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
  14 /*
  15  *      Changes:
  16  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  17  *                              remove ip6_null_entry from the top of
  18  *                              routing table.
  19  *      Ville Nuorvala:         Fixed routing subtrees.
  20  */
  21
  22 #define pr_fmt(fmt) "IPv6: " fmt
  23
  24 #include <linux/errno.h>
  25 #include <linux/types.h>
  26 #include <linux/net.h>
  27 #include <linux/route.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/in6.h>
  30 #include <linux/init.h>
  31 #include <linux/list.h>
  32 #include <linux/slab.h>
  33
  34 #include <net/ipv6.h>
  35 #include <net/ndisc.h>
  36 #include <net/addrconf.h>
  37
  38 #include <net/ip6_fib.h>
  39 #include <net/ip6_route.h>
  40
  41 #define RT6_DEBUG 2
  42
  43 #if RT6_DEBUG >= 3
  44 #define RT6_TRACE(x...) pr_debug(x)
  45 #else
  46 #define RT6_TRACE(x...) do { ; } while (0)
  47 #endif
  48
  49 static struct kmem_cache * fib6_node_kmem __read_mostly;
  50
  51 enum fib_walk_state_t
  52 {
  53 #ifdef CONFIG_IPV6_SUBTREES
  54         FWS_S,
  55 #endif
  56         FWS_L,
  57         FWS_R,
  58         FWS_C,
  59         FWS_U
  60 };
  61
  62 struct fib6_cleaner_t
  63 {
  64         struct fib6_walker_t w;
  65         struct net *net;
  66         int (*func)(struct rt6_info *, void *arg);
  67         void *arg;
  68 };
  69
  70 static DEFINE_RWLOCK(fib6_walker_lock);
  71
  72 #ifdef CONFIG_IPV6_SUBTREES
  73 #define FWS_INIT FWS_S
  74 #else
  75 #define FWS_INIT FWS_L
  76 #endif
  77
  78 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
  79                               struct rt6_info *rt);
  80 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
  81 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
  82 static int fib6_walk(struct fib6_walker_t *w);
  83 static int fib6_walk_continue(struct fib6_walker_t *w);
  84
  85 /*
  86  *      A routing update causes an increase of the serial number on the
  87  *      affected subtree. This allows for cached routes to be asynchronously
  88  *      tested when modifications are made to the destination cache as a
  89  *      result of redirects, path MTU changes, etc.
  90  */
  91
  92 static __u32 rt_sernum;
  93
  94 static void fib6_gc_timer_cb(unsigned long arg);
  95
  96 static LIST_HEAD(fib6_walkers);
  97 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
  98
  99 static inline void fib6_walker_link(struct fib6_walker_t *w)
 100 {
 101         write_lock_bh(&fib6_walker_lock);
 102         list_add(&w->lh, &fib6_walkers);
 103         write_unlock_bh(&fib6_walker_lock);
 104 }
 105
 106 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
 107 {
 108         write_lock_bh(&fib6_walker_lock);
 109         list_del(&w->lh);
 110         write_unlock_bh(&fib6_walker_lock);
 111 }
 112 static __inline__ u32 fib6_new_sernum(void)
 113 {
 114         u32 n = ++rt_sernum;
 115         if ((__s32)n <= 0)
 116                 rt_sernum = n = 1;
 117         return n;
 118 }
 119
 120 /*
 121  *      Auxiliary address test functions for the radix tree.
 122  *
 123  *      These assume a 32bit processor (although it will work on
 124  *      64bit processors)
 125  */
 126
 127 /*
 128  *      test bit
 129  */
 130 #if defined(__LITTLE_ENDIAN)
 131 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 132 #else
 133 # define BITOP_BE32_SWIZZLE     0
 134 #endif
 135
 136 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
 137 {
 138         const __be32 *addr = token;
 139         /*
 140          * Here,
 141          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 142          * is optimized version of
 143          *      htonl(1 << ((~fn_bit)&0x1F))
 144          * See include/asm-generic/bitops/le.h.
 145          */
 146         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 147                addr[fn_bit >> 5];
 148 }
 149
 150 static __inline__ struct fib6_node * node_alloc(void)
 151 {
 152         struct fib6_node *fn;
 153
 154         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 155
 156         return fn;
 157 }
 158
 159 static __inline__ void node_free(struct fib6_node * fn)
 160 {
 161         kmem_cache_free(fib6_node_kmem, fn);
 162 }
 163
 164 static __inline__ void rt6_release(struct rt6_info *rt)
 165 {
 166         if (atomic_dec_and_test(&rt->rt6i_ref))
 167                 dst_free(&rt->dst);
 168 }
 169
 170 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 171 {
 172         unsigned int h;
 173
 174         /*
 175          * Initialize table lock at a single place to give lockdep a key,
 176          * tables aren't visible prior to being linked to the list.
 177          */
 178         rwlock_init(&tb->tb6_lock);
 179
 180         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 181
 182         /*
 183          * No protection necessary, this is the only list mutatation
 184          * operation, tables never disappear once they exist.
