lib/libc/db/hash/hash.c

   1 /*      $NetBSD: hash.c,v 1.32 2012/03/13 21:13:32 christos 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 1.32 2012/03/13 21:13:32 christos 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 DEBUG1
 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                 hashp->BSHIFT = __log2((uint32_t)hashp->BSIZE);
 314         }
 315
 316         if (info) {
 317                 if (info->bsize) {
 318                         /* Round pagesize up to power of 2 */
 319                         hashp->BSHIFT = __log2(info->bsize);
 320                         hashp->BSIZE = 1 << hashp->BSHIFT;
 321                         if (hashp->BSIZE > MAX_BSIZE) {
 322                                 errno = EINVAL;
 323                                 return (NULL);
 324                         }
 325                 }
 326                 if (info->ffactor)
 327                         hashp->FFACTOR = info->ffactor;
 328                 if (info->hash)
 329                         hashp->hash = info->hash;
 330                 if (info->nelem)
 331                         nelem = info->nelem;
 332                 if (info->lorder) {
 333                         if (info->lorder != BIG_ENDIAN &&
 334                             info->lorder != LITTLE_ENDIAN) {
 335                                 errno = EINVAL;
 336                                 return (NULL);
 337                         }
 338                         hashp->LORDER = info->lorder;
 339                 }
 340         }
 341         /* init_htab should destroy the table and set errno if it fails */
 342         if (init_htab(hashp, (size_t)nelem))
 343                 return (NULL);
 344         else
 345                 return (hashp);
 346 }
 347 /*
 348  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
 349  * the table and set errno, so we just pass the error information along.
 350  *
 351  * Returns 0 on No Error
 352  */
 353 static int
 354 init_htab(HTAB *hashp, size_t nelem)
 355 {
 356         int nbuckets;
 357         uint32_t nsegs;
 358         int l2;
 359
 360         /*
 361          * Divide number of elements by the fill factor and determine a
 362          * desired number of buckets.  Allocate space for the next greater
 363          * power of two number of buckets.
 364          */
 365         nelem = (nelem - 1) / hashp->FFACTOR + 1;
 366
 367         _DBFIT(nelem, uint32_t);
 368         l2 = __log2(MAX((uint32_t)nelem, 2));
 369         nbuckets = 1 << l2;
 370
 371         hashp->SPARES[l2] = l2 + 1;
 372         hashp->SPARES[l2 + 1] = l2 + 1;
 373         hashp->OVFL_POINT = l2;
 374         hashp->LAST_FREED = 2;
 375
 376         /* First bitmap page is at: splitpoint l2 page offset 1 */
 377         if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0))
 378                 return (-1);
 379
 380         hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
 381         hashp->HIGH_MASK = (nbuckets << 1) - 1;
 382         /* LINTED constant in conditional context */
 383         hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
 384             hashp->BSHIFT) + 1;
 385
 386         nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
 387         nsegs = 1 << __log2(nsegs);
 388
 389         if (nsegs > (uint32_t)hashp->DSIZE)
 390                 hashp->DSIZE = nsegs;
 391         return (alloc_segs(hashp, (int)nsegs));
 392 }
 393
 394 /********************** DESTROY/CLOSE ROUTINES ************************/
 395
 396 /*
 397  * Flushes any changes to the file if necessary and destroys the hashp
 398  * structure, freeing all allocated space.
 399  */
 400 static int
 401 hdestroy(HTAB *hashp)
 402 {
 403         int i, save_errno;
 404
 405         save_errno = 0;
 406
 407 #ifdef HASH_STATISTICS
 408         (void)fprintf(stderr, "hdestroy: accesses %d collisions %d\n",
 409             hash_accesses, hash_collisions);
 410         (void)fprintf(stderr, "hdestroy: expansions %d\n",
 411             hash_expansions);
 412         (void)fprintf(stderr, "hdestroy: overflows %d\n",
 413             hash_overflows);
 414         (void)fprintf(stderr, "keys %d maxp %d segmentcount %d\n",
 415             hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
 416
 417         for (i = 0; i < NCACHED; i++)
 418                 (void)fprintf(stderr,
 419                     "spares[%d] = %d\n", i, hashp->SPARES[i]);
 420 #endif
 421         /*
 422          * Call on buffer manager to free buffers, and if required,
 423          * write them to disk.
