db/hash/hash.c

   1 /*-
   2  * Copyright (c) 1990, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Margo Seltzer.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36
  37 #if defined(LIBC_SCCS) && !defined(lint)
  38 static char sccsid[] = "@(#)hash.c      8.9 (Berkeley) 6/16/94";
  39 #endif /* LIBC_SCCS and not lint */
  40
  41 #include <sys/param.h>
  42 #include <sys/stat.h>
  43
  44 #include <errno.h>
  45 #include <fcntl.h>
  46 #include <stdio.h>
  47 #include <stdlib.h>
  48 #include <string.h>
  49 #include <unistd.h>
  50 #ifdef DEBUG
  51 #include <assert.h>
  52 #endif
  53
  54 #include <db.h>
  55 #include "hash.h"
  56 #include "page.h"
  57 #include "extern.h"
  58
  59 static int   alloc_segs __P((HTAB *, int));
  60 static int   flush_meta __P((HTAB *));
  61 static int   hash_access __P((HTAB *, ACTION, DBT *, DBT *));
  62 static int   hash_close __P((DB *));
  63 static int   hash_delete __P((const DB *, const DBT *, u_int32_t));
  64 static int   hash_fd __P((const DB *));
  65 static int   hash_get __P((const DB *, const DBT *, DBT *, u_int32_t));
  66 static int   hash_put __P((const DB *, DBT *, const DBT *, u_int32_t));
  67 static void *hash_realloc __P((SEGMENT **, int, int));
  68 static int   hash_seq __P((const DB *, DBT *, DBT *, u_int32_t));
  69 static int   hash_sync __P((const DB *, u_int32_t));
  70 static int   hdestroy __P((HTAB *));
  71 static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *));
  72 static int   init_htab __P((HTAB *, int));
  73 #if BYTE_ORDER == LITTLE_ENDIAN
  74 static void  swap_header __P((HTAB *));
  75 static void  swap_header_copy __P((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 extern DB *
  96 __hash_open(file, flags, mode, info, dflags)
  97         const char *file;
  98         int flags, mode, dflags;
  99         const HASHINFO *info;   /* Special directives for create */
 100 {
 101         HTAB *hashp;
 102         struct stat statbuf;
 103         DB *dbp;
 104         int bpages, hdrsize, new_table, nsegs, save_errno;
 105
 106         if ((flags & O_ACCMODE) == O_WRONLY) {
 107                 errno = EINVAL;
 108                 return (NULL);
 109         }
 110
 111         if (!(hashp = (HTAB *)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                 (void)fcntl(hashp->fp, F_SETFD, 1);
 134         }
 135         if (new_table) {
 136                 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
 137                         RETURN_ERROR(errno, error1);
 138         } else {
 139                 /* Table already exists */
 140                 if (info && info->hash)
 141                         hashp->hash = info->hash;
 142                 else
 143                         hashp->hash = __default_hash;
 144
 145                 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
 146 #if BYTE_ORDER == LITTLE_ENDIAN
 147                 swap_header(hashp);
 148 #endif
 149                 if (hdrsize == -1)
 150                         RETURN_ERROR(errno, error1);
 151                 if (hdrsize != sizeof(HASHHDR))
 152                         RETURN_ERROR(EFTYPE, error1);
 153                 /* Verify file type, versions and hash function */
 154                 if (hashp->MAGIC != HASHMAGIC)
 155                         RETURN_ERROR(EFTYPE, error1);
 156 #define OLDHASHVERSION  1
 157                 if (hashp->VERSION != HASHVERSION &&
 158                     hashp->VERSION != OLDHASHVERSION)
 159                         RETURN_ERROR(EFTYPE, error1);
 160                 if (hashp->hash(CHARKEY, sizeof(CHARKEY))
 161                     != (u_int32_t) hashp->H_CHARKEY)
 162                         RETURN_ERROR(EFTYPE, error1);
 163                 /*
 164                  * Figure out how many segments we need.  Max_Bucket is the
 165                  * maximum bucket number, so the number of buckets is
 166                  * max_bucket + 1.