 185          */
 186         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 187 }
 188
 189 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 190
 191 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 192 {
 193         struct fib6_table *table;
 194
 195         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 196         if (table) {
 197                 table->tb6_id = id;
 198                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
 199                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 200                 inet_peer_base_init(&table->tb6_peers);
 201         }
 202
 203         return table;
 204 }
 205
 206 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 207 {
 208         struct fib6_table *tb;
 209
 210         if (id == 0)
 211                 id = RT6_TABLE_MAIN;
 212         tb = fib6_get_table(net, id);
 213         if (tb)
 214                 return tb;
 215
 216         tb = fib6_alloc_table(net, id);
 217         if (tb)
 218                 fib6_link_table(net, tb);
 219
 220         return tb;
 221 }
 222
 223 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 224 {
 225         struct fib6_table *tb;
 226         struct hlist_head *head;
 227         unsigned int h;
 228
 229         if (id == 0)
 230                 id = RT6_TABLE_MAIN;
 231         h = id & (FIB6_TABLE_HASHSZ - 1);
 232         rcu_read_lock();
 233         head = &net->ipv6.fib_table_hash[h];
 234         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 235                 if (tb->tb6_id == id) {
 236                         rcu_read_unlock();
 237                         return tb;
 238                 }
 239         }
 240         rcu_read_unlock();
 241
 242         return NULL;
 243 }
 244
 245 static void __net_init fib6_tables_init(struct net *net)
 246 {
 247         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 248         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 249 }
 250 #else
 251
 252 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 253 {
 254         return fib6_get_table(net, id);
 255 }
 256
 257 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 258 {
 259           return net->ipv6.fib6_main_tbl;
 260 }
 261
 262 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 263                                    int flags, pol_lookup_t lookup)
 264 {
 265         return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
 266 }
 267
 268 static void __net_init fib6_tables_init(struct net *net)
 269 {
 270         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 271 }
 272
 273 #endif
 274
 275 static int fib6_dump_node(struct fib6_walker_t *w)
 276 {
 277         int res;
 278         struct rt6_info *rt;
 279
 280         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
 281                 res = rt6_dump_route(rt, w->args);
 282                 if (res < 0) {
 283                         /* Frame is full, suspend walking */
 284                         w->leaf = rt;
 285                         return 1;
 286                 }
 287                 WARN_ON(res == 0);
 288         }
 289         w->leaf = NULL;
 290         return 0;
 291 }
 292
 293 static void fib6_dump_end(struct netlink_callback *cb)
 294 {
 295         struct fib6_walker_t *w = (void*)cb->args[2];
 296
 297         if (w) {
 298                 if (cb->args[4]) {
 299                         cb->args[4] = 0;
 300                         fib6_walker_unlink(w);
 301                 }
 302                 cb->args[2] = 0;
 303                 kfree(w);
 304         }
 305         cb->done = (void*)cb->args[3];
 306         cb->args[1] = 3;
 307 }
 308
 309 static int fib6_dump_done(struct netlink_callback *cb)
 310 {
 311         fib6_dump_end(cb);
 312         return cb->done ? cb->done(cb) : 0;
 313 }
 314
 315 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 316                            struct netlink_callback *cb)
 317 {
 318         struct fib6_walker_t *w;
 319         int res;
 320
 321         w = (void *)cb->args[2];
 322         w->root = &table->tb6_root;
 323
 324         if (cb->args[4] == 0) {
 325                 w->count = 0;
 326                 w->skip = 0;
 327
 328                 read_lock_bh(&table->tb6_lock);
 329                 res = fib6_walk(w);
 330                 read_unlock_bh(&table->tb6_lock);
 331                 if (res > 0) {
 332                         cb->args[4] = 1;
 333                         cb->args[5] = w->root->fn_sernum;
 334                 }
 335         } else {
 336                 if (cb->args[5] != w->root->fn_sernum) {
 337                         /* Begin at the root if the tree changed */
 338                         cb->args[5] = w->root->fn_sernum;
 339                         w->state = FWS_INIT;
 340                         w->node = w->root;
 341                         w->skip = w->count;
 342                 } else
 343                         w->skip = 0;
 344
 345                 read_lock_bh(&table->tb6_lock);
 346                 res = fib6_walk_continue(w);
 347                 read_unlock_bh(&table->tb6_lock);
 348                 if (res <= 0) {
 349                         fib6_walker_unlink(w);
 350                         cb->args[4] = 0;
 351                 }
 352         }
 353
 354         return res;
 355 }
 356
 357 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 358 {
 359         struct net *net = sock_net(skb->sk);
 360         unsigned int h, s_h;
 361         unsigned int e = 0, s_e;
 362         struct rt6_rtnl_dump_arg arg;
 363         struct fib6_walker_t *w;
 364         struct fib6_table *tb;
 365         struct hlist_head *head;
 366         int res = 0;
 367
 368         s_h = cb->args[0];
 369         s_e = cb->args[1];
 370
 371         w = (void *)cb->args[2];
 372         if (!w) {
 373                 /* New dump:
 374                  *
 375                  * 1. hook callback destructor.
 376                  */
 377                 cb->args[3] = (long)cb->done;
 378                 cb->done = fib6_dump_done;
 379
 380                 /*
 381                  * 2. allocate and initialize walker.
 382                  */
 383                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 384                 if (!w)
 385                         return -ENOMEM;
 386                 w->func = fib6_dump_node;
 387                 cb->args[2] = (long)w;
 388         }
 389
 390         arg.skb = skb;
 391         arg.cb = cb;
 392         arg.net = net;
 393         w->args = &arg;
 394
 395         rcu_read_lock();
 396         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 397                 e = 0;
 398                 head = &net->ipv6.fib_table_hash[h];
 399                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 400                         if (e < s_e)
 401                                 goto next;
 402                         res = fib6_dump_table(tb, skb, cb);
 403                         if (res != 0)
 404                                 goto out;
 405 next:
 406                         e++;
 407                 }
 408         }
 409 out:
 410         rcu_read_unlock();
 411         cb->args[1] = e;
 412         cb->args[0] = h;
 413
 414         res = res < 0 ? res : skb->len;
 415         if (res <= 0)
 416                 fib6_dump_end(cb);
 417         return res;
 418 }
 419
 420 /*
 421  *      Routing Table
 422  *
 423  *      return the appropriate node for a routing tree "add" operation
 424  *      by either creating and inserting or by returning an existing
 425  *      node.
 426  */
 427
 428 static struct fib6_node * fib6_add_1(struct fib6_node *root, void *addr,
 429                                      int addrlen, int plen,
 430                                      int offset, int allow_create,
 431                                      int replace_required)
 432 {
 433         struct fib6_node *fn, *in, *ln;
 434         struct fib6_node *pn = NULL;
 435         struct rt6key *key;
 436         int     bit;
 437         __be32  dir = 0;
 438         __u32   sernum = fib6_new_sernum();
 439
 440         RT6_TRACE("fib6_add_1\n");
 441
 442         /* insert node in tree */
 443
 444         fn = root;
 445
 446         do {
 447                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
 448
 449                 /*
 450                  *      Prefix match
 451                  */
 452                 if (plen < fn->fn_bit ||
 453                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 454                         if (!allow_create) {
 455                                 if (replace_required) {
 456                                         pr_warn("Can't replace route, no match found\n");
 457                                         return ERR_PTR(-ENOENT);
 458                                 }
 459                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 460                         }
 461                         goto insert_above;
 462                 }
 463
 464                 /*
 465                  *      Exact match ?
 466                  */
 467
 468                 if (plen == fn->fn_bit) {
 469                         /* clean up an intermediate node */
 470                         if (!(fn->fn_flags & RTN_RTINFO)) {
 471                                 rt6_release(fn->leaf);
 472                                 fn->leaf = NULL;
 473                         }
 474
 475                         fn->fn_sernum = sernum;
 476
 477                         return fn;
 478                 }
 479
 480                 /*
 481                  *      We have more bits to go
 482                  */
 483
 484                 /* Try to walk down on tree. */
 485                 fn->fn_sernum = sernum;
 486                 dir = addr_bit_set(addr, fn->fn_bit);
 487                 pn = fn;
 488                 fn = dir ? fn->right: fn->left;
 489         } while (fn);
 490
 491         if (!allow_create) {
 492                 /* We should not create new node because
 493                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 494                  * I assume it is safe to require NLM_F_CREATE when
 495                  * REPLACE flag is used! Later we may want to remove the
 496                  * check for replace_required, because according
 497                  * to netlink specification, NLM_F_CREATE
 498                  * MUST be specified if new route is created.
