net/ipv6/ip6_fib.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      Forwarding Information Database
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <[email protected]>
   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         struct rt6_info **fallback_ins = NULL;
 697         int replace = (info->nlh &&
 698                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 699         int add = (!info->nlh ||
 700                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 701         int found = 0;
 702         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 703         int err;
 704
 705         ins = &fn->leaf;
 706
 707         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 708                 /*
 709                  *      Search for duplicates
 710                  */
 711
 712                 if (iter->rt6i_metric == rt->rt6i_metric) {
 713                         /*
 714                          *      Same priority level
 715                          */
 716                         if (info->nlh &&
 717                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 718                                 return -EEXIST;
 719                         if (replace) {
 720                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 721                                         found++;
 722                                         break;
 723                                 }
 724                                 if (rt_can_ecmp)
 725                                         fallback_ins = fallback_ins ?: ins;
 726                                 goto next_iter;
 727                         }
 728
 729                         if (iter->dst.dev == rt->dst.dev &&
 730                             iter->rt6i_idev == rt->rt6i_idev &&
 731                             ipv6_addr_equal(&iter->rt6i_gateway,
 732                                             &rt->rt6i_gateway)) {
 733                                 if (rt->rt6i_nsiblings)
 734                                         rt->rt6i_nsiblings = 0;
 735                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 736                                         return -EEXIST;
 737                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 738                                         rt6_clean_expires(iter);
 739                                 else
 740                                         rt6_set_expires(iter, rt->dst.expires);
 741                                 return -EEXIST;
 742                         }
 743                         /* If we have the same destination and the same metric,
 744                          * but not the same gateway, then the route we try to
 745                          * add is sibling to this route, increment our counter
 746                          * of siblings, and later we will add our route to the
 747                          * list.
 748                          * Only static routes (which don't have flag
 749                          * RTF_EXPIRES) are used for ECMPv6.
 750                          *
 751                          * To avoid long list, we only had siblings if the
 752                          * route have a gateway.
 753                          */
 754                         if (rt_can_ecmp &&
 755                             rt6_qualify_for_ecmp(iter))
 756                                 rt->rt6i_nsiblings++;
 757                 }
 758
 759                 if (iter->rt6i_metric > rt->rt6i_metric)
 760                         break;
 761
 762 next_iter:
 763                 ins = &iter->dst.rt6_next;
 764         }
 765
 766         if (fallback_ins && !found) {
 767                 /* No ECMP-able route found, replace first non-ECMP one */
 768                 ins = fallback_ins;
 769                 iter = *ins;
 770                 found++;
 771         }
 772
 773         /* Reset round-robin state, if necessary */
 774         if (ins == &fn->leaf)
 775                 fn->rr_ptr = NULL;
 776
 777         /* Link this route to others same route. */
 778         if (rt->rt6i_nsiblings) {
 779                 unsigned int rt6i_nsiblings;
 780                 struct rt6_info *sibling, *temp_sibling;
 781
 782                 /* Find the first route that have the same metric */
 783                 sibling = fn->leaf;
 784                 while (sibling) {
 785                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 786                             rt6_qualify_for_ecmp(sibling)) {
 787                                 list_add_tail(&rt->rt6i_siblings,
 788                                               &sibling->rt6i_siblings);
 789                                 break;
 790                         }
 791                         sibling = sibling->dst.rt6_next;
 792                 }
 793                 /* For each sibling in the list, increment the counter of
 794                  * siblings. BUG() if counters does not match, list of siblings
 795                  * is broken!
 796                  */
 797                 rt6i_nsiblings = 0;
 798                 list_for_each_entry_safe(sibling, temp_sibling,
 799                                          &rt->rt6i_siblings, rt6i_siblings) {
 800                         sibling->rt6i_nsiblings++;
 801                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 802                         rt6i_nsiblings++;
 803                 }
 804                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 805         }
 806
 807         /*
 808          *      insert node
 809          */
 810         if (!replace) {
 811                 if (!add)
 812                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 813
 814 add:
 815                 err = fib6_commit_metrics(&rt->dst, mxc);
 816                 if (err)
 817                         return err;
 818
 819                 rt->dst.rt6_next = iter;
 820                 *ins = rt;
 821                 rt->rt6i_node = fn;
 822                 atomic_inc(&rt->rt6i_ref);
 823                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 824                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 825
 826                 if (!(fn->fn_flags & RTN_RTINFO)) {
 827                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 828                         fn->fn_flags |= RTN_RTINFO;
 829                 }
 830
 831         } else {
 832                 int nsiblings;
 833
 834                 if (!found) {
 835                         if (add)
 836                                 goto add;
 837                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
 838                         return -ENOENT;
 839                 }
 840
 841                 err = fib6_commit_metrics(&rt->dst, mxc);
 842                 if (err)
 843                         return err;
 844
 845                 *ins = rt;
 846                 rt->rt6i_node = fn;
 847                 rt->dst.rt6_next = iter->dst.rt6_next;
 848                 atomic_inc(&rt->rt6i_ref);
 849                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 850                 if (!(fn->fn_flags & RTN_RTINFO)) {
 851                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 852                         fn->fn_flags |= RTN_RTINFO;
 853                 }
 854                 nsiblings = iter->rt6i_nsiblings;
 855                 fib6_purge_rt(iter, fn, info->nl_net);
 856                 rt6_release(iter);
 857
 858                 if (nsiblings) {
 859                         /* Replacing an ECMP route, remove all siblings */
 860                         ins = &rt->dst.rt6_next;
 861                         iter = *ins;
 862                         while (iter) {
 863                                 if (rt6_qualify_for_ecmp(iter)) {
 864                                         *ins = iter->dst.rt6_next;
 865                                         fib6_purge_rt(iter, fn, info->nl_net);
 866                                         rt6_release(iter);
 867                                         nsiblings--;
 868                                 } else {
 869                                         ins = &iter->dst.rt6_next;
 870                                 }
 871                                 iter = *ins;
 872                         }
 873                         WARN_ON(nsiblings != 0);
 874                 }
 875         }
 876
 877         return 0;
 878 }
 879
 880 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
 881 {
 882         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
 883             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
 884                 mod_timer(&net->ipv6.ip6_fib_timer,
 885                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 886 }
 887
 888 void fib6_force_start_gc(struct net *net)
 889 {
 890         if (!timer_pending(&net->ipv6.ip6_fib_timer))
 891                 mod_timer(&net->ipv6.ip6_fib_timer,
 892                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 893 }
 894
 895 /*
 896  *      Add routing information to the routing tree.
