net/ipv6/ip6_fib.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      Forwarding Information Database
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Changes:
  14  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  15  *                              remove ip6_null_entry from the top of
  16  *                              routing table.
  17  *      Ville Nuorvala:         Fixed routing subtrees.
  18  */
  19
  20 #define pr_fmt(fmt) "IPv6: " fmt
  21
  22 #include <linux/errno.h>
  23 #include <linux/types.h>
  24 #include <linux/net.h>
  25 #include <linux/route.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/in6.h>
  28 #include <linux/init.h>
  29 #include <linux/list.h>
  30 #include <linux/slab.h>
  31
  32 #include <net/ipv6.h>
  33 #include <net/ndisc.h>
  34 #include <net/addrconf.h>
  35
  36 #include <net/ip6_fib.h>
  37 #include <net/ip6_route.h>
  38
  39 #define RT6_DEBUG 2
  40
  41 #if RT6_DEBUG >= 3
  42 #define RT6_TRACE(x...) pr_debug(x)
  43 #else
  44 #define RT6_TRACE(x...) do { ; } while (0)
  45 #endif
  46
  47 static struct kmem_cache *fib6_node_kmem __read_mostly;
  48
  49 struct fib6_cleaner {
  50         struct fib6_walker w;
  51         struct net *net;
  52         int (*func)(struct rt6_info *, void *arg);
  53         int sernum;
  54         void *arg;
  55 };
  56
  57 static DEFINE_RWLOCK(fib6_walker_lock);
  58
  59 #ifdef CONFIG_IPV6_SUBTREES
  60 #define FWS_INIT FWS_S
  61 #else
  62 #define FWS_INIT FWS_L
  63 #endif
  64
  65 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
  66 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
  67 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
  68 static int fib6_walk(struct fib6_walker *w);
  69 static int fib6_walk_continue(struct fib6_walker *w);
  70
  71 /*
  72  *      A routing update causes an increase of the serial number on the
  73  *      affected subtree. This allows for cached routes to be asynchronously
  74  *      tested when modifications are made to the destination cache as a
  75  *      result of redirects, path MTU changes, etc.
  76  */
  77
  78 static void fib6_gc_timer_cb(unsigned long arg);
  79
  80 static LIST_HEAD(fib6_walkers);
  81 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
  82
  83 static void fib6_walker_link(struct fib6_walker *w)
  84 {
  85         write_lock_bh(&fib6_walker_lock);
  86         list_add(&w->lh, &fib6_walkers);
  87         write_unlock_bh(&fib6_walker_lock);
  88 }
  89
  90 static void fib6_walker_unlink(struct fib6_walker *w)
  91 {
  92         write_lock_bh(&fib6_walker_lock);
  93         list_del(&w->lh);
  94         write_unlock_bh(&fib6_walker_lock);
  95 }
  96
  97 static int fib6_new_sernum(struct net *net)
  98 {
  99         int new, old;
 100
 101         do {
 102                 old = atomic_read(&net->ipv6.fib6_sernum);
 103                 new = old < INT_MAX ? old + 1 : 1;
 104         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 105                                 old, new) != old);
 106         return new;
 107 }
 108
 109 enum {
 110         FIB6_NO_SERNUM_CHANGE = 0,
 111 };
 112
 113 /*
 114  *      Auxiliary address test functions for the radix tree.
 115  *
 116  *      These assume a 32bit processor (although it will work on
 117  *      64bit processors)
 118  */
 119
 120 /*
 121  *      test bit
 122  */
 123 #if defined(__LITTLE_ENDIAN)
 124 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 125 #else
 126 # define BITOP_BE32_SWIZZLE     0
 127 #endif
 128
 129 static __be32 addr_bit_set(const void *token, int fn_bit)
 130 {
 131         const __be32 *addr = token;
 132         /*
 133          * Here,
 134          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 135          * is optimized version of
 136          *      htonl(1 << ((~fn_bit)&0x1F))
 137          * See include/asm-generic/bitops/le.h.
 138          */
 139         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 140                addr[fn_bit >> 5];
 141 }
 142
 143 static struct fib6_node *node_alloc(void)
 144 {
 145         struct fib6_node *fn;
 146
 147         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 148
 149         return fn;
 150 }
 151
 152 static void node_free(struct fib6_node *fn)
 153 {
 154         kmem_cache_free(fib6_node_kmem, fn);
 155 }
 156
 157 static void rt6_release(struct rt6_info *rt)
 158 {
 159         if (atomic_dec_and_test(&rt->rt6i_ref))
 160                 dst_free(&rt->dst);
 161 }
 162
 163 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 164 {
 165         unsigned int h;
 166
 167         /*
 168          * Initialize table lock at a single place to give lockdep a key,
 169          * tables aren't visible prior to being linked to the list.
 170          */
 171         rwlock_init(&tb->tb6_lock);
 172
 173         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 174
 175         /*
 176          * No protection necessary, this is the only list mutatation
 177          * operation, tables never disappear once they exist.
 178          */
 179         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 180 }
 181
 182 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 183
 184 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 185 {
 186         struct fib6_table *table;
 187
 188         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 189         if (table) {
 190                 table->tb6_id = id;
 191                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
 192                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 193                 inet_peer_base_init(&table->tb6_peers);
 194         }
 195
 196         return table;
 197 }
 198
 199 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 200 {
 201         struct fib6_table *tb;
 202
 203         if (id == 0)
 204                 id = RT6_TABLE_MAIN;
 205         tb = fib6_get_table(net, id);
 206         if (tb)
 207                 return tb;
 208
 209         tb = fib6_alloc_table(net, id);
 210         if (tb)
 211                 fib6_link_table(net, tb);
 212
 213         return tb;
 214 }
 215
 216 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 217 {
 218         struct fib6_table *tb;
 219         struct hlist_head *head;
 220         unsigned int h;
 221
 222         if (id == 0)
 223                 id = RT6_TABLE_MAIN;
 224         h = id & (FIB6_TABLE_HASHSZ - 1);
 225         rcu_read_lock();
 226         head = &net->ipv6.fib_table_hash[h];
 227         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 228                 if (tb->tb6_id == id) {
 229                         rcu_read_unlock();
 230                         return tb;
 231                 }
 232         }
 233         rcu_read_unlock();
 234
 235         return NULL;
 236 }
 237
 238 static void __net_init fib6_tables_init(struct net *net)
 239 {
 240         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 241         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 242 }
 243 #else
 244
 245 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 246 {
 247         return fib6_get_table(net, id);
 248 }
 249
 250 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 251 {
 252           return net->ipv6.fib6_main_tbl;
 253 }
 254
 255 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 256                                    int flags, pol_lookup_t lookup)
 257 {
 258         return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
 259 }
 260
 261 static void __net_init fib6_tables_init(struct net *net)
 262 {
 263         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 264 }
 265
 266 #endif
 267
 268 static int fib6_dump_node(struct fib6_walker *w)
 269 {
 270         int res;
 271         struct rt6_info *rt;
 272
 273         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
 274                 res = rt6_dump_route(rt, w->args);
 275                 if (res < 0) {
 276                         /* Frame is full, suspend walking */
 277                         w->leaf = rt;
 278                         return 1;
 279                 }
 280         }
 281         w->leaf = NULL;
 282         return 0;
 283 }
 284
 285 static void fib6_dump_end(struct netlink_callback *cb)
 286 {
 287         struct fib6_walker *w = (void *)cb->args[2];
 288
 289         if (w) {
 290                 if (cb->args[4]) {