 424          */
 425         if (__buf_free(hashp, 1, hashp->save_file))
 426                 save_errno = errno;
 427         if (hashp->dir) {
 428                 free(*hashp->dir);      /* Free initial segments */
 429                 /* Free extra segments */
 430                 while (hashp->exsegs--)
 431                         free(hashp->dir[--hashp->nsegs]);
 432                 free(hashp->dir);
 433         }
 434         if (flush_meta(hashp) && !save_errno)
 435                 save_errno = errno;
 436         /* Free Bigmaps */
 437         for (i = 0; i < hashp->nmaps; i++)
 438                 if (hashp->mapp[i])
 439                         free(hashp->mapp[i]);
 440
 441         if (hashp->fp != -1)
 442                 (void)close(hashp->fp);
 443
 444         free(hashp);
 445
 446         if (save_errno) {
 447                 errno = save_errno;
 448                 return (ERROR);
 449         }
 450         return (SUCCESS);
 451 }
 452 /*
 453  * Write modified pages to disk
 454  *
 455  * Returns:
 456  *       0 == OK
 457  *      -1 ERROR
 458  */
 459 static int
 460 hash_sync(const DB *dbp, uint32_t flags)
 461 {
 462         HTAB *hashp;
 463
 464         if (flags != 0) {
 465                 errno = EINVAL;
 466                 return (ERROR);
 467         }
 468
 469         if (!dbp)
 470                 return (ERROR);
 471
 472         hashp = dbp->internal;
 473         if (!hashp->save_file)
 474                 return (0);
 475         if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
 476                 return (ERROR);
 477         hashp->new_file = 0;
 478         return (0);
 479 }
 480
 481 /*
 482  * Returns:
 483  *       0 == OK
 484  *      -1 indicates that errno should be set
 485  */
 486 static int
 487 flush_meta(HTAB *hashp)
 488 {
 489         HASHHDR *whdrp;
 490 #if BYTE_ORDER == LITTLE_ENDIAN
 491         HASHHDR whdr;
 492 #endif
 493         int fp, i;
 494         ssize_t wsize;
 495
 496         if (!hashp->save_file)
 497                 return (0);
 498         hashp->MAGIC = HASHMAGIC;
 499         hashp->VERSION = HASHVERSION;
 500         hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
 501
 502         fp = hashp->fp;
 503         whdrp = &hashp->hdr;
 504 #if BYTE_ORDER == LITTLE_ENDIAN
 505         whdrp = &whdr;
 506         swap_header_copy(&hashp->hdr, whdrp);
 507 #endif
 508         if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
 509                 return (-1);
 510         else
 511                 if (wsize != sizeof(HASHHDR)) {
 512                         errno = EFTYPE;
 513                         hashp->err = errno;
 514                         return (-1);
 515                 }
 516         for (i = 0; i < NCACHED; i++)
 517                 if (hashp->mapp[i])
 518                         if (__put_page(hashp, (char *)(void *)hashp->mapp[i],
 519                                 (u_int)hashp->BITMAPS[i], 0, 1))
 520                                 return (-1);
 521         return (0);
 522 }
 523
 524 /*******************************SEARCH ROUTINES *****************************/
 525 /*
 526  * All the access routines return
 527  *
 528  * Returns:
 529  *       0 on SUCCESS
 530  *       1 to indicate an external ERROR (i.e. key not found, etc)
 531  *      -1 to indicate an internal ERROR (i.e. out of memory, etc)
 532  */
 533 static int
 534 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag)
 535 {
 536         HTAB *hashp;
 537
 538         hashp = dbp->internal;
 539         if (flag) {
 540                 hashp->err = errno = EINVAL;
 541                 return (ERROR);
 542         }
 543         return (hash_access(hashp, HASH_GET, __UNCONST(key), data));
 544 }
 545
 546 static int
 547 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag)
 548 {
 549         HTAB *hashp;
 550
 551         hashp = dbp->internal;
 552         if (flag && flag != R_NOOVERWRITE) {
 553                 hashp->err = errno = EINVAL;
 554                 return (ERROR);
 555         }
 556         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 557                 hashp->err = errno = EPERM;
 558                 return (ERROR);
 559         }
 560         /* LINTED const castaway */
 561         return (hash_access(hashp, flag == R_NOOVERWRITE ?
