iter->w);
2598         spin_unlock_bh(&iter->tbl->tb6_lock);
2599         if (r > 0) {
2600                 return iter->w.leaf;
2601         } else if (r < 0) {
2602                 fib6_walker_unlink(net, &iter->w);
2603                 return NULL;
2604         }
2605         fib6_walker_unlink(net, &iter->w);
2606
2607         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2608         if (!iter->tbl)
2609                 return NULL;
2610
2611         ipv6_route_seq_setup_walk(iter, net);
2612         goto iter_table;
2613 }
2614
2615 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2616         __acquires(RCU_BH)
2617 {
2618         struct net *net = seq_file_net(seq);
2619         struct ipv6_route_iter *iter = seq->private;
2620
2621         rcu_read_lock_bh();
2622         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2623         iter->skip = *pos;
2624
2625         if (iter->tbl) {
2626                 loff_t p = 0;
2627
2628                 ipv6_route_seq_setup_walk(iter, net);
2629                 return ipv6_route_seq_next(seq, NULL, &p);
2630         } else {
2631                 return NULL;
2632         }
2633 }
2634
2635 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2636 {
2637         struct fib6_walker *w = &iter->w;
2638         return w->node && !(w->state == FWS_U && w->node == w->root);
2639 }
2640
2641 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2642         __releases(RCU_BH)
2643 {
2644         struct net *net = seq_file_net(seq);
2645         struct ipv6_route_iter *iter = seq->private;
2646
2647         if (ipv6_route_iter_active(iter))
2648                 fib6_walker_unlink(net, &iter->w);
2649
2650         rcu_read_unlock_bh();
2651 }
2652
2653 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2654 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2655                                     struct bpf_iter_meta *meta,
2656                                     void *v)
2657 {
2658         struct bpf_iter__ipv6_route ctx;
2659
2660         ctx.meta = meta;
2661         ctx.rt = v;
2662         return bpf_iter_run_prog(prog, &ctx);
2663 }
2664
2665 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2666 {
2667         struct ipv6_route_iter *iter = seq->private;
2668         struct bpf_iter_meta meta;
2669         struct bpf_prog *prog;
2670         int ret;
2671
2672         meta.seq = seq;
2673         prog = bpf_iter_get_info(&meta, false);
2674         if (!prog)
2675                 return ipv6_route_native_seq_show(seq, v);
2676
2677         ret = ipv6_route_prog_seq_show(prog, &meta, v);
2678         iter->w.leaf = NULL;
2679
2680         return ret;
2681 }
2682
2683 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2684 {
2685         struct bpf_iter_meta meta;
2686         struct bpf_prog *prog;
2687
2688         if (!v) {
2689                 meta.seq = seq;
2690                 prog = bpf_iter_get_info(&meta, true);
2691                 if (prog)
2692                         (void)ipv6_route_prog_seq_show(prog, &meta, v);
2693         }
2694
2695         ipv6_route_native_seq_stop(seq, v);
2696 }
2697 #else
2698 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2699 {
2700         return ipv6_route_native_seq_show(seq, v);
2701 }
2702
2703 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2704 {
2705         ipv6_route_native_seq_stop(seq, v);
2706 }
2707 #endif
2708
2709 const struct seq_operations ipv6_route_seq_ops = {
2710         .start  = ipv6_route_seq_start,
2711         .next   = ipv6_route_seq_next,
2712         .stop   = ipv6_route_seq_stop,
2713         .show   = ipv6_route_seq_show
2714 };
2715 #endif /* CONFIG_PROC_FS */