 167                  */
 168                 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
 169                          hashp->SGSIZE;
 170                 hashp->nsegs = 0;
 171                 if (alloc_segs(hashp, nsegs))
 172                         /*
 173                          * If alloc_segs fails, table will have been destroyed
 174                          * and errno will have been set.
 175                          */
 176                         return (NULL);
 177                 /* Read in bitmaps */
 178                 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
 179                     (hashp->BSIZE << BYTE_SHIFT) - 1) >>
 180                     (hashp->BSHIFT + BYTE_SHIFT);
 181
 182                 hashp->nmaps = bpages;
 183                 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
 184         }
 185
 186         /* Initialize Buffer Manager */
 187         if (info && info->cachesize)
 188                 __buf_init(hashp, info->cachesize);
 189         else
 190                 __buf_init(hashp, DEF_BUFSIZE);
 191
 192         hashp->new_file = new_table;
 193         hashp->save_file = file && (hashp->flags & O_ACCMODE) != O_RDONLY;
 194         hashp->cbucket = -1;
 195         if (!(dbp = (DB *)malloc(sizeof(DB)))) {
 196                 save_errno = errno;
 197                 hdestroy(hashp);
 198                 errno = save_errno;
 199                 return (NULL);
 200         }
 201         dbp->internal = hashp;
 202         dbp->close = hash_close;
 203         dbp->del = hash_delete;
 204         dbp->fd = hash_fd;
 205         dbp->get = hash_get;
 206         dbp->put = hash_put;
 207         dbp->seq = hash_seq;
 208         dbp->sync = hash_sync;
 209         dbp->type = DB_HASH;
 210
 211 #ifdef DEBUG
 212         (void)fprintf(stderr,
 213 "%s\n%s%x\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",
 214             "init_htab:",
 215             "TABLE POINTER   ", hashp,
 216             "BUCKET SIZE     ", hashp->BSIZE,
 217             "BUCKET SHIFT    ", hashp->BSHIFT,
 218             "DIRECTORY SIZE  ", hashp->DSIZE,
 219             "SEGMENT SIZE    ", hashp->SGSIZE,
 220             "SEGMENT SHIFT   ", hashp->SSHIFT,
 221             "FILL FACTOR     ", hashp->FFACTOR,
 222             "MAX BUCKET      ", hashp->MAX_BUCKET,
 223             "OVFL POINT      ", hashp->OVFL_POINT,
 224             "LAST FREED      ", hashp->LAST_FREED,
 225             "HIGH MASK       ", hashp->HIGH_MASK,
 226             "LOW  MASK       ", hashp->LOW_MASK,
 227             "NSEGS           ", hashp->nsegs,
 228             "NKEYS           ", hashp->NKEYS);
 229 #endif
 230 #ifdef HASH_STATISTICS
 231         hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
 232 #endif
 233         return (dbp);
 234
 235 error1:
 236         if (hashp != NULL)
 237                 (void)close(hashp->fp);
 238
 239 error0:
 240         free(hashp);
 241         errno = save_errno;
 242         return (NULL);
 243 }
 244
 245 static int
 246 hash_close(dbp)
 247         DB *dbp;
 248 {
 249         HTAB *hashp;
 250         int retval;
 251
 252         if (!dbp)
 253                 return (ERROR);
 254
 255         hashp = (HTAB *)dbp->internal;
 256         retval = hdestroy(hashp);
 257         free(dbp);
 258         return (retval);
 259 }
 260
 261 static int
 262 hash_fd(dbp)
 263         const DB *dbp;
 264 {
 265         HTAB *hashp;
 266
 267         if (!dbp)
 268                 return (ERROR);
 269
 270         hashp = (HTAB *)dbp->internal;
 271         if (hashp->fp == -1) {
 272                 errno = ENOENT;
 273                 return (-1);
 274         }
 275         return (hashp->fp);
 276 }
 277
 278 /************************** LOCAL CREATION ROUTINES **********************/
 279 static HTAB *
 280 init_hash(hashp, file, info)
 281         HTAB *hashp;
 282         const char *file;
 283         HASHINFO *info;
 284 {
 285 #ifdef _STATBUF_ST_BLKSIZE
 286         struct stat statbuf;
 287 #endif
 288         int nelem;
 289
 290         nelem = 1;
