lib/libc/db/hash/hash.c

   1 /*      $NetBSD: hash.c,v 1.31 2009/02/12 06:35:54 lukem Exp $  */
   2
   3 /*-
   4  * Copyright (c) 1990, 1993, 1994
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * This code is derived from software contributed to Berkeley by
   8  * Margo Seltzer.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. Neither the name of the University nor the names of its contributors
  19  *    may be used to endorse or promote products derived from this software
  20  *    without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34
  35 #if HAVE_NBTOOL_CONFIG_H
  36 #include "nbtool_config.h"
  37 #endif
  38
  39 #include <sys/cdefs.h>
  40 __RCSID("$NetBSD: hash.c,v 1.31 2009/02/12 06:35:54 lukem Exp $");
  41
  42 #include "namespace.h"
  43 #include <sys/param.h>
  44 #include <sys/stat.h>
  45
  46 #include <errno.h>
  47 #include <fcntl.h>
  48 #include <stdio.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <unistd.h>
  52 #include <assert.h>
  53
  54 #include <db.h>
  55 #include "hash.h"
  56 #include "page.h"
  57 #include "extern.h"
  58
  59 static int   alloc_segs(HTAB *, int);
  60 static int   flush_meta(HTAB *);
  61 static int   hash_access(HTAB *, ACTION, DBT *, DBT *);
  62 static int   hash_close(DB *);
  63 static int   hash_delete(const DB *, const DBT *, uint32_t);
  64 static int   hash_fd(const DB *);
  65 static int   hash_get(const DB *, const DBT *, DBT *, uint32_t);
  66 static int   hash_put(const DB *, DBT *, const DBT *, uint32_t);
  67 static void *hash_realloc(SEGMENT **, size_t, size_t);
  68 static int   hash_seq(const DB *, DBT *, DBT *, uint32_t);
  69 static int   hash_sync(const DB *, uint32_t);
  70 static int   hdestroy(HTAB *);
  71 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
  72 static int   init_htab(HTAB *, size_t);
  73 #if BYTE_ORDER == LITTLE_ENDIAN
  74 static void  swap_header(HTAB *);
  75 static void  swap_header_copy(HASHHDR *, HASHHDR *);
  76 #endif
  77
  78 /* Fast arithmetic, relying on powers of 2, */
  79 #define MOD(x, y)               ((x) & ((y) - 1))
  80
  81 #define RETURN_ERROR(ERR, LOC)  { save_errno = ERR; goto LOC; }
  82
  83 /* Return values */
  84 #define SUCCESS  (0)
  85 #define ERROR   (-1)
  86 #define ABNORMAL (1)
  87
  88 #ifdef HASH_STATISTICS
  89 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
  90 #endif
  91
  92 /************************** INTERFACE ROUTINES ***************************/
  93 /* OPEN/CLOSE */
  94
  95 /* ARGSUSED */
  96 DB *
  97 __hash_open(const char *file, int flags, mode_t mode, const HASHINFO *info,
  98     int dflags)
  99 {
 100         HTAB *hashp;
 101         struct stat statbuf;
 102         DB *dbp;
 103         int bpages, new_table, nsegs, save_errno;
 104         ssize_t hdrsize;
 105
 106         if ((flags & O_ACCMODE) == O_WRONLY) {
 107                 errno = EINVAL;
 108                 return (NULL);
 109         }
 110
 111         if (!(hashp = calloc(1, sizeof(HTAB))))
 112                 return (NULL);
 113         hashp->fp = -1;
 114
 115         /*
 116          * Even if user wants write only, we need to be able to read
 117          * the actual file, so we need to open it read/write. But, the
 118          * field in the hashp structure needs to be accurate so that
 119          * we can check accesses.