 499                  * That would keep IPv6 consistent with IPv4
 500                  */
 501                 if (replace_required) {
 502                         pr_warn("Can't replace route, no match found\n");
 503                         return ERR_PTR(-ENOENT);
 504                 }
 505                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 506         }
 507         /*
 508          *      We walked to the bottom of tree.
 509          *      Create new leaf node without children.
 510          */
 511
 512         ln = node_alloc();
 513
 514         if (!ln)
 515                 return ERR_PTR(-ENOMEM);
 516         ln->fn_bit = plen;
 517
 518         ln->parent = pn;
 519         ln->fn_sernum = sernum;
 520
 521         if (dir)
 522                 pn->right = ln;
 523         else
 524                 pn->left  = ln;
 525
 526         return ln;
 527
 528
 529 insert_above:
 530         /*
 531          * split since we don't have a common prefix anymore or
 532          * we have a less significant route.
 533          * we've to insert an intermediate node on the list
 534          * this new node will point to the one we need to create
 535          * and the current
 536          */
 537
 538         pn = fn->parent;
 539
 540         /* find 1st bit in difference between the 2 addrs.
 541
 542            See comment in __ipv6_addr_diff: bit may be an invalid value,
 543            but if it is >= plen, the value is ignored in any case.
 544          */
 545
 546         bit = __ipv6_addr_diff(addr, &key->addr, addrlen);
 547
 548         /*
 549          *              (intermediate)[in]
 550          *                /        \
 551          *      (new leaf node)[ln] (old node)[fn]
 552          */
 553         if (plen > bit) {
 554                 in = node_alloc();
 555                 ln = node_alloc();
 556
 557                 if (!in || !ln) {
 558                         if (in)
 559                                 node_free(in);
 560                         if (ln)
 561                                 node_free(ln);
 562                         return ERR_PTR(-ENOMEM);
 563                 }
 564
 565                 /*
 566                  * new intermediate node.
 567                  * RTN_RTINFO will
 568                  * be off since that an address that chooses one of
 569                  * the branches would not match less specific routes
 570                  * in the other branch
 571                  */
 572
 573                 in->fn_bit = bit;
 574
 575                 in->parent = pn;
 576                 in->leaf = fn->leaf;
 577                 atomic_inc(&in->leaf->rt6i_ref);
 578
 579                 in->fn_sernum = sernum;
 580
 581                 /* update parent pointer */
 582                 if (dir)
 583                         pn->right = in;
 584                 else
 585                         pn->left  = in;
 586
 587                 ln->fn_bit = plen;
 588
 589                 ln->parent = in;
 590                 fn->parent = in;
 591
 592                 ln->fn_sernum = sernum;
 593
 594                 if (addr_bit_set(addr, bit)) {
 595                         in->right = ln;
 596                         in->left  = fn;
 597                 } else {
 598                         in->left  = ln;
 599                         in->right = fn;
 600                 }
 601         } else { /* plen <= bit */
 602
 603                 /*
 604                  *              (new leaf node)[ln]
 605                  *                /        \
 606                  *           (old node)[fn] NULL
 607                  */
 608
 609                 ln = node_alloc();
 610
 611                 if (!ln)
 612                         return ERR_PTR(-ENOMEM);
 613
 614                 ln->fn_bit = plen;
 615
 616                 ln->parent = pn;
 617
 618                 ln->fn_sernum = sernum;
 619
 620                 if (dir)
 621                         pn->right = ln;
 622                 else
 623                         pn->left  = ln;
 624
 625                 if (addr_bit_set(&key->addr, plen))
 626                         ln->right = fn;
 627                 else
 628                         ln->left  = fn;
 629
 630                 fn->parent = ln;
 631         }
 632         return ln;
 633 }
 634
 635 static inline bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 636 {
 637         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 638                RTF_GATEWAY;
 639 }
 640
 641 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 642                           struct net *net)
 643 {
 644         if (atomic_read(&rt->rt6i_ref) != 1) {
 645                 /* This route is used as dummy address holder in some split
 646                  * nodes. It is not leaked, but it still holds other resources,
 647                  * which must be released in time. So, scan ascendant nodes
 648                  * and replace dummy references to this route with references
 649                  * to still alive ones.
 650                  */
 651                 while (fn) {
 652                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
 653                                 fn->leaf = fib6_find_prefix(net, fn);
 654                                 atomic_inc(&fn->leaf->rt6i_ref);
 655                                 rt6_release(rt);
 656                         }
 657                         fn = fn->parent;
 658                 }
 659                 /* No more references are possible at this point. */
 660                 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
 661         }
 662 }
 663
 664 /*
 665  *      Insert routing information in a node.
 666  */
 667
 668 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 669                             struct nl_info *info)
 670 {
 671         struct rt6_info *iter = NULL;
 672         struct rt6_info **ins;
 673         int replace = (info->nlh &&
 674                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 675         int add = (!info->nlh ||
 676                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 677         int found = 0;
 678         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 679
 680         ins = &fn->leaf;
 681
 682         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 683                 /*
 684                  *      Search for duplicates
 685                  */
 686
 687                 if (iter->rt6i_metric == rt->rt6i_metric) {
 688                         /*
 689                          *      Same priority level
 690                          */
 691                         if (info->nlh &&
 692                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 693                                 return -EEXIST;
 694                         if (replace) {
 695                                 found++;
 696                                 break;
 697                         }
 698
 699                         if (iter->dst.dev == rt->dst.dev &&
 700                             iter->rt6i_idev == rt->rt6i_idev &&
 701                             ipv6_addr_equal(&iter->rt6i_gateway,
 702                                             &rt->rt6i_gateway)) {
 703                                 if (rt->rt6i_nsiblings)
 704                                         rt->rt6i_nsiblings = 0;
 705                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 706                                         return -EEXIST;
 707                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 708                                         rt6_clean_expires(iter);
 709                                 else
 710                                         rt6_set_expires(iter, rt->dst.expires);
 711                                 return -EEXIST;
 712                         }
 713                         /* If we have the same destination and the same metric,
 714                          * but not the same gateway, then the route we try to
 715                          * add is sibling to this route, increment our counter
 716                          * of siblings, and later we will add our route to the
 717                          * list.