 291         hashp->NKEYS = 0;
 292         hashp->LORDER = BYTE_ORDER;
 293         hashp->BSIZE = DEF_BUCKET_SIZE;
 294         hashp->BSHIFT = DEF_BUCKET_SHIFT;
 295         hashp->SGSIZE = DEF_SEGSIZE;
 296         hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
 297         hashp->DSIZE = DEF_DIRSIZE;
 298         hashp->FFACTOR = DEF_FFACTOR;
 299         hashp->hash = __default_hash;
 300         memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
 301         memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
 302
 303         /* Fix bucket size to be optimal for file system */
 304 #ifdef _STATBUF_ST_BLKSIZE
 305         if (file != NULL) {
 306                 if (stat(file, &statbuf))
 307                         return (NULL);
 308                 hashp->BSIZE = statbuf.st_blksize;
 309                 hashp->BSHIFT = __hash_log2(hashp->BSIZE);
 310         }
 311 #endif
 312
 313         if (info) {
 314                 if (info->bsize) {
 315                         /* Round pagesize up to power of 2 */
 316                         hashp->BSHIFT = __hash_log2(info->bsize);
 317                         hashp->BSIZE = 1 << hashp->BSHIFT;
 318                         if (hashp->BSIZE > MAX_BSIZE) {
 319                                 errno = EINVAL;
 320                                 return (NULL);
 321                         }
 322                 }
 323                 if (info->ffactor)
 324                         hashp->FFACTOR = info->ffactor;
 325                 if (info->hash)
 326                         hashp->hash = info->hash;
 327                 if (info->nelem)
 328                         nelem = info->nelem;
 329                 if (info->lorder) {
 330                         if (info->lorder != BIG_ENDIAN &&
 331                             info->lorder != LITTLE_ENDIAN) {
 332                                 errno = EINVAL;
 333                                 return (NULL);
 334                         }
 335                         hashp->LORDER = info->lorder;
 336                 }
 337         }
 338         /* init_htab should destroy the table and set errno if it fails */
 339         if (init_htab(hashp, nelem))
 340                 return (NULL);
 341         else
 342                 return (hashp);
 343 }
 344 /*
 345  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
 346  * the table and set errno, so we just pass the error information along.
 347  *
 348  * Returns 0 on No Error
 349  */
 350 static int
 351 init_htab(hashp, nelem)
 352         HTAB *hashp;
 353         int nelem;
 354 {
 355         register int nbuckets, nsegs;
 356         int l2;
 357
 358         /*
 359          * Divide number of elements by the fill factor and determine a
 360          * desired number of buckets.  Allocate space for the next greater
 361          * power of two number of buckets.
 362          */
 363         nelem = (nelem - 1) / hashp->FFACTOR + 1;
 364
 365         l2 = __hash_log2(MAX(nelem, 2));
 366         nbuckets = 1 << l2;
 367
 368         hashp->SPARES[l2] = l2 + 1;
 369         hashp->SPARES[l2 + 1] = l2 + 1;
 370         hashp->OVFL_POINT = l2;
 371         hashp->LAST_FREED = 2;
 372
 373         /* First bitmap page is at: splitpoint l2 page offset 1 */
 374         if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
 375                 return (-1);
 376
 377         hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
 378         hashp->HIGH_MASK = (nbuckets << 1) - 1;
 379         hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
 380             hashp->BSHIFT) + 1;
 381
 382         nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
 383         nsegs = 1 << __hash_log2(nsegs);
 384
 385         if (nsegs > hashp->DSIZE)
 386                 hashp->DSIZE = nsegs;
 387         return (alloc_segs(hashp, nsegs));
 388 }
 389
 390 /********************** DESTROY/CLOSE ROUTINES ************************/
 391
 392 /*
 393  * Flushes any changes to the file if necessary and destroys the hashp
 394  * structure, freeing all allocated space.