 897  *      <destination addr>/<source addr>
 898  *      with source addr info in sub-trees
 899  */
 900
 901 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
 902              struct nl_info *info, struct mx6_config *mxc)
 903 {
 904         struct fib6_node *fn, *pn = NULL;
 905         int err = -ENOMEM;
 906         int allow_create = 1;
 907         int replace_required = 0;
 908         int sernum = fib6_new_sernum(info->nl_net);
 909
 910         if (info->nlh) {
 911                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
 912                         allow_create = 0;
 913                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
 914                         replace_required = 1;
 915         }
 916         if (!allow_create && !replace_required)
 917                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 918
 919         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
 920                         offsetof(struct rt6_info, rt6i_dst), allow_create,
 921                         replace_required, sernum);
 922         if (IS_ERR(fn)) {
 923                 err = PTR_ERR(fn);
 924                 fn = NULL;
 925                 goto out;
 926         }
 927
 928         pn = fn;
 929
 930 #ifdef CONFIG_IPV6_SUBTREES
 931         if (rt->rt6i_src.plen) {
 932                 struct fib6_node *sn;
 933
 934                 if (!fn->subtree) {
 935                         struct fib6_node *sfn;
 936
 937                         /*
 938                          * Create subtree.
 939                          *
 940                          *              fn[main tree]
 941                          *              |
 942                          *              sfn[subtree root]
 943                          *                 \
 944                          *                  sn[new leaf node]
 945                          */
 946
 947                         /* Create subtree root node */
 948                         sfn = node_alloc();
 949                         if (!sfn)
 950                                 goto st_failure;
 951
 952                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
 953                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
 954                         sfn->fn_flags = RTN_ROOT;
 955                         sfn->fn_sernum = sernum;
 956
 957                         /* Now add the first leaf node to new subtree */
 958
 959                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
 960                                         rt->rt6i_src.plen,
 961                                         offsetof(struct rt6_info, rt6i_src),
 962                                         allow_create, replace_required, sernum);
 963
 964                         if (IS_ERR(sn)) {
 965                                 /* If it is failed, discard just allocated
 966                                    root, and then (in st_failure) stale node
 967                                    in main tree.
 968                                  */
 969                                 node_free(sfn);
 970                                 err = PTR_ERR(sn);
 971                                 goto st_failure;
 972                         }
 973
 974                         /* Now link new subtree to main tree */
 975                         sfn->parent = fn;
 976                         fn->subtree = sfn;
 977                 } else {
 978                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
 979                                         rt->rt6i_src.plen,
 980                                         offsetof(struct rt6_info, rt6i_src),
 981                                         allow_create, replace_required, sernum);
 982
 983                         if (IS_ERR(sn)) {
 984                                 err = PTR_ERR(sn);
 985                                 goto st_failure;
 986                         }
 987                 }
 988
 989                 if (!fn->leaf) {
 990                         fn->leaf = rt;
 991                         atomic_inc(&rt->rt6i_ref);
 992                 }
 993                 fn = sn;
 994         }
 995 #endif
 996
 997         err = fib6_add_rt2node(fn, rt, info, mxc);
 998         if (!err) {
 999                 fib6_start_gc(info->nl_net, rt);
1000                 if (!(rt->rt6i_flags & RTF_CACHE))
1001                         fib6_prune_clones(info->nl_net, pn);
1002         }
1003
1004 out:
1005         if (err) {
1006 #ifdef CONFIG_IPV6_SUBTREES
1007                 /*
1008                  * If fib6_add_1 has cleared the old leaf pointer in the
1009                  * super-tree leaf node we have to find a new one for it.