 291                         cb->args[4] = 0;
 292                         fib6_walker_unlink(w);
 293                 }
 294                 cb->args[2] = 0;
 295                 kfree(w);
 296         }
 297         cb->done = (void *)cb->args[3];
 298         cb->args[1] = 3;
 299 }
 300
 301 static int fib6_dump_done(struct netlink_callback *cb)
 302 {
 303         fib6_dump_end(cb);
 304         return cb->done ? cb->done(cb) : 0;
 305 }
 306
 307 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 308                            struct netlink_callback *cb)
 309 {
 310         struct fib6_walker *w;
 311         int res;
 312
 313         w = (void *)cb->args[2];
 314         w->root = &table->tb6_root;
 315
 316         if (cb->args[4] == 0) {
 317                 w->count = 0;
 318                 w->skip = 0;
 319
 320                 read_lock_bh(&table->tb6_lock);
 321                 res = fib6_walk(w);
 322                 read_unlock_bh(&table->tb6_lock);
 323                 if (res > 0) {
 324                         cb->args[4] = 1;
 325                         cb->args[5] = w->root->fn_sernum;
 326                 }
 327         } else {
 328                 if (cb->args[5] != w->root->fn_sernum) {
 329                         /* Begin at the root if the tree changed */
 330                         cb->args[5] = w->root->fn_sernum;
 331                         w->state = FWS_INIT;
 332                         w->node = w->root;
 333                         w->skip = w->count;
 334                 } else
 335                         w->skip = 0;
 336
 337                 read_lock_bh(&table->tb6_lock);
 338                 res = fib6_walk_continue(w);
 339                 read_unlock_bh(&table->tb6_lock);
 340                 if (res <= 0) {
 341                         fib6_walker_unlink(w);
 342                         cb->args[4] = 0;
 343                 }
 344         }
 345
 346         return res;
 347 }
 348
 349 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 350 {
 351         struct net *net = sock_net(skb->sk);
 352         unsigned int h, s_h;
 353         unsigned int e = 0, s_e;
 354         struct rt6_rtnl_dump_arg arg;
 355         struct fib6_walker *w;
 356         struct fib6_table *tb;
 357         struct hlist_head *head;
 358         int res = 0;
 359
 360         s_h = cb->args[0];
 361         s_e = cb->args[1];
 362
 363         w = (void *)cb->args[2];
 364         if (!w) {
 365                 /* New dump:
 366                  *
 367                  * 1. hook callback destructor.
 368                  */
 369                 cb->args[3] = (long)cb->done;
 370                 cb->done = fib6_dump_done;
 371
 372                 /*
 373                  * 2. allocate and initialize walker.
 374                  */
 375                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 376                 if (!w)
 377                         return -ENOMEM;
 378                 w->func = fib6_dump_node;
 379                 cb->args[2] = (long)w;
 380         }
 381
 382         arg.skb = skb;
 383         arg.cb = cb;
 384         arg.net = net;
 385         w->args = &arg;
 386
 387         rcu_read_lock();
 388         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 389                 e = 0;
 390                 head = &net->ipv6.fib_table_hash[h];
 391                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 392                         if (e < s_e)
 393                                 goto next;
 394                         res = fib6_dump_table(tb, skb, cb);
 395                         if (res != 0)
 396                                 goto out;
 397 next:
 398                         e++;
 399                 }
 400         }
 401 out:
 402         rcu_read_unlock();
 403         cb->args[1] = e;
 404         cb->args[0] = h;
 405
 406         res = res < 0 ? res : skb->len;
 407         if (res <= 0)
 408                 fib6_dump_end(cb);
 409         return res;
 410 }
 411
 412 /*
 413  *      Routing Table
 414  *
 415  *      return the appropriate node for a routing tree "add" operation
 416  *      by either creating and inserting or by returning an existing
 417  *      node.
 418  */
 419
 420 static struct fib6_node *fib6_add_1(struct fib6_node *root,
 421                                      struct in6_addr *addr, int plen,
 422                                      int offset, int allow_create,
 423                                      int replace_required, int sernum)
 424 {
 425         struct fib6_node *fn, *in, *ln;
 426         struct fib6_node *pn = NULL;
 427         struct rt6key *key;
 428         int     bit;
 429         __be32  dir = 0;
 430
 431         RT6_TRACE("fib6_add_1\n");
 432
 433         /* insert node in tree */
 434
 435         fn = root;
 436
 437         do {
 438                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
 439
 440                 /*
 441                  *      Prefix match
 442                  */
 443                 if (plen < fn->fn_bit ||
 444                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 445                         if (!allow_create) {
 446                                 if (replace_required) {
 447                                         pr_warn("Can't replace route, no match found\n");
 448                                         return ERR_PTR(-ENOENT);
 449                                 }
 450                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 451                         }
 452                         goto insert_above;
 453                 }
 454
 455                 /*
 456                  *      Exact match ?
 457                  */
 458
 459                 if (plen == fn->fn_bit) {
 460                         /* clean up an intermediate node */
 461                         if (!(fn->fn_flags & RTN_RTINFO)) {
 462                                 rt6_release(fn->leaf);
 463                                 fn->leaf = NULL;
 464                         }
 465
 466                         fn->fn_sernum = sernum;
 467
 468                         return fn;
 469                 }
 470
 471                 /*
 472                  *      We have more bits to go
 473                  */
 474
 475                 /* Try to walk down on tree. */
 476                 fn->fn_sernum = sernum;
 477                 dir = addr_bit_set(addr, fn->fn_bit);
 478                 pn = fn;
 479                 fn = dir ? fn->right : fn->left;
 480         } while (fn);
 481
 482         if (!allow_create) {
 483                 /* We should not create new node because
 484                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 485                  * I assume it is safe to require NLM_F_CREATE when
 486                  * REPLACE flag is used! Later we may want to remove the
 487                  * check for replace_required, because according
 488                  * to netlink specification, NLM_F_CREATE
 489                  * MUST be specified if new route is created.
 490                  * That would keep IPv6 consistent with IPv4
 491                  */
 492                 if (replace_required) {
 493                         pr_warn("Can't replace route, no match found\n");
 494                         return ERR_PTR(-ENOENT);
 495                 }
 496                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 497         }
 498         /*
 499          *      We walked to the bottom of tree.
 500          *      Create new leaf node without children.
 501          */
 502
 503         ln = node_alloc();
 504
 505         if (!ln)
 506                 return ERR_PTR(-ENOMEM);
 507         ln->fn_bit = plen;
 508
 509         ln->parent = pn;
 510         ln->fn_sernum = sernum;
 511
 512         if (dir)
 513                 pn->right = ln;
 514         else
 515                 pn->left  = ln;
 516
 517         return ln;
 518
 519
 520 insert_above:
 521         /*
 522          * split since we don't have a common prefix anymore or
 523          * we have a less significant route.
 524          * we've to insert an intermediate node on the list
 525          * this new node will point to the one we need to create
 526          * and the current
 527          */
 528
 529         pn = fn->parent;
 530
 531         /* find 1st bit in difference between the 2 addrs.
