dbm/src/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. ***REMOVED*** - see
  17  *    ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change
  18  * 4. 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 defined(LIBC_SCCS) && !defined(lint)
  36 static char sccsid[] = "@(#)hash.c      8.9 (Berkeley) 6/16/94";
  37 #endif /* LIBC_SCCS and not lint */
  38
  39 #include "watcomfx.h"
  40
  41 #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh) && !defined(XP_OS2_VACPP)
  42 #include <sys/param.h>
  43 #endif
  44
  45 #if !defined(macintosh)
  46 #ifdef XP_OS2_EMX
  47 #include <sys/types.h>
  48 #endif
  49 #include <sys/stat.h>
  50 #endif
  51
  52 #if defined(macintosh)
  53 #include <unix.h>
  54 #include <unistd.h>
  55 #endif
  56
  57 #include <errno.h>
  58 #include <fcntl.h>
  59 #include <stdio.h>
  60 #include <stdlib.h>
  61 #include <string.h>
  62
  63 #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh) && !defined(XP_OS2_VACPP)
  64 #include <unistd.h>
  65 #endif
  66 #if defined(_WIN32) || defined(_WINDOWS)
  67 #include <windows.h>
  68 #endif
  69
  70 #include <assert.h>
  71
  72 #include "mcom_db.h"
  73 #include "hash.h"
  74 #include "page.h"
  75
  76 /*
  77 #include "extern.h"
  78 */
  79 static int   alloc_segs __P((HTAB *, int));
  80 static int   flush_meta __P((HTAB *));
  81 static int   hash_access __P((HTAB *, ACTION, DBT *, DBT *));
  82 static int   hash_close __P((DB *));
  83 static int   hash_delete __P((const DB *, const DBT *, uint));
  84 static int   hash_fd __P((const DB *));
  85 static int   hash_get __P((const DB *, const DBT *, DBT *, uint));
  86 static int   hash_put __P((const DB *, DBT *, const DBT *, uint));
  87 static void *hash_realloc __P((SEGMENT **, size_t, size_t));
  88 static int   hash_seq __P((const DB *, DBT *, DBT *, uint));
  89 static int   hash_sync __P((const DB *, uint));
  90 static int   hdestroy __P((HTAB *));
  91 static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *));
  92 static int   init_htab __P((HTAB *, int));
  93 #if BYTE_ORDER == LITTLE_ENDIAN
  94 static void  swap_header __P((HTAB *));
  95 static void  swap_header_copy __P((HASHHDR *, HASHHDR *));
  96 #endif
  97
  98 /* Fast arithmetic, relying on powers of 2, */
  99 #define MOD(x, y)               ((x) & ((y) - 1))
 100
 101 #define RETURN_ERROR(ERR, LOC)  { save_errno = ERR; goto LOC; }
 102
 103 /* Return values */
 104 #define SUCCESS  (0)
 105 #define DBM_ERROR       (-1)
 106 #define ABNORMAL (1)
 107
 108 #ifdef HASH_STATISTICS
 109 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
 110 #endif
 111
 112 /* A new Lou (montulli@mozilla.com) routine.
 113  *
 114  * The database is screwed.
 115  *
 116  * This closes the file, flushing buffers as appropriate.
 117  */
 118 static void
 119 __remove_database(DB *dbp)
 120 {
 121         HTAB *hashp = (HTAB *)dbp->internal;
 122
 123         assert(0);
 124
 125         if (!hashp)
 126                 return;
 127         hdestroy(hashp);
 128         dbp->internal = NULL;
 129 }
 130
 131 /************************** INTERFACE ROUTINES ***************************/
 132 /* OPEN/CLOSE */
 133
 134
 135 extern DB *
 136 __hash_open(const char *file, int flags, int mode, const HASHINFO *info, int dflags)
 137 {
 138         HTAB *hashp=NULL;
 139         struct stat statbuf;
 140         DB *dbp;
 141         int bpages, hdrsize, new_table, nsegs, save_errno;
 142
 143         if ((flags & O_ACCMODE) == O_WRONLY) {
 144                 errno = EINVAL;
 145                 return NULL;
 146         }
 147
 148         /* zero the statbuffer so that
 149          * we can check it for a non-zero
 150          * date to see if stat succeeded
 151          */
 152         memset(&statbuf, 0, sizeof(struct stat));
 153
 154         if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB)))) {
 155                 errno = ENOMEM;
 156                 return NULL;
 157         }
 158         hashp->fp = NO_FILE;
 159         if(file)
 160                 hashp->filename = strdup(file);
 161
 162         /*
 163          * Even if user wants write only, we need to be able to read
 164          * the actual file, so we need to open it read/write. But, the
 165          * field in the hashp structure needs to be accurate so that
 166          * we can check accesses.