 395  */
 396 static int
 397 hdestroy(hashp)
 398         HTAB *hashp;
 399 {
 400         int i, save_errno;
 401
 402         save_errno = 0;
 403
 404 #ifdef HASH_STATISTICS
 405         (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
 406             hash_accesses, hash_collisions);
 407         (void)fprintf(stderr, "hdestroy: expansions %ld\n",
 408             hash_expansions);
 409         (void)fprintf(stderr, "hdestroy: overflows %ld\n",
 410             hash_overflows);
 411         (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
 412             hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
 413
 414         for (i = 0; i < NCACHED; i++)
 415                 (void)fprintf(stderr,
 416                     "spares[%d] = %d\n", i, hashp->SPARES[i]);
 417 #endif
 418         /*
 419          * Call on buffer manager to free buffers, and if required,
 420          * write them to disk.
 421          */
 422         if (__buf_free(hashp, 1, hashp->save_file))
 423                 save_errno = errno;
 424         if (hashp->dir) {
 425                 free(*hashp->dir);      /* Free initial segments */
 426                 /* Free extra segments */
 427                 while (hashp->exsegs--)
 428                         free(hashp->dir[--hashp->nsegs]);
 429                 free(hashp->dir);
 430         }
 431         if (flush_meta(hashp) && !save_errno)
 432                 save_errno = errno;
 433         /* Free Bigmaps */
 434         for (i = 0; i < hashp->nmaps; i++)
 435                 if (hashp->mapp[i])
 436                         free(hashp->mapp[i]);
 437
 438         if (hashp->fp != -1)
 439                 (void)close(hashp->fp);
 440
 441         free(hashp);
 442
 443         if (save_errno) {
 444                 errno = save_errno;
 445                 return (ERROR);
 446         }
 447         return (SUCCESS);
 448 }
 449 /*
 450  * Write modified pages to disk
 451  *
 452  * Returns:
 453  *       0 == OK
 454  *      -1 ERROR
 455  */
 456 static int
 457 hash_sync(dbp, flags)
 458         const DB *dbp;
 459         u_int32_t flags;
 460 {
 461         HTAB *hashp;
 462
 463         if (flags != 0) {
 464                 errno = EINVAL;
 465                 return (ERROR);
 466         }
 467
 468         if (!dbp)
 469                 return (ERROR);
 470
 471         hashp = (HTAB *)dbp->internal;
 472         if (!hashp->save_file)
 473                 return (0);
 474         if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
 475                 return (ERROR);
 476         hashp->new_file = 0;
 477         return (0);
 478 }
 479
 480 /*
 481  * Returns:
 482  *       0 == OK
 483  *      -1 indicates that errno should be set
 484  */
 485 static int
 486 flush_meta(hashp)
 487         HTAB *hashp;
 488 {
 489         HASHHDR *whdrp;
 490 #if BYTE_ORDER == LITTLE_ENDIAN
 491         HASHHDR whdr;
 492 #endif
 493         int fp, i, wsize;
 494
 495         if (!hashp->save_file)
 496                 return (0);
 497         hashp->MAGIC = HASHMAGIC;
 498         hashp->VERSION = HASHVERSION;
 499         hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
 500
 501         fp = hashp->fp;
 502         whdrp = &hashp->hdr;
 503 #if BYTE_ORDER == LITTLE_ENDIAN
 504         whdrp = &whdr;
 505         swap_header_copy(&hashp->hdr, whdrp);
 506 #endif
 507         if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
 508             ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
 509                 return (-1);
 510         else
 511                 if (wsize != sizeof(HASHHDR)) {
 512                         errno = EFTYPE;
 513                         hashp->errnum = errno;
 514                         return (-1);
 515                 }
 516         for (i = 0; i < NCACHED; i++)
 517                 if (hashp->mapp[i])
 518                         if (__put_page(hashp, (char *)hashp->mapp[i],
 519                                 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(dbp, key, data, flag)
 535         const DB *dbp;
 536         const DBT *key;
 537         DBT *data;
 538         u_int32_t flag;
 539 {
 540         HTAB *hashp;
 541
 542         hashp = (HTAB *)dbp->internal;
 543         if (flag) {
 544                 hashp->errnum = errno = EINVAL;
 545                 return (ERROR);
 546         }
 547         return (hash_access(hashp, HASH_GET, (DBT *)key, data));
 548 }
 549
 550 static int
 551 hash_put(dbp, key, data, flag)
 552         const DB *dbp;
 553         DBT *key;
 554         const DBT *data;
 555         u_int32_t flag;
 556 {
 557         HTAB *hashp;
 558
 559         hashp = (HTAB *)dbp->internal;
 560         if (flag && flag != R_NOOVERWRITE) {
 561                 hashp->errnum = errno = EINVAL;
 562                 return (ERROR);
 563         }
 564         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 565                 hashp->errnum = errno = EPERM;
 566                 return (ERROR);
 567         }
 568         return (hash_access(hashp, flag == R_NOOVERWRITE ?