 562             HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data)));
 563 }
 564
 565 static int
 566 hash_delete(const DB *dbp, const DBT *key, uint32_t flag)
 567 {
 568         HTAB *hashp;
 569
 570         hashp = dbp->internal;
 571         if (flag && flag != R_CURSOR) {
 572                 hashp->err = errno = EINVAL;
 573                 return (ERROR);
 574         }
 575         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 576                 hashp->err = errno = EPERM;
 577                 return (ERROR);
 578         }
 579         return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL);
 580 }
 581
 582 /*
 583  * Assume that hashp has been set in wrapper routine.
 584  */
 585 static int
 586 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
 587 {
 588         BUFHEAD *rbufp;
 589         BUFHEAD *bufp, *save_bufp;
 590         uint16_t *bp;
 591         int n, ndx, off;
 592         size_t size;
 593         char *kp;
 594         uint16_t pageno;
 595
 596 #ifdef HASH_STATISTICS
 597         hash_accesses++;
 598 #endif
 599
 600         off = hashp->BSIZE;
 601         size = key->size;
 602         kp = (char *)key->data;
 603         rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0);
 604         if (!rbufp)
 605                 return (ERROR);
 606         save_bufp = rbufp;
 607
 608         /* Pin the bucket chain */
 609         rbufp->flags |= BUF_PIN;
 610         for (bp = (uint16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
 611                 if (bp[1] >= REAL_KEY) {
 612                         /* Real key/data pair */
 613                         if (size == (size_t)(off - *bp) &&
 614                             memcmp(kp, rbufp->page + *bp, size) == 0)
 615                                 goto found;
 616                         off = bp[1];
 617 #ifdef HASH_STATISTICS
 618                         hash_collisions++;
 619 #endif
 620                         bp += 2;
 621                         ndx += 2;
 622                 } else if (bp[1] == OVFLPAGE) {
 623                         rbufp = __get_buf(hashp, (uint32_t)*bp, rbufp, 0);
 624                         if (!rbufp) {
 625                                 save_bufp->flags &= ~BUF_PIN;
 626                                 return (ERROR);
 627                         }
 628                         /* FOR LOOP INIT */
 629                         bp = (uint16_t *)(void *)rbufp->page;
 630                         n = *bp++;
 631                         ndx = 1;
 632                         off = hashp->BSIZE;
 633                 } else if (bp[1] < REAL_KEY) {
 634                         if ((ndx =
 635                             __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0)
 636                                 goto found;
 637                         if (ndx == -2) {
 638                                 bufp = rbufp;
 639                                 if (!(pageno =
 640                                     __find_last_page(hashp, &bufp))) {
 641                                         ndx = 0;
 642                                         rbufp = bufp;
 643                                         break;  /* FOR */
 644                                 }
 645                                 rbufp = __get_buf(hashp, (uint32_t)pageno,
 646                                     bufp, 0);
 647                                 if (!rbufp) {
 648                                         save_bufp->flags &= ~BUF_PIN;
 649                                         return (ERROR);
 650                                 }
 651                                 /* FOR LOOP INIT */
 652                                 bp = (uint16_t *)(void *)rbufp->page;
 653                                 n = *bp++;
 654                                 ndx = 1;
 655                                 off = hashp->BSIZE;
 656                         } else {
 657                                 save_bufp->flags &= ~BUF_PIN;
 658                                 return (ERROR);
 659                         }
 660                 }
 661
 662         /* Not found */
 663         switch (action) {
 664         case HASH_PUT:
 665         case HASH_PUTNEW:
 666                 if (__addel(hashp, rbufp, key, val)) {
