newlib/libc/search/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. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #include <sys/param.h>
  34 #if defined(LIBC_SCCS) && !defined(lint)
  35 static char sccsid[] = "@(#)hash.c      8.9 (Berkeley) 6/16/94";
  36 #endif /* LIBC_SCCS and not lint */
  37 #include <sys/cdefs.h>
  38 #include <sys/types.h>
  39
  40 #include <sys/stat.h>
  41
  42 #include <reent.h>
  43 #include <errno.h>
  44 #include <fcntl.h>
  45 #include <stdio.h>
  46 #include <stdlib.h>
  47 #include <string.h>
  48 #include <unistd.h>
  49 #ifdef DEBUG
  50 #include <assert.h>
  51 #endif
  52
  53 #define __DBINTERFACE_PRIVATE   /* activate prototypes from db_local.h */
  54 #include "db_local.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 *, u_int);
  64 static int   hash_fd(const DB *);
  65 static int   hash_get(const DB *, const DBT *, DBT *, u_int);
  66 static int   hash_put(const DB *, DBT *, const DBT *, u_int);
  67 static void *hash_realloc(SEGMENT **, int, int);
  68 static int   hash_seq(const DB *, DBT *, DBT *, u_int);
  69 static int   hash_sync(const DB *, u_int);
  70 static int   hdestroy(HTAB *);
  71 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
  72 static int   init_htab(HTAB *, int);
  73 #if (BYTE_ORDER == LITTLE_ENDIAN)
  74 static void  swap_header(HTAB *);
  75 static void  swap_header_copy(HASHHDR *, HASHHDR *);
  76 #endif
  77
  78 /* Macros for min/max.  */
  79 #ifndef MIN
  80 #define MIN(a,b) (((a)<(b))?(a):(b))
  81 #endif
  82 #ifndef MAX
  83 #define MAX(a,b) (((a)>(b))?(a):(b))
  84 #endif
  85
  86 /* Fast arithmetic, relying on powers of 2, */
  87 #define MOD(x, y)               ((x) & ((y) - 1))
  88
  89 #define RETURN_ERROR(ERR, LOC)  { save_errno = ERR; goto LOC; }
  90
  91 /* Return values */
  92 #define SUCCESS  (0)
  93 #define ERROR   (-1)
  94 #define ABNORMAL (1)
  95
  96 #ifdef HASH_STATISTICS
  97 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
  98 #endif
  99
 100 /************************** INTERFACE ROUTINES ***************************/
 101 /* OPEN/CLOSE */
 102
 103 extern DB *
 104 __hash_open (const char *file,
 105         int flags,
 106         int mode,
 107         int dflags,
 108         const HASHINFO *info)   /* Special directives for create */
 109 {
 110         HTAB *hashp;
 111
 112 #ifdef __USE_INTERNAL_STAT64
 113         struct stat64 statbuf;
 114 #else
 115         struct stat statbuf;
 116 #endif
 117         DB *dbp;
 118         int bpages, hdrsize, new_table, nsegs, save_errno;
 119
 120         if ((flags & O_ACCMODE) == O_WRONLY) {
 121                 errno = EINVAL;
 122                 return (NULL);
 123         }
 124
 125         if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
 126                 return (NULL);
 127         hashp->fp = -1;
 128
 129         /*
 130          * Even if user wants write only, we need to be able to read
 131          * the actual file, so we need to open it read/write. But, the
 132          * field in the hashp structure needs to be accurate so that
 133          * we can check accesses.
