linux/net/ipv4/fib_trie.c
<<
>>
Prefs
   1/*
   2 *   This program is free software; you can redistribute it and/or
   3 *   modify it under the terms of the GNU General Public License
   4 *   as published by the Free Software Foundation; either version
   5 *   2 of the License, or (at your option) any later version.
   6 *
   7 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
   8 *     & Swedish University of Agricultural Sciences.
   9 *
  10 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
  11 *     Agricultural Sciences.
  12 *
  13 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
  14 *
  15 * This work is based on the LPC-trie which is originally described in:
  16 *
  17 * An experimental study of compression methods for dynamic tries
  18 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
  19 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
  20 *
  21 *
  22 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
  23 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
  24 *
  25 *
  26 * Code from fib_hash has been reused which includes the following header:
  27 *
  28 *
  29 * INET         An implementation of the TCP/IP protocol suite for the LINUX
  30 *              operating system.  INET is implemented using the  BSD Socket
  31 *              interface as the means of communication with the user level.
  32 *
  33 *              IPv4 FIB: lookup engine and maintenance routines.
  34 *
  35 *
  36 * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  37 *
  38 *              This program is free software; you can redistribute it and/or
  39 *              modify it under the terms of the GNU General Public License
  40 *              as published by the Free Software Foundation; either version
  41 *              2 of the License, or (at your option) any later version.
  42 *
  43 * Substantial contributions to this work comes from:
  44 *
  45 *              David S. Miller, <davem@davemloft.net>
  46 *              Stephen Hemminger <shemminger@osdl.org>
  47 *              Paul E. McKenney <paulmck@us.ibm.com>
  48 *              Patrick McHardy <kaber@trash.net>
  49 */
  50
  51#define VERSION "0.409"
  52
  53#include <asm/uaccess.h>
  54#include <linux/bitops.h>
  55#include <linux/types.h>
  56#include <linux/kernel.h>
  57#include <linux/mm.h>
  58#include <linux/string.h>
  59#include <linux/socket.h>
  60#include <linux/sockios.h>
  61#include <linux/errno.h>
  62#include <linux/in.h>
  63#include <linux/inet.h>
  64#include <linux/inetdevice.h>
  65#include <linux/netdevice.h>
  66#include <linux/if_arp.h>
  67#include <linux/proc_fs.h>
  68#include <linux/rcupdate.h>
  69#include <linux/skbuff.h>
  70#include <linux/netlink.h>
  71#include <linux/init.h>
  72#include <linux/list.h>
  73#include <linux/slab.h>
  74#include <linux/export.h>
  75#include <linux/vmalloc.h>
  76#include <linux/notifier.h>
  77#include <net/net_namespace.h>
  78#include <net/ip.h>
  79#include <net/protocol.h>
  80#include <net/route.h>
  81#include <net/tcp.h>
  82#include <net/sock.h>
  83#include <net/ip_fib.h>
  84#include <trace/events/fib.h>
  85#include "fib_lookup.h"
  86
  87static BLOCKING_NOTIFIER_HEAD(fib_chain);
  88
  89int register_fib_notifier(struct notifier_block *nb)
  90{
  91        return blocking_notifier_chain_register(&fib_chain, nb);
  92}
  93EXPORT_SYMBOL(register_fib_notifier);
  94
  95int unregister_fib_notifier(struct notifier_block *nb)
  96{
  97        return blocking_notifier_chain_unregister(&fib_chain, nb);
  98}
  99EXPORT_SYMBOL(unregister_fib_notifier);
 100
 101int call_fib_notifiers(struct net *net, enum fib_event_type event_type,
 102                       struct fib_notifier_info *info)
 103{
 104        info->net = net;
 105        return blocking_notifier_call_chain(&fib_chain, event_type, info);
 106}
 107
 108static int call_fib_entry_notifiers(struct net *net,
 109                                    enum fib_event_type event_type, u32 dst,
 110                                    int dst_len, struct fib_info *fi,
 111                                    u8 tos, u8 type, u32 tb_id, u32 nlflags)
 112{
 113        struct fib_entry_notifier_info info = {
 114                .dst = dst,
 115                .dst_len = dst_len,
 116                .fi = fi,
 117                .tos = tos,
 118                .type = type,
 119                .tb_id = tb_id,
 120                .nlflags = nlflags,
 121        };
 122        return call_fib_notifiers(net, event_type, &info.info);
 123}
 124
 125#define MAX_STAT_DEPTH 32
 126
 127#define KEYLENGTH       (8*sizeof(t_key))
 128#define KEY_MAX         ((t_key)~0)
 129
 130typedef unsigned int t_key;
 131
 132#define IS_TRIE(n)      ((n)->pos >= KEYLENGTH)
 133#define IS_TNODE(n)     ((n)->bits)
 134#define IS_LEAF(n)      (!(n)->bits)
 135
 136struct key_vector {
 137        t_key key;
 138        unsigned char pos;              /* 2log(KEYLENGTH) bits needed */
 139        unsigned char bits;             /* 2log(KEYLENGTH) bits needed */
 140        unsigned char slen;
 141        union {
 142                /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 143                struct hlist_head leaf;
 144                /* This array is valid if (pos | bits) > 0 (TNODE) */
 145                struct key_vector __rcu *tnode[0];
 146        };
 147};
 148
 149struct tnode {
 150        struct rcu_head rcu;
 151        t_key empty_children;           /* KEYLENGTH bits needed */
 152        t_key full_children;            /* KEYLENGTH bits needed */
 153        struct key_vector __rcu *parent;
 154        struct key_vector kv[1];
 155#define tn_bits kv[0].bits
 156};
 157
 158#define TNODE_SIZE(n)   offsetof(struct tnode, kv[0].tnode[n])
 159#define LEAF_SIZE       TNODE_SIZE(1)
 160
 161#ifdef CONFIG_IP_FIB_TRIE_STATS
 162struct trie_use_stats {
 163        unsigned int gets;
 164        unsigned int backtrack;
 165        unsigned int semantic_match_passed;
 166        unsigned int semantic_match_miss;
 167        unsigned int null_node_hit;
 168        unsigned int resize_node_skipped;
 169};
 170#endif
 171
 172struct trie_stat {
 173        unsigned int totdepth;
 174        unsigned int maxdepth;
 175        unsigned int tnodes;
 176        unsigned int leaves;
 177        unsigned int nullpointers;
 178        unsigned int prefixes;
 179        unsigned int nodesizes[MAX_STAT_DEPTH];
 180};
 181
 182struct trie {
 183        struct key_vector kv[1];
 184#ifdef CONFIG_IP_FIB_TRIE_STATS
 185        struct trie_use_stats __percpu *stats;
 186#endif
 187};
 188
 189static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 190static size_t tnode_free_size;
 191
 192/*
 193 * synchronize_rcu after call_rcu for that many pages; it should be especially
 194 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 195 * obtained experimentally, aiming to avoid visible slowdown.
 196 */
 197static const int sync_pages = 128;
 198
 199static struct kmem_cache *fn_alias_kmem __read_mostly;
 200static struct kmem_cache *trie_leaf_kmem __read_mostly;
 201
 202static inline struct tnode *tn_info(struct key_vector *kv)
 203{
 204        return container_of(kv, struct tnode, kv[0]);
 205}
 206
 207/* caller must hold RTNL */
 208#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 209#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 210
 211/* caller must hold RCU read lock or RTNL */
 212#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 213#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 214
 215/* wrapper for rcu_assign_pointer */
 216static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 217{
 218        if (n)
 219                rcu_assign_pointer(tn_info(n)->parent, tp);
 220}
 221
 222#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 223
 224/* This provides us with the number of children in this node, in the case of a
 225 * leaf this will return 0 meaning none of the children are accessible.
 226 */
 227static inline unsigned long child_length(const struct key_vector *tn)
 228{
 229        return (1ul << tn->bits) & ~(1ul);
 230}
 231
 232#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 233
 234static inline unsigned long get_index(t_key key, struct key_vector *kv)
 235{
 236        unsigned long index = key ^ kv->key;
 237
 238        if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 239                return 0;
 240
 241        return index >> kv->pos;
 242}
 243
 244/* To understand this stuff, an understanding of keys and all their bits is
 245 * necessary. Every node in the trie has a key associated with it, but not
 246 * all of the bits in that key are significant.
 247 *
 248 * Consider a node 'n' and its parent 'tp'.
 249 *
 250 * If n is a leaf, every bit in its key is significant. Its presence is
 251 * necessitated by path compression, since during a tree traversal (when
 252 * searching for a leaf - unless we are doing an insertion) we will completely
 253 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 254 * a potentially successful search, that we have indeed been walking the
 255 * correct key path.
 256 *
 257 * Note that we can never "miss" the correct key in the tree if present by
 258 * following the wrong path. Path compression ensures that segments of the key
 259 * that are the same for all keys with a given prefix are skipped, but the
 260 * skipped part *is* identical for each node in the subtrie below the skipped
 261 * bit! trie_insert() in this implementation takes care of that.
 262 *
 263 * if n is an internal node - a 'tnode' here, the various parts of its key
 264 * have many different meanings.
 265 *
 266 * Example:
 267 * _________________________________________________________________
 268 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 269 * -----------------------------------------------------------------
 270 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 271 *
 272 * _________________________________________________________________
 273 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 274 * -----------------------------------------------------------------
 275 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 276 *
 277 * tp->pos = 22
 278 * tp->bits = 3
 279 * n->pos = 13
 280 * n->bits = 4
 281 *
 282 * First, let's just ignore the bits that come before the parent tp, that is
 283 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 284 * point we do not use them for anything.
 285 *
 286 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 287 * index into the parent's child array. That is, they will be used to find
 288 * 'n' among tp's children.
 289 *
 290 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
 291 * for the node n.
 292 *
 293 * All the bits we have seen so far are significant to the node n. The rest
 294 * of the bits are really not needed or indeed known in n->key.
 295 *
 296 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 297 * n's child array, and will of course be different for each child.
 298 *
 299 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
 300 * at this point.