 718                          * Only static routes (which don't have flag
 719                          * RTF_EXPIRES) are used for ECMPv6.
 720                          *
 721                          * To avoid long list, we only had siblings if the
 722                          * route have a gateway.
 723                          */
 724                         if (rt_can_ecmp &&
 725                             rt6_qualify_for_ecmp(iter))
 726                                 rt->rt6i_nsiblings++;
 727                 }
 728
 729                 if (iter->rt6i_metric > rt->rt6i_metric)
 730                         break;
 731
 732                 ins = &iter->dst.rt6_next;
 733         }
 734
 735         /* Reset round-robin state, if necessary */
 736         if (ins == &fn->leaf)
 737                 fn->rr_ptr = NULL;
 738
 739         /* Link this route to others same route. */
 740         if (rt->rt6i_nsiblings) {
 741                 unsigned int rt6i_nsiblings;
 742                 struct rt6_info *sibling, *temp_sibling;
 743
 744                 /* Find the first route that have the same metric */
 745                 sibling = fn->leaf;
 746                 while (sibling) {
 747                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 748                             rt6_qualify_for_ecmp(sibling)) {
 749                                 list_add_tail(&rt->rt6i_siblings,
 750                                               &sibling->rt6i_siblings);
 751                                 break;
 752                         }
 753                         sibling = sibling->dst.rt6_next;
 754                 }
 755                 /* For each sibling in the list, increment the counter of
 756                  * siblings. BUG() if counters does not match, list of siblings
 757                  * is broken!
 758                  */
 759                 rt6i_nsiblings = 0;
 760                 list_for_each_entry_safe(sibling, temp_sibling,
 761                                          &rt->rt6i_siblings, rt6i_siblings) {
 762                         sibling->rt6i_nsiblings++;
 763                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 764                         rt6i_nsiblings++;
 765                 }
 766                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 767         }
 768
 769         /*
 770          *      insert node
 771          */
 772         if (!replace) {
 773                 if (!add)
 774                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 775
 776 add:
 777                 rt->dst.rt6_next = iter;
 778                 *ins = rt;
 779                 rt->rt6i_node = fn;
 780                 atomic_inc(&rt->rt6i_ref);
 781                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 782                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 783
 784                 if (!(fn->fn_flags & RTN_RTINFO)) {
 785                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 786                         fn->fn_flags |= RTN_RTINFO;
 787                 }
 788
 789         } else {
 790                 if (!found) {
 791                         if (add)
 792                                 goto add;
 793                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
 794                         return -ENOENT;
 795                 }
 796                 *ins = rt;
 797                 rt->rt6i_node = fn;
 798                 rt->dst.rt6_next = iter->dst.rt6_next;
 799                 atomic_inc(&rt->rt6i_ref);
 800                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 801                 if (!(fn->fn_flags & RTN_RTINFO)) {
 802                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 803                         fn->fn_flags |= RTN_RTINFO;
 804                 }
 805                 fib6_purge_rt(iter, fn, info->nl_net);
 806                 rt6_release(iter);
 807         }
 808
 809         return 0;
 810 }
 811
 812 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
 813 {
 814         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
 815             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
 816                 mod_timer(&net->ipv6.ip6_fib_timer,
 817                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 818 }
 819
 820 void fib6_force_start_gc(struct net *net)
 821 {
 822         if (!timer_pending(&net->ipv6.ip6_fib_timer))
 823                 mod_timer(&net->ipv6.ip6_fib_timer,
 824                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 825 }
 826
 827 /*
 828  *      Add routing information to the routing tree.
 829  *      <destination addr>/<source addr>
 830  *      with source addr info in sub-trees
 831  */
 832
 833 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info)
 834 {
 835         struct fib6_node *fn, *pn = NULL;
 836         int err = -ENOMEM;
 837         int allow_create = 1;
 838         int replace_required = 0;
 839
 840         if (info->nlh) {
 841                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
 842                         allow_create = 0;
 843                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
 844                         replace_required = 1;
 845         }
 846         if (!allow_create && !replace_required)
 847                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 848
 849         fn = fib6_add_1(root, &rt->rt6i_dst.addr, sizeof(struct in6_addr),
 850                         rt->rt6i_dst.plen, offsetof(struct rt6_info, rt6i_dst),
 851                         allow_create, replace_required);
 852         if (IS_ERR(fn)) {
 853                 err = PTR_ERR(fn);
 854                 fn = NULL;
 855                 goto out;
 856         }
 857
 858         pn = fn;
 859
 860 #ifdef CONFIG_IPV6_SUBTREES
 861         if (rt->rt6i_src.plen) {
 862                 struct fib6_node *sn;
 863
 864                 if (!fn->subtree) {
 865                         struct fib6_node *sfn;
 866
 867                         /*
 868                          * Create subtree.
 869                          *
 870                          *              fn[main tree]
 871                          *              |
 872                          *              sfn[subtree root]
 873                          *                 \
 874                          *                  sn[new leaf node]
 875                          */
 876
 877                         /* Create subtree root node */
 878                         sfn = node_alloc();
 879                         if (!sfn)
 880                                 goto st_failure;
 881
 882                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
 883                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
 884                         sfn->fn_flags = RTN_ROOT;
 885                         sfn->fn_sernum = fib6_new_sernum();
 886
 887                         /* Now add the first leaf node to new subtree */
 888
 889                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
 890                                         sizeof(struct in6_addr), rt->rt6i_src.plen,
 891                                         offsetof(struct rt6_info, rt6i_src),
 892                                         allow_create, replace_required);
 893
 894                         if (IS_ERR(sn)) {
 895                                 /* If it is failed, discard just allocated
 896                                    root, and then (in st_failure) stale node
 897                                    in main tree.
 898                                  */
 899                                 node_free(sfn);
 900                                 err = PTR_ERR(sn);
 901                                 goto st_failure;
 902                         }
 903
 904                         /* Now link new subtree to main tree */
 905                         sfn->parent = fn;
 906                         fn->subtree = sfn;
 907                 } else {
 908                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
 909                                         sizeof(struct in6_addr), rt->rt6i_src.plen,
 910                                         offsetof(struct rt6_info, rt6i_src),
 911                                         allow_create, replace_required);
 912
 913                         if (IS_ERR(sn)) {
 914                                 err = PTR_ERR(sn);
 915                                 goto st_failure;
 916                         }
 917                 }
 918
 919                 if (!fn->leaf) {
 920                         fn->leaf = rt;
 921                         atomic_inc(&rt->rt6i_ref);
 922                 }
 923                 fn = sn;
 924         }
 925 #endif
 926
 927         err = fib6_add_rt2node(fn, rt, info);
 928         if (!err) {
 929                 fib6_start_gc(info->nl_net, rt);
 930                 if (!(rt->rt6i_flags & RTF_CACHE))
 931                         fib6_prune_clones(info->nl_net, pn, rt);
 932         }
 933
 934 out:
 935         if (err) {
 936 #ifdef CONFIG_IPV6_SUBTREES
 937                 /*
 938                  * If fib6_add_1 has cleared the old leaf pointer in the
 939                  * super-tree leaf node we have to find a new one for it.