 120          */
 121         hashp->flags = flags;
 122
 123         new_table = 0;
 124         if (!file || (flags & O_TRUNC) ||
 125             (stat(file, &statbuf) && (errno == ENOENT))) {
 126                 if (errno == ENOENT)
 127                         errno = 0; /* Just in case someone looks at errno */
 128                 new_table = 1;
 129         }
 130         if (file) {
 131                 if ((hashp->fp = open(file, flags, mode)) == -1)
 132                         RETURN_ERROR(errno, error0);
 133                 if (fcntl(hashp->fp, F_SETFD, FD_CLOEXEC) == -1)
 134                         RETURN_ERROR(errno, error1);
 135                 if (fstat(hashp->fp, &statbuf) == -1)
 136                         RETURN_ERROR(errno, error1);
 137                 new_table |= statbuf.st_size == 0;
 138         }
 139         if (new_table) {
 140                 if (!(hashp = init_hash(hashp, file, info)))
 141                         RETURN_ERROR(errno, error1);
 142         } else {
 143                 /* Table already exists */
 144                 if (info && info->hash)
 145                         hashp->hash = info->hash;
 146                 else
 147                         hashp->hash = __default_hash;
 148
 149                 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
 150 #if BYTE_ORDER == LITTLE_ENDIAN
 151                 swap_header(hashp);
 152 #endif
 153                 if (hdrsize == -1)
 154                         RETURN_ERROR(errno, error1);
 155                 if (hdrsize != sizeof(HASHHDR))
 156                         RETURN_ERROR(EFTYPE, error1);
 157                 /* Verify file type, versions and hash function */
 158                 if (hashp->MAGIC != HASHMAGIC)
 159                         RETURN_ERROR(EFTYPE, error1);
 160 #define OLDHASHVERSION  1
 161                 if (hashp->VERSION != HASHVERSION &&
 162                     hashp->VERSION != OLDHASHVERSION)
 163                         RETURN_ERROR(EFTYPE, error1);
 164                 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) !=
 165                     (uint32_t)hashp->H_CHARKEY)
 166                         RETURN_ERROR(EFTYPE, error1);
 167                 /*
 168                  * Figure out how many segments we need.  Max_Bucket is the
 169                  * maximum bucket number, so the number of buckets is
 170                  * max_bucket + 1.
 171                  */
 172                 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
 173                          hashp->SGSIZE;
 174                 hashp->nsegs = 0;
 175                 if (alloc_segs(hashp, nsegs))
 176                         /*
 177                          * If alloc_segs fails, table will have been destroyed
 178                          * and errno will have been set.
 179                          */
 180                         return (NULL);
 181                 /* Read in bitmaps */
 182                 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
 183                     (unsigned int)(hashp->BSIZE << BYTE_SHIFT) - 1) >>
 184                     (hashp->BSHIFT + BYTE_SHIFT);
 185
 186                 hashp->nmaps = bpages;
 187                 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(uint32_t *));
 188         }
 189
 190         /* Initialize Buffer Manager */
 191         if (info && info->cachesize)
 192                 __buf_init(hashp, info->cachesize);
 193         else
 194                 __buf_init(hashp, DEF_BUFSIZE);
 195
 196         hashp->new_file = new_table;
 197         hashp->save_file = file && (hashp->flags & O_RDWR);
 198         hashp->cbucket = -1;
 199         if (!(dbp = malloc(sizeof(DB)))) {
 200                 save_errno = errno;
 201                 hdestroy(hashp);
 202                 errno = save_errno;
 203                 return (NULL);
 204         }
 205         dbp->internal = hashp;
 206         dbp->close = hash_close;
 207         dbp->del = hash_delete;
 208         dbp->fd = hash_fd;
 209         dbp->get = hash_get;
 210         dbp->put = hash_put;
 211         dbp->seq = hash_seq;
 212         dbp->sync = hash_sync;
 213         dbp->type = DB_HASH;
 214
 215 #ifdef DEBUG
 216         (void)fprintf(stderr,
 217 "%s\n%s%p\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%x\n%s%x\n%s%d\n%s%d\n",
 218             "init_htab:",
 219             "TABLE POINTER   ", hashp,
 220             "BUCKET SIZE     ", hashp->BSIZE,
 221             "BUCKET SHIFT    ", hashp->BSHIFT,
 222             "DIRECTORY SIZE  ", hashp->DSIZE,
 223             "SEGMENT SIZE    ", hashp->SGSIZE,
 224             "SEGMENT SHIFT   ", hashp->SSHIFT,
 225             "FILL FACTOR     ", hashp->FFACTOR,
 226             "MAX BUCKET      ", hashp->MAX_BUCKET,