 301 */
 302
 303static const int halve_threshold = 25;
 304static const int inflate_threshold = 50;
 305static const int halve_threshold_root = 15;
 306static const int inflate_threshold_root = 30;
 307
 308static void __alias_free_mem(struct rcu_head *head)
 309{
 310        struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 311        kmem_cache_free(fn_alias_kmem, fa);
 312}
 313
 314static inline void alias_free_mem_rcu(struct fib_alias *fa)
 315{
 316        call_rcu(&fa->rcu, __alias_free_mem);
 317}
 318
 319#define TNODE_KMALLOC_MAX \
 320        ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 321#define TNODE_VMALLOC_MAX \
 322        ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 323
 324static void __node_free_rcu(struct rcu_head *head)
 325{
 326        struct tnode *n = container_of(head, struct tnode, rcu);
 327
 328        if (!n->tn_bits)
 329                kmem_cache_free(trie_leaf_kmem, n);
 330        else
 331                kvfree(n);
 332}
 333
 334#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 335
 336static struct tnode *tnode_alloc(int bits)
 337{
 338        size_t size;
 339
 340        /* verify bits is within bounds */
 341        if (bits > TNODE_VMALLOC_MAX)
 342                return NULL;
 343
 344        /* determine size and verify it is non-zero and didn't overflow */
 345        size = TNODE_SIZE(1ul << bits);
 346
 347        if (size <= PAGE_SIZE)
 348                return kzalloc(size, GFP_KERNEL);
 349        else
 350                return vzalloc(size);
 351}
 352
 353static inline void empty_child_inc(struct key_vector *n)
 354{
 355        ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
 356}
 357
 358static inline void empty_child_dec(struct key_vector *n)
 359{
 360        tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
 361}
 362
 363static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 364{
 365        struct key_vector *l;
 366        struct tnode *kv;
 367
 368        kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 369        if (!kv)
 370                return NULL;
 371
 372        /* initialize key vector */
 373        l = kv->kv;
 374        l->key = key;
 375        l->pos = 0;
 376        l->bits = 0;
 377        l->slen = fa->fa_slen;
 378
 379        /* link leaf to fib alias */
 380        INIT_HLIST_HEAD(&l->leaf);
 381        hlist_add_head(&fa->fa_list, &l->leaf);
 382
 383        return l;
 384}
 385
 386static struct key_vector *tnode_new(t_key key, int pos, int bits)
 387{
 388        unsigned int shift = pos + bits;
 389        struct key_vector *tn;
 390        struct tnode *tnode;
 391
 392        /* verify bits and pos their msb bits clear and values are valid */
 393        BUG_ON(!bits || (shift > KEYLENGTH));
 394
 395        tnode = tnode_alloc(bits);
 396        if (!tnode)
 397                return NULL;
 398
 399        pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 400                 sizeof(struct key_vector *) << bits);
 401
 402        if (bits == KEYLENGTH)
 403                tnode->full_children = 1;
 404        else
 405                tnode->empty_children = 1ul << bits;
 406
 407        tn = tnode->kv;
 408        tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 409        tn->pos = pos;
 410        tn->bits = bits;
 411        tn->slen = pos;
 412
 413        return tn;
 414}
 415
 416/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 417 * and no bits are skipped. See discussion in dyntree paper p. 6
 418 */
 419static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 420{
 421        return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 422}
 423
 424/* Add a child at position i overwriting the old value.
 425 * Update the value of full_children and empty_children.
 426 */
 427static void put_child(struct key_vector *tn, unsigned long i,
 428                      struct key_vector *n)
 429{
 430        struct key_vector *chi = get_child(tn, i);
 431        int isfull, wasfull;
 432
 433        BUG_ON(i >= child_length(tn));
 434
 435        /* update emptyChildren, overflow into fullChildren */
 436        if (!n && chi)
 437                empty_child_inc(tn);
 438        if (n && !chi)
 439                empty_child_dec(tn);
 440
 441        /* update fullChildren */
 442        wasfull = tnode_full(tn, chi);
 443        isfull = tnode_full(tn, n);
 444
 445        if (wasfull && !isfull)
 446                tn_info(tn)->full_children--;
 447        else if (!wasfull && isfull)
 448                tn_info(tn)->full_children++;
 449
 450        if (n && (tn->slen < n->slen))
 451                tn->slen = n->slen;
 452
 453        rcu_assign_pointer(tn->tnode[i], n);
 454}
 455
 456static void update_children(struct key_vector *tn)
 457{
 458        unsigned long i;
 459
 460        /* update all of the child parent pointers */
 461        for (i = child_length(tn); i;) {
 462                struct key_vector *inode = get_child(tn, --i);
 463
 464                if (!inode)
 465                        continue;
 466
 467                /* Either update the children of a tnode that
 468                 * already belongs to us or update the child
 469                 * to point to ourselves.
 470                 */
 471                if (node_parent(inode) == tn)
 472                        update_children(inode);
 473                else
 474                        node_set_parent(inode, tn);
 475        }
 476}
 477
 478static inline void put_child_root(struct key_vector *tp, t_key key,
 479                                  struct key_vector *n)
 480{
 481        if (IS_TRIE(tp))
 482                rcu_assign_pointer(tp->tnode[0], n);
 483        else
 484                put_child(tp, get_index(key, tp), n);
 485}
 486
 487static inline void tnode_free_init(struct key_vector *tn)
 488{
 489        tn_info(tn)->rcu.next = NULL;
 490}
 491
 492static inline void tnode_free_append(struct key_vector *tn,
 493                                     struct key_vector *n)
 494{
 495        tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 496        tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 497}
 498
 499static void tnode_free(struct key_vector *tn)
 500{
 501        struct callback_head *head = &tn_info(tn)->rcu;
 502
 503        while (head) {
 504                head = head->next;
 505                tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 506                node_free(tn);
 507
 508                tn = container_of(head, struct tnode, rcu)->kv;
 509        }
 510
 511        if (tnode_free_size >= PAGE_SIZE * sync_pages) {
 512                tnode_free_size = 0;
 513                synchronize_rcu();
 514        }
 515}
 516
 517static struct key_vector *replace(struct trie *t,
 518                                  struct key_vector *oldtnode,
 519                                  struct key_vector *tn)
 520{
 521        struct key_vector *tp = node_parent(oldtnode);
 522        unsigned long i;
 523
 524        /* setup the parent pointer out of and back into this node */
 525        NODE_INIT_PARENT(tn, tp);
 526        put_child_root(tp, tn->key, tn);
 527
 528        /* update all of the child parent pointers */
 529        update_children(tn);
 530
 531        /* all pointers should be clean so we are done */
 532        tnode_free(oldtnode);
 533
 534        /* resize children now that oldtnode is freed */
 535        for (i = child_length(tn); i;) {
 536                struct key_vector *inode = get_child(tn, --i);
 537
 538                /* resize child node */
 539                if (tnode_full(tn, inode))
 540                        tn = resize(t, inode);
 541        }
 542
 543        return tp;
 544}
 545
 546static struct key_vector *inflate(struct trie *t,
 547                                  struct key_vector *oldtnode)
 548{
 549        struct key_vector *tn;
 550        unsigned long i;
 551        t_key m;
 552
 553        pr_debug("In inflate\n");
 554
 555        tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 556        if (!tn)
 557                goto notnode;
 558
 559        /* prepare oldtnode to be freed */
 560        tnode_free_init(oldtnode);
 561
 562        /* Assemble all of the pointers in our cluster, in this case that
 563         * represents all of the pointers out of our allocated nodes that
 564         * point to existing tnodes and the links between our allocated
 565         * nodes.
 566         */
 567        for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 568                struct key_vector *inode = get_child(oldtnode, --i);
 569                struct key_vector *node0, *node1;
 570                unsigned long j, k;
 571
 572                /* An empty child */
 573                if (!inode)
 574                        continue;
 575
 576                /* A leaf or an internal node with skipped bits */
 577                if (!tnode_full(oldtnode, inode)) {
 578                        put_child(tn, get_index(inode->key, tn), inode);
 579                        continue;
 580                }
 581
 582                /* drop the node in the old tnode free list */
 583                tnode_free_append(oldtnode, inode);
 584
 585                /* An internal node with two children */
 586                if (inode->bits == 1) {
 587                        put_child(tn, 2 * i + 1, get_child(inode, 1));
 588                        put_child(tn, 2 * i, get_child(inode, 0));
 589                        continue;
 590                }
 591
 592                /* We will replace this node 'inode' with two new
 593                 * ones, 'node0' and 'node1', each with half of the
 594                 * original children. The two new nodes will have
 595                 * a position one bit further down the key and this
 596                 * means that the "significant" part of their keys
 597                 * (see the discussion near the top of this file)
 598                 * will differ by one bit, which will be "0" in
 599                 * node0's key and "1" in node1's key. Since we are
 600                 * moving the key position by one step, the bit that
 601                 * we are moving away from - the bit at position
 602                 * (tn->pos) - is the one that will differ between
 603                 * node0 and node1. So... we synthesize that bit in the
 604                 * two new keys.
 605                 */
 606                node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 607                if (!node1)
 608                        goto nomem;
 609                node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 610
 611                tnode_free_append(tn, node1);
 612                if (!node0)
 613                        goto nomem;
 614                tnode_free_append(tn, node0);
 615
 616                /* populate child pointers in new nodes */
 617                for (k = child_length(inode), j = k / 2; j;) {
 618                        put_child(node1, --j, get_child(inode, --k));
 619                        put_child(node0, j, get_child(inode, j));
 620                        put_child(node1, --j, get_child(inode, --k));
 621                        put_child(node0, j, get_child(inode, j));
 622                }
 623
 624                /* link new nodes to parent */
 625                NODE_INIT_PARENT(node1, tn);
 626                NODE_INIT_PARENT(node0, tn);
 627
 628                /* link parent to nodes */
 629                put_child(tn, 2 * i + 1, node1);
 630                put_child(tn, 2 * i, node0);
 631        }
 632
 633        /* setup the parent pointers into and out of this node */
 634        return replace(t, oldtnode, tn);
 635nomem:
 636        /* all pointers should be clean so we are done */
 637        tnode_free(tn);
 638notnode:
 639        return NULL;
 640}
 641
 642static struct key_vector *halve(struct trie *t,
 643                                struct key_vector *oldtnode)
 644{
 645        struct key_vector *tn;
 646        unsigned long i;
 647
 648        pr_debug("In halve\n");
 649
 650        tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 651        if (!tn)
 652                goto notnode;
 653
 654        /* prepare oldtnode to be freed */
 655        tnode_free_init(oldtnode);
 656
 657        /* Assemble all of the pointers in our cluster, in this case that
 658         * represents all of the pointers out of our allocated nodes that
 659         * point to existing tnodes and the links between our allocated
 660         * nodes.