 134          */
 135         hashp->flags = flags;
 136
 137         new_table = 0;
 138         if (!file || (flags & O_TRUNC) ||
 139 #ifdef __USE_INTERNAL_STAT64
 140             (_stat64_r(_REENT, file, &statbuf) && (errno == ENOENT))) {
 141 #else
 142             (_stat_r(_REENT, file, &statbuf) && (errno == ENOENT))) {
 143 #endif
 144                 if (errno == ENOENT)
 145                         errno = 0; /* Just in case someone looks at errno */
 146                 new_table = 1;
 147         }
 148         if (file) {
 149                 if ((hashp->fp = open(file, flags, mode)) == -1)
 150                         RETURN_ERROR(errno, error0);
 151
 152                 /* if the .db file is empty, and we had permission to create
 153                    a new .db file, then reinitialize the database */
 154                 if ((flags & O_CREAT) &&
 155 #ifdef __USE_INTERNAL_STAT64
 156                      _fstat64_r(_REENT, hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
 157 #else
 158                      _fstat_r(_REENT, hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
 159 #endif
 160                         new_table = 1;
 161
 162 #ifdef HAVE_FCNTL
 163                 (void)fcntl(hashp->fp, F_SETFD, 1);
 164 #endif
 165         }
 166         if (new_table) {
 167                 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
 168                         RETURN_ERROR(errno, error1);
 169         } else {
 170                 /* Table already exists */
 171                 if (info && info->hash)
 172                         hashp->hash = info->hash;
 173                 else
 174                         hashp->hash = __default_hash;
 175
 176                 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
 177 #if (BYTE_ORDER == LITTLE_ENDIAN)
 178                 swap_header(hashp);
 179 #endif
 180                 if (hdrsize == -1)
 181                         RETURN_ERROR(errno, error1);
 182                 if (hdrsize != sizeof(HASHHDR))
 183                         RETURN_ERROR(EFTYPE, error1);
 184                 /* Verify file type, versions and hash function */
 185                 if (hashp->MAGIC != HASHMAGIC)
 186                         RETURN_ERROR(EFTYPE, error1);
 187 #define OLDHASHVERSION  1
 188                 if (hashp->HASH_VERSION != HASHVERSION &&
 189                     hashp->HASH_VERSION != OLDHASHVERSION)
 190                         RETURN_ERROR(EFTYPE, error1);
 191                 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
 192                         RETURN_ERROR(EFTYPE, error1);
 193                 /* Check bucket size isn't too big for target int. */
 194                 if (hashp->BSIZE > INT_MAX)
 195                         RETURN_ERROR(EFTYPE, error1);
 196                 /*
 197                  * Figure out how many segments we need.  Max_Bucket is the
 198                  * maximum bucket number, so the number of buckets is
 199                  * max_bucket + 1.
 200                  */
 201                 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
 202                          hashp->SGSIZE;
 203                 hashp->nsegs = 0;
 204                 if (alloc_segs(hashp, nsegs))
 205                         /*
 206                          * If alloc_segs fails, table will have been destroyed
 207                          * and errno will have been set.
 208                          */
 209                         return (NULL);
 210                 /* Read in bitmaps */
 211                 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
 212                     (hashp->BSIZE << BYTE_SHIFT) - 1) >>
 213                     (hashp->BSHIFT + BYTE_SHIFT);
 214
 215                 hashp->nmaps = bpages;
 216                 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(__uint32_t *));
 217         }
 218
 219         /* Initialize Buffer Manager */
 220         if (info && info->cachesize)
 221                 __buf_init(hashp, info->cachesize);
 222         else
 223                 __buf_init(hashp, DEF_BUFSIZE);
 224
 225         hashp->new_file = new_table;
 226         hashp->save_file = file && (hashp->flags & O_RDWR);
 227         hashp->cbucket = -1;
 228         if (!(dbp = (DB *)malloc(sizeof(DB)))) {
 229                 save_errno = errno;
 230                 hdestroy(hashp);
 231                 errno = save_errno;
 232                 return (NULL);
 233         }
 234         dbp->internal = hashp;
 235         dbp->close = hash_close;
 236         dbp->del = hash_delete;
 237         dbp->fd = hash_fd;
 238         dbp->get = hash_get;
 239         dbp->put = hash_put;
 240         dbp->seq = hash_seq;
 241         dbp->sync = hash_sync;
 242         dbp->type = DB_HASH;
 243
 244 #ifdef DEBUG
 245         (void)fprintf(stderr,
 246 "%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",
 247             "init_htab:",
 248             "TABLE POINTER   ", hashp,
 249             "BUCKET SIZE     ", hashp->BSIZE,