 227             "OVFL POINT      ", hashp->OVFL_POINT,
 228             "LAST FREED      ", hashp->LAST_FREED,
 229             "HIGH MASK       ", hashp->HIGH_MASK,
 230             "LOW  MASK       ", hashp->LOW_MASK,
 231             "NSEGS           ", hashp->nsegs,
 232             "NKEYS           ", hashp->NKEYS);
 233 #endif
 234 #ifdef HASH_STATISTICS
 235         hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
 236 #endif
 237         return (dbp);
 238
 239 error1:
 240         if (hashp != NULL)
 241                 (void)close(hashp->fp);
 242
 243 error0:
 244         free(hashp);
 245         errno = save_errno;
 246         return (NULL);
 247 }
 248
 249 static int
 250 hash_close(DB *dbp)
 251 {
 252         HTAB *hashp;
 253         int retval;
 254
 255         if (!dbp)
 256                 return (ERROR);
 257
 258         hashp = dbp->internal;
 259         retval = hdestroy(hashp);
 260         free(dbp);
 261         return (retval);
 262 }
 263
 264 static int
 265 hash_fd(const DB *dbp)
 266 {
 267         HTAB *hashp;
 268
 269         if (!dbp)
 270                 return (ERROR);
 271
 272         hashp = dbp->internal;
 273         if (hashp->fp == -1) {
 274                 errno = ENOENT;
 275                 return (-1);
 276         }
 277         return (hashp->fp);
 278 }
 279
 280 /************************** LOCAL CREATION ROUTINES **********************/
 281 static HTAB *
 282 init_hash(HTAB *hashp, const char *file, const HASHINFO *info)
 283 {
 284         struct stat statbuf;
 285         int nelem;
 286
 287         nelem = 1;
 288         hashp->NKEYS = 0;
 289         hashp->LORDER = BYTE_ORDER;
 290         hashp->BSIZE = DEF_BUCKET_SIZE;
 291         hashp->BSHIFT = DEF_BUCKET_SHIFT;
 292         hashp->SGSIZE = DEF_SEGSIZE;
 293         hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
 294         hashp->DSIZE = DEF_DIRSIZE;
 295         hashp->FFACTOR = DEF_FFACTOR;
 296         hashp->hash = __default_hash;
 297         memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
 298         memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
 299
 300         /* Fix bucket size to be optimal for file system */
 301         if (file != NULL) {
 302                 if (stat(file, &statbuf))
 303                         return (NULL);
 304 #ifdef __minix
 305                 if (statbuf.st_blksize == 0) {
 306                         /* 0 in 2 cases: upgrade from old to new struct stat or
 307                          * there is a bug in underlying fs.
 308                          */
 309                         hashp->BSIZE = MIN(MINIX_ST_BLKSIZE, MAX_BSIZE);
 310                 } else
 311 #endif
 312                         hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
 313
 314                 hashp->BSHIFT = __log2((uint32_t)hashp->BSIZE);
 315         }
 316
 317         if (info) {
 318                 if (info->bsize) {
 319                         /* Round pagesize up to power of 2 */
 320                         hashp->BSHIFT = __log2(info->bsize);
 321                         hashp->BSIZE = 1 << hashp->BSHIFT;
 322                         if (hashp->BSIZE > MAX_BSIZE) {
 323                                 errno = EINVAL;
 324                                 return (NULL);
 325                         }
 326                 }
 327                 if (info->ffactor)
 328                         hashp->FFACTOR = info->ffactor;
 329                 if (info->hash)
 330                         hashp->hash = info->hash;
 331                 if (info->nelem)
 332                         nelem = info->nelem;
 333                 if (info->lorder) {
 334                         if (info->lorder != BIG_ENDIAN &&
 335                             info->lorder != LITTLE_ENDIAN) {
 336                                 errno = EINVAL;
 337                                 return (NULL);
 338                         }
 339                         hashp->LORDER = info->lorder;
 340                 }
 341         }
 342         /* init_htab should destroy the table and set errno if it fails */
 343         if (init_htab(hashp, (size_t)nelem))
 344                 return (NULL);
 345         else
 346                 return (hashp);
 347 }
 348 /*
 349  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
 350  * the table and set errno, so we just pass the error information along.
 351  *
 352  * Returns 0 on No Error
 353  */
 354 static int
 355 init_htab(HTAB *hashp, size_t nelem)
 356 {
 357         int nbuckets;
 358         uint32_t nsegs;
 359         int l2;
 360
 361         /*
 362          * Divide number of elements by the fill factor and determine a
 363          * desired number of buckets.  Allocate space for the next greater
 364          * power of two number of buckets.
