linux/net/ipv4/fib_trie.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *
   4 *   Robert Olsson <robert.olsson@its.uu.se> Uppsala Universitet
   5 *     & Swedish University of Agricultural Sciences.
   6 *
   7 *   Jens Laas <jens.laas@data.slu.se> Swedish University of
   8 *     Agricultural Sciences.
   9 *
  10 *   Hans Liss <hans.liss@its.uu.se>  Uppsala Universitet
  11 *
  12 * This work is based on the LPC-trie which is originally described in:
  13 *
  14 * An experimental study of compression methods for dynamic tries
  15 * Stefan Nilsson and Matti Tikkanen. Algorithmica, 33(1):19-33, 2002.
  16 * http://www.csc.kth.se/~snilsson/software/dyntrie2/
  17 *
  18 * IP-address lookup using LC-tries. Stefan Nilsson and Gunnar Karlsson
  19 * IEEE Journal on Selected Areas in Communications, 17(6):1083-1092, June 1999
  20 *
  21 * Code from fib_hash has been reused which includes the following header:
  22 *
  23 * INET         An implementation of the TCP/IP protocol suite for the LINUX
  24 *              operating system.  INET is implemented using the  BSD Socket
  25 *              interface as the means of communication with the user level.
  26 *
  27 *              IPv4 FIB: lookup engine and maintenance routines.
  28 *
  29 * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  30 *
  31 * Substantial contributions to this work comes from:
  32 *
  33 *              David S. Miller, <davem@davemloft.net>
  34 *              Stephen Hemminger <shemminger@osdl.org>
  35 *              Paul E. McKenney <paulmck@us.ibm.com>
  36 *              Patrick McHardy <kaber@trash.net>
  37 */
  38
  39#define VERSION "0.409"
  40
  41#include <linux/cache.h>
  42#include <linux/uaccess.h>
  43#include <linux/bitops.h>
  44#include <linux/types.h>
  45#include <linux/kernel.h>
  46#include <linux/mm.h>
  47#include <linux/string.h>
  48#include <linux/socket.h>
  49#include <linux/sockios.h>
  50#include <linux/errno.h>
  51#include <linux/in.h>
  52#include <linux/inet.h>
  53#include <linux/inetdevice.h>
  54#include <linux/netdevice.h>
  55#include <linux/if_arp.h>
  56#include <linux/proc_fs.h>
  57#include <linux/rcupdate.h>
  58#include <linux/skbuff.h>
  59#include <linux/netlink.h>
  60#include <linux/init.h>
  61#include <linux/list.h>
  62#include <linux/slab.h>
  63#include <linux/export.h>
  64#include <linux/vmalloc.h>
  65#include <linux/notifier.h>
  66#include <net/net_namespace.h>
  67#include <net/ip.h>
  68#include <net/protocol.h>
  69#include <net/route.h>
  70#include <net/tcp.h>
  71#include <net/sock.h>
  72#include <net/ip_fib.h>
  73#include <net/fib_notifier.h>
  74#include <trace/events/fib.h>
  75#include "fib_lookup.h"
  76
  77static int call_fib_entry_notifier(struct notifier_block *nb, struct net *net,
  78                                   enum fib_event_type event_type, u32 dst,
  79                                   int dst_len, struct fib_alias *fa)
  80{
  81        struct fib_entry_notifier_info info = {
  82                .dst = dst,
  83                .dst_len = dst_len,
  84                .fi = fa->fa_info,
  85                .tos = fa->fa_tos,
  86                .type = fa->fa_type,
  87                .tb_id = fa->tb_id,
  88        };
  89        return call_fib4_notifier(nb, net, event_type, &info.info);
  90}
  91
  92static int call_fib_entry_notifiers(struct net *net,
  93                                    enum fib_event_type event_type, u32 dst,
  94                                    int dst_len, struct fib_alias *fa,
  95                                    struct netlink_ext_ack *extack)
  96{
  97        struct fib_entry_notifier_info info = {
  98                .info.extack = extack,
  99                .dst = dst,
 100                .dst_len = dst_len,
 101                .fi = fa->fa_info,
 102                .tos = fa->fa_tos,
 103                .type = fa->fa_type,
 104                .tb_id = fa->tb_id,
 105        };
 106        return call_fib4_notifiers(net, event_type, &info.info);
 107}
 108
 109#define MAX_STAT_DEPTH 32
 110
 111#define KEYLENGTH       (8*sizeof(t_key))
 112#define KEY_MAX         ((t_key)~0)
 113
 114typedef unsigned int t_key;
 115
 116#define IS_TRIE(n)      ((n)->pos >= KEYLENGTH)
 117#define IS_TNODE(n)     ((n)->bits)
 118#define IS_LEAF(n)      (!(n)->bits)
 119
 120struct key_vector {
 121        t_key key;
 122        unsigned char pos;              /* 2log(KEYLENGTH) bits needed */
 123        unsigned char bits;             /* 2log(KEYLENGTH) bits needed */
 124        unsigned char slen;
 125        union {
 126                /* This list pointer if valid if (pos | bits) == 0 (LEAF) */
 127                struct hlist_head leaf;
 128                /* This array is valid if (pos | bits) > 0 (TNODE) */
 129                struct key_vector __rcu *tnode[0];
 130        };
 131};
 132
 133struct tnode {
 134        struct rcu_head rcu;
 135        t_key empty_children;           /* KEYLENGTH bits needed */
 136        t_key full_children;            /* KEYLENGTH bits needed */
 137        struct key_vector __rcu *parent;
 138        struct key_vector kv[1];
 139#define tn_bits kv[0].bits
 140};
 141
 142#define TNODE_SIZE(n)   offsetof(struct tnode, kv[0].tnode[n])
 143#define LEAF_SIZE       TNODE_SIZE(1)
 144
 145#ifdef CONFIG_IP_FIB_TRIE_STATS
 146struct trie_use_stats {
 147        unsigned int gets;
 148        unsigned int backtrack;
 149        unsigned int semantic_match_passed;
 150        unsigned int semantic_match_miss;
 151        unsigned int null_node_hit;
 152        unsigned int resize_node_skipped;
 153};
 154#endif
 155
 156struct trie_stat {
 157        unsigned int totdepth;
 158        unsigned int maxdepth;
 159        unsigned int tnodes;
 160        unsigned int leaves;
 161        unsigned int nullpointers;
 162        unsigned int prefixes;
 163        unsigned int nodesizes[MAX_STAT_DEPTH];
 164};
 165
 166struct trie {
 167        struct key_vector kv[1];
 168#ifdef CONFIG_IP_FIB_TRIE_STATS
 169        struct trie_use_stats __percpu *stats;
 170#endif
 171};
 172
 173static struct key_vector *resize(struct trie *t, struct key_vector *tn);
 174static unsigned int tnode_free_size;
 175
 176/*
 177 * synchronize_rcu after call_rcu for outstanding dirty memory; it should be
 178 * especially useful before resizing the root node with PREEMPT_NONE configs;
 179 * the value was obtained experimentally, aiming to avoid visible slowdown.
 180 */
 181unsigned int sysctl_fib_sync_mem = 512 * 1024;
 182unsigned int sysctl_fib_sync_mem_min = 64 * 1024;
 183unsigned int sysctl_fib_sync_mem_max = 64 * 1024 * 1024;
 184
 185static struct kmem_cache *fn_alias_kmem __ro_after_init;
 186static struct kmem_cache *trie_leaf_kmem __ro_after_init;
 187
 188static inline struct tnode *tn_info(struct key_vector *kv)
 189{
 190        return container_of(kv, struct tnode, kv[0]);
 191}
 192
 193/* caller must hold RTNL */
 194#define node_parent(tn) rtnl_dereference(tn_info(tn)->parent)
 195#define get_child(tn, i) rtnl_dereference((tn)->tnode[i])
 196
 197/* caller must hold RCU read lock or RTNL */
 198#define node_parent_rcu(tn) rcu_dereference_rtnl(tn_info(tn)->parent)
 199#define get_child_rcu(tn, i) rcu_dereference_rtnl((tn)->tnode[i])
 200
 201/* wrapper for rcu_assign_pointer */
 202static inline void node_set_parent(struct key_vector *n, struct key_vector *tp)
 203{
 204        if (n)
 205                rcu_assign_pointer(tn_info(n)->parent, tp);
 206}
 207
 208#define NODE_INIT_PARENT(n, p) RCU_INIT_POINTER(tn_info(n)->parent, p)
 209
 210/* This provides us with the number of children in this node, in the case of a
 211 * leaf this will return 0 meaning none of the children are accessible.
 212 */
 213static inline unsigned long child_length(const struct key_vector *tn)
 214{
 215        return (1ul << tn->bits) & ~(1ul);
 216}
 217
 218#define get_cindex(key, kv) (((key) ^ (kv)->key) >> (kv)->pos)
 219
 220static inline unsigned long get_index(t_key key, struct key_vector *kv)
 221{
 222        unsigned long index = key ^ kv->key;
 223
 224        if ((BITS_PER_LONG <= KEYLENGTH) && (KEYLENGTH == kv->pos))
 225                return 0;
 226
 227        return index >> kv->pos;
 228}
 229
 230/* To understand this stuff, an understanding of keys and all their bits is
 231 * necessary. Every node in the trie has a key associated with it, but not
 232 * all of the bits in that key are significant.
 233 *
 234 * Consider a node 'n' and its parent 'tp'.
 235 *
 236 * If n is a leaf, every bit in its key is significant. Its presence is
 237 * necessitated by path compression, since during a tree traversal (when
 238 * searching for a leaf - unless we are doing an insertion) we will completely
 239 * ignore all skipped bits we encounter. Thus we need to verify, at the end of
 240 * a potentially successful search, that we have indeed been walking the
 241 * correct key path.
 242 *
 243 * Note that we can never "miss" the correct key in the tree if present by
 244 * following the wrong path. Path compression ensures that segments of the key
 245 * that are the same for all keys with a given prefix are skipped, but the
 246 * skipped part *is* identical for each node in the subtrie below the skipped
 247 * bit! trie_insert() in this implementation takes care of that.
 248 *
 249 * if n is an internal node - a 'tnode' here, the various parts of its key
 250 * have many different meanings.
 251 *
 252 * Example:
 253 * _________________________________________________________________
 254 * | i | i | i | i | i | i | i | N | N | N | S | S | S | S | S | C |
 255 * -----------------------------------------------------------------
 256 *  31  30  29  28  27  26  25  24  23  22  21  20  19  18  17  16
 257 *
 258 * _________________________________________________________________
 259 * | C | C | C | u | u | u | u | u | u | u | u | u | u | u | u | u |
 260 * -----------------------------------------------------------------
 261 *  15  14  13  12  11  10   9   8   7   6   5   4   3   2   1   0
 262 *
 263 * tp->pos = 22
 264 * tp->bits = 3
 265 * n->pos = 13
 266 * n->bits = 4
 267 *
 268 * First, let's just ignore the bits that come before the parent tp, that is
 269 * the bits from (tp->pos + tp->bits) to 31. They are *known* but at this
 270 * point we do not use them for anything.
 271 *
 272 * The bits from (tp->pos) to (tp->pos + tp->bits - 1) - "N", above - are the
 273 * index into the parent's child array. That is, they will be used to find
 274 * 'n' among tp's children.
 275 *
 276 * The bits from (n->pos + n->bits) to (tp->pos - 1) - "S" - are skipped bits
 277 * for the node n.
 278 *
 279 * All the bits we have seen so far are significant to the node n. The rest
 280 * of the bits are really not needed or indeed known in n->key.
 281 *
 282 * The bits from (n->pos) to (n->pos + n->bits - 1) - "C" - are the index into
 283 * n's child array, and will of course be different for each child.
 284 *
 285 * The rest of the bits, from 0 to (n->pos -1) - "u" - are completely unknown
 286 * at this point.