 250             "BUCKET SHIFT    ", hashp->BSHIFT,
 251             "DIRECTORY SIZE  ", hashp->DSIZE,
 252             "SEGMENT SIZE    ", hashp->SGSIZE,
 253             "SEGMENT SHIFT   ", hashp->SSHIFT,
 254             "FILL FACTOR     ", hashp->FFACTOR,
 255             "MAX BUCKET      ", hashp->MAX_BUCKET,
 256             "OVFL POINT      ", hashp->OVFL_POINT,
 257             "LAST FREED      ", hashp->LAST_FREED,
 258             "HIGH MASK       ", hashp->HIGH_MASK,
 259             "LOW  MASK       ", hashp->LOW_MASK,
 260             "NSEGS           ", hashp->nsegs,
 261             "NKEYS           ", hashp->NKEYS);
 262 #endif
 263 #ifdef HASH_STATISTICS
 264         hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
 265 #endif
 266         return (dbp);
 267
 268 error1:
 269         if (hashp != NULL)
 270                 (void)close(hashp->fp);
 271
 272 error0:
 273         free(hashp);
 274         errno = save_errno;
 275         return (NULL);
 276 }
 277
 278 static int
 279 hash_close(
 280         DB *dbp
 281 )
 282 {
 283         HTAB *hashp;
 284         int retval;
 285
 286         if (!dbp)
 287                 return (ERROR);
 288
 289         hashp = (HTAB *)dbp->internal;
 290         retval = hdestroy(hashp);
 291         free(dbp);
 292         return (retval);
 293 }
 294
 295 static int
 296 hash_fd(
 297         const DB *dbp
 298 )
 299 {
 300         HTAB *hashp;
 301
 302         if (!dbp)
 303                 return (ERROR);
 304
 305         hashp = (HTAB *)dbp->internal;
 306         if (hashp->fp == -1) {
 307                 errno = ENOENT;
 308                 return (-1);
 309         }
 310         return (hashp->fp);
 311 }
 312
 313 /************************** LOCAL CREATION ROUTINES **********************/
 314 static HTAB *
 315 init_hash(
 316         HTAB *hashp,
 317         const char *file,
 318         const HASHINFO *info
 319 )
 320 {
 321 #ifdef __USE_INTERNAL_STAT64
 322         struct stat64 statbuf;
 323 #else
 324         struct stat statbuf;
 325 #endif
 326         int nelem;
 327
 328         nelem = 1;
 329         hashp->NKEYS = 0;
 330        hashp->LORDER = DB_BYTE_ORDER;
 331         hashp->BSIZE = DEF_BUCKET_SIZE;
 332         hashp->BSHIFT = DEF_BUCKET_SHIFT;
 333         hashp->SGSIZE = DEF_SEGSIZE;
 334         hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
 335         hashp->DSIZE = DEF_DIRSIZE;
 336         hashp->FFACTOR = DEF_FFACTOR;
 337         hashp->hash = __default_hash;
 338         memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
 339         memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
 340
 341         /* Fix bucket size to be optimal for file system */
 342         if (file != NULL) {
 343 #ifdef __USE_INTERNAL_STAT64
 344                 if (_stat64_r(_REENT, file, &statbuf))
 345 #else
 346                 if (_stat_r(_REENT, file, &statbuf))
 347 #endif
 348                         return (NULL);
 349                 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
 350                 hashp->BSHIFT = __log2(hashp->BSIZE);
 351         }
 352
 353         if (info) {
 354                 if (info->bsize) {
 355                         /* Round pagesize up to power of 2 */
 356                         hashp->BSHIFT = __log2(info->bsize);
 357                         hashp->BSIZE = 1 << hashp->BSHIFT;
 358                         if (hashp->BSIZE > MAX_BSIZE) {
 359                                 errno = EINVAL;
 360                                 return (NULL);
 361                         }
 362                 }
 363                 if (info->ffactor)
 364                         hashp->FFACTOR = info->ffactor;
 365                 if (info->hash)
 366                         hashp->hash = info->hash;
 367                 if (info->nelem)
 368                         nelem = info->nelem;
 369                 if (info->lorder) {
 370                        if (info->lorder != DB_BIG_ENDIAN &&
 371                            info->lorder != DB_LITTLE_ENDIAN) {
 372                                 errno = EINVAL;
 373                                 return (NULL);
 374                         }
 375                         hashp->LORDER = info->lorder;
 376                 }
 377         }
 378         /* init_htab should destroy the table and set errno if it fails */
 379         if (init_htab(hashp, nelem))
 380                 return (NULL);
 381         else
 382                 return (hashp);
 383 }
 384 /*
 385  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
 386  * the table and set errno, so we just pass the error information along.
 387  *
 388  * Returns 0 on No Error
 389  */
 390 static int
 391 init_htab(
 392         HTAB *hashp,
 393         int nelem
 394 )
 395 {
 396         int nbuckets, nsegs;
 397         int l2;
 398
 399         /*
 400          * Divide number of elements by the fill factor and determine a
 401          * desired number of buckets.  Allocate space for the next greater
 402          * power of two number of buckets.
