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/bpf.h>
  19 #include <linux/errno.h>
  20 #include <linux/types.h>
  21 #include <linux/net.h>
  22 #include <linux/route.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/in6.h>
  25 #include <linux/init.h>
  26 #include <linux/list.h>
  27 #include <linux/slab.h>
  28
  29 #include <net/ip.h>
  30 #include <net/ipv6.h>
  31 #include <net/ndisc.h>
  32 #include <net/addrconf.h>
  33 #include <net/lwtunnel.h>
  34 #include <net/fib_notifier.h>
  35
  36 #include <net/ip_fib.h>
  37 #include <net/ip6_fib.h>
  38 #include <net/ip6_route.h>
  39
  40 static struct kmem_cache *fib6_node_kmem __read_mostly;
  41
  42 struct fib6_cleaner {
  43         struct fib6_walker w;
  44         struct net *net;
  45         int (*func)(struct fib6_info *, void *arg);
  46         int sernum;
  47         void *arg;
  48         bool skip_notify;
  49 };
  50
  51 #ifdef CONFIG_IPV6_SUBTREES
  52 #define FWS_INIT FWS_S
  53 #else
  54 #define FWS_INIT FWS_L
  55 #endif
  56
  57 static struct fib6_info *fib6_find_prefix(struct net *net,
  58                                          struct fib6_table *table,
  59                                          struct fib6_node *fn);
  60 static struct fib6_node *fib6_repair_tree(struct net *net,
  61                                           struct fib6_table *table,
  62                                           struct fib6_node *fn);
  63 static int fib6_walk(struct net *net, struct fib6_walker *w);
  64 static int fib6_walk_continue(struct fib6_walker *w);
  65
  66 /*
  67  *      A routing update causes an increase of the serial number on the
  68  *      affected subtree. This allows for cached routes to be asynchronously
  69  *      tested when modifications are made to the destination cache as a
  70  *      result of redirects, path MTU changes, etc.
  71  */
  72
  73 static void fib6_gc_timer_cb(struct timer_list *t);
  74
  75 #define FOR_WALKERS(net, w) \
  76         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  77
  78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  79 {
  80         write_lock_bh(&net->ipv6.fib6_walker_lock);
  81         list_add(&w->lh, &net->ipv6.fib6_walkers);
  82         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  83 }
  84
  85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  86 {
  87         write_lock_bh(&net->ipv6.fib6_walker_lock);
  88         list_del(&w->lh);
  89         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  90 }
  91
  92 static int fib6_new_sernum(struct net *net)
  93 {
  94         int new, old = atomic_read(&net->ipv6.fib6_sernum);
  95
  96         do {
  97                 new = old < INT_MAX ? old + 1 : 1;
  98         } while (!atomic_try_cmpxchg(&net->ipv6.fib6_sernum, &old, new));
  99
 100         return new;
 101 }
 102
 103 enum {
 104         FIB6_NO_SERNUM_CHANGE = 0,
 105 };
 106
 107 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 108 {
 109         struct fib6_node *fn;
 110
 111         fn = rcu_dereference_protected(f6i->fib6_node,
 112                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 113         if (fn)
 114                 WRITE_ONCE(fn->fn_sernum, fib6_new_sernum(net));
 115 }
 116
 117 /*
 118  *      Auxiliary address test functions for the radix tree.
 119  *
 120  *      These assume a 32bit processor (although it will work on
 121  *      64bit processors)
 122  */
 123
 124 /*
 125  *      test bit
 126  */
 127 #if defined(__LITTLE_ENDIAN)
 128 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 129 #else
 130 # define BITOP_BE32_SWIZZLE     0
 131 #endif
 132
 133 static __be32 addr_bit_set(const void *token, int fn_bit)
 134 {
 135         const __be32 *addr = token;
 136         /*
 137          * Here,
 138          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 139          * is optimized version of
 140          *      htonl(1 << ((~fn_bit)&0x1F))
 141          * See include/asm-generic/bitops/le.h.
 142          */
 143         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 144                addr[fn_bit >> 5];
 145 }
 146
 147 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh)
 148 {
 149         struct fib6_info *f6i;
 150         size_t sz = sizeof(*f6i);
 151
 152         if (with_fib6_nh)
 153                 sz += sizeof(struct fib6_nh);
 154
 155         f6i = kzalloc(sz, gfp_flags);
 156         if (!f6i)
 157                 return NULL;
 158
 159         /* fib6_siblings is a union with nh_list, so this initializes both */
 160         INIT_LIST_HEAD(&f6i->fib6_siblings);
 161         refcount_set(&f6i->fib6_ref, 1);
 162
 163         INIT_HLIST_NODE(&f6i->gc_link);
 164
 165         return f6i;
 166 }
 167
 168 void fib6_info_destroy_rcu(struct rcu_head *head)
 169 {
 170         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 171
 172         WARN_ON(f6i->fib6_node);
 173
 174         if (f6i->nh)
 175                 nexthop_put(f6i->nh);
 176         else
 177                 fib6_nh_release(f6i->fib6_nh);
 178
 179         ip_fib_metrics_put(f6i->fib6_metrics);
 180         kfree(f6i);
 181 }
 182 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 183
 184 static struct fib6_node *node_alloc(struct net *net)
 185 {
 186         struct fib6_node *fn;
 187
 188         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 189         if (fn)
 190                 net->ipv6.rt6_stats->fib_nodes++;
 191
 192         return fn;
 193 }
 194
 195 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 196 {
 197         kmem_cache_free(fib6_node_kmem, fn);
 198         net->ipv6.rt6_stats->fib_nodes--;
 199 }
 200
 201 static void node_free_rcu(struct rcu_head *head)
 202 {
 203         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 204
 205         kmem_cache_free(fib6_node_kmem, fn);
 206 }
 207
 208 static void node_free(struct net *net, struct fib6_node *fn)
 209 {
 210         call_rcu(&fn->rcu, node_free_rcu);
 211         net->ipv6.rt6_stats->fib_nodes--;
 212 }
 213
 214 static void fib6_free_table(struct fib6_table *table)
 215 {
 216         inetpeer_invalidate_tree(&table->tb6_peers);
 217         kfree(table);
 218 }
 219
 220 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 221 {
 222         unsigned int h;
 223
 224         /*
 225          * Initialize table lock at a single place to give lockdep a key,
 226          * tables aren't visible prior to being linked to the list.
 227          */
 228         spin_lock_init(&tb->tb6_lock);
 229         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 230
 231         /*
 232          * No protection necessary, this is the only list mutatation
 233          * operation, tables never disappear once they exist.
 234          */
 235         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 236 }
 237
 238 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 239
 240 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 241 {
 242         struct fib6_table *table;
 243
 244         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 245         if (table) {
 246                 table->tb6_id = id;
 247                 rcu_assign_pointer(table->tb6_root.leaf,
 248                                    net->ipv6.fib6_null_entry);
 249                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 250                 inet_peer_base_init(&table->tb6_peers);
 251                 INIT_HLIST_HEAD(&table->tb6_gc_hlist);
 252         }
 253
 254         return table;
 255 }
 256
 257 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 258 {
 259         struct fib6_table *tb;
 260
 261         if (id == 0)
 262                 id = RT6_TABLE_MAIN;
 263         tb = fib6_get_table(net, id);
 264         if (tb)
 265                 return tb;
 266
 267         tb = fib6_alloc_table(net, id);
 268         if (tb)
 269                 fib6_link_table(net, tb);
 270
 271         return tb;
 272 }
 273 EXPORT_SYMBOL_GPL(fib6_new_table);
 274
 275 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 276 {
 277         struct fib6_table *tb;
 278         struct hlist_head *head;
 279         unsigned int h;
 280
 281         if (id == 0)
 282                 id = RT6_TABLE_MAIN;
 283         h = id & (FIB6_TABLE_HASHSZ - 1);
 284         rcu_read_lock();
 285         head = &net->ipv6.fib_table_hash[h];
 286         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 287                 if (tb->tb6_id == id) {
 288                         rcu_read_unlock();
 289                         return tb;
 290                 }
 291         }
 292         rcu_read_unlock();
 293
 294         return NULL;
 295 }
 296 EXPORT_SYMBOL_GPL(fib6_get_table);
 297
 298 static void __net_init fib6_tables_init(struct net *net)
 299 {
 300         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 301         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 302 }
 303 #else
 304
 305 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 306 {
 307         return fib6_get_table(net, id);
 308 }
 309
 310 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 311 {
 312           return net->ipv6.fib6_main_tbl;
 313 }
 314
 315 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 316                                    const struct sk_buff *skb,
 317                                    int flags, pol_lookup_t lookup)
 318 {
 319         struct rt6_info *rt;
 320
 321         rt = pol_lookup_func(lookup,
 322                         net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 323         if (rt->dst.error == -EAGAIN) {
 324                 ip6_rt_put_flags(rt, flags);
 325                 rt = net->ipv6.ip6_null_entry;
 326                 if (!(flags & RT6_LOOKUP_F_DST_NOREF))
 327                         dst_hold(&rt->dst);
 328         }
 329
 330         return &rt->dst;
 331 }
 332
 333 /* called with rcu lock held; no reference taken on fib6_info */
 334 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 335                 struct fib6_result *res, int flags)
 336 {
 337         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
 338                                  res, flags);
 339 }
 340
 341 static void __net_init fib6_tables_init(struct net *net)
 342 {
 343         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 344 }
 345
 346 #endif
 347
 348 unsigned int fib6_tables_seq_read(struct net *net)
 349 {
 350         unsigned int h, fib_seq = 0;
 351
 352         rcu_read_lock();
 353         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 354                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 355                 struct fib6_table *tb;
 356
 357                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 358                         fib_seq += tb->fib_seq;
 359         }
 360         rcu_read_unlock();
 361
 362         return fib_seq;
 363 }
 364
 365 static int call_fib6_entry_notifier(struct notifier_block *nb,
 366                                     enum fib_event_type event_type,
 367                                     struct fib6_info *rt,
 368                                     struct netlink_ext_ack *extack)
 369 {
 370         struct fib6_entry_notifier_info info = {
 371                 .info.extack = extack,
 372                 .rt = rt,
 373         };
 374
 375         return call_fib6_notifier(nb, event_type, &info.info);
 376 }
 377
 378 static int call_fib6_multipath_entry_notifier(struct notifier_block *nb,
 379                                               enum fib_event_type event_type,
 380                                               struct fib6_info *rt,
 381                                               unsigned int nsiblings,
 382                                               struct netlink_ext_ack *extack)
 383 {
 384         struct fib6_entry_notifier_info info = {
 385                 .info.extack = extack,
 386                 .rt = rt,
 387                 .nsiblings = nsiblings,
 388         };
 389
 390         return call_fib6_notifier(nb, event_type, &info.info);
 391 }
 392
 393 int call_fib6_entry_notifiers(struct net *net,
 394                               enum fib_event_type event_type,