 287 */
 288
 289static const int halve_threshold = 25;
 290static const int inflate_threshold = 50;
 291static const int halve_threshold_root = 15;
 292static const int inflate_threshold_root = 30;
 293
 294static void __alias_free_mem(struct rcu_head *head)
 295{
 296        struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
 297        kmem_cache_free(fn_alias_kmem, fa);
 298}
 299
 300static inline void alias_free_mem_rcu(struct fib_alias *fa)
 301{
 302        call_rcu(&fa->rcu, __alias_free_mem);
 303}
 304
 305#define TNODE_KMALLOC_MAX \
 306        ilog2((PAGE_SIZE - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 307#define TNODE_VMALLOC_MAX \
 308        ilog2((SIZE_MAX - TNODE_SIZE(0)) / sizeof(struct key_vector *))
 309
 310static void __node_free_rcu(struct rcu_head *head)
 311{
 312        struct tnode *n = container_of(head, struct tnode, rcu);
 313
 314        if (!n->tn_bits)
 315                kmem_cache_free(trie_leaf_kmem, n);
 316        else
 317                kvfree(n);
 318}
 319
 320#define node_free(n) call_rcu(&tn_info(n)->rcu, __node_free_rcu)
 321
 322static struct tnode *tnode_alloc(int bits)
 323{
 324        size_t size;
 325
 326        /* verify bits is within bounds */
 327        if (bits > TNODE_VMALLOC_MAX)
 328                return NULL;
 329
 330        /* determine size and verify it is non-zero and didn't overflow */
 331        size = TNODE_SIZE(1ul << bits);
 332
 333        if (size <= PAGE_SIZE)
 334                return kzalloc(size, GFP_KERNEL);
 335        else
 336                return vzalloc(size);
 337}
 338
 339static inline void empty_child_inc(struct key_vector *n)
 340{
 341        ++tn_info(n)->empty_children ? : ++tn_info(n)->full_children;
 342}
 343
 344static inline void empty_child_dec(struct key_vector *n)
 345{
 346        tn_info(n)->empty_children-- ? : tn_info(n)->full_children--;
 347}
 348
 349static struct key_vector *leaf_new(t_key key, struct fib_alias *fa)
 350{
 351        struct key_vector *l;
 352        struct tnode *kv;
 353
 354        kv = kmem_cache_alloc(trie_leaf_kmem, GFP_KERNEL);
 355        if (!kv)
 356                return NULL;
 357
 358        /* initialize key vector */
 359        l = kv->kv;
 360        l->key = key;
 361        l->pos = 0;
 362        l->bits = 0;
 363        l->slen = fa->fa_slen;
 364
 365        /* link leaf to fib alias */
 366        INIT_HLIST_HEAD(&l->leaf);
 367        hlist_add_head(&fa->fa_list, &l->leaf);
 368
 369        return l;
 370}
 371
 372static struct key_vector *tnode_new(t_key key, int pos, int bits)
 373{
 374        unsigned int shift = pos + bits;
 375        struct key_vector *tn;
 376        struct tnode *tnode;
 377
 378        /* verify bits and pos their msb bits clear and values are valid */
 379        BUG_ON(!bits || (shift > KEYLENGTH));
 380
 381        tnode = tnode_alloc(bits);
 382        if (!tnode)
 383                return NULL;
 384
 385        pr_debug("AT %p s=%zu %zu\n", tnode, TNODE_SIZE(0),
 386                 sizeof(struct key_vector *) << bits);
 387
 388        if (bits == KEYLENGTH)
 389                tnode->full_children = 1;
 390        else
 391                tnode->empty_children = 1ul << bits;
 392
 393        tn = tnode->kv;
 394        tn->key = (shift < KEYLENGTH) ? (key >> shift) << shift : 0;
 395        tn->pos = pos;
 396        tn->bits = bits;
 397        tn->slen = pos;
 398
 399        return tn;
 400}
 401
 402/* Check whether a tnode 'n' is "full", i.e. it is an internal node
 403 * and no bits are skipped. See discussion in dyntree paper p. 6
 404 */
 405static inline int tnode_full(struct key_vector *tn, struct key_vector *n)
 406{
 407        return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
 408}
 409
 410/* Add a child at position i overwriting the old value.
 411 * Update the value of full_children and empty_children.
 412 */
 413static void put_child(struct key_vector *tn, unsigned long i,
 414                      struct key_vector *n)
 415{
 416        struct key_vector *chi = get_child(tn, i);
 417        int isfull, wasfull;
 418
 419        BUG_ON(i >= child_length(tn));
 420
 421        /* update emptyChildren, overflow into fullChildren */
 422        if (!n && chi)
 423                empty_child_inc(tn);
 424        if (n && !chi)
 425                empty_child_dec(tn);
 426
 427        /* update fullChildren */
 428        wasfull = tnode_full(tn, chi);
 429        isfull = tnode_full(tn, n);
 430
 431        if (wasfull && !isfull)
 432                tn_info(tn)->full_children--;
 433        else if (!wasfull && isfull)
 434                tn_info(tn)->full_children++;
 435
 436        if (n && (tn->slen < n->slen))
 437                tn->slen = n->slen;
 438
 439        rcu_assign_pointer(tn->tnode[i], n);
 440}
 441
 442static void update_children(struct key_vector *tn)
 443{
 444        unsigned long i;
 445
 446        /* update all of the child parent pointers */
 447        for (i = child_length(tn); i;) {
 448                struct key_vector *inode = get_child(tn, --i);
 449
 450                if (!inode)
 451                        continue;
 452
 453                /* Either update the children of a tnode that
 454                 * already belongs to us or update the child
 455                 * to point to ourselves.
 456                 */
 457                if (node_parent(inode) == tn)
 458                        update_children(inode);
 459                else
 460                        node_set_parent(inode, tn);
 461        }
 462}
 463
 464static inline void put_child_root(struct key_vector *tp, t_key key,
 465                                  struct key_vector *n)
 466{
 467        if (IS_TRIE(tp))
 468                rcu_assign_pointer(tp->tnode[0], n);
 469        else
 470                put_child(tp, get_index(key, tp), n);
 471}
 472
 473static inline void tnode_free_init(struct key_vector *tn)
 474{
 475        tn_info(tn)->rcu.next = NULL;
 476}
 477
 478static inline void tnode_free_append(struct key_vector *tn,
 479                                     struct key_vector *n)
 480{
 481        tn_info(n)->rcu.next = tn_info(tn)->rcu.next;
 482        tn_info(tn)->rcu.next = &tn_info(n)->rcu;
 483}
 484
 485static void tnode_free(struct key_vector *tn)
 486{
 487        struct callback_head *head = &tn_info(tn)->rcu;
 488
 489        while (head) {
 490                head = head->next;
 491                tnode_free_size += TNODE_SIZE(1ul << tn->bits);
 492                node_free(tn);
 493
 494                tn = container_of(head, struct tnode, rcu)->kv;
 495        }
 496
 497        if (tnode_free_size >= sysctl_fib_sync_mem) {
 498                tnode_free_size = 0;
 499                synchronize_rcu();
 500        }
 501}
 502
 503static struct key_vector *replace(struct trie *t,
 504                                  struct key_vector *oldtnode,
 505                                  struct key_vector *tn)
 506{
 507        struct key_vector *tp = node_parent(oldtnode);
 508        unsigned long i;
 509
 510        /* setup the parent pointer out of and back into this node */
 511        NODE_INIT_PARENT(tn, tp);
 512        put_child_root(tp, tn->key, tn);
 513
 514        /* update all of the child parent pointers */
 515        update_children(tn);
 516
 517        /* all pointers should be clean so we are done */
 518        tnode_free(oldtnode);
 519
 520        /* resize children now that oldtnode is freed */
 521        for (i = child_length(tn); i;) {
 522                struct key_vector *inode = get_child(tn, --i);
 523
 524                /* resize child node */
 525                if (tnode_full(tn, inode))
 526                        tn = resize(t, inode);
 527        }
 528
 529        return tp;
 530}
 531
 532static struct key_vector *inflate(struct trie *t,
 533                                  struct key_vector *oldtnode)
 534{
 535        struct key_vector *tn;
 536        unsigned long i;
 537        t_key m;
 538
 539        pr_debug("In inflate\n");
 540
 541        tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
 542        if (!tn)
 543                goto notnode;
 544
 545        /* prepare oldtnode to be freed */
 546        tnode_free_init(oldtnode);
 547
 548        /* Assemble all of the pointers in our cluster, in this case that
 549         * represents all of the pointers out of our allocated nodes that
 550         * point to existing tnodes and the links between our allocated
 551         * nodes.
 552         */
 553        for (i = child_length(oldtnode), m = 1u << tn->pos; i;) {
 554                struct key_vector *inode = get_child(oldtnode, --i);
 555                struct key_vector *node0, *node1;
 556                unsigned long j, k;
 557
 558                /* An empty child */
 559                if (!inode)
 560                        continue;
 561
 562                /* A leaf or an internal node with skipped bits */
 563                if (!tnode_full(oldtnode, inode)) {
 564                        put_child(tn, get_index(inode->key, tn), inode);
 565                        continue;
 566                }
 567
 568                /* drop the node in the old tnode free list */
 569                tnode_free_append(oldtnode, inode);
 570
 571                /* An internal node with two children */
 572                if (inode->bits == 1) {
 573                        put_child(tn, 2 * i + 1, get_child(inode, 1));
 574                        put_child(tn, 2 * i, get_child(inode, 0));
 575                        continue;
 576                }
 577
 578                /* We will replace this node 'inode' with two new
 579                 * ones, 'node0' and 'node1', each with half of the
 580                 * original children. The two new nodes will have
 581                 * a position one bit further down the key and this
 582                 * means that the "significant" part of their keys
 583                 * (see the discussion near the top of this file)
 584                 * will differ by one bit, which will be "0" in
 585                 * node0's key and "1" in node1's key. Since we are
 586                 * moving the key position by one step, the bit that
 587                 * we are moving away from - the bit at position
 588                 * (tn->pos) - is the one that will differ between
 589                 * node0 and node1. So... we synthesize that bit in the
 590                 * two new keys.
 591                 */
 592                node1 = tnode_new(inode->key | m, inode->pos, inode->bits - 1);
 593                if (!node1)
 594                        goto nomem;
 595                node0 = tnode_new(inode->key, inode->pos, inode->bits - 1);
 596
 597                tnode_free_append(tn, node1);
 598                if (!node0)
 599                        goto nomem;
 600                tnode_free_append(tn, node0);
 601
 602                /* populate child pointers in new nodes */
 603                for (k = child_length(inode), j = k / 2; j;) {
 604                        put_child(node1, --j, get_child(inode, --k));
 605                        put_child(node0, j, get_child(inode, j));
 606                        put_child(node1, --j, get_child(inode, --k));
 607                        put_child(node0, j, get_child(inode, j));
 608                }
 609
 610                /* link new nodes to parent */
 611                NODE_INIT_PARENT(node1, tn);
 612                NODE_INIT_PARENT(node0, tn);
 613
 614                /* link parent to nodes */
 615                put_child(tn, 2 * i + 1, node1);
 616                put_child(tn, 2 * i, node0);
 617        }
 618
 619        /* setup the parent pointers into and out of this node */
 620        return replace(t, oldtnode, tn);
 621nomem:
 622        /* all pointers should be clean so we are done */
 623        tnode_free(tn);
 624notnode:
 625        return NULL;
 626}
 627
 628static struct key_vector *halve(struct trie *t,
 629                                struct key_vector *oldtnode)
 630{
 631        struct key_vector *tn;
 632        unsigned long i;
 633
 634        pr_debug("In halve\n");
 635
 636        tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
 637        if (!tn)
 638                goto notnode;
 639
 640        /* prepare oldtnode to be freed */
 641        tnode_free_init(oldtnode);
 642
 643        /* Assemble all of the pointers in our cluster, in this case that
 644         * represents all of the pointers out of our allocated nodes that
 645         * point to existing tnodes and the links between our allocated
 646         * nodes.
