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 <net/net_namespace.h>
  77#include <net/ip.h>
  78#include <net/protocol.h>
  79#include <net/route.h>
  80#include <net/tcp.h>
  81#include <net/sock.h>
  82#include <net/ip_fib.h>
  83#include <net/switchdev.h>
  84#include <trace/events/fib.h>
  85#include "fib_lookup.h"
  86
  87#define MAX_STAT_DEPTH 32
  88
  89#define KEYLENGTH       (8*sizeof(t_key))
  90#define KEY_MAX         ((t_key)~0)
  91
  92typedef unsigned int t_key;
  93
  94#define IS_TRIE(n)      ((n)->pos >= KEYLENGTH)
  95#define IS_TNODE(n)     ((n)->bits)
  96#define IS_LEAF(n)      (!(n)->bits)
  97
  98struct key_vector {
  99        t_key key;
 100        unsigned char pos;              /* 2log(KEYLENGTH) bits needed */
 101        unsigned char bits;             /* 2log(KEYLENGTH) bits needed */
 102        unsigned char slen;
 103        union {
 104                /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 105                struct hlist_head leaf;
 106                /* This array is valid if (pos | bits) > 0 (TNODE) */
 107                struct key_vector __rcu *tnode[0];
 108        };
 109};
 110
 111struct tnode {
 112        struct rcu_head rcu;
 113        t_key empty_children;           /* KEYLENGTH bits needed */
 114        t_key full_children;            /* KEYLENGTH bits needed */
 115        struct key_vector __rcu *parent;
 116        struct key_vector kv[1];
 117#define tn_bits kv[0].bits
 118};
 119
 120#define TNODE_SIZE(n)   offsetof(struct tnode, kv[0].tnode[n])
 121#define LEAF_SIZE       TNODE_SIZE(1)
 122
 123#ifdef CONFIG_IP_FIB_TRIE_STATS
 124struct trie_use_stats {
 125        unsigned int gets;
 126        unsigned int backtrack;
 127        unsigned int semantic_match_passed;
 128        unsigned int semantic_match_miss;
 129        unsigned int null_node_hit;
 130        unsigned int resize_node_skipped;
 131};
 132#endif
 133
 134struct trie_stat {
 135        unsigned int totdepth;
 136        unsigned int maxdepth;
 137        unsigned int tnodes;
 138        unsigned int leaves;
 139        unsigned int nullpointers;
 140        unsigned int prefixes;
 141        unsigned int nodesizes[MAX_STAT_DEPTH];
 142};
 143
 144struct trie {
 145        struct key_vector kv[1];
 146#ifdef CONFIG_IP_FIB_TRIE_STATS
 147        struct trie_use_stats __percpu *stats;
 148#endif
 149};
 150
 151static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 152static size_t tnode_free_size;
 153
 154/*
 155 * synchronize_rcu after call_rcu for that many pages; it should be especially
 156 * useful before resizing the root node with PREEMPT_NONE configs; the value was
 157 * obtained experimentally, aiming to avoid visible slowdown.
 158 */
 159static const int sync_pages = 128;
 160
 161static struct kmem_cache *fn_alias_kmem __read_mostly;
 162static struct kmem_cache *trie_leaf_kmem __read_mostly;
 163
 164static inline struct tnode *tn_info(struct key_vector *kv)
 165{
 166        return container_of(kv, struct tnode, kv[0]);
 167}
 168
 169/* caller must hold RTNL */
 170#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 171#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 172
 173/* caller must hold RCU read lock or RTNL */
 174#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 175#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 176
 177/* wrapper for rcu_assign_pointer */
 178static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 179{
 180        if (n)
 181                rcu_assign_pointer(tn_info(n)->parent, tp);
 182}
 183
 184#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 185
 186/* This provides us with the number of children in this node, in the case of a
 187 * leaf this will return 0 meaning none of the children are accessible.
 188 */
 189static inline unsigned long child_length(const struct key_vector *tn)
 190{
 191        return (1ul << tn->bits) & ~(1ul);
 192}
 193
 194#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 195
 196static inline unsigned long get_index(t_key key, struct key_vector *kv)
 197{
 198        unsigned long index = key ^ kv->key;
 199
 200        if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 201                return 0;
 202
 203        return index >> kv->pos;
 204}
 205
 206/* To understand this stuff, an understanding of keys and all their bits is
 207 * necessary. Every node in the trie has a key associated with it, but not
 208 * all of the bits in that key are significant.
 209 *
 210 * Consider a node 'n' and its parent 'tp'.
 211 *
 212 * If n is a leaf, every bit in its key is significant. Its presence is
 213 * necessitated by path compression, since during a tree traversal (when
 214 * searching for a leaf - unless we are doing an insertion) we will completely
 215 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 216 * a potentially successful search, that we have indeed been walking the
 217 * correct key path.
 218 *
 219 * Note that we can never "miss" the correct key in the tree if present by
 220 * following the wrong path. Path compression ensures that segments of the key
 221 * that are the same for all keys with a given prefix are skipped, but the
 222 * skipped part *is* identical for each node in the subtrie below the skipped
 223 * bit! trie_insert() in this implementation takes care of that.
 224 *
 225 * if n is an internal node - a 'tnode' here, the various parts of its key
 226 * have many different meanings.
 227 *
 228 * Example:
 229 * _________________________________________________________________
 230 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 231 * -----------------------------------------------------------------
 232 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 233 *
 234 * _________________________________________________________________
 235 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 236 * -----------------------------------------------------------------
 237 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 238 *
 239 * tp->pos = 22
 240 * tp->bits = 3
 241 * n->pos = 13
 242 * n->bits = 4
 243 *
 244 * First, let's just ignore the bits that come before the parent tp, that is
 245 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 246 * point we do not use them for anything.
 247 *
 248 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 249 * index into the parent's child array. That is, they will be used to find
 250 * 'n' among tp's children.
 251 *
 252 * The bits from (n->pos + n->bits) to (tn->pos - 1) - "S" - are skipped bits
 253 * for the node n.
 254 *
 255 * All the bits we have seen so far are significant to the node n. The rest
 256 * of the bits are really not needed or indeed known in n->key.
 257 *
 258 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 259 * n's child array, and will of course be different for each child.
 260 *
 261 * The rest of the bits, from 0 to (n->pos + n->bits), are completely unknown
 262 * at this point.
 263 */
 264
 265static const int halve_threshold = 25;
 266static const int inflate_threshold = 50;
 267static const int halve_threshold_root = 15;
 268static const int inflate_threshold_root = 30;
 269
 270static void __alias_free_mem(struct rcu_head *head)
 271{
 272        struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 273        kmem_cache_free(fn_alias_kmem, fa);
 274}
 275
 276static inline void alias_free_mem_rcu(struct fib_alias *fa)
 277{
 278        call_rcu(&fa->rcu, __alias_free_mem);
 279}
 280
 281#define TNODE_KMALLOC_MAX \
 282        ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 283#define TNODE_VMALLOC_MAX \
 284        ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 285
 286static void __node_free_rcu(struct rcu_head *head)
 287{
 288        struct tnode *n = container_of(head, struct tnode, rcu);
 289
 290        if (!n->tn_bits)
 291                kmem_cache_free(trie_leaf_kmem, n);
 292        else
 293                kvfree(n);
 294}
 295
 296#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 297
 298static struct tnode *tnode_alloc(int bits)
 299{
 300        size_t size;
 301
 302        /* verify bits is within bounds */
 303        if (bits > TNODE_VMALLOC_MAX)
 304                return NULL;
 305
 306        /* determine size and verify it is non-zero and didn't overflow */
 307        size = TNODE_SIZE(1ul << bits);
 308
 309        if (size <= PAGE_SIZE)
 310                return kzalloc(size, GFP_KERNEL);
 311        else
 312                return vzalloc(size);
 313}
 314
 315static inline void empty_child_inc(struct key_vector *n)
 316{
 317        ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
 318}
 319
 320static inline void empty_child_dec(struct key_vector *n)
 321{
 322        tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
 323}
 324
 325static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 326{
 327        struct key_vector *l;
 328        struct tnode *kv;
 329
 330        kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 331        if (!kv)
 332                return NULL;
 333
 334        /* initialize key vector */
 335        l = kv->kv;
 336        l->key = key;
 337        l->pos = 0;
 338        l->bits = 0;
 339        l->slen = fa->fa_slen;
 340
 341        /* link leaf to fib alias */
 342        INIT_HLIST_HEAD(&l->leaf);
 343        hlist_add_head(&fa->fa_list, &l->leaf);
 344
 345        return l;
 346}
 347
 348static struct key_vector *tnode_new(t_key key, int pos, int bits)
 349{
 350        unsigned int shift = pos + bits;
 351        struct key_vector *tn;
 352        struct tnode *tnode;
 353
 354        /* verify bits and pos their msb bits clear and values are valid */
 355        BUG_ON(!bits || (shift > KEYLENGTH));
 356
 357        tnode = tnode_alloc(bits);
 358        if (!tnode)
 359                return NULL;
 360
 361        pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 362                 sizeof(struct key_vector *) << bits);
 363
 364        if (bits == KEYLENGTH)
 365                tnode->full_children = 1;
 366        else
 367                tnode->empty_children = 1ul << bits;
 368
 369        tn = tnode->kv;
 370        tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 371        tn->pos = pos;
 372        tn->bits = bits;
 373        tn->slen = pos;
 374
 375        return tn;
 376}
 377
 378/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 379 * and no bits are skipped. See discussion in dyntree paper p. 6
 380 */
 381static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 382{
 383        return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 384}
 385
 386/* Add a child at position i overwriting the old value.
 387 * Update the value of full_children and empty_children.