 395                               struct fib6_info *rt,
 396                               struct netlink_ext_ack *extack)
 397 {
 398         struct fib6_entry_notifier_info info = {
 399                 .info.extack = extack,
 400                 .rt = rt,
 401         };
 402
 403         rt->fib6_table->fib_seq++;
 404         return call_fib6_notifiers(net, event_type, &info.info);
 405 }
 406
 407 int call_fib6_multipath_entry_notifiers(struct net *net,
 408                                         enum fib_event_type event_type,
 409                                         struct fib6_info *rt,
 410                                         unsigned int nsiblings,
 411                                         struct netlink_ext_ack *extack)
 412 {
 413         struct fib6_entry_notifier_info info = {
 414                 .info.extack = extack,
 415                 .rt = rt,
 416                 .nsiblings = nsiblings,
 417         };
 418
 419         rt->fib6_table->fib_seq++;
 420         return call_fib6_notifiers(net, event_type, &info.info);
 421 }
 422
 423 int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
 424 {
 425         struct fib6_entry_notifier_info info = {
 426                 .rt = rt,
 427                 .nsiblings = rt->fib6_nsiblings,
 428         };
 429
 430         rt->fib6_table->fib_seq++;
 431         return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE, &info.info);
 432 }
 433
 434 struct fib6_dump_arg {
 435         struct net *net;
 436         struct notifier_block *nb;
 437         struct netlink_ext_ack *extack;
 438 };
 439
 440 static int fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 441 {
 442         enum fib_event_type fib_event = FIB_EVENT_ENTRY_REPLACE;
 443         int err;
 444
 445         if (!rt || rt == arg->net->ipv6.fib6_null_entry)
 446                 return 0;
 447
 448         if (rt->fib6_nsiblings)
 449                 err = call_fib6_multipath_entry_notifier(arg->nb, fib_event,
 450                                                          rt,
 451                                                          rt->fib6_nsiblings,
 452                                                          arg->extack);
 453         else
 454                 err = call_fib6_entry_notifier(arg->nb, fib_event, rt,
 455                                                arg->extack);
 456
 457         return err;
 458 }
 459
 460 static int fib6_node_dump(struct fib6_walker *w)
 461 {
 462         int err;
 463
 464         err = fib6_rt_dump(w->leaf, w->args);
 465         w->leaf = NULL;
 466         return err;
 467 }
 468
 469 static int fib6_table_dump(struct net *net, struct fib6_table *tb,
 470                            struct fib6_walker *w)
 471 {
 472         int err;
 473
 474         w->root = &tb->tb6_root;
 475         spin_lock_bh(&tb->tb6_lock);
 476         err = fib6_walk(net, w);
 477         spin_unlock_bh(&tb->tb6_lock);
 478         return err;
 479 }
 480
 481 /* Called with rcu_read_lock() */
 482 int fib6_tables_dump(struct net *net, struct notifier_block *nb,
 483                      struct netlink_ext_ack *extack)
 484 {
 485         struct fib6_dump_arg arg;
 486         struct fib6_walker *w;
 487         unsigned int h;
 488         int err = 0;
 489
 490         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 491         if (!w)
 492                 return -ENOMEM;
 493
 494         w->func = fib6_node_dump;
 495         arg.net = net;
 496         arg.nb = nb;
 497         arg.extack = extack;
 498         w->args = &arg;
 499
 500         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 501                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 502                 struct fib6_table *tb;
 503
 504                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 505                         err = fib6_table_dump(net, tb, w);
 506                         if (err)
 507                                 goto out;
 508                 }
 509         }
 510
 511 out:
 512         kfree(w);
 513
 514         /* The tree traversal function should never return a positive value. */
 515         return err > 0 ? -EINVAL : err;
 516 }
 517
 518 static int fib6_dump_node(struct fib6_walker *w)
 519 {
 520         int res;
 521         struct fib6_info *rt;
 522
 523         for_each_fib6_walker_rt(w) {
 524                 res = rt6_dump_route(rt, w->args, w->skip_in_node);
 525                 if (res >= 0) {
 526                         /* Frame is full, suspend walking */
 527                         w->leaf = rt;
 528
 529                         /* We'll restart from this node, so if some routes were
 530                          * already dumped, skip them next time.
 531                          */
 532                         w->skip_in_node += res;
 533
 534                         return 1;
 535                 }
 536                 w->skip_in_node = 0;
 537
 538                 /* Multipath routes are dumped in one route with the
 539                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 540                  * last sibling of this route (no need to dump the
 541                  * sibling routes again)
 542                  */
 543                 if (rt->fib6_nsiblings)
 544                         rt = list_last_entry(&rt->fib6_siblings,
 545                                              struct fib6_info,
 546                                              fib6_siblings);
 547         }
 548         w->leaf = NULL;
 549         return 0;
 550 }
 551
 552 static void fib6_dump_end(struct netlink_callback *cb)
 553 {
 554         struct net *net = sock_net(cb->skb->sk);
 555         struct fib6_walker *w = (void *)cb->args[2];
 556
 557         if (w) {
 558                 if (cb->args[4]) {
 559                         cb->args[4] = 0;
 560                         fib6_walker_unlink(net, w);
 561                 }
 562                 cb->args[2] = 0;
 563                 kfree(w);
 564         }
 565         cb->done = (void *)cb->args[3];
 566         cb->args[1] = 3;
 567 }
 568
 569 static int fib6_dump_done(struct netlink_callback *cb)
 570 {
 571         fib6_dump_end(cb);
 572         return cb->done ? cb->done(cb) : 0;
 573 }
 574
 575 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 576                            struct netlink_callback *cb)
 577 {
 578         struct net *net = sock_net(skb->sk);
 579         struct fib6_walker *w;
 580         int res;
 581
 582         w = (void *)cb->args[2];
 583         w->root = &table->tb6_root;
 584
 585         if (cb->args[4] == 0) {
 586                 w->count = 0;
 587                 w->skip = 0;
 588                 w->skip_in_node = 0;
 589
 590                 spin_lock_bh(&table->tb6_lock);
 591                 res = fib6_walk(net, w);
 592                 spin_unlock_bh(&table->tb6_lock);
 593                 if (res > 0) {
 594                         cb->args[4] = 1;
 595                         cb->args[5] = READ_ONCE(w->root->fn_sernum);
 596                 }
 597         } else {
 598                 int sernum = READ_ONCE(w->root->fn_sernum);
 599                 if (cb->args[5] != sernum) {
 600                         /* Begin at the root if the tree changed */
 601                         cb->args[5] = sernum;
 602                         w->state = FWS_INIT;
 603                         w->node = w->root;
 604                         w->skip = w->count;
 605                         w->skip_in_node = 0;
 606                 } else
 607                         w->skip = 0;
 608
 609                 spin_lock_bh(&table->tb6_lock);
 610                 res = fib6_walk_continue(w);
 611                 spin_unlock_bh(&table->tb6_lock);
 612                 if (res <= 0) {
 613                         fib6_walker_unlink(net, w);
 614                         cb->args[4] = 0;
 615                 }
 616         }
 617
 618         return res;
 619 }
 620
 621 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 622 {
 623         struct rt6_rtnl_dump_arg arg = {
 624                 .filter.dump_exceptions = true,
 625                 .filter.dump_routes = true,
 626                 .filter.rtnl_held = false,
 627         };
 628         const struct nlmsghdr *nlh = cb->nlh;
 629         struct net *net = sock_net(skb->sk);
 630         unsigned int e = 0, s_e;
 631         struct hlist_head *head;
 632         struct fib6_walker *w;
 633         struct fib6_table *tb;
 634         unsigned int h, s_h;
 635         int err = 0;
 636
 637         rcu_read_lock();
 638         if (cb->strict_check) {
 639                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 640                 if (err < 0)
 641                         goto unlock;
 642         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 643                 struct rtmsg *rtm = nlmsg_data(nlh);
 644
 645                 if (rtm->rtm_flags & RTM_F_PREFIX)
 646                         arg.filter.flags = RTM_F_PREFIX;
 647         }
 648
 649         w = (void *)cb->args[2];
 650         if (!w) {
 651                 /* New dump:
 652                  *
 653                  * 1. allocate and initialize walker.
 654                  */
 655                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 656                 if (!w) {
 657                         err = -ENOMEM;
 658                         goto unlock;
 659                 }
 660                 w->func = fib6_dump_node;
 661                 cb->args[2] = (long)w;
 662
 663                 /* 2. hook callback destructor.
 664                  */
 665                 cb->args[3] = (long)cb->done;
 666                 cb->done = fib6_dump_done;
 667
 668         }
 669
 670         arg.skb = skb;
 671         arg.cb = cb;
 672         arg.net = net;
 673         w->args = &arg;
 674
 675         if (arg.filter.table_id) {
 676                 tb = fib6_get_table(net, arg.filter.table_id);
 677                 if (!tb) {
 678                         if (rtnl_msg_family(cb->nlh) != PF_INET6)
 679                                 goto unlock;
 680
 681                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 682                         err = -ENOENT;
 683                         goto unlock;
 684                 }
 685
 686                 if (!cb->args[0]) {
 687                         err = fib6_dump_table(tb, skb, cb);
 688                         if (!err)
 689                                 cb->args[0] = 1;
 690                 }
 691                 goto unlock;
 692         }
 693
 694         s_h = cb->args[0];
 695         s_e = cb->args[1];
 696
 697         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 698                 e = 0;
 699                 head = &net->ipv6.fib_table_hash[h];
 700                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 701                         if (e < s_e)
 702                                 goto next;
 703                         err = fib6_dump_table(tb, skb, cb);
 704                         if (err != 0)
 705                                 goto out;
 706 next:
 707                         e++;
 708                 }
 709         }
 710 out:
 711         cb->args[1] = e;
 712         cb->args[0] = h;
 713
 714 unlock:
 715         rcu_read_unlock();
 716         if (err <= 0)
 717                 fib6_dump_end(cb);
 718         return err;
 719 }
 720
 721 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 722 {
 723         if (!f6i)
 724                 return;
 725
 726         if (f6i->fib6_metrics == &dst_default_metrics) {
 727                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 728
 729                 if (!p)
 730                         return;
 731
 732                 refcount_set(&p->refcnt, 1);
 733                 f6i->fib6_metrics = p;
 734         }
 735
 736         f6i->fib6_metrics->metrics[metric - 1] = val;
 737 }
 738
 739 /*
 740  *      Routing Table
 741  *
 742  *      return the appropriate node for a routing tree "add" operation
 743  *      by either creating and inserting or by returning an existing
 744  *      node.