 647         */
 648        for (i = child_length(oldtnode); i;) {
 649                struct key_vector *node1 = get_child(oldtnode, --i);
 650                struct key_vector *node0 = get_child(oldtnode, --i);
 651                struct key_vector *inode;
 652
 653                /* At least one of the children is empty */
 654                if (!node1 || !node0) {
 655                        put_child(tn, i / 2, node1 ? : node0);
 656                        continue;
 657                }
 658
 659                /* Two nonempty children */
 660                inode = tnode_new(node0->key, oldtnode->pos, 1);
 661                if (!inode)
 662                        goto nomem;
 663                tnode_free_append(tn, inode);
 664
 665                /* initialize pointers out of node */
 666                put_child(inode, 1, node1);
 667                put_child(inode, 0, node0);
 668                NODE_INIT_PARENT(inode, tn);
 669
 670                /* link parent to node */
 671                put_child(tn, i / 2, inode);
 672        }
 673
 674        /* setup the parent pointers into and out of this node */
 675        return replace(t, oldtnode, tn);
 676nomem:
 677        /* all pointers should be clean so we are done */
 678        tnode_free(tn);
 679notnode:
 680        return NULL;
 681}
 682
 683static struct key_vector *collapse(struct trie *t,
 684                                   struct key_vector *oldtnode)
 685{
 686        struct key_vector *n, *tp;
 687        unsigned long i;
 688
 689        /* scan the tnode looking for that one child that might still exist */
 690        for (n = NULL, i = child_length(oldtnode); !n && i;)
 691                n = get_child(oldtnode, --i);
 692
 693        /* compress one level */
 694        tp = node_parent(oldtnode);
 695        put_child_root(tp, oldtnode->key, n);
 696        node_set_parent(n, tp);
 697
 698        /* drop dead node */
 699        node_free(oldtnode);
 700
 701        return tp;
 702}
 703
 704static unsigned char update_suffix(struct key_vector *tn)
 705{
 706        unsigned char slen = tn->pos;
 707        unsigned long stride, i;
 708        unsigned char slen_max;
 709
 710        /* only vector 0 can have a suffix length greater than or equal to
 711         * tn->pos + tn->bits, the second highest node will have a suffix
 712         * length at most of tn->pos + tn->bits - 1
 713         */
 714        slen_max = min_t(unsigned char, tn->pos + tn->bits - 1, tn->slen);
 715
 716        /* search though the list of children looking for nodes that might
 717         * have a suffix greater than the one we currently have.  This is
 718         * why we start with a stride of 2 since a stride of 1 would
 719         * represent the nodes with suffix length equal to tn->pos
 720         */
 721        for (i = 0, stride = 0x2ul ; i < child_length(tn); i += stride) {
 722                struct key_vector *n = get_child(tn, i);
 723
 724                if (!n || (n->slen <= slen))
 725                        continue;
 726
 727                /* update stride and slen based on new value */
 728                stride <<= (n->slen - slen);
 729                slen = n->slen;
 730                i &= ~(stride - 1);
 731
 732                /* stop searching if we have hit the maximum possible value */
 733                if (slen >= slen_max)
 734                        break;
 735        }
 736
 737        tn->slen = slen;
 738
 739        return slen;
 740}
 741
 742/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
 743 * the Helsinki University of Technology and Matti Tikkanen of Nokia
 744 * Telecommunications, page 6:
 745 * "A node is doubled if the ratio of non-empty children to all
 746 * children in the *doubled* node is at least 'high'."
 747 *
 748 * 'high' in this instance is the variable 'inflate_threshold'. It
 749 * is expressed as a percentage, so we multiply it with
 750 * child_length() and instead of multiplying by 2 (since the
 751 * child array will be doubled by inflate()) and multiplying
 752 * the left-hand side by 100 (to handle the percentage thing) we
 753 * multiply the left-hand side by 50.
 754 *
 755 * The left-hand side may look a bit weird: child_length(tn)
 756 * - tn->empty_children is of course the number of non-null children
 757 * in the current node. tn->full_children is the number of "full"
 758 * children, that is non-null tnodes with a skip value of 0.
 759 * All of those will be doubled in the resulting inflated tnode, so
 760 * we just count them one extra time here.
 761 *
 762 * A clearer way to write this would be:
 763 *
 764 * to_be_doubled = tn->full_children;
 765 * not_to_be_doubled = child_length(tn) - tn->empty_children -
 766 *     tn->full_children;
 767 *
 768 * new_child_length = child_length(tn) * 2;
 769 *
 770 * new_fill_factor = 100 * (not_to_be_doubled + 2*to_be_doubled) /
 771 *      new_child_length;
 772 * if (new_fill_factor >= inflate_threshold)
 773 *
 774 * ...and so on, tho it would mess up the while () loop.
 775 *
 776 * anyway,
 777 * 100 * (not_to_be_doubled + 2*to_be_doubled) / new_child_length >=
 778 *      inflate_threshold
 779 *
 780 * avoid a division:
 781 * 100 * (not_to_be_doubled + 2*to_be_doubled) >=
 782 *      inflate_threshold * new_child_length
 783 *
 784 * expand not_to_be_doubled and to_be_doubled, and shorten:
 785 * 100 * (child_length(tn) - tn->empty_children +
 786 *    tn->full_children) >= inflate_threshold * new_child_length
 787 *
 788 * expand new_child_length:
 789 * 100 * (child_length(tn) - tn->empty_children +
 790 *    tn->full_children) >=
 791 *      inflate_threshold * child_length(tn) * 2
 792 *
 793 * shorten again:
 794 * 50 * (tn->full_children + child_length(tn) -
 795 *    tn->empty_children) >= inflate_threshold *
 796 *    child_length(tn)
 797 *
 798 */
 799static inline bool should_inflate(struct key_vector *tp, struct key_vector *tn)
 800{
 801        unsigned long used = child_length(tn);
 802        unsigned long threshold = used;
 803
 804        /* Keep root node larger */
 805        threshold *= IS_TRIE(tp) ? inflate_threshold_root : inflate_threshold;
 806        used -= tn_info(tn)->empty_children;
 807        used += tn_info(tn)->full_children;
 808
 809        /* if bits == KEYLENGTH then pos = 0, and will fail below */
 810
 811        return (used > 1) && tn->pos && ((50 * used) >= threshold);
 812}
 813
 814static inline bool should_halve(struct key_vector *tp, struct key_vector *tn)
 815{
 816        unsigned long used = child_length(tn);
 817        unsigned long threshold = used;
 818
 819        /* Keep root node larger */
 820        threshold *= IS_TRIE(tp) ? halve_threshold_root : halve_threshold;
 821        used -= tn_info(tn)->empty_children;
 822
 823        /* if bits == KEYLENGTH then used = 100% on wrap, and will fail below */
 824
 825        return (used > 1) && (tn->bits > 1) && ((100 * used) < threshold);
 826}
 827
 828static inline bool should_collapse(struct key_vector *tn)
 829{
 830        unsigned long used = child_length(tn);
 831
 832        used -= tn_info(tn)->empty_children;
 833
 834        /* account for bits == KEYLENGTH case */
 835        if ((tn->bits == KEYLENGTH) && tn_info(tn)->full_children)
 836                used -= KEY_MAX;
 837
 838        /* One child or none, time to drop us from the trie */
 839        return used < 2;
 840}
 841
 842#define MAX_WORK 10
 843static struct key_vector *resize(struct trie *t, struct key_vector *tn)
 844{
 845#ifdef CONFIG_IP_FIB_TRIE_STATS
 846        struct trie_use_stats __percpu *stats = t->stats;
 847#endif
 848        struct key_vector *tp = node_parent(tn);
 849        unsigned long cindex = get_index(tn->key, tp);
 850        int max_work = MAX_WORK;
 851
 852        pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
 853                 tn, inflate_threshold, halve_threshold);
 854
 855        /* track the tnode via the pointer from the parent instead of
 856         * doing it ourselves.  This way we can let RCU fully do its
 857         * thing without us interfering
 858         */
 859        BUG_ON(tn != get_child(tp, cindex));
 860
 861        /* Double as long as the resulting node has a number of
 862         * nonempty nodes that are above the threshold.
 863         */
 864        while (should_inflate(tp, tn) && max_work) {
 865                tp = inflate(t, tn);
 866                if (!tp) {
 867#ifdef CONFIG_IP_FIB_TRIE_STATS
 868                        this_cpu_inc(stats->resize_node_skipped);
 869#endif
 870                        break;
 871                }
 872
 873                max_work--;
 874                tn = get_child(tp, cindex);
 875        }
 876
 877        /* update parent in case inflate failed */
 878        tp = node_parent(tn);
 879
 880        /* Return if at least one inflate is run */
 881        if (max_work != MAX_WORK)
 882                return tp;
 883
 884        /* Halve as long as the number of empty children in this
 885         * node is above threshold.
 886         */
 887        while (should_halve(tp, tn) && max_work) {
 888                tp = halve(t, tn);
 889                if (!tp) {
 890#ifdef CONFIG_IP_FIB_TRIE_STATS
 891                        this_cpu_inc(stats->resize_node_skipped);
 892#endif
 893                        break;
 894                }
 895
 896                max_work--;
 897                tn = get_child(tp, cindex);
 898        }
 899
 900        /* Only one child remains */
 901        if (should_collapse(tn))
 902                return collapse(t, tn);
 903
 904        /* update parent in case halve failed */
 905        return node_parent(tn);
 906}
 907
 908static void node_pull_suffix(struct key_vector *tn, unsigned char slen)
 909{
 910        unsigned char node_slen = tn->slen;
 911
 912        while ((node_slen > tn->pos) && (node_slen > slen)) {
 913                slen = update_suffix(tn);
 914                if (node_slen == slen)
 915                        break;
 916
 917                tn = node_parent(tn);
 918                node_slen = tn->slen;
 919        }
 920}
 921
 922static void node_push_suffix(struct key_vector *tn, unsigned char slen)
 923{
 924        while (tn->slen < slen) {
 925                tn->slen = slen;
 926                tn = node_parent(tn);
 927        }
 928}
 929
 930/* rcu_read_lock needs to be hold by caller from readside */
 931static struct key_vector *fib_find_node(struct trie *t,
 932                                        struct key_vector **tp, u32 key)
 933{
 934        struct key_vector *pn, *n = t->kv;
 935        unsigned long index = 0;
 936
 937        do {
 938                pn = n;
 939                n = get_child_rcu(n, index);
 940
 941                if (!n)
 942                        break;
 943
 944                index = get_cindex(key, n);
 945
 946                /* This bit of code is a bit tricky but it combines multiple
 947                 * checks into a single check.  The prefix consists of the
 948                 * prefix plus zeros for the bits in the cindex. The index
 949                 * is the difference between the key and this value.  From
 950                 * this we can actually derive several pieces of data.
 951                 *   if (index >= (1ul << bits))
 952                 *     we have a mismatch in skip bits and failed
 953                 *   else
 954                 *     we know the value is cindex
 955                 *
 956                 * This check is safe even if bits == KEYLENGTH due to the
 957                 * fact that we can only allocate a node with 32 bits if a
 958                 * long is greater than 32 bits.
 959                 */
 960                if (index >= (1ul << n->bits)) {
 961                        n = NULL;
 962                        break;
 963                }
 964
 965                /* keep searching until we find a perfect match leaf or NULL */
 966        } while (IS_TNODE(n));
 967
 968        *tp = pn;
 969
 970        return n;
 971}
 972
 973/* Return the first fib alias matching TOS with
 974 * priority less than or equal to PRIO.
 975 */
 976static struct fib_alias *fib_find_alias(struct hlist_head *fah, u8 slen,
 977                                        u8 tos, u32 prio, u32 tb_id)
 978{
 979        struct fib_alias *fa;
 980
 981        if (!fah)
 982                return NULL;
 983
 984        hlist_for_each_entry(fa, fah, fa_list) {
 985                if (fa->fa_slen < slen)
 986                        continue;
 987                if (fa->fa_slen != slen)
 988                        break;
 989                if (fa->tb_id > tb_id)
 990                        continue;
 991                if (fa->tb_id != tb_id)
 992                        break;
 993                if (fa->fa_tos > tos)
 994                        continue;
 995                if (fa->fa_info->fib_priority >= prio || fa->fa_tos < tos)
 996                        return fa;
 997        }
 998
 999        return NULL;
1000}
1001
1002static void trie_rebalance(struct trie *t, struct key_vector *tn)
1003{
1004        while (!IS_TRIE(tn))
1005                tn = resize(t, tn);
1006}
1007
1008static int fib_insert_node(struct trie *t, struct key_vector *tp,
1009                           struct fib_alias *new, t_key key)
1010{
1011        struct key_vector *n, *l;
1012
1013        l = leaf_new(key, new);
1014        if (!l)
1015                goto noleaf;
1016
1017        /* retrieve child from parent node */
1018        n = get_child(tp, get_index(key, tp));
1019
1020        /* Case 2: n is a LEAF or a TNODE and the key doesn't match.