 532
 533            See comment in __ipv6_addr_diff: bit may be an invalid value,
 534            but if it is >= plen, the value is ignored in any case.
 535          */
 536
 537         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 538
 539         /*
 540          *              (intermediate)[in]
 541          *                /        \
 542          *      (new leaf node)[ln] (old node)[fn]
 543          */
 544         if (plen > bit) {
 545                 in = node_alloc();
 546                 ln = node_alloc();
 547
 548                 if (!in || !ln) {
 549                         if (in)
 550                                 node_free(in);
 551                         if (ln)
 552                                 node_free(ln);
 553                         return ERR_PTR(-ENOMEM);
 554                 }
 555
 556                 /*
 557                  * new intermediate node.
 558                  * RTN_RTINFO will
 559                  * be off since that an address that chooses one of
 560                  * the branches would not match less specific routes
 561                  * in the other branch
 562                  */
 563
 564                 in->fn_bit = bit;
 565
 566                 in->parent = pn;
 567                 in->leaf = fn->leaf;
 568                 atomic_inc(&in->leaf->rt6i_ref);
 569
 570                 in->fn_sernum = sernum;
 571
 572                 /* update parent pointer */
 573                 if (dir)
 574                         pn->right = in;
 575                 else
 576                         pn->left  = in;
 577
 578                 ln->fn_bit = plen;
 579
 580                 ln->parent = in;
 581                 fn->parent = in;
 582
 583                 ln->fn_sernum = sernum;
 584
 585                 if (addr_bit_set(addr, bit)) {
 586                         in->right = ln;
 587                         in->left  = fn;
 588                 } else {
 589                         in->left  = ln;
 590                         in->right = fn;
 591                 }
 592         } else { /* plen <= bit */
 593
 594                 /*
 595                  *              (new leaf node)[ln]
 596                  *                /        \
 597                  *           (old node)[fn] NULL
 598                  */
 599
 600                 ln = node_alloc();
 601
 602                 if (!ln)
 603                         return ERR_PTR(-ENOMEM);
 604
 605                 ln->fn_bit = plen;
 606
 607                 ln->parent = pn;
 608
 609                 ln->fn_sernum = sernum;
 610
 611                 if (dir)
 612                         pn->right = ln;
 613                 else
 614                         pn->left  = ln;
 615
 616                 if (addr_bit_set(&key->addr, plen))
 617                         ln->right = fn;
 618                 else
 619                         ln->left  = fn;
 620
 621                 fn->parent = ln;
 622         }
 623         return ln;
 624 }
 625
 626 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 627 {
 628         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 629                RTF_GATEWAY;
 630 }
 631
 632 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 633 {
 634         int i;
 635
 636         for (i = 0; i < RTAX_MAX; i++) {
 637                 if (test_bit(i, mxc->mx_valid))
 638                         mp[i] = mxc->mx[i];
 639         }
 640 }
 641
 642 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 643 {
 644         if (!mxc->mx)
 645                 return 0;
 646
 647         if (dst->flags & DST_HOST) {
 648                 u32 *mp = dst_metrics_write_ptr(dst);
 649
 650                 if (unlikely(!mp))
 651                         return -ENOMEM;
 652
 653                 fib6_copy_metrics(mp, mxc);
 654         } else {
 655                 dst_init_metrics(dst, mxc->mx, false);
 656
 657                 /* We've stolen mx now. */
 658                 mxc->mx = NULL;
 659         }
 660
 661         return 0;
 662 }
 663
 664 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 665                           struct net *net)
 666 {
 667         if (atomic_read(&rt->rt6i_ref) != 1) {
 668                 /* This route is used as dummy address holder in some split
 669                  * nodes. It is not leaked, but it still holds other resources,
 670                  * which must be released in time. So, scan ascendant nodes
 671                  * and replace dummy references to this route with references
 672                  * to still alive ones.
 673                  */
 674                 while (fn) {
 675                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
 676                                 fn->leaf = fib6_find_prefix(net, fn);
 677                                 atomic_inc(&fn->leaf->rt6i_ref);
 678                                 rt6_release(rt);
 679                         }
 680                         fn = fn->parent;
 681                 }
 682                 /* No more references are possible at this point. */
 683                 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
 684         }
 685 }
 686
 687 /*
 688  *      Insert routing information in a node.
 689  */
 690
 691 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 692                             struct nl_info *info, struct mx6_config *mxc)
 693 {
 694         struct rt6_info *iter = NULL;
 695         struct rt6_info **ins;
 696         int replace = (info->nlh &&
 697                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 698         int add = (!info->nlh ||
 699                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 700         int found = 0;
 701         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 702         int err;
 703
 704         ins = &fn->leaf;
 705
 706         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 707                 /*
 708                  *      Search for duplicates
 709                  */
 710
 711                 if (iter->rt6i_metric == rt->rt6i_metric) {
 712                         /*
 713                          *      Same priority level
 714                          */
 715                         if (info->nlh &&
 716                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 717                                 return -EEXIST;
 718                         if (replace) {
 719                                 found++;
 720                                 break;
 721                         }
 722
 723                         if (iter->dst.dev == rt->dst.dev &&
 724                             iter->rt6i_idev == rt->rt6i_idev &&
 725                             ipv6_addr_equal(&iter->rt6i_gateway,
 726                                             &rt->rt6i_gateway)) {
 727                                 if (rt->rt6i_nsiblings)
 728                                         rt->rt6i_nsiblings = 0;
 729                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 730                                         return -EEXIST;
 731                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 732                                         rt6_clean_expires(iter);
 733                                 else
 734                                         rt6_set_expires(iter, rt->dst.expires);
 735                                 return -EEXIST;
 736                         }
 737                         /* If we have the same destination and the same metric,
 738                          * but not the same gateway, then the route we try to
 739                          * add is sibling to this route, increment our counter
 740                          * of siblings, and later we will add our route to the
 741                          * list.
 742                          * Only static routes (which don't have flag
 743                          * RTF_EXPIRES) are used for ECMPv6.
 744                          *
 745                          * To avoid long list, we only had siblings if the
 746                          * route have a gateway.
 747                          */
 748                         if (rt_can_ecmp &&
 749                             rt6_qualify_for_ecmp(iter))
 750                                 rt->rt6i_nsiblings++;
 751                 }
 752
 753                 if (iter->rt6i_metric > rt->rt6i_metric)
 754                         break;
 755
 756                 ins = &iter->dst.rt6_next;
 757         }
 758
 759         /* Reset round-robin state, if necessary */
 760         if (ins == &fn->leaf)
 761                 fn->rr_ptr = NULL;
 762
 763         /* Link this route to others same route. */
 764         if (rt->rt6i_nsiblings) {
 765                 unsigned int rt6i_nsiblings;
 766                 struct rt6_info *sibling, *temp_sibling;
 767
 768                 /* Find the first route that have the same metric */
 769                 sibling = fn->leaf;
 770                 while (sibling) {
 771                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 772                             rt6_qualify_for_ecmp(sibling)) {
 773                                 list_add_tail(&rt->rt6i_siblings,
 774                                               &sibling->rt6i_siblings);
 775                                 break;
 776                         }
 777                         sibling = sibling->dst.rt6_next;
 778                 }
 779                 /* For each sibling in the list, increment the counter of
 780                  * siblings. BUG() if counters does not match, list of siblings
 781                  * is broken!