 940                  */
 941                 if (pn != fn && pn->leaf == rt) {
 942                         pn->leaf = NULL;
 943                         atomic_dec(&rt->rt6i_ref);
 944                 }
 945                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
 946                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
 947 #if RT6_DEBUG >= 2
 948                         if (!pn->leaf) {
 949                                 WARN_ON(pn->leaf == NULL);
 950                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
 951                         }
 952 #endif
 953                         atomic_inc(&pn->leaf->rt6i_ref);
 954                 }
 955 #endif
 956                 dst_free(&rt->dst);
 957         }
 958         return err;
 959
 960 #ifdef CONFIG_IPV6_SUBTREES
 961         /* Subtree creation failed, probably main tree node
 962            is orphan. If it is, shoot it.
 963          */
 964 st_failure:
 965         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
 966                 fib6_repair_tree(info->nl_net, fn);
 967         dst_free(&rt->dst);
 968         return err;
 969 #endif
 970 }
 971
 972 /*
 973  *      Routing tree lookup
 974  *
 975  */
 976
 977 struct lookup_args {
 978         int                     offset;         /* key offset on rt6_info       */
 979         const struct in6_addr   *addr;          /* search key                   */
 980 };
 981
 982 static struct fib6_node * fib6_lookup_1(struct fib6_node *root,
 983                                         struct lookup_args *args)
 984 {
 985         struct fib6_node *fn;
 986         __be32 dir;
 987
 988         if (unlikely(args->offset == 0))
 989                 return NULL;
 990
 991         /*
 992          *      Descend on a tree
 993          */
 994
 995         fn = root;
 996
 997         for (;;) {
 998                 struct fib6_node *next;
 999
1000                 dir = addr_bit_set(args->addr, fn->fn_bit);
1001
1002                 next = dir ? fn->right : fn->left;
1003
1004                 if (next) {
1005                         fn = next;
1006                         continue;
1007                 }
1008                 break;
1009         }
1010
1011         while (fn) {
1012                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1013                         struct rt6key *key;
1014
1015                         key = (struct rt6key *) ((u8 *) fn->leaf +
1016                                                  args->offset);
1017
1018                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1019 #ifdef CONFIG_IPV6_SUBTREES
1020                                 if (fn->subtree) {
1021                                         struct fib6_node *sfn;
1022                                         sfn = fib6_lookup_1(fn->subtree,
1023                                                             args + 1);
1024                                         if (!sfn)
1025                                                 goto backtrack;
1026                                         fn = sfn;
1027                                 }
1028 #endif
1029                                 if (fn->fn_flags & RTN_RTINFO)
1030                                         return fn;
1031                         }
1032                 }
1033 #ifdef CONFIG_IPV6_SUBTREES
1034 backtrack:
1035 #endif
1036                 if (fn->fn_flags & RTN_ROOT)
1037                         break;
1038
1039                 fn = fn->parent;
1040         }
1041
1042         return NULL;
1043 }
1044
1045 struct fib6_node * fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1046                                const struct in6_addr *saddr)
1047 {
1048         struct fib6_node *fn;
1049         struct lookup_args args[] = {
1050                 {
1051                         .offset = offsetof(struct rt6_info, rt6i_dst),
1052                         .addr = daddr,
1053                 },
1054 #ifdef CONFIG_IPV6_SUBTREES
1055                 {
1056                         .offset = offsetof(struct rt6_info, rt6i_src),
1057                         .addr = saddr,
1058                 },
1059 #endif
1060                 {
1061                         .offset = 0,    /* sentinel */
1062                 }
1063         };
1064
1065         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1066         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1067                 fn = root;
1068
1069         return fn;
1070 }
1071
1072 /*
1073  *      Get node with specified destination prefix (and source prefix,
1074  *      if subtrees are used)
1075  */
1076
1077
1078 static struct fib6_node * fib6_locate_1(struct fib6_node *root,
1079                                         const struct in6_addr *addr,
1080                                         int plen, int offset)
1081 {
1082         struct fib6_node *fn;
1083
1084         for (fn = root; fn ; ) {
1085                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1086
1087                 /*
1088                  *      Prefix match
1089                  */
1090                 if (plen < fn->fn_bit ||
1091                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1092                         return NULL;
1093
1094                 if (plen == fn->fn_bit)
1095                         return fn;
1096
1097                 /*
1098                  *      We have more bits to go
1099                  */
1100                 if (addr_bit_set(addr, fn->fn_bit))
1101                         fn = fn->right;
1102                 else
1103                         fn = fn->left;
1104         }
1105         return NULL;
1106 }
1107
1108 struct fib6_node * fib6_locate(struct fib6_node *root,
1109                                const struct in6_addr *daddr, int dst_len,
1110                                const struct in6_addr *saddr, int src_len)
1111 {
1112         struct fib6_node *fn;
1113
1114         fn = fib6_locate_1(root, daddr, dst_len,
1115                            offsetof(struct rt6_info, rt6i_dst));
1116
1117 #ifdef CONFIG_IPV6_SUBTREES
1118         if (src_len) {
1119                 WARN_ON(saddr == NULL);
1120                 if (fn && fn->subtree)
1121                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1122                                            offsetof(struct rt6_info, rt6i_src));
1123         }
1124 #endif
1125
1126         if (fn && fn->fn_flags & RTN_RTINFO)
1127                 return fn;
1128
1129         return NULL;
1130 }
1131
1132
1133 /*
1134  *      Deletion
1135  *
1136  */
1137
1138 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1139 {
1140         if (fn->fn_flags & RTN_ROOT)
1141                 return net->ipv6.ip6_null_entry;
1142
1143         while (fn) {
1144                 if (fn->left)
1145                         return fn->left->leaf;
1146                 if (fn->right)
1147                         return fn->right->leaf;
1148
1149                 fn = FIB6_SUBTREE(fn);
1150         }
1151         return NULL;
1152 }
1153
1154 /*
1155  *      Called to trim the tree of intermediate nodes when possible. "fn"
1156  *      is the node we want to try and remove.