 667                         save_bufp->flags &= ~BUF_PIN;
 668                         return (ERROR);
 669                 } else {
 670                         save_bufp->flags &= ~BUF_PIN;
 671                         return (SUCCESS);
 672                 }
 673         case HASH_GET:
 674         case HASH_DELETE:
 675         default:
 676                 save_bufp->flags &= ~BUF_PIN;
 677                 return (ABNORMAL);
 678         }
 679
 680 found:
 681         switch (action) {
 682         case HASH_PUTNEW:
 683                 save_bufp->flags &= ~BUF_PIN;
 684                 return (ABNORMAL);
 685         case HASH_GET:
 686                 bp = (uint16_t *)(void *)rbufp->page;
 687                 if (bp[ndx + 1] < REAL_KEY) {
 688                         if (__big_return(hashp, rbufp, ndx, val, 0))
 689                                 return (ERROR);
 690                 } else {
 691                         val->data = (uint8_t *)rbufp->page + (int)bp[ndx + 1];
 692                         val->size = bp[ndx] - bp[ndx + 1];
 693                 }
 694                 break;
 695         case HASH_PUT:
 696                 if ((__delpair(hashp, rbufp, ndx)) ||
 697                     (__addel(hashp, rbufp, key, val))) {
 698                         save_bufp->flags &= ~BUF_PIN;
 699                         return (ERROR);
 700                 }
 701                 break;
 702         case HASH_DELETE:
 703                 if (__delpair(hashp, rbufp, ndx))
 704                         return (ERROR);
 705                 break;
 706         default:
 707                 abort();
 708         }
 709         save_bufp->flags &= ~BUF_PIN;
 710         return (SUCCESS);
 711 }
 712
 713 static int
 714 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag)
 715 {
 716         uint32_t bucket;
 717         BUFHEAD *bufp = NULL; /* XXX: gcc */
 718         HTAB *hashp;
 719         uint16_t *bp, ndx;
 720
 721         hashp = dbp->internal;
 722         if (flag && flag != R_FIRST && flag != R_NEXT) {
 723                 hashp->err = errno = EINVAL;
 724                 return (ERROR);
 725         }
 726 #ifdef HASH_STATISTICS
 727         hash_accesses++;
 728 #endif
 729         if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
 730                 hashp->cbucket = 0;
 731                 hashp->cndx = 1;
 732                 hashp->cpage = NULL;
 733         }
 734
 735         for (bp = NULL; !bp || !bp[0]; ) {
 736                 if (!(bufp = hashp->cpage)) {
 737                         for (bucket = hashp->cbucket;
 738                             bucket <= (uint32_t)hashp->MAX_BUCKET;
 739                             bucket++, hashp->cndx = 1) {
 740                                 bufp = __get_buf(hashp, bucket, NULL, 0);
 741                                 if (!bufp)
 742                                         return (ERROR);
 743                                 hashp->cpage = bufp;
 744                                 bp = (uint16_t *)(void *)bufp->page;
 745                                 if (bp[0])
 746                                         break;
 747                         }
 748                         hashp->cbucket = bucket;
 749                         if (hashp->cbucket > hashp->MAX_BUCKET) {
 750                                 hashp->cbucket = -1;
 751                                 return (ABNORMAL);
 752                         }
 753                 } else
 754                         bp = (uint16_t *)(void *)hashp->cpage->page;
 755
 756                 _DIAGASSERT(bp != NULL);
 757                 _DIAGASSERT(bufp != NULL);
 758                 while (bp[hashp->cndx + 1] == OVFLPAGE) {
 759                         bufp = hashp->cpage =
 760                             __get_buf(hashp, (uint32_t)bp[hashp->cndx], bufp,
 761                                 0);
 762                         if (!bufp)
 763                                 return (ERROR);
 764                         bp = (uint16_t *)(void *)(bufp->page);
 765                         hashp->cndx = 1;
 766                 }
 767                 if (!bp[0]) {
 768                         hashp->cpage = NULL;
 769                         ++hashp->cbucket;
 770                 }
 771         }