 782                  */
 783                 rt6i_nsiblings = 0;
 784                 list_for_each_entry_safe(sibling, temp_sibling,
 785                                          &rt->rt6i_siblings, rt6i_siblings) {
 786                         sibling->rt6i_nsiblings++;
 787                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 788                         rt6i_nsiblings++;
 789                 }
 790                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 791         }
 792
 793         /*
 794          *      insert node
 795          */
 796         if (!replace) {
 797                 if (!add)
 798                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 799
 800 add:
 801                 err = fib6_commit_metrics(&rt->dst, mxc);
 802                 if (err)
 803                         return err;
 804
 805                 rt->dst.rt6_next = iter;
 806                 *ins = rt;
 807                 rt->rt6i_node = fn;
 808                 atomic_inc(&rt->rt6i_ref);
 809                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 810                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 811
 812                 if (!(fn->fn_flags & RTN_RTINFO)) {
 813                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 814                         fn->fn_flags |= RTN_RTINFO;
 815                 }
 816
 817         } else {
 818                 if (!found) {
 819                         if (add)
 820                                 goto add;
 821                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
 822                         return -ENOENT;
 823                 }
 824
 825                 err = fib6_commit_metrics(&rt->dst, mxc);
 826                 if (err)
 827                         return err;
 828
 829                 *ins = rt;
 830                 rt->rt6i_node = fn;
 831                 rt->dst.rt6_next = iter->dst.rt6_next;
 832                 atomic_inc(&rt->rt6i_ref);
 833                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 834                 if (!(fn->fn_flags & RTN_RTINFO)) {
 835                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 836                         fn->fn_flags |= RTN_RTINFO;
 837                 }
 838                 fib6_purge_rt(iter, fn, info->nl_net);
 839                 rt6_release(iter);
 840         }
 841
 842         return 0;
 843 }
 844
 845 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
 846 {
 847         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
 848             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
 849                 mod_timer(&net->ipv6.ip6_fib_timer,
 850                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 851 }
 852
 853 void fib6_force_start_gc(struct net *net)
 854 {
 855         if (!timer_pending(&net->ipv6.ip6_fib_timer))
 856                 mod_timer(&net->ipv6.ip6_fib_timer,
 857                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 858 }
 859
 860 /*
 861  *      Add routing information to the routing tree.
 862  *      <destination addr>/<source addr>
 863  *      with source addr info in sub-trees
 864  */
 865
 866 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
 867              struct nl_info *info, struct mx6_config *mxc)
 868 {
 869         struct fib6_node *fn, *pn = NULL;
 870         int err = -ENOMEM;
 871         int allow_create = 1;
 872         int replace_required = 0;
 873         int sernum = fib6_new_sernum(info->nl_net);
 874
 875         if (info->nlh) {
 876                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
 877                         allow_create = 0;
 878                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
 879                         replace_required = 1;
 880         }
 881         if (!allow_create && !replace_required)
 882                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 883
 884         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
 885                         offsetof(struct rt6_info, rt6i_dst), allow_create,
 886                         replace_required, sernum);
 887         if (IS_ERR(fn)) {
 888                 err = PTR_ERR(fn);
 889                 fn = NULL;
 890                 goto out;
 891         }
 892
 893         pn = fn;
 894
 895 #ifdef CONFIG_IPV6_SUBTREES
 896         if (rt->rt6i_src.plen) {
 897                 struct fib6_node *sn;
 898
 899                 if (!fn->subtree) {
 900                         struct fib6_node *sfn;
 901
 902                         /*
 903                          * Create subtree.
 904                          *
 905                          *              fn[main tree]
 906                          *              |
 907                          *              sfn[subtree root]
 908                          *                 \
 909                          *                  sn[new leaf node]
 910                          */
 911
 912                         /* Create subtree root node */
 913                         sfn = node_alloc();
 914                         if (!sfn)
 915                                 goto st_failure;
 916
 917                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
 918                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
 919                         sfn->fn_flags = RTN_ROOT;
 920                         sfn->fn_sernum = sernum;
 921
 922                         /* Now add the first leaf node to new subtree */
 923
 924                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
 925                                         rt->rt6i_src.plen,
 926                                         offsetof(struct rt6_info, rt6i_src),
 927                                         allow_create, replace_required, sernum);
 928
 929                         if (IS_ERR(sn)) {
 930                                 /* If it is failed, discard just allocated
 931                                    root, and then (in st_failure) stale node
 932                                    in main tree.
 933                                  */
 934                                 node_free(sfn);
 935                                 err = PTR_ERR(sn);
 936                                 goto st_failure;
 937                         }
 938
 939                         /* Now link new subtree to main tree */
 940                         sfn->parent = fn;
 941                         fn->subtree = sfn;
 942                 } else {
 943                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
 944                                         rt->rt6i_src.plen,
 945                                         offsetof(struct rt6_info, rt6i_src),
 946                                         allow_create, replace_required, sernum);
 947
 948                         if (IS_ERR(sn)) {
 949                                 err = PTR_ERR(sn);
 950                                 goto st_failure;
 951                         }
 952                 }
 953
 954                 if (!fn->leaf) {
 955                         fn->leaf = rt;
 956                         atomic_inc(&rt->rt6i_ref);
 957                 }
 958                 fn = sn;
 959         }
 960 #endif
 961
 962         err = fib6_add_rt2node(fn, rt, info, mxc);
 963         if (!err) {
 964                 fib6_start_gc(info->nl_net, rt);
 965                 if (!(rt->rt6i_flags & RTF_CACHE))
 966                         fib6_prune_clones(info->nl_net, pn);
 967         }
 968
 969 out:
 970         if (err) {
 971 #ifdef CONFIG_IPV6_SUBTREES
 972                 /*
 973                  * If fib6_add_1 has cleared the old leaf pointer in the
 974                  * super-tree leaf node we have to find a new one for it.
 975                  */
 976                 if (pn != fn && pn->leaf == rt) {
 977                         pn->leaf = NULL;
 978                         atomic_dec(&rt->rt6i_ref);
 979                 }
 980                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
 981                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
 982 #if RT6_DEBUG >= 2
 983                         if (!pn->leaf) {
 984                                 WARN_ON(pn->leaf == NULL);
 985                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
 986                         }
 987 #endif
 988                         atomic_inc(&pn->leaf->rt6i_ref);
 989                 }
 990 #endif
 991                 dst_free(&rt->dst);
 992         }
 993         return err;
 994
 995 #ifdef CONFIG_IPV6_SUBTREES
 996         /* Subtree creation failed, probably main tree node
 997            is orphan. If it is, shoot it.