 403          */
 404         nelem = (nelem - 1) / hashp->FFACTOR + 1;
 405
 406         l2 = __log2(MAX(nelem, 2));
 407         nbuckets = 1 << l2;
 408
 409         hashp->SPARES[l2] = l2 + 1;
 410         hashp->SPARES[l2 + 1] = l2 + 1;
 411         hashp->OVFL_POINT = l2;
 412         hashp->LAST_FREED = 2;
 413
 414         /* First bitmap page is at: splitpoint l2 page offset 1 */
 415         if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
 416                 return (-1);
 417
 418         hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
 419         hashp->HIGH_MASK = (nbuckets << 1) - 1;
 420         hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
 421             hashp->BSHIFT) + 1;
 422
 423         nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
 424         nsegs = 1 << __log2(nsegs);
 425
 426         if (nsegs > hashp->DSIZE)
 427                 hashp->DSIZE = nsegs;
 428         return (alloc_segs(hashp, nsegs));
 429 }
 430
 431 /********************** DESTROY/CLOSE ROUTINES ************************/
 432
 433 /*
 434  * Flushes any changes to the file if necessary and destroys the hashp
 435  * structure, freeing all allocated space.
 436  */
 437 static int
 438 hdestroy(
 439         HTAB *hashp
 440 )
 441 {
 442         int i, save_errno;
 443
 444         save_errno = 0;
 445
 446 #ifdef HASH_STATISTICS
 447         (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
 448             hash_accesses, hash_collisions);
 449         (void)fprintf(stderr, "hdestroy: expansions %ld\n",
 450             hash_expansions);
 451         (void)fprintf(stderr, "hdestroy: overflows %ld\n",
 452             hash_overflows);
 453         (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
 454             hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
 455
 456         for (i = 0; i < NCACHED; i++)
 457                 (void)fprintf(stderr,
 458                     "spares[%d] = %d\n", i, hashp->SPARES[i]);
 459 #endif
 460         /*
 461          * Call on buffer manager to free buffers, and if required,
 462          * write them to disk.
 463          */
 464         if (__buf_free(hashp, 1, hashp->save_file))
 465                 save_errno = errno;
 466         if (hashp->dir) {
 467                 free(*hashp->dir);      /* Free initial segments */
 468                 /* Free extra segments */
 469                 while (hashp->exsegs--)
 470                         free(hashp->dir[--hashp->nsegs]);
 471                 free(hashp->dir);
 472         }
 473         if (flush_meta(hashp) && !save_errno)
 474                 save_errno = errno;
 475         /* Free Bigmaps */
 476         for (i = 0; i < hashp->nmaps; i++)
 477                 if (hashp->mapp[i])
 478                         free(hashp->mapp[i]);
 479
 480         if (hashp->fp != -1)
 481                 (void)close(hashp->fp);
 482
 483         free(hashp);
 484
 485         if (save_errno) {
 486                 errno = save_errno;
 487                 return (ERROR);
 488         }
 489         return (SUCCESS);
 490 }
 491 /*
 492  * Write modified pages to disk
 493  *
 494  * Returns:
 495  *       0 == OK
 496  *      -1 ERROR
 497  */
 498 static int
 499 hash_sync(
 500         const DB *dbp,
 501         u_int flags
 502 )
 503 {
 504         HTAB *hashp;
 505
 506         if (flags != 0) {
 507                 errno = EINVAL;
 508                 return (ERROR);
 509         }
 510
 511         if (!dbp)
 512                 return (ERROR);
 513
 514         hashp = (HTAB *)dbp->internal;
 515         if (!hashp->save_file)
 516                 return (0);
 517         if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
 518                 return (ERROR);
 519         hashp->new_file = 0;
 520         return (0);
 521 }
 522
 523 /*
 524  * Returns:
 525  *       0 == OK
 526  *      -1 indicates that errno should be set
 527  */
 528 static int
 529 flush_meta(
 530         HTAB *hashp
 531 )
 532 {
 533         HASHHDR *whdrp;
 534 #if (BYTE_ORDER == LITTLE_ENDIAN)
 535         HASHHDR whdr;
 536 #endif
 537         int fp, i, wsize;
 538
 539         if (!hashp->save_file)
 540                 return (0);
 541         hashp->MAGIC = HASHMAGIC;
 542         hashp->HASH_VERSION = HASHVERSION;
 543         hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
 544
 545         fp = hashp->fp;
 546         whdrp = &hashp->hdr;
 547 #if (BYTE_ORDER == LITTLE_ENDIAN)
 548         whdrp = &whdr;
 549         swap_header_copy(&hashp->hdr, whdrp);
 550 #endif
 551         if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
 552             ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
 553                 return (-1);
 554         else
 555                 if (wsize != sizeof(HASHHDR)) {
 556                         errno = EFTYPE;