 745  */
 746
 747 static struct fib6_node *fib6_add_1(struct net *net,
 748                                     struct fib6_table *table,
 749                                     struct fib6_node *root,
 750                                     struct in6_addr *addr, int plen,
 751                                     int offset, int allow_create,
 752                                     int replace_required,
 753                                     struct netlink_ext_ack *extack)
 754 {
 755         struct fib6_node *fn, *in, *ln;
 756         struct fib6_node *pn = NULL;
 757         struct rt6key *key;
 758         int     bit;
 759         __be32  dir = 0;
 760
 761         /* insert node in tree */
 762
 763         fn = root;
 764
 765         do {
 766                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 767                                             lockdep_is_held(&table->tb6_lock));
 768                 key = (struct rt6key *)((u8 *)leaf + offset);
 769
 770                 /*
 771                  *      Prefix match
 772                  */
 773                 if (plen < fn->fn_bit ||
 774                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 775                         if (!allow_create) {
 776                                 if (replace_required) {
 777                                         NL_SET_ERR_MSG(extack,
 778                                                        "Can not replace route - no match found");
 779                                         pr_warn("Can't replace route, no match found\n");
 780                                         return ERR_PTR(-ENOENT);
 781                                 }
 782                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 783                         }
 784                         goto insert_above;
 785                 }
 786
 787                 /*
 788                  *      Exact match ?
 789                  */
 790
 791                 if (plen == fn->fn_bit) {
 792                         /* clean up an intermediate node */
 793                         if (!(fn->fn_flags & RTN_RTINFO)) {
 794                                 RCU_INIT_POINTER(fn->leaf, NULL);
 795                                 fib6_info_release(leaf);
 796                         /* remove null_entry in the root node */
 797                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 798                                    rcu_access_pointer(fn->leaf) ==
 799                                    net->ipv6.fib6_null_entry) {
 800                                 RCU_INIT_POINTER(fn->leaf, NULL);
 801                         }
 802
 803                         return fn;
 804                 }
 805
 806                 /*
 807                  *      We have more bits to go
 808                  */
 809
 810                 /* Try to walk down on tree. */
 811                 dir = addr_bit_set(addr, fn->fn_bit);
 812                 pn = fn;
 813                 fn = dir ?
 814                      rcu_dereference_protected(fn->right,
 815                                         lockdep_is_held(&table->tb6_lock)) :
 816                      rcu_dereference_protected(fn->left,
 817                                         lockdep_is_held(&table->tb6_lock));
 818         } while (fn);
 819
 820         if (!allow_create) {
 821                 /* We should not create new node because
 822                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 823                  * I assume it is safe to require NLM_F_CREATE when
 824                  * REPLACE flag is used! Later we may want to remove the
 825                  * check for replace_required, because according
 826                  * to netlink specification, NLM_F_CREATE
 827                  * MUST be specified if new route is created.
 828                  * That would keep IPv6 consistent with IPv4
 829                  */
 830                 if (replace_required) {
 831                         NL_SET_ERR_MSG(extack,
 832                                        "Can not replace route - no match found");
 833                         pr_warn("Can't replace route, no match found\n");
 834                         return ERR_PTR(-ENOENT);
 835                 }
 836                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 837         }
 838         /*
 839          *      We walked to the bottom of tree.
 840          *      Create new leaf node without children.
 841          */
 842
 843         ln = node_alloc(net);
 844
 845         if (!ln)
 846                 return ERR_PTR(-ENOMEM);
 847         ln->fn_bit = plen;
 848         RCU_INIT_POINTER(ln->parent, pn);
 849
 850         if (dir)
 851                 rcu_assign_pointer(pn->right, ln);
 852         else
 853                 rcu_assign_pointer(pn->left, ln);
 854
 855         return ln;
 856
 857
 858 insert_above:
 859         /*
 860          * split since we don't have a common prefix anymore or
 861          * we have a less significant route.
 862          * we've to insert an intermediate node on the list
 863          * this new node will point to the one we need to create
 864          * and the current
 865          */
 866
 867         pn = rcu_dereference_protected(fn->parent,
 868                                        lockdep_is_held(&table->tb6_lock));
 869
 870         /* find 1st bit in difference between the 2 addrs.
 871
 872            See comment in __ipv6_addr_diff: bit may be an invalid value,
 873            but if it is >= plen, the value is ignored in any case.
 874          */
 875
 876         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 877
 878         /*
 879          *              (intermediate)[in]
 880          *                /        \
 881          *      (new leaf node)[ln] (old node)[fn]
 882          */
 883         if (plen > bit) {
 884                 in = node_alloc(net);
 885                 ln = node_alloc(net);
 886
 887                 if (!in || !ln) {
 888                         if (in)
 889                                 node_free_immediate(net, in);
 890                         if (ln)
 891                                 node_free_immediate(net, ln);
 892                         return ERR_PTR(-ENOMEM);
 893                 }
 894
 895                 /*
 896                  * new intermediate node.
 897                  * RTN_RTINFO will
 898                  * be off since that an address that chooses one of
 899                  * the branches would not match less specific routes
 900                  * in the other branch
 901                  */
 902
 903                 in->fn_bit = bit;
 904
 905                 RCU_INIT_POINTER(in->parent, pn);
 906                 in->leaf = fn->leaf;
 907                 fib6_info_hold(rcu_dereference_protected(in->leaf,
 908                                 lockdep_is_held(&table->tb6_lock)));
 909
 910                 /* update parent pointer */
 911                 if (dir)
 912                         rcu_assign_pointer(pn->right, in);
 913                 else
 914                         rcu_assign_pointer(pn->left, in);
 915
 916                 ln->fn_bit = plen;
 917
 918                 RCU_INIT_POINTER(ln->parent, in);
 919                 rcu_assign_pointer(fn->parent, in);
 920
 921                 if (addr_bit_set(addr, bit)) {
 922                         rcu_assign_pointer(in->right, ln);
 923                         rcu_assign_pointer(in->left, fn);
 924                 } else {
 925                         rcu_assign_pointer(in->left, ln);
 926                         rcu_assign_pointer(in->right, fn);
 927                 }
 928         } else { /* plen <= bit */
 929
 930                 /*
 931                  *              (new leaf node)[ln]
 932                  *                /        \
 933                  *           (old node)[fn] NULL
 934                  */
 935
 936                 ln = node_alloc(net);
 937
 938                 if (!ln)
 939                         return ERR_PTR(-ENOMEM);
 940
 941                 ln->fn_bit = plen;
 942
 943                 RCU_INIT_POINTER(ln->parent, pn);
 944
 945                 if (addr_bit_set(&key->addr, plen))
 946                         RCU_INIT_POINTER(ln->right, fn);
 947                 else
 948                         RCU_INIT_POINTER(ln->left, fn);
 949
 950                 rcu_assign_pointer(fn->parent, ln);
 951
 952                 if (dir)
 953                         rcu_assign_pointer(pn->right, ln);
 954                 else
 955                         rcu_assign_pointer(pn->left, ln);
 956         }
 957         return ln;
 958 }
 959
 960 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
 961                                   const struct fib6_info *match,
 962                                   const struct fib6_table *table)
 963 {
 964         int cpu;
 965
 966         if (!fib6_nh->rt6i_pcpu)
 967                 return;
 968
 969         /* release the reference to this fib entry from
 970          * all of its cached pcpu routes
 971          */
 972         for_each_possible_cpu(cpu) {
 973                 struct rt6_info **ppcpu_rt;
 974                 struct rt6_info *pcpu_rt;
 975
 976                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
 977                 pcpu_rt = *ppcpu_rt;
 978
 979                 /* only dropping the 'from' reference if the cached route
 980                  * is using 'match'. The cached pcpu_rt->from only changes
 981                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
 982                  * change from one fib6_info reference to another
 983                  */
 984                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
 985                         struct fib6_info *from;
 986
 987                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 988                         fib6_info_release(from);
 989                 }
 990         }
 991 }
 992
 993 struct fib6_nh_pcpu_arg {
 994         struct fib6_info        *from;
 995         const struct fib6_table *table;
 996 };
 997
 998 static int fib6_nh_drop_pcpu_from(struct fib6_nh *nh, void *_arg)
 999 {
1000         struct fib6_nh_pcpu_arg *arg = _arg;
1001
1002         __fib6_drop_pcpu_from(nh, arg->from, arg->table);
1003         return 0;
1004 }
1005
1006 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
1007                                 const struct fib6_table *table)
1008 {
1009         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
1010          * while we are cleaning them here.
1011          */
1012         f6i->fib6_destroying = 1;
1013         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
1014
1015         if (f6i->nh) {
1016                 struct fib6_nh_pcpu_arg arg = {
1017                         .from = f6i,
1018                         .table = table
1019                 };
1020
1021                 nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_drop_pcpu_from,
1022                                          &arg);
1023         } else {
1024                 struct fib6_nh *fib6_nh;
1025
1026                 fib6_nh = f6i->fib6_nh;
1027                 __fib6_drop_pcpu_from(fib6_nh, f6i, table);
1028         }
1029 }
1030
1031 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
1032                           struct net *net)
1033 {
1034         struct fib6_table *table = rt->fib6_table;
1035
1036         /* Flush all cached dst in exception table */
1037         rt6_flush_exceptions(rt);
1038         fib6_drop_pcpu_from(rt, table);
1039
1040         if (rt->nh && !list_empty(&rt->nh_list))
1041                 list_del_init(&rt->nh_list);
1042
1043         if (refcount_read(&rt->fib6_ref) != 1) {
1044                 /* This route is used as dummy address holder in some split
1045                  * nodes. It is not leaked, but it still holds other resources,
1046                  * which must be released in time. So, scan ascendant nodes
1047                  * and replace dummy references to this route with references
1048                  * to still alive ones.
1049                  */
1050                 while (fn) {
1051                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1052                                             lockdep_is_held(&table->tb6_lock));
1053                         struct fib6_info *new_leaf;
1054                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
1055                                 new_leaf = fib6_find_prefix(net, table, fn);
1056                                 fib6_info_hold(new_leaf);
1057
1058                                 rcu_assign_pointer(fn->leaf, new_leaf);
1059                                 fib6_info_release(rt);
1060                         }
1061                         fn = rcu_dereference_protected(fn->parent,
1062                                     lockdep_is_held(&table->tb6_lock));
1063                 }
1064         }
1065
1066         fib6_clean_expires(rt);
1067         fib6_remove_gc_list(rt);
1068 }
1069
1070 /*
1071  *      Insert routing information in a node.