1010                  */
1011                 if (pn != fn && pn->leaf == rt) {
1012                         pn->leaf = NULL;
1013                         atomic_dec(&rt->rt6i_ref);
1014                 }
1015                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1016                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
1017 #if RT6_DEBUG >= 2
1018                         if (!pn->leaf) {
1019                                 WARN_ON(pn->leaf == NULL);
1020                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1021                         }
1022 #endif
1023                         atomic_inc(&pn->leaf->rt6i_ref);
1024                 }
1025 #endif
1026                 dst_free(&rt->dst);
1027         }
1028         return err;
1029
1030 #ifdef CONFIG_IPV6_SUBTREES
1031         /* Subtree creation failed, probably main tree node
1032            is orphan. If it is, shoot it.
1033          */
1034 st_failure:
1035         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1036                 fib6_repair_tree(info->nl_net, fn);
1037         dst_free(&rt->dst);
1038         return err;
1039 #endif
1040 }
1041
1042 /*
1043  *      Routing tree lookup
1044  *
1045  */
1046
1047 struct lookup_args {
1048         int                     offset;         /* key offset on rt6_info       */
1049         const struct in6_addr   *addr;          /* search key                   */
1050 };
1051
1052 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1053                                        struct lookup_args *args)
1054 {
1055         struct fib6_node *fn;
1056         __be32 dir;
1057
1058         if (unlikely(args->offset == 0))
1059                 return NULL;
1060
1061         /*
1062          *      Descend on a tree
1063          */
1064
1065         fn = root;
1066
1067         for (;;) {
1068                 struct fib6_node *next;
1069
1070                 dir = addr_bit_set(args->addr, fn->fn_bit);
1071
1072                 next = dir ? fn->right : fn->left;
1073
1074                 if (next) {
1075                         fn = next;
1076                         continue;
1077                 }
1078                 break;
1079         }
1080
1081         while (fn) {
1082                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1083                         struct rt6key *key;
1084
1085                         key = (struct rt6key *) ((u8 *) fn->leaf +
1086                                                  args->offset);
1087
1088                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1089 #ifdef CONFIG_IPV6_SUBTREES
1090                                 if (fn->subtree) {
1091                                         struct fib6_node *sfn;
1092                                         sfn = fib6_lookup_1(fn->subtree,
1093                                                             args + 1);
1094                                         if (!sfn)
1095                                                 goto backtrack;
1096                                         fn = sfn;
1097                                 }
1098 #endif
1099                                 if (fn->fn_flags & RTN_RTINFO)
1100                                         return fn;
1101                         }
1102                 }
1103 #ifdef CONFIG_IPV6_SUBTREES
1104 backtrack:
1105 #endif
1106                 if (fn->fn_flags & RTN_ROOT)
1107                         break;
1108
1109                 fn = fn->parent;
1110         }
1111
1112         return NULL;
1113 }
1114
1115 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1116                               const struct in6_addr *saddr)
1117 {
1118         struct fib6_node *fn;
1119         struct lookup_args args[] = {
1120                 {
1121                         .offset = offsetof(struct rt6_info, rt6i_dst),
1122                         .addr = daddr,
1123                 },
1124 #ifdef CONFIG_IPV6_SUBTREES
1125                 {
1126                         .offset = offsetof(struct rt6_info, rt6i_src),
1127                         .addr = saddr,
1128                 },
1129 #endif
1130                 {
1131                         .offset = 0,    /* sentinel */
1132                 }
1133         };
1134
1135         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1136         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1137                 fn = root;
1138
1139         return fn;
1140 }
1141
1142 /*
1143  *      Get node with specified destination prefix (and source prefix,
1144  *      if subtrees are used)
1145  */
1146
1147
1148 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1149                                        const struct in6_addr *addr,
1150                                        int plen, int offset)
1151 {
1152         struct fib6_node *fn;
1153
1154         for (fn = root; fn ; ) {
1155                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1156
1157                 /*
1158                  *      Prefix match
1159                  */
1160                 if (plen < fn->fn_bit ||
1161                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1162                         return NULL;
1163
1164                 if (plen == fn->fn_bit)
1165                         return fn;
1166
1167                 /*
1168                  *      We have more bits to go
1169                  */
1170                 if (addr_bit_set(addr, fn->fn_bit))
1171                         fn = fn->right;
1172                 else
1173                         fn = fn->left;
1174         }
1175         return NULL;
1176 }
1177
1178 struct fib6_node *fib6_locate(struct fib6_node *root,
1179                               const struct in6_addr *daddr, int dst_len,
1180                               const struct in6_addr *saddr, int src_len)
1181 {
1182         struct fib6_node *fn;
1183
1184         fn = fib6_locate_1(root, daddr, dst_len,
1185                            offsetof(struct rt6_info, rt6i_dst));
1186
1187 #ifdef CONFIG_IPV6_SUBTREES
1188         if (src_len) {
1189                 WARN_ON(saddr == NULL);
1190                 if (fn && fn->subtree)
1191                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1192                                            offsetof(struct rt6_info, rt6i_src));
1193         }
1194 #endif
1195
1196         if (fn && fn->fn_flags & RTN_RTINFO)
1197                 return fn;
1198
1199         return NULL;
1200 }
1201
1202
1203 /*
1204  *      Deletion
1205  *
1206  */
1207
1208 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1209 {
1210         if (fn->fn_flags & RTN_ROOT)
1211                 return net->ipv6.ip6_null_entry;
1212
1213         while (fn) {
1214                 if (fn->left)
1215                         return fn->left->leaf;
1216                 if (fn->right)
1217                         return fn->right->leaf;
1218
1219                 fn = FIB6_SUBTREE(fn);
1220         }
1221         return NULL;
1222 }
1223
1224 /*
1225  *      Called to trim the tree of intermediate nodes when possible. "fn"
1226  *      is the node we want to try and remove.