1157  */
1158
1159 static struct fib6_node *fib6_repair_tree(struct net *net,
1160                                            struct fib6_node *fn)
1161 {
1162         int children;
1163         int nstate;
1164         struct fib6_node *child, *pn;
1165         struct fib6_walker_t *w;
1166         int iter = 0;
1167
1168         for (;;) {
1169                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1170                 iter++;
1171
1172                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1173                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1174                 WARN_ON(fn->leaf != NULL);
1175
1176                 children = 0;
1177                 child = NULL;
1178                 if (fn->right) child = fn->right, children |= 1;
1179                 if (fn->left) child = fn->left, children |= 2;
1180
1181                 if (children == 3 || FIB6_SUBTREE(fn)
1182 #ifdef CONFIG_IPV6_SUBTREES
1183                     /* Subtree root (i.e. fn) may have one child */
1184                     || (children && fn->fn_flags & RTN_ROOT)
1185 #endif
1186                     ) {
1187                         fn->leaf = fib6_find_prefix(net, fn);
1188 #if RT6_DEBUG >= 2
1189                         if (!fn->leaf) {
1190                                 WARN_ON(!fn->leaf);
1191                                 fn->leaf = net->ipv6.ip6_null_entry;
1192                         }
1193 #endif
1194                         atomic_inc(&fn->leaf->rt6i_ref);
1195                         return fn->parent;
1196                 }
1197
1198                 pn = fn->parent;
1199 #ifdef CONFIG_IPV6_SUBTREES
1200                 if (FIB6_SUBTREE(pn) == fn) {
1201                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1202                         FIB6_SUBTREE(pn) = NULL;
1203                         nstate = FWS_L;
1204                 } else {
1205                         WARN_ON(fn->fn_flags & RTN_ROOT);
1206 #endif
1207                         if (pn->right == fn) pn->right = child;
1208                         else if (pn->left == fn) pn->left = child;
1209 #if RT6_DEBUG >= 2
1210                         else
1211                                 WARN_ON(1);
1212 #endif
1213                         if (child)
1214                                 child->parent = pn;
1215                         nstate = FWS_R;
1216 #ifdef CONFIG_IPV6_SUBTREES
1217                 }
1218 #endif
1219
1220                 read_lock(&fib6_walker_lock);
1221                 FOR_WALKERS(w) {
1222                         if (!child) {
1223                                 if (w->root == fn) {
1224                                         w->root = w->node = NULL;
1225                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1226                                 } else if (w->node == fn) {
1227                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1228                                         w->node = pn;
1229                                         w->state = nstate;
1230                                 }
1231                         } else {
1232                                 if (w->root == fn) {
1233                                         w->root = child;
1234                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1235                                 }
1236                                 if (w->node == fn) {
1237                                         w->node = child;
1238                                         if (children&2) {
1239                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1240                                                 w->state = w->state>=FWS_R ? FWS_U : FWS_INIT;
1241                                         } else {
1242                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1243                                                 w->state = w->state>=FWS_C ? FWS_U : FWS_INIT;
1244                                         }
1245                                 }
1246                         }
1247                 }
1248                 read_unlock(&fib6_walker_lock);
1249
1250                 node_free(fn);
1251                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1252                         return pn;
1253
1254                 rt6_release(pn->leaf);
1255                 pn->leaf = NULL;
1256                 fn = pn;
1257         }
1258 }
1259
1260 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1261                            struct nl_info *info)
1262 {
1263         struct fib6_walker_t *w;
1264         struct rt6_info *rt = *rtp;
1265         struct net *net = info->nl_net;
1266
1267         RT6_TRACE("fib6_del_route\n");
1268
1269         /* Unlink it */
1270         *rtp = rt->dst.rt6_next;
1271         rt->rt6i_node = NULL;
1272         net->ipv6.rt6_stats->fib_rt_entries--;
1273         net->ipv6.rt6_stats->fib_discarded_routes++;
1274
1275         /* Reset round-robin state, if necessary */
1276         if (fn->rr_ptr == rt)
1277                 fn->rr_ptr = NULL;
1278
1279         /* Remove this entry from other siblings */
1280         if (rt->rt6i_nsiblings) {
1281                 struct rt6_info *sibling, *next_sibling;
1282
1283                 list_for_each_entry_safe(sibling, next_sibling,
1284                                          &rt->rt6i_siblings, rt6i_siblings)
1285                         sibling->rt6i_nsiblings--;
1286                 rt->rt6i_nsiblings = 0;
1287                 list_del_init(&rt->rt6i_siblings);
1288         }
1289
1290         /* Adjust walkers */
1291         read_lock(&fib6_walker_lock);
1292         FOR_WALKERS(w) {
1293                 if (w->state == FWS_C && w->leaf == rt) {
1294                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1295                         w->leaf = rt->dst.rt6_next;
1296                         if (!w->leaf)
1297                                 w->state = FWS_U;
1298                 }
1299         }
1300         read_unlock(&fib6_walker_lock);
1301
1302         rt->dst.rt6_next = NULL;
1303
1304         /* If it was last route, expunge its radix tree node */
1305         if (!fn->leaf) {
1306                 fn->fn_flags &= ~RTN_RTINFO;
1307                 net->ipv6.rt6_stats->fib_route_nodes--;
1308                 fn = fib6_repair_tree(net, fn);
1309         }
1310
1311         fib6_purge_rt(rt, fn, net);
1312
1313         inet6_rt_notify(RTM_DELROUTE, rt, info);
1314         rt6_release(rt);
1315 }
1316
1317 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1318 {
1319         struct net *net = info->nl_net;
1320         struct fib6_node *fn = rt->rt6i_node;
1321         struct rt6_info **rtp;
1322
1323 #if RT6_DEBUG >= 2
1324         if (rt->dst.obsolete>0) {
1325                 WARN_ON(fn != NULL);
1326                 return -ENOENT;
1327         }
1328 #endif
1329         if (!fn || rt == net->ipv6.ip6_null_entry)
1330                 return -ENOENT;
1331
1332         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1333
1334         if (!(rt->rt6i_flags & RTF_CACHE)) {
1335                 struct fib6_node *pn = fn;
1336 #ifdef CONFIG_IPV6_SUBTREES
1337                 /* clones of this route might be in another subtree */
1338                 if (rt->rt6i_src.plen) {
1339                         while (!(pn->fn_flags & RTN_ROOT))
1340                                 pn = pn->parent;
1341                         pn = pn->parent;
1342                 }
1343 #endif
1344                 fib6_prune_clones(info->nl_net, pn, rt);
1345         }
1346
1347         /*
1348          *      Walk the leaf entries looking for ourself
1349          */
1350
1351         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1352                 if (*rtp == rt) {
1353                         fib6_del_route(fn, rtp, info);
1354                         return 0;
1355                 }
1356         }
1357         return -ENOENT;
1358 }
1359
1360 /*
1361  *      Tree traversal function.