 661         */
 662        for (i = child_length(oldtnode); i;) {
 663                struct key_vector *node1 = get_child(oldtnode, --i);
 664                struct key_vector *node0 = get_child(oldtnode, --i);
 665                struct key_vector *inode;
 666
 667                /* At least one of the children is empty */
 668                if (!node1 || !node0) {
 669                        put_child(tn, i / 2, node1 ? : node0);
 670                        continue;
 671                }
 672
 673                /* Two nonempty children */
 674                inode = tnode_new(node0->key, oldtnode->pos, 1);
 675                if (!inode)
 676                        goto nomem;
 677                tnode_free_append(tn, inode);
 678
 679                /* initialize pointers out of node */
 680                put_child(inode, 1, node1);
 681                put_child(inode, 0, node0);
 682                NODE_INIT_PARENT(inode, tn);
 683
 684                /* link parent to node */
 685                put_child(tn, i / 2, inode);
 686        }
 687
 688        /* setup the parent pointers into and out of this node */
 689        return replace(t, oldtnode, tn);
 690nomem:
 691        /* all pointers should be clean so we are done */
 692        tnode_free(tn);
 693notnode:
 694        return NULL;
 695}
 696
 697static struct key_vector *collapse(struct trie *t,
 698                                   struct key_vector *oldtnode)
 699{
 700        struct key_vector *n, *tp;
 701        unsigned long i;
 702
 703        /* scan the tnode looking for that one child that might still exist */
 704        for (n = NULL, i = child_length(oldtnode); !n && i;)
 705                n = get_child(oldtnode, --i);
 706
 707        /* compress one level */
 708        tp = node_parent(oldtnode);
 709        put_child_root(tp, oldtnode->key, n);
 710        node_set_parent(n, tp);
 711
 712        /* drop dead node */
 713        node_free(oldtnode);
 714
 715        return tp;
 716}
 717
 718static unsigned char update_suffix(struct key_vector *tn)
 719{
 720        unsigned char slen = tn->pos;
 721        unsigned long stride, i;
 722        unsigned char slen_max;
 723
 724        /* only vector 0 can have a suffix length greater than or equal to
 725         * tn->pos + tn->bits, the second highest node will have a suffix
 726         * length at most of tn->pos + tn->bits - 1
 727         */
 728        slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
 729
 730        /* search though the list of children looking for nodes that might
 731         * have a suffix greater than the one we currently have.  This is
 732         * why we start with a stride of 2 since a stride of 1 would
 733         * represent the nodes with suffix length equal to tn->pos
 734         */
 735        for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 736                struct key_vector *n = get_child(tn, i);
 737
 738                if (!n || (n->slen <= slen))
 739                        continue;
 740
 741                /* update stride and slen based on new value */
 742                stride <<= (n->slen - slen);
 743                slen = n->slen;
 744                i &= ~(stride - 1);
 745
 746                /* stop searching if we have hit the maximum possible value */
 747                if (slen >= slen_max)
 748                        break;
 749        }
 750
 751        tn->slen = slen;
 752
 753        return slen;
 754}
 755
 756/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 757 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 758 * Telecommunications, page 6:
 759 * "A node is doubled if the ratio of non-empty children to all
 760 * children in the *doubled* node is at least 'high'."
 761 *
 762 * 'high' in this instance is the variable 'inflate_threshold'. It
 763 * is expressed as a percentage, so we multiply it with
 764 * child_length() and instead of multiplying by 2 (since the
 765 * child array will be doubled by inflate()) and multiplying
 766 * the left-hand side by 100 (to handle the percentage thing) we
 767 * multiply the left-hand side by 50.
 768 *
 769 * The left-hand side may look a bit weird: child_length(tn)
 770 * - tn->empty_children is of course the number of non-null children
 771 * in the current node. tn->full_children is the number of "full"
 772 * children, that is non-null tnodes with a skip value of 0.
 773 * All of those will be doubled in the resulting inflated tnode, so
 774 * we just count them one extra time here.
 775 *
 776 * A clearer way to write this would be:
 777 *
 778 * to_be_doubled = tn->full_children;
 779 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 780 *     tn->full_children;
 781 *
 782 * new_child_length = child_length(tn) * 2;
 783 *
 784 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 785 *      new_child_length;
 786 * if (new_fill_factor >= inflate_threshold)
 787 *
 788 * ...and so on, tho it would mess up the while () loop.
 789 *
 790 * anyway,
 791 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 792 *      inflate_threshold
 793 *
 794 * avoid a division:
 795 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 796 *      inflate_threshold * new_child_length
 797 *
 798 * expand not_to_be_doubled and to_be_doubled, and shorten:
 799 * 100 * (child_length(tn) - tn->empty_children +
 800 *    tn->full_children) >= inflate_threshold * new_child_length
 801 *
 802 * expand new_child_length:
 803 * 100 * (child_length(tn) - tn->empty_children +
 804 *    tn->full_children) >=
 805 *      inflate_threshold * child_length(tn) * 2
 806 *
 807 * shorten again:
 808 * 50 * (tn->full_children + child_length(tn) -
 809 *    tn->empty_children) >= inflate_threshold *
 810 *    child_length(tn)
 811 *
 812 */
 813static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 814{
 815        unsigned long used = child_length(tn);
 816        unsigned long threshold = used;
 817
 818        /* Keep root node larger */
 819        threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 820        used -= tn_info(tn)->empty_children;
 821        used += tn_info(tn)->full_children;
 822
 823        /* if bits == KEYLENGTH then pos = 0, and will fail below */
 824
 825        return (used > 1) && tn->pos && ((50 * used) >= threshold);
 826}
 827
 828static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 829{
 830        unsigned long used = child_length(tn);
 831        unsigned long threshold = used;
 832
 833        /* Keep root node larger */
 834        threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 835        used -= tn_info(tn)->empty_children;
 836
 837        /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 838
 839        return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 840}
 841
 842static inline bool should_collapse(struct key_vector *tn)
 843{
 844        unsigned long used = child_length(tn);
 845
 846        used -= tn_info(tn)->empty_children;
 847
 848        /* account for bits == KEYLENGTH case */
 849        if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 850                used -= KEY_MAX;
 851
 852        /* One child or none, time to drop us from the trie */
 853        return used < 2;
 854}
 855
 856#define MAX_WORK 10
 857static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 858{
 859#ifdef CONFIG_IP_FIB_TRIE_STATS
 860        struct trie_use_stats __percpu *stats = t->stats;
 861#endif
 862        struct key_vector *tp = node_parent(tn);
 863        unsigned long cindex = get_index(tn->key, tp);
 864        int max_work = MAX_WORK;
 865
 866        pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 867                 tn, inflate_threshold, halve_threshold);
 868
 869        /* track the tnode via the pointer from the parent instead of
 870         * doing it ourselves.  This way we can let RCU fully do its
 871         * thing without us interfering
 872         */
 873        BUG_ON(tn != get_child(tp, cindex));
 874
 875        /* Double as long as the resulting node has a number of
 876         * nonempty nodes that are above the threshold.
 877         */
 878        while (should_inflate(tp, tn) && max_work) {
 879                tp = inflate(t, tn);
 880                if (!tp) {
 881#ifdef CONFIG_IP_FIB_TRIE_STATS
 882                        this_cpu_inc(stats->resize_node_skipped);
 883#endif
 884                        break;
 885                }
 886
 887                max_work--;
 888                tn = get_child(tp, cindex);
 889        }
 890
 891        /* update parent in case inflate failed */
 892        tp = node_parent(tn);
 893
 894        /* Return if at least one inflate is run */
 895        if (max_work != MAX_WORK)
 896                return tp;
 897
 898        /* Halve as long as the number of empty children in this
 899         * node is above threshold.
 900         */
 901        while (should_halve(tp, tn) && max_work) {
 902                tp = halve(t, tn);
 903                if (!tp) {
 904#ifdef CONFIG_IP_FIB_TRIE_STATS
 905                        this_cpu_inc(stats->resize_node_skipped);
 906#endif
 907                        break;
 908                }
 909
 910                max_work--;
 911                tn = get_child(tp, cindex);
 912        }
 913
 914        /* Only one child remains */
 915        if (should_collapse(tn))
 916                return collapse(t, tn);
 917
 918        /* update parent in case halve failed */
 919        return node_parent(tn);
 920}
 921
 922static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
 923{
 924        unsigned char node_slen = tn->slen;
 925
 926        while ((node_slen > tn->pos) && (node_slen > slen)) {
 927                slen = update_suffix(tn);
 928                if (node_slen == slen)
 929                        break;
 930
 931                tn = node_parent(tn);
 932                node_slen = tn->slen;
 933        }
 934}
 935
 936static void node_push_suffix(struct key_vector *tn, unsigned char slen)
 937{
 938        while (tn->slen < slen) {
 939                tn->slen = slen;
 940                tn = node_parent(tn);
 941        }
 942}
 943
 944/* rcu_read_lock needs to be hold by caller from readside */
 945static struct key_vector *fib_find_node(struct trie *t,
 946                                        struct key_vector **tp, u32 key)
 947{
 948        struct key_vector *pn, *n = t->kv;
 949        unsigned long index = 0;
 950
 951        do {
 952                pn = n;
 953                n = get_child_rcu(n, index);
 954
 955                if (!n)
 956                        break;
 957
 958                index = get_cindex(key, n);
 959
 960                /* This bit of code is a bit tricky but it combines multiple
 961                 * checks into a single check.  The prefix consists of the
 962                 * prefix plus zeros for the bits in the cindex. The index
 963                 * is the difference between the key and this value.  From
 964                 * this we can actually derive several pieces of data.
 965                 *   if (index >= (1ul << bits))
 966                 *     we have a mismatch in skip bits and failed
 967                 *   else
 968                 *     we know the value is cindex
 969                 *
 970                 * This check is safe even if bits == KEYLENGTH due to the
 971                 * fact that we can only allocate a node with 32 bits if a
 972                 * long is greater than 32 bits.
 973                 */
 974                if (index >= (1ul << n->bits)) {
 975                        n = NULL;
 976                        break;
 977                }
 978
 979                /* keep searching until we find a perfect match leaf or NULL */
 980        } while (IS_TNODE(n));
 981
 982        *tp = pn;
 983
 984        return n;
 985}
 986
 987/* Return the first fib alias matching TOS with
 988 * priority less than or equal to PRIO.