 365          */
 366         nelem = (nelem - 1) / hashp->FFACTOR + 1;
 367
 368         _DBFIT(nelem, uint32_t);
 369         l2 = __log2(MAX((uint32_t)nelem, 2));
 370         nbuckets = 1 << l2;
 371
 372         hashp->SPARES[l2] = l2 + 1;
 373         hashp->SPARES[l2 + 1] = l2 + 1;
 374         hashp->OVFL_POINT = l2;
 375         hashp->LAST_FREED = 2;
 376
 377         /* First bitmap page is at: splitpoint l2 page offset 1 */
 378         if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0))
 379                 return (-1);
 380
 381         hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
 382         hashp->HIGH_MASK = (nbuckets << 1) - 1;
 383         /* LINTED constant in conditional context */
 384         hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
 385             hashp->BSHIFT) + 1;
 386
 387         nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
 388         nsegs = 1 << __log2(nsegs);
 389
 390         if (nsegs > (uint32_t)hashp->DSIZE)
 391                 hashp->DSIZE = nsegs;
 392         return (alloc_segs(hashp, (int)nsegs));
 393 }
 394
 395 /********************** DESTROY/CLOSE ROUTINES ************************/
 396
 397 /*
 398  * Flushes any changes to the file if necessary and destroys the hashp
 399  * structure, freeing all allocated space.
 400  */
 401 static int
 402 hdestroy(HTAB *hashp)
 403 {
 404         int i, save_errno;
 405
 406         save_errno = 0;
 407
 408 #ifdef HASH_STATISTICS
 409         (void)fprintf(stderr, "hdestroy: accesses %d collisions %d\n",
 410             hash_accesses, hash_collisions);
 411         (void)fprintf(stderr, "hdestroy: expansions %d\n",
 412             hash_expansions);
 413         (void)fprintf(stderr, "hdestroy: overflows %d\n",
 414             hash_overflows);
 415         (void)fprintf(stderr, "keys %d maxp %d segmentcount %d\n",
 416             hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
 417
 418         for (i = 0; i < NCACHED; i++)
 419                 (void)fprintf(stderr,
 420                     "spares[%d] = %d\n", i, hashp->SPARES[i]);
 421 #endif
 422         /*
 423          * Call on buffer manager to free buffers, and if required,
 424          * write them to disk.
 425          */
 426         if (__buf_free(hashp, 1, hashp->save_file))
 427                 save_errno = errno;
 428         if (hashp->dir) {
 429                 free(*hashp->dir);      /* Free initial segments */
 430                 /* Free extra segments */
 431                 while (hashp->exsegs--)
 432                         free(hashp->dir[--hashp->nsegs]);
 433                 free(hashp->dir);
 434         }
 435         if (flush_meta(hashp) && !save_errno)
 436                 save_errno = errno;
 437         /* Free Bigmaps */
 438         for (i = 0; i < hashp->nmaps; i++)
 439                 if (hashp->mapp[i])
 440                         free(hashp->mapp[i]);
 441
 442         if (hashp->fp != -1)
 443                 (void)close(hashp->fp);
 444
 445         free(hashp);
 446
 447         if (save_errno) {
 448                 errno = save_errno;
 449                 return (ERROR);
 450         }
 451         return (SUCCESS);
 452 }
 453 /*
 454  * Write modified pages to disk
 455  *
 456  * Returns:
 457  *       0 == OK
 458  *      -1 ERROR
 459  */
 460 static int
 461 hash_sync(const DB *dbp, uint32_t flags)
 462 {
 463         HTAB *hashp;
 464
 465         if (flags != 0) {
 466                 errno = EINVAL;
 467                 return (ERROR);
 468         }
 469
 470         if (!dbp)
 471                 return (ERROR);
 472
 473         hashp = dbp->internal;
 474         if (!hashp->save_file)
 475                 return (0);
 476         if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
 477                 return (ERROR);
 478         hashp->new_file = 0;
 479         return (0);
 480 }
 481
 482 /*
 483  * Returns:
 484  *       0 == OK
 485  *      -1 indicates that errno should be set
 486  */
 487 static int
 488 flush_meta(HTAB *hashp)
 489 {
 490         HASHHDR *whdrp;
 491 #if BYTE_ORDER == LITTLE_ENDIAN
 492         HASHHDR whdr;
 493 #endif
 494         int fp, i;
 495         ssize_t wsize;
 496
 497         if (!hashp->save_file)
 498                 return (0);
 499         hashp->MAGIC = HASHMAGIC;
 500         hashp->VERSION = HASHVERSION;
 501         hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
 502
 503         fp = hashp->fp;
 504         whdrp = &hashp->hdr;
 505 #if BYTE_ORDER == LITTLE_ENDIAN
 506         whdrp = &whdr;
 507         swap_header_copy(&hashp->hdr, whdrp);
 508 #endif
 509         if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
 510                 return (-1);
 511         else
 512                 if (wsize != sizeof(HASHHDR)) {