 557                         hashp->error = errno;
 558                         return (-1);
 559                 }
 560         for (i = 0; i < NCACHED; i++)
 561                 if (hashp->mapp[i])
 562                         if (__put_page(hashp, (char *)hashp->mapp[i],
 563                                 hashp->BITMAPS[i], 0, 1))
 564                                 return (-1);
 565         return (0);
 566 }
 567
 568 /*******************************SEARCH ROUTINES *****************************/
 569 /*
 570  * All the access routines return
 571  *
 572  * Returns:
 573  *       0 on SUCCESS
 574  *       1 to indicate an external ERROR (i.e. key not found, etc)
 575  *      -1 to indicate an internal ERROR (i.e. out of memory, etc)
 576  */
 577 static int
 578 hash_get(
 579         const DB *dbp,
 580         const DBT *key,
 581         DBT *data,
 582         u_int flag
 583 )
 584 {
 585         HTAB *hashp;
 586
 587         hashp = (HTAB *)dbp->internal;
 588         if (flag) {
 589                 hashp->error = errno = EINVAL;
 590                 return (ERROR);
 591         }
 592         return (hash_access(hashp, HASH_GET, (DBT *)key, data));
 593 }
 594
 595 static int
 596 hash_put(
 597         const DB *dbp,
 598         DBT *key,
 599         const DBT *data,
 600         u_int flag
 601 )
 602 {
 603         HTAB *hashp;
 604
 605         hashp = (HTAB *)dbp->internal;
 606         if (flag && flag != R_NOOVERWRITE) {
 607                 hashp->error = EINVAL;
 608                 errno = EINVAL;
 609                 return (ERROR);
 610         }
 611         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 612                 hashp->error = errno = EPERM;
 613                 return (ERROR);
 614         }
 615         return (hash_access(hashp, flag == R_NOOVERWRITE ?
 616             HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
 617 }
 618
 619 static int
 620 hash_delete(
 621         const DB *dbp,
 622         const DBT *key,
 623         u_int flag              /* Ignored */
 624 )
 625 {
 626         HTAB *hashp;
 627
 628         hashp = (HTAB *)dbp->internal;
 629         if (flag && flag != R_CURSOR) {
 630                 hashp->error = errno = EINVAL;
 631                 return (ERROR);
 632         }
 633         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 634                 hashp->error = errno = EPERM;
 635                 return (ERROR);
 636         }
 637         return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
 638 }
 639
 640 /*
 641  * Assume that hashp has been set in wrapper routine.
 642  */
 643 static int
 644 hash_access(
 645         HTAB *hashp,
 646         ACTION action,
 647         DBT *key,
 648         DBT *val
 649 )
 650 {
 651         BUFHEAD *rbufp;
 652         BUFHEAD *bufp, *save_bufp;
 653         __uint16_t *bp;
 654         int n, ndx, off, size;
 655         char *kp;
 656         __uint16_t pageno;
 657
 658 #ifdef HASH_STATISTICS
 659         hash_accesses++;
 660 #endif
 661
 662         off = hashp->BSIZE;
 663         size = key->size;
 664         kp = (char *)key->data;
 665         rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
 666         if (!rbufp)
 667                 return (ERROR);
 668         save_bufp = rbufp;
 669
 670         /* Pin the bucket chain */
 671         rbufp->flags |= BUF_PIN;
 672         for (bp = (__uint16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
 673                 if (bp[1] >= REAL_KEY) {
 674                         /* Real key/data pair */
 675                         if (size == off - *bp &&
 676                             memcmp(kp, rbufp->page + *bp, size) == 0)
 677                                 goto found;
 678                         off = bp[1];
 679 #ifdef HASH_STATISTICS
 680                         hash_collisions++;
 681 #endif
 682                         bp += 2;
 683                         ndx += 2;
 684                 } else if (bp[1] == OVFLPAGE) {
 685                         rbufp = __get_buf(hashp, *bp, rbufp, 0);
 686                         if (!rbufp) {
 687                                 save_bufp->flags &= ~BUF_PIN;
 688                                 return (ERROR);
 689                         }
 690                         /* FOR LOOP INIT */
 691                         bp = (__uint16_t *)rbufp->page;
 692                         n = *bp++;
 693                         ndx = 1;
 694                         off = hashp->BSIZE;
 695                 } else if (bp[1] < REAL_KEY) {
 696                         if ((ndx =
 697                             __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
 698                                 goto found;
 699                         if (ndx == -2) {
 700                                 bufp = rbufp;