 388 */
 389static void put_child(struct key_vector *tn, unsigned long i,
 390                      struct key_vector *n)
 391{
 392        struct key_vector *chi = get_child(tn, i);
 393        int isfull, wasfull;
 394
 395        BUG_ON(i >= child_length(tn));
 396
 397        /* update emptyChildren, overflow into fullChildren */
 398        if (!n && chi)
 399                empty_child_inc(tn);
 400        if (n && !chi)
 401                empty_child_dec(tn);
 402
 403        /* update fullChildren */
 404        wasfull = tnode_full(tn, chi);
 405        isfull = tnode_full(tn, n);
 406
 407        if (wasfull && !isfull)
 408                tn_info(tn)->full_children--;
 409        else if (!wasfull && isfull)
 410                tn_info(tn)->full_children++;
 411
 412        if (n && (tn->slen < n->slen))
 413                tn->slen = n->slen;
 414
 415        rcu_assign_pointer(tn->tnode[i], n);
 416}
 417
 418static void update_children(struct key_vector *tn)
 419{
 420        unsigned long i;
 421
 422        /* update all of the child parent pointers */
 423        for (i = child_length(tn); i;) {
 424                struct key_vector *inode = get_child(tn, --i);
 425
 426                if (!inode)
 427                        continue;
 428
 429                /* Either update the children of a tnode that
 430                 * already belongs to us or update the child
 431                 * to point to ourselves.
 432                 */
 433                if (node_parent(inode) == tn)
 434                        update_children(inode);
 435                else
 436                        node_set_parent(inode, tn);
 437        }
 438}
 439
 440static inline void put_child_root(struct key_vector *tp, t_key key,
 441                                  struct key_vector *n)
 442{
 443        if (IS_TRIE(tp))
 444                rcu_assign_pointer(tp->tnode[0], n);
 445        else
 446                put_child(tp, get_index(key, tp), n);
 447}
 448
 449static inline void tnode_free_init(struct key_vector *tn)
 450{
 451        tn_info(tn)->rcu.next = NULL;
 452}
 453
 454static inline void tnode_free_append(struct key_vector *tn,
 455                                     struct key_vector *n)
 456{
 457        tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 458        tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 459}
 460
 461static void tnode_free(struct key_vector *tn)
 462{
 463        struct callback_head *head = &tn_info(tn)->rcu;
 464
 465        while (head) {
 466                head = head->next;
 467                tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 468                node_free(tn);
 469
 470                tn = container_of(head, struct tnode, rcu)->kv;
 471        }
 472
 473        if (tnode_free_size >= PAGE_SIZE * sync_pages) {
 474                tnode_free_size = 0;
 475                synchronize_rcu();
 476        }
 477}
 478
 479static struct key_vector *replace(struct trie *t,
 480                                  struct key_vector *oldtnode,
 481                                  struct key_vector *tn)
 482{
 483        struct key_vector *tp = node_parent(oldtnode);
 484        unsigned long i;
 485
 486        /* setup the parent pointer out of and back into this node */
 487        NODE_INIT_PARENT(tn, tp);
 488        put_child_root(tp, tn->key, tn);
 489
 490        /* update all of the child parent pointers */
 491        update_children(tn);
 492
 493        /* all pointers should be clean so we are done */
 494        tnode_free(oldtnode);
 495
 496        /* resize children now that oldtnode is freed */
 497        for (i = child_length(tn); i;) {
 498                struct key_vector *inode = get_child(tn, --i);
 499
 500                /* resize child node */
 501                if (tnode_full(tn, inode))
 502                        tn = resize(t, inode);
 503        }
 504
 505        return tp;
 506}
 507
 508static struct key_vector *inflate(struct trie *t,
 509                                  struct key_vector *oldtnode)
 510{
 511        struct key_vector *tn;
 512        unsigned long i;
 513        t_key m;
 514
 515        pr_debug("In inflate\n");
 516
 517        tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 518        if (!tn)
 519                goto notnode;
 520
 521        /* prepare oldtnode to be freed */
 522        tnode_free_init(oldtnode);
 523
 524        /* Assemble all of the pointers in our cluster, in this case that
 525         * represents all of the pointers out of our allocated nodes that
 526         * point to existing tnodes and the links between our allocated
 527         * nodes.
 528         */
 529        for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 530                struct key_vector *inode = get_child(oldtnode, --i);
 531                struct key_vector *node0, *node1;
 532                unsigned long j, k;
 533
 534                /* An empty child */
 535                if (!inode)
 536                        continue;
 537
 538                /* A leaf or an internal node with skipped bits */
 539                if (!tnode_full(oldtnode, inode)) {
 540                        put_child(tn, get_index(inode->key, tn), inode);
 541                        continue;
 542                }
 543
 544                /* drop the node in the old tnode free list */
 545                tnode_free_append(oldtnode, inode);
 546
 547                /* An internal node with two children */
 548                if (inode->bits == 1) {
 549                        put_child(tn, 2 * i + 1, get_child(inode, 1));
 550                        put_child(tn, 2 * i, get_child(inode, 0));
 551                        continue;
 552                }
 553
 554                /* We will replace this node 'inode' with two new
 555                 * ones, 'node0' and 'node1', each with half of the
 556                 * original children. The two new nodes will have
 557                 * a position one bit further down the key and this
 558                 * means that the "significant" part of their keys
 559                 * (see the discussion near the top of this file)
 560                 * will differ by one bit, which will be "0" in
 561                 * node0's key and "1" in node1's key. Since we are
 562                 * moving the key position by one step, the bit that
 563                 * we are moving away from - the bit at position
 564                 * (tn->pos) - is the one that will differ between
 565                 * node0 and node1. So... we synthesize that bit in the
 566                 * two new keys.
 567                 */
 568                node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 569                if (!node1)
 570                        goto nomem;
 571                node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 572
 573                tnode_free_append(tn, node1);
 574                if (!node0)
 575                        goto nomem;
 576                tnode_free_append(tn, node0);
 577
 578                /* populate child pointers in new nodes */
 579                for (k = child_length(inode), j = k / 2; j;) {
 580                        put_child(node1, --j, get_child(inode, --k));
 581                        put_child(node0, j, get_child(inode, j));
 582                        put_child(node1, --j, get_child(inode, --k));
 583                        put_child(node0, j, get_child(inode, j));
 584                }
 585
 586                /* link new nodes to parent */
 587                NODE_INIT_PARENT(node1, tn);
 588                NODE_INIT_PARENT(node0, tn);
 589
 590                /* link parent to nodes */
 591                put_child(tn, 2 * i + 1, node1);
 592                put_child(tn, 2 * i, node0);
 593        }
 594
 595        /* setup the parent pointers into and out of this node */
 596        return replace(t, oldtnode, tn);
 597nomem:
 598        /* all pointers should be clean so we are done */
 599        tnode_free(tn);
 600notnode:
 601        return NULL;
 602}
 603
 604static struct key_vector *halve(struct trie *t,
 605                                struct key_vector *oldtnode)
 606{
 607        struct key_vector *tn;
 608        unsigned long i;
 609
 610        pr_debug("In halve\n");
 611
 612        tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 613        if (!tn)
 614                goto notnode;
 615
 616        /* prepare oldtnode to be freed */
 617        tnode_free_init(oldtnode);
 618
 619        /* Assemble all of the pointers in our cluster, in this case that
 620         * represents all of the pointers out of our allocated nodes that
 621         * point to existing tnodes and the links between our allocated
 622         * nodes.
 623         */
 624        for (i = child_length(oldtnode); i;) {
 625                struct key_vector *node1 = get_child(oldtnode, --i);
 626                struct key_vector *node0 = get_child(oldtnode, --i);
 627                struct key_vector *inode;
 628
 629                /* At least one of the children is empty */
 630                if (!node1 || !node0) {
 631                        put_child(tn, i / 2, node1 ? : node0);
 632                        continue;
 633                }
 634
 635                /* Two nonempty children */
 636                inode = tnode_new(node0->key, oldtnode->pos, 1);
 637                if (!inode)
 638                        goto nomem;
 639                tnode_free_append(tn, inode);
 640
 641                /* initialize pointers out of node */
 642                put_child(inode, 1, node1);
 643                put_child(inode, 0, node0);
 644                NODE_INIT_PARENT(inode, tn);
 645
 646                /* link parent to node */
 647                put_child(tn, i / 2, inode);
 648        }
 649
 650        /* setup the parent pointers into and out of this node */
 651        return replace(t, oldtnode, tn);
 652nomem:
 653        /* all pointers should be clean so we are done */
 654        tnode_free(tn);
 655notnode:
 656        return NULL;
 657}
 658
 659static struct key_vector *collapse(struct trie *t,
 660                                   struct key_vector *oldtnode)
 661{
 662        struct key_vector *n, *tp;
 663        unsigned long i;
 664
 665        /* scan the tnode looking for that one child that might still exist */
 666        for (n = NULL, i = child_length(oldtnode); !n && i;)
 667                n = get_child(oldtnode, --i);
 668
 669        /* compress one level */
 670        tp = node_parent(oldtnode);
 671        put_child_root(tp, oldtnode->key, n);
 672        node_set_parent(n, tp);
 673
 674        /* drop dead node */
 675        node_free(oldtnode);
 676
 677        return tp;
 678}
 679
 680static unsigned char update_suffix(struct key_vector *tn)
 681{
 682        unsigned char slen = tn->pos;
 683        unsigned long stride, i;
 684
 685        /* search though the list of children looking for nodes that might
 686         * have a suffix greater than the one we currently have.  This is
 687         * why we start with a stride of 2 since a stride of 1 would
 688         * represent the nodes with suffix length equal to tn->pos
 689         */
 690        for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 691                struct key_vector *n = get_child(tn, i);
 692
 693                if (!n || (n->slen <= slen))
 694                        continue;
 695
 696                /* update stride and slen based on new value */
 697                stride <<= (n->slen - slen);
 698                slen = n->slen;
 699                i &= ~(stride - 1);
 700
 701                /* if slen covers all but the last bit we can stop here
 702                 * there will be nothing longer than that since only node
 703                 * 0 and 1 << (bits - 1) could have that as their suffix
 704                 * length.
 705                 */
 706                if ((slen + 1) >= (tn->pos + tn->bits))
 707                        break;
 708        }
 709
 710        tn->slen = slen;
 711
 712        return slen;
 713}
 714
 715/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 716 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 717 * Telecommunications, page 6:
 718 * "A node is doubled if the ratio of non-empty children to all
 719 * children in the *doubled* node is at least 'high'."