 569             HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
 570 }
 571
 572 static int
 573 hash_delete(dbp, key, flag)
 574         const DB *dbp;
 575         const DBT *key;
 576         u_int32_t flag;         /* Ignored */
 577 {
 578         HTAB *hashp;
 579
 580         hashp = (HTAB *)dbp->internal;
 581         if (flag && flag != R_CURSOR) {
 582                 hashp->errnum = errno = EINVAL;
 583                 return (ERROR);
 584         }
 585         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 586                 hashp->errnum = errno = EPERM;
 587                 return (ERROR);
 588         }
 589         return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
 590 }
 591
 592 /*
 593  * Assume that hashp has been set in wrapper routine.
 594  */
 595 static int
 596 hash_access(hashp, action, key, val)
 597         HTAB *hashp;
 598         ACTION action;
 599         DBT *key, *val;
 600 {
 601         register BUFHEAD *rbufp;
 602         BUFHEAD *bufp, *save_bufp;
 603         register u_int16_t *bp;
 604         register int n, ndx, off, size;
 605         register char *kp;
 606         u_int16_t pageno;
 607
 608 #ifdef HASH_STATISTICS
 609         hash_accesses++;
 610 #endif
 611
 612         off = hashp->BSIZE;
 613         size = key->size;
 614         kp = (char *)key->data;
 615         rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
 616         if (!rbufp)
 617                 return (ERROR);
 618         save_bufp = rbufp;
 619
 620         /* Pin the bucket chain */
 621         rbufp->flags |= BUF_PIN;
 622         for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
 623                 if (bp[1] >= REAL_KEY) {
 624                         /* Real key/data pair */
 625                         if (size == off - *bp &&
 626                             memcmp(kp, rbufp->page + *bp, size) == 0)
 627                                 goto found;
 628                         off = bp[1];
 629 #ifdef HASH_STATISTICS
 630                         hash_collisions++;
 631 #endif
 632                         bp += 2;
 633                         ndx += 2;
 634                 } else if (bp[1] == OVFLPAGE) {
 635                         rbufp = __get_buf(hashp, *bp, rbufp, 0);
 636                         if (!rbufp) {
 637                                 save_bufp->flags &= ~BUF_PIN;
 638                                 return (ERROR);
 639                         }
 640                         /* FOR LOOP INIT */
 641                         bp = (u_int16_t *)rbufp->page;
 642                         n = *bp++;
 643                         ndx = 1;
 644                         off = hashp->BSIZE;
 645                 } else if (bp[1] < REAL_KEY) {
 646                         if ((ndx =
 647                             __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
 648                                 goto found;
 649                         if (ndx == -2) {
 650                                 bufp = rbufp;
 651                                 if (!(pageno =
 652                                     __find_last_page(hashp, &bufp))) {
 653                                         ndx = 0;
 654                                         rbufp = bufp;
 655                                         break;  /* FOR */
 656                                 }
 657                                 rbufp = __get_buf(hashp, pageno, bufp, 0);
 658                                 if (!rbufp) {
 659                                         save_bufp->flags &= ~BUF_PIN;
 660                                         return (ERROR);
 661                                 }
 662                                 /* FOR LOOP INIT */
 663                                 bp = (u_int16_t *)rbufp->page;
 664                                 n = *bp++;
 665                                 ndx = 1;
 666                                 off = hashp->BSIZE;
 667                         } else {
 668                                 save_bufp->flags &= ~BUF_PIN;
 669                                 return (ERROR);
 670                         }
 671                 }
 672
 673         /* Not found */
 674         switch (action) {
 675         case HASH_PUT:
 676         case HASH_PUTNEW:
 677                 if (__addel(hashp, rbufp, key, val)) {
 678                         save_bufp->flags &= ~BUF_PIN;
 679                         return (ERROR);
 680                 } else {
 681                         save_bufp->flags &= ~BUF_PIN;
 682                         return (SUCCESS);
 683                 }
 684         case HASH_GET:
 685         case HASH_DELETE:
 686         default:
 687                 save_bufp->flags &= ~BUF_PIN;
 688                 return (ABNORMAL);
 689         }
 690
 691 found:
 692         switch (action) {
 693         case HASH_PUTNEW:
 694                 save_bufp->flags &= ~BUF_PIN;
 695                 return (ABNORMAL);
 696         case HASH_GET:
 697                 bp = (u_int16_t *)rbufp->page;
 698                 if (bp[ndx + 1] < REAL_KEY) {
 699                         if (__big_return(hashp, rbufp, ndx, val, 0))
 700                                 return (ERROR);
 701                 } else {
 702                         val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
 703                         val->size = bp[ndx] - bp[ndx + 1];
 704                 }
 705                 break;
 706         case HASH_PUT:
 707                 if ((__delpair(hashp, rbufp, ndx)) ||
 708                     (__addel(hashp, rbufp, key, val))) {
 709                         save_bufp->flags &= ~BUF_PIN;
 710                         return (ERROR);
 711                 }
 712                 break;
 713         case HASH_DELETE:
 714                 if (__delpair(hashp, rbufp, ndx))
 715                         return (ERROR);
 716                 break;
 717         default:
 718                 abort();
 719         }
 720         save_bufp->flags &= ~BUF_PIN;
 721         return (SUCCESS);
 722 }
 723
 724 static int
 725 hash_seq(dbp, key, data, flag)
 726         const DB *dbp;
 727         DBT *key, *data;
 728         u_int32_t flag;
 729 {
 730         register u_int32_t bucket;
 731         register BUFHEAD *bufp;
 732         HTAB *hashp;
 733         u_int16_t *bp, ndx;
 734
 735         hashp = (HTAB *)dbp->internal;
 736         if (flag && flag != R_FIRST && flag != R_NEXT) {
 737                 hashp->errnum = errno = EINVAL;
 738                 return (ERROR);
 739         }
 740 #ifdef HASH_STATISTICS
 741         hash_accesses++;
 742 #endif
 743         if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
 744                 hashp->cbucket = 0;
 745                 hashp->cndx = 1;
 746                 hashp->cpage = NULL;
 747         }
 748
 749         for (bp = NULL; !bp || !bp[0]; ) {
 750                 if (!(bufp = hashp->cpage)) {
 751                         for (bucket = hashp->cbucket;
 752                             bucket <= (u_int32_t) hashp->MAX_BUCKET;
 753                             bucket++, hashp->cndx = 1) {
 754                                 bufp = __get_buf(hashp, bucket, NULL, 0);
 755                                 if (!bufp)
 756                                         return (ERROR);
 757                                 hashp->cpage = bufp;
 758                                 bp = (u_int16_t *)bufp->page;
 759                                 if (bp[0])
 760                                         break;
 761                         }
 762                         hashp->cbucket = bucket;
 763                         if (hashp->cbucket > hashp->MAX_BUCKET) {
 764                                 hashp->cbucket = -1;
 765                                 return (ABNORMAL);
 766                         }
 767                 } else
 768                         bp = (u_int16_t *)hashp->cpage->page;
 769
 770 #ifdef DEBUG
 771                 assert(bp);
 772                 assert(bufp);
 773 #endif
 774                 while (bp[hashp->cndx + 1] == OVFLPAGE) {
 775                         bufp = hashp->cpage =
 776                             __get_buf(hashp, bp[hashp->cndx], bufp, 0);
 777                         if (!bufp)
 778                                 return (ERROR);
 779                         bp = (u_int16_t *)(bufp->page);
 780                         hashp->cndx = 1;
 781                 }
 782                 if (!bp[0]) {
 783                         hashp->cpage = NULL;