 989 */
 990static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
 991                                        u8 tos, u32 prio, u32 tb_id)
 992{
 993        struct fib_alias *fa;
 994
 995        if (!fah)
 996                return NULL;
 997
 998        hlist_for_each_entry(fa, fah, fa_list) {
 999                if (fa->fa_slen < slen)
1000                        continue;
1001                if (fa->fa_slen != slen)
1002                        break;
1003                if (fa->tb_id > tb_id)
1004                        continue;
1005                if (fa->tb_id != tb_id)
1006                        break;
1007                if (fa->fa_tos > tos)
1008                        continue;
1009                if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
1010                        return fa;
1011        }
1012
1013        return NULL;
1014}
1015
1016static void trie_rebalance(struct trie *t, struct key_vector *tn)
1017{
1018        while (!IS_TRIE(tn))
1019                tn = resize(t, tn);
1020}
1021
1022static int fib_insert_node(struct trie *t, struct key_vector *tp,
1023                           struct fib_alias *new, t_key key)
1024{
1025        struct key_vector *n, *l;
1026
1027        l = leaf_new(key, new);
1028        if (!l)
1029                goto noleaf;
1030
1031        /* retrieve child from parent node */
1032        n = get_child(tp, get_index(key, tp));
1033
1034        /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1035         *
1036         *  Add a new tnode here
1037         *  first tnode need some special handling
1038         *  leaves us in position for handling as case 3
1039         */
1040        if (n) {
1041                struct key_vector *tn;
1042
1043                tn = tnode_new(key, __fls(key ^ n->key), 1);
1044                if (!tn)
1045                        goto notnode;
1046
1047                /* initialize routes out of node */
1048                NODE_INIT_PARENT(tn, tp);
1049                put_child(tn, get_index(key, tn) ^ 1, n);
1050
1051                /* start adding routes into the node */
1052                put_child_root(tp, key, tn);
1053                node_set_parent(n, tn);
1054
1055                /* parent now has a NULL spot where the leaf can go */
1056                tp = tn;
1057        }
1058
1059        /* Case 3: n is NULL, and will just insert a new leaf */
1060        node_push_suffix(tp, new->fa_slen);
1061        NODE_INIT_PARENT(l, tp);
1062        put_child_root(tp, key, l);
1063        trie_rebalance(t, tp);
1064
1065        return 0;
1066notnode:
1067        node_free(l);
1068noleaf:
1069        return -ENOMEM;
1070}
1071
1072static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1073                            struct key_vector *l, struct fib_alias *new,
1074                            struct fib_alias *fa, t_key key)
1075{
1076        if (!l)
1077                return fib_insert_node(t, tp, new, key);
1078
1079        if (fa) {
1080                hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1081        } else {
1082                struct fib_alias *last;
1083
1084                hlist_for_each_entry(last, &l->leaf, fa_list) {
1085                        if (new->fa_slen < last->fa_slen)
1086                                break;
1087                        if ((new->fa_slen == last->fa_slen) &&
1088                            (new->tb_id > last->tb_id))
1089                                break;
1090                        fa = last;
1091                }
1092
1093                if (fa)
1094                        hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1095                else
1096                        hlist_add_head_rcu(&new->fa_list, &l->leaf);
1097        }
1098
1099        /* if we added to the tail node then we need to update slen */
1100        if (l->slen < new->fa_slen) {
1101                l->slen = new->fa_slen;
1102                node_push_suffix(tp, new->fa_slen);
1103        }
1104
1105        return 0;
1106}
1107
1108/* Caller must hold RTNL. */
1109int fib_table_insert(struct net *net, struct fib_table *tb,
1110                     struct fib_config *cfg)
1111{
1112        struct trie *t = (struct trie *)tb->tb_data;
1113        struct fib_alias *fa, *new_fa;
1114        struct key_vector *l, *tp;
1115        u16 nlflags = NLM_F_EXCL;
1116        struct fib_info *fi;
1117        u8 plen = cfg->fc_dst_len;
1118        u8 slen = KEYLENGTH - plen;
1119        u8 tos = cfg->fc_tos;
1120        u32 key;
1121        int err;
1122
1123        if (plen > KEYLENGTH)
1124                return -EINVAL;
1125
1126        key = ntohl(cfg->fc_dst);
1127
1128        pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1129
1130        if ((plen < KEYLENGTH) && (key << plen))
1131                return -EINVAL;
1132
1133        fi = fib_create_info(cfg);
1134        if (IS_ERR(fi)) {
1135                err = PTR_ERR(fi);
1136                goto err;
1137        }
1138
1139        l = fib_find_node(t, &tp, key);
1140        fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1141                                tb->tb_id) : NULL;
1142
1143        /* Now fa, if non-NULL, points to the first fib alias
1144         * with the same keys [prefix,tos,priority], if such key already
1145         * exists or to the node before which we will insert new one.
1146         *
1147         * If fa is NULL, we will need to allocate a new one and
1148         * insert to the tail of the section matching the suffix length
1149         * of the new alias.
1150         */
1151
1152        if (fa && fa->fa_tos == tos &&
1153            fa->fa_info->fib_priority == fi->fib_priority) {
1154                struct fib_alias *fa_first, *fa_match;
1155
1156                err = -EEXIST;
1157                if (cfg->fc_nlflags & NLM_F_EXCL)
1158                        goto out;
1159
1160                nlflags &= ~NLM_F_EXCL;
1161
1162                /* We have 2 goals:
1163                 * 1. Find exact match for type, scope, fib_info to avoid
1164                 * duplicate routes
1165                 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1166                 */
1167                fa_match = NULL;
1168                fa_first = fa;
1169                hlist_for_each_entry_from(fa, fa_list) {
1170                        if ((fa->fa_slen != slen) ||
1171                            (fa->tb_id != tb->tb_id) ||
1172                            (fa->fa_tos != tos))
1173                                break;
1174                        if (fa->fa_info->fib_priority != fi->fib_priority)
1175                                break;
1176                        if (fa->fa_type == cfg->fc_type &&
1177                            fa->fa_info == fi) {
1178                                fa_match = fa;
1179                                break;
1180                        }
1181                }
1182
1183                if (cfg->fc_nlflags & NLM_F_REPLACE) {
1184                        struct fib_info *fi_drop;
1185                        u8 state;
1186
1187                        nlflags |= NLM_F_REPLACE;
1188                        fa = fa_first;
1189                        if (fa_match) {
1190                                if (fa == fa_match)
1191                                        err = 0;
1192                                goto out;
1193                        }
1194                        err = -ENOBUFS;
1195                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1196                        if (!new_fa)
1197                                goto out;
1198
1199                        fi_drop = fa->fa_info;
1200                        new_fa->fa_tos = fa->fa_tos;
1201                        new_fa->fa_info = fi;
1202                        new_fa->fa_type = cfg->fc_type;
1203                        state = fa->fa_state;
1204                        new_fa->fa_state = state & ~FA_S_ACCESSED;
1205                        new_fa->fa_slen = fa->fa_slen;
1206                        new_fa->tb_id = tb->tb_id;
1207                        new_fa->fa_default = -1;
1208
1209                        hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1210
1211                        alias_free_mem_rcu(fa);
1212
1213                        fib_release_info(fi_drop);
1214                        if (state & FA_S_ACCESSED)
1215                                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1216
1217                        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD,
1218                                                 key, plen, fi,
1219                                                 new_fa->fa_tos, cfg->fc_type,
1220                                                 tb->tb_id, cfg->fc_nlflags);
1221                        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1222                                tb->tb_id, &cfg->fc_nlinfo, nlflags);
1223
1224                        goto succeeded;
1225                }
1226                /* Error if we find a perfect match which
1227                 * uses the same scope, type, and nexthop
1228                 * information.
1229                 */
1230                if (fa_match)
1231                        goto out;
1232
1233                if (cfg->fc_nlflags & NLM_F_APPEND)
1234                        nlflags |= NLM_F_APPEND;
1235                else
1236                        fa = fa_first;
1237        }
1238        err = -ENOENT;
1239        if (!(cfg->fc_nlflags & NLM_F_CREATE))
1240                goto out;
1241
1242        nlflags |= NLM_F_CREATE;
1243        err = -ENOBUFS;
1244        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1245        if (!new_fa)
1246                goto out;
1247
1248        new_fa->fa_info = fi;
1249        new_fa->fa_tos = tos;
1250        new_fa->fa_type = cfg->fc_type;
1251        new_fa->fa_state = 0;
1252        new_fa->fa_slen = slen;
1253        new_fa->tb_id = tb->tb_id;
1254        new_fa->fa_default = -1;
1255
1256        /* Insert new entry to the list. */
1257        err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1258        if (err)
1259                goto out_free_new_fa;
1260
1261        if (!plen)
1262                tb->tb_num_default++;
1263
1264        rt_cache_flush(cfg->fc_nlinfo.nl_net);
1265        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_ADD, key, plen, fi, tos,
1266                                 cfg->fc_type, tb->tb_id, cfg->fc_nlflags);
1267        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1268                  &cfg->fc_nlinfo, nlflags);
1269succeeded:
1270        return 0;
1271
1272out_free_new_fa:
1273        kmem_cache_free(fn_alias_kmem, new_fa);
1274out:
1275        fib_release_info(fi);
1276err:
1277        return err;
1278}
1279
1280static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1281{
1282        t_key prefix = n->key;
1283
1284        return (key ^ prefix) & (prefix | -prefix);
1285}
1286
1287/* should be called with rcu_read_lock */
1288int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1289                     struct fib_result *res, int fib_flags)
1290{
1291        struct trie *t = (struct trie *) tb->tb_data;
1292#ifdef CONFIG_IP_FIB_TRIE_STATS
1293        struct trie_use_stats __percpu *stats = t->stats;
1294#endif
1295        const t_key key = ntohl(flp->daddr);
1296        struct key_vector *n, *pn;
1297        struct fib_alias *fa;
1298        unsigned long index;
1299        t_key cindex;
1300
1301        trace_fib_table_lookup(tb->tb_id, flp);
1302
1303        pn = t->kv;
1304        cindex = 0;
1305
1306        n = get_child_rcu(pn, cindex);
1307        if (!n)
1308                return -EAGAIN;
1309
1310#ifdef CONFIG_IP_FIB_TRIE_STATS
1311        this_cpu_inc(stats->gets);
1312#endif
1313
1314        /* Step 1: Travel to the longest prefix match in the trie */
1315        for (;;) {
1316                index = get_cindex(key, n);
1317
1318                /* This bit of code is a bit tricky but it combines multiple
1319                 * checks into a single check.  The prefix consists of the
1320                 * prefix plus zeros for the "bits" in the prefix. The index
1321                 * is the difference between the key and this value.  From
1322                 * this we can actually derive several pieces of data.