 513                         errno = EFTYPE;
 514                         hashp->err = errno;
 515                         return (-1);
 516                 }
 517         for (i = 0; i < NCACHED; i++)
 518                 if (hashp->mapp[i])
 519                         if (__put_page(hashp, (char *)(void *)hashp->mapp[i],
 520                                 (u_int)hashp->BITMAPS[i], 0, 1))
 521                                 return (-1);
 522         return (0);
 523 }
 524
 525 /*******************************SEARCH ROUTINES *****************************/
 526 /*
 527  * All the access routines return
 528  *
 529  * Returns:
 530  *       0 on SUCCESS
 531  *       1 to indicate an external ERROR (i.e. key not found, etc)
 532  *      -1 to indicate an internal ERROR (i.e. out of memory, etc)
 533  */
 534 static int
 535 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag)
 536 {
 537         HTAB *hashp;
 538
 539         hashp = dbp->internal;
 540         if (flag) {
 541                 hashp->err = errno = EINVAL;
 542                 return (ERROR);
 543         }
 544         return (hash_access(hashp, HASH_GET, __UNCONST(key), data));
 545 }
 546
 547 static int
 548 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag)
 549 {
 550         HTAB *hashp;
 551
 552         hashp = dbp->internal;
 553         if (flag && flag != R_NOOVERWRITE) {
 554                 hashp->err = errno = EINVAL;
 555                 return (ERROR);
 556         }
 557         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 558                 hashp->err = errno = EPERM;
 559                 return (ERROR);
 560         }
 561         /* LINTED const castaway */
 562         return (hash_access(hashp, flag == R_NOOVERWRITE ?
 563             HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data)));
 564 }
 565
 566 static int
 567 hash_delete(const DB *dbp, const DBT *key, uint32_t flag)
 568 {
 569         HTAB *hashp;
 570
 571         hashp = dbp->internal;
 572         if (flag && flag != R_CURSOR) {
 573                 hashp->err = errno = EINVAL;
 574                 return (ERROR);
 575         }
 576         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 577                 hashp->err = errno = EPERM;
 578                 return (ERROR);
 579         }
 580         return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL);
 581 }
 582
 583 /*
 584  * Assume that hashp has been set in wrapper routine.
 585  */
 586 static int
 587 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
 588 {
 589         BUFHEAD *rbufp;
 590         BUFHEAD *bufp, *save_bufp;
 591         uint16_t *bp;
 592         int n, ndx, off;
 593         size_t size;
 594         char *kp;
 595         uint16_t pageno;
 596
 597 #ifdef HASH_STATISTICS
 598         hash_accesses++;
 599 #endif
 600
 601         off = hashp->BSIZE;
 602         size = key->size;
 603         kp = (char *)key->data;
 604         rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0);
 605         if (!rbufp)
 606                 return (ERROR);
 607         save_bufp = rbufp;
 608
 609         /* Pin the bucket chain */
 610         rbufp->flags |= BUF_PIN;
 611         for (bp = (uint16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
 612                 if (bp[1] >= REAL_KEY) {
 613                         /* Real key/data pair */
 614                         if (size == (size_t)(off - *bp) &&
 615                             memcmp(kp, rbufp->page + *bp, size) == 0)
 616                                 goto found;
 617                         off = bp[1];
 618 #ifdef HASH_STATISTICS
 619                         hash_collisions++;
 620 #endif
 621                         bp += 2;
 622                         ndx += 2;
 623                 } else if (bp[1] == OVFLPAGE) {
 624                         rbufp = __get_buf(hashp, (uint32_t)*bp, rbufp, 0);
 625                         if (!rbufp) {
 626                                 save_bufp->flags &= ~BUF_PIN;
 627                                 return (ERROR);
 628                         }
 629                         /* FOR LOOP INIT */
 630                         bp = (uint16_t *)(void *)rbufp->page;
 631                         n = *bp++;
 632                         ndx = 1;
 633                         off = hashp->BSIZE;
 634                 } else if (bp[1] < REAL_KEY) {
 635                         if ((ndx =
 636                             __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0)
 637                                 goto found;
 638                         if (ndx == -2) {
 639                                 bufp = rbufp;
 640                                 if (!(pageno =
 641                                     __find_last_page(hashp, &bufp))) {
 642                                         ndx = 0;