1072  */
1073
1074 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
1075                             struct nl_info *info,
1076                             struct netlink_ext_ack *extack)
1077 {
1078         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
1079                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1080         struct fib6_info *iter = NULL;
1081         struct fib6_info __rcu **ins;
1082         struct fib6_info __rcu **fallback_ins = NULL;
1083         int replace = (info->nlh &&
1084                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
1085         int add = (!info->nlh ||
1086                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
1087         int found = 0;
1088         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
1089         bool notify_sibling_rt = false;
1090         u16 nlflags = NLM_F_EXCL;
1091         int err;
1092
1093         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
1094                 nlflags |= NLM_F_APPEND;
1095
1096         ins = &fn->leaf;
1097
1098         for (iter = leaf; iter;
1099              iter = rcu_dereference_protected(iter->fib6_next,
1100                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
1101                 /*
1102                  *      Search for duplicates
1103                  */
1104
1105                 if (iter->fib6_metric == rt->fib6_metric) {
1106                         /*
1107                          *      Same priority level
1108                          */
1109                         if (info->nlh &&
1110                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1111                                 return -EEXIST;
1112
1113                         nlflags &= ~NLM_F_EXCL;
1114                         if (replace) {
1115                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1116                                         found++;
1117                                         break;
1118                                 }
1119                                 fallback_ins = fallback_ins ?: ins;
1120                                 goto next_iter;
1121                         }
1122
1123                         if (rt6_duplicate_nexthop(iter, rt)) {
1124                                 if (rt->fib6_nsiblings)
1125                                         rt->fib6_nsiblings = 0;
1126                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1127                                         return -EEXIST;
1128                                 if (!(rt->fib6_flags & RTF_EXPIRES)) {
1129                                         fib6_clean_expires(iter);
1130                                         fib6_remove_gc_list(iter);
1131                                 } else {
1132                                         fib6_set_expires(iter, rt->expires);
1133                                         fib6_add_gc_list(iter);
1134                                 }
1135
1136                                 if (rt->fib6_pmtu)
1137                                         fib6_metric_set(iter, RTAX_MTU,
1138                                                         rt->fib6_pmtu);
1139                                 return -EEXIST;
1140                         }
1141                         /* If we have the same destination and the same metric,
1142                          * but not the same gateway, then the route we try to
1143                          * add is sibling to this route, increment our counter
1144                          * of siblings, and later we will add our route to the
1145                          * list.
1146                          * Only static routes (which don't have flag
1147                          * RTF_EXPIRES) are used for ECMPv6.
1148                          *
1149                          * To avoid long list, we only had siblings if the
1150                          * route have a gateway.
1151                          */
1152                         if (rt_can_ecmp &&
1153                             rt6_qualify_for_ecmp(iter))
1154                                 rt->fib6_nsiblings++;
1155                 }
1156
1157                 if (iter->fib6_metric > rt->fib6_metric)
1158                         break;
1159
1160 next_iter:
1161                 ins = &iter->fib6_next;
1162         }
1163
1164         if (fallback_ins && !found) {
1165                 /* No matching route with same ecmp-able-ness found, replace
1166                  * first matching route
1167                  */
1168                 ins = fallback_ins;
1169                 iter = rcu_dereference_protected(*ins,
1170                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1171                 found++;
1172         }
1173
1174         /* Reset round-robin state, if necessary */
1175         if (ins == &fn->leaf)
1176                 fn->rr_ptr = NULL;
1177
1178         /* Link this route to others same route. */
1179         if (rt->fib6_nsiblings) {
1180                 unsigned int fib6_nsiblings;
1181                 struct fib6_info *sibling, *temp_sibling;
1182
1183                 /* Find the first route that have the same metric */
1184                 sibling = leaf;
1185                 notify_sibling_rt = true;
1186                 while (sibling) {
1187                         if (sibling->fib6_metric == rt->fib6_metric &&
1188                             rt6_qualify_for_ecmp(sibling)) {
1189                                 list_add_tail(&rt->fib6_siblings,
1190                                               &sibling->fib6_siblings);
1191                                 break;
1192                         }
1193                         sibling = rcu_dereference_protected(sibling->fib6_next,
1194                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1195                         notify_sibling_rt = false;
1196                 }
1197                 /* For each sibling in the list, increment the counter of
1198                  * siblings. BUG() if counters does not match, list of siblings
1199                  * is broken!
1200                  */
1201                 fib6_nsiblings = 0;
1202                 list_for_each_entry_safe(sibling, temp_sibling,
1203                                          &rt->fib6_siblings, fib6_siblings) {
1204                         sibling->fib6_nsiblings++;
1205                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1206                         fib6_nsiblings++;
1207                 }
1208                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1209                 rt6_multipath_rebalance(temp_sibling);
1210         }
1211
1212         /*
1213          *      insert node
1214          */
1215         if (!replace) {
1216                 if (!add)
1217                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1218
1219 add:
1220                 nlflags |= NLM_F_CREATE;
1221
1222                 /* The route should only be notified if it is the first
1223                  * route in the node or if it is added as a sibling
1224                  * route to the first route in the node.
1225                  */
1226                 if (!info->skip_notify_kernel &&
1227                     (notify_sibling_rt || ins == &fn->leaf)) {
1228                         enum fib_event_type fib_event;
1229
1230                         if (notify_sibling_rt)
1231                                 fib_event = FIB_EVENT_ENTRY_APPEND;
1232                         else
1233                                 fib_event = FIB_EVENT_ENTRY_REPLACE;
1234                         err = call_fib6_entry_notifiers(info->nl_net,
1235                                                         fib_event, rt,
1236                                                         extack);
1237                         if (err) {
1238                                 struct fib6_info *sibling, *next_sibling;
1239
1240                                 /* If the route has siblings, then it first
1241                                  * needs to be unlinked from them.
1242                                  */
1243                                 if (!rt->fib6_nsiblings)
1244                                         return err;
1245
1246                                 list_for_each_entry_safe(sibling, next_sibling,
1247                                                          &rt->fib6_siblings,
1248                                                          fib6_siblings)
1249                                         sibling->fib6_nsiblings--;
1250                                 rt->fib6_nsiblings = 0;
1251                                 list_del_init(&rt->fib6_siblings);
1252                                 rt6_multipath_rebalance(next_sibling);
1253                                 return err;
1254                         }
1255                 }
1256
1257                 rcu_assign_pointer(rt->fib6_next, iter);
1258                 fib6_info_hold(rt);
1259                 rcu_assign_pointer(rt->fib6_node, fn);
1260                 rcu_assign_pointer(*ins, rt);
1261                 if (!info->skip_notify)
1262                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1263                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1264
1265                 if (!(fn->fn_flags & RTN_RTINFO)) {
1266                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1267                         fn->fn_flags |= RTN_RTINFO;
1268                 }
1269
1270         } else {
1271                 int nsiblings;
1272
1273                 if (!found) {
1274                         if (add)
1275                                 goto add;
1276                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1277                         return -ENOENT;
1278                 }
1279
1280                 if (!info->skip_notify_kernel && ins == &fn->leaf) {
1281                         err = call_fib6_entry_notifiers(info->nl_net,
1282                                                         FIB_EVENT_ENTRY_REPLACE,
1283                                                         rt, extack);
1284                         if (err)
1285                                 return err;
1286                 }
1287
1288                 fib6_info_hold(rt);
1289                 rcu_assign_pointer(rt->fib6_node, fn);
1290                 rt->fib6_next = iter->fib6_next;
1291                 rcu_assign_pointer(*ins, rt);
1292                 if (!info->skip_notify)
1293                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1294                 if (!(fn->fn_flags & RTN_RTINFO)) {
1295                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1296                         fn->fn_flags |= RTN_RTINFO;
1297                 }
1298                 nsiblings = iter->fib6_nsiblings;
1299                 iter->fib6_node = NULL;
1300                 fib6_purge_rt(iter, fn, info->nl_net);
1301                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1302                         fn->rr_ptr = NULL;
1303                 fib6_info_release(iter);
1304
1305                 if (nsiblings) {
1306                         /* Replacing an ECMP route, remove all siblings */
1307                         ins = &rt->fib6_next;
1308                         iter = rcu_dereference_protected(*ins,
1309                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1310                         while (iter) {
1311                                 if (iter->fib6_metric > rt->fib6_metric)
1312                                         break;
1313                                 if (rt6_qualify_for_ecmp(iter)) {
1314                                         *ins = iter->fib6_next;
1315                                         iter->fib6_node = NULL;
1316                                         fib6_purge_rt(iter, fn, info->nl_net);
1317                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1318                                                 fn->rr_ptr = NULL;
1319                                         fib6_info_release(iter);
1320                                         nsiblings--;
1321                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1322                                 } else {
1323                                         ins = &iter->fib6_next;
1324                                 }
1325                                 iter = rcu_dereference_protected(*ins,
1326                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1327                         }
1328                         WARN_ON(nsiblings != 0);
1329                 }
1330         }
1331
1332         return 0;
1333 }
1334
1335 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1336 {
1337         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1338             (rt->fib6_flags & RTF_EXPIRES))
1339                 mod_timer(&net->ipv6.ip6_fib_timer,
1340                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1341 }
1342
1343 void fib6_force_start_gc(struct net *net)
1344 {
1345         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1346                 mod_timer(&net->ipv6.ip6_fib_timer,
1347                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1348 }
1349
1350 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1351                                            int sernum)
1352 {
1353         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1354                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1355
1356         /* paired with smp_rmb() in fib6_get_cookie_safe() */
1357         smp_wmb();
1358         while (fn) {
1359                 WRITE_ONCE(fn->fn_sernum, sernum);
1360                 fn = rcu_dereference_protected(fn->parent,
1361                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1362         }
1363 }
1364
1365 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1366 {
1367         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1368 }
1369
1370 /* allow ipv4 to update sernum via ipv6_stub */
1371 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1372 {
1373         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1374         fib6_update_sernum_upto_root(net, f6i);
1375         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1376 }
1377
1378 /*
1379  *      Add routing information to the routing tree.
1380  *      <destination addr>/<source addr>
1381  *      with source addr info in sub-trees
1382  *      Need to own table->tb6_lock
1383  */
1384
1385 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1386              struct nl_info *info, struct netlink_ext_ack *extack)
1387 {
1388         struct fib6_table *table = rt->fib6_table;
1389         struct fib6_node *fn;
1390 #ifdef CONFIG_IPV6_SUBTREES
1391         struct fib6_node *pn = NULL;
1392 #endif
1393         int err = -ENOMEM;
1394         int allow_create = 1;
1395         int replace_required = 0;
1396
1397         if (info->nlh) {
1398                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1399                         allow_create = 0;
1400                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1401                         replace_required = 1;
1402         }
1403         if (!allow_create && !replace_required)
1404                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1405
1406         fn = fib6_add_1(info->nl_net, table, root,
1407                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1408                         offsetof(struct fib6_info, fib6_dst), allow_create,
1409                         replace_required, extack);
1410         if (IS_ERR(fn)) {
1411                 err = PTR_ERR(fn);
1412                 fn = NULL;
1413                 goto out;
1414         }
1415
1416 #ifdef CONFIG_IPV6_SUBTREES
1417         pn = fn;
1418
1419         if (rt->fib6_src.plen) {
1420                 struct fib6_node *sn;
1421
1422                 if (!rcu_access_pointer(fn->subtree)) {
1423                         struct fib6_node *sfn;
1424
1425                         /*
1426                          * Create subtree.