 784                         ++hashp->cbucket;
 785                 }
 786         }
 787         ndx = hashp->cndx;
 788         if (bp[ndx + 1] < REAL_KEY) {
 789                 if (__big_keydata(hashp, bufp, key, data, 1))
 790                         return (ERROR);
 791         } else {
 792                 key->data = (u_char *)hashp->cpage->page + bp[ndx];
 793                 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
 794                 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
 795                 data->size = bp[ndx] - bp[ndx + 1];
 796                 ndx += 2;
 797                 if (ndx > bp[0]) {
 798                         hashp->cpage = NULL;
 799                         hashp->cbucket++;
 800                         hashp->cndx = 1;
 801                 } else
 802                         hashp->cndx = ndx;
 803         }
 804         return (SUCCESS);
 805 }
 806
 807 /********************************* UTILITIES ************************/
 808
 809 /*
 810  * Returns:
 811  *       0 ==> OK
 812  *      -1 ==> Error
 813  */
 814 extern int
 815 __expand_table(hashp)
 816         HTAB *hashp;
 817 {
 818         u_int32_t old_bucket, new_bucket;
 819         int dirsize, new_segnum, spare_ndx;
 820
 821 #ifdef HASH_STATISTICS
 822         hash_expansions++;
 823 #endif
 824         new_bucket = ++hashp->MAX_BUCKET;
 825         old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
 826
 827         new_segnum = new_bucket >> hashp->SSHIFT;
 828
 829         /* Check if we need a new segment */
 830         if (new_segnum >= hashp->nsegs) {
 831                 /* Check if we need to expand directory */
 832                 if (new_segnum >= hashp->DSIZE) {
 833                         /* Reallocate directory */
 834                         dirsize = hashp->DSIZE * sizeof(SEGMENT *);
 835                         if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
 836                                 return (-1);
 837                         hashp->DSIZE = dirsize << 1;
 838                 }
 839                 if ((hashp->dir[new_segnum] =
 840                     (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
 841                         return (-1);
 842                 hashp->exsegs++;
 843                 hashp->nsegs++;
 844         }
 845         /*
 846          * If the split point is increasing (MAX_BUCKET's log base 2
 847          * * increases), we need to copy the current contents of the spare
 848          * split bucket to the next bucket.
 849          */
 850         spare_ndx = __hash_log2(hashp->MAX_BUCKET + 1);
 851         if (spare_ndx > hashp->OVFL_POINT) {
 852                 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
 853                 hashp->OVFL_POINT = spare_ndx;
 854         }
 855
 856         if (new_bucket > (u_int32_t) hashp->HIGH_MASK) {
 857                 /* Starting a new doubling */
 858                 hashp->LOW_MASK = hashp->HIGH_MASK;
 859                 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
 860         }
 861         /* Relocate records to the new bucket */
 862         return (__split_page(hashp, old_bucket, new_bucket));
 863 }
 864
 865 /*
 866  * If realloc guarantees that the pointer is not destroyed if the realloc
 867  * fails, then this routine can go away.
 868  */
 869 static void *
 870 hash_realloc(p_ptr, oldsize, newsize)
 871         SEGMENT **p_ptr;
 872         int oldsize, newsize;
 873 {
 874         register void *p;
 875
 876         if (p = malloc(newsize)) {
 877                 memmove(p, *p_ptr, oldsize);
 878                 memset((char *)p + oldsize, 0, newsize - oldsize);
 879                 free(*p_ptr);
 880                 *p_ptr = p;
 881         }
 882         return (p);
 883 }
 884
 885 extern u_int32_t
 886 __call_hash(hashp, k, len)
 887         HTAB *hashp;
 888         char *k;
 889         int len;
 890 {
 891         int n, bucket;
 892
 893         n = hashp->hash(k, len);
 894         bucket = n & hashp->HIGH_MASK;
 895         if (bucket > hashp->MAX_BUCKET)
 896                 bucket = bucket & hashp->LOW_MASK;
 897         return (bucket);
 898 }
 899
 900 /*
 901  * Allocate segment table.  On error, destroy the table and set errno.