1323                 *   if (index >= (1ul << bits))
1324                 *     we have a mismatch in skip bits and failed
1325                 *   else
1326                 *     we know the value is cindex
1327                 *
1328                 * This check is safe even if bits == KEYLENGTH due to the
1329                 * fact that we can only allocate a node with 32 bits if a
1330                 * long is greater than 32 bits.
1331                 */
1332                if (index >= (1ul << n->bits))
1333                        break;
1334
1335                /* we have found a leaf. Prefixes have already been compared */
1336                if (IS_LEAF(n))
1337                        goto found;
1338
1339                /* only record pn and cindex if we are going to be chopping
1340                 * bits later.  Otherwise we are just wasting cycles.
1341                 */
1342                if (n->slen > n->pos) {
1343                        pn = n;
1344                        cindex = index;
1345                }
1346
1347                n = get_child_rcu(n, index);
1348                if (unlikely(!n))
1349                        goto backtrace;
1350        }
1351
1352        /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1353        for (;;) {
1354                /* record the pointer where our next node pointer is stored */
1355                struct key_vector __rcu **cptr = n->tnode;
1356
1357                /* This test verifies that none of the bits that differ
1358                 * between the key and the prefix exist in the region of
1359                 * the lsb and higher in the prefix.
1360                 */
1361                if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1362                        goto backtrace;
1363
1364                /* exit out and process leaf */
1365                if (unlikely(IS_LEAF(n)))
1366                        break;
1367
1368                /* Don't bother recording parent info.  Since we are in
1369                 * prefix match mode we will have to come back to wherever
1370                 * we started this traversal anyway
1371                 */
1372
1373                while ((n = rcu_dereference(*cptr)) == NULL) {
1374backtrace:
1375#ifdef CONFIG_IP_FIB_TRIE_STATS
1376                        if (!n)
1377                                this_cpu_inc(stats->null_node_hit);
1378#endif
1379                        /* If we are at cindex 0 there are no more bits for
1380                         * us to strip at this level so we must ascend back
1381                         * up one level to see if there are any more bits to
1382                         * be stripped there.
1383                         */
1384                        while (!cindex) {
1385                                t_key pkey = pn->key;
1386
1387                                /* If we don't have a parent then there is
1388                                 * nothing for us to do as we do not have any
1389                                 * further nodes to parse.
1390                                 */
1391                                if (IS_TRIE(pn))
1392                                        return -EAGAIN;
1393#ifdef CONFIG_IP_FIB_TRIE_STATS
1394                                this_cpu_inc(stats->backtrack);
1395#endif
1396                                /* Get Child's index */
1397                                pn = node_parent_rcu(pn);
1398                                cindex = get_index(pkey, pn);
1399                        }
1400
1401                        /* strip the least significant bit from the cindex */
1402                        cindex &= cindex - 1;
1403
1404                        /* grab pointer for next child node */
1405                        cptr = &pn->tnode[cindex];
1406                }
1407        }
1408
1409found:
1410        /* this line carries forward the xor from earlier in the function */
1411        index = key ^ n->key;
1412
1413        /* Step 3: Process the leaf, if that fails fall back to backtracing */
1414        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1415                struct fib_info *fi = fa->fa_info;
1416                int nhsel, err;
1417
1418                if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1419                        if (index >= (1ul << fa->fa_slen))
1420                                continue;
1421                }
1422                if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1423                        continue;
1424                if (fi->fib_dead)
1425                        continue;
1426                if (fa->fa_info->fib_scope < flp->flowi4_scope)
1427                        continue;
1428                fib_alias_accessed(fa);
1429                err = fib_props[fa->fa_type].error;
1430                if (unlikely(err < 0)) {
1431#ifdef CONFIG_IP_FIB_TRIE_STATS
1432                        this_cpu_inc(stats->semantic_match_passed);
1433#endif
1434                        return err;
1435                }
1436                if (fi->fib_flags & RTNH_F_DEAD)
1437                        continue;
1438                for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1439                        const struct fib_nh *nh = &fi->fib_nh[nhsel];
1440                        struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1441
1442                        if (nh->nh_flags & RTNH_F_DEAD)
1443                                continue;
1444                        if (in_dev &&
1445                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1446                            nh->nh_flags & RTNH_F_LINKDOWN &&
1447                            !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1448                                continue;
1449                        if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1450                                if (flp->flowi4_oif &&
1451                                    flp->flowi4_oif != nh->nh_oif)
1452                                        continue;
1453                        }
1454
1455                        if (!(fib_flags & FIB_LOOKUP_NOREF))
1456                                atomic_inc(&fi->fib_clntref);
1457
1458                        res->prefixlen = KEYLENGTH - fa->fa_slen;
1459                        res->nh_sel = nhsel;
1460                        res->type = fa->fa_type;
1461                        res->scope = fi->fib_scope;
1462                        res->fi = fi;
1463                        res->table = tb;
1464                        res->fa_head = &n->leaf;
1465#ifdef CONFIG_IP_FIB_TRIE_STATS
1466                        this_cpu_inc(stats->semantic_match_passed);
1467#endif
1468                        trace_fib_table_lookup_nh(nh);
1469
1470                        return err;
1471                }
1472        }
1473#ifdef CONFIG_IP_FIB_TRIE_STATS
1474        this_cpu_inc(stats->semantic_match_miss);
1475#endif
1476        goto backtrace;
1477}
1478EXPORT_SYMBOL_GPL(fib_table_lookup);
1479
1480static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1481                             struct key_vector *l, struct fib_alias *old)
1482{
1483        /* record the location of the previous list_info entry */
1484        struct hlist_node **pprev = old->fa_list.pprev;
1485        struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1486
1487        /* remove the fib_alias from the list */
1488        hlist_del_rcu(&old->fa_list);
1489
1490        /* if we emptied the list this leaf will be freed and we can sort
1491         * out parent suffix lengths as a part of trie_rebalance
1492         */
1493        if (hlist_empty(&l->leaf)) {
1494                if (tp->slen == l->slen)
1495                        node_pull_suffix(tp, tp->pos);
1496                put_child_root(tp, l->key, NULL);
1497                node_free(l);
1498                trie_rebalance(t, tp);
1499                return;
1500        }
1501
1502        /* only access fa if it is pointing at the last valid hlist_node */
1503        if (*pprev)
1504                return;
1505
1506        /* update the trie with the latest suffix length */
1507        l->slen = fa->fa_slen;
1508        node_pull_suffix(tp, fa->fa_slen);
1509}
1510
1511/* Caller must hold RTNL. */
1512int fib_table_delete(struct net *net, struct fib_table *tb,
1513                     struct fib_config *cfg)
1514{
1515        struct trie *t = (struct trie *) tb->tb_data;
1516        struct fib_alias *fa, *fa_to_delete;
1517        struct key_vector *l, *tp;
1518        u8 plen = cfg->fc_dst_len;
1519        u8 slen = KEYLENGTH - plen;
1520        u8 tos = cfg->fc_tos;
1521        u32 key;
1522
1523        if (plen > KEYLENGTH)
1524                return -EINVAL;
1525
1526        key = ntohl(cfg->fc_dst);
1527
1528        if ((plen < KEYLENGTH) && (key << plen))
1529                return -EINVAL;
1530
1531        l = fib_find_node(t, &tp, key);
1532        if (!l)
1533                return -ESRCH;
1534
1535        fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1536        if (!fa)
1537                return -ESRCH;
1538
1539        pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1540
1541        fa_to_delete = NULL;
1542        hlist_for_each_entry_from(fa, fa_list) {
1543                struct fib_info *fi = fa->fa_info;
1544
1545                if ((fa->fa_slen != slen) ||
1546                    (fa->tb_id != tb->tb_id) ||
1547                    (fa->fa_tos != tos))
1548                        break;
1549
1550                if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1551                    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1552                     fa->fa_info->fib_scope == cfg->fc_scope) &&
1553                    (!cfg->fc_prefsrc ||
1554                     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1555                    (!cfg->fc_protocol ||
1556                     fi->fib_protocol == cfg->fc_protocol) &&
1557                    fib_nh_match(cfg, fi) == 0) {
1558                        fa_to_delete = fa;
1559                        break;
1560                }
1561        }
1562
1563        if (!fa_to_delete)
1564                return -ESRCH;
1565
1566        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1567                                 fa_to_delete->fa_info, tos, cfg->fc_type,
1568                                 tb->tb_id, 0);
1569        rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1570                  &cfg->fc_nlinfo, 0);
1571
1572        if (!plen)
1573                tb->tb_num_default--;
1574
1575        fib_remove_alias(t, tp, l, fa_to_delete);
1576
1577        if (fa_to_delete->fa_state & FA_S_ACCESSED)
1578                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1579
1580        fib_release_info(fa_to_delete->fa_info);
1581        alias_free_mem_rcu(fa_to_delete);
1582        return 0;
1583}
1584
1585/* Scan for the next leaf starting at the provided key value */
1586static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1587{
1588        struct key_vector *pn, *n = *tn;
1589        unsigned long cindex;
1590
1591        /* this loop is meant to try and find the key in the trie */
1592        do {
1593                /* record parent and next child index */
1594                pn = n;
1595                cindex = (key > pn->key) ? get_index(key, pn) : 0;
1596
1597                if (cindex >> pn->bits)
1598                        break;
1599
1600                /* descend into the next child */
1601                n = get_child_rcu(pn, cindex++);
1602                if (!n)
1603                        break;
1604
1605                /* guarantee forward progress on the keys */
1606                if (IS_LEAF(n) && (n->key >= key))
1607                        goto found;
1608        } while (IS_TNODE(n));
1609
1610        /* this loop will search for the next leaf with a greater key */
1611        while (!IS_TRIE(pn)) {
1612                /* if we exhausted the parent node we will need to climb */
1613                if (cindex >= (1ul << pn->bits)) {
1614                        t_key pkey = pn->key;
1615
1616                        pn = node_parent_rcu(pn);
1617                        cindex = get_index(pkey, pn) + 1;
1618                        continue;
1619                }
1620