1021         *
1022         *  Add a new tnode here
1023         *  first tnode need some special handling
1024         *  leaves us in position for handling as case 3
1025         */
1026        if (n) {
1027                struct key_vector *tn;
1028
1029                tn = tnode_new(key, __fls(key ^ n->key), 1);
1030                if (!tn)
1031                        goto notnode;
1032
1033                /* initialize routes out of node */
1034                NODE_INIT_PARENT(tn, tp);
1035                put_child(tn, get_index(key, tn) ^ 1, n);
1036
1037                /* start adding routes into the node */
1038                put_child_root(tp, key, tn);
1039                node_set_parent(n, tn);
1040
1041                /* parent now has a NULL spot where the leaf can go */
1042                tp = tn;
1043        }
1044
1045        /* Case 3: n is NULL, and will just insert a new leaf */
1046        node_push_suffix(tp, new->fa_slen);
1047        NODE_INIT_PARENT(l, tp);
1048        put_child_root(tp, key, l);
1049        trie_rebalance(t, tp);
1050
1051        return 0;
1052notnode:
1053        node_free(l);
1054noleaf:
1055        return -ENOMEM;
1056}
1057
1058/* fib notifier for ADD is sent before calling fib_insert_alias with
1059 * the expectation that the only possible failure ENOMEM
1060 */
1061static int fib_insert_alias(struct trie *t, struct key_vector *tp,
1062                            struct key_vector *l, struct fib_alias *new,
1063                            struct fib_alias *fa, t_key key)
1064{
1065        if (!l)
1066                return fib_insert_node(t, tp, new, key);
1067
1068        if (fa) {
1069                hlist_add_before_rcu(&new->fa_list, &fa->fa_list);
1070        } else {
1071                struct fib_alias *last;
1072
1073                hlist_for_each_entry(last, &l->leaf, fa_list) {
1074                        if (new->fa_slen < last->fa_slen)
1075                                break;
1076                        if ((new->fa_slen == last->fa_slen) &&
1077                            (new->tb_id > last->tb_id))
1078                                break;
1079                        fa = last;
1080                }
1081
1082                if (fa)
1083                        hlist_add_behind_rcu(&new->fa_list, &fa->fa_list);
1084                else
1085                        hlist_add_head_rcu(&new->fa_list, &l->leaf);
1086        }
1087
1088        /* if we added to the tail node then we need to update slen */
1089        if (l->slen < new->fa_slen) {
1090                l->slen = new->fa_slen;
1091                node_push_suffix(tp, new->fa_slen);
1092        }
1093
1094        return 0;
1095}
1096
1097static bool fib_valid_key_len(u32 key, u8 plen, struct netlink_ext_ack *extack)
1098{
1099        if (plen > KEYLENGTH) {
1100                NL_SET_ERR_MSG(extack, "Invalid prefix length");
1101                return false;
1102        }
1103
1104        if ((plen < KEYLENGTH) && (key << plen)) {
1105                NL_SET_ERR_MSG(extack,
1106                               "Invalid prefix for given prefix length");
1107                return false;
1108        }
1109
1110        return true;
1111}
1112
1113/* Caller must hold RTNL. */
1114int fib_table_insert(struct net *net, struct fib_table *tb,
1115                     struct fib_config *cfg, struct netlink_ext_ack *extack)
1116{
1117        enum fib_event_type event = FIB_EVENT_ENTRY_ADD;
1118        struct trie *t = (struct trie *)tb->tb_data;
1119        struct fib_alias *fa, *new_fa;
1120        struct key_vector *l, *tp;
1121        u16 nlflags = NLM_F_EXCL;
1122        struct fib_info *fi;
1123        u8 plen = cfg->fc_dst_len;
1124        u8 slen = KEYLENGTH - plen;
1125        u8 tos = cfg->fc_tos;
1126        u32 key;
1127        int err;
1128
1129        key = ntohl(cfg->fc_dst);
1130
1131        if (!fib_valid_key_len(key, plen, extack))
1132                return -EINVAL;
1133
1134        pr_debug("Insert table=%u %08x/%d\n", tb->tb_id, key, plen);
1135
1136        fi = fib_create_info(cfg, extack);
1137        if (IS_ERR(fi)) {
1138                err = PTR_ERR(fi);
1139                goto err;
1140        }
1141
1142        l = fib_find_node(t, &tp, key);
1143        fa = l ? fib_find_alias(&l->leaf, slen, tos, fi->fib_priority,
1144                                tb->tb_id) : NULL;
1145
1146        /* Now fa, if non-NULL, points to the first fib alias
1147         * with the same keys [prefix,tos,priority], if such key already
1148         * exists or to the node before which we will insert new one.
1149         *
1150         * If fa is NULL, we will need to allocate a new one and
1151         * insert to the tail of the section matching the suffix length
1152         * of the new alias.
1153         */
1154
1155        if (fa && fa->fa_tos == tos &&
1156            fa->fa_info->fib_priority == fi->fib_priority) {
1157                struct fib_alias *fa_first, *fa_match;
1158
1159                err = -EEXIST;
1160                if (cfg->fc_nlflags & NLM_F_EXCL)
1161                        goto out;
1162
1163                nlflags &= ~NLM_F_EXCL;
1164
1165                /* We have 2 goals:
1166                 * 1. Find exact match for type, scope, fib_info to avoid
1167                 * duplicate routes
1168                 * 2. Find next 'fa' (or head), NLM_F_APPEND inserts before it
1169                 */
1170                fa_match = NULL;
1171                fa_first = fa;
1172                hlist_for_each_entry_from(fa, fa_list) {
1173                        if ((fa->fa_slen != slen) ||
1174                            (fa->tb_id != tb->tb_id) ||
1175                            (fa->fa_tos != tos))
1176                                break;
1177                        if (fa->fa_info->fib_priority != fi->fib_priority)
1178                                break;
1179                        if (fa->fa_type == cfg->fc_type &&
1180                            fa->fa_info == fi) {
1181                                fa_match = fa;
1182                                break;
1183                        }
1184                }
1185
1186                if (cfg->fc_nlflags & NLM_F_REPLACE) {
1187                        struct fib_info *fi_drop;
1188                        u8 state;
1189
1190                        nlflags |= NLM_F_REPLACE;
1191                        fa = fa_first;
1192                        if (fa_match) {
1193                                if (fa == fa_match)
1194                                        err = 0;
1195                                goto out;
1196                        }
1197                        err = -ENOBUFS;
1198                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1199                        if (!new_fa)
1200                                goto out;
1201
1202                        fi_drop = fa->fa_info;
1203                        new_fa->fa_tos = fa->fa_tos;
1204                        new_fa->fa_info = fi;
1205                        new_fa->fa_type = cfg->fc_type;
1206                        state = fa->fa_state;
1207                        new_fa->fa_state = state & ~FA_S_ACCESSED;
1208                        new_fa->fa_slen = fa->fa_slen;
1209                        new_fa->tb_id = tb->tb_id;
1210                        new_fa->fa_default = -1;
1211
1212                        err = call_fib_entry_notifiers(net,
1213                                                       FIB_EVENT_ENTRY_REPLACE,
1214                                                       key, plen, new_fa,
1215                                                       extack);
1216                        if (err)
1217                                goto out_free_new_fa;
1218
1219                        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen,
1220                                  tb->tb_id, &cfg->fc_nlinfo, nlflags);
1221
1222                        hlist_replace_rcu(&fa->fa_list, &new_fa->fa_list);
1223
1224                        alias_free_mem_rcu(fa);
1225
1226                        fib_release_info(fi_drop);
1227                        if (state & FA_S_ACCESSED)
1228                                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1229
1230                        goto succeeded;
1231                }
1232                /* Error if we find a perfect match which
1233                 * uses the same scope, type, and nexthop
1234                 * information.
1235                 */
1236                if (fa_match)
1237                        goto out;
1238
1239                if (cfg->fc_nlflags & NLM_F_APPEND) {
1240                        event = FIB_EVENT_ENTRY_APPEND;
1241                        nlflags |= NLM_F_APPEND;
1242                } else {
1243                        fa = fa_first;
1244                }
1245        }
1246        err = -ENOENT;
1247        if (!(cfg->fc_nlflags & NLM_F_CREATE))
1248                goto out;
1249
1250        nlflags |= NLM_F_CREATE;
1251        err = -ENOBUFS;
1252        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1253        if (!new_fa)
1254                goto out;
1255
1256        new_fa->fa_info = fi;
1257        new_fa->fa_tos = tos;
1258        new_fa->fa_type = cfg->fc_type;
1259        new_fa->fa_state = 0;
1260        new_fa->fa_slen = slen;
1261        new_fa->tb_id = tb->tb_id;
1262        new_fa->fa_default = -1;
1263
1264        err = call_fib_entry_notifiers(net, event, key, plen, new_fa, extack);
1265        if (err)
1266                goto out_free_new_fa;
1267
1268        /* Insert new entry to the list. */
1269        err = fib_insert_alias(t, tp, l, new_fa, fa, key);
1270        if (err)
1271                goto out_fib_notif;
1272
1273        if (!plen)
1274                tb->tb_num_default++;
1275
1276        rt_cache_flush(cfg->fc_nlinfo.nl_net);
1277        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, new_fa->tb_id,
1278                  &cfg->fc_nlinfo, nlflags);
1279succeeded:
1280        return 0;
1281
1282out_fib_notif:
1283        /* notifier was sent that entry would be added to trie, but
1284         * the add failed and need to recover. Only failure for
1285         * fib_insert_alias is ENOMEM.
1286         */
1287        NL_SET_ERR_MSG(extack, "Failed to insert route into trie");
1288        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key,
1289                                 plen, new_fa, NULL);
1290out_free_new_fa:
1291        kmem_cache_free(fn_alias_kmem, new_fa);
1292out:
1293        fib_release_info(fi);
1294err:
1295        return err;
1296}
1297
1298static inline t_key prefix_mismatch(t_key key, struct key_vector *n)
1299{
1300        t_key prefix = n->key;
1301
1302        return (key ^ prefix) & (prefix | -prefix);
1303}
1304
1305/* should be called with rcu_read_lock */
1306int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
1307                     struct fib_result *res, int fib_flags)
1308{
1309        struct trie *t = (struct trie *) tb->tb_data;
1310#ifdef CONFIG_IP_FIB_TRIE_STATS
1311        struct trie_use_stats __percpu *stats = t->stats;
1312#endif
1313        const t_key key = ntohl(flp->daddr);
1314        struct key_vector *n, *pn;
1315        struct fib_alias *fa;
1316        unsigned long index;
1317        t_key cindex;
1318
1319        pn = t->kv;
1320        cindex = 0;
1321
1322        n = get_child_rcu(pn, cindex);
1323        if (!n) {
1324                trace_fib_table_lookup(tb->tb_id, flp, NULL, -EAGAIN);
1325                return -EAGAIN;
1326        }
1327
1328#ifdef CONFIG_IP_FIB_TRIE_STATS
1329        this_cpu_inc(stats->gets);
1330#endif
1331
1332        /* Step 1: Travel to the longest prefix match in the trie */
1333        for (;;) {
1334                index = get_cindex(key, n);
1335
1336                /* This bit of code is a bit tricky but it combines multiple
1337                 * checks into a single check.  The prefix consists of the
1338                 * prefix plus zeros for the "bits" in the prefix. The index
1339                 * is the difference between the key and this value.  From
1340                 * this we can actually derive several pieces of data.