1227  */
1228
1229 static struct fib6_node *fib6_repair_tree(struct net *net,
1230                                            struct fib6_node *fn)
1231 {
1232         int children;
1233         int nstate;
1234         struct fib6_node *child, *pn;
1235         struct fib6_walker *w;
1236         int iter = 0;
1237
1238         for (;;) {
1239                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1240                 iter++;
1241
1242                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1243                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1244                 WARN_ON(fn->leaf);
1245
1246                 children = 0;
1247                 child = NULL;
1248                 if (fn->right)
1249                         child = fn->right, children |= 1;
1250                 if (fn->left)
1251                         child = fn->left, children |= 2;
1252
1253                 if (children == 3 || FIB6_SUBTREE(fn)
1254 #ifdef CONFIG_IPV6_SUBTREES
1255                     /* Subtree root (i.e. fn) may have one child */
1256                     || (children && fn->fn_flags & RTN_ROOT)
1257 #endif
1258                     ) {
1259                         fn->leaf = fib6_find_prefix(net, fn);
1260 #if RT6_DEBUG >= 2
1261                         if (!fn->leaf) {
1262                                 WARN_ON(!fn->leaf);
1263                                 fn->leaf = net->ipv6.ip6_null_entry;
1264                         }
1265 #endif
1266                         atomic_inc(&fn->leaf->rt6i_ref);
1267                         return fn->parent;
1268                 }
1269
1270                 pn = fn->parent;
1271 #ifdef CONFIG_IPV6_SUBTREES
1272                 if (FIB6_SUBTREE(pn) == fn) {
1273                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1274                         FIB6_SUBTREE(pn) = NULL;
1275                         nstate = FWS_L;
1276                 } else {
1277                         WARN_ON(fn->fn_flags & RTN_ROOT);
1278 #endif
1279                         if (pn->right == fn)
1280                                 pn->right = child;
1281                         else if (pn->left == fn)
1282                                 pn->left = child;
1283 #if RT6_DEBUG >= 2
1284                         else
1285                                 WARN_ON(1);
1286 #endif
1287                         if (child)
1288                                 child->parent = pn;
1289                         nstate = FWS_R;
1290 #ifdef CONFIG_IPV6_SUBTREES
1291                 }
1292 #endif
1293
1294                 read_lock(&fib6_walker_lock);
1295                 FOR_WALKERS(w) {
1296                         if (!child) {
1297                                 if (w->root == fn) {
1298                                         w->root = w->node = NULL;
1299                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1300                                 } else if (w->node == fn) {
1301                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1302                                         w->node = pn;
1303                                         w->state = nstate;
1304                                 }
1305                         } else {
1306                                 if (w->root == fn) {
1307                                         w->root = child;
1308                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1309                                 }
1310                                 if (w->node == fn) {
1311                                         w->node = child;
1312                                         if (children&2) {
1313                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1314                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1315                                         } else {
1316                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1317                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1318                                         }
1319                                 }
1320                         }
1321                 }
1322                 read_unlock(&fib6_walker_lock);
1323
1324                 node_free(fn);
1325                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1326                         return pn;
1327
1328                 rt6_release(pn->leaf);
1329                 pn->leaf = NULL;
1330                 fn = pn;
1331         }
1332 }
1333
1334 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1335                            struct nl_info *info)
1336 {
1337         struct fib6_walker *w;
1338         struct rt6_info *rt = *rtp;
1339         struct net *net = info->nl_net;
1340
1341         RT6_TRACE("fib6_del_route\n");
1342
1343         /* Unlink it */
1344         *rtp = rt->dst.rt6_next;
1345         rt->rt6i_node = NULL;
1346         net->ipv6.rt6_stats->fib_rt_entries--;
1347         net->ipv6.rt6_stats->fib_discarded_routes++;
1348
1349         /* Reset round-robin state, if necessary */
1350         if (fn->rr_ptr == rt)
1351                 fn->rr_ptr = NULL;
1352
1353         /* Remove this entry from other siblings */
1354         if (rt->rt6i_nsiblings) {
1355                 struct rt6_info *sibling, *next_sibling;
1356
1357                 list_for_each_entry_safe(sibling, next_sibling,
1358                                          &rt->rt6i_siblings, rt6i_siblings)
1359                         sibling->rt6i_nsiblings--;
1360                 rt->rt6i_nsiblings = 0;
1361                 list_del_init(&rt->rt6i_siblings);
1362         }
1363
1364         /* Adjust walkers */
1365         read_lock(&fib6_walker_lock);
1366         FOR_WALKERS(w) {