 720 *
 721 * 'high' in this instance is the variable 'inflate_threshold'. It
 722 * is expressed as a percentage, so we multiply it with
 723 * child_length() and instead of multiplying by 2 (since the
 724 * child array will be doubled by inflate()) and multiplying
 725 * the left-hand side by 100 (to handle the percentage thing) we
 726 * multiply the left-hand side by 50.
 727 *
 728 * The left-hand side may look a bit weird: child_length(tn)
 729 * - tn->empty_children is of course the number of non-null children
 730 * in the current node. tn->full_children is the number of "full"
 731 * children, that is non-null tnodes with a skip value of 0.
 732 * All of those will be doubled in the resulting inflated tnode, so
 733 * we just count them one extra time here.
 734 *
 735 * A clearer way to write this would be:
 736 *
 737 * to_be_doubled = tn->full_children;
 738 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 739 *     tn->full_children;
 740 *
 741 * new_child_length = child_length(tn) * 2;
 742 *
 743 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 744 *      new_child_length;
 745 * if (new_fill_factor >= inflate_threshold)
 746 *
 747 * ...and so on, tho it would mess up the while () loop.
 748 *
 749 * anyway,
 750 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 751 *      inflate_threshold
 752 *
 753 * avoid a division:
 754 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 755 *      inflate_threshold * new_child_length
 756 *
 757 * expand not_to_be_doubled and to_be_doubled, and shorten:
 758 * 100 * (child_length(tn) - tn->empty_children +
 759 *    tn->full_children) >= inflate_threshold * new_child_length
 760 *
 761 * expand new_child_length:
 762 * 100 * (child_length(tn) - tn->empty_children +
 763 *    tn->full_children) >=
 764 *      inflate_threshold * child_length(tn) * 2
 765 *
 766 * shorten again:
 767 * 50 * (tn->full_children + child_length(tn) -
 768 *    tn->empty_children) >= inflate_threshold *
 769 *    child_length(tn)
 770 *
 771 */
 772static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 773{
 774        unsigned long used = child_length(tn);
 775        unsigned long threshold = used;
 776
 777        /* Keep root node larger */
 778        threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 779        used -= tn_info(tn)->empty_children;
 780        used += tn_info(tn)->full_children;
 781
 782        /* if bits == KEYLENGTH then pos = 0, and will fail below */
 783
 784        return (used > 1) && tn->pos && ((50 * used) >= threshold);
 785}
 786
 787static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 788{
 789        unsigned long used = child_length(tn);
 790        unsigned long threshold = used;
 791
 792        /* Keep root node larger */
 793        threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 794        used -= tn_info(tn)->empty_children;
 795
 796        /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 797
 798        return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 799}
 800
 801static inline bool should_collapse(struct key_vector *tn)
 802{
 803        unsigned long used = child_length(tn);
 804
 805        used -= tn_info(tn)->empty_children;
 806
 807        /* account for bits == KEYLENGTH case */
 808        if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 809                used -= KEY_MAX;
 810
 811        /* One child or none, time to drop us from the trie */
 812        return used < 2;
 813}
 814
 815#define MAX_WORK 10
 816static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 817{
 818#ifdef CONFIG_IP_FIB_TRIE_STATS
 819        struct trie_use_stats __percpu *stats = t->stats;
 820#endif
 821        struct key_vector *tp = node_parent(tn);
 822        unsigned long cindex = get_index(tn->key, tp);
 823        int max_work = MAX_WORK;
 824
 825        pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 826                 tn, inflate_threshold, halve_threshold);
 827
 828        /* track the tnode via the pointer from the parent instead of
 829         * doing it ourselves.  This way we can let RCU fully do its
 830         * thing without us interfering
 831         */
 832        BUG_ON(tn != get_child(tp, cindex));
 833
 834        /* Double as long as the resulting node has a number of
 835         * nonempty nodes that are above the threshold.
 836         */
 837        while (should_inflate(tp, tn) && max_work) {
 838                tp = inflate(t, tn);
 839                if (!tp) {
 840#ifdef CONFIG_IP_FIB_TRIE_STATS
 841                        this_cpu_inc(stats->resize_node_skipped);
 842#endif
 843                        break;
 844                }
 845
 846                max_work--;
 847                tn = get_child(tp, cindex);
 848        }
 849
 850        /* update parent in case inflate failed */
 851        tp = node_parent(tn);
 852
 853        /* Return if at least one inflate is run */
 854        if (max_work != MAX_WORK)
 855                return tp;
 856
 857        /* Halve as long as the number of empty children in this
 858         * node is above threshold.
 859         */
 860        while (should_halve(tp, tn) && max_work) {
 861                tp = halve(t, tn);
 862                if (!tp) {
 863#ifdef CONFIG_IP_FIB_TRIE_STATS
 864                        this_cpu_inc(stats->resize_node_skipped);
 865#endif
 866                        break;
 867                }
 868
 869                max_work--;
 870                tn = get_child(tp, cindex);
 871        }
 872
 873        /* Only one child remains */
 874        if (should_collapse(tn))
 875                return collapse(t, tn);
 876
 877        /* update parent in case halve failed */
 878        tp = node_parent(tn);
 879
 880        /* Return if at least one deflate was run */
 881        if (max_work != MAX_WORK)
 882                return tp;
 883
 884        /* push the suffix length to the parent node */
 885        if (tn->slen > tn->pos) {
 886                unsigned char slen = update_suffix(tn);
 887
 888                if (slen > tp->slen)
 889                        tp->slen = slen;
 890        }
 891
 892        return tp;
 893}
 894
 895static void leaf_pull_suffix(struct key_vector *tp, struct key_vector *l)
 896{
 897        while ((tp->slen > tp->pos) && (tp->slen > l->slen)) {
 898                if (update_suffix(tp) > l->slen)
 899                        break;
 900                tp = node_parent(tp);
 901        }
 902}
 903
 904static void leaf_push_suffix(struct key_vector *tn, struct key_vector *l)
 905{
 906        /* if this is a new leaf then tn will be NULL and we can sort
 907         * out parent suffix lengths as a part of trie_rebalance
 908         */
 909        while (tn->slen < l->slen) {
 910                tn->slen = l->slen;
 911                tn = node_parent(tn);
 912        }
 913}
 914
 915/* rcu_read_lock needs to be hold by caller from readside */
 916static struct key_vector *fib_find_node(struct trie *t,
 917                                        struct key_vector **tp, u32 key)
 918{
 919        struct key_vector *pn, *n = t->kv;
 920        unsigned long index = 0;
 921
 922        do {
 923                pn = n;
 924                n = get_child_rcu(n, index);
 925
 926                if (!n)
 927                        break;
 928
 929                index = get_cindex(key, n);
 930
 931                /* This bit of code is a bit tricky but it combines multiple
 932                 * checks into a single check.  The prefix consists of the
 933                 * prefix plus zeros for the bits in the cindex. The index
 934                 * is the difference between the key and this value.  From
 935                 * this we can actually derive several pieces of data.
 936                 *   if (index >= (1ul << bits))
 937                 *     we have a mismatch in skip bits and failed
 938                 *   else
 939                 *     we know the value is cindex
 940                 *
 941                 * This check is safe even if bits == KEYLENGTH due to the
 942                 * fact that we can only allocate a node with 32 bits if a
 943                 * long is greater than 32 bits.
 944                 */
 945                if (index >= (1ul << n->bits)) {
 946                        n = NULL;
 947                        break;
 948                }
 949
 950                /* keep searching until we find a perfect match leaf or NULL */
 951        } while (IS_TNODE(n));
 952
 953        *tp = pn;
 954
 955        return n;
 956}
 957
 958/* Return the first fib alias matching TOS with
 959 * priority less than or equal to PRIO.
 960 */
 961static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
 962                                        u8 tos, u32 prio, u32 tb_id)
 963{
 964        struct fib_alias *fa;
 965
 966        if (!fah)
 967                return NULL;
 968
 969        hlist_for_each_entry(fa, fah, fa_list) {
 970                if (fa->fa_slen < slen)
 971                        continue;
 972                if (fa->fa_slen != slen)
 973                        break;
 974                if (fa->tb_id > tb_id)
 975                        continue;
 976                if (fa->tb_id != tb_id)
 977                        break;
 978                if (fa->fa_tos > tos)
 979                        continue;
 980                if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
 981                        return fa;
 982        }
 983
 984        return NULL;
 985}
 986
 987static void trie_rebalance(struct trie *t, struct key_vector *tn)
 988{
 989        while (!IS_TRIE(tn))
 990                tn = resize(t, tn);
 991}
 992
 993static int fib_insert_node(struct trie *t, struct key_vector *tp,
 994                           struct fib_alias *new, t_key key)
 995{
 996        struct key_vector *n, *l;
 997
 998        l = leaf_new(key, new);
 999        if (!l)
1000                goto noleaf;
1001
1002        /* retrieve child from parent node */
1003        n = get_child(tp, get_index(key, tp));
1004
1005        /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1006         *
1007         *  Add a new tnode here
1008         *  first tnode need some special handling
1009         *  leaves us in position for handling as case 3
1010         */
1011        if (n) {
1012                struct key_vector *tn;
1013
1014                tn = tnode_new(key, __fls(key ^ n->key), 1);
1015                if (!tn)
1016                        goto notnode;
1017
1018                /* initialize routes out of node */
1019                NODE_INIT_PARENT(tn, tp);
1020                put_child(tn, get_index(key, tn) ^ 1, n);
1021
1022                /* start adding routes into the node */
1023                put_child_root(tp, key, tn);
1024                node_set_parent(n, tn);
1025
1026                /* parent now has a NULL spot where the leaf can go */
1027                tp = tn;
1028        }
1029
1030        /* Case 3: n is NULL, and will just insert a new leaf */
1031        NODE_INIT_PARENT(l, tp);
1032        put_child_root(tp, key, l);
1033        trie_rebalance(t, tp);
1034
1035        return 0;
1036notnode:
1037        node_free(l);
1038noleaf:
1039        return -ENOMEM;
1040}
1041
1042static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1043                            struct key_vector *l, struct fib_alias *new,
1044                            struct fib_alias *fa, t_key key)
1045{
1046        if (!l)
1047                return fib_insert_node(t, tp, new, key);
1048
1049        if (fa) {
1050                hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1051        } else {
1052                struct fib_alias *last;
1053
1054                hlist_for_each_entry(last, &l->leaf, fa_list) {
1055                        if (new->fa_slen < last->fa_slen)
1056                                break;
1057                        if ((new->fa_slen == last->fa_slen) &&
1058                            (new->tb_id > last->tb_id))
1059                                break;
1060                        fa = last;
1061                }
1062
1063                if (fa)
1064                        hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1065                else
1066                        hlist_add_head_rcu(&new->fa_list, &l->leaf);
1067        }
1068
1069        /* if we added to the tail node then we need to update slen */
1070        if (l->slen < new->fa_slen) {
1071                l->slen = new->fa_slen;
1072                leaf_push_suffix(tp, l);
1073        }
1074
1075        return 0;
1076}
1077
1078/* Caller must hold RTNL. */
1079int fib_table_insert(struct fib_table *tb, struct fib_config *cfg)
1080{
1081        struct trie *t = (struct trie *)tb->tb_data;
1082        struct fib_alias *fa, *new_fa;
1083        struct key_vector *l, *tp;
1084        unsigned int nlflags = 0;
1085        struct fib_info *fi;
1086        u8 plen = cfg->fc_dst_len;
1087        u8 slen = KEYLENGTH - plen;
1088        u8 tos = cfg->fc_tos;
1089        u32 key;
1090        int err;
1091
1092        if (plen > KEYLENGTH)
1093                return -EINVAL;
1094
1095        key = ntohl(cfg->fc_dst);
1096
1097        pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1098
1099        if ((plen < KEYLENGTH) && (key << plen))
1100                return -EINVAL;
1101
1102        fi = fib_create_info(cfg);
1103        if (IS_ERR(fi)) {
1104                err = PTR_ERR(fi);
1105                goto err;
1106        }
1107
1108        l = fib_find_node(t, &tp, key);
1109        fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1110                                tb->tb_id) : NULL;
1111
1112        /* Now fa, if non-NULL, points to the first fib alias
1113         * with the same keys [prefix,tos,priority], if such key already
1114         * exists or to the node before which we will insert new one.