1341                 *   if (index >= (1ul << bits))
1342                 *     we have a mismatch in skip bits and failed
1343                 *   else
1344                 *     we know the value is cindex
1345                 *
1346                 * This check is safe even if bits == KEYLENGTH due to the
1347                 * fact that we can only allocate a node with 32 bits if a
1348                 * long is greater than 32 bits.
1349                 */
1350                if (index >= (1ul << n->bits))
1351                        break;
1352
1353                /* we have found a leaf. Prefixes have already been compared */
1354                if (IS_LEAF(n))
1355                        goto found;
1356
1357                /* only record pn and cindex if we are going to be chopping
1358                 * bits later.  Otherwise we are just wasting cycles.
1359                 */
1360                if (n->slen > n->pos) {
1361                        pn = n;
1362                        cindex = index;
1363                }
1364
1365                n = get_child_rcu(n, index);
1366                if (unlikely(!n))
1367                        goto backtrace;
1368        }
1369
1370        /* Step 2: Sort out leaves and begin backtracing for longest prefix */
1371        for (;;) {
1372                /* record the pointer where our next node pointer is stored */
1373                struct key_vector __rcu **cptr = n->tnode;
1374
1375                /* This test verifies that none of the bits that differ
1376                 * between the key and the prefix exist in the region of
1377                 * the lsb and higher in the prefix.
1378                 */
1379                if (unlikely(prefix_mismatch(key, n)) || (n->slen == n->pos))
1380                        goto backtrace;
1381
1382                /* exit out and process leaf */
1383                if (unlikely(IS_LEAF(n)))
1384                        break;
1385
1386                /* Don't bother recording parent info.  Since we are in
1387                 * prefix match mode we will have to come back to wherever
1388                 * we started this traversal anyway
1389                 */
1390
1391                while ((n = rcu_dereference(*cptr)) == NULL) {
1392backtrace:
1393#ifdef CONFIG_IP_FIB_TRIE_STATS
1394                        if (!n)
1395                                this_cpu_inc(stats->null_node_hit);
1396#endif
1397                        /* If we are at cindex 0 there are no more bits for
1398                         * us to strip at this level so we must ascend back
1399                         * up one level to see if there are any more bits to
1400                         * be stripped there.
1401                         */
1402                        while (!cindex) {
1403                                t_key pkey = pn->key;
1404
1405                                /* If we don't have a parent then there is
1406                                 * nothing for us to do as we do not have any
1407                                 * further nodes to parse.
1408                                 */
1409                                if (IS_TRIE(pn)) {
1410                                        trace_fib_table_lookup(tb->tb_id, flp,
1411                                                               NULL, -EAGAIN);
1412                                        return -EAGAIN;
1413                                }
1414#ifdef CONFIG_IP_FIB_TRIE_STATS
1415                                this_cpu_inc(stats->backtrack);
1416#endif
1417                                /* Get Child's index */
1418                                pn = node_parent_rcu(pn);
1419                                cindex = get_index(pkey, pn);
1420                        }
1421
1422                        /* strip the least significant bit from the cindex */
1423                        cindex &= cindex - 1;
1424
1425                        /* grab pointer for next child node */
1426                        cptr = &pn->tnode[cindex];
1427                }
1428        }
1429
1430found:
1431        /* this line carries forward the xor from earlier in the function */
1432        index = key ^ n->key;
1433
1434        /* Step 3: Process the leaf, if that fails fall back to backtracing */
1435        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
1436                struct fib_info *fi = fa->fa_info;
1437                int nhsel, err;
1438
1439                if ((BITS_PER_LONG > KEYLENGTH) || (fa->fa_slen < KEYLENGTH)) {
1440                        if (index >= (1ul << fa->fa_slen))
1441                                continue;
1442                }
1443                if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1444                        continue;
1445                if (fi->fib_dead)
1446                        continue;
1447                if (fa->fa_info->fib_scope < flp->flowi4_scope)
1448                        continue;
1449                fib_alias_accessed(fa);
1450                err = fib_props[fa->fa_type].error;
1451                if (unlikely(err < 0)) {
1452#ifdef CONFIG_IP_FIB_TRIE_STATS
1453                        this_cpu_inc(stats->semantic_match_passed);
1454#endif
1455                        trace_fib_table_lookup(tb->tb_id, flp, NULL, err);
1456                        return err;
1457                }
1458                if (fi->fib_flags & RTNH_F_DEAD)
1459                        continue;
1460                for (nhsel = 0; nhsel < fi->fib_nhs; nhsel++) {
1461                        struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel);
1462
1463                        if (nhc->nhc_flags & RTNH_F_DEAD)
1464                                continue;
1465                        if (ip_ignore_linkdown(nhc->nhc_dev) &&
1466                            nhc->nhc_flags & RTNH_F_LINKDOWN &&
1467                            !(fib_flags & FIB_LOOKUP_IGNORE_LINKSTATE))
1468                                continue;
1469                        if (!(flp->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) {
1470                                if (flp->flowi4_oif &&
1471                                    flp->flowi4_oif != nhc->nhc_oif)
1472                                        continue;
1473                        }
1474
1475                        if (!(fib_flags & FIB_LOOKUP_NOREF))
1476                                refcount_inc(&fi->fib_clntref);
1477
1478                        res->prefix = htonl(n->key);
1479                        res->prefixlen = KEYLENGTH - fa->fa_slen;
1480                        res->nh_sel = nhsel;
1481                        res->nhc = nhc;
1482                        res->type = fa->fa_type;
1483                        res->scope = fi->fib_scope;
1484                        res->fi = fi;
1485                        res->table = tb;
1486                        res->fa_head = &n->leaf;
1487#ifdef CONFIG_IP_FIB_TRIE_STATS
1488                        this_cpu_inc(stats->semantic_match_passed);
1489#endif
1490                        trace_fib_table_lookup(tb->tb_id, flp, nhc, err);
1491
1492                        return err;
1493                }
1494        }
1495#ifdef CONFIG_IP_FIB_TRIE_STATS
1496        this_cpu_inc(stats->semantic_match_miss);
1497#endif
1498        goto backtrace;
1499}
1500EXPORT_SYMBOL_GPL(fib_table_lookup);
1501
1502static void fib_remove_alias(struct trie *t, struct key_vector *tp,
1503                             struct key_vector *l, struct fib_alias *old)
1504{
1505        /* record the location of the previous list_info entry */
1506        struct hlist_node **pprev = old->fa_list.pprev;
1507        struct fib_alias *fa = hlist_entry(pprev, typeof(*fa), fa_list.next);
1508
1509        /* remove the fib_alias from the list */
1510        hlist_del_rcu(&old->fa_list);
1511
1512        /* if we emptied the list this leaf will be freed and we can sort
1513         * out parent suffix lengths as a part of trie_rebalance
1514         */
1515        if (hlist_empty(&l->leaf)) {
1516                if (tp->slen == l->slen)
1517                        node_pull_suffix(tp, tp->pos);
1518                put_child_root(tp, l->key, NULL);
1519                node_free(l);
1520                trie_rebalance(t, tp);
1521                return;
1522        }
1523
1524        /* only access fa if it is pointing at the last valid hlist_node */
1525        if (*pprev)
1526                return;
1527
1528        /* update the trie with the latest suffix length */
1529        l->slen = fa->fa_slen;
1530        node_pull_suffix(tp, fa->fa_slen);
1531}
1532
1533/* Caller must hold RTNL. */
1534int fib_table_delete(struct net *net, struct fib_table *tb,
1535                     struct fib_config *cfg, struct netlink_ext_ack *extack)
1536{
1537        struct trie *t = (struct trie *) tb->tb_data;
1538        struct fib_alias *fa, *fa_to_delete;
1539        struct key_vector *l, *tp;
1540        u8 plen = cfg->fc_dst_len;
1541        u8 slen = KEYLENGTH - plen;
1542        u8 tos = cfg->fc_tos;
1543        u32 key;
1544
1545        key = ntohl(cfg->fc_dst);
1546
1547        if (!fib_valid_key_len(key, plen, extack))
1548                return -EINVAL;
1549
1550        l = fib_find_node(t, &tp, key);
1551        if (!l)
1552                return -ESRCH;
1553
1554        fa = fib_find_alias(&l->leaf, slen, tos, 0, tb->tb_id);
1555        if (!fa)
1556                return -ESRCH;
1557
1558        pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
1559
1560        fa_to_delete = NULL;
1561        hlist_for_each_entry_from(fa, fa_list) {
1562                struct fib_info *fi = fa->fa_info;
1563
1564                if ((fa->fa_slen != slen) ||
1565                    (fa->tb_id != tb->tb_id) ||
1566                    (fa->fa_tos != tos))
1567                        break;
1568
1569                if ((!cfg->fc_type || fa->fa_type == cfg->fc_type) &&
1570                    (cfg->fc_scope == RT_SCOPE_NOWHERE ||
1571                     fa->fa_info->fib_scope == cfg->fc_scope) &&
1572                    (!cfg->fc_prefsrc ||
1573                     fi->fib_prefsrc == cfg->fc_prefsrc) &&
1574                    (!cfg->fc_protocol ||
1575                     fi->fib_protocol == cfg->fc_protocol) &&
1576                    fib_nh_match(cfg, fi, extack) == 0 &&
1577                    fib_metrics_match(cfg, fi)) {
1578                        fa_to_delete = fa;
1579                        break;
1580                }
1581        }
1582
1583        if (!fa_to_delete)
1584                return -ESRCH;
1585
1586        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, key, plen,
1587                                 fa_to_delete, extack);
1588        rtmsg_fib(RTM_DELROUTE, htonl(key), fa_to_delete, plen, tb->tb_id,
1589                  &cfg->fc_nlinfo, 0);
1590
1591        if (!plen)
1592                tb->tb_num_default--;
1593
1594        fib_remove_alias(t, tp, l, fa_to_delete);
1595
1596        if (fa_to_delete->fa_state & FA_S_ACCESSED)
1597                rt_cache_flush(cfg->fc_nlinfo.nl_net);
1598
1599        fib_release_info(fa_to_delete->fa_info);
1600        alias_free_mem_rcu(fa_to_delete);
1601        return 0;
1602}
1603
1604/* Scan for the next leaf starting at the provided key value */
1605static struct key_vector *leaf_walk_rcu(struct key_vector **tn, t_key key)
1606{
1607        struct key_vector *pn, *n = *tn;
1608        unsigned long cindex;
1609
1610        /* this loop is meant to try and find the key in the trie */
1611        do {
1612                /* record parent and next child index */
1613                pn = n;
1614                cindex = (key > pn->key) ? get_index(key, pn) : 0;
1615
1616                if (cindex >> pn->bits)
1617                        break;
1618
1619                /* descend into the next child */
1620                n = get_child_rcu(pn, cindex++);
1621                if (!n)
1622                        break;
1623
1624                /* guarantee forward progress on the keys */
1625                if (IS_LEAF(n) && (n->key >= key))
1626                        goto found;
1627        } while (IS_TNODE(n));
1628
1629        /* this loop will search for the next leaf with a greater key */
1630        while (!IS_TRIE(pn)) {
1631                /* if we exhausted the parent node we will need to climb */
1632                if (cindex >= (1ul << pn->bits)) {
1633                        t_key pkey = pn->key;
1634
1635                        pn = node_parent_rcu(pn);
1636                        cindex = get_index(pkey, pn) + 1;
1637                        continue;
1638                }
1639
1640                /* grab the next available node */
1641                n = get_child_rcu(pn, cindex++);
1642                if (!n)
1643                        continue;
1644
1645                /* no need to compare keys since we bumped the index */
1646                if (IS_LEAF(n))
1647                        goto found;
1648
1649                /* Rescan start scanning in new node */
1650                pn = n;
1651                cindex = 0;
1652        }
1653
1654        *tn = pn;
1655        return NULL; /* Root of trie */
1656found:
1657        /* if we are at the limit for keys just return NULL for the tnode */
1658        *tn = pn;
1659        return n;
1660}
1661
1662static void fib_trie_free(struct fib_table *tb)
1663{
1664        struct trie *t = (struct trie *)tb->tb_data;
1665        struct key_vector *pn = t->kv;
1666        unsigned long cindex = 1;
1667        struct hlist_node *tmp;
1668        struct fib_alias *fa;
1669
1670        /* walk trie in reverse order and free everything */
1671        for (;;) {
1672                struct key_vector *n;
1673
1674                if (!(cindex--)) {
1675                        t_key pkey = pn->key;
1676
1677                        if (IS_TRIE(pn))
1678                                break;
1679