1362  *
1363  *      Certainly, it is not interrupt safe.
1364  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1365  *      It means, that we can modify tree during walking
1366  *      and use this function for garbage collection, clone pruning,
1367  *      cleaning tree when a device goes down etc. etc.
1368  *
1369  *      It guarantees that every node will be traversed,
1370  *      and that it will be traversed only once.
1371  *
1372  *      Callback function w->func may return:
1373  *      0 -> continue walking.
1374  *      positive value -> walking is suspended (used by tree dumps,
1375  *      and probably by gc, if it will be split to several slices)
1376  *      negative value -> terminate walking.
1377  *
1378  *      The function itself returns:
1379  *      0   -> walk is complete.
1380  *      >0  -> walk is incomplete (i.e. suspended)
1381  *      <0  -> walk is terminated by an error.
1382  */
1383
1384 static int fib6_walk_continue(struct fib6_walker_t *w)
1385 {
1386         struct fib6_node *fn, *pn;
1387
1388         for (;;) {
1389                 fn = w->node;
1390                 if (!fn)
1391                         return 0;
1392
1393                 if (w->prune && fn != w->root &&
1394                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1395                         w->state = FWS_C;
1396                         w->leaf = fn->leaf;
1397                 }
1398                 switch (w->state) {
1399 #ifdef CONFIG_IPV6_SUBTREES
1400                 case FWS_S:
1401                         if (FIB6_SUBTREE(fn)) {
1402                                 w->node = FIB6_SUBTREE(fn);
1403                                 continue;
1404                         }
1405                         w->state = FWS_L;
1406 #endif
1407                 case FWS_L:
1408                         if (fn->left) {
1409                                 w->node = fn->left;
1410                                 w->state = FWS_INIT;
1411                                 continue;
1412                         }
1413                         w->state = FWS_R;
1414                 case FWS_R:
1415                         if (fn->right) {
1416                                 w->node = fn->right;
1417                                 w->state = FWS_INIT;
1418                                 continue;
1419                         }
1420                         w->state = FWS_C;
1421                         w->leaf = fn->leaf;
1422                 case FWS_C:
1423                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1424                                 int err;
1425
1426                                 if (w->skip) {
1427                                         w->skip--;
1428                                         goto skip;
1429                                 }
1430
1431                                 err = w->func(w);
1432                                 if (err)
1433                                         return err;
1434
1435                                 w->count++;
1436                                 continue;
1437                         }
1438 skip:
1439                         w->state = FWS_U;
1440                 case FWS_U:
1441                         if (fn == w->root)
1442                                 return 0;
1443                         pn = fn->parent;
1444                         w->node = pn;
1445 #ifdef CONFIG_IPV6_SUBTREES
1446                         if (FIB6_SUBTREE(pn) == fn) {
1447                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1448                                 w->state = FWS_L;
1449                                 continue;
1450                         }
1451 #endif
1452                         if (pn->left == fn) {
1453                                 w->state = FWS_R;
1454                                 continue;
1455                         }
1456                         if (pn->right == fn) {
1457                                 w->state = FWS_C;
1458                                 w->leaf = w->node->leaf;
1459                                 continue;
1460                         }
1461 #if RT6_DEBUG >= 2
1462                         WARN_ON(1);
1463 #endif
1464                 }
1465         }
1466 }
1467
1468 static int fib6_walk(struct fib6_walker_t *w)
1469 {
1470         int res;
1471
1472         w->state = FWS_INIT;
1473         w->node = w->root;
1474
1475         fib6_walker_link(w);
1476         res = fib6_walk_continue(w);
1477         if (res <= 0)
1478                 fib6_walker_unlink(w);
1479         return res;
1480 }
1481
1482 static int fib6_clean_node(struct fib6_walker_t *w)
1483 {
1484         int res;
1485         struct rt6_info *rt;
1486         struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1487         struct nl_info info = {
1488                 .nl_net = c->net,
1489         };
1490
1491         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1492                 res = c->func(rt, c->arg);
1493                 if (res < 0) {
1494                         w->leaf = rt;
1495                         res = fib6_del(rt, &info);
1496                         if (res) {
1497 #if RT6_DEBUG >= 2
1498                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1499                                          __func__, rt, rt->rt6i_node, res);
1500 #endif
1501                                 continue;
1502                         }
1503                         return 0;
1504                 }
1505                 WARN_ON(res != 0);
1506         }
1507         w->leaf = rt;
1508         return 0;
1509 }
1510
1511 /*
1512  *      Convenient frontend to tree walker.
1513  *
1514  *      func is called on each route.
1515  *              It may return -1 -> delete this route.
1516  *                            0  -> continue walking
1517  *
1518  *      prune==1 -> only immediate children of node (certainly,
1519  *      ignoring pure split nodes) will be scanned.
1520  */
1521
1522 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1523                             int (*func)(struct rt6_info *, void *arg),
1524                             int prune, void *arg)
1525 {
1526         struct fib6_cleaner_t c;
1527
1528         c.w.root = root;
1529         c.w.func = fib6_clean_node;
1530         c.w.prune = prune;
1531         c.w.count = 0;
1532         c.w.skip = 0;
1533         c.func = func;
1534         c.arg = arg;
1535         c.net = net;
1536
1537         fib6_walk(&c.w);
1538 }
1539
1540 void fib6_clean_all_ro(struct net *net, int (*func)(struct rt6_info *, void *arg),
1541                     int prune, void *arg)
1542 {
1543         struct fib6_table *table;
1544         struct hlist_head *head;
1545         unsigned int h;
1546
1547         rcu_read_lock();
1548         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1549                 head = &net->ipv6.fib_table_hash[h];
1550                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1551                         read_lock_bh(&table->tb6_lock);
1552                         fib6_clean_tree(net, &table->tb6_root,
1553                                         func, prune, arg);
1554                         read_unlock_bh(&table->tb6_lock);
1555                 }
1556         }
1557         rcu_read_unlock();
1558 }
1559 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1560                     int prune, void *arg)
1561 {
1562         struct fib6_table *table;
1563         struct hlist_head *head;
1564         unsigned int h;
1565
1566         rcu_read_lock();
1567         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1568                 head = &net->ipv6.fib_table_hash[h];
1569                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1570                         write_lock_bh(&table->tb6_lock);
1571                         fib6_clean_tree(net, &table->tb6_root,
1572                                         func, prune, arg);
1573                         write_unlock_bh(&table->tb6_lock);
1574                 }
1575         }
1576         rcu_read_unlock();
1577 }
1578
1579 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1580 {
1581         if (rt->rt6i_flags & RTF_CACHE) {
1582                 RT6_TRACE("pruning clone %p\n", rt);
1583                 return -1;
1584         }
1585
1586         return 0;
1587 }
1588
1589 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1590                               struct rt6_info *rt)
1591 {
1592         fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1593 }
1594
1595 /*
1596  *      Garbage collection
1597  */
1598
1599 static struct fib6_gc_args
1600 {
1601         int                     timeout;
1602         int                     more;
1603 } gc_args;
1604
1605 static int fib6_age(struct rt6_info *rt, void *arg)
1606 {
1607         unsigned long now = jiffies;
1608
1609         /*
1610          *      check addrconf expiration here.