 772         ndx = hashp->cndx;
 773         if (bp[ndx + 1] < REAL_KEY) {
 774                 if (__big_keydata(hashp, bufp, key, data, 1))
 775                         return (ERROR);
 776         } else {
 777                 if (hashp->cpage == NULL)
 778                         return (ERROR);
 779                 key->data = (uint8_t *)hashp->cpage->page + bp[ndx];
 780                 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
 781                 data->data = (uint8_t *)hashp->cpage->page + bp[ndx + 1];
 782                 data->size = bp[ndx] - bp[ndx + 1];
 783                 ndx += 2;
 784                 if (ndx > bp[0]) {
 785                         hashp->cpage = NULL;
 786                         hashp->cbucket++;
 787                         hashp->cndx = 1;
 788                 } else
 789                         hashp->cndx = ndx;
 790         }
 791         return (SUCCESS);
 792 }
 793
 794 /********************************* UTILITIES ************************/
 795
 796 /*
 797  * Returns:
 798  *       0 ==> OK
 799  *      -1 ==> Error
 800  */
 801 int
 802 __expand_table(HTAB *hashp)
 803 {
 804         uint32_t old_bucket, new_bucket;
 805         int new_segnum, spare_ndx;
 806         size_t dirsize;
 807
 808 #ifdef HASH_STATISTICS
 809         hash_expansions++;
 810 #endif
 811         new_bucket = ++hashp->MAX_BUCKET;
 812         old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
 813
 814         new_segnum = new_bucket >> hashp->SSHIFT;
 815
 816         /* Check if we need a new segment */
 817         if (new_segnum >= hashp->nsegs) {
 818                 /* Check if we need to expand directory */
 819                 if (new_segnum >= hashp->DSIZE) {
 820                         /* Reallocate directory */
 821                         dirsize = hashp->DSIZE * sizeof(SEGMENT *);
 822                         if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
 823                                 return (-1);
 824                         dirsize <<= 1;
 825                         _DBFIT(dirsize, uint32_t);
 826                         hashp->DSIZE = (uint32_t)dirsize;
 827                 }
 828                 if ((hashp->dir[new_segnum] =
 829                     calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
 830                         return (-1);
 831                 hashp->exsegs++;
 832                 hashp->nsegs++;
 833         }
 834         /*
 835          * If the split point is increasing (MAX_BUCKET's log base 2
 836          * * increases), we need to copy the current contents of the spare
 837          * split bucket to the next bucket.
 838          */
 839         spare_ndx = __log2((uint32_t)(hashp->MAX_BUCKET + 1));
 840         if (spare_ndx > hashp->OVFL_POINT) {
 841                 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
 842                 hashp->OVFL_POINT = spare_ndx;
 843         }
 844
 845         if (new_bucket > (uint32_t)hashp->HIGH_MASK) {
 846                 /* Starting a new doubling */
 847                 hashp->LOW_MASK = hashp->HIGH_MASK;
 848                 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
 849         }
 850         /* Relocate records to the new bucket */
 851         return (__split_page(hashp, old_bucket, new_bucket));
 852 }
 853
 854 /*
 855  * If realloc guarantees that the pointer is not destroyed if the realloc
 856  * fails, then this routine can go away.
 857  */
 858 static void *
 859 hash_realloc(SEGMENT **p_ptr, size_t oldsize, size_t newsize)
 860 {
 861         void *p;
 862
 863         if ((p = malloc(newsize)) != NULL) {
 864                 memmove(p, *p_ptr, oldsize);
 865                 memset((char *)p + oldsize, 0, newsize - oldsize);
 866                 free(*p_ptr);
 867                 *p_ptr = p;
 868         }
 869         return (p);
 870 }
 871
 872 uint32_t
 873 __call_hash(HTAB *hashp, char *k, int len)
 874 {
 875         int n, bucket;
 876
 877         n = hashp->hash(k, (size_t)len);
 878         bucket = n & hashp->HIGH_MASK;
 879         if (bucket > hashp->MAX_BUCKET)
 880                 bucket = bucket & hashp->LOW_MASK;
 881         return (bucket);
 882 }
 883
 884 /*
 885  * Allocate segment table.  On error, destroy the table and set errno.