1680                        n = pn;
1681                        pn = node_parent(pn);
1682
1683                        /* drop emptied tnode */
1684                        put_child_root(pn, n->key, NULL);
1685                        node_free(n);
1686
1687                        cindex = get_index(pkey, pn);
1688
1689                        continue;
1690                }
1691
1692                /* grab the next available node */
1693                n = get_child(pn, cindex);
1694                if (!n)
1695                        continue;
1696
1697                if (IS_TNODE(n)) {
1698                        /* record pn and cindex for leaf walking */
1699                        pn = n;
1700                        cindex = 1ul << n->bits;
1701
1702                        continue;
1703                }
1704
1705                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1706                        hlist_del_rcu(&fa->fa_list);
1707                        alias_free_mem_rcu(fa);
1708                }
1709
1710                put_child_root(pn, n->key, NULL);
1711                node_free(n);
1712        }
1713
1714#ifdef CONFIG_IP_FIB_TRIE_STATS
1715        free_percpu(t->stats);
1716#endif
1717        kfree(tb);
1718}
1719
1720struct fib_table *fib_trie_unmerge(struct fib_table *oldtb)
1721{
1722        struct trie *ot = (struct trie *)oldtb->tb_data;
1723        struct key_vector *l, *tp = ot->kv;
1724        struct fib_table *local_tb;
1725        struct fib_alias *fa;
1726        struct trie *lt;
1727        t_key key = 0;
1728
1729        if (oldtb->tb_data == oldtb->__data)
1730                return oldtb;
1731
1732        local_tb = fib_trie_table(RT_TABLE_LOCAL, NULL);
1733        if (!local_tb)
1734                return NULL;
1735
1736        lt = (struct trie *)local_tb->tb_data;
1737
1738        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1739                struct key_vector *local_l = NULL, *local_tp;
1740
1741                hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1742                        struct fib_alias *new_fa;
1743
1744                        if (local_tb->tb_id != fa->tb_id)
1745                                continue;
1746
1747                        /* clone fa for new local table */
1748                        new_fa = kmem_cache_alloc(fn_alias_kmem, GFP_KERNEL);
1749                        if (!new_fa)
1750                                goto out;
1751
1752                        memcpy(new_fa, fa, sizeof(*fa));
1753
1754                        /* insert clone into table */
1755                        if (!local_l)
1756                                local_l = fib_find_node(lt, &local_tp, l->key);
1757
1758                        if (fib_insert_alias(lt, local_tp, local_l, new_fa,
1759                                             NULL, l->key)) {
1760                                kmem_cache_free(fn_alias_kmem, new_fa);
1761                                goto out;
1762                        }
1763                }
1764
1765                /* stop loop if key wrapped back to 0 */
1766                key = l->key + 1;
1767                if (key < l->key)
1768                        break;
1769        }
1770
1771        return local_tb;
1772out:
1773        fib_trie_free(local_tb);
1774
1775        return NULL;
1776}
1777
1778/* Caller must hold RTNL */
1779void fib_table_flush_external(struct fib_table *tb)
1780{
1781        struct trie *t = (struct trie *)tb->tb_data;
1782        struct key_vector *pn = t->kv;
1783        unsigned long cindex = 1;
1784        struct hlist_node *tmp;
1785        struct fib_alias *fa;
1786
1787        /* walk trie in reverse order */
1788        for (;;) {
1789                unsigned char slen = 0;
1790                struct key_vector *n;
1791
1792                if (!(cindex--)) {
1793                        t_key pkey = pn->key;
1794
1795                        /* cannot resize the trie vector */
1796                        if (IS_TRIE(pn))
1797                                break;
1798
1799                        /* update the suffix to address pulled leaves */
1800                        if (pn->slen > pn->pos)
1801                                update_suffix(pn);
1802
1803                        /* resize completed node */
1804                        pn = resize(t, pn);
1805                        cindex = get_index(pkey, pn);
1806
1807                        continue;
1808                }
1809
1810                /* grab the next available node */
1811                n = get_child(pn, cindex);
1812                if (!n)
1813                        continue;
1814
1815                if (IS_TNODE(n)) {
1816                        /* record pn and cindex for leaf walking */
1817                        pn = n;
1818                        cindex = 1ul << n->bits;
1819
1820                        continue;
1821                }
1822
1823                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1824                        /* if alias was cloned to local then we just
1825                         * need to remove the local copy from main
1826                         */
1827                        if (tb->tb_id != fa->tb_id) {
1828                                hlist_del_rcu(&fa->fa_list);
1829                                alias_free_mem_rcu(fa);
1830                                continue;
1831                        }
1832
1833                        /* record local slen */
1834                        slen = fa->fa_slen;
1835                }
1836
1837                /* update leaf slen */
1838                n->slen = slen;
1839
1840                if (hlist_empty(&n->leaf)) {
1841                        put_child_root(pn, n->key, NULL);
1842                        node_free(n);
1843                }
1844        }
1845}
1846
1847/* Caller must hold RTNL. */
1848int fib_table_flush(struct net *net, struct fib_table *tb, bool flush_all)
1849{
1850        struct trie *t = (struct trie *)tb->tb_data;
1851        struct key_vector *pn = t->kv;
1852        unsigned long cindex = 1;
1853        struct hlist_node *tmp;
1854        struct fib_alias *fa;
1855        int found = 0;
1856
1857        /* walk trie in reverse order */
1858        for (;;) {
1859                unsigned char slen = 0;
1860                struct key_vector *n;
1861
1862                if (!(cindex--)) {
1863                        t_key pkey = pn->key;
1864
1865                        /* cannot resize the trie vector */
1866                        if (IS_TRIE(pn))
1867                                break;
1868
1869                        /* update the suffix to address pulled leaves */
1870                        if (pn->slen > pn->pos)
1871                                update_suffix(pn);
1872
1873                        /* resize completed node */
1874                        pn = resize(t, pn);
1875                        cindex = get_index(pkey, pn);
1876
1877                        continue;
1878                }
1879
1880                /* grab the next available node */
1881                n = get_child(pn, cindex);
1882                if (!n)
1883                        continue;
1884
1885                if (IS_TNODE(n)) {
1886                        /* record pn and cindex for leaf walking */
1887                        pn = n;
1888                        cindex = 1ul << n->bits;
1889
1890                        continue;
1891                }
1892
1893                hlist_for_each_entry_safe(fa, tmp, &n->leaf, fa_list) {
1894                        struct fib_info *fi = fa->fa_info;
1895
1896                        if (!fi || tb->tb_id != fa->tb_id ||
1897                            (!(fi->fib_flags & RTNH_F_DEAD) &&
1898                             !fib_props[fa->fa_type].error)) {
1899                                slen = fa->fa_slen;
1900                                continue;
1901                        }
1902
1903                        /* Do not flush error routes if network namespace is
1904                         * not being dismantled
1905                         */
1906                        if (!flush_all && fib_props[fa->fa_type].error) {
1907                                slen = fa->fa_slen;
1908                                continue;
1909                        }
1910
1911                        call_fib_entry_notifiers(net, FIB_EVENT_ENTRY_DEL,
1912                                                 n->key,
1913                                                 KEYLENGTH - fa->fa_slen, fa,
1914                                                 NULL);
1915                        hlist_del_rcu(&fa->fa_list);
1916                        fib_release_info(fa->fa_info);
1917                        alias_free_mem_rcu(fa);
1918                        found++;
1919                }
1920
1921                /* update leaf slen */
1922                n->slen = slen;
1923
1924                if (hlist_empty(&n->leaf)) {
1925                        put_child_root(pn, n->key, NULL);
1926                        node_free(n);
1927                }
1928        }
1929
1930        pr_debug("trie_flush found=%d\n", found);
1931        return found;
1932}
1933
1934static void fib_leaf_notify(struct net *net, struct key_vector *l,
1935                            struct fib_table *tb, struct notifier_block *nb)
1936{
1937        struct fib_alias *fa;
1938
1939        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
1940                struct fib_info *fi = fa->fa_info;
1941
1942                if (!fi)
1943                        continue;
1944
1945                /* local and main table can share the same trie,
1946                 * so don't notify twice for the same entry.
1947                 */
1948                if (tb->tb_id != fa->tb_id)
1949                        continue;
1950
1951                call_fib_entry_notifier(nb, net, FIB_EVENT_ENTRY_ADD, l->key,
1952                                        KEYLENGTH - fa->fa_slen, fa);
1953        }
1954}
1955
1956static void fib_table_notify(struct net *net, struct fib_table *tb,
1957                             struct notifier_block *nb)
1958{
1959        struct trie *t = (struct trie *)tb->tb_data;
1960        struct key_vector *l, *tp = t->kv;
1961        t_key key = 0;
1962
1963        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
1964                fib_leaf_notify(net, l, tb, nb);
1965
1966                key = l->key + 1;
1967                /* stop in case of wrap around */
1968                if (key < l->key)
1969                        break;
1970        }
1971}
1972
1973void fib_notify(struct net *net, struct notifier_block *nb)
1974{
1975        unsigned int h;
1976
1977        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
1978                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
1979                struct fib_table *tb;
1980
1981                hlist_for_each_entry_rcu(tb, head, tb_hlist)
1982                        fib_table_notify(net, tb, nb);
1983        }
1984}
1985
1986static void __trie_free_rcu(struct rcu_head *head)
1987{
1988        struct fib_table *tb = container_of(head, struct fib_table, rcu);
1989#ifdef CONFIG_IP_FIB_TRIE_STATS
1990        struct trie *t = (struct trie *)tb->tb_data;
1991
1992        if (tb->tb_data == tb->__data)
1993                free_percpu(t->stats);
1994#endif /* CONFIG_IP_FIB_TRIE_STATS */
1995        kfree(tb);
1996}
1997
1998void fib_free_table(struct fib_table *tb)
1999{
2000        call_rcu(&tb->rcu, __trie_free_rcu);
2001}
2002
2003static int fn_trie_dump_leaf(struct key_vector *l, struct fib_table *tb,
2004                             struct sk_buff *skb, struct netlink_callback *cb,
2005                             struct fib_dump_filter *filter)
2006{
2007        unsigned int flags = NLM_F_MULTI;
2008        __be32 xkey = htonl(l->key);
2009        struct fib_alias *fa;
2010        int i, s_i;
2011
2012        if (filter->filter_set)
2013                flags |= NLM_F_DUMP_FILTERED;
2014
2015        s_i = cb->args[4];
2016        i = 0;
2017
2018        /* rcu_read_lock is hold by caller */
2019        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2020                int err;
2021
2022                if (i < s_i)
2023                        goto next;
2024
2025                if (tb->tb_id != fa->tb_id)
2026                        goto next;
2027
2028                if (filter->filter_set) {
2029                        if (filter->rt_type && fa->fa_type != filter->rt_type)
2030                                goto next;
2031
2032                        if ((filter->protocol &&
2033                             fa->fa_info->fib_protocol != filter->protocol))
2034                                goto next;
2035
2036                        if (filter->dev &&
2037                            !fib_info_nh_uses_dev(fa->fa_info, filter->dev))
2038                                goto next;
2039                }
2040
2041                err = fib_dump_info(skb, NETLINK_CB(cb->skb).portid,
2042                                    cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2043                                    tb->tb_id, fa->fa_type,
2044                                    xkey, KEYLENGTH - fa->fa_slen,
2045                                    fa->fa_tos, fa->fa_info, flags);
2046                if (err < 0) {
2047                        cb->args[4] = i;
2048                        return err;
2049                }
2050next:
2051                i++;
2052        }
2053
2054        cb->args[4] = i;
2055        return skb->len;
2056}
2057
2058/* rcu_read_lock needs to be hold by caller from readside */
2059int fib_table_dump(struct fib_table *tb, struct sk_buff *skb,
2060                   struct netlink_callback *cb, struct fib_dump_filter *filter)
2061{
2062        struct trie *t = (struct trie *)tb->tb_data;
2063        struct key_vector *l, *tp = t->kv;
2064        /* Dump starting at last key.