 701                                 if (!(pageno =
 702                                     __find_last_page(hashp, &bufp))) {
 703                                         ndx = 0;
 704                                         rbufp = bufp;
 705                                         break;  /* FOR */
 706                                 }
 707                                 rbufp = __get_buf(hashp, pageno, bufp, 0);
 708                                 if (!rbufp) {
 709                                         save_bufp->flags &= ~BUF_PIN;
 710                                         return (ERROR);
 711                                 }
 712                                 /* FOR LOOP INIT */
 713                                 bp = (__uint16_t *)rbufp->page;
 714                                 n = *bp++;
 715                                 ndx = 1;
 716                                 off = hashp->BSIZE;
 717                         } else {
 718                                 save_bufp->flags &= ~BUF_PIN;
 719                                 return (ERROR);
 720                         }
 721                 }
 722
 723         /* Not found */
 724         switch (action) {
 725         case HASH_PUT:
 726         case HASH_PUTNEW:
 727                 if (__addel(hashp, rbufp, key, val)) {
 728                         save_bufp->flags &= ~BUF_PIN;
 729                         return (ERROR);
 730                 } else {
 731                         save_bufp->flags &= ~BUF_PIN;
 732                         return (SUCCESS);
 733                 }
 734         case HASH_GET:
 735         case HASH_DELETE:
 736         default:
 737                 save_bufp->flags &= ~BUF_PIN;
 738                 return (ABNORMAL);
 739         }
 740
 741 found:
 742         switch (action) {
 743         case HASH_PUTNEW:
 744                 save_bufp->flags &= ~BUF_PIN;
 745                 return (ABNORMAL);
 746         case HASH_GET:
 747                 bp = (__uint16_t *)rbufp->page;
 748                 if (bp[ndx + 1] < REAL_KEY) {
 749                         if (__big_return(hashp, rbufp, ndx, val, 0))
 750                                 return (ERROR);
 751                 } else {
 752                         val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
 753                         val->size = bp[ndx] - bp[ndx + 1];
 754                 }
 755                 break;
 756         case HASH_PUT:
 757                 if ((__delpair(hashp, rbufp, ndx)) ||
 758                     (__addel(hashp, rbufp, key, val))) {
 759                         save_bufp->flags &= ~BUF_PIN;
 760                         return (ERROR);
 761                 }
 762                 break;
 763         case HASH_DELETE:
 764                 if (__delpair(hashp, rbufp, ndx))
 765                         return (ERROR);
 766                 break;
 767         default:
 768                 abort();
 769         }
 770         save_bufp->flags &= ~BUF_PIN;
 771         return (SUCCESS);
 772 }
 773
 774 static int
 775 hash_seq(
 776         const DB *dbp,
 777         DBT *key,
 778         DBT *data,
 779         u_int flag
 780 )
 781 {
 782         __uint32_t bucket;
 783         BUFHEAD *bufp;
 784         HTAB *hashp;
 785         __uint16_t *bp, ndx;
 786
 787         hashp = (HTAB *)dbp->internal;
 788         if (flag && flag != R_FIRST && flag != R_NEXT) {
 789                 hashp->error = errno = EINVAL;
 790                 return (ERROR);
 791         }
 792 #ifdef HASH_STATISTICS
 793         hash_accesses++;
 794 #endif
 795         if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
 796                 hashp->cbucket = 0;
 797                 hashp->cndx = 1;
 798                 hashp->cpage = NULL;
 799         }
 800
 801         for (bp = NULL; !bp || !bp[0]; ) {
 802                 if (!(bufp = hashp->cpage)) {
 803                         for (bucket = hashp->cbucket;
 804                             bucket <= hashp->MAX_BUCKET;
 805                             bucket++, hashp->cndx = 1) {
 806                                 bufp = __get_buf(hashp, bucket, NULL, 0);
 807                                 if (!bufp)
 808                                         return (ERROR);
 809                                 hashp->cpage = bufp;
 810                                 bp = (__uint16_t *)bufp->page;
 811                                 if (bp[0])
 812                                         break;
 813                         }
 814                         hashp->cbucket = bucket;
 815                         if (hashp->cbucket > hashp->MAX_BUCKET) {
 816                                 hashp->cbucket = -1;
 817                                 return (ABNORMAL);
 818                         }
 819                 } else
 820                         bp = (__uint16_t *)hashp->cpage->page;