 998          */
 999 st_failure:
1000         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1001                 fib6_repair_tree(info->nl_net, fn);
1002         dst_free(&rt->dst);
1003         return err;
1004 #endif
1005 }
1006
1007 /*
1008  *      Routing tree lookup
1009  *
1010  */
1011
1012 struct lookup_args {
1013         int                     offset;         /* key offset on rt6_info       */
1014         const struct in6_addr   *addr;          /* search key                   */
1015 };
1016
1017 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1018                                        struct lookup_args *args)
1019 {
1020         struct fib6_node *fn;
1021         __be32 dir;
1022
1023         if (unlikely(args->offset == 0))
1024                 return NULL;
1025
1026         /*
1027          *      Descend on a tree
1028          */
1029
1030         fn = root;
1031
1032         for (;;) {
1033                 struct fib6_node *next;
1034
1035                 dir = addr_bit_set(args->addr, fn->fn_bit);
1036
1037                 next = dir ? fn->right : fn->left;
1038
1039                 if (next) {
1040                         fn = next;
1041                         continue;
1042                 }
1043                 break;
1044         }
1045
1046         while (fn) {
1047                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1048                         struct rt6key *key;
1049
1050                         key = (struct rt6key *) ((u8 *) fn->leaf +
1051                                                  args->offset);
1052
1053                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1054 #ifdef CONFIG_IPV6_SUBTREES
1055                                 if (fn->subtree) {
1056                                         struct fib6_node *sfn;
1057                                         sfn = fib6_lookup_1(fn->subtree,
1058                                                             args + 1);
1059                                         if (!sfn)
1060                                                 goto backtrack;
1061                                         fn = sfn;
1062                                 }
1063 #endif
1064                                 if (fn->fn_flags & RTN_RTINFO)
1065                                         return fn;
1066                         }
1067                 }
1068 #ifdef CONFIG_IPV6_SUBTREES
1069 backtrack:
1070 #endif
1071                 if (fn->fn_flags & RTN_ROOT)
1072                         break;
1073
1074                 fn = fn->parent;
1075         }
1076
1077         return NULL;
1078 }
1079
1080 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1081                               const struct in6_addr *saddr)
1082 {
1083         struct fib6_node *fn;
1084         struct lookup_args args[] = {
1085                 {
1086                         .offset = offsetof(struct rt6_info, rt6i_dst),
1087                         .addr = daddr,
1088                 },
1089 #ifdef CONFIG_IPV6_SUBTREES
1090                 {
1091                         .offset = offsetof(struct rt6_info, rt6i_src),
1092                         .addr = saddr,
1093                 },
1094 #endif
1095                 {
1096                         .offset = 0,    /* sentinel */
1097                 }
1098         };
1099
1100         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1101         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1102                 fn = root;
1103
1104         return fn;
1105 }
1106
1107 /*
1108  *      Get node with specified destination prefix (and source prefix,
1109  *      if subtrees are used)
1110  */
1111
1112
1113 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1114                                        const struct in6_addr *addr,
1115                                        int plen, int offset)
1116 {
1117         struct fib6_node *fn;
1118
1119         for (fn = root; fn ; ) {
1120                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1121
1122                 /*
1123                  *      Prefix match
1124                  */
1125                 if (plen < fn->fn_bit ||
1126                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1127                         return NULL;
1128
1129                 if (plen == fn->fn_bit)
1130                         return fn;
1131
1132                 /*
1133                  *      We have more bits to go
1134                  */
1135                 if (addr_bit_set(addr, fn->fn_bit))
1136                         fn = fn->right;
1137                 else
1138                         fn = fn->left;
1139         }
1140         return NULL;
1141 }
1142
1143 struct fib6_node *fib6_locate(struct fib6_node *root,
1144                               const struct in6_addr *daddr, int dst_len,
1145                               const struct in6_addr *saddr, int src_len)
1146 {
1147         struct fib6_node *fn;
1148
1149         fn = fib6_locate_1(root, daddr, dst_len,
1150                            offsetof(struct rt6_info, rt6i_dst));
1151
1152 #ifdef CONFIG_IPV6_SUBTREES
1153         if (src_len) {
1154                 WARN_ON(saddr == NULL);
1155                 if (fn && fn->subtree)
1156                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1157                                            offsetof(struct rt6_info, rt6i_src));
1158         }
1159 #endif
1160
1161         if (fn && fn->fn_flags & RTN_RTINFO)
1162                 return fn;
1163
1164         return NULL;
1165 }
1166
1167
1168 /*
1169  *      Deletion
1170  *
1171  */
1172
1173 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1174 {
1175         if (fn->fn_flags & RTN_ROOT)
1176                 return net->ipv6.ip6_null_entry;
1177
1178         while (fn) {
1179                 if (fn->left)
1180                         return fn->left->leaf;
1181                 if (fn->right)
1182                         return fn->right->leaf;
1183
1184                 fn = FIB6_SUBTREE(fn);
1185         }
1186         return NULL;
1187 }
1188
1189 /*
1190  *      Called to trim the tree of intermediate nodes when possible. "fn"
1191  *      is the node we want to try and remove.
1192  */
1193
1194 static struct fib6_node *fib6_repair_tree(struct net *net,
1195                                            struct fib6_node *fn)
1196 {
1197         int children;
1198         int nstate;
1199         struct fib6_node *child, *pn;
1200         struct fib6_walker *w;
1201         int iter = 0;
1202
1203         for (;;) {
1204                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1205                 iter++;
1206
1207                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1208                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1209                 WARN_ON(fn->leaf != NULL);
1210
1211                 children = 0;
1212                 child = NULL;
1213                 if (fn->right)
1214                         child = fn->right, children |= 1;
1215                 if (fn->left)
1216                         child = fn->left, children |= 2;
1217
1218                 if (children == 3 || FIB6_SUBTREE(fn)
1219 #ifdef CONFIG_IPV6_SUBTREES
1220                     /* Subtree root (i.e. fn) may have one child */
1221                     || (children && fn->fn_flags & RTN_ROOT)
1222 #endif
1223                     ) {
1224                         fn->leaf = fib6_find_prefix(net, fn);
1225 #if RT6_DEBUG >= 2
1226                         if (!fn->leaf) {
1227                                 WARN_ON(!fn->leaf);
1228                                 fn->leaf = net->ipv6.ip6_null_entry;
1229                         }
1230 #endif
1231                         atomic_inc(&fn->leaf->rt6i_ref);
1232                         return fn->parent;
1233                 }
1234
1235                 pn = fn->parent;
1236 #ifdef CONFIG_IPV6_SUBTREES
1237                 if (FIB6_SUBTREE(pn) == fn) {
1238                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1239                         FIB6_SUBTREE(pn) = NULL;
1240                         nstate = FWS_L;
1241                 } else {
1242                         WARN_ON(fn->fn_flags & RTN_ROOT);
1243 #endif
1244                         if (pn->right == fn)
1245                                 pn->right = child;
1246                         else if (pn->left == fn)
1247                                 pn->left = child;
1248 #if RT6_DEBUG >= 2
1249                         else
1250                                 WARN_ON(1);
1251 #endif
1252                         if (child)
1253                                 child->parent = pn;
1254                         nstate = FWS_R;
1255 #ifdef CONFIG_IPV6_SUBTREES
1256                 }
1257 #endif
1258
1259                 read_lock(&fib6_walker_lock);
1260                 FOR_WALKERS(w) {
1261                         if (!child) {
1262                                 if (w->root == fn) {
1263                                         w->root = w->node = NULL;
1264                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1265                                 } else if (w->node == fn) {
1266                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1267                                         w->node = pn;
1268                                         w->state = nstate;
1269                                 }
1270                         } else {
1271                                 if (w->root == fn) {
1272                                         w->root = child;
1273                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1274                                 }
1275                                 if (w->node == fn) {
1276                                         w->node = child;
1277                                         if (children&2) {
1278                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1279                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1280                                         } else {