 167          */
 168         hashp->flags = flags;
 169
 170         new_table = 0;
 171         if (!file || (flags & O_TRUNC)  || (stat(file, &statbuf)  && (errno == ENOENT)))
 172         {
 173                 if (errno == ENOENT)
 174                         errno = 0; /* Just in case someone looks at errno */
 175                 new_table = 1;
 176         }
 177         else if(statbuf.st_mtime && statbuf.st_size == 0)
 178         {
 179                 /* check for a zero length file and delete it
 180                  * if it exists
 181                  */
 182                 new_table = 1;
 183         }
 184         hashp->file_size = statbuf.st_size;
 185
 186         if (file) {
 187 #if defined(_WIN32) || defined(_WINDOWS) || defined (macintosh)  || defined(XP_OS2)
 188                 if ((hashp->fp = DBFILE_OPEN(file, flags | O_BINARY, mode)) == -1)
 189                         RETURN_ERROR(errno, error1);
 190 #else
 191                 if ((hashp->fp = open(file, flags, mode)) == -1)
 192                         RETURN_ERROR(errno, error1);
 193                 (void)fcntl(hashp->fp, F_SETFD, 1);
 194 #endif
 195         }
 196         if (new_table) {
 197                 if (!init_hash(hashp, file, (HASHINFO *)info))
 198                         RETURN_ERROR(errno, error1);
 199         } else {
 200                 /* Table already exists */
 201                 if (info && info->hash)
 202                         hashp->hash = info->hash;
 203                 else
 204                         hashp->hash = __default_hash;
 205
 206                 hdrsize = read(hashp->fp, (char *)&hashp->hdr, sizeof(HASHHDR));
 207                 if (hdrsize == -1)
 208                         RETURN_ERROR(errno, error1);
 209                 if (hdrsize != sizeof(HASHHDR))
 210                         RETURN_ERROR(EFTYPE, error1);
 211 #if BYTE_ORDER == LITTLE_ENDIAN
 212                 swap_header(hashp);
 213 #endif
 214                 /* Verify file type, versions and hash function */
 215                 if (hashp->MAGIC != HASHMAGIC)
 216                         RETURN_ERROR(EFTYPE, error1);
 217 #define OLDHASHVERSION  1
 218                 if (hashp->VERSION != HASHVERSION &&
 219                     hashp->VERSION != OLDHASHVERSION)
 220                         RETURN_ERROR(EFTYPE, error1);
 221                 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
 222                         RETURN_ERROR(EFTYPE, error1);
 223                 if (hashp->NKEYS < 0) /* Old bad database. */
 224                         RETURN_ERROR(EFTYPE, error1);
 225
 226                 /*
 227                  * Figure out how many segments we need.  Max_Bucket is the
 228                  * maximum bucket number, so the number of buckets is
 229                  * max_bucket + 1.