1367                 if (w->state == FWS_C && w->leaf == rt) {
1368                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1369                         w->leaf = rt->dst.rt6_next;
1370                         if (!w->leaf)
1371                                 w->state = FWS_U;
1372                 }
1373         }
1374         read_unlock(&fib6_walker_lock);
1375
1376         rt->dst.rt6_next = NULL;
1377
1378         /* If it was last route, expunge its radix tree node */
1379         if (!fn->leaf) {
1380                 fn->fn_flags &= ~RTN_RTINFO;
1381                 net->ipv6.rt6_stats->fib_route_nodes--;
1382                 fn = fib6_repair_tree(net, fn);
1383         }
1384
1385         fib6_purge_rt(rt, fn, net);
1386
1387         inet6_rt_notify(RTM_DELROUTE, rt, info);
1388         rt6_release(rt);
1389 }
1390
1391 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1392 {
1393         struct net *net = info->nl_net;
1394         struct fib6_node *fn = rt->rt6i_node;
1395         struct rt6_info **rtp;
1396
1397 #if RT6_DEBUG >= 2
1398         if (rt->dst.obsolete > 0) {
1399                 WARN_ON(fn);
1400                 return -ENOENT;
1401         }
1402 #endif
1403         if (!fn || rt == net->ipv6.ip6_null_entry)
1404                 return -ENOENT;
1405
1406         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1407
1408         if (!(rt->rt6i_flags & RTF_CACHE)) {
1409                 struct fib6_node *pn = fn;
1410 #ifdef CONFIG_IPV6_SUBTREES
1411                 /* clones of this route might be in another subtree */
1412                 if (rt->rt6i_src.plen) {
1413                         while (!(pn->fn_flags & RTN_ROOT))
1414                                 pn = pn->parent;
1415                         pn = pn->parent;
1416                 }
1417 #endif
1418                 fib6_prune_clones(info->nl_net, pn);
1419         }
1420
1421         /*
1422          *      Walk the leaf entries looking for ourself
1423          */
1424
1425         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1426                 if (*rtp == rt) {
1427                         fib6_del_route(fn, rtp, info);
1428                         return 0;
1429                 }
1430         }
1431         return -ENOENT;
1432 }
1433
1434 /*
1435  *      Tree traversal function.
1436  *
1437  *      Certainly, it is not interrupt safe.
1438  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1439  *      It means, that we can modify tree during walking
1440  *      and use this function for garbage collection, clone pruning,
1441  *      cleaning tree when a device goes down etc. etc.
1442  *
1443  *      It guarantees that every node will be traversed,
1444  *      and that it will be traversed only once.
1445  *
1446  *      Callback function w->func may return:
1447  *      0 -> continue walking.
1448  *      positive value -> walking is suspended (used by tree dumps,
1449  *      and probably by gc, if it will be split to several slices)
1450  *      negative value -> terminate walking.
1451  *
1452  *      The function itself returns:
1453  *      0   -> walk is complete.
1454  *      >0  -> walk is incomplete (i.e. suspended)
1455  *      <0  -> walk is terminated by an error.
1456  */
1457
1458 static int fib6_walk_continue(struct fib6_walker *w)
1459 {
1460         struct fib6_node *fn, *pn;
1461
1462         for (;;) {
1463                 fn = w->node;
1464                 if (!fn)
1465                         return 0;
1466
1467                 if (w->prune && fn != w->root &&
1468                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1469                         w->state = FWS_C;
1470                         w->leaf = fn->leaf;
1471                 }
1472                 switch (w->state) {
1473 #ifdef CONFIG_IPV6_SUBTREES
1474                 case FWS_S:
1475                         if (FIB6_SUBTREE(fn)) {
1476                                 w->node = FIB6_SUBTREE(fn);
1477                                 continue;
1478                         }
1479                         w->state = FWS_L;
1480 #endif
1481                 case FWS_L:
1482                         if (fn->left) {
1483                                 w->node = fn->left;
1484                                 w->state = FWS_INIT;
1485                                 continue;
1486                         }
1487                         w->state = FWS_R;
1488                 case FWS_R:
1489                         if (fn->right) {
1490                                 w->node = fn->right;
1491                                 w->state = FWS_INIT;
1492                                 continue;
1493                         }
1494                         w->state = FWS_C;
1495                         w->leaf = fn->leaf;
1496                 case FWS_C:
1497                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1498                                 int err;
1499
1500                                 if (w->skip) {
1501                                         w->skip--;
1502                                         goto skip;
1503                                 }
1504
1505                                 err = w->func(w);
1506                                 if (err)
1507                                         return err;
1508
1509                                 w->count++;
1510                                 continue;
1511                         }
1512 skip:
1513                         w->state = FWS_U;
1514                 case FWS_U:
1515                         if (fn == w->root)
1516                                 return 0;
1517                         pn = fn->parent;
1518                         w->node = pn;
1519 #ifdef CONFIG_IPV6_SUBTREES