 643                                         rbufp = bufp;
 644                                         break;  /* FOR */
 645                                 }
 646                                 rbufp = __get_buf(hashp, (uint32_t)pageno,
 647                                     bufp, 0);
 648                                 if (!rbufp) {
 649                                         save_bufp->flags &= ~BUF_PIN;
 650                                         return (ERROR);
 651                                 }
 652                                 /* FOR LOOP INIT */
 653                                 bp = (uint16_t *)(void *)rbufp->page;
 654                                 n = *bp++;
 655                                 ndx = 1;
 656                                 off = hashp->BSIZE;
 657                         } else {
 658                                 save_bufp->flags &= ~BUF_PIN;
 659                                 return (ERROR);
 660                         }
 661                 }
 662
 663         /* Not found */
 664         switch (action) {
 665         case HASH_PUT:
 666         case HASH_PUTNEW:
 667                 if (__addel(hashp, rbufp, key, val)) {
 668                         save_bufp->flags &= ~BUF_PIN;
 669                         return (ERROR);
 670                 } else {
 671                         save_bufp->flags &= ~BUF_PIN;
 672                         return (SUCCESS);
 673                 }
 674         case HASH_GET:
 675         case HASH_DELETE:
 676         default:
 677                 save_bufp->flags &= ~BUF_PIN;
 678                 return (ABNORMAL);
 679         }
 680
 681 found:
 682         switch (action) {
 683         case HASH_PUTNEW:
 684                 save_bufp->flags &= ~BUF_PIN;
 685                 return (ABNORMAL);
 686         case HASH_GET:
 687                 bp = (uint16_t *)(void *)rbufp->page;
 688                 if (bp[ndx + 1] < REAL_KEY) {
 689                         if (__big_return(hashp, rbufp, ndx, val, 0))
 690                                 return (ERROR);
 691                 } else {
 692                         val->data = (uint8_t *)rbufp->page + (int)bp[ndx + 1];
 693                         val->size = bp[ndx] - bp[ndx + 1];
 694                 }
 695                 break;
 696         case HASH_PUT:
 697                 if ((__delpair(hashp, rbufp, ndx)) ||
 698                     (__addel(hashp, rbufp, key, val))) {
 699                         save_bufp->flags &= ~BUF_PIN;
 700                         return (ERROR);
 701                 }
 702                 break;
 703         case HASH_DELETE:
 704                 if (__delpair(hashp, rbufp, ndx))
 705                         return (ERROR);
 706                 break;
 707         default:
 708                 abort();
 709         }
 710         save_bufp->flags &= ~BUF_PIN;
 711         return (SUCCESS);
 712 }
 713
 714 static int
 715 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag)
 716 {
 717         uint32_t bucket;
 718         BUFHEAD *bufp = NULL; /* XXX: gcc */
 719         HTAB *hashp;
 720         uint16_t *bp, ndx;
 721
 722         hashp = dbp->internal;
 723         if (flag && flag != R_FIRST && flag != R_NEXT) {
 724                 hashp->err = errno = EINVAL;
 725                 return (ERROR);
 726         }
 727 #ifdef HASH_STATISTICS
 728         hash_accesses++;
 729 #endif
 730         if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
 731                 hashp->cbucket = 0;
 732                 hashp->cndx = 1;
 733                 hashp->cpage = NULL;
 734         }
 735
 736         for (bp = NULL; !bp || !bp[0]; ) {
 737                 if (!(bufp = hashp->cpage)) {
 738                         for (bucket = hashp->cbucket;
 739                             bucket <= (uint32_t)hashp->MAX_BUCKET;
 740                             bucket++, hashp->cndx = 1) {
 741                                 bufp = __get_buf(hashp, bucket, NULL, 0);
 742                                 if (!bufp)
 743                                         return (ERROR);
 744                                 hashp->cpage = bufp;
 745                                 bp = (uint16_t *)(void *)bufp->page;
 746                                 if (bp[0])
 747                                         break;
 748                         }
 749                         hashp->cbucket = bucket;
 750                         if (hashp->cbucket > hashp->MAX_BUCKET) {
 751                                 hashp->cbucket = -1;
 752                                 return (ABNORMAL);
 753                         }
 754                 } else
 755                         bp = (uint16_t *)(void *)hashp->cpage->page;
 756
 757                 _DIAGASSERT(bp != NULL);
 758                 _DIAGASSERT(bufp != NULL);