1621                /* grab the next available node */
1622                n = get_child_rcu(pn, cindex++);
1623                if (!n)
1624                        continue;
1625
1626                /* no need to compare keys since we bumped the index */
1627                if (IS_LEAF(n))
1628                        goto found;
1629
1630                /* Rescan start scanning in new node */
1631                pn = n;
1632                cindex = 0;
1633        }
1634
1635        *tn = pn;
1636        return NULL; /* Root of trie */
1637found:
1638        /* if we are at the limit for keys just return NULL for the tnode */
1639        *tn = pn;
1640        return n;
1641}
1642
1643static void fib_trie_free(struct fib_table *tb)
1644{
1645        struct trie *t = (struct trie *)tb->tb_data;
1646        struct key_vector *pn = t->kv;
1647        unsigned long cindex = 1;
1648        struct hlist_node *tmp;
1649        struct fib_alias *fa;
1650
1651        /* walk trie in reverse order and free everything */
1652        for (;;) {
1653                struct key_vector *n;
1654
1655                if (!(cindex--)) {
1656                        t_key pkey = pn->key;
1657
1658                        if (IS_TRIE(pn))
1659                                break;
1660
1661                        n = pn;
1662                        pn = node_parent(pn);
1663
1664                        /* drop emptied tnode */
1665                        put_child_root(pn, n->key, NULL);
1666                        node_free(n);
1667
1668                        cindex = get_index(pkey, pn);
1669
1670                        continue;
1671                }
1672
1673                /* grab the next available node */
1674                n = get_child(pn, cindex);
1675                if (!n)
1676                        continue;
1677
1678                if (IS_TNODE(n)) {
1679                        /* record pn and cindex for leaf walking */
1680                        pn = n;
1681                        cindex = 1ul << n->bits;
1682
1683                        continue;
1684                }
1685
1686                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1687                        hlist_del_rcu(&fa->fa_list);
1688                        alias_free_mem_rcu(fa);
1689                }
1690
1691                put_child_root(pn, n->key, NULL);
1692                node_free(n);
1693        }
1694
1695#ifdef CONFIG_IP_FIB_TRIE_STATS
1696        free_percpu(t->stats);
1697#endif
1698        kfree(tb);
1699}
1700
1701struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1702{
1703        struct trie *ot = (struct trie *)oldtb->tb_data;
1704        struct key_vector *l, *tp = ot->kv;
1705        struct fib_table *local_tb;
1706        struct fib_alias *fa;
1707        struct trie *lt;
1708        t_key key = 0;
1709
1710        if (oldtb->tb_data == oldtb->__data)
1711                return oldtb;
1712
1713        local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1714        if (!local_tb)
1715                return NULL;
1716
1717        lt = (struct trie *)local_tb->tb_data;
1718
1719        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1720                struct key_vector *local_l = NULL, *local_tp;
1721
1722                hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1723                        struct fib_alias *new_fa;
1724
1725                        if (local_tb->tb_id != fa->tb_id)
1726                                continue;
1727
1728                        /* clone fa for new local table */
1729                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1730                        if (!new_fa)
1731                                goto out;
1732
1733                        memcpy(new_fa, fa, sizeof(*fa));
1734
1735                        /* insert clone into table */
1736                        if (!local_l)
1737                                local_l = fib_find_node(lt, &local_tp, l->key);
1738
1739                        if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1740                                             NULL, l->key)) {
1741                                kmem_cache_free(fn_alias_kmem, new_fa);
1742                                goto out;
1743                        }
1744                }
1745
1746                /* stop loop if key wrapped back to 0 */
1747                key = l->key + 1;
1748                if (key < l->key)
1749                        break;
1750        }
1751
1752        return local_tb;
1753out:
1754        fib_trie_free(local_tb);
1755
1756        return NULL;
1757}
1758
1759/* Caller must hold RTNL */
1760void fib_table_flush_external(struct fib_table *tb)
1761{
1762        struct trie *t = (struct trie *)tb->tb_data;
1763        struct key_vector *pn = t->kv;
1764        unsigned long cindex = 1;
1765        struct hlist_node *tmp;
1766        struct fib_alias *fa;
1767
1768        /* walk trie in reverse order */
1769        for (;;) {
1770                unsigned char slen = 0;
1771                struct key_vector *n;
1772
1773                if (!(cindex--)) {
1774                        t_key pkey = pn->key;
1775
1776                        /* cannot resize the trie vector */
1777                        if (IS_TRIE(pn))
1778                                break;
1779
1780                        /* update the suffix to address pulled leaves */
1781                        if (pn->slen > pn->pos)
1782                                update_suffix(pn);
1783
1784                        /* resize completed node */
1785                        pn = resize(t, pn);
1786                        cindex = get_index(pkey, pn);
1787
1788                        continue;
1789                }
1790
1791                /* grab the next available node */
1792                n = get_child(pn, cindex);
1793                if (!n)
1794                        continue;
1795
1796                if (IS_TNODE(n)) {
1797                        /* record pn and cindex for leaf walking */
1798                        pn = n;
1799                        cindex = 1ul << n->bits;
1800
1801                        continue;
1802                }
1803
1804                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1805                        /* if alias was cloned to local then we just
1806                         * need to remove the local copy from main
1807                         */
1808                        if (tb->tb_id != fa->tb_id) {
1809                                hlist_del_rcu(&fa->fa_list);
1810                                alias_free_mem_rcu(fa);
1811                                continue;
1812                        }
1813
1814                        /* record local slen */
1815                        slen = fa->fa_slen;
1816                }
1817
1818                /* update leaf slen */
1819                n->slen = slen;
1820
1821                if (hlist_empty(&n->leaf)) {
1822                        put_child_root(pn, n->key, NULL);
1823                        node_free(n);
1824                }
1825        }
1826}
1827
1828/* Caller must hold RTNL. */
1829int fib_table_flush(struct net *net, struct fib_table *tb)
1830{
1831        struct trie *t = (struct trie *)tb->tb_data;
1832        struct key_vector *pn = t->kv;
1833        unsigned long cindex = 1;
1834        struct hlist_node *tmp;
1835        struct fib_alias *fa;
1836        int found = 0;
1837
1838        /* walk trie in reverse order */
1839        for (;;) {
1840                unsigned char slen = 0;
1841                struct key_vector *n;
1842
1843                if (!(cindex--)) {
1844                        t_key pkey = pn->key;
1845
1846                        /* cannot resize the trie vector */
1847                        if (IS_TRIE(pn))
1848                                break;
1849
1850                        /* update the suffix to address pulled leaves */
1851                        if (pn->slen > pn->pos)
1852                                update_suffix(pn);
1853
1854                        /* resize completed node */
1855                        pn = resize(t, pn);
1856                        cindex = get_index(pkey, pn);
1857
1858                        continue;
1859                }
1860
1861                /* grab the next available node */
1862                n = get_child(pn, cindex);
1863                if (!n)
1864                        continue;
1865
1866                if (IS_TNODE(n)) {
1867                        /* record pn and cindex for leaf walking */
1868                        pn = n;
1869                        cindex = 1ul << n->bits;
1870
1871                        continue;
1872                }
1873
1874                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1875                        struct fib_info *fi = fa->fa_info;
1876
1877                        if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1878                                slen = fa->fa_slen;
1879                                continue;
1880                        }
1881
1882                        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1883                                                 n->key,
1884                                                 KEYLENGTH - fa->fa_slen,
1885                                                 fi, fa->fa_tos, fa->fa_type,
1886                                                 tb->tb_id, 0);
1887                        hlist_del_rcu(&fa->fa_list);
1888                        fib_release_info(fa->fa_info);
1889                        alias_free_mem_rcu(fa);
1890                        found++;
1891                }
1892
1893                /* update leaf slen */
1894                n->slen = slen;
1895
1896                if (hlist_empty(&n->leaf)) {
1897                        put_child_root(pn, n->key, NULL);
1898                        node_free(n);
1899                }
1900        }
1901
1902        pr_debug("trie_flush found=%d\n", found);
1903        return found;
1904}
1905
1906static void __trie_free_rcu(struct rcu_head *head)
1907{
1908        struct fib_table *tb = container_of(head, struct fib_table, rcu);
1909#ifdef CONFIG_IP_FIB_TRIE_STATS
1910        struct trie *t = (struct trie *)tb->tb_data;
1911
1912        if (tb->tb_data == tb->__data)
1913                free_percpu(t->stats);
1914#endif /* CONFIG_IP_FIB_TRIE_STATS */
1915        kfree(tb);
1916}
1917
1918void fib_free_table(struct fib_table *tb)
1919{
1920        call_rcu(&tb->rcu, __trie_free_rcu);
1921}
1922
1923static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1924                             struct sk_buff *skb, struct netlink_callback *cb)
1925{
1926        __be32 xkey = htonl(l->key);
1927        struct fib_alias *fa;
1928        int i, s_i;
1929
1930        s_i = cb->args[4];
1931        i = 0;
1932
1933        /* rcu_read_lock is hold by caller */
1934        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1935                if (i < s_i) {
1936                        i++;
1937                        continue;
1938                }
1939
1940                if (tb->tb_id != fa->tb_id) {
1941                        i++;
1942                        continue;
1943                }
1944
1945                if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1946                                  cb->nlh->nlmsg_seq,
1947                                  RTM_NEWROUTE,
1948                                  tb->tb_id,
1949                                  fa->fa_type,
1950                                  xkey,
1951                                  KEYLENGTH - fa->fa_slen,
1952                                  fa->fa_tos,
1953                                  fa->fa_info, NLM_F_MULTI) < 0) {
1954                        cb->args[4] = i;
1955                        return -1;
1956                }
1957                i++;
1958        }
1959
1960        cb->args[4] = i;
1961        return skb->len;
1962}
1963
1964/* rcu_read_lock needs to be hold by caller from readside */
1965int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1966                   struct netlink_callback *cb)
1967{
1968        struct trie *t = (struct trie *)tb->tb_data;
1969        struct key_vector *l, *tp = t->kv;
1970        /* Dump starting at last key.
1971         * Note: 0.0.0.0/0 (ie default) is first key.