1281                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1282                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1283                                         }
1284                                 }
1285                         }
1286                 }
1287                 read_unlock(&fib6_walker_lock);
1288
1289                 node_free(fn);
1290                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1291                         return pn;
1292
1293                 rt6_release(pn->leaf);
1294                 pn->leaf = NULL;
1295                 fn = pn;
1296         }
1297 }
1298
1299 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1300                            struct nl_info *info)
1301 {
1302         struct fib6_walker *w;
1303         struct rt6_info *rt = *rtp;
1304         struct net *net = info->nl_net;
1305
1306         RT6_TRACE("fib6_del_route\n");
1307
1308         /* Unlink it */
1309         *rtp = rt->dst.rt6_next;
1310         rt->rt6i_node = NULL;
1311         net->ipv6.rt6_stats->fib_rt_entries--;
1312         net->ipv6.rt6_stats->fib_discarded_routes++;
1313
1314         /* Reset round-robin state, if necessary */
1315         if (fn->rr_ptr == rt)
1316                 fn->rr_ptr = NULL;
1317
1318         /* Remove this entry from other siblings */
1319         if (rt->rt6i_nsiblings) {
1320                 struct rt6_info *sibling, *next_sibling;
1321
1322                 list_for_each_entry_safe(sibling, next_sibling,
1323                                          &rt->rt6i_siblings, rt6i_siblings)
1324                         sibling->rt6i_nsiblings--;
1325                 rt->rt6i_nsiblings = 0;
1326                 list_del_init(&rt->rt6i_siblings);
1327         }
1328
1329         /* Adjust walkers */
1330         read_lock(&fib6_walker_lock);
1331         FOR_WALKERS(w) {
1332                 if (w->state == FWS_C && w->leaf == rt) {
1333                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1334                         w->leaf = rt->dst.rt6_next;
1335                         if (!w->leaf)
1336                                 w->state = FWS_U;
1337                 }
1338         }
1339         read_unlock(&fib6_walker_lock);
1340
1341         rt->dst.rt6_next = NULL;
1342
1343         /* If it was last route, expunge its radix tree node */
1344         if (!fn->leaf) {
1345                 fn->fn_flags &= ~RTN_RTINFO;
1346                 net->ipv6.rt6_stats->fib_route_nodes--;
1347                 fn = fib6_repair_tree(net, fn);
1348         }
1349
1350         fib6_purge_rt(rt, fn, net);
1351
1352         inet6_rt_notify(RTM_DELROUTE, rt, info);
1353         rt6_release(rt);
1354 }
1355
1356 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1357 {
1358         struct net *net = info->nl_net;
1359         struct fib6_node *fn = rt->rt6i_node;
1360         struct rt6_info **rtp;
1361
1362 #if RT6_DEBUG >= 2
1363         if (rt->dst.obsolete > 0) {
1364                 WARN_ON(fn != NULL);
1365                 return -ENOENT;
1366         }
1367 #endif
1368         if (!fn || rt == net->ipv6.ip6_null_entry)
1369                 return -ENOENT;
1370
1371         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1372
1373         if (!(rt->rt6i_flags & RTF_CACHE)) {
1374                 struct fib6_node *pn = fn;
1375 #ifdef CONFIG_IPV6_SUBTREES
1376                 /* clones of this route might be in another subtree */
1377                 if (rt->rt6i_src.plen) {
1378                         while (!(pn->fn_flags & RTN_ROOT))
1379                                 pn = pn->parent;
1380                         pn = pn->parent;
1381                 }
1382 #endif
1383                 fib6_prune_clones(info->nl_net, pn);
1384         }
1385
1386         /*
1387          *      Walk the leaf entries looking for ourself
1388          */
1389
1390         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1391                 if (*rtp == rt) {
1392                         fib6_del_route(fn, rtp, info);
1393                         return 0;
1394                 }
1395         }
1396         return -ENOENT;
1397 }
1398
1399 /*
1400  *      Tree traversal function.
1401  *
1402  *      Certainly, it is not interrupt safe.
1403  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1404  *      It means, that we can modify tree during walking
1405  *      and use this function for garbage collection, clone pruning,
1406  *      cleaning tree when a device goes down etc. etc.
1407  *
1408  *      It guarantees that every node will be traversed,
1409  *      and that it will be traversed only once.
1410  *
1411  *      Callback function w->func may return:
1412  *      0 -> continue walking.
1413  *      positive value -> walking is suspended (used by tree dumps,
1414  *      and probably by gc, if it will be split to several slices)
1415  *      negative value -> terminate walking.
1416  *
1417  *      The function itself returns:
1418  *      0   -> walk is complete.
1419  *      >0  -> walk is incomplete (i.e. suspended)
1420  *      <0  -> walk is terminated by an error.
1421  */
1422
1423 static int fib6_walk_continue(struct fib6_walker *w)
1424 {
1425         struct fib6_node *fn, *pn;
1426
1427         for (;;) {
1428                 fn = w->node;
1429                 if (!fn)
1430                         return 0;
1431
1432                 if (w->prune && fn != w->root &&
1433                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1434                         w->state = FWS_C;
1435                         w->leaf = fn->leaf;
1436                 }
1437                 switch (w->state) {
1438 #ifdef CONFIG_IPV6_SUBTREES
1439                 case FWS_S:
1440                         if (FIB6_SUBTREE(fn)) {
1441                                 w->node = FIB6_SUBTREE(fn);
1442                                 continue;
1443                         }
1444                         w->state = FWS_L;
1445 #endif
1446                 case FWS_L:
1447                         if (fn->left) {
1448                                 w->node = fn->left;
1449                                 w->state = FWS_INIT;
1450                                 continue;
1451                         }
1452                         w->state = FWS_R;
1453                 case FWS_R:
1454                         if (fn->right) {
1455                                 w->node = fn->right;
1456                                 w->state = FWS_INIT;
1457                                 continue;
1458                         }
1459                         w->state = FWS_C;
1460                         w->leaf = fn->leaf;
1461                 case FWS_C:
1462                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1463                                 int err;
1464
1465                                 if (w->skip) {
1466                                         w->skip--;
1467                                         goto skip;
1468                                 }
1469
1470                                 err = w->func(w);
1471                                 if (err)
1472                                         return err;
1473
1474                                 w->count++;
1475                                 continue;
1476                         }
1477 skip:
1478                         w->state = FWS_U;
1479                 case FWS_U:
1480                         if (fn == w->root)
1481                                 return 0;
1482                         pn = fn->parent;
1483                         w->node = pn;
1484 #ifdef CONFIG_IPV6_SUBTREES
1485                         if (FIB6_SUBTREE(pn) == fn) {
1486                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1487                                 w->state = FWS_L;
1488                                 continue;
1489                         }
1490 #endif
1491                         if (pn->left == fn) {
1492                                 w->state = FWS_R;
1493                                 continue;
1494                         }
1495                         if (pn->right == fn) {
1496                                 w->state = FWS_C;
1497                                 w->leaf = w->node->leaf;
1498                                 continue;
1499                         }
1500 #if RT6_DEBUG >= 2
1501                         WARN_ON(1);
1502 #endif
1503                 }
1504         }
1505 }
1506
1507 static int fib6_walk(struct fib6_walker *w)
1508 {
1509         int res;
1510
1511         w->state = FWS_INIT;
1512         w->node = w->root;
1513
1514         fib6_walker_link(w);
1515         res = fib6_walk_continue(w);
1516         if (res <= 0)
1517                 fib6_walker_unlink(w);
1518         return res;
1519 }
1520
1521 static int fib6_clean_node(struct fib6_walker *w)
1522 {
1523         int res;
1524         struct rt6_info *rt;
1525         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1526         struct nl_info info = {
1527                 .nl_net = c->net,
1528         };
1529
1530         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1531             w->node->fn_sernum != c->sernum)
1532                 w->node->fn_sernum = c->sernum;
1533
1534         if (!c->func) {
1535                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1536                 w->leaf = NULL;
1537                 return 0;
1538         }
1539
1540         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1541                 res = c->func(rt, c->arg);
1542                 if (res < 0) {
1543                         w->leaf = rt;
1544                         res = fib6_del(rt, &info);
1545                         if (res) {
1546 #if RT6_DEBUG >= 2
1547                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1548                                          __func__, rt, rt->rt6i_node, res);
1549 #endif
1550                                 continue;
1551                         }
1552                         return 0;
1553                 }
1554                 WARN_ON(res != 0);
1555         }
1556         w->leaf = rt;
1557         return 0;
1558 }
1559
1560 /*
1561  *      Convenient frontend to tree walker.