 759                 while (bp[hashp->cndx + 1] == OVFLPAGE) {
 760                         bufp = hashp->cpage =
 761                             __get_buf(hashp, (uint32_t)bp[hashp->cndx], bufp,
 762                                 0);
 763                         if (!bufp)
 764                                 return (ERROR);
 765                         bp = (uint16_t *)(void *)(bufp->page);
 766                         hashp->cndx = 1;
 767                 }
 768                 if (!bp[0]) {
 769                         hashp->cpage = NULL;
 770                         ++hashp->cbucket;
 771                 }
 772         }
 773         ndx = hashp->cndx;
 774         if (bp[ndx + 1] < REAL_KEY) {
 775                 if (__big_keydata(hashp, bufp, key, data, 1))
 776                         return (ERROR);
 777         } else {
 778                 if (hashp->cpage == NULL)
 779                         return (ERROR);
 780                 key->data = (uint8_t *)hashp->cpage->page + bp[ndx];
 781                 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
 782                 data->data = (uint8_t *)hashp->cpage->page + bp[ndx + 1];
 783                 data->size = bp[ndx] - bp[ndx + 1];
 784                 ndx += 2;
 785                 if (ndx > bp[0]) {
 786                         hashp->cpage = NULL;
 787                         hashp->cbucket++;
 788                         hashp->cndx = 1;
 789                 } else
 790                         hashp->cndx = ndx;
 791         }
 792         return (SUCCESS);
 793 }
 794
 795 /********************************* UTILITIES ************************/
 796
 797 /*
 798  * Returns:
 799  *       0 ==> OK
 800  *      -1 ==> Error
 801  */
 802 int
 803 __expand_table(HTAB *hashp)
 804 {
 805         uint32_t old_bucket, new_bucket;
 806         int new_segnum, spare_ndx;
 807         size_t dirsize;
 808
 809 #ifdef HASH_STATISTICS
 810         hash_expansions++;
 811 #endif
 812         new_bucket = ++hashp->MAX_BUCKET;
 813         old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
 814
 815         new_segnum = new_bucket >> hashp->SSHIFT;
 816
 817         /* Check if we need a new segment */
 818         if (new_segnum >= hashp->nsegs) {
 819                 /* Check if we need to expand directory */
 820                 if (new_segnum >= hashp->DSIZE) {
 821                         /* Reallocate directory */
 822                         dirsize = hashp->DSIZE * sizeof(SEGMENT *);
 823                         if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
 824                                 return (-1);
 825                         hashp->DSIZE = dirsize << 1;
 826                 }
 827                 if ((hashp->dir[new_segnum] =
 828                     calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
 829                         return (-1);
 830                 hashp->exsegs++;
 831                 hashp->nsegs++;
 832         }
 833         /*
 834          * If the split point is increasing (MAX_BUCKET's log base 2
 835          * * increases), we need to copy the current contents of the spare
 836          * split bucket to the next bucket.
 837          */
 838         spare_ndx = __log2((uint32_t)(hashp->MAX_BUCKET + 1));
 839         if (spare_ndx > hashp->OVFL_POINT) {
 840                 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
 841                 hashp->OVFL_POINT = spare_ndx;
 842         }
 843
 844         if (new_bucket > (uint32_t)hashp->HIGH_MASK) {
 845                 /* Starting a new doubling */
 846                 hashp->LOW_MASK = hashp->HIGH_MASK;
 847                 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
 848         }
 849         /* Relocate records to the new bucket */
 850         return (__split_page(hashp, old_bucket, new_bucket));
 851 }
 852
 853 /*
 854  * If realloc guarantees that the pointer is not destroyed if the realloc
 855  * fails, then this routine can go away.
 856  */
 857 static void *
 858 hash_realloc(SEGMENT **p_ptr, size_t oldsize, size_t newsize)
 859 {
 860         void *p;
 861
 862         if ((p = malloc(newsize)) != NULL) {
 863                 memmove(p, *p_ptr, oldsize);
 864                 memset((char *)p + oldsize, 0, newsize - oldsize);
 865                 free(*p_ptr);
 866                 *p_ptr = p;
 867         }
 868         return (p);
 869 }
 870
 871 uint32_t
 872 __call_hash(HTAB *hashp, char *k, int len)
 873 {
 874         int n, bucket;
 875
 876         n = hashp->hash(k, (size_t)len);
 877         bucket = n & hashp->HIGH_MASK;
 878         if (bucket > hashp->MAX_BUCKET)
 879                 bucket = bucket & hashp->LOW_MASK;
 880         return (bucket);
 881 }
 882
 883 /*
 884  * Allocate segment table.  On error, destroy the table and set errno.