2065         * Note: 0.0.0.0/0 (ie default) is first key.
2066         */
2067        int count = cb->args[2];
2068        t_key key = cb->args[3];
2069
2070        while ((l = leaf_walk_rcu(&tp, key)) != NULL) {
2071                int err;
2072
2073                err = fn_trie_dump_leaf(l, tb, skb, cb, filter);
2074                if (err < 0) {
2075                        cb->args[3] = key;
2076                        cb->args[2] = count;
2077                        return err;
2078                }
2079
2080                ++count;
2081                key = l->key + 1;
2082
2083                memset(&cb->args[4], 0,
2084                       sizeof(cb->args) - 4*sizeof(cb->args[0]));
2085
2086                /* stop loop if key wrapped back to 0 */
2087                if (key < l->key)
2088                        break;
2089        }
2090
2091        cb->args[3] = key;
2092        cb->args[2] = count;
2093
2094        return skb->len;
2095}
2096
2097void __init fib_trie_init(void)
2098{
2099        fn_alias_kmem = kmem_cache_create("ip_fib_alias",
2100                                          sizeof(struct fib_alias),
2101                                          0, SLAB_PANIC, NULL);
2102
2103        trie_leaf_kmem = kmem_cache_create("ip_fib_trie",
2104                                           LEAF_SIZE,
2105                                           0, SLAB_PANIC, NULL);
2106}
2107
2108struct fib_table *fib_trie_table(u32 id, struct fib_table *alias)
2109{
2110        struct fib_table *tb;
2111        struct trie *t;
2112        size_t sz = sizeof(*tb);
2113
2114        if (!alias)
2115                sz += sizeof(struct trie);
2116
2117        tb = kzalloc(sz, GFP_KERNEL);
2118        if (!tb)
2119                return NULL;
2120
2121        tb->tb_id = id;
2122        tb->tb_num_default = 0;
2123        tb->tb_data = (alias ? alias->__data : tb->__data);
2124
2125        if (alias)
2126                return tb;
2127
2128        t = (struct trie *) tb->tb_data;
2129        t->kv[0].pos = KEYLENGTH;
2130        t->kv[0].slen = KEYLENGTH;
2131#ifdef CONFIG_IP_FIB_TRIE_STATS
2132        t->stats = alloc_percpu(struct trie_use_stats);
2133        if (!t->stats) {
2134                kfree(tb);
2135                tb = NULL;
2136        }
2137#endif
2138
2139        return tb;
2140}
2141
2142#ifdef CONFIG_PROC_FS
2143/* Depth first Trie walk iterator */
2144struct fib_trie_iter {
2145        struct seq_net_private p;
2146        struct fib_table *tb;
2147        struct key_vector *tnode;
2148        unsigned int index;
2149        unsigned int depth;
2150};
2151
2152static struct key_vector *fib_trie_get_next(struct fib_trie_iter *iter)
2153{
2154        unsigned long cindex = iter->index;
2155        struct key_vector *pn = iter->tnode;
2156        t_key pkey;
2157
2158        pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
2159                 iter->tnode, iter->index, iter->depth);
2160
2161        while (!IS_TRIE(pn)) {
2162                while (cindex < child_length(pn)) {
2163                        struct key_vector *n = get_child_rcu(pn, cindex++);
2164
2165                        if (!n)
2166                                continue;
2167
2168                        if (IS_LEAF(n)) {
2169                                iter->tnode = pn;
2170                                iter->index = cindex;
2171                        } else {
2172                                /* push down one level */
2173                                iter->tnode = n;
2174                                iter->index = 0;
2175                                ++iter->depth;
2176                        }
2177
2178                        return n;
2179                }
2180
2181                /* Current node exhausted, pop back up */
2182                pkey = pn->key;
2183                pn = node_parent_rcu(pn);
2184                cindex = get_index(pkey, pn) + 1;
2185                --iter->depth;
2186        }
2187
2188        /* record root node so further searches know we are done */
2189        iter->tnode = pn;
2190        iter->index = 0;
2191
2192        return NULL;
2193}
2194
2195static struct key_vector *fib_trie_get_first(struct fib_trie_iter *iter,
2196                                             struct trie *t)
2197{
2198        struct key_vector *n, *pn;
2199
2200        if (!t)
2201                return NULL;
2202
2203        pn = t->kv;
2204        n = rcu_dereference(pn->tnode[0]);
2205        if (!n)
2206                return NULL;
2207
2208        if (IS_TNODE(n)) {
2209                iter->tnode = n;
2210                iter->index = 0;
2211                iter->depth = 1;
2212        } else {
2213                iter->tnode = pn;
2214                iter->index = 0;
2215                iter->depth = 0;
2216        }
2217
2218        return n;
2219}
2220
2221static void trie_collect_stats(struct trie *t, struct trie_stat *s)
2222{
2223        struct key_vector *n;
2224        struct fib_trie_iter iter;
2225
2226        memset(s, 0, sizeof(*s));
2227
2228        rcu_read_lock();
2229        for (n = fib_trie_get_first(&iter, t); n; n = fib_trie_get_next(&iter)) {
2230                if (IS_LEAF(n)) {
2231                        struct fib_alias *fa;
2232
2233                        s->leaves++;
2234                        s->totdepth += iter.depth;
2235                        if (iter.depth > s->maxdepth)
2236                                s->maxdepth = iter.depth;
2237
2238                        hlist_for_each_entry_rcu(fa, &n->leaf, fa_list)
2239                                ++s->prefixes;
2240                } else {
2241                        s->tnodes++;
2242                        if (n->bits < MAX_STAT_DEPTH)
2243                                s->nodesizes[n->bits]++;
2244                        s->nullpointers += tn_info(n)->empty_children;
2245                }
2246        }
2247        rcu_read_unlock();
2248}
2249
2250/*
2251 *      This outputs /proc/net/fib_triestats
2252 */
2253static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
2254{
2255        unsigned int i, max, pointers, bytes, avdepth;
2256
2257        if (stat->leaves)
2258                avdepth = stat->totdepth*100 / stat->leaves;
2259        else
2260                avdepth = 0;
2261
2262        seq_printf(seq, "\tAver depth:     %u.%02d\n",
2263                   avdepth / 100, avdepth % 100);
2264        seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
2265
2266        seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
2267        bytes = LEAF_SIZE * stat->leaves;
2268
2269        seq_printf(seq, "\tPrefixes:       %u\n", stat->prefixes);
2270        bytes += sizeof(struct fib_alias) * stat->prefixes;
2271
2272        seq_printf(seq, "\tInternal nodes: %u\n\t", stat->tnodes);
2273        bytes += TNODE_SIZE(0) * stat->tnodes;
2274
2275        max = MAX_STAT_DEPTH;
2276        while (max > 0 && stat->nodesizes[max-1] == 0)
2277                max--;
2278
2279        pointers = 0;
2280        for (i = 1; i < max; i++)
2281                if (stat->nodesizes[i] != 0) {
2282                        seq_printf(seq, "  %u: %u",  i, stat->nodesizes[i]);
2283                        pointers += (1<<i) * stat->nodesizes[i];
2284                }
2285        seq_putc(seq, '\n');
2286        seq_printf(seq, "\tPointers: %u\n", pointers);
2287
2288        bytes += sizeof(struct key_vector *) * pointers;
2289        seq_printf(seq, "Null ptrs: %u\n", stat->nullpointers);
2290        seq_printf(seq, "Total size: %u  kB\n", (bytes + 1023) / 1024);
2291}
2292
2293#ifdef CONFIG_IP_FIB_TRIE_STATS
2294static void trie_show_usage(struct seq_file *seq,
2295                            const struct trie_use_stats __percpu *stats)
2296{
2297        struct trie_use_stats s = { 0 };
2298        int cpu;
2299
2300        /* loop through all of the CPUs and gather up the stats */
2301        for_each_possible_cpu(cpu) {
2302                const struct trie_use_stats *pcpu = per_cpu_ptr(stats, cpu);
2303
2304                s.gets += pcpu->gets;
2305                s.backtrack += pcpu->backtrack;
2306                s.semantic_match_passed += pcpu->semantic_match_passed;
2307                s.semantic_match_miss += pcpu->semantic_match_miss;
2308                s.null_node_hit += pcpu->null_node_hit;
2309                s.resize_node_skipped += pcpu->resize_node_skipped;
2310        }
2311
2312        seq_printf(seq, "\nCounters:\n---------\n");
2313        seq_printf(seq, "gets = %u\n", s.gets);
2314        seq_printf(seq, "backtracks = %u\n", s.backtrack);
2315        seq_printf(seq, "semantic match passed = %u\n",
2316                   s.semantic_match_passed);
2317        seq_printf(seq, "semantic match miss = %u\n", s.semantic_match_miss);
2318        seq_printf(seq, "null node hit= %u\n", s.null_node_hit);
2319        seq_printf(seq, "skipped node resize = %u\n\n", s.resize_node_skipped);
2320}
2321#endif /*  CONFIG_IP_FIB_TRIE_STATS */
2322
2323static void fib_table_print(struct seq_file *seq, struct fib_table *tb)
2324{
2325        if (tb->tb_id == RT_TABLE_LOCAL)
2326                seq_puts(seq, "Local:\n");
2327        else if (tb->tb_id == RT_TABLE_MAIN)
2328                seq_puts(seq, "Main:\n");
2329        else
2330                seq_printf(seq, "Id %d:\n", tb->tb_id);
2331}
2332
2333
2334static int fib_triestat_seq_show(struct seq_file *seq, void *v)
2335{
2336        struct net *net = (struct net *)seq->private;
2337        unsigned int h;
2338
2339        seq_printf(seq,
2340                   "Basic info: size of leaf:"
2341                   " %zd bytes, size of tnode: %zd bytes.\n",
2342                   LEAF_SIZE, TNODE_SIZE(0));
2343
2344        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2345                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2346                struct fib_table *tb;
2347
2348                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2349                        struct trie *t = (struct trie *) tb->tb_data;
2350                        struct trie_stat stat;
2351
2352                        if (!t)
2353                                continue;
2354
2355                        fib_table_print(seq, tb);
2356
2357                        trie_collect_stats(t, &stat);
2358                        trie_show_stats(seq, &stat);
2359#ifdef CONFIG_IP_FIB_TRIE_STATS
2360                        trie_show_usage(seq, t->stats);
2361#endif
2362                }
2363        }
2364
2365        return 0;
2366}
2367
2368static struct key_vector *fib_trie_get_idx(struct seq_file *seq, loff_t pos)
2369{
2370        struct fib_trie_iter *iter = seq->private;
2371        struct net *net = seq_file_net(seq);
2372        loff_t idx = 0;
2373        unsigned int h;
2374
2375        for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
2376                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2377                struct fib_table *tb;
2378
2379                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2380                        struct key_vector *n;
2381
2382                        for (n = fib_trie_get_first(iter,
2383                                                    (struct trie *) tb->tb_data);
2384                             n; n = fib_trie_get_next(iter))
2385                                if (pos == idx++) {
2386                                        iter->tb = tb;
2387                                        return n;
2388                                }
2389                }
2390        }
2391
2392        return NULL;
2393}
2394
2395static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
2396        __acquires(RCU)
2397{
2398        rcu_read_lock();
2399        return fib_trie_get_idx(seq, *pos);
2400}
2401
2402static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2403{
2404        struct fib_trie_iter *iter = seq->private;