1520                         if (FIB6_SUBTREE(pn) == fn) {
1521                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1522                                 w->state = FWS_L;
1523                                 continue;
1524                         }
1525 #endif
1526                         if (pn->left == fn) {
1527                                 w->state = FWS_R;
1528                                 continue;
1529                         }
1530                         if (pn->right == fn) {
1531                                 w->state = FWS_C;
1532                                 w->leaf = w->node->leaf;
1533                                 continue;
1534                         }
1535 #if RT6_DEBUG >= 2
1536                         WARN_ON(1);
1537 #endif
1538                 }
1539         }
1540 }
1541
1542 static int fib6_walk(struct fib6_walker *w)
1543 {
1544         int res;
1545
1546         w->state = FWS_INIT;
1547         w->node = w->root;
1548
1549         fib6_walker_link(w);
1550         res = fib6_walk_continue(w);
1551         if (res <= 0)
1552                 fib6_walker_unlink(w);
1553         return res;
1554 }
1555
1556 static int fib6_clean_node(struct fib6_walker *w)
1557 {
1558         int res;
1559         struct rt6_info *rt;
1560         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1561         struct nl_info info = {
1562                 .nl_net = c->net,
1563         };
1564
1565         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1566             w->node->fn_sernum != c->sernum)
1567                 w->node->fn_sernum = c->sernum;
1568
1569         if (!c->func) {
1570                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1571                 w->leaf = NULL;
1572                 return 0;
1573         }
1574
1575         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1576                 res = c->func(rt, c->arg);
1577                 if (res < 0) {
1578                         w->leaf = rt;
1579                         res = fib6_del(rt, &info);
1580                         if (res) {
1581 #if RT6_DEBUG >= 2
1582                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1583                                          __func__, rt, rt->rt6i_node, res);
1584 #endif
1585                                 continue;
1586                         }
1587                         return 0;
1588                 }
1589                 WARN_ON(res != 0);
1590         }
1591         w->leaf = rt;
1592         return 0;
1593 }
1594
1595 /*
1596  *      Convenient frontend to tree walker.
1597  *
1598  *      func is called on each route.
1599  *              It may return -1 -> delete this route.
1600  *                            0  -> continue walking
1601  *
1602  *      prune==1 -> only immediate children of node (certainly,
1603  *      ignoring pure split nodes) will be scanned.
1604  */
1605
1606 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1607                             int (*func)(struct rt6_info *, void *arg),
1608                             bool prune, int sernum, void *arg)
1609 {
1610         struct fib6_cleaner c;
1611
1612         c.w.root = root;
1613         c.w.func = fib6_clean_node;
1614         c.w.prune = prune;
1615         c.w.count = 0;
1616         c.w.skip = 0;
1617         c.func = func;
1618         c.sernum = sernum;
1619         c.arg = arg;
1620         c.net = net;
1621
1622         fib6_walk(&c.w);
1623 }
1624
1625 static void __fib6_clean_all(struct net *net,
1626                              int (*func)(struct rt6_info *, void *),
1627                              int sernum, void *arg)
1628 {
1629         struct fib6_table *table;
1630         struct hlist_head *head;
1631         unsigned int h;
1632
1633         rcu_read_lock();
1634         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1635                 head = &net->ipv6.fib_table_hash[h];
1636                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1637                         write_lock_bh(&table->tb6_lock);
1638                         fib6_clean_tree(net, &table->tb6_root,
1639                                         func, false, sernum, arg);
1640                         write_unlock_bh(&table->tb6_lock);
1641                 }
1642         }
1643         rcu_read_unlock();
1644 }
1645
1646 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1647                     void *arg)
1648 {
1649         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1650 }
1651
1652 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1653 {
1654         if (rt->rt6i_flags & RTF_CACHE) {
1655                 RT6_TRACE("pruning clone %p\n", rt);
1656                 return -1;
1657         }
1658
1659         return 0;
1660 }
1661
1662 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1663 {
1664         fib6_clean_tree(net, fn, fib6_prune_clone, true,
1665                         FIB6_NO_SERNUM_CHANGE, NULL);
1666 }
1667
1668 static void fib6_flush_trees(struct net *net)
1669 {
1670         int new_sernum = fib6_new_sernum(net);
1671
1672         __fib6_clean_all(net, NULL, new_sernum, NULL);
1673 }
1674
1675 /*
1676  *      Garbage collection
1677  */
1678
1679 static struct fib6_gc_args
1680 {
1681         int                     timeout;
1682         int                     more;
1683 } gc_args;
1684
1685 static int fib6_age(struct rt6_info *rt, void *arg)
1686 {
1687         unsigned long now = jiffies;
1688
1689         /*
1690          *      check addrconf expiration here.
1691          *      Routes are expired even if they are in use.
1692          *
1693          *      Also age clones. Note, that clones are aged out
1694          *      only if they are not in use now.