1115         *
1116         * If fa is NULL, we will need to allocate a new one and
1117         * insert to the tail of the section matching the suffix length
1118         * of the new alias.
1119         */
1120
1121        if (fa && fa->fa_tos == tos &&
1122            fa->fa_info->fib_priority == fi->fib_priority) {
1123                struct fib_alias *fa_first, *fa_match;
1124
1125                err = -EEXIST;
1126                if (cfg->fc_nlflags & NLM_F_EXCL)
1127                        goto out;
1128
1129                /* We have 2 goals:
1130                 * 1. Find exact match for type, scope, fib_info to avoid
1131                 * duplicate routes
1132                 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1133                 */
1134                fa_match = NULL;
1135                fa_first = fa;
1136                hlist_for_each_entry_from(fa, fa_list) {
1137                        if ((fa->fa_slen != slen) ||
1138                            (fa->tb_id != tb->tb_id) ||
1139                            (fa->fa_tos != tos))
1140                                break;
1141                        if (fa->fa_info->fib_priority != fi->fib_priority)
1142                                break;
1143                        if (fa->fa_type == cfg->fc_type &&
1144                            fa->fa_info == fi) {
1145                                fa_match = fa;
1146                                break;
1147                        }
1148                }
1149
1150                if (cfg->fc_nlflags & NLM_F_REPLACE) {
1151                        struct fib_info *fi_drop;
1152                        u8 state;
1153
1154                        fa = fa_first;
1155                        if (fa_match) {
1156                                if (fa == fa_match)
1157                                        err = 0;
1158                                goto out;
1159                        }
1160                        err = -ENOBUFS;
1161                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1162                        if (!new_fa)
1163                                goto out;
1164
1165                        fi_drop = fa->fa_info;
1166                        new_fa->fa_tos = fa->fa_tos;
1167                        new_fa->fa_info = fi;
1168                        new_fa->fa_type = cfg->fc_type;
1169                        state = fa->fa_state;
1170                        new_fa->fa_state = state & ~FA_S_ACCESSED;
1171                        new_fa->fa_slen = fa->fa_slen;
1172                        new_fa->tb_id = tb->tb_id;
1173                        new_fa->fa_default = -1;
1174
1175                        err = switchdev_fib_ipv4_add(key, plen, fi,
1176                                                     new_fa->fa_tos,
1177                                                     cfg->fc_type,
1178                                                     cfg->fc_nlflags,
1179                                                     tb->tb_id);
1180                        if (err) {
1181                                switchdev_fib_ipv4_abort(fi);
1182                                kmem_cache_free(fn_alias_kmem, new_fa);
1183                                goto out;
1184                        }
1185
1186                        hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1187
1188                        alias_free_mem_rcu(fa);
1189
1190                        fib_release_info(fi_drop);
1191                        if (state & FA_S_ACCESSED)
1192                                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1193                        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1194                                tb->tb_id, &cfg->fc_nlinfo, NLM_F_REPLACE);
1195
1196                        goto succeeded;
1197                }
1198                /* Error if we find a perfect match which
1199                 * uses the same scope, type, and nexthop
1200                 * information.
1201                 */
1202                if (fa_match)
1203                        goto out;
1204
1205                if (cfg->fc_nlflags & NLM_F_APPEND)
1206                        nlflags = NLM_F_APPEND;
1207                else
1208                        fa = fa_first;
1209        }
1210        err = -ENOENT;
1211        if (!(cfg->fc_nlflags & NLM_F_CREATE))
1212                goto out;
1213
1214        err = -ENOBUFS;
1215        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1216        if (!new_fa)
1217                goto out;
1218
1219        new_fa->fa_info = fi;
1220        new_fa->fa_tos = tos;
1221        new_fa->fa_type = cfg->fc_type;
1222        new_fa->fa_state = 0;
1223        new_fa->fa_slen = slen;
1224        new_fa->tb_id = tb->tb_id;
1225        new_fa->fa_default = -1;
1226
1227        /* (Optionally) offload fib entry to switch hardware. */
1228        err = switchdev_fib_ipv4_add(key, plen, fi, tos, cfg->fc_type,
1229                                     cfg->fc_nlflags, tb->tb_id);
1230        if (err) {
1231                switchdev_fib_ipv4_abort(fi);
1232                goto out_free_new_fa;
1233        }
1234
1235        /* Insert new entry to the list. */
1236        err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1237        if (err)
1238                goto out_sw_fib_del;
1239
1240        if (!plen)
1241                tb->tb_num_default++;
1242
1243        rt_cache_flush(cfg->fc_nlinfo.nl_net);
1244        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1245                  &cfg->fc_nlinfo, nlflags);
1246succeeded:
1247        return 0;
1248
1249out_sw_fib_del:
1250        switchdev_fib_ipv4_del(key, plen, fi, tos, cfg->fc_type, tb->tb_id);
1251out_free_new_fa:
1252        kmem_cache_free(fn_alias_kmem, new_fa);
1253out:
1254        fib_release_info(fi);
1255err:
1256        return err;
1257}
1258
1259static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1260{
1261        t_key prefix = n->key;
1262
1263        return (key ^ prefix) & (prefix | -prefix);
1264}
1265
1266/* should be called with rcu_read_lock */
1267int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1268                     struct fib_result *res, int fib_flags)
1269{
1270        struct trie *t = (struct trie *) tb->tb_data;
1271#ifdef CONFIG_IP_FIB_TRIE_STATS
1272        struct trie_use_stats __percpu *stats = t->stats;
1273#endif
1274        const t_key key = ntohl(flp->daddr);
1275        struct key_vector *n, *pn;
1276        struct fib_alias *fa;
1277        unsigned long index;
1278        t_key cindex;
1279
1280        trace_fib_table_lookup(tb->tb_id, flp);
1281
1282        pn = t->kv;
1283        cindex = 0;
1284
1285        n = get_child_rcu(pn, cindex);
1286        if (!n)
1287                return -EAGAIN;
1288
1289#ifdef CONFIG_IP_FIB_TRIE_STATS
1290        this_cpu_inc(stats->gets);
1291#endif
1292
1293        /* Step 1: Travel to the longest prefix match in the trie */
1294        for (;;) {
1295                index = get_cindex(key, n);
1296
1297                /* This bit of code is a bit tricky but it combines multiple
1298                 * checks into a single check.  The prefix consists of the
1299                 * prefix plus zeros for the "bits" in the prefix. The index
1300                 * is the difference between the key and this value.  From
1301                 * this we can actually derive several pieces of data.
1302                 *   if (index >= (1ul << bits))
1303                 *     we have a mismatch in skip bits and failed
1304                 *   else
1305                 *     we know the value is cindex
1306                 *
1307                 * This check is safe even if bits == KEYLENGTH due to the
1308                 * fact that we can only allocate a node with 32 bits if a
1309                 * long is greater than 32 bits.
1310                 */
1311                if (index >= (1ul << n->bits))
1312                        break;
1313
1314                /* we have found a leaf. Prefixes have already been compared */
1315                if (IS_LEAF(n))
1316                        goto found;
1317
1318                /* only record pn and cindex if we are going to be chopping
1319                 * bits later.  Otherwise we are just wasting cycles.
1320                 */
1321                if (n->slen > n->pos) {
1322                        pn = n;
1323                        cindex = index;
1324                }
1325
1326                n = get_child_rcu(n, index);
1327                if (unlikely(!n))
1328                        goto backtrace;
1329        }
1330
1331        /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1332        for (;;) {
1333                /* record the pointer where our next node pointer is stored */
1334                struct key_vector __rcu **cptr = n->tnode;
1335
1336                /* This test verifies that none of the bits that differ
1337                 * between the key and the prefix exist in the region of
1338                 * the lsb and higher in the prefix.