 230                  */
 231                 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
 232                          hashp->SGSIZE;
 233                 hashp->nsegs = 0;
 234                 if (alloc_segs(hashp, nsegs))
 235                         /* If alloc_segs fails, errno will have been set.  */
 236                         RETURN_ERROR(errno, error1);
 237                 /* Read in bitmaps */
 238                 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
 239                     (hashp->BSIZE << BYTE_SHIFT) - 1) >>
 240                     (hashp->BSHIFT + BYTE_SHIFT);
 241
 242                 hashp->nmaps = bpages;
 243                 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(uint32 *));
 244         }
 245
 246         /* Initialize Buffer Manager */
 247         if (info && info->cachesize)
 248                 __buf_init(hashp, (int32) info->cachesize);
 249         else
 250                 __buf_init(hashp, DEF_BUFSIZE);
 251
 252         hashp->new_file = new_table;
 253 #ifdef macintosh
 254         hashp->save_file = file && !(hashp->flags & O_RDONLY);
 255 #else
 256         hashp->save_file = file && (hashp->flags & O_RDWR);
 257 #endif
 258         hashp->cbucket = -1;
 259         if (!(dbp = (DB *)malloc(sizeof(DB)))) {
 260                 RETURN_ERROR(ENOMEM, error1);
 261         }
 262         dbp->internal = hashp;
 263         dbp->close = hash_close;
 264         dbp->del = hash_delete;
 265         dbp->fd = hash_fd;
 266         dbp->get = hash_get;
 267         dbp->put = hash_put;
 268         dbp->seq = hash_seq;
 269         dbp->sync = hash_sync;
 270         dbp->type = DB_HASH;
 271
 272 #ifdef HASH_STATISTICS
 273         hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
 274 #endif
 275         return (dbp);
 276
 277 error1:
 278         hdestroy(hashp);
 279         errno = save_errno;
 280         return (NULL);
 281 }
 282
 283 static int
 284 hash_close(DB *dbp)
 285 {
 286         HTAB *hashp;
 287         int retval;
 288
 289         if (!dbp)
 290                 return (DBM_ERROR);
 291
 292         hashp = (HTAB *)dbp->internal;
 293         if(!hashp)
 294                 return (DBM_ERROR);
 295
 296         retval = hdestroy(hashp);
 297         free(dbp);
 298         return (retval);
 299 }
 300
 301 static int hash_fd(const DB *dbp)
 302 {
 303         HTAB *hashp;
 304
 305         if (!dbp)
 306                 return (DBM_ERROR);
 307
 308         hashp = (HTAB *)dbp->internal;
 309         if(!hashp)
 310                 return (DBM_ERROR);
 311
 312         if (hashp->fp == -1) {
 313                 errno = ENOENT;
 314                 return (-1);
 315         }
 316         return (hashp->fp);
 317 }
 318
 319 /************************** LOCAL CREATION ROUTINES **********************/
 320 static HTAB *
 321 init_hash(HTAB *hashp, const char *file, HASHINFO *info)
 322 {
 323         struct stat statbuf;
 324         int nelem;
 325
 326         nelem = 1;
 327         hashp->NKEYS = 0;
 328         hashp->LORDER = BYTE_ORDER;
 329         hashp->BSIZE = DEF_BUCKET_SIZE;
 330         hashp->BSHIFT = DEF_BUCKET_SHIFT;
 331         hashp->SGSIZE = DEF_SEGSIZE;
 332         hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
 333         hashp->DSIZE = DEF_DIRSIZE;
 334         hashp->FFACTOR = DEF_FFACTOR;
 335         hashp->hash = __default_hash;
 336         memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
 337         memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
 338
 339         /* Fix bucket size to be optimal for file system */
 340         if (file != NULL) {
 341                 if (stat(file, &statbuf))
 342                         return (NULL);
 343
 344 #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh) && !defined(VMS) && !defined(XP_OS2)
 345 #if defined(__QNX__) && !defined(__QNXNTO__)
 346                 hashp->BSIZE = 512; /* preferred blk size on qnx4 */
 347 #else
 348                 hashp->BSIZE = statbuf.st_blksize;
 349 #endif
 350
 351                 /* new code added by Lou to reduce block
 352                  * size down below MAX_BSIZE
 353                  */
 354                 if (hashp->BSIZE > MAX_BSIZE)
 355                         hashp->BSIZE = MAX_BSIZE;
 356 #endif
 357                 hashp->BSHIFT = __log2((uint32)hashp->BSIZE);
 358         }
 359
 360         if (info) {
 361                 if (info->bsize) {
 362                         /* Round pagesize up to power of 2 */
 363                         hashp->BSHIFT = __log2(info->bsize);
 364                         hashp->BSIZE = 1 << hashp->BSHIFT;
 365                         if (hashp->BSIZE > MAX_BSIZE) {
 366                                 errno = EINVAL;
 367                                 return (NULL);
 368                         }
 369                 }
 370                 if (info->ffactor)
 371                         hashp->FFACTOR = info->ffactor;
 372                 if (info->hash)
 373                         hashp->hash = info->hash;
 374                 if (info->nelem)
 375                         nelem = info->nelem;
 376                 if (info->lorder) {
 377                         if (info->lorder != BIG_ENDIAN &&
 378                             info->lorder != LITTLE_ENDIAN) {
 379                                 errno = EINVAL;
 380                                 return (NULL);
 381                         }
 382                         hashp->LORDER = info->lorder;
 383                 }
 384         }
 385         /* init_htab sets errno if it fails */
 386         if (init_htab(hashp, nelem))
 387                 return (NULL);
 388         else
 389                 return (hashp);
 390 }
 391 /*
 392  * This calls alloc_segs which may run out of memory.  Alloc_segs will
 393  * set errno, so we just pass the error information along.