2405        struct net *net = seq_file_net(seq);
2406        struct fib_table *tb = iter->tb;
2407        struct hlist_node *tb_node;
2408        unsigned int h;
2409        struct key_vector *n;
2410
2411        ++*pos;
2412        /* next node in same table */
2413        n = fib_trie_get_next(iter);
2414        if (n)
2415                return n;
2416
2417        /* walk rest of this hash chain */
2418        h = tb->tb_id & (FIB_TABLE_HASHSZ - 1);
2419        while ((tb_node = rcu_dereference(hlist_next_rcu(&tb->tb_hlist)))) {
2420                tb = hlist_entry(tb_node, struct fib_table, tb_hlist);
2421                n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2422                if (n)
2423                        goto found;
2424        }
2425
2426        /* new hash chain */
2427        while (++h < FIB_TABLE_HASHSZ) {
2428                struct hlist_head *head = &net->ipv4.fib_table_hash[h];
2429                hlist_for_each_entry_rcu(tb, head, tb_hlist) {
2430                        n = fib_trie_get_first(iter, (struct trie *) tb->tb_data);
2431                        if (n)
2432                                goto found;
2433                }
2434        }
2435        return NULL;
2436
2437found:
2438        iter->tb = tb;
2439        return n;
2440}
2441
2442static void fib_trie_seq_stop(struct seq_file *seq, void *v)
2443        __releases(RCU)
2444{
2445        rcu_read_unlock();
2446}
2447
2448static void seq_indent(struct seq_file *seq, int n)
2449{
2450        while (n-- > 0)
2451                seq_puts(seq, "   ");
2452}
2453
2454static inline const char *rtn_scope(char *buf, size_t len, enum rt_scope_t s)
2455{
2456        switch (s) {
2457        case RT_SCOPE_UNIVERSE: return "universe";
2458        case RT_SCOPE_SITE:     return "site";
2459        case RT_SCOPE_LINK:     return "link";
2460        case RT_SCOPE_HOST:     return "host";
2461        case RT_SCOPE_NOWHERE:  return "nowhere";
2462        default:
2463                snprintf(buf, len, "scope=%d", s);
2464                return buf;
2465        }
2466}
2467
2468static const char *const rtn_type_names[__RTN_MAX] = {
2469        [RTN_UNSPEC] = "UNSPEC",
2470        [RTN_UNICAST] = "UNICAST",
2471        [RTN_LOCAL] = "LOCAL",
2472        [RTN_BROADCAST] = "BROADCAST",
2473        [RTN_ANYCAST] = "ANYCAST",
2474        [RTN_MULTICAST] = "MULTICAST",
2475        [RTN_BLACKHOLE] = "BLACKHOLE",
2476        [RTN_UNREACHABLE] = "UNREACHABLE",
2477        [RTN_PROHIBIT] = "PROHIBIT",
2478        [RTN_THROW] = "THROW",
2479        [RTN_NAT] = "NAT",
2480        [RTN_XRESOLVE] = "XRESOLVE",
2481};
2482
2483static inline const char *rtn_type(char *buf, size_t len, unsigned int t)
2484{
2485        if (t < __RTN_MAX && rtn_type_names[t])
2486                return rtn_type_names[t];
2487        snprintf(buf, len, "type %u", t);
2488        return buf;
2489}
2490
2491/* Pretty print the trie */
2492static int fib_trie_seq_show(struct seq_file *seq, void *v)
2493{
2494        const struct fib_trie_iter *iter = seq->private;
2495        struct key_vector *n = v;
2496
2497        if (IS_TRIE(node_parent_rcu(n)))
2498                fib_table_print(seq, iter->tb);
2499
2500        if (IS_TNODE(n)) {
2501                __be32 prf = htonl(n->key);
2502
2503                seq_indent(seq, iter->depth-1);
2504                seq_printf(seq, "  +-- %pI4/%zu %u %u %u\n",
2505                           &prf, KEYLENGTH - n->pos - n->bits, n->bits,
2506                           tn_info(n)->full_children,
2507                           tn_info(n)->empty_children);
2508        } else {
2509                __be32 val = htonl(n->key);
2510                struct fib_alias *fa;
2511
2512                seq_indent(seq, iter->depth);
2513                seq_printf(seq, "  |-- %pI4\n", &val);
2514
2515                hlist_for_each_entry_rcu(fa, &n->leaf, fa_list) {
2516                        char buf1[32], buf2[32];
2517
2518                        seq_indent(seq, iter->depth + 1);
2519                        seq_printf(seq, "  /%zu %s %s",
2520                                   KEYLENGTH - fa->fa_slen,
2521                                   rtn_scope(buf1, sizeof(buf1),
2522                                             fa->fa_info->fib_scope),
2523                                   rtn_type(buf2, sizeof(buf2),
2524                                            fa->fa_type));
2525                        if (fa->fa_tos)
2526                                seq_printf(seq, " tos=%d", fa->fa_tos);
2527                        seq_putc(seq, '\n');
2528                }
2529        }
2530
2531        return 0;
2532}
2533
2534static const struct seq_operations fib_trie_seq_ops = {
2535        .start  = fib_trie_seq_start,
2536        .next   = fib_trie_seq_next,
2537        .stop   = fib_trie_seq_stop,
2538        .show   = fib_trie_seq_show,
2539};
2540
2541struct fib_route_iter {
2542        struct seq_net_private p;
2543        struct fib_table *main_tb;
2544        struct key_vector *tnode;
2545        loff_t  pos;
2546        t_key   key;
2547};
2548
2549static struct key_vector *fib_route_get_idx(struct fib_route_iter *iter,
2550                                            loff_t pos)
2551{
2552        struct key_vector *l, **tp = &iter->tnode;
2553        t_key key;
2554
2555        /* use cached location of previously found key */
2556        if (iter->pos > 0 && pos >= iter->pos) {
2557                key = iter->key;
2558        } else {
2559                iter->pos = 1;
2560                key = 0;
2561        }
2562
2563        pos -= iter->pos;
2564
2565        while ((l = leaf_walk_rcu(tp, key)) && (pos-- > 0)) {
2566                key = l->key + 1;
2567                iter->pos++;
2568                l = NULL;
2569
2570                /* handle unlikely case of a key wrap */
2571                if (!key)
2572                        break;
2573        }
2574
2575        if (l)
2576                iter->key = l->key;     /* remember it */
2577        else
2578                iter->pos = 0;          /* forget it */
2579
2580        return l;
2581}
2582
2583static void *fib_route_seq_start(struct seq_file *seq, loff_t *pos)
2584        __acquires(RCU)
2585{
2586        struct fib_route_iter *iter = seq->private;
2587        struct fib_table *tb;
2588        struct trie *t;
2589
2590        rcu_read_lock();
2591
2592        tb = fib_get_table(seq_file_net(seq), RT_TABLE_MAIN);
2593        if (!tb)
2594                return NULL;
2595
2596        iter->main_tb = tb;
2597        t = (struct trie *)tb->tb_data;
2598        iter->tnode = t->kv;
2599
2600        if (*pos != 0)
2601                return fib_route_get_idx(iter, *pos);
2602
2603        iter->pos = 0;
2604        iter->key = KEY_MAX;
2605
2606        return SEQ_START_TOKEN;
2607}
2608
2609static void *fib_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2610{
2611        struct fib_route_iter *iter = seq->private;
2612        struct key_vector *l = NULL;
2613        t_key key = iter->key + 1;
2614
2615        ++*pos;
2616
2617        /* only allow key of 0 for start of sequence */
2618        if ((v == SEQ_START_TOKEN) || key)
2619                l = leaf_walk_rcu(&iter->tnode, key);
2620
2621        if (l) {
2622                iter->key = l->key;
2623                iter->pos++;
2624        } else {
2625                iter->pos = 0;
2626        }
2627
2628        return l;
2629}
2630
2631static void fib_route_seq_stop(struct seq_file *seq, void *v)
2632        __releases(RCU)
2633{
2634        rcu_read_unlock();
2635}
2636
2637static unsigned int fib_flag_trans(int type, __be32 mask, const struct fib_info *fi)
2638{
2639        unsigned int flags = 0;
2640
2641        if (type == RTN_UNREACHABLE || type == RTN_PROHIBIT)
2642                flags = RTF_REJECT;
2643        if (fi && fi->fib_nh->fib_nh_gw4)
2644                flags |= RTF_GATEWAY;
2645        if (mask == htonl(0xFFFFFFFF))
2646                flags |= RTF_HOST;
2647        flags |= RTF_UP;
2648        return flags;
2649}
2650
2651/*
2652 *      This outputs /proc/net/route.
2653 *      The format of the file is not supposed to be changed
2654 *      and needs to be same as fib_hash output to avoid breaking
2655 *      legacy utilities
2656 */
2657static int fib_route_seq_show(struct seq_file *seq, void *v)
2658{
2659        struct fib_route_iter *iter = seq->private;
2660        struct fib_table *tb = iter->main_tb;
2661        struct fib_alias *fa;
2662        struct key_vector *l = v;
2663        __be32 prefix;
2664
2665        if (v == SEQ_START_TOKEN) {
2666                seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
2667                           "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
2668                           "\tWindow\tIRTT");
2669                return 0;
2670        }
2671
2672        prefix = htonl(l->key);
2673
2674        hlist_for_each_entry_rcu(fa, &l->leaf, fa_list) {
2675                const struct fib_info *fi = fa->fa_info;
2676                __be32 mask = inet_make_mask(KEYLENGTH - fa->fa_slen);
2677                unsigned int flags = fib_flag_trans(fa->fa_type, mask, fi);
2678
2679                if ((fa->fa_type == RTN_BROADCAST) ||
2680                    (fa->fa_type == RTN_MULTICAST))
2681                        continue;
2682
2683                if (fa->tb_id != tb->tb_id)
2684                        continue;
2685
2686                seq_setwidth(seq, 127);
2687
2688                if (fi)
2689                        seq_printf(seq,
2690                                   "%s\t%08X\t%08X\t%04X\t%d\t%u\t"
2691                                   "%d\t%08X\t%d\t%u\t%u",
2692                                   fi->fib_dev ? fi->fib_dev->name : "*",
2693                                   prefix,
2694                                   fi->fib_nh->fib_nh_gw4, flags, 0, 0,
2695                                   fi->fib_priority,
2696                                   mask,
2697                                   (fi->fib_advmss ?
2698                                    fi->fib_advmss + 40 : 0),
2699                                   fi->fib_window,
2700                                   fi->fib_rtt >> 3);
2701                else
2702                        seq_printf(seq,
2703                                   "*\t%08X\t%08X\t%04X\t%d\t%u\t"
2704                                   "%d\t%08X\t%d\t%u\t%u",
2705                                   prefix, 0, flags, 0, 0, 0,
2706                                   mask, 0, 0, 0);
2707
2708                seq_pad(seq, '\n');
2709        }
2710
2711        return 0;
2712}
2713
2714static const struct seq_operations fib_route_seq_ops = {
2715        .start  = fib_route_seq_start,
2716        .next   = fib_route_seq_next,
2717        .stop   = fib_route_seq_stop,
2718        .show   = fib_route_seq_show,
2719};
2720
2721int __net_init fib_proc_init(struct net *net)
2722{
2723        if (!proc_create_net("fib_trie", 0444, net->proc_net, &fib_trie_seq_ops,
2724                        sizeof(struct fib_trie_iter)))
2725                goto out1;
2726
2727        if (!proc_create_net_single("fib_triestat", 0444, net->proc_net,
2728                        fib_triestat_seq_show, NULL))
2729                goto out2;
2730
2731        if (!proc_create_net("route", 0444, net->proc_net, &fib_route_seq_ops,
2732                        sizeof(struct fib_route_iter)))
2733                goto out3;
2734
2735        return 0;
2736
2737out3:
2738        remove_proc_entry("fib_triestat", net->proc_net);
2739out2:
2740        remove_proc_entry("fib_trie", net->proc_net);
2741out1:
2742        return -ENOMEM;
2743}
2744
2745void __net_exit fib_proc_exit(struct net *net)
2746{
2747        remove_proc_entry("fib_trie", net->proc_net);
2748        remove_proc_entry("fib_triestat", net->proc_net);
2749        remove_proc_entry("route", net->proc_net);
2750}
2751
2752#endif /* CONFIG_PROC_FS */
2753