1339                 */
1340                if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1341                        goto backtrace;
1342
1343                /* exit out and process leaf */
1344                if (unlikely(IS_LEAF(n)))
1345                        break;
1346
1347                /* Don't bother recording parent info.  Since we are in
1348                 * prefix match mode we will have to come back to wherever
1349                 * we started this traversal anyway
1350                 */
1351
1352                while ((n = rcu_dereference(*cptr)) == NULL) {
1353backtrace:
1354#ifdef CONFIG_IP_FIB_TRIE_STATS
1355                        if (!n)
1356                                this_cpu_inc(stats->null_node_hit);
1357#endif
1358                        /* If we are at cindex 0 there are no more bits for
1359                         * us to strip at this level so we must ascend back
1360                         * up one level to see if there are any more bits to
1361                         * be stripped there.
1362                         */
1363                        while (!cindex) {
1364                                t_key pkey = pn->key;
1365
1366                                /* If we don't have a parent then there is
1367                                 * nothing for us to do as we do not have any
1368                                 * further nodes to parse.
1369                                 */
1370                                if (IS_TRIE(pn))
1371                                        return -EAGAIN;
1372#ifdef CONFIG_IP_FIB_TRIE_STATS
1373                                this_cpu_inc(stats->backtrack);
1374#endif
1375                                /* Get Child's index */
1376                                pn = node_parent_rcu(pn);
1377                                cindex = get_index(pkey, pn);
1378                        }
1379
1380                        /* strip the least significant bit from the cindex */
1381                        cindex &= cindex - 1;
1382
1383                        /* grab pointer for next child node */
1384                        cptr = &pn->tnode[cindex];
1385                }
1386        }
1387
1388found:
1389        /* this line carries forward the xor from earlier in the function */
1390        index = key ^ n->key;
1391
1392        /* Step 3: Process the leaf, if that fails fall back to backtracing */
1393        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1394                struct fib_info *fi = fa->fa_info;
1395                int nhsel, err;
1396
1397                if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1398                        if (index >= (1ul << fa->fa_slen))
1399                                continue;
1400                }
1401                if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1402                        continue;
1403                if (fi->fib_dead)
1404                        continue;
1405                if (fa->fa_info->fib_scope < flp->flowi4_scope)
1406                        continue;
1407                fib_alias_accessed(fa);
1408                err = fib_props[fa->fa_type].error;
1409                if (unlikely(err < 0)) {
1410#ifdef CONFIG_IP_FIB_TRIE_STATS
1411                        this_cpu_inc(stats->semantic_match_passed);
1412#endif
1413                        return err;
1414                }
1415                if (fi->fib_flags & RTNH_F_DEAD)
1416                        continue;
1417                for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1418                        const struct fib_nh *nh = &fi->fib_nh[nhsel];
1419                        struct in_device *in_dev = __in_dev_get_rcu(nh->nh_dev);
1420
1421                        if (nh->nh_flags & RTNH_F_DEAD)
1422                                continue;
1423                        if (in_dev &&
1424                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1425                            nh->nh_flags & RTNH_F_LINKDOWN &&
1426                            !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1427                                continue;
1428                        if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1429                                if (flp->flowi4_oif &&
1430                                    flp->flowi4_oif != nh->nh_oif)
1431                                        continue;
1432                        }
1433
1434                        if (!(fib_flags & FIB_LOOKUP_NOREF))
1435                                atomic_inc(&fi->fib_clntref);
1436
1437                        res->prefixlen = KEYLENGTH - fa->fa_slen;
1438                        res->nh_sel = nhsel;
1439                        res->type = fa->fa_type;
1440                        res->scope = fi->fib_scope;
1441                        res->fi = fi;
1442                        res->table = tb;
1443                        res->fa_head = &n->leaf;
1444#ifdef CONFIG_IP_FIB_TRIE_STATS
1445                        this_cpu_inc(stats->semantic_match_passed);
1446#endif
1447                        trace_fib_table_lookup_nh(nh);
1448
1449                        return err;
1450                }
1451        }
1452#ifdef CONFIG_IP_FIB_TRIE_STATS
1453        this_cpu_inc(stats->semantic_match_miss);
1454#endif
1455        goto backtrace;
1456}
1457EXPORT_SYMBOL_GPL(fib_table_lookup);
1458
1459static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1460                             struct key_vector *l, struct fib_alias *old)
1461{
1462        /* record the location of the previous list_info entry */
1463        struct hlist_node **pprev = old->fa_list.pprev;
1464        struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1465
1466        /* remove the fib_alias from the list */
1467        hlist_del_rcu(&old->fa_list);
1468
1469        /* if we emptied the list this leaf will be freed and we can sort
1470         * out parent suffix lengths as a part of trie_rebalance
1471         */
1472        if (hlist_empty(&l->leaf)) {
1473                put_child_root(tp, l->key, NULL);
1474                node_free(l);
1475                trie_rebalance(t, tp);
1476                return;
1477        }
1478
1479        /* only access fa if it is pointing at the last valid hlist_node */
1480        if (*pprev)
1481                return;
1482
1483        /* update the trie with the latest suffix length */
1484        l->slen = fa->fa_slen;
1485        leaf_pull_suffix(tp, l);
1486}
1487
1488/* Caller must hold RTNL. */
1489int fib_table_delete(struct fib_table *tb, struct fib_config *cfg)
1490{
1491        struct trie *t = (struct trie *) tb->tb_data;
1492        struct fib_alias *fa, *fa_to_delete;
1493        struct key_vector *l, *tp;
1494        u8 plen = cfg->fc_dst_len;
1495        u8 slen = KEYLENGTH - plen;
1496        u8 tos = cfg->fc_tos;
1497        u32 key;
1498
1499        if (plen > KEYLENGTH)
1500                return -EINVAL;
1501
1502        key = ntohl(cfg->fc_dst);
1503
1504        if ((plen < KEYLENGTH) && (key << plen))
1505                return -EINVAL;
1506
1507        l = fib_find_node(t, &tp, key);
1508        if (!l)
1509                return -ESRCH;
1510
1511        fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1512        if (!fa)
1513                return -ESRCH;
1514
1515        pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1516
1517        fa_to_delete = NULL;
1518        hlist_for_each_entry_from(fa, fa_list) {
1519                struct fib_info *fi = fa->fa_info;
1520
1521                if ((fa->fa_slen != slen) ||
1522                    (fa->tb_id != tb->tb_id) ||
1523                    (fa->fa_tos != tos))
1524                        break;
1525
1526                if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1527                    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1528                     fa->fa_info->fib_scope == cfg->fc_scope) &&
1529                    (!cfg->fc_prefsrc ||
1530                     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1531                    (!cfg->fc_protocol ||
1532                     fi->fib_protocol == cfg->fc_protocol) &&
1533                    fib_nh_match(cfg, fi) == 0) {
1534                        fa_to_delete = fa;
1535                        break;
1536                }
1537        }
1538
1539        if (!fa_to_delete)
1540                return -ESRCH;
1541
1542        switchdev_fib_ipv4_del(key, plen, fa_to_delete->fa_info, tos,
1543                               cfg->fc_type, tb->tb_id);
1544
1545        rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1546                  &cfg->fc_nlinfo, 0);
1547
1548        if (!plen)
1549                tb->tb_num_default--;
1550
1551        fib_remove_alias(t, tp, l, fa_to_delete);
1552
1553        if (fa_to_delete->fa_state & FA_S_ACCESSED)
1554                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1555
1556        fib_release_info(fa_to_delete->fa_info);
1557        alias_free_mem_rcu(fa_to_delete);
1558        return 0;
1559}
1560
1561/* Scan for the next leaf starting at the provided key value */
1562static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1563{
1564        struct key_vector *pn, *n = *tn;
1565        unsigned long cindex;
1566
1567        /* this loop is meant to try and find the key in the trie */
1568        do {
1569                /* record parent and next child index */
1570                pn = n;
1571                cindex = (key > pn->key) ? get_index(key, pn) : 0;
1572
1573                if (cindex >> pn->bits)
1574                        break;
1575
1576                /* descend into the next child */
1577                n = get_child_rcu(pn, cindex++);
1578                if (!n)
1579                        break;
1580
1581                /* guarantee forward progress on the keys */
1582                if (IS_LEAF(n) && (n->key >= key))
1583                        goto found;
1584        } while (IS_TNODE(n));
1585
1586        /* this loop will search for the next leaf with a greater key */
1587        while (!IS_TRIE(pn)) {
1588                /* if we exhausted the parent node we will need to climb */
1589                if (cindex >= (1ul << pn->bits)) {
1590                        t_key pkey = pn->key;
1591
1592                        pn = node_parent_rcu(pn);
1593                        cindex = get_index(pkey, pn) + 1;
1594                        continue;
1595                }
1596
1597                /* grab the next available node */
1598                n = get_child_rcu(pn, cindex++);
1599                if (!n)
1600                        continue;
1601
1602                /* no need to compare keys since we bumped the index */
1603                if (IS_LEAF(n))
1604                        goto found;
1605
1606                /* Rescan start scanning in new node */
1607                pn = n;
1608                cindex = 0;
1609        }
1610
1611        *tn = pn;
1612        return NULL; /* Root of trie */
1613found:
1614        /* if we are at the limit for keys just return NULL for the tnode */
1615        *tn = pn;
1616        return n;
1617}
1618
1619static void fib_trie_free(struct fib_table *tb)
1620{
1621        struct trie *t = (struct trie *)tb->tb_data;
1622        struct key_vector *pn = t->kv;