 885  *
 886  * Returns 0 on success
 887  */
 888 static int
 889 alloc_segs(HTAB *hashp, int nsegs)
 890 {
 891         int i;
 892         SEGMENT store;
 893
 894         int save_errno;
 895
 896         hashp->dir = calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *));
 897         if (hashp->dir == NULL) {
 898                 save_errno = errno;
 899                 (void)hdestroy(hashp);
 900                 errno = save_errno;
 901                 return (-1);
 902         }
 903         hashp->nsegs = nsegs;
 904         if (nsegs == 0)
 905                 return 0;
 906         /* Allocate segments */
 907         store = calloc((size_t)(nsegs << hashp->SSHIFT), sizeof(SEGMENT));
 908         if (store == NULL) {
 909                 save_errno = errno;
 910                 (void)hdestroy(hashp);
 911                 errno = save_errno;
 912                 return (-1);
 913         }
 914         for (i = 0; i < nsegs; i++)
 915                 hashp->dir[i] = &store[i << hashp->SSHIFT];
 916         return (0);
 917 }
 918
 919 #if BYTE_ORDER == LITTLE_ENDIAN
 920 /*
 921  * Hashp->hdr needs to be byteswapped.
 922  */
 923 static void
 924 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
 925 {
 926         size_t i;
 927
 928         P_32_COPY(srcp->magic, destp->magic);
 929         P_32_COPY(srcp->version, destp->version);
 930         P_32_COPY(srcp->lorder, destp->lorder);
 931         P_32_COPY(srcp->bsize, destp->bsize);
 932         P_32_COPY(srcp->bshift, destp->bshift);
 933         P_32_COPY(srcp->dsize, destp->dsize);
 934         P_32_COPY(srcp->ssize, destp->ssize);
 935         P_32_COPY(srcp->sshift, destp->sshift);
 936         P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
 937         P_32_COPY(srcp->last_freed, destp->last_freed);
 938         P_32_COPY(srcp->max_bucket, destp->max_bucket);
 939         P_32_COPY(srcp->high_mask, destp->high_mask);
 940         P_32_COPY(srcp->low_mask, destp->low_mask);
 941         P_32_COPY(srcp->ffactor, destp->ffactor);
 942         P_32_COPY(srcp->nkeys, destp->nkeys);
 943         P_32_COPY(srcp->hdrpages, destp->hdrpages);
 944         P_32_COPY(srcp->h_charkey, destp->h_charkey);
 945         for (i = 0; i < NCACHED; i++) {
 946                 P_32_COPY(srcp->spares[i], destp->spares[i]);
 947                 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
 948         }
 949 }
 950
 951 static void
 952 swap_header(HTAB *hashp)
 953 {
 954         HASHHDR *hdrp;
 955         size_t i;
 956
 957         hdrp = &hashp->hdr;
 958
 959         M_32_SWAP(hdrp->magic);
 960         M_32_SWAP(hdrp->version);
 961         M_32_SWAP(hdrp->lorder);
 962         M_32_SWAP(hdrp->bsize);
 963         M_32_SWAP(hdrp->bshift);
 964         M_32_SWAP(hdrp->dsize);
 965         M_32_SWAP(hdrp->ssize);
 966         M_32_SWAP(hdrp->sshift);
 967         M_32_SWAP(hdrp->ovfl_point);
 968         M_32_SWAP(hdrp->last_freed);
 969         M_32_SWAP(hdrp->max_bucket);
 970         M_32_SWAP(hdrp->high_mask);
 971         M_32_SWAP(hdrp->low_mask);
 972         M_32_SWAP(hdrp->ffactor);
 973         M_32_SWAP(hdrp->nkeys);
 974         M_32_SWAP(hdrp->hdrpages);
 975         M_32_SWAP(hdrp->h_charkey);
 976         for (i = 0; i < NCACHED; i++) {
 977                 M_32_SWAP(hdrp->spares[i]);
 978                 M_16_SWAP(hdrp->bitmaps[i]);
 979         }
 980 }
 981 #endif