1427                          *
1428                          *              fn[main tree]
1429                          *              |
1430                          *              sfn[subtree root]
1431                          *                 \
1432                          *                  sn[new leaf node]
1433                          */
1434
1435                         /* Create subtree root node */
1436                         sfn = node_alloc(info->nl_net);
1437                         if (!sfn)
1438                                 goto failure;
1439
1440                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1441                         rcu_assign_pointer(sfn->leaf,
1442                                            info->nl_net->ipv6.fib6_null_entry);
1443                         sfn->fn_flags = RTN_ROOT;
1444
1445                         /* Now add the first leaf node to new subtree */
1446
1447                         sn = fib6_add_1(info->nl_net, table, sfn,
1448                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1449                                         offsetof(struct fib6_info, fib6_src),
1450                                         allow_create, replace_required, extack);
1451
1452                         if (IS_ERR(sn)) {
1453                                 /* If it is failed, discard just allocated
1454                                    root, and then (in failure) stale node
1455                                    in main tree.
1456                                  */
1457                                 node_free_immediate(info->nl_net, sfn);
1458                                 err = PTR_ERR(sn);
1459                                 goto failure;
1460                         }
1461
1462                         /* Now link new subtree to main tree */
1463                         rcu_assign_pointer(sfn->parent, fn);
1464                         rcu_assign_pointer(fn->subtree, sfn);
1465                 } else {
1466                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1467                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1468                                         offsetof(struct fib6_info, fib6_src),
1469                                         allow_create, replace_required, extack);
1470
1471                         if (IS_ERR(sn)) {
1472                                 err = PTR_ERR(sn);
1473                                 goto failure;
1474                         }
1475                 }
1476
1477                 if (!rcu_access_pointer(fn->leaf)) {
1478                         if (fn->fn_flags & RTN_TL_ROOT) {
1479                                 /* put back null_entry for root node */
1480                                 rcu_assign_pointer(fn->leaf,
1481                                             info->nl_net->ipv6.fib6_null_entry);
1482                         } else {
1483                                 fib6_info_hold(rt);
1484                                 rcu_assign_pointer(fn->leaf, rt);
1485                         }
1486                 }
1487                 fn = sn;
1488         }
1489 #endif
1490
1491         err = fib6_add_rt2node(fn, rt, info, extack);
1492         if (!err) {
1493                 if (rt->nh)
1494                         list_add(&rt->nh_list, &rt->nh->f6i_list);
1495                 __fib6_update_sernum_upto_root(rt, fib6_new_sernum(info->nl_net));
1496
1497                 if (rt->fib6_flags & RTF_EXPIRES)
1498                         fib6_add_gc_list(rt);
1499
1500                 fib6_start_gc(info->nl_net, rt);
1501         }
1502
1503 out:
1504         if (err) {
1505 #ifdef CONFIG_IPV6_SUBTREES
1506                 /*
1507                  * If fib6_add_1 has cleared the old leaf pointer in the
1508                  * super-tree leaf node we have to find a new one for it.
1509                  */
1510                 if (pn != fn) {
1511                         struct fib6_info *pn_leaf =
1512                                 rcu_dereference_protected(pn->leaf,
1513                                     lockdep_is_held(&table->tb6_lock));
1514                         if (pn_leaf == rt) {
1515                                 pn_leaf = NULL;
1516                                 RCU_INIT_POINTER(pn->leaf, NULL);
1517                                 fib6_info_release(rt);
1518                         }
1519                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1520                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1521                                                            pn);
1522                                 if (!pn_leaf)
1523                                         pn_leaf =
1524                                             info->nl_net->ipv6.fib6_null_entry;
1525                                 fib6_info_hold(pn_leaf);
1526                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1527                         }
1528                 }
1529 #endif
1530                 goto failure;
1531         } else if (fib6_requires_src(rt)) {
1532                 fib6_routes_require_src_inc(info->nl_net);
1533         }
1534         return err;
1535
1536 failure:
1537         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1538          * 1. fn is an intermediate node and we failed to add the new
1539          * route to it in both subtree creation failure and fib6_add_rt2node()
1540          * failure case.
1541          * 2. fn is the root node in the table and we fail to add the first
1542          * default route to it.
1543          */
1544         if (fn &&
1545             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1546              (fn->fn_flags & RTN_TL_ROOT &&
1547               !rcu_access_pointer(fn->leaf))))
1548                 fib6_repair_tree(info->nl_net, table, fn);
1549         return err;
1550 }
1551
1552 /*
1553  *      Routing tree lookup
1554  *
1555  */
1556
1557 struct lookup_args {
1558         int                     offset;         /* key offset on fib6_info */
1559         const struct in6_addr   *addr;          /* search key                   */
1560 };
1561
1562 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1563                                             struct lookup_args *args)
1564 {
1565         struct fib6_node *fn;
1566         __be32 dir;
1567
1568         if (unlikely(args->offset == 0))
1569                 return NULL;
1570
1571         /*
1572          *      Descend on a tree
1573          */
1574
1575         fn = root;
1576
1577         for (;;) {
1578                 struct fib6_node *next;
1579
1580                 dir = addr_bit_set(args->addr, fn->fn_bit);
1581
1582                 next = dir ? rcu_dereference(fn->right) :
1583                              rcu_dereference(fn->left);
1584
1585                 if (next) {
1586                         fn = next;
1587                         continue;
1588                 }
1589                 break;
1590         }
1591
1592         while (fn) {
1593                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1594
1595                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1596                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1597                         struct rt6key *key;
1598
1599                         if (!leaf)
1600                                 goto backtrack;
1601
1602                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1603
1604                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1605 #ifdef CONFIG_IPV6_SUBTREES
1606                                 if (subtree) {
1607                                         struct fib6_node *sfn;
1608                                         sfn = fib6_node_lookup_1(subtree,
1609                                                                  args + 1);
1610                                         if (!sfn)
1611                                                 goto backtrack;
1612                                         fn = sfn;
1613                                 }
1614 #endif
1615                                 if (fn->fn_flags & RTN_RTINFO)
1616                                         return fn;
1617                         }
1618                 }
1619 backtrack:
1620                 if (fn->fn_flags & RTN_ROOT)
1621                         break;
1622
1623                 fn = rcu_dereference(fn->parent);
1624         }
1625
1626         return NULL;
1627 }
1628
1629 /* called with rcu_read_lock() held
1630  */
1631 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1632                                    const struct in6_addr *daddr,
1633                                    const struct in6_addr *saddr)
1634 {
1635         struct fib6_node *fn;
1636         struct lookup_args args[] = {
1637                 {
1638                         .offset = offsetof(struct fib6_info, fib6_dst),
1639                         .addr = daddr,
1640                 },
1641 #ifdef CONFIG_IPV6_SUBTREES
1642                 {
1643                         .offset = offsetof(struct fib6_info, fib6_src),
1644                         .addr = saddr,
1645                 },
1646 #endif
1647                 {
1648                         .offset = 0,    /* sentinel */
1649                 }
1650         };
1651
1652         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1653         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1654                 fn = root;
1655
1656         return fn;
1657 }
1658
1659 /*
1660  *      Get node with specified destination prefix (and source prefix,
1661  *      if subtrees are used)
1662  *      exact_match == true means we try to find fn with exact match of
1663  *      the passed in prefix addr
1664  *      exact_match == false means we try to find fn with longest prefix
1665  *      match of the passed in prefix addr. This is useful for finding fn
1666  *      for cached route as it will be stored in the exception table under
1667  *      the node with longest prefix length.
1668  */
1669
1670
1671 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1672                                        const struct in6_addr *addr,
1673                                        int plen, int offset,
1674                                        bool exact_match)
1675 {
1676         struct fib6_node *fn, *prev = NULL;
1677
1678         for (fn = root; fn ; ) {
1679                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1680                 struct rt6key *key;
1681
1682                 /* This node is being deleted */
1683                 if (!leaf) {
1684                         if (plen <= fn->fn_bit)
1685                                 goto out;
1686                         else
1687                                 goto next;
1688                 }
1689
1690                 key = (struct rt6key *)((u8 *)leaf + offset);
1691
1692                 /*
1693                  *      Prefix match
1694                  */
1695                 if (plen < fn->fn_bit ||
1696                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1697                         goto out;
1698
1699                 if (plen == fn->fn_bit)
1700                         return fn;
1701
1702                 if (fn->fn_flags & RTN_RTINFO)
1703                         prev = fn;
1704
1705 next:
1706                 /*
1707                  *      We have more bits to go
1708                  */
1709                 if (addr_bit_set(addr, fn->fn_bit))
1710                         fn = rcu_dereference(fn->right);
1711                 else
1712                         fn = rcu_dereference(fn->left);
1713         }
1714 out:
1715         if (exact_match)
1716                 return NULL;
1717         else
1718                 return prev;
1719 }
1720
1721 struct fib6_node *fib6_locate(struct fib6_node *root,
1722                               const struct in6_addr *daddr, int dst_len,
1723                               const struct in6_addr *saddr, int src_len,
1724                               bool exact_match)
1725 {
1726         struct fib6_node *fn;
1727
1728         fn = fib6_locate_1(root, daddr, dst_len,
1729                            offsetof(struct fib6_info, fib6_dst),
1730                            exact_match);
1731
1732 #ifdef CONFIG_IPV6_SUBTREES
1733         if (src_len) {
1734                 WARN_ON(saddr == NULL);
1735                 if (fn) {
1736                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1737
1738                         if (subtree) {
1739                                 fn = fib6_locate_1(subtree, saddr, src_len,
1740                                            offsetof(struct fib6_info, fib6_src),
1741                                            exact_match);
1742                         }
1743                 }
1744         }
1745 #endif
1746
1747         if (fn && fn->fn_flags & RTN_RTINFO)
1748                 return fn;
1749
1750         return NULL;
1751 }
1752
1753
1754 /*
1755  *      Deletion
1756  *
1757  */
1758
1759 static struct fib6_info *fib6_find_prefix(struct net *net,
1760                                          struct fib6_table *table,
1761                                          struct fib6_node *fn)
1762 {
1763         struct fib6_node *child_left, *child_right;
1764
1765         if (fn->fn_flags & RTN_ROOT)
1766                 return net->ipv6.fib6_null_entry;
1767
1768         while (fn) {
1769                 child_left = rcu_dereference_protected(fn->left,
1770                                     lockdep_is_held(&table->tb6_lock));
1771                 child_right = rcu_dereference_protected(fn->right,
1772                                     lockdep_is_held(&table->tb6_lock));
1773                 if (child_left)
1774                         return rcu_dereference_protected(child_left->leaf,
1775                                         lockdep_is_held(&table->tb6_lock));
1776                 if (child_right)
1777                         return rcu_dereference_protected(child_right->leaf,
1778                                         lockdep_is_held(&table->tb6_lock));
1779
1780                 fn = FIB6_SUBTREE(fn);
1781         }
1782         return NULL;
1783 }
1784
1785 /*
1786  *      Called to trim the tree of intermediate nodes when possible. "fn"
1787  *      is the node we want to try and remove.