1623        unsigned long cindex = 1;
1624        struct hlist_node *tmp;
1625        struct fib_alias *fa;
1626
1627        /* walk trie in reverse order and free everything */
1628        for (;;) {
1629                struct key_vector *n;
1630
1631                if (!(cindex--)) {
1632                        t_key pkey = pn->key;
1633
1634                        if (IS_TRIE(pn))
1635                                break;
1636
1637                        n = pn;
1638                        pn = node_parent(pn);
1639
1640                        /* drop emptied tnode */
1641                        put_child_root(pn, n->key, NULL);
1642                        node_free(n);
1643
1644                        cindex = get_index(pkey, pn);
1645
1646                        continue;
1647                }
1648
1649                /* grab the next available node */
1650                n = get_child(pn, cindex);
1651                if (!n)
1652                        continue;
1653
1654                if (IS_TNODE(n)) {
1655                        /* record pn and cindex for leaf walking */
1656                        pn = n;
1657                        cindex = 1ul << n->bits;
1658
1659                        continue;
1660                }
1661
1662                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1663                        hlist_del_rcu(&fa->fa_list);
1664                        alias_free_mem_rcu(fa);
1665                }
1666
1667                put_child_root(pn, n->key, NULL);
1668                node_free(n);
1669        }
1670
1671#ifdef CONFIG_IP_FIB_TRIE_STATS
1672        free_percpu(t->stats);
1673#endif
1674        kfree(tb);
1675}
1676
1677struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1678{
1679        struct trie *ot = (struct trie *)oldtb->tb_data;
1680        struct key_vector *l, *tp = ot->kv;
1681        struct fib_table *local_tb;
1682        struct fib_alias *fa;
1683        struct trie *lt;
1684        t_key key = 0;
1685
1686        if (oldtb->tb_data == oldtb->__data)
1687                return oldtb;
1688
1689        local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1690        if (!local_tb)
1691                return NULL;
1692
1693        lt = (struct trie *)local_tb->tb_data;
1694
1695        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1696                struct key_vector *local_l = NULL, *local_tp;
1697
1698                hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1699                        struct fib_alias *new_fa;
1700
1701                        if (local_tb->tb_id != fa->tb_id)
1702                                continue;
1703
1704                        /* clone fa for new local table */
1705                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1706                        if (!new_fa)
1707                                goto out;
1708
1709                        memcpy(new_fa, fa, sizeof(*fa));
1710
1711                        /* insert clone into table */
1712                        if (!local_l)
1713                                local_l = fib_find_node(lt, &local_tp, l->key);
1714
1715                        if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1716                                             NULL, l->key))
1717                                goto out;
1718                }
1719
1720                /* stop loop if key wrapped back to 0 */
1721                key = l->key + 1;
1722                if (key < l->key)
1723                        break;
1724        }
1725
1726        return local_tb;
1727out:
1728        fib_trie_free(local_tb);
1729
1730        return NULL;
1731}
1732
1733/* Caller must hold RTNL */
1734void fib_table_flush_external(struct fib_table *tb)
1735{
1736        struct trie *t = (struct trie *)tb->tb_data;
1737        struct key_vector *pn = t->kv;
1738        unsigned long cindex = 1;
1739        struct hlist_node *tmp;
1740        struct fib_alias *fa;
1741
1742        /* walk trie in reverse order */
1743        for (;;) {
1744                unsigned char slen = 0;
1745                struct key_vector *n;
1746
1747                if (!(cindex--)) {
1748                        t_key pkey = pn->key;
1749
1750                        /* cannot resize the trie vector */
1751                        if (IS_TRIE(pn))
1752                                break;
1753
1754                        /* resize completed node */
1755                        pn = resize(t, pn);
1756                        cindex = get_index(pkey, pn);
1757
1758                        continue;
1759                }
1760
1761                /* grab the next available node */
1762                n = get_child(pn, cindex);
1763                if (!n)
1764                        continue;
1765
1766                if (IS_TNODE(n)) {
1767                        /* record pn and cindex for leaf walking */
1768                        pn = n;
1769                        cindex = 1ul << n->bits;
1770
1771                        continue;
1772                }
1773
1774                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1775                        struct fib_info *fi = fa->fa_info;
1776
1777                        /* if alias was cloned to local then we just
1778                         * need to remove the local copy from main
1779                         */
1780                        if (tb->tb_id != fa->tb_id) {
1781                                hlist_del_rcu(&fa->fa_list);
1782                                alias_free_mem_rcu(fa);
1783                                continue;
1784                        }
1785
1786                        /* record local slen */
1787                        slen = fa->fa_slen;
1788
1789                        if (!fi || !(fi->fib_flags & RTNH_F_OFFLOAD))
1790                                continue;
1791
1792                        switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1793                                               fi, fa->fa_tos, fa->fa_type,
1794                                               tb->tb_id);
1795                }
1796
1797                /* update leaf slen */
1798                n->slen = slen;
1799
1800                if (hlist_empty(&n->leaf)) {
1801                        put_child_root(pn, n->key, NULL);
1802                        node_free(n);
1803                }
1804        }
1805}
1806
1807/* Caller must hold RTNL. */
1808int fib_table_flush(struct fib_table *tb)
1809{
1810        struct trie *t = (struct trie *)tb->tb_data;
1811        struct key_vector *pn = t->kv;
1812        unsigned long cindex = 1;
1813        struct hlist_node *tmp;
1814        struct fib_alias *fa;
1815        int found = 0;
1816
1817        /* walk trie in reverse order */
1818        for (;;) {
1819                unsigned char slen = 0;
1820                struct key_vector *n;
1821
1822                if (!(cindex--)) {
1823                        t_key pkey = pn->key;
1824
1825                        /* cannot resize the trie vector */
1826                        if (IS_TRIE(pn))
1827                                break;
1828
1829                        /* resize completed node */
1830                        pn = resize(t, pn);
1831                        cindex = get_index(pkey, pn);
1832
1833                        continue;
1834                }
1835
1836                /* grab the next available node */
1837                n = get_child(pn, cindex);
1838                if (!n)
1839                        continue;
1840
1841                if (IS_TNODE(n)) {
1842                        /* record pn and cindex for leaf walking */
1843                        pn = n;
1844                        cindex = 1ul << n->bits;
1845
1846                        continue;
1847                }
1848
1849                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1850                        struct fib_info *fi = fa->fa_info;
1851
1852                        if (!fi || !(fi->fib_flags & RTNH_F_DEAD)) {
1853                                slen = fa->fa_slen;
1854                                continue;
1855                        }
1856
1857                        switchdev_fib_ipv4_del(n->key, KEYLENGTH - fa->fa_slen,
1858                                               fi, fa->fa_tos, fa->fa_type,
1859                                               tb->tb_id);
1860                        hlist_del_rcu(&fa->fa_list);
1861                        fib_release_info(fa->fa_info);
1862                        alias_free_mem_rcu(fa);
1863                        found++;
1864                }
1865
1866                /* update leaf slen */
1867                n->slen = slen;
1868
1869                if (hlist_empty(&n->leaf)) {
1870                        put_child_root(pn, n->key, NULL);
1871                        node_free(n);
1872                }
1873        }
1874
1875        pr_debug("trie_flush found=%d\n", found);
1876        return found;
1877}
1878
1879static void __trie_free_rcu(struct rcu_head *head)
1880{
1881        struct fib_table *tb = container_of(head, struct fib_table, rcu);
1882#ifdef CONFIG_IP_FIB_TRIE_STATS
1883        struct trie *t = (struct trie *)tb->tb_data;
1884
1885        if (tb->tb_data == tb->__data)
1886                free_percpu(t->stats);
1887#endif /* CONFIG_IP_FIB_TRIE_STATS */
1888        kfree(tb);
1889}
1890
1891void fib_free_table(struct fib_table *tb)
1892{
1893        call_rcu(&tb->rcu, __trie_free_rcu);
1894}
1895
1896static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
1897                             struct sk_buff *skb, struct netlink_callback *cb)
1898{
1899        __be32 xkey = htonl(l->key);
1900        struct fib_alias *fa;
1901        int i, s_i;
1902
1903        s_i = cb->args[4];
1904        i = 0;
1905
1906        /* rcu_read_lock is hold by caller */
1907        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1908                if (i < s_i) {
1909                        i++;
1910                        continue;
1911                }
1912
1913                if (tb->tb_id != fa->tb_id) {
1914                        i++;
1915                        continue;
1916                }
1917
1918                if (fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
1919                                  cb->nlh->nlmsg_seq,
1920                                  RTM_NEWROUTE,
1921                                  tb->tb_id,
1922                                  fa->fa_type,
1923                                  xkey,
1924                                  KEYLENGTH - fa->fa_slen,
1925                                  fa->fa_tos,
1926                                  fa->fa_info, NLM_F_MULTI) < 0) {
1927                        cb->args[4] = i;
1928                        return -1;
1929                }
1930                i++;
1931        }
1932
1933        cb->args[4] = i;
1934        return skb->len;
1935}
1936
1937/* rcu_read_lock needs to be hold by caller from readside */
1938int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
1939                   struct netlink_callback *cb)
1940{
1941        struct trie *t = (struct trie *)tb->tb_data;
1942        struct key_vector *l, *tp = t->kv;
1943        /* Dump starting at last key.
1944         * Note: 0.0.0.0/0 (ie default) is first key.