1972         */
1973        int count = cb->args[2];
1974        t_key key = cb->args[3];
1975
1976        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1977                if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1978                        cb->args[3] = key;
1979                        cb->args[2] = count;
1980                        return -1;
1981                }
1982
1983                ++count;
1984                key = l->key + 1;
1985
1986                memset(&cb->args[4], 0,
1987                       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1988
1989                /* stop loop if key wrapped back to 0 */
1990                if (key < l->key)
1991                        break;
1992        }
1993
1994        cb->args[3] = key;
1995        cb->args[2] = count;
1996
1997        return skb->len;
1998}
1999
2000void __init fib_trie_init(void)
2001{
2002        fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2003                                          sizeof(struct fib_alias),
2004                                          0, SLAB_PANIC, NULL);
2005
2006        trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2007                                           LEAF_SIZE,
2008                                           0, SLAB_PANIC, NULL);
2009}
2010
2011struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2012{
2013        struct fib_table *tb;
2014        struct trie *t;
2015        size_t sz = sizeof(*tb);
2016
2017        if (!alias)
2018                sz += sizeof(struct trie);
2019
2020        tb = kzalloc(sz, GFP_KERNEL);
2021        if (!tb)
2022                return NULL;
2023
2024        tb->tb_id = id;
2025        tb->tb_num_default = 0;
2026        tb->tb_data = (alias ? alias->__data : tb->__data);
2027
2028        if (alias)
2029                return tb;
2030
2031        t = (struct trie *) tb->tb_data;
2032        t->kv[0].pos = KEYLENGTH;
2033        t->kv[0].slen = KEYLENGTH;
2034#ifdef CONFIG_IP_FIB_TRIE_STATS
2035        t->stats = alloc_percpu(struct trie_use_stats);
2036        if (!t->stats) {
2037                kfree(tb);
2038                tb = NULL;
2039        }
2040#endif
2041
2042        return tb;
2043}
2044
2045#ifdef CONFIG_PROC_FS
2046/* Depth first Trie walk iterator */
2047struct fib_trie_iter {
2048        struct seq_net_private p;
2049        struct fib_table *tb;
2050        struct key_vector *tnode;
2051        unsigned int index;
2052        unsigned int depth;
2053};
2054
2055static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2056{
2057        unsigned long cindex = iter->index;
2058        struct key_vector *pn = iter->tnode;
2059        t_key pkey;
2060
2061        pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2062                 iter->tnode, iter->index, iter->depth);
2063
2064        while (!IS_TRIE(pn)) {
2065                while (cindex < child_length(pn)) {
2066                        struct key_vector *n = get_child_rcu(pn, cindex++);
2067
2068                        if (!n)
2069                                continue;
2070
2071                        if (IS_LEAF(n)) {
2072                                iter->tnode = pn;
2073                                iter->index = cindex;
2074                        } else {
2075                                /* push down one level */
2076                                iter->tnode = n;
2077                                iter->index = 0;
2078                                ++iter->depth;
2079                        }
2080
2081                        return n;
2082                }
2083
2084                /* Current node exhausted, pop back up */
2085                pkey = pn->key;
2086                pn = node_parent_rcu(pn);
2087                cindex = get_index(pkey, pn) + 1;
2088                --iter->depth;
2089        }
2090
2091        /* record root node so further searches know we are done */
2092        iter->tnode = pn;
2093        iter->index = 0;
2094
2095        return NULL;
2096}
2097
2098static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2099                                             struct trie *t)
2100{
2101        struct key_vector *n, *pn;
2102
2103        if (!t)
2104                return NULL;
2105
2106        pn = t->kv;
2107        n = rcu_dereference(pn->tnode[0]);
2108        if (!n)
2109                return NULL;
2110
2111        if (IS_TNODE(n)) {
2112                iter->tnode = n;
2113                iter->index = 0;
2114                iter->depth = 1;
2115        } else {
2116                iter->tnode = pn;
2117                iter->index = 0;
2118                iter->depth = 0;
2119        }
2120
2121        return n;
2122}
2123
2124static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2125{
2126        struct key_vector *n;
2127        struct fib_trie_iter iter;
2128
2129        memset(s, 0, sizeof(*s));
2130
2131        rcu_read_lock();
2132        for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2133                if (IS_LEAF(n)) {
2134                        struct fib_alias *fa;
2135
2136                        s->leaves++;
2137                        s->totdepth += iter.depth;
2138                        if (iter.depth > s->maxdepth)
2139                                s->maxdepth = iter.depth;
2140
2141                        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2142                                ++s->prefixes;
2143                } else {
2144                        s->tnodes++;
2145                        if (n->bits < MAX_STAT_DEPTH)
2146                                s->nodesizes[n->bits]++;
2147                        s->nullpointers += tn_info(n)->empty_children;
2148                }
2149        }
2150        rcu_read_unlock();
2151}
2152
2153/*
2154 *      This outputs /proc/net/fib_triestats
2155 */
2156static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2157{
2158        unsigned int i, max, pointers, bytes, avdepth;
2159
2160        if (stat->leaves)
2161                avdepth = stat->totdepth*100 / stat->leaves;
2162        else
2163                avdepth = 0;
2164
2165        seq_printf(seq, "\tAver depth:     %u.%02d\n",
2166                   avdepth / 100, avdepth % 100);
2167        seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2168
2169        seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2170        bytes = LEAF_SIZE * stat->leaves;
2171
2172        seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2173        bytes += sizeof(struct fib_alias) * stat->prefixes;
2174
2175        seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2176        bytes += TNODE_SIZE(0) * stat->tnodes;
2177
2178        max = MAX_STAT_DEPTH;
2179        while (max > 0 && stat->nodesizes[max-1] == 0)
2180                max--;
2181
2182        pointers = 0;
2183        for (i = 1; i < max; i++)
2184                if (stat->nodesizes[i] != 0) {
2185                        seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2186                        pointers += (1<<i) * stat->nodesizes[i];
2187                }
2188        seq_putc(seq, '\n');
2189        seq_printf(seq, "\tPointers: %u\n", pointers);
2190
2191        bytes += sizeof(struct key_vector *) * pointers;
2192        seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2193        seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2194}
2195
2196#ifdef CONFIG_IP_FIB_TRIE_STATS
2197static void trie_show_usage(struct seq_file *seq,
2198                            const struct trie_use_stats __percpu *stats)
2199{
2200        struct trie_use_stats s = { 0 };
2201        int cpu;
2202
2203        /* loop through all of the CPUs and gather up the stats */
2204        for_each_possible_cpu(cpu) {
2205                const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2206
2207                s.gets += pcpu->gets;
2208                s.backtrack += pcpu->backtrack;
2209                s.semantic_match_passed += pcpu->semantic_match_passed;
2210                s.semantic_match_miss += pcpu->semantic_match_miss;
2211                s.null_node_hit += pcpu->null_node_hit;
2212                s.resize_node_skipped += pcpu->resize_node_skipped;
2213        }
2214
2215        seq_printf(seq, "\nCounters:\n---------\n");
2216        seq_printf(seq, "gets = %u\n", s.gets);
2217        seq_printf(seq, "backtracks = %u\n", s.backtrack);
2218        seq_printf(seq, "semantic match passed = %u\n",
2219                   s.semantic_match_passed);
2220        seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2221        seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2222        seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2223}
2224#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2225
2226static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2227{
2228        if (tb->tb_id == RT_TABLE_LOCAL)
2229                seq_puts(seq, "Local:\n");
2230        else if (tb->tb_id == RT_TABLE_MAIN)
2231                seq_puts(seq, "Main:\n");
2232        else
2233                seq_printf(seq, "Id %d:\n", tb->tb_id);
2234}
2235
2236
2237static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2238{
2239        struct net *net = (struct net *)seq->private;
2240        unsigned int h;
2241
2242        seq_printf(seq,
2243                   "Basic info: size of leaf:"
2244                   " %Zd bytes, size of tnode: %Zd bytes.\n",
2245                   LEAF_SIZE, TNODE_SIZE(0));
2246
2247        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2248                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2249                struct fib_table *tb;
2250
2251                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2252                        struct trie *t = (struct trie *) tb->tb_data;
2253                        struct trie_stat stat;
2254
2255                        if (!t)
2256                                continue;
2257
2258                        fib_table_print(seq, tb);
2259
2260                        trie_collect_stats(t, &stat);
2261                        trie_show_stats(seq, &stat);
2262#ifdef CONFIG_IP_FIB_TRIE_STATS
2263                        trie_show_usage(seq, t->stats);
2264#endif
2265                }
2266        }
2267
2268        return 0;
2269}
2270
2271static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2272{
2273        return single_open_net(inode, file, fib_triestat_seq_show);
2274}
2275
2276static const struct file_operations fib_triestat_fops = {
2277        .owner  = THIS_MODULE,
2278        .open   = fib_triestat_seq_open,
2279        .read   = seq_read,
2280        .llseek = seq_lseek,
2281        .release = single_release_net,
2282};
2283
2284static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2285{
2286        struct fib_trie_iter *iter = seq->private;
2287        struct net *net = seq_file_net(seq);
2288        loff_t idx = 0;
2289        unsigned int h;
2290
2291        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2292                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2293                struct fib_table *tb;
2294
2295                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2296                        struct key_vector *n;
2297
2298                        for (n = fib_trie_get_first(iter,
2299                                                    (struct trie *) tb->tb_data);
2300                             n; n = fib_trie_get_next(iter))
2301                                if (pos == idx++) {
2302                                        iter->tb = tb;
2303                                        return n;
2304                                }
2305                }
2306        }
2307
2308        return NULL;
2309}
2310
2311static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2312        __acquires(RCU)
2313{
2314        rcu_read_lock();
2315        return fib_trie_get_idx(seq, *pos);
2316}
2317
2318static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2319{
2320        struct fib_trie_iter *iter = seq->private;
2321        struct net *net = seq_file_net(seq);
2322        struct fib_table *tb = iter->tb;
2323        struct hlist_node *tb_node;
2324        unsigned int h;
2325        struct key_vector *n;
2326
2327        ++*pos;
2328        /* next node in same table */
2329        n = fib_trie_get_next(iter);