1788  *      Need to own table->tb6_lock
1789  */
1790
1791 static struct fib6_node *fib6_repair_tree(struct net *net,
1792                                           struct fib6_table *table,
1793                                           struct fib6_node *fn)
1794 {
1795         int children;
1796         int nstate;
1797         struct fib6_node *child;
1798         struct fib6_walker *w;
1799         int iter = 0;
1800
1801         /* Set fn->leaf to null_entry for root node. */
1802         if (fn->fn_flags & RTN_TL_ROOT) {
1803                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1804                 return fn;
1805         }
1806
1807         for (;;) {
1808                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1809                                             lockdep_is_held(&table->tb6_lock));
1810                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1811                                             lockdep_is_held(&table->tb6_lock));
1812                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1813                                             lockdep_is_held(&table->tb6_lock));
1814                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1815                                             lockdep_is_held(&table->tb6_lock));
1816                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1817                                             lockdep_is_held(&table->tb6_lock));
1818                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1819                                             lockdep_is_held(&table->tb6_lock));
1820                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1821                                             lockdep_is_held(&table->tb6_lock));
1822                 struct fib6_info *new_fn_leaf;
1823
1824                 pr_debug("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1825                 iter++;
1826
1827                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1828                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1829                 WARN_ON(fn_leaf);
1830
1831                 children = 0;
1832                 child = NULL;
1833                 if (fn_r) {
1834                         child = fn_r;
1835                         children |= 1;
1836                 }
1837                 if (fn_l) {
1838                         child = fn_l;
1839                         children |= 2;
1840                 }
1841
1842                 if (children == 3 || FIB6_SUBTREE(fn)
1843 #ifdef CONFIG_IPV6_SUBTREES
1844                     /* Subtree root (i.e. fn) may have one child */
1845                     || (children && fn->fn_flags & RTN_ROOT)
1846 #endif
1847                     ) {
1848                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1849 #if RT6_DEBUG >= 2
1850                         if (!new_fn_leaf) {
1851                                 WARN_ON(!new_fn_leaf);
1852                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1853                         }
1854 #endif
1855                         fib6_info_hold(new_fn_leaf);
1856                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1857                         return pn;
1858                 }
1859
1860 #ifdef CONFIG_IPV6_SUBTREES
1861                 if (FIB6_SUBTREE(pn) == fn) {
1862                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1863                         RCU_INIT_POINTER(pn->subtree, NULL);
1864                         nstate = FWS_L;
1865                 } else {
1866                         WARN_ON(fn->fn_flags & RTN_ROOT);
1867 #endif
1868                         if (pn_r == fn)
1869                                 rcu_assign_pointer(pn->right, child);
1870                         else if (pn_l == fn)
1871                                 rcu_assign_pointer(pn->left, child);
1872 #if RT6_DEBUG >= 2
1873                         else
1874                                 WARN_ON(1);
1875 #endif
1876                         if (child)
1877                                 rcu_assign_pointer(child->parent, pn);
1878                         nstate = FWS_R;
1879 #ifdef CONFIG_IPV6_SUBTREES
1880                 }
1881 #endif
1882
1883                 read_lock(&net->ipv6.fib6_walker_lock);
1884                 FOR_WALKERS(net, w) {
1885                         if (!child) {
1886                                 if (w->node == fn) {
1887                                         pr_debug("W %p adjusted by delnode 1, s=%d/%d\n",
1888                                                  w, w->state, nstate);
1889                                         w->node = pn;
1890                                         w->state = nstate;
1891                                 }
1892                         } else {
1893                                 if (w->node == fn) {
1894                                         w->node = child;
1895                                         if (children&2) {
1896                                                 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1897                                                          w, w->state);
1898                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1899                                         } else {
1900                                                 pr_debug("W %p adjusted by delnode 2, s=%d\n",
1901                                                          w, w->state);
1902                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1903                                         }
1904                                 }
1905                         }
1906                 }
1907                 read_unlock(&net->ipv6.fib6_walker_lock);
1908
1909                 node_free(net, fn);
1910                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1911                         return pn;
1912
1913                 RCU_INIT_POINTER(pn->leaf, NULL);
1914                 fib6_info_release(pn_leaf);
1915                 fn = pn;
1916         }
1917 }
1918
1919 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1920                            struct fib6_info __rcu **rtp, struct nl_info *info)
1921 {
1922         struct fib6_info *leaf, *replace_rt = NULL;
1923         struct fib6_walker *w;
1924         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1925                                     lockdep_is_held(&table->tb6_lock));
1926         struct net *net = info->nl_net;
1927         bool notify_del = false;
1928
1929         /* If the deleted route is the first in the node and it is not part of
1930          * a multipath route, then we need to replace it with the next route
1931          * in the node, if exists.
1932          */
1933         leaf = rcu_dereference_protected(fn->leaf,
1934                                          lockdep_is_held(&table->tb6_lock));
1935         if (leaf == rt && !rt->fib6_nsiblings) {
1936                 if (rcu_access_pointer(rt->fib6_next))
1937                         replace_rt = rcu_dereference_protected(rt->fib6_next,
1938                                             lockdep_is_held(&table->tb6_lock));
1939                 else
1940                         notify_del = true;
1941         }
1942
1943         /* Unlink it */
1944         *rtp = rt->fib6_next;
1945         rt->fib6_node = NULL;
1946         net->ipv6.rt6_stats->fib_rt_entries--;
1947         net->ipv6.rt6_stats->fib_discarded_routes++;
1948
1949         /* Reset round-robin state, if necessary */
1950         if (rcu_access_pointer(fn->rr_ptr) == rt)
1951                 fn->rr_ptr = NULL;
1952
1953         /* Remove this entry from other siblings */
1954         if (rt->fib6_nsiblings) {
1955                 struct fib6_info *sibling, *next_sibling;
1956
1957                 /* The route is deleted from a multipath route. If this
1958                  * multipath route is the first route in the node, then we need
1959                  * to emit a delete notification. Otherwise, we need to skip
1960                  * the notification.
1961                  */
1962                 if (rt->fib6_metric == leaf->fib6_metric &&
1963                     rt6_qualify_for_ecmp(leaf))
1964                         notify_del = true;
1965                 list_for_each_entry_safe(sibling, next_sibling,
1966                                          &rt->fib6_siblings, fib6_siblings)
1967                         sibling->fib6_nsiblings--;
1968                 rt->fib6_nsiblings = 0;
1969                 list_del_init(&rt->fib6_siblings);
1970                 rt6_multipath_rebalance(next_sibling);
1971         }
1972
1973         /* Adjust walkers */
1974         read_lock(&net->ipv6.fib6_walker_lock);
1975         FOR_WALKERS(net, w) {
1976                 if (w->state == FWS_C && w->leaf == rt) {
1977                         pr_debug("walker %p adjusted by delroute\n", w);
1978                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1979                                             lockdep_is_held(&table->tb6_lock));
1980                         if (!w->leaf)
1981                                 w->state = FWS_U;
1982                 }
1983         }
1984         read_unlock(&net->ipv6.fib6_walker_lock);
1985
1986         /* If it was last route, call fib6_repair_tree() to:
1987          * 1. For root node, put back null_entry as how the table was created.
1988          * 2. For other nodes, expunge its radix tree node.
1989          */
1990         if (!rcu_access_pointer(fn->leaf)) {
1991                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1992                         fn->fn_flags &= ~RTN_RTINFO;
1993                         net->ipv6.rt6_stats->fib_route_nodes--;
1994                 }
1995                 fn = fib6_repair_tree(net, table, fn);
1996         }
1997
1998         fib6_purge_rt(rt, fn, net);
1999
2000         if (!info->skip_notify_kernel) {
2001                 if (notify_del)
2002                         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
2003                                                   rt, NULL);
2004                 else if (replace_rt)
2005                         call_fib6_entry_notifiers_replace(net, replace_rt);
2006         }
2007         if (!info->skip_notify)
2008                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
2009
2010         fib6_info_release(rt);
2011 }
2012
2013 /* Need to own table->tb6_lock */
2014 int fib6_del(struct fib6_info *rt, struct nl_info *info)
2015 {
2016         struct net *net = info->nl_net;
2017         struct fib6_info __rcu **rtp;
2018         struct fib6_info __rcu **rtp_next;
2019         struct fib6_table *table;
2020         struct fib6_node *fn;
2021
2022         if (rt == net->ipv6.fib6_null_entry)
2023                 return -ENOENT;
2024
2025         table = rt->fib6_table;
2026         fn = rcu_dereference_protected(rt->fib6_node,
2027                                        lockdep_is_held(&table->tb6_lock));
2028         if (!fn)
2029                 return -ENOENT;
2030
2031         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
2032
2033         /*
2034          *      Walk the leaf entries looking for ourself
2035          */
2036
2037         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
2038                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
2039                                         lockdep_is_held(&table->tb6_lock));
2040                 if (rt == cur) {
2041                         if (fib6_requires_src(cur))
2042                                 fib6_routes_require_src_dec(info->nl_net);
2043                         fib6_del_route(table, fn, rtp, info);
2044                         return 0;
2045                 }
2046                 rtp_next = &cur->fib6_next;
2047         }
2048         return -ENOENT;
2049 }
2050
2051 /*
2052  *      Tree traversal function.
2053  *
2054  *      Certainly, it is not interrupt safe.
2055  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
2056  *      It means, that we can modify tree during walking
2057  *      and use this function for garbage collection, clone pruning,
2058  *      cleaning tree when a device goes down etc. etc.
2059  *
2060  *      It guarantees that every node will be traversed,
2061  *      and that it will be traversed only once.
2062  *
2063  *      Callback function w->func may return:
2064  *      0 -> continue walking.