1611          *      Routes are expired even if they are in use.
1612          *
1613          *      Also age clones. Note, that clones are aged out
1614          *      only if they are not in use now.
1615          */
1616
1617         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1618                 if (time_after(now, rt->dst.expires)) {
1619                         RT6_TRACE("expiring %p\n", rt);
1620                         return -1;
1621                 }
1622                 gc_args.more++;
1623         } else if (rt->rt6i_flags & RTF_CACHE) {
1624                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1625                     time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1626                         RT6_TRACE("aging clone %p\n", rt);
1627                         return -1;
1628                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1629                         struct neighbour *neigh;
1630                         __u8 neigh_flags = 0;
1631
1632                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1633                         if (neigh) {
1634                                 neigh_flags = neigh->flags;
1635                                 neigh_release(neigh);
1636                         }
1637                         if (!(neigh_flags & NTF_ROUTER)) {
1638                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1639                                           rt);
1640                                 return -1;
1641                         }
1642                 }
1643                 gc_args.more++;
1644         }
1645
1646         return 0;
1647 }
1648
1649 static DEFINE_SPINLOCK(fib6_gc_lock);
1650
1651 void fib6_run_gc(unsigned long expires, struct net *net)
1652 {
1653         if (expires != ~0UL) {
1654                 spin_lock_bh(&fib6_gc_lock);
1655                 gc_args.timeout = expires ? (int)expires :
1656                         net->ipv6.sysctl.ip6_rt_gc_interval;
1657         } else {
1658                 if (!spin_trylock_bh(&fib6_gc_lock)) {
1659                         mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1660                         return;
1661                 }
1662                 gc_args.timeout = net->ipv6.sysctl.ip6_rt_gc_interval;
1663         }
1664
1665         gc_args.more = icmp6_dst_gc();
1666
1667         fib6_clean_all(net, fib6_age, 0, NULL);
1668
1669         if (gc_args.more)
1670                 mod_timer(&net->ipv6.ip6_fib_timer,
1671                           round_jiffies(jiffies
1672                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1673         else
1674                 del_timer(&net->ipv6.ip6_fib_timer);
1675         spin_unlock_bh(&fib6_gc_lock);
1676 }
1677
1678 static void fib6_gc_timer_cb(unsigned long arg)
1679 {
1680         fib6_run_gc(0, (struct net *)arg);
1681 }
1682
1683 static int __net_init fib6_net_init(struct net *net)
1684 {
1685         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1686
1687         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1688
1689         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1690         if (!net->ipv6.rt6_stats)
1691                 goto out_timer;
1692
1693         /* Avoid false sharing : Use at least a full cache line */
1694         size = max_t(size_t, size, L1_CACHE_BYTES);
1695
1696         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1697         if (!net->ipv6.fib_table_hash)
1698                 goto out_rt6_stats;
1699
1700         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1701                                           GFP_KERNEL);
1702         if (!net->ipv6.fib6_main_tbl)
1703                 goto out_fib_table_hash;
1704
1705         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1706         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1707         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1708                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1709         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1710
1711 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1712         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1713                                            GFP_KERNEL);
1714         if (!net->ipv6.fib6_local_tbl)
1715                 goto out_fib6_main_tbl;
1716         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1717         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1718         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1719                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1720         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1721 #endif
1722         fib6_tables_init(net);
1723
1724         return 0;
1725
1726 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1727 out_fib6_main_tbl:
1728         kfree(net->ipv6.fib6_main_tbl);
1729 #endif
1730 out_fib_table_hash:
1731         kfree(net->ipv6.fib_table_hash);
1732 out_rt6_stats:
1733         kfree(net->ipv6.rt6_stats);
1734 out_timer:
1735         return -ENOMEM;
1736  }
1737
1738 static void fib6_net_exit(struct net *net)
1739 {
1740         rt6_ifdown(net, NULL);
1741         del_timer_sync(&net->ipv6.ip6_fib_timer);
1742
1743 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1744         inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1745         kfree(net->ipv6.fib6_local_tbl);
1746 #endif
1747         inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1748         kfree(net->ipv6.fib6_main_tbl);
1749         kfree(net->ipv6.fib_table_hash);
1750         kfree(net->ipv6.rt6_stats);
1751 }
1752
1753 static struct pernet_operations fib6_net_ops = {
1754         .init = fib6_net_init,
1755         .exit = fib6_net_exit,
1756 };
1757
1758 int __init fib6_init(void)
1759 {
1760         int ret = -ENOMEM;
1761
1762         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1763                                            sizeof(struct fib6_node),
1764                                            0, SLAB_HWCACHE_ALIGN,
1765                                            NULL);
1766         if (!fib6_node_kmem)
1767                 goto out;
1768
1769         ret = register_pernet_subsys(&fib6_net_ops);
1770         if (ret)
1771                 goto out_kmem_cache_create;
1772
1773         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1774                               NULL);
1775         if (ret)
1776                 goto out_unregister_subsys;
1777 out:
1778         return ret;
1779
1780 out_unregister_subsys:
1781         unregister_pernet_subsys(&fib6_net_ops);
1782 out_kmem_cache_create:
1783         kmem_cache_destroy(fib6_node_kmem);
1784         goto out;
1785 }
1786
1787 void fib6_gc_cleanup(void)
1788 {
1789         unregister_pernet_subsys(&fib6_net_ops);
1790         kmem_cache_destroy(fib6_node_kmem);
1791 }