 821
 822 #ifdef DEBUG
 823                 assert(bp);
 824                 assert(bufp);
 825 #endif
 826                 while (bp[hashp->cndx + 1] == OVFLPAGE) {
 827                         bufp = hashp->cpage =
 828                             __get_buf(hashp, bp[hashp->cndx], bufp, 0);
 829                         if (!bufp)
 830                                 return (ERROR);
 831                         bp = (__uint16_t *)(bufp->page);
 832                         hashp->cndx = 1;
 833                 }
 834                 if (!bp[0]) {
 835                         hashp->cpage = NULL;
 836                         ++hashp->cbucket;
 837                 }
 838         }
 839         ndx = hashp->cndx;
 840         if (bp[ndx + 1] < REAL_KEY) {
 841                 if (__big_keydata(hashp, bufp, key, data, 1))
 842                         return (ERROR);
 843         } else {
 844                 key->data = (u_char *)hashp->cpage->page + bp[ndx];
 845                 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
 846                 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
 847                 data->size = bp[ndx] - bp[ndx + 1];
 848                 ndx += 2;
 849                 if (ndx > bp[0]) {
 850                         hashp->cpage = NULL;
 851                         hashp->cbucket++;
 852                         hashp->cndx = 1;
 853                 } else
 854                         hashp->cndx = ndx;
 855         }
 856         return (SUCCESS);
 857 }
 858
 859 /********************************* UTILITIES ************************/
 860
 861 /*
 862  * Returns:
 863  *       0 ==> OK
 864  *      -1 ==> Error
 865  */
 866 extern int
 867 __expand_table(
 868         HTAB *hashp
 869 )
 870 {
 871         __uint32_t old_bucket, new_bucket;
 872         int dirsize, new_segnum, spare_ndx;
 873
 874 #ifdef HASH_STATISTICS
 875         hash_expansions++;
 876 #endif
 877         new_bucket = ++hashp->MAX_BUCKET;
 878         old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
 879
 880         new_segnum = new_bucket >> hashp->SSHIFT;
 881
 882         /* Check if we need a new segment */
 883         if (new_segnum >= hashp->nsegs) {
 884                 /* Check if we need to expand directory */
 885                 if (new_segnum >= hashp->DSIZE) {
 886                         /* Reallocate directory */
 887                         dirsize = hashp->DSIZE * sizeof(SEGMENT *);
 888                         if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
 889                                 return (-1);
 890                         hashp->DSIZE = dirsize << 1;
 891                 }
 892                 if ((hashp->dir[new_segnum] =
 893                     (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
 894                         return (-1);
 895                 hashp->exsegs++;
 896                 hashp->nsegs++;
 897         }
 898         /*
 899          * If the split point is increasing (MAX_BUCKET's log base 2
 900          * * increases), we need to copy the current contents of the spare
 901          * split bucket to the next bucket.
 902          */
 903         spare_ndx = __log2(hashp->MAX_BUCKET + 1);
 904         if (spare_ndx > hashp->OVFL_POINT) {
 905                 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
 906                 hashp->OVFL_POINT = spare_ndx;
 907         }
 908
 909         if (new_bucket > hashp->HIGH_MASK) {
 910                 /* Starting a new doubling */
 911                 hashp->LOW_MASK = hashp->HIGH_MASK;
 912                 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
 913         }
 914         /* Relocate records to the new bucket */
 915         return (__split_page(hashp, old_bucket, new_bucket));
 916 }
 917
 918 /*
 919  * If realloc guarantees that the pointer is not destroyed if the realloc
 920  * fails, then this routine can go away.
 921  */
 922 static void *
 923 hash_realloc(
 924         SEGMENT **p_ptr,
 925         int oldsize,
 926         int newsize
 927 )
 928 {
 929         void *p;
 930
 931         if ( (p = malloc(newsize)) ) {
 932                 memmove(p, *p_ptr, oldsize);
 933                 memset((char *)p + oldsize, 0, newsize - oldsize);
 934                 free(*p_ptr);
 935                 *p_ptr = p;
 936         }
 937         return (p);
 938 }
 939
 940 extern __uint32_t
 941 __call_hash(
 942         HTAB *hashp,
 943         char *k,
 944         int len
 945 )
 946 {
 947         int n, bucket;
 948
 949         n = hashp->hash(k, len);
 950         bucket = n & hashp->HIGH_MASK;
 951         if (bucket > hashp->MAX_BUCKET)
 952                 bucket = bucket & hashp->LOW_MASK;
 953         return (bucket);
 954 }
 955
 956 /*
 957  * Allocate segment table.  On error, destroy the table and set errno.