1945         */
1946        int count = cb->args[2];
1947        t_key key = cb->args[3];
1948
1949        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1950                if (fn_trie_dump_leaf(l, tb, skb, cb) < 0) {
1951                        cb->args[3] = key;
1952                        cb->args[2] = count;
1953                        return -1;
1954                }
1955
1956                ++count;
1957                key = l->key + 1;
1958
1959                memset(&cb->args[4], 0,
1960                       sizeof(cb->args) - 4*sizeof(cb->args[0]));
1961
1962                /* stop loop if key wrapped back to 0 */
1963                if (key < l->key)
1964                        break;
1965        }
1966
1967        cb->args[3] = key;
1968        cb->args[2] = count;
1969
1970        return skb->len;
1971}
1972
1973void __init fib_trie_init(void)
1974{
1975        fn_alias_kmem = kmem_cache_create("ip_fib_alias",
1976                                          sizeof(struct fib_alias),
1977                                          0, SLAB_PANIC, NULL);
1978
1979        trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
1980                                           LEAF_SIZE,
1981                                           0, SLAB_PANIC, NULL);
1982}
1983
1984struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
1985{
1986        struct fib_table *tb;
1987        struct trie *t;
1988        size_t sz = sizeof(*tb);
1989
1990        if (!alias)
1991                sz += sizeof(struct trie);
1992
1993        tb = kzalloc(sz, GFP_KERNEL);
1994        if (!tb)
1995                return NULL;
1996
1997        tb->tb_id = id;
1998        tb->tb_num_default = 0;
1999        tb->tb_data = (alias ? alias->__data : tb->__data);
2000
2001        if (alias)
2002                return tb;
2003
2004        t = (struct trie *) tb->tb_data;
2005        t->kv[0].pos = KEYLENGTH;
2006        t->kv[0].slen = KEYLENGTH;
2007#ifdef CONFIG_IP_FIB_TRIE_STATS
2008        t->stats = alloc_percpu(struct trie_use_stats);
2009        if (!t->stats) {
2010                kfree(tb);
2011                tb = NULL;
2012        }
2013#endif
2014
2015        return tb;
2016}
2017
2018#ifdef CONFIG_PROC_FS
2019/* Depth first Trie walk iterator */
2020struct fib_trie_iter {
2021        struct seq_net_private p;
2022        struct fib_table *tb;
2023        struct key_vector *tnode;
2024        unsigned int index;
2025        unsigned int depth;
2026};
2027
2028static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2029{
2030        unsigned long cindex = iter->index;
2031        struct key_vector *pn = iter->tnode;
2032        t_key pkey;
2033
2034        pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2035                 iter->tnode, iter->index, iter->depth);
2036
2037        while (!IS_TRIE(pn)) {
2038                while (cindex < child_length(pn)) {
2039                        struct key_vector *n = get_child_rcu(pn, cindex++);
2040
2041                        if (!n)
2042                                continue;
2043
2044                        if (IS_LEAF(n)) {
2045                                iter->tnode = pn;
2046                                iter->index = cindex;
2047                        } else {
2048                                /* push down one level */
2049                                iter->tnode = n;
2050                                iter->index = 0;
2051                                ++iter->depth;
2052                        }
2053
2054                        return n;
2055                }
2056
2057                /* Current node exhausted, pop back up */
2058                pkey = pn->key;
2059                pn = node_parent_rcu(pn);
2060                cindex = get_index(pkey, pn) + 1;
2061                --iter->depth;
2062        }
2063
2064        /* record root node so further searches know we are done */
2065        iter->tnode = pn;
2066        iter->index = 0;
2067
2068        return NULL;
2069}
2070
2071static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2072                                             struct trie *t)
2073{
2074        struct key_vector *n, *pn;
2075
2076        if (!t)
2077                return NULL;
2078
2079        pn = t->kv;
2080        n = rcu_dereference(pn->tnode[0]);
2081        if (!n)
2082                return NULL;
2083
2084        if (IS_TNODE(n)) {
2085                iter->tnode = n;
2086                iter->index = 0;
2087                iter->depth = 1;
2088        } else {
2089                iter->tnode = pn;
2090                iter->index = 0;
2091                iter->depth = 0;
2092        }
2093
2094        return n;
2095}
2096
2097static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2098{
2099        struct key_vector *n;
2100        struct fib_trie_iter iter;
2101
2102        memset(s, 0, sizeof(*s));
2103
2104        rcu_read_lock();
2105        for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2106                if (IS_LEAF(n)) {
2107                        struct fib_alias *fa;
2108
2109                        s->leaves++;
2110                        s->totdepth += iter.depth;
2111                        if (iter.depth > s->maxdepth)
2112                                s->maxdepth = iter.depth;
2113
2114                        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2115                                ++s->prefixes;
2116                } else {
2117                        s->tnodes++;
2118                        if (n->bits < MAX_STAT_DEPTH)
2119                                s->nodesizes[n->bits]++;
2120                        s->nullpointers += tn_info(n)->empty_children;
2121                }
2122        }
2123        rcu_read_unlock();
2124}
2125
2126/*
2127 *      This outputs /proc/net/fib_triestats
2128 */
2129static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2130{
2131        unsigned int i, max, pointers, bytes, avdepth;
2132
2133        if (stat->leaves)
2134                avdepth = stat->totdepth*100 / stat->leaves;
2135        else
2136                avdepth = 0;
2137
2138        seq_printf(seq, "\tAver depth:     %u.%02d\n",
2139                   avdepth / 100, avdepth % 100);
2140        seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2141
2142        seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2143        bytes = LEAF_SIZE * stat->leaves;
2144
2145        seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2146        bytes += sizeof(struct fib_alias) * stat->prefixes;
2147
2148        seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2149        bytes += TNODE_SIZE(0) * stat->tnodes;
2150
2151        max = MAX_STAT_DEPTH;
2152        while (max > 0 && stat->nodesizes[max-1] == 0)
2153                max--;
2154
2155        pointers = 0;
2156        for (i = 1; i < max; i++)
2157                if (stat->nodesizes[i] != 0) {
2158                        seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2159                        pointers += (1<<i) * stat->nodesizes[i];
2160                }
2161        seq_putc(seq, '\n');
2162        seq_printf(seq, "\tPointers: %u\n", pointers);
2163
2164        bytes += sizeof(struct key_vector *) * pointers;
2165        seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2166        seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2167}
2168
2169#ifdef CONFIG_IP_FIB_TRIE_STATS
2170static void trie_show_usage(struct seq_file *seq,
2171                            const struct trie_use_stats __percpu *stats)
2172{
2173        struct trie_use_stats s = { 0 };
2174        int cpu;
2175
2176        /* loop through all of the CPUs and gather up the stats */
2177        for_each_possible_cpu(cpu) {
2178                const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2179
2180                s.gets += pcpu->gets;
2181                s.backtrack += pcpu->backtrack;
2182                s.semantic_match_passed += pcpu->semantic_match_passed;
2183                s.semantic_match_miss += pcpu->semantic_match_miss;
2184                s.null_node_hit += pcpu->null_node_hit;
2185                s.resize_node_skipped += pcpu->resize_node_skipped;
2186        }
2187
2188        seq_printf(seq, "\nCounters:\n---------\n");
2189        seq_printf(seq, "gets = %u\n", s.gets);
2190        seq_printf(seq, "backtracks = %u\n", s.backtrack);
2191        seq_printf(seq, "semantic match passed = %u\n",
2192                   s.semantic_match_passed);
2193        seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2194        seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2195        seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2196}
2197#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2198
2199static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2200{
2201        if (tb->tb_id == RT_TABLE_LOCAL)
2202                seq_puts(seq, "Local:\n");
2203        else if (tb->tb_id == RT_TABLE_MAIN)
2204                seq_puts(seq, "Main:\n");
2205        else
2206                seq_printf(seq, "Id %d:\n", tb->tb_id);
2207}
2208
2209
2210static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2211{
2212        struct net *net = (struct net *)seq->private;
2213        unsigned int h;
2214
2215        seq_printf(seq,
2216                   "Basic info: size of leaf:"
2217                   " %Zd bytes, size of tnode: %Zd bytes.\n",
2218                   LEAF_SIZE, TNODE_SIZE(0));
2219
2220        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2221                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2222                struct fib_table *tb;
2223
2224                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2225                        struct trie *t = (struct trie *) tb->tb_data;
2226                        struct trie_stat stat;
2227
2228                        if (!t)
2229                                continue;
2230
2231                        fib_table_print(seq, tb);
2232
2233                        trie_collect_stats(t, &stat);
2234                        trie_show_stats(seq, &stat);
2235#ifdef CONFIG_IP_FIB_TRIE_STATS
2236                        trie_show_usage(seq, t->stats);
2237#endif
2238                }
2239        }
2240
2241        return 0;
2242}
2243
2244static int fib_triestat_seq_open(struct inode *inode, struct file *file)
2245{
2246        return single_open_net(inode, file, fib_triestat_seq_show);
2247}
2248
2249static const struct file_operations fib_triestat_fops = {
2250        .owner  = THIS_MODULE,
2251        .open   = fib_triestat_seq_open,
2252        .read   = seq_read,
2253        .llseek = seq_lseek,
2254        .release = single_release_net,
2255};
2256
2257static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2258{
2259        struct fib_trie_iter *iter = seq->private;
2260        struct net *net = seq_file_net(seq);
2261        loff_t idx = 0;
2262        unsigned int h;
2263
2264        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2265                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2266                struct fib_table *tb;
2267
2268                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2269                        struct key_vector *n;
2270
2271                        for (n = fib_trie_get_first(iter,
2272                                                    (struct trie *) tb->tb_data);
2273                             n; n = fib_trie_get_next(iter))
2274                                if (pos == idx++) {
2275                                        iter->tb = tb;
2276                                        return n;
2277                                }
2278                }
2279        }
2280
2281        return NULL;
2282}
2283
2284static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2285        __acquires(RCU)
2286{
2287        rcu_read_lock();
2288        return fib_trie_get_idx(seq, *pos);
2289}
2290
2291static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2292{
2293        struct fib_trie_iter *iter = seq->private;
2294        struct net *net = seq_file_net(seq);
2295        struct fib_table *tb = iter->tb;
2296        struct hlist_node *tb_node;
2297        unsigned int h;
2298        struct key_vector *n;
2299
2300        ++*pos;
2301        /* next node in same table */
2302        n = fib_trie_get_next(iter);
2303        if (n)
2304                return n;
2305
2306        /* walk rest of this hash chain */