1695          */
1696
1697         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1698                 if (time_after(now, rt->dst.expires)) {
1699                         RT6_TRACE("expiring %p\n", rt);
1700                         return -1;
1701                 }
1702                 gc_args.more++;
1703         } else if (rt->rt6i_flags & RTF_CACHE) {
1704                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1705                     time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1706                         RT6_TRACE("aging clone %p\n", rt);
1707                         return -1;
1708                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1709                         struct neighbour *neigh;
1710                         __u8 neigh_flags = 0;
1711
1712                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1713                         if (neigh) {
1714                                 neigh_flags = neigh->flags;
1715                                 neigh_release(neigh);
1716                         }
1717                         if (!(neigh_flags & NTF_ROUTER)) {
1718                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1719                                           rt);
1720                                 return -1;
1721                         }
1722                 }
1723                 gc_args.more++;
1724         }
1725
1726         return 0;
1727 }
1728
1729 static DEFINE_SPINLOCK(fib6_gc_lock);
1730
1731 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1732 {
1733         unsigned long now;
1734
1735         if (force) {
1736                 spin_lock_bh(&fib6_gc_lock);
1737         } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1738                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1739                 return;
1740         }
1741         gc_args.timeout = expires ? (int)expires :
1742                           net->ipv6.sysctl.ip6_rt_gc_interval;
1743
1744         gc_args.more = icmp6_dst_gc();
1745
1746         fib6_clean_all(net, fib6_age, NULL);
1747         now = jiffies;
1748         net->ipv6.ip6_rt_last_gc = now;
1749
1750         if (gc_args.more)
1751                 mod_timer(&net->ipv6.ip6_fib_timer,
1752                           round_jiffies(now
1753                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1754         else
1755                 del_timer(&net->ipv6.ip6_fib_timer);
1756         spin_unlock_bh(&fib6_gc_lock);
1757 }
1758
1759 static void fib6_gc_timer_cb(unsigned long arg)
1760 {
1761         fib6_run_gc(0, (struct net *)arg, true);
1762 }
1763
1764 static int __net_init fib6_net_init(struct net *net)
1765 {
1766         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1767
1768         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1769
1770         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1771         if (!net->ipv6.rt6_stats)
1772                 goto out_timer;
1773
1774         /* Avoid false sharing : Use at least a full cache line */
1775         size = max_t(size_t, size, L1_CACHE_BYTES);
1776
1777         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1778         if (!net->ipv6.fib_table_hash)
1779                 goto out_rt6_stats;
1780
1781         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1782                                           GFP_KERNEL);
1783         if (!net->ipv6.fib6_main_tbl)
1784                 goto out_fib_table_hash;
1785
1786         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1787         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1788         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1789                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1790         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1791
1792 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1793         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1794                                            GFP_KERNEL);
1795         if (!net->ipv6.fib6_local_tbl)
1796                 goto out_fib6_main_tbl;
1797         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1798         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1799         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1800                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1801         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1802 #endif
1803         fib6_tables_init(net);
1804
1805         return 0;
1806
1807 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1808 out_fib6_main_tbl:
1809         kfree(net->ipv6.fib6_main_tbl);
1810 #endif
1811 out_fib_table_hash:
1812         kfree(net->ipv6.fib_table_hash);
1813 out_rt6_stats:
1814         kfree(net->ipv6.rt6_stats);
1815 out_timer:
1816         return -ENOMEM;
1817 }
1818
1819 static void fib6_net_exit(struct net *net)
1820 {
1821         rt6_ifdown(net, NULL);
1822         del_timer_sync(&net->ipv6.ip6_fib_timer);
1823
1824 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1825         inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1826         kfree(net->ipv6.fib6_local_tbl);
1827 #endif
1828         inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1829         kfree(net->ipv6.fib6_main_tbl);
1830         kfree(net->ipv6.fib_table_hash);
1831         kfree(net->ipv6.rt6_stats);
1832 }
1833
1834 static struct pernet_operations fib6_net_ops = {
1835         .init = fib6_net_init,
1836         .exit = fib6_net_exit,
1837 };
1838
1839 int __init fib6_init(void)
1840 {
1841         int ret = -ENOMEM;
1842
1843         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1844                                            sizeof(struct fib6_node),
1845                                            0, SLAB_HWCACHE_ALIGN,
1846                                            NULL);
1847         if (!fib6_node_kmem)
1848                 goto out;
1849
1850         ret = register_pernet_subsys(&fib6_net_ops);
1851         if (ret)
1852                 goto out_kmem_cache_create;
1853
1854         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1855                               NULL);
1856         if (ret)
1857                 goto out_unregister_subsys;
1858
1859         __fib6_flush_trees = fib6_flush_trees;
1860 out:
1861         return ret;
1862
1863 out_unregister_subsys:
1864         unregister_pernet_subsys(&fib6_net_ops);
1865 out_kmem_cache_create:
1866         kmem_cache_destroy(fib6_node_kmem);
1867         goto out;
1868 }
1869
1870 void fib6_gc_cleanup(void)
1871 {
1872         unregister_pernet_subsys(&fib6_net_ops);
1873         kmem_cache_destroy(fib6_node_kmem);
1874 }
1875
1876 #ifdef CONFIG_PROC_FS
1877
1878 struct ipv6_route_iter {
1879         struct seq_net_private p;
1880         struct fib6_walker w;
1881         loff_t skip;
1882         struct fib6_table *tbl;
1883         int sernum;
1884 };
1885
1886 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1887 {
1888         struct rt6_info *rt = v;
1889         struct ipv6_route_iter *iter = seq->private;
1890
1891         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1892
1893 #ifdef CONFIG_IPV6_SUBTREES
1894         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1895 #else
1896         seq_puts(seq, "00000000000000000000000000000000 00 ");
1897 #endif
1898         if (rt->rt6i_flags & RTF_GATEWAY)
1899                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1900         else
1901                 seq_puts(seq, "00000000000000000000000000000000");
1902
1903         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1904                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1905                    rt->dst.__use, rt->rt6i_flags,
1906                    rt->dst.dev ? rt->dst.dev->name : "");
1907         iter->w.leaf = NULL;
1908         return 0;
1909 }
1910
1911 static int ipv6_route_yield(struct fib6_walker *w)
1912 {
1913         struct ipv6_route_iter *iter = w->args;
1914
1915         if (!iter->skip)
1916                 return 1;
1917
1918         do {
1919                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1920                 iter->skip--;
1921                 if (!iter->skip && iter->w.leaf)
1922                         return 1;
1923         } while (iter->w.leaf);
1924
1925         return 0;
1926 }
1927
1928 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1929 {
1930         memset(&iter->w, 0, sizeof(iter->w));
1931         iter->w.func = ipv6_route_yield;
1932         iter->w.root = &iter->tbl->tb6_root;
1933         iter->w.state = FWS_INIT;
1934         iter->w.node = iter->w.root;
1935         iter->w.args = iter;
1936         iter->sernum = iter->w.root->fn_sernum;
1937         INIT_LIST_HEAD(&iter->w.lh);
1938         fib6_walker_link(&iter->w);
1939 }
1940
1941 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1942                                                     struct net *net)
1943 {
1944         unsigned int h;
1945         struct hlist_node *node;
1946
1947         if (tbl) {
1948                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1949                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1950         } else {
1951                 h = 0;
1952                 node = NULL;
1953         }
1954
1955         while (!node && h < FIB6_TABLE_HASHSZ) {
1956                 node = rcu_dereference_bh(
1957                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1958         }
1959         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1960 }
1961
1962 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1963 {
1964         if (iter->sernum != iter->w.root->fn_sernum) {
1965                 iter->sernum = iter->w.root->fn_sernum;
1966                 iter->w.state = FWS_INIT;
1967                 iter->w.node = iter->w.root;
1968                 WARN_ON(iter->w.skip);
1969                 iter->w.skip = iter->w.count;
1970         }
1971 }
1972
1973 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1974 {
1975         int r;
1976         struct rt6_info *n;
1977         struct net *net = seq_file_net(seq);
1978         struct ipv6_route_iter *iter = seq->private;
1979
1980         if (!v)
1981                 goto iter_table;
1982
1983         n = ((struct rt6_info *)v)->dst.rt6_next;
1984         if (n) {
1985                 ++*pos;
1986                 return n;
1987         }
1988
1989 iter_table:
1990         ipv6_route_check_sernum(iter);
1991         read_lock(&iter->tbl->tb6_lock);
1992         r = fib6_walk_continue(&iter->w);
1993         read_unlock(&iter->tbl->tb6_lock);
1994         if (r > 0) {
1995                 if (v)
1996                         ++*pos;
1997                 return iter->w.leaf;
1998         } else if (r < 0) {
1999                 fib6_walker_unlink(&iter->w);
2000                 return NULL;
2001         }
2002         fib6_walker_unlink(&iter->w);
2003
2004         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2005         if (!iter->tbl)
2006                 return NULL;
2007
2008         ipv6_route_seq_setup_walk(iter);
2009         goto iter_table;
2010 }
2011
2012 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2013         __acquires(RCU_BH)
2014 {
2015         struct net *net = seq_file_net(seq);
2016         struct ipv6_route_iter *iter = seq->private;
2017
2018         rcu_read_lock_bh();
2019         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2020         iter->skip = *pos;
2021
2022         if (iter->tbl) {
2023                 ipv6_route_seq_setup_walk(iter);
2024                 return ipv6_route_seq_next(seq, NULL, pos);
2025         } else {
2026                 return NULL;
2027         }
2028 }
2029
2030 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2031 {
2032         struct fib6_walker *w = &iter->w;
2033         return w->node && !(w->state == FWS_U && w->node == w->root);
2034 }
2035
2036 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2037         __releases(RCU_BH)
2038 {
2039         struct ipv6_route_iter *iter = seq->private;
2040
2041         if (ipv6_route_iter_active(iter))
2042                 fib6_walker_unlink(&iter->w);
2043
2044         rcu_read_unlock_bh();
2045 }
2046
2047 static const struct seq_operations ipv6_route_seq_ops = {
2048         .start  = ipv6_route_seq_start,
2049         .next   = ipv6_route_seq_next,
2050         .stop   = ipv6_route_seq_stop,
2051         .show   = ipv6_route_seq_show
2052 };
2053
2054 int ipv6_route_open(struct inode *inode, struct file *file)
2055 {
2056         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2057                             sizeof(struct ipv6_route_iter));
2058 }
2059
2060 #endif /* CONFIG_PROC_FS */