2330        if (n)
2331                return n;
2332
2333        /* walk rest of this hash chain */
2334        h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2335        while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2336                tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2337                n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2338                if (n)
2339                        goto found;
2340        }
2341
2342        /* new hash chain */
2343        while (++h < FIB_TABLE_HASHSZ) {
2344                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2345                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2346                        n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2347                        if (n)
2348                                goto found;
2349                }
2350        }
2351        return NULL;
2352
2353found:
2354        iter->tb = tb;
2355        return n;
2356}
2357
2358static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2359        __releases(RCU)
2360{
2361        rcu_read_unlock();
2362}
2363
2364static void seq_indent(struct seq_file *seq, int n)
2365{
2366        while (n-- > 0)
2367                seq_puts(seq, "   ");
2368}
2369
2370static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2371{
2372        switch (s) {
2373        case RT_SCOPE_UNIVERSE: return "universe";
2374        case RT_SCOPE_SITE:     return "site";
2375        case RT_SCOPE_LINK:     return "link";
2376        case RT_SCOPE_HOST:     return "host";
2377        case RT_SCOPE_NOWHERE:  return "nowhere";
2378        default:
2379                snprintf(buf, len, "scope=%d", s);
2380                return buf;
2381        }
2382}
2383
2384static const char *const rtn_type_names[__RTN_MAX] = {
2385        [RTN_UNSPEC] = "UNSPEC",
2386        [RTN_UNICAST] = "UNICAST",
2387        [RTN_LOCAL] = "LOCAL",
2388        [RTN_BROADCAST] = "BROADCAST",
2389        [RTN_ANYCAST] = "ANYCAST",
2390        [RTN_MULTICAST] = "MULTICAST",
2391        [RTN_BLACKHOLE] = "BLACKHOLE",
2392        [RTN_UNREACHABLE] = "UNREACHABLE",
2393        [RTN_PROHIBIT] = "PROHIBIT",
2394        [RTN_THROW] = "THROW",
2395        [RTN_NAT] = "NAT",
2396        [RTN_XRESOLVE] = "XRESOLVE",
2397};
2398
2399static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2400{
2401        if (t < __RTN_MAX && rtn_type_names[t])
2402                return rtn_type_names[t];
2403        snprintf(buf, len, "type %u", t);
2404        return buf;
2405}
2406
2407/* Pretty print the trie */
2408static int fib_trie_seq_show(struct seq_file *seq, void *v)
2409{
2410        const struct fib_trie_iter *iter = seq->private;
2411        struct key_vector *n = v;
2412
2413        if (IS_TRIE(node_parent_rcu(n)))
2414                fib_table_print(seq, iter->tb);
2415
2416        if (IS_TNODE(n)) {
2417                __be32 prf = htonl(n->key);
2418
2419                seq_indent(seq, iter->depth-1);
2420                seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2421                           &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2422                           tn_info(n)->full_children,
2423                           tn_info(n)->empty_children);
2424        } else {
2425                __be32 val = htonl(n->key);
2426                struct fib_alias *fa;
2427
2428                seq_indent(seq, iter->depth);
2429                seq_printf(seq, "  |-- %pI4\n", &val);
2430
2431                hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2432                        char buf1[32], buf2[32];
2433
2434                        seq_indent(seq, iter->depth + 1);
2435                        seq_printf(seq, "  /%zu %s %s",
2436                                   KEYLENGTH - fa->fa_slen,
2437                                   rtn_scope(buf1, sizeof(buf1),
2438                                             fa->fa_info->fib_scope),
2439                                   rtn_type(buf2, sizeof(buf2),
2440                                            fa->fa_type));
2441                        if (fa->fa_tos)
2442                                seq_printf(seq, " tos=%d", fa->fa_tos);
2443                        seq_putc(seq, '\n');
2444                }
2445        }
2446
2447        return 0;
2448}
2449
2450static const struct seq_operations fib_trie_seq_ops = {
2451        .start  = fib_trie_seq_start,
2452        .next   = fib_trie_seq_next,
2453        .stop   = fib_trie_seq_stop,
2454        .show   = fib_trie_seq_show,
2455};
2456
2457static int fib_trie_seq_open(struct inode *inode, struct file *file)
2458{
2459        return seq_open_net(inode, file, &fib_trie_seq_ops,
2460                            sizeof(struct fib_trie_iter));
2461}
2462
2463static const struct file_operations fib_trie_fops = {
2464        .owner  = THIS_MODULE,
2465        .open   = fib_trie_seq_open,
2466        .read   = seq_read,
2467        .llseek = seq_lseek,
2468        .release = seq_release_net,
2469};
2470
2471struct fib_route_iter {
2472        struct seq_net_private p;
2473        struct fib_table *main_tb;
2474        struct key_vector *tnode;
2475        loff_t  pos;
2476        t_key   key;
2477};
2478
2479static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2480                                            loff_t pos)
2481{
2482        struct key_vector *l, **tp = &iter->tnode;
2483        t_key key;
2484
2485        /* use cached location of previously found key */
2486        if (iter->pos > 0 && pos >= iter->pos) {
2487                key = iter->key;
2488        } else {
2489                iter->pos = 1;
2490                key = 0;
2491        }
2492
2493        pos -= iter->pos;
2494
2495        while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2496                key = l->key + 1;
2497                iter->pos++;
2498                l = NULL;
2499
2500                /* handle unlikely case of a key wrap */
2501                if (!key)
2502                        break;
2503        }
2504
2505        if (l)
2506                iter->key = l->key;     /* remember it */
2507        else
2508                iter->pos = 0;          /* forget it */
2509
2510        return l;
2511}
2512
2513static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2514        __acquires(RCU)
2515{
2516        struct fib_route_iter *iter = seq->private;
2517        struct fib_table *tb;
2518        struct trie *t;
2519
2520        rcu_read_lock();
2521
2522        tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2523        if (!tb)
2524                return NULL;
2525
2526        iter->main_tb = tb;
2527        t = (struct trie *)tb->tb_data;
2528        iter->tnode = t->kv;
2529
2530        if (*pos != 0)
2531                return fib_route_get_idx(iter, *pos);
2532
2533        iter->pos = 0;
2534        iter->key = KEY_MAX;
2535
2536        return SEQ_START_TOKEN;
2537}
2538
2539static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2540{
2541        struct fib_route_iter *iter = seq->private;
2542        struct key_vector *l = NULL;
2543        t_key key = iter->key + 1;
2544
2545        ++*pos;
2546
2547        /* only allow key of 0 for start of sequence */
2548        if ((v == SEQ_START_TOKEN) || key)
2549                l = leaf_walk_rcu(&iter->tnode, key);
2550
2551        if (l) {
2552                iter->key = l->key;
2553                iter->pos++;
2554        } else {
2555                iter->pos = 0;
2556        }
2557
2558        return l;
2559}
2560
2561static void fib_route_seq_stop(struct seq_file *seq, void *v)
2562        __releases(RCU)
2563{
2564        rcu_read_unlock();
2565}
2566
2567static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2568{
2569        unsigned int flags = 0;
2570
2571        if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2572                flags = RTF_REJECT;
2573        if (fi && fi->fib_nh->nh_gw)
2574                flags |= RTF_GATEWAY;
2575        if (mask == htonl(0xFFFFFFFF))
2576                flags |= RTF_HOST;
2577        flags |= RTF_UP;
2578        return flags;
2579}
2580
2581/*
2582 *      This outputs /proc/net/route.
2583 *      The format of the file is not supposed to be changed
2584 *      and needs to be same as fib_hash output to avoid breaking
2585 *      legacy utilities
2586 */
2587static int fib_route_seq_show(struct seq_file *seq, void *v)
2588{
2589        struct fib_route_iter *iter = seq->private;
2590        struct fib_table *tb = iter->main_tb;
2591        struct fib_alias *fa;
2592        struct key_vector *l = v;
2593        __be32 prefix;
2594
2595        if (v == SEQ_START_TOKEN) {
2596                seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2597                           "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2598                           "\tWindow\tIRTT");
2599                return 0;
2600        }
2601
2602        prefix = htonl(l->key);
2603
2604        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2605                const struct fib_info *fi = fa->fa_info;
2606                __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2607                unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2608
2609                if ((fa->fa_type == RTN_BROADCAST) ||
2610                    (fa->fa_type == RTN_MULTICAST))
2611                        continue;
2612
2613                if (fa->tb_id != tb->tb_id)
2614                        continue;
2615
2616                seq_setwidth(seq, 127);
2617
2618                if (fi)
2619                        seq_printf(seq,
2620                                   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2621                                   "%d\t%08X\t%d\t%u\t%u",
2622                                   fi->fib_dev ? fi->fib_dev->name : "*",
2623                                   prefix,
2624                                   fi->fib_nh->nh_gw, flags, 0, 0,
2625                                   fi->fib_priority,
2626                                   mask,
2627                                   (fi->fib_advmss ?
2628                                    fi->fib_advmss + 40 : 0),
2629                                   fi->fib_window,
2630                                   fi->fib_rtt >> 3);
2631                else
2632                        seq_printf(seq,
2633                                   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2634                                   "%d\t%08X\t%d\t%u\t%u",
2635                                   prefix, 0, flags, 0, 0, 0,
2636                                   mask, 0, 0, 0);
2637
2638                seq_pad(seq, '\n');
2639        }
2640
2641        return 0;
2642}
2643
2644static const struct seq_operations fib_route_seq_ops = {
2645        .start  = fib_route_seq_start,
2646        .next   = fib_route_seq_next,
2647        .stop   = fib_route_seq_stop,
2648        .show   = fib_route_seq_show,
2649};
2650
2651static int fib_route_seq_open(struct inode *inode, struct file *file)
2652{
2653        return seq_open_net(inode, file, &fib_route_seq_ops,
2654                            sizeof(struct fib_route_iter));
2655}
2656
2657static const struct file_operations fib_route_fops = {
2658        .owner  = THIS_MODULE,
2659        .open   = fib_route_seq_open,
2660        .read   = seq_read,
2661        .llseek = seq_lseek,
2662        .release = seq_release_net,
2663};
2664
2665int __net_init fib_proc_init(struct net *net)
2666{
2667        if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2668                goto out1;
2669
2670        if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2671                         &fib_triestat_fops))
2672                goto out2;
2673
2674        if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2675                goto out3;
2676
2677        return 0;
2678
2679out3:
2680        remove_proc_entry("fib_triestat", net->proc_net);
2681out2:
2682        remove_proc_entry("fib_trie", net->proc_net);
2683out1:
2684        return -ENOMEM;
2685}
2686
2687void __net_exit fib_proc_exit(struct net *net)
2688{
2689        remove_proc_entry("fib_trie", net->proc_net);
2690        remove_proc_entry("fib_triestat", net->proc_net);
2691        remove_proc_entry("route", net->proc_net);
2692}
2693
2694#endif /* CONFIG_PROC_FS */
2695