 394  *
 395  * Returns 0 on No Error
 396  */
 397 static int
 398 init_htab(HTAB *hashp, int nelem)
 399 {
 400         register int nbuckets, nsegs;
 401         int l2;
 402
 403         /*
 404          * Divide number of elements by the fill factor and determine a
 405          * desired number of buckets.  Allocate space for the next greater
 406          * power of two number of buckets.
 407          */
 408         nelem = (nelem - 1) / hashp->FFACTOR + 1;
 409
 410         l2 = __log2((uint32)PR_MAX(nelem, 2));
 411         nbuckets = 1 << l2;
 412
 413         hashp->SPARES[l2] = l2 + 1;
 414         hashp->SPARES[l2 + 1] = l2 + 1;
 415         hashp->OVFL_POINT = l2;
 416         hashp->LAST_FREED = 2;
 417
 418         /* First bitmap page is at: splitpoint l2 page offset 1 */
 419         if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0))
 420                 return (-1);
 421
 422         hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
 423         hashp->HIGH_MASK = (nbuckets << 1) - 1;
 424         hashp->HDRPAGES = ((PR_MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
 425             hashp->BSHIFT) + 1;
 426
 427         nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
 428         nsegs = 1 << __log2((uint32)nsegs);
 429
 430         if (nsegs > hashp->DSIZE)
 431                 hashp->DSIZE = nsegs;
 432         return (alloc_segs(hashp, nsegs));
 433 }
 434
 435 /********************** DESTROY/CLOSE ROUTINES ************************/
 436
 437 /*
 438  * Flushes any changes to the file if necessary and destroys the hashp
 439  * structure, freeing all allocated space.
 440  */
 441 static int
 442 hdestroy(HTAB *hashp)
 443 {
 444         int i, save_errno;
 445
 446         save_errno = 0;
 447
 448 #ifdef HASH_STATISTICS
 449         (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
 450             hash_accesses, hash_collisions);
 451         (void)fprintf(stderr, "hdestroy: expansions %ld\n",
 452             hash_expansions);
 453         (void)fprintf(stderr, "hdestroy: overflows %ld\n",
 454             hash_overflows);
 455         (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
 456             hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
 457
 458         for (i = 0; i < NCACHED; i++)
 459                 (void)fprintf(stderr,
 460                     "spares[%d] = %d\n", i, hashp->SPARES[i]);
 461 #endif
 462         /*
 463          * Call on buffer manager to free buffers, and if required,
 464          * write them to disk.
 465          */
 466         if (__buf_free(hashp, 1, hashp->save_file))
 467                 save_errno = errno;
 468         if (hashp->dir) {
 469                 free(*hashp->dir);      /* Free initial segments */
 470                 /* Free extra segments */
 471                 while (hashp->exsegs--)
 472                         free(hashp->dir[--hashp->nsegs]);
 473                 free(hashp->dir);
 474         }
 475         if (flush_meta(hashp) && !save_errno)
 476                 save_errno = errno;
 477         /* Free Bigmaps */
 478         for (i = 0; i < hashp->nmaps; i++)
 479                 if (hashp->mapp[i])
 480                         free(hashp->mapp[i]);
 481
 482         if (hashp->fp != -1)
 483                 (void)close(hashp->fp);
 484
 485         if(hashp->filename) {
 486 #if defined(_WIN32) || defined(_WINDOWS) || defined(XP_OS2)
 487                 if (hashp->is_temp)
 488                         (void)unlink(hashp->filename);
 489 #endif
 490                 free(hashp->filename);
 491         }
 492         if (hashp->tmp_buf)
 493             free(hashp->tmp_buf);
 494         if (hashp->tmp_key)
 495             free(hashp->tmp_key);
 496         free(hashp);
 497         if (save_errno) {
 498                 errno = save_errno;
 499                 return (DBM_ERROR);
 500         }
 501         return (SUCCESS);
 502 }
 503
 504 #if defined(_WIN32) || defined(_WINDOWS)
 505 /*
 506  * Close and reopen file to force file length update on windows.