2307        h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2308        while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2309                tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2310                n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2311                if (n)
2312                        goto found;
2313        }
2314
2315        /* new hash chain */
2316        while (++h < FIB_TABLE_HASHSZ) {
2317                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2318                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2319                        n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2320                        if (n)
2321                                goto found;
2322                }
2323        }
2324        return NULL;
2325
2326found:
2327        iter->tb = tb;
2328        return n;
2329}
2330
2331static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2332        __releases(RCU)
2333{
2334        rcu_read_unlock();
2335}
2336
2337static void seq_indent(struct seq_file *seq, int n)
2338{
2339        while (n-- > 0)
2340                seq_puts(seq, "   ");
2341}
2342
2343static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2344{
2345        switch (s) {
2346        case RT_SCOPE_UNIVERSE: return "universe";
2347        case RT_SCOPE_SITE:     return "site";
2348        case RT_SCOPE_LINK:     return "link";
2349        case RT_SCOPE_HOST:     return "host";
2350        case RT_SCOPE_NOWHERE:  return "nowhere";
2351        default:
2352                snprintf(buf, len, "scope=%d", s);
2353                return buf;
2354        }
2355}
2356
2357static const char *const rtn_type_names[__RTN_MAX] = {
2358        [RTN_UNSPEC] = "UNSPEC",
2359        [RTN_UNICAST] = "UNICAST",
2360        [RTN_LOCAL] = "LOCAL",
2361        [RTN_BROADCAST] = "BROADCAST",
2362        [RTN_ANYCAST] = "ANYCAST",
2363        [RTN_MULTICAST] = "MULTICAST",
2364        [RTN_BLACKHOLE] = "BLACKHOLE",
2365        [RTN_UNREACHABLE] = "UNREACHABLE",
2366        [RTN_PROHIBIT] = "PROHIBIT",
2367        [RTN_THROW] = "THROW",
2368        [RTN_NAT] = "NAT",
2369        [RTN_XRESOLVE] = "XRESOLVE",
2370};
2371
2372static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2373{
2374        if (t < __RTN_MAX && rtn_type_names[t])
2375                return rtn_type_names[t];
2376        snprintf(buf, len, "type %u", t);
2377        return buf;
2378}
2379
2380/* Pretty print the trie */
2381static int fib_trie_seq_show(struct seq_file *seq, void *v)
2382{
2383        const struct fib_trie_iter *iter = seq->private;
2384        struct key_vector *n = v;
2385
2386        if (IS_TRIE(node_parent_rcu(n)))
2387                fib_table_print(seq, iter->tb);
2388
2389        if (IS_TNODE(n)) {
2390                __be32 prf = htonl(n->key);
2391
2392                seq_indent(seq, iter->depth-1);
2393                seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2394                           &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2395                           tn_info(n)->full_children,
2396                           tn_info(n)->empty_children);
2397        } else {
2398                __be32 val = htonl(n->key);
2399                struct fib_alias *fa;
2400
2401                seq_indent(seq, iter->depth);
2402                seq_printf(seq, "  |-- %pI4\n", &val);
2403
2404                hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2405                        char buf1[32], buf2[32];
2406
2407                        seq_indent(seq, iter->depth + 1);
2408                        seq_printf(seq, "  /%zu %s %s",
2409                                   KEYLENGTH - fa->fa_slen,
2410                                   rtn_scope(buf1, sizeof(buf1),
2411                                             fa->fa_info->fib_scope),
2412                                   rtn_type(buf2, sizeof(buf2),
2413                                            fa->fa_type));
2414                        if (fa->fa_tos)
2415                                seq_printf(seq, " tos=%d", fa->fa_tos);
2416                        seq_putc(seq, '\n');
2417                }
2418        }
2419
2420        return 0;
2421}
2422
2423static const struct seq_operations fib_trie_seq_ops = {
2424        .start  = fib_trie_seq_start,
2425        .next   = fib_trie_seq_next,
2426        .stop   = fib_trie_seq_stop,
2427        .show   = fib_trie_seq_show,
2428};
2429
2430static int fib_trie_seq_open(struct inode *inode, struct file *file)
2431{
2432        return seq_open_net(inode, file, &fib_trie_seq_ops,
2433                            sizeof(struct fib_trie_iter));
2434}
2435
2436static const struct file_operations fib_trie_fops = {
2437        .owner  = THIS_MODULE,
2438        .open   = fib_trie_seq_open,
2439        .read   = seq_read,
2440        .llseek = seq_lseek,
2441        .release = seq_release_net,
2442};
2443
2444struct fib_route_iter {
2445        struct seq_net_private p;
2446        struct fib_table *main_tb;
2447        struct key_vector *tnode;
2448        loff_t  pos;
2449        t_key   key;
2450};
2451
2452static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2453                                            loff_t pos)
2454{
2455        struct fib_table *tb = iter->main_tb;
2456        struct key_vector *l, **tp = &iter->tnode;
2457        struct trie *t;
2458        t_key key;
2459
2460        /* use cache location of next-to-find key */
2461        if (iter->pos > 0 && pos >= iter->pos) {
2462                pos -= iter->pos;
2463                key = iter->key;
2464        } else {
2465                t = (struct trie *)tb->tb_data;
2466                iter->tnode = t->kv;
2467                iter->pos = 0;
2468                key = 0;
2469        }
2470
2471        while ((l = leaf_walk_rcu(tp, key)) != NULL) {
2472                key = l->key + 1;
2473                iter->pos++;
2474
2475                if (--pos <= 0)
2476                        break;
2477
2478                l = NULL;
2479
2480                /* handle unlikely case of a key wrap */
2481                if (!key)
2482                        break;
2483        }
2484
2485        if (l)
2486                iter->key = key;        /* remember it */
2487        else
2488                iter->pos = 0;          /* forget it */
2489
2490        return l;
2491}
2492
2493static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2494        __acquires(RCU)
2495{
2496        struct fib_route_iter *iter = seq->private;
2497        struct fib_table *tb;
2498        struct trie *t;
2499
2500        rcu_read_lock();
2501
2502        tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2503        if (!tb)
2504                return NULL;
2505
2506        iter->main_tb = tb;
2507
2508        if (*pos != 0)
2509                return fib_route_get_idx(iter, *pos);
2510
2511        t = (struct trie *)tb->tb_data;
2512        iter->tnode = t->kv;
2513        iter->pos = 0;
2514        iter->key = 0;
2515
2516        return SEQ_START_TOKEN;
2517}
2518
2519static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2520{
2521        struct fib_route_iter *iter = seq->private;
2522        struct key_vector *l = NULL;
2523        t_key key = iter->key;
2524
2525        ++*pos;
2526
2527        /* only allow key of 0 for start of sequence */
2528        if ((v == SEQ_START_TOKEN) || key)
2529                l = leaf_walk_rcu(&iter->tnode, key);
2530
2531        if (l) {
2532                iter->key = l->key + 1;
2533                iter->pos++;
2534        } else {
2535                iter->pos = 0;
2536        }
2537
2538        return l;
2539}
2540
2541static void fib_route_seq_stop(struct seq_file *seq, void *v)
2542        __releases(RCU)
2543{
2544        rcu_read_unlock();
2545}
2546
2547static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2548{
2549        unsigned int flags = 0;
2550
2551        if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2552                flags = RTF_REJECT;
2553        if (fi && fi->fib_nh->nh_gw)
2554                flags |= RTF_GATEWAY;
2555        if (mask == htonl(0xFFFFFFFF))
2556                flags |= RTF_HOST;
2557        flags |= RTF_UP;
2558        return flags;
2559}
2560
2561/*
2562 *      This outputs /proc/net/route.
2563 *      The format of the file is not supposed to be changed
2564 *      and needs to be same as fib_hash output to avoid breaking
2565 *      legacy utilities
2566 */
2567static int fib_route_seq_show(struct seq_file *seq, void *v)
2568{
2569        struct fib_route_iter *iter = seq->private;
2570        struct fib_table *tb = iter->main_tb;
2571        struct fib_alias *fa;
2572        struct key_vector *l = v;
2573        __be32 prefix;
2574
2575        if (v == SEQ_START_TOKEN) {
2576                seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2577                           "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2578                           "\tWindow\tIRTT");
2579                return 0;
2580        }
2581
2582        prefix = htonl(l->key);
2583
2584        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2585                const struct fib_info *fi = fa->fa_info;
2586                __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2587                unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2588
2589                if ((fa->fa_type == RTN_BROADCAST) ||
2590                    (fa->fa_type == RTN_MULTICAST))
2591                        continue;
2592
2593                if (fa->tb_id != tb->tb_id)
2594                        continue;
2595
2596                seq_setwidth(seq, 127);
2597
2598                if (fi)
2599                        seq_printf(seq,
2600                                   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2601                                   "%d\t%08X\t%d\t%u\t%u",
2602                                   fi->fib_dev ? fi->fib_dev->name : "*",
2603                                   prefix,
2604                                   fi->fib_nh->nh_gw, flags, 0, 0,
2605                                   fi->fib_priority,
2606                                   mask,
2607                                   (fi->fib_advmss ?
2608                                    fi->fib_advmss + 40 : 0),
2609                                   fi->fib_window,
2610                                   fi->fib_rtt >> 3);
2611                else
2612                        seq_printf(seq,
2613                                   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2614                                   "%d\t%08X\t%d\t%u\t%u",
2615                                   prefix, 0, flags, 0, 0, 0,
2616                                   mask, 0, 0, 0);
2617
2618                seq_pad(seq, '\n');
2619        }
2620
2621        return 0;
2622}
2623
2624static const struct seq_operations fib_route_seq_ops = {
2625        .start  = fib_route_seq_start,
2626        .next   = fib_route_seq_next,
2627        .stop   = fib_route_seq_stop,
2628        .show   = fib_route_seq_show,
2629};
2630
2631static int fib_route_seq_open(struct inode *inode, struct file *file)
2632{
2633        return seq_open_net(inode, file, &fib_route_seq_ops,
2634                            sizeof(struct fib_route_iter));
2635}
2636
2637static const struct file_operations fib_route_fops = {
2638        .owner  = THIS_MODULE,
2639        .open   = fib_route_seq_open,
2640        .read   = seq_read,
2641        .llseek = seq_lseek,
2642        .release = seq_release_net,
2643};
2644
2645int __net_init fib_proc_init(struct net *net)
2646{
2647        if (!proc_create("fib_trie", S_IRUGO, net->proc_net, &fib_trie_fops))
2648                goto out1;
2649
2650        if (!proc_create("fib_triestat", S_IRUGO, net->proc_net,
2651                         &fib_triestat_fops))
2652                goto out2;
2653
2654        if (!proc_create("route", S_IRUGO, net->proc_net, &fib_route_fops))
2655                goto out3;
2656
2657        return 0;
2658
2659out3:
2660        remove_proc_entry("fib_triestat", net->proc_net);
2661out2:
2662        remove_proc_entry("fib_trie", net->proc_net);
2663out1:
2664        return -ENOMEM;
2665}
2666
2667void __net_exit fib_proc_exit(struct net *net)
2668{
2669        remove_proc_entry("fib_trie", net->proc_net);
2670        remove_proc_entry("fib_triestat", net->proc_net);
2671        remove_proc_entry("route", net->proc_net);
2672}
2673
2674#endif /* CONFIG_PROC_FS */
2675