1562  *
1563  *      func is called on each route.
1564  *              It may return -1 -> delete this route.
1565  *                            0  -> continue walking
1566  *
1567  *      prune==1 -> only immediate children of node (certainly,
1568  *      ignoring pure split nodes) will be scanned.
1569  */
1570
1571 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1572                             int (*func)(struct rt6_info *, void *arg),
1573                             bool prune, int sernum, void *arg)
1574 {
1575         struct fib6_cleaner c;
1576
1577         c.w.root = root;
1578         c.w.func = fib6_clean_node;
1579         c.w.prune = prune;
1580         c.w.count = 0;
1581         c.w.skip = 0;
1582         c.func = func;
1583         c.sernum = sernum;
1584         c.arg = arg;
1585         c.net = net;
1586
1587         fib6_walk(&c.w);
1588 }
1589
1590 static void __fib6_clean_all(struct net *net,
1591                              int (*func)(struct rt6_info *, void *),
1592                              int sernum, void *arg)
1593 {
1594         struct fib6_table *table;
1595         struct hlist_head *head;
1596         unsigned int h;
1597
1598         rcu_read_lock();
1599         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1600                 head = &net->ipv6.fib_table_hash[h];
1601                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1602                         write_lock_bh(&table->tb6_lock);
1603                         fib6_clean_tree(net, &table->tb6_root,
1604                                         func, false, sernum, arg);
1605                         write_unlock_bh(&table->tb6_lock);
1606                 }
1607         }
1608         rcu_read_unlock();
1609 }
1610
1611 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1612                     void *arg)
1613 {
1614         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1615 }
1616
1617 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1618 {
1619         if (rt->rt6i_flags & RTF_CACHE) {
1620                 RT6_TRACE("pruning clone %p\n", rt);
1621                 return -1;
1622         }
1623
1624         return 0;
1625 }
1626
1627 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1628 {
1629         fib6_clean_tree(net, fn, fib6_prune_clone, true,
1630                         FIB6_NO_SERNUM_CHANGE, NULL);
1631 }
1632
1633 static void fib6_flush_trees(struct net *net)
1634 {
1635         int new_sernum = fib6_new_sernum(net);
1636
1637         __fib6_clean_all(net, NULL, new_sernum, NULL);
1638 }
1639
1640 /*
1641  *      Garbage collection
1642  */
1643
1644 static struct fib6_gc_args
1645 {
1646         int                     timeout;
1647         int                     more;
1648 } gc_args;
1649
1650 static int fib6_age(struct rt6_info *rt, void *arg)
1651 {
1652         unsigned long now = jiffies;
1653
1654         /*
1655          *      check addrconf expiration here.
1656          *      Routes are expired even if they are in use.
1657          *
1658          *      Also age clones. Note, that clones are aged out
1659          *      only if they are not in use now.
1660          */
1661
1662         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1663                 if (time_after(now, rt->dst.expires)) {
1664                         RT6_TRACE("expiring %p\n", rt);
1665                         return -1;
1666                 }
1667                 gc_args.more++;
1668         } else if (rt->rt6i_flags & RTF_CACHE) {
1669                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1670                     time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1671                         RT6_TRACE("aging clone %p\n", rt);
1672                         return -1;
1673                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1674                         struct neighbour *neigh;
1675                         __u8 neigh_flags = 0;
1676
1677                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1678                         if (neigh) {
1679                                 neigh_flags = neigh->flags;
1680                                 neigh_release(neigh);
1681                         }
1682                         if (!(neigh_flags & NTF_ROUTER)) {
1683                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1684                                           rt);
1685                                 return -1;
1686                         }
1687                 }
1688                 gc_args.more++;
1689         }
1690
1691         return 0;
1692 }
1693
1694 static DEFINE_SPINLOCK(fib6_gc_lock);
1695
1696 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1697 {
1698         unsigned long now;
1699
1700         if (force) {
1701                 spin_lock_bh(&fib6_gc_lock);
1702         } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1703                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1704                 return;
1705         }
1706         gc_args.timeout = expires ? (int)expires :
1707                           net->ipv6.sysctl.ip6_rt_gc_interval;
1708
1709         gc_args.more = icmp6_dst_gc();
1710
1711         fib6_clean_all(net, fib6_age, NULL);
1712         now = jiffies;
1713         net->ipv6.ip6_rt_last_gc = now;
1714
1715         if (gc_args.more)
1716                 mod_timer(&net->ipv6.ip6_fib_timer,
1717                           round_jiffies(now
1718                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1719         else
1720                 del_timer(&net->ipv6.ip6_fib_timer);
1721         spin_unlock_bh(&fib6_gc_lock);
1722 }
1723
1724 static void fib6_gc_timer_cb(unsigned long arg)
1725 {
1726         fib6_run_gc(0, (struct net *)arg, true);
1727 }
1728
1729 static int __net_init fib6_net_init(struct net *net)
1730 {
1731         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1732
1733         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1734
1735         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1736         if (!net->ipv6.rt6_stats)
1737                 goto out_timer;
1738
1739         /* Avoid false sharing : Use at least a full cache line */
1740         size = max_t(size_t, size, L1_CACHE_BYTES);
1741
1742         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1743         if (!net->ipv6.fib_table_hash)
1744                 goto out_rt6_stats;
1745
1746         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1747                                           GFP_KERNEL);
1748         if (!net->ipv6.fib6_main_tbl)
1749                 goto out_fib_table_hash;
1750
1751         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1752         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1753         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1754                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1755         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1756
1757 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1758         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1759                                            GFP_KERNEL);
1760         if (!net->ipv6.fib6_local_tbl)
1761                 goto out_fib6_main_tbl;
1762         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1763         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1764         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1765                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1766         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1767 #endif
1768         fib6_tables_init(net);
1769
1770         return 0;
1771
1772 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1773 out_fib6_main_tbl:
1774         kfree(net->ipv6.fib6_main_tbl);
1775 #endif
1776 out_fib_table_hash:
1777         kfree(net->ipv6.fib_table_hash);
1778 out_rt6_stats:
1779         kfree(net->ipv6.rt6_stats);
1780 out_timer:
1781         return -ENOMEM;
1782 }
1783
1784 static void fib6_net_exit(struct net *net)
1785 {
1786         rt6_ifdown(net, NULL);
1787         del_timer_sync(&net->ipv6.ip6_fib_timer);
1788
1789 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1790         inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1791         kfree(net->ipv6.fib6_local_tbl);
1792 #endif
1793         inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1794         kfree(net->ipv6.fib6_main_tbl);
1795         kfree(net->ipv6.fib_table_hash);
1796         kfree(net->ipv6.rt6_stats);
1797 }
1798
1799 static struct pernet_operations fib6_net_ops = {
1800         .init = fib6_net_init,
1801         .exit = fib6_net_exit,
1802 };
1803
1804 int __init fib6_init(void)
1805 {
1806         int ret = -ENOMEM;
1807
1808         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1809                                            sizeof(struct fib6_node),
1810                                            0, SLAB_HWCACHE_ALIGN,
1811                                            NULL);
1812         if (!fib6_node_kmem)
1813                 goto out;
1814
1815         ret = register_pernet_subsys(&fib6_net_ops);
1816         if (ret)
1817                 goto out_kmem_cache_create;
1818
1819         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1820                               NULL);
1821         if (ret)
1822                 goto out_unregister_subsys;
1823
1824         __fib6_flush_trees = fib6_flush_trees;
1825 out:
1826         return ret;
1827
1828 out_unregister_subsys:
1829         unregister_pernet_subsys(&fib6_net_ops);
1830 out_kmem_cache_create:
1831         kmem_cache_destroy(fib6_node_kmem);
1832         goto out;
1833 }
1834
1835 void fib6_gc_cleanup(void)
1836 {
1837         unregister_pernet_subsys(&fib6_net_ops);
1838         kmem_cache_destroy(fib6_node_kmem);
1839 }
1840
1841 #ifdef CONFIG_PROC_FS
1842
1843 struct ipv6_route_iter {
1844         struct seq_net_private p;
1845         struct fib6_walker w;
1846         loff_t skip;
1847         struct fib6_table *tbl;
1848         int sernum;
1849 };
1850
1851 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1852 {
1853         struct rt6_info *rt = v;
1854         struct ipv6_route_iter *iter = seq->private;
1855
1856         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1857
1858 #ifdef CONFIG_IPV6_SUBTREES
1859         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1860 #else
1861         seq_puts(seq, "00000000000000000000000000000000 00 ");
1862 #endif
1863         if (rt->rt6i_flags & RTF_GATEWAY)
1864                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1865         else
1866                 seq_puts(seq, "00000000000000000000000000000000");
1867
1868         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1869                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1870                    rt->dst.__use, rt->rt6i_flags,
1871                    rt->dst.dev ? rt->dst.dev->name : "");
1872         iter->w.leaf = NULL;
1873         return 0;
1874 }
1875
1876 static int ipv6_route_yield(struct fib6_walker *w)
1877 {
1878         struct ipv6_route_iter *iter = w->args;
1879
1880         if (!iter->skip)
1881                 return 1;
1882
1883         do {
1884                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1885                 iter->skip--;
1886                 if (!iter->skip && iter->w.leaf)
1887                         return 1;
1888         } while (iter->w.leaf);
1889
1890         return 0;
1891 }
1892
1893 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1894 {
1895         memset(&iter->w, 0, sizeof(iter->w));
1896         iter->w.func = ipv6_route_yield;
1897         iter->w.root = &iter->tbl->tb6_root;
1898         iter->w.state = FWS_INIT;
1899         iter->w.node = iter->w.root;
1900         iter->w.args = iter;
1901         iter->sernum = iter->w.root->fn_sernum;
1902         INIT_LIST_HEAD(&iter->w.lh);
1903         fib6_walker_link(&iter->w);
1904 }
1905
1906 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1907                                                     struct net *net)
1908 {
1909         unsigned int h;
1910         struct hlist_node *node;
1911
1912         if (tbl) {
1913                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1914                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1915         } else {
1916                 h = 0;
1917                 node = NULL;
1918         }
1919
1920         while (!node && h < FIB6_TABLE_HASHSZ) {
1921                 node = rcu_dereference_bh(
1922                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1923         }
1924         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1925 }
1926
1927 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1928 {
1929         if (iter->sernum != iter->w.root->fn_sernum) {
1930                 iter->sernum = iter->w.root->fn_sernum;
1931                 iter->w.state = FWS_INIT;
1932                 iter->w.node = iter->w.root;
1933                 WARN_ON(iter->w.skip);
1934                 iter->w.skip = iter->w.count;
1935         }
1936 }
1937
1938 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1939 {
1940         int r;
1941         struct rt6_info *n;
1942         struct net *net = seq_file_net(seq);
1943         struct ipv6_route_iter *iter = seq->private;
1944
1945         if (!v)
1946                 goto iter_table;
1947
1948         n = ((struct rt6_info *)v)->dst.rt6_next;
1949         if (n) {
1950                 ++*pos;
1951                 return n;
1952         }
1953
1954 iter_table:
1955         ipv6_route_check_sernum(iter);
1956         read_lock(&iter->tbl->tb6_lock);
1957         r = fib6_walk_continue(&iter->w);
1958         read_unlock(&iter->tbl->tb6_lock);
1959         if (r > 0) {
1960                 if (v)
1961                         ++*pos;
1962                 return iter->w.leaf;
1963         } else if (r < 0) {
1964                 fib6_walker_unlink(&iter->w);
1965                 return NULL;
1966         }
1967         fib6_walker_unlink(&iter->w);
1968
1969         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
1970         if (!iter->tbl)
1971                 return NULL;
1972
1973         ipv6_route_seq_setup_walk(iter);
1974         goto iter_table;
1975 }
1976
1977 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
1978         __acquires(RCU_BH)
1979 {
1980         struct net *net = seq_file_net(seq);
1981         struct ipv6_route_iter *iter = seq->private;
1982
1983         rcu_read_lock_bh();
1984         iter->tbl = ipv6_route_seq_next_table(NULL, net);
1985         iter->skip = *pos;
1986
1987         if (iter->tbl) {
1988                 ipv6_route_seq_setup_walk(iter);
1989                 return ipv6_route_seq_next(seq, NULL, pos);
1990         } else {
1991                 return NULL;
1992         }
1993 }
1994
1995 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
1996 {
1997         struct fib6_walker *w = &iter->w;
1998         return w->node && !(w->state == FWS_U && w->node == w->root);
1999 }
2000
2001 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2002         __releases(RCU_BH)
2003 {
2004         struct ipv6_route_iter *iter = seq->private;
2005
2006         if (ipv6_route_iter_active(iter))
2007                 fib6_walker_unlink(&iter->w);
2008
2009         rcu_read_unlock_bh();
2010 }
2011
2012 static const struct seq_operations ipv6_route_seq_ops = {
2013         .start  = ipv6_route_seq_start,
2014         .next   = ipv6_route_seq_next,
2015         .stop   = ipv6_route_seq_stop,
2016         .show   = ipv6_route_seq_show
2017 };
2018
2019 int ipv6_route_open(struct inode *inode, struct file *file)
2020 {
2021         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2022                             sizeof(struct ipv6_route_iter));
2023 }
2024
2025 #endif /* CONFIG_PROC_FS */