 958  *
 959  * Returns 0 on success
 960  */
 961 static int
 962 alloc_segs(
 963         HTAB *hashp,
 964         int nsegs
 965 )
 966 {
 967         int i;
 968         SEGMENT store;
 969
 970         int save_errno;
 971
 972         if ((hashp->dir =
 973             (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
 974                 save_errno = errno;
 975                 (void)hdestroy(hashp);
 976                 errno = save_errno;
 977                 return (-1);
 978         }
 979         /* Allocate segments */
 980         if ((store =
 981             (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
 982                 save_errno = errno;
 983                 (void)hdestroy(hashp);
 984                 errno = save_errno;
 985                 return (-1);
 986         }
 987         for (i = 0; i < nsegs; i++, hashp->nsegs++)
 988                 hashp->dir[i] = &store[i << hashp->SSHIFT];
 989         return (0);
 990 }
 991
 992 #if (BYTE_ORDER == LITTLE_ENDIAN)
 993 /*
 994  * Hashp->hdr needs to be byteswapped.
 995  */
 996 static void
 997 swap_header_copy(
 998         HASHHDR *srcp,
 999         HASHHDR *destp
1000 )
1001 {
1002         int i;
1003
1004         P_32_COPY(srcp->magic, destp->magic);
1005         P_32_COPY(srcp->version, destp->version);
1006         P_32_COPY(srcp->lorder, destp->lorder);
1007         P_32_COPY(srcp->bsize, destp->bsize);
1008         P_32_COPY(srcp->bshift, destp->bshift);
1009         P_32_COPY(srcp->dsize, destp->dsize);
1010         P_32_COPY(srcp->ssize, destp->ssize);
1011         P_32_COPY(srcp->sshift, destp->sshift);
1012         P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
1013         P_32_COPY(srcp->last_freed, destp->last_freed);
1014         P_32_COPY(srcp->max_bucket, destp->max_bucket);
1015         P_32_COPY(srcp->high_mask, destp->high_mask);
1016         P_32_COPY(srcp->low_mask, destp->low_mask);
1017         P_32_COPY(srcp->ffactor, destp->ffactor);
1018         P_32_COPY(srcp->nkeys, destp->nkeys);
1019         P_32_COPY(srcp->hdrpages, destp->hdrpages);
1020         P_32_COPY(srcp->h_charkey, destp->h_charkey);
1021         for (i = 0; i < NCACHED; i++) {
1022                 P_32_COPY(srcp->spares[i], destp->spares[i]);
1023                 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
1024         }
1025 }
1026
1027 static void
1028 swap_header(
1029         HTAB *hashp
1030 )
1031 {
1032         HASHHDR *hdrp;
1033         int i;
1034
1035         hdrp = &hashp->hdr;
1036
1037         M_32_SWAP(hdrp->magic);
1038         M_32_SWAP(hdrp->version);
1039         M_32_SWAP(hdrp->lorder);
1040         M_32_SWAP(hdrp->bsize);
1041         M_32_SWAP(hdrp->bshift);
1042         M_32_SWAP(hdrp->dsize);
1043         M_32_SWAP(hdrp->ssize);
1044         M_32_SWAP(hdrp->sshift);
1045         M_32_SWAP(hdrp->ovfl_point);
1046         M_32_SWAP(hdrp->last_freed);
1047         M_32_SWAP(hdrp->max_bucket);
1048         M_32_SWAP(hdrp->high_mask);
1049         M_32_SWAP(hdrp->low_mask);
1050         M_32_SWAP(hdrp->ffactor);
1051         M_32_SWAP(hdrp->nkeys);
1052         M_32_SWAP(hdrp->hdrpages);
1053         M_32_SWAP(hdrp->h_charkey);
1054         for (i = 0; i < NCACHED; i++) {
1055                 M_32_SWAP(hdrp->spares[i]);
1056                 M_16_SWAP(hdrp->bitmaps[i]);
1057         }
1058 }
1059 #endif