2065  *      positive value -> walking is suspended (used by tree dumps,
2066  *      and probably by gc, if it will be split to several slices)
2067  *      negative value -> terminate walking.
2068  *
2069  *      The function itself returns:
2070  *      0   -> walk is complete.
2071  *      >0  -> walk is incomplete (i.e. suspended)
2072  *      <0  -> walk is terminated by an error.
2073  *
2074  *      This function is called with tb6_lock held.
2075  */
2076
2077 static int fib6_walk_continue(struct fib6_walker *w)
2078 {
2079         struct fib6_node *fn, *pn, *left, *right;
2080
2081         /* w->root should always be table->tb6_root */
2082         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
2083
2084         for (;;) {
2085                 fn = w->node;
2086                 if (!fn)
2087                         return 0;
2088
2089                 switch (w->state) {
2090 #ifdef CONFIG_IPV6_SUBTREES
2091                 case FWS_S:
2092                         if (FIB6_SUBTREE(fn)) {
2093                                 w->node = FIB6_SUBTREE(fn);
2094                                 continue;
2095                         }
2096                         w->state = FWS_L;
2097                         fallthrough;
2098 #endif
2099                 case FWS_L:
2100                         left = rcu_dereference_protected(fn->left, 1);
2101                         if (left) {
2102                                 w->node = left;
2103                                 w->state = FWS_INIT;
2104                                 continue;
2105                         }
2106                         w->state = FWS_R;
2107                         fallthrough;
2108                 case FWS_R:
2109                         right = rcu_dereference_protected(fn->right, 1);
2110                         if (right) {
2111                                 w->node = right;
2112                                 w->state = FWS_INIT;
2113                                 continue;
2114                         }
2115                         w->state = FWS_C;
2116                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
2117                         fallthrough;
2118                 case FWS_C:
2119                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
2120                                 int err;
2121
2122                                 if (w->skip) {
2123                                         w->skip--;
2124                                         goto skip;
2125                                 }
2126
2127                                 err = w->func(w);
2128                                 if (err)
2129                                         return err;
2130
2131                                 w->count++;
2132                                 continue;
2133                         }
2134 skip:
2135                         w->state = FWS_U;
2136                         fallthrough;
2137                 case FWS_U:
2138                         if (fn == w->root)
2139                                 return 0;
2140                         pn = rcu_dereference_protected(fn->parent, 1);
2141                         left = rcu_dereference_protected(pn->left, 1);
2142                         right = rcu_dereference_protected(pn->right, 1);
2143                         w->node = pn;
2144 #ifdef CONFIG_IPV6_SUBTREES
2145                         if (FIB6_SUBTREE(pn) == fn) {
2146                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
2147                                 w->state = FWS_L;
2148                                 continue;
2149                         }
2150 #endif
2151                         if (left == fn) {
2152                                 w->state = FWS_R;
2153                                 continue;
2154                         }
2155                         if (right == fn) {
2156                                 w->state = FWS_C;
2157                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
2158                                 continue;
2159                         }
2160 #if RT6_DEBUG >= 2
2161                         WARN_ON(1);
2162 #endif
2163                 }
2164         }
2165 }
2166
2167 static int fib6_walk(struct net *net, struct fib6_walker *w)
2168 {
2169         int res;
2170
2171         w->state = FWS_INIT;
2172         w->node = w->root;
2173
2174         fib6_walker_link(net, w);
2175         res = fib6_walk_continue(w);
2176         if (res <= 0)
2177                 fib6_walker_unlink(net, w);
2178         return res;
2179 }
2180
2181 static int fib6_clean_node(struct fib6_walker *w)
2182 {
2183         int res;
2184         struct fib6_info *rt;
2185         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
2186         struct nl_info info = {
2187                 .nl_net = c->net,
2188                 .skip_notify = c->skip_notify,
2189         };
2190
2191         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
2192             READ_ONCE(w->node->fn_sernum) != c->sernum)
2193                 WRITE_ONCE(w->node->fn_sernum, c->sernum);
2194
2195         if (!c->func) {
2196                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
2197                 w->leaf = NULL;
2198                 return 0;
2199         }
2200
2201         for_each_fib6_walker_rt(w) {
2202                 res = c->func(rt, c->arg);
2203                 if (res == -1) {
2204                         w->leaf = rt;
2205                         res = fib6_del(rt, &info);
2206                         if (res) {
2207 #if RT6_DEBUG >= 2
2208                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2209                                          __func__, rt,
2210                                          rcu_access_pointer(rt->fib6_node),
2211                                          res);
2212 #endif
2213                                 continue;
2214                         }
2215                         return 0;
2216                 } else if (res == -2) {
2217                         if (WARN_ON(!rt->fib6_nsiblings))
2218                                 continue;
2219                         rt = list_last_entry(&rt->fib6_siblings,
2220                                              struct fib6_info, fib6_siblings);
2221                         continue;
2222                 }
2223                 WARN_ON(res != 0);
2224         }
2225         w->leaf = rt;
2226         return 0;
2227 }
2228
2229 /*
2230  *      Convenient frontend to tree walker.
2231  *
2232  *      func is called on each route.
2233  *              It may return -2 -> skip multipath route.
2234  *                            -1 -> delete this route.
2235  *                            0  -> continue walking
2236  */
2237
2238 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2239                             int (*func)(struct fib6_info *, void *arg),
2240                             int sernum, void *arg, bool skip_notify)
2241 {
2242         struct fib6_cleaner c;
2243
2244         c.w.root = root;
2245         c.w.func = fib6_clean_node;
2246         c.w.count = 0;
2247         c.w.skip = 0;
2248         c.w.skip_in_node = 0;
2249         c.func = func;
2250         c.sernum = sernum;
2251         c.arg = arg;
2252         c.net = net;
2253         c.skip_notify = skip_notify;
2254
2255         fib6_walk(net, &c.w);
2256 }
2257
2258 static void __fib6_clean_all(struct net *net,
2259                              int (*func)(struct fib6_info *, void *),
2260                              int sernum, void *arg, bool skip_notify)
2261 {
2262         struct fib6_table *table;
2263         struct hlist_head *head;
2264         unsigned int h;
2265
2266         rcu_read_lock();
2267         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2268                 head = &net->ipv6.fib_table_hash[h];
2269                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2270                         spin_lock_bh(&table->tb6_lock);
2271                         fib6_clean_tree(net, &table->tb6_root,
2272                                         func, sernum, arg, skip_notify);
2273                         spin_unlock_bh(&table->tb6_lock);
2274                 }
2275         }
2276         rcu_read_unlock();
2277 }
2278
2279 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2280                     void *arg)
2281 {
2282         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2283 }
2284
2285 void fib6_clean_all_skip_notify(struct net *net,
2286                                 int (*func)(struct fib6_info *, void *),
2287                                 void *arg)
2288 {
2289         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2290 }
2291
2292 static void fib6_flush_trees(struct net *net)
2293 {
2294         int new_sernum = fib6_new_sernum(net);
2295
2296         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2297 }
2298
2299 /*
2300  *      Garbage collection
2301  */
2302
2303 static int fib6_age(struct fib6_info *rt, struct fib6_gc_args *gc_args)
2304 {
2305         unsigned long now = jiffies;
2306
2307         /*
2308          *      check addrconf expiration here.
2309          *      Routes are expired even if they are in use.
2310          */
2311
2312         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2313                 if (time_after(now, rt->expires)) {
2314                         pr_debug("expiring %p\n", rt);
2315                         return -1;
2316                 }
2317                 gc_args->more++;
2318         }
2319
2320         /*      Also age clones in the exception table.
2321          *      Note, that clones are aged out
2322          *      only if they are not in use now.