 886  *
 887  * Returns 0 on success
 888  */
 889 static int
 890 alloc_segs(HTAB *hashp, int nsegs)
 891 {
 892         int i;
 893         SEGMENT store;
 894
 895         int save_errno;
 896
 897         hashp->dir = calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *));
 898         if (hashp->dir == NULL) {
 899                 save_errno = errno;
 900                 (void)hdestroy(hashp);
 901                 errno = save_errno;
 902                 return (-1);
 903         }
 904         hashp->nsegs = nsegs;
 905         if (nsegs == 0)
 906                 return 0;
 907         /* Allocate segments */
 908         store = calloc((size_t)(nsegs << hashp->SSHIFT), sizeof(SEGMENT));
 909         if (store == NULL) {
 910                 save_errno = errno;
 911                 (void)hdestroy(hashp);
 912                 errno = save_errno;
 913                 return (-1);
 914         }
 915         for (i = 0; i < nsegs; i++)
 916                 hashp->dir[i] = &store[i << hashp->SSHIFT];
 917         return (0);
 918 }
 919
 920 #if BYTE_ORDER == LITTLE_ENDIAN
 921 /*
 922  * Hashp->hdr needs to be byteswapped.
 923  */
 924 static void
 925 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
 926 {
 927         size_t i;
 928
 929         P_32_COPY(srcp->magic, destp->magic);
 930         P_32_COPY(srcp->version, destp->version);
 931         P_32_COPY(srcp->lorder, destp->lorder);
 932         P_32_COPY(srcp->bsize, destp->bsize);
 933         P_32_COPY(srcp->bshift, destp->bshift);
 934         P_32_COPY(srcp->dsize, destp->dsize);
 935         P_32_COPY(srcp->ssize, destp->ssize);
 936         P_32_COPY(srcp->sshift, destp->sshift);
 937         P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
 938         P_32_COPY(srcp->last_freed, destp->last_freed);
 939         P_32_COPY(srcp->max_bucket, destp->max_bucket);
 940         P_32_COPY(srcp->high_mask, destp->high_mask);
 941         P_32_COPY(srcp->low_mask, destp->low_mask);
 942         P_32_COPY(srcp->ffactor, destp->ffactor);
 943         P_32_COPY(srcp->nkeys, destp->nkeys);
 944         P_32_COPY(srcp->hdrpages, destp->hdrpages);
 945         P_32_COPY(srcp->h_charkey, destp->h_charkey);
 946         for (i = 0; i < NCACHED; i++) {
 947                 P_32_COPY(srcp->spares[i], destp->spares[i]);
 948                 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
 949         }
 950 }
 951
 952 static void
 953 swap_header(HTAB *hashp)
 954 {
 955         HASHHDR *hdrp;
 956         size_t i;
 957
 958         hdrp = &hashp->hdr;
 959
 960         M_32_SWAP(hdrp->magic);
 961         M_32_SWAP(hdrp->version);
 962         M_32_SWAP(hdrp->lorder);
 963         M_32_SWAP(hdrp->bsize);
 964         M_32_SWAP(hdrp->bshift);
 965         M_32_SWAP(hdrp->dsize);
 966         M_32_SWAP(hdrp->ssize);
 967         M_32_SWAP(hdrp->sshift);
 968         M_32_SWAP(hdrp->ovfl_point);
 969         M_32_SWAP(hdrp->last_freed);
 970         M_32_SWAP(hdrp->max_bucket);
 971         M_32_SWAP(hdrp->high_mask);
 972         M_32_SWAP(hdrp->low_mask);
 973         M_32_SWAP(hdrp->ffactor);
 974         M_32_SWAP(hdrp->nkeys);
 975         M_32_SWAP(hdrp->hdrpages);
 976         M_32_SWAP(hdrp->h_charkey);
 977         for (i = 0; i < NCACHED; i++) {
 978                 M_32_SWAP(hdrp->spares[i]);
 979                 M_16_SWAP(hdrp->bitmaps[i]);
 980         }
 981 }
 982 #endif