 902  *
 903  * Returns 0 on success
 904  */
 905 static int
 906 alloc_segs(hashp, nsegs)
 907         HTAB *hashp;
 908         int nsegs;
 909 {
 910         register int i;
 911         register SEGMENT store;
 912
 913         int save_errno;
 914
 915         if ((hashp->dir =
 916             (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
 917                 save_errno = errno;
 918                 (void)hdestroy(hashp);
 919                 errno = save_errno;
 920                 return (-1);
 921         }
 922         /* Allocate segments */
 923         if ((store =
 924             (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
 925                 save_errno = errno;
 926                 (void)hdestroy(hashp);
 927                 errno = save_errno;
 928                 return (-1);
 929         }
 930         for (i = 0; i < nsegs; i++, hashp->nsegs++)
 931                 hashp->dir[i] = &store[i << hashp->SSHIFT];
 932         return (0);
 933 }
 934
 935 #if BYTE_ORDER == LITTLE_ENDIAN
 936 /*
 937  * Hashp->hdr needs to be byteswapped.
 938  */
 939 static void
 940 swap_header_copy(srcp, destp)
 941         HASHHDR *srcp, *destp;
 942 {
 943         int i;
 944
 945         P_32_COPY(srcp->magic, destp->magic);
 946         P_32_COPY(srcp->version, destp->version);
 947         P_32_COPY(srcp->lorder, destp->lorder);
 948         P_32_COPY(srcp->bsize, destp->bsize);
 949         P_32_COPY(srcp->bshift, destp->bshift);
 950         P_32_COPY(srcp->dsize, destp->dsize);
 951         P_32_COPY(srcp->ssize, destp->ssize);
 952         P_32_COPY(srcp->sshift, destp->sshift);
 953         P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
 954         P_32_COPY(srcp->last_freed, destp->last_freed);
 955         P_32_COPY(srcp->max_bucket, destp->max_bucket);
 956         P_32_COPY(srcp->high_mask, destp->high_mask);
 957         P_32_COPY(srcp->low_mask, destp->low_mask);
 958         P_32_COPY(srcp->ffactor, destp->ffactor);
 959         P_32_COPY(srcp->nkeys, destp->nkeys);
 960         P_32_COPY(srcp->hdrpages, destp->hdrpages);
 961         P_32_COPY(srcp->h_charkey, destp->h_charkey);
 962         for (i = 0; i < NCACHED; i++) {
 963                 P_32_COPY(srcp->spares[i], destp->spares[i]);
 964                 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
 965         }
 966 }
 967
 968 static void
 969 swap_header(hashp)
 970         HTAB *hashp;
 971 {
 972         HASHHDR *hdrp;
 973         int i;
 974
 975         hdrp = &hashp->hdr;
 976
 977         M_32_SWAP(hdrp->magic);
 978         M_32_SWAP(hdrp->version);
 979         M_32_SWAP(hdrp->lorder);
 980         M_32_SWAP(hdrp->bsize);
 981         M_32_SWAP(hdrp->bshift);
 982         M_32_SWAP(hdrp->dsize);
 983         M_32_SWAP(hdrp->ssize);
 984         M_32_SWAP(hdrp->sshift);
 985         M_32_SWAP(hdrp->ovfl_point);
 986         M_32_SWAP(hdrp->last_freed);
 987         M_32_SWAP(hdrp->max_bucket);
 988         M_32_SWAP(hdrp->high_mask);
 989         M_32_SWAP(hdrp->low_mask);
 990         M_32_SWAP(hdrp->ffactor);
 991         M_32_SWAP(hdrp->nkeys);
 992         M_32_SWAP(hdrp->hdrpages);
 993         M_32_SWAP(hdrp->h_charkey);
 994         for (i = 0; i < NCACHED; i++) {
 995                 M_32_SWAP(hdrp->spares[i]);
 996                 M_16_SWAP(hdrp->bitmaps[i]);
 997         }
 998 }
 999 #endif