2323          */
2324         rt6_age_exceptions(rt, gc_args, now);
2325
2326         return 0;
2327 }
2328
2329 static void fib6_gc_table(struct net *net,
2330                           struct fib6_table *tb6,
2331                           struct fib6_gc_args *gc_args)
2332 {
2333         struct fib6_info *rt;
2334         struct hlist_node *n;
2335         struct nl_info info = {
2336                 .nl_net = net,
2337                 .skip_notify = false,
2338         };
2339
2340         hlist_for_each_entry_safe(rt, n, &tb6->tb6_gc_hlist, gc_link)
2341                 if (fib6_age(rt, gc_args) == -1)
2342                         fib6_del(rt, &info);
2343 }
2344
2345 static void fib6_gc_all(struct net *net, struct fib6_gc_args *gc_args)
2346 {
2347         struct fib6_table *table;
2348         struct hlist_head *head;
2349         unsigned int h;
2350
2351         rcu_read_lock();
2352         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2353                 head = &net->ipv6.fib_table_hash[h];
2354                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2355                         spin_lock_bh(&table->tb6_lock);
2356
2357                         fib6_gc_table(net, table, gc_args);
2358
2359                         spin_unlock_bh(&table->tb6_lock);
2360                 }
2361         }
2362         rcu_read_unlock();
2363 }
2364
2365 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2366 {
2367         struct fib6_gc_args gc_args;
2368         unsigned long now;
2369
2370         if (force) {
2371                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2372         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2373                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2374                 return;
2375         }
2376         gc_args.timeout = expires ? (int)expires :
2377                           net->ipv6.sysctl.ip6_rt_gc_interval;
2378         gc_args.more = 0;
2379
2380         fib6_gc_all(net, &gc_args);
2381         now = jiffies;
2382         net->ipv6.ip6_rt_last_gc = now;
2383
2384         if (gc_args.more)
2385                 mod_timer(&net->ipv6.ip6_fib_timer,
2386                           round_jiffies(now
2387                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2388         else
2389                 del_timer(&net->ipv6.ip6_fib_timer);
2390         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2391 }
2392
2393 static void fib6_gc_timer_cb(struct timer_list *t)
2394 {
2395         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2396
2397         fib6_run_gc(0, arg, true);
2398 }
2399
2400 static int __net_init fib6_net_init(struct net *net)
2401 {
2402         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2403         int err;
2404
2405         err = fib6_notifier_init(net);
2406         if (err)
2407                 return err;
2408
2409         /* Default to 3-tuple */
2410         net->ipv6.sysctl.multipath_hash_fields =
2411                 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
2412
2413         spin_lock_init(&net->ipv6.fib6_gc_lock);
2414         rwlock_init(&net->ipv6.fib6_walker_lock);
2415         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2416         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2417
2418         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2419         if (!net->ipv6.rt6_stats)
2420                 goto out_notifier;
2421
2422         /* Avoid false sharing : Use at least a full cache line */
2423         size = max_t(size_t, size, L1_CACHE_BYTES);
2424
2425         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2426         if (!net->ipv6.fib_table_hash)
2427                 goto out_rt6_stats;
2428
2429         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2430                                           GFP_KERNEL);
2431         if (!net->ipv6.fib6_main_tbl)
2432                 goto out_fib_table_hash;
2433
2434         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2435         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2436                            net->ipv6.fib6_null_entry);
2437         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2438                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2439         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2440         INIT_HLIST_HEAD(&net->ipv6.fib6_main_tbl->tb6_gc_hlist);
2441
2442 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2443         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2444                                            GFP_KERNEL);
2445         if (!net->ipv6.fib6_local_tbl)
2446                 goto out_fib6_main_tbl;
2447         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2448         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2449                            net->ipv6.fib6_null_entry);
2450         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2451                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2452         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2453         INIT_HLIST_HEAD(&net->ipv6.fib6_local_tbl->tb6_gc_hlist);
2454 #endif
2455         fib6_tables_init(net);
2456
2457         return 0;
2458
2459 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2460 out_fib6_main_tbl:
2461         kfree(net->ipv6.fib6_main_tbl);
2462 #endif
2463 out_fib_table_hash:
2464         kfree(net->ipv6.fib_table_hash);
2465 out_rt6_stats:
2466         kfree(net->ipv6.rt6_stats);
2467 out_notifier:
2468         fib6_notifier_exit(net);
2469         return -ENOMEM;
2470 }
2471
2472 static void fib6_net_exit(struct net *net)
2473 {
2474         unsigned int i;
2475
2476         del_timer_sync(&net->ipv6.ip6_fib_timer);
2477
2478         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2479                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2480                 struct hlist_node *tmp;
2481                 struct fib6_table *tb;
2482
2483                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2484                         hlist_del(&tb->tb6_hlist);
2485                         fib6_free_table(tb);
2486                 }
2487         }
2488
2489         kfree(net->ipv6.fib_table_hash);
2490         kfree(net->ipv6.rt6_stats);
2491         fib6_notifier_exit(net);
2492 }
2493
2494 static struct pernet_operations fib6_net_ops = {
2495         .init = fib6_net_init,
2496         .exit = fib6_net_exit,
2497 };
2498
2499 int __init fib6_init(void)
2500 {
2501         int ret = -ENOMEM;
2502
2503         fib6_node_kmem = KMEM_CACHE(fib6_node,
2504                                     SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT);
2505         if (!fib6_node_kmem)
2506                 goto out;
2507
2508         ret = register_pernet_subsys(&fib6_net_ops);
2509         if (ret)
2510                 goto out_kmem_cache_create;
2511
2512         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2513                                    inet6_dump_fib, RTNL_FLAG_DUMP_UNLOCKED);
2514         if (ret)
2515                 goto out_unregister_subsys;
2516
2517         __fib6_flush_trees = fib6_flush_trees;
2518 out:
2519         return ret;
2520
2521 out_unregister_subsys:
2522         unregister_pernet_subsys(&fib6_net_ops);
2523 out_kmem_cache_create:
2524         kmem_cache_destroy(fib6_node_kmem);
2525         goto out;
2526 }
2527
2528 void fib6_gc_cleanup(void)
2529 {
2530         unregister_pernet_subsys(&fib6_net_ops);
2531         kmem_cache_destroy(fib6_node_kmem);
2532 }
2533
2534 #ifdef CONFIG_PROC_FS
2535 static int ipv6_route_native_seq_show(struct seq_file *seq, void *v)
2536 {
2537         struct fib6_info *rt = v;
2538         struct ipv6_route_iter *iter = seq->private;
2539         struct fib6_nh *fib6_nh = rt->fib6_nh;
2540         unsigned int flags = rt->fib6_flags;
2541         const struct net_device *dev;
2542
2543         if (rt->nh)
2544                 fib6_nh = nexthop_fib6_nh(rt->nh);
2545
2546         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2547
2548 #ifdef CONFIG_IPV6_SUBTREES
2549         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2550 #else
2551         seq_puts(seq, "00000000000000000000000000000000 00 ");
2552 #endif
2553         if (fib6_nh->fib_nh_gw_family) {
2554                 flags |= RTF_GATEWAY;
2555                 seq_printf(seq, "%pi6", &fib6_nh->fib_nh_gw6);
2556         } else {
2557                 seq_puts(seq, "00000000000000000000000000000000");
2558         }
2559
2560         dev = fib6_nh->fib_nh_dev;
2561         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2562                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2563                    flags, dev ? dev->name : "");
2564         iter->w.leaf = NULL;
2565         return 0;
2566 }
2567
2568 static int ipv6_route_yield(struct fib6_walker *w)
2569 {
2570         struct ipv6_route_iter *iter = w->args;
2571
2572         if (!iter->skip)
2573                 return 1;
2574
2575         do {
2576                 iter->w.leaf = rcu_dereference_protected(
2577                                 iter->w.leaf->fib6_next,
2578                                 lockdep_is_held(&iter->tbl->tb6_lock));
2579                 iter->skip--;
2580                 if (!iter->skip && iter->w.leaf)
2581                         return 1;
2582         } while (iter->w.leaf);
2583
2584         return 0;
2585 }
2586
2587 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2588                                       struct net *net)
2589 {
2590         memset(&iter->w, 0, sizeof(iter->w));
2591         iter->w.func = ipv6_route_yield;
2592         iter->w.root = &iter->tbl->tb6_root;
2593         iter->w.state = FWS_INIT;
2594         iter->w.node = iter->w.root;
2595         iter->w.args = iter;
2596         iter->sernum = READ_ONCE(iter->w.root->fn_sernum);
2597         INIT_LIST_HEAD(&iter->w.lh);
2598         fib6_walker_link(net, &iter->w);
2599 }
2600
2601 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2602                                                     struct net *net)
2603 {
2604         unsigned int h;
2605         struct hlist_node *node;
2606
2607         if (tbl) {
2608                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2609                 node = rcu_dereference(hlist_next_rcu(&tbl->tb6_hlist));
2610         } else {
2611                 h = 0;
2612                 node = NULL;
2613         }
2614
2615         while (!node && h < FIB6_TABLE_HASHSZ) {
2616                 node = rcu_dereference(
2617                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2618         }
2619         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2620 }
2621
2622 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2623 {
2624         int sernum = READ_ONCE(iter->w.root->fn_sernum);
2625
2626         if (iter->sernum != sernum) {
2627                 iter->sernum = sernum;
2628                 iter->w.state = FWS_INIT;
2629                 iter->w.node = iter->w.root;
2630                 WARN_ON(iter->w.skip);
2631                 iter->w.skip = iter->w.count;
2632         }
2633 }
2634
2635 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2636 {
2637         int r;
2638         struct fib6_info *n;
2639         struct net *net = seq_file_net(seq);
2640         struct ipv6_route_iter *iter = seq->private;
2641
2642         ++(*pos);
2643         if (!v)
2644                 goto iter_table;
2645
2646         n = rcu_dereference(((struct fib6_info *)v)->fib6_next);
2647         if (n)
2648                 return n;
2649
2650 iter_table:
2651         ipv6_route_check_sernum(iter);
2652         spin_lock_bh(&iter->tbl->tb6_lock);
2653         r = fib6_walk_continue(&iter->w);
2654         spin_unlock_bh(&iter->tbl->tb6_lock);
2655         if (r > 0) {
2656                 return iter->w.leaf;
2657         } else if (r < 0) {
2658                 fib6_walker_unlink(net, &iter->w);
2659                 return NULL;
2660         }
2661         fib6_walker_unlink(net, &iter->w);
2662
2663         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2664         if (!iter->tbl)
2665                 return NULL;
2666
2667         ipv6_route_seq_setup_walk(iter, net);
2668         goto iter_table;
2669 }
2670
2671 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2672         __acquires(RCU)
2673 {
2674         struct net *net = seq_file_net(seq);
2675         struct ipv6_route_iter *iter = seq->private;
2676
2677         rcu_read_lock();
2678         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2679         iter->skip = *pos;
2680
2681         if (iter->tbl) {
2682                 loff_t p = 0;
2683
2684                 ipv6_route_seq_setup_walk(iter, net);
2685                 return ipv6_route_seq_next(seq, NULL, &p);
2686         } else {
2687                 return NULL;
2688         }
2689 }
2690
2691 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2692 {
2693         struct fib6_walker *w = &iter->w;
2694         return w->node && !(w->state == FWS_U && w->node == w->root);
2695 }
2696
2697 static void ipv6_route_native_seq_stop(struct seq_file *seq, void *v)
2698         __releases(RCU)
2699 {
2700         struct net *net = seq_file_net(seq);
2701         struct ipv6_route_iter *iter = seq->private;
2702
2703         if (ipv6_route_iter_active(iter))
2704                 fib6_walker_unlink(net, &iter->w);
2705
2706         rcu_read_unlock();
2707 }
2708
2709 #if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
2710 static int ipv6_route_prog_seq_show(struct bpf_prog *prog,
2711                                     struct bpf_iter_meta *meta,
2712                                     void *v)
2713 {
2714         struct bpf_iter__ipv6_route ctx;
2715
2716         ctx.meta = meta;
2717         ctx.rt = v;
2718         return bpf_iter_run_prog(prog, &ctx);
2719 }
2720
2721 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2722 {
2723         struct ipv6_route_iter *iter = seq->private;
2724         struct bpf_iter_meta meta;
2725         struct bpf_prog *prog;
2726         int ret;
2727
2728         meta.seq = seq;
2729         prog = bpf_iter_get_info(&meta, false);
2730         if (!prog)
2731                 return ipv6_route_native_seq_show(seq, v);
2732
2733         ret = ipv6_route_prog_seq_show(prog, &meta, v);
2734         iter->w.leaf = NULL;
2735
2736         return ret;
2737 }
2738
2739 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2740 {
2741         struct bpf_iter_meta meta;
2742         struct bpf_prog *prog;
2743
2744         if (!v) {
2745                 meta.seq = seq;
2746                 prog = bpf_iter_get_info(&meta, true);
2747                 if (prog)
2748                         (void)ipv6_route_prog_seq_show(prog, &meta, v);
2749         }
2750
2751         ipv6_route_native_seq_stop(seq, v);
2752 }
2753 #else
2754 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2755 {
2756         return ipv6_route_native_seq_show(seq, v);
2757 }
2758
2759 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2760 {
2761         ipv6_route_native_seq_stop(seq, v);
2762 }
2763 #endif
2764
2765 const struct seq_operations ipv6_route_seq_ops = {
2766         .start  = ipv6_route_seq_start,
2767         .next   = ipv6_route_seq_next,
2768         .stop   = ipv6_route_seq_stop,
2769         .show   = ipv6_route_seq_show
2770 };
2771 #endif /* CONFIG_PROC_FS */