 507  *
 508  * Returns:
 509  *       0 == OK
 510  *      -1 DBM_ERROR
 511  */
 512 static int
 513 update_EOF(HTAB *hashp)
 514 {
 515 #if defined(DBM_REOPEN_ON_FLUSH)
 516         char *      file       = hashp->filename;
 517         off_t       file_size;
 518         int         flags;
 519         int         mode       = -1;
 520         struct stat statbuf;
 521
 522         memset(&statbuf, 0, sizeof statbuf);
 523
 524         /* make sure we won't lose the file by closing it. */
 525         if (!file || (stat(file, &statbuf)  && (errno == ENOENT)))  {
 526                 /* pretend we did it. */
 527                 return 0;
 528         }
 529
 530         (void)close(hashp->fp);
 531
 532         flags = hashp->flags & ~(O_TRUNC | O_CREAT | O_EXCL);
 533
 534         if ((hashp->fp = DBFILE_OPEN(file, flags | O_BINARY, mode)) == -1)
 535                 return -1;
 536         file_size = lseek(hashp->fp, (off_t)0, SEEK_END);
 537         if (file_size == -1)
 538                 return -1;
 539         hashp->file_size = file_size;
 540         return 0;
 541 #else
 542         int    fd        = hashp->fp;
 543         off_t  file_size = lseek(fd, (off_t)0, SEEK_END);
 544         HANDLE handle    = (HANDLE)_get_osfhandle(fd);
 545         BOOL   cool      = FlushFileBuffers(handle);
 546 #ifdef DEBUG3
 547         if (!cool) {
 548                 DWORD err = GetLastError();
 549                 (void)fprintf(stderr,
 550                         "FlushFileBuffers failed, last error = %d, 0x%08x\n",
 551                         err, err);
 552         }
 553 #endif
 554         if (file_size == -1)
 555                 return -1;
 556         hashp->file_size = file_size;
 557         return cool ? 0 : -1;
 558 #endif
 559 }
 560 #endif
 561
 562 /*
 563  * Write modified pages to disk
 564  *
 565  * Returns:
 566  *       0 == OK
 567  *      -1 DBM_ERROR
 568  */
 569 static int
 570 hash_sync(const DB *dbp, uint flags)
 571 {
 572         HTAB *hashp;
 573
 574         if (flags != 0) {
 575                 errno = EINVAL;
 576                 return (DBM_ERROR);
 577         }
 578
 579         if (!dbp)
 580                 return (DBM_ERROR);
 581
 582         hashp = (HTAB *)dbp->internal;
 583         if(!hashp)
 584                 return (DBM_ERROR);
 585
 586         if (!hashp->save_file)
 587                 return (0);
 588         if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
 589                 return (DBM_ERROR);
 590 #if defined(_WIN32) || defined(_WINDOWS)
 591         if (hashp->updateEOF && hashp->filename && !hashp->is_temp) {
 592                 int status = update_EOF(hashp);
 593                 hashp->updateEOF = 0;
 594                 if (status)
 595                         return status;
 596         }
 597 #endif
 598         hashp->new_file = 0;
 599         return (0);
 600 }
 601
 602 /*
 603  * Returns:
 604  *       0 == OK
 605  *      -1 indicates that errno should be set
 606  */
 607 static int
 608 flush_meta(HTAB *hashp)
 609 {
 610         HASHHDR *whdrp;
 611 #if BYTE_ORDER == LITTLE_ENDIAN
 612         HASHHDR whdr;
 613 #endif
 614         int fp, i, wsize;
 615
 616         if (!hashp->save_file)
 617                 return (0);
 618         hashp->MAGIC = HASHMAGIC;
 619         hashp->VERSION = HASHVERSION;
 620         hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
 621
 622         fp = hashp->fp;
 623         whdrp = &hashp->hdr;
 624 #if BYTE_ORDER == LITTLE_ENDIAN
 625         whdrp = &whdr;
 626         swap_header_copy(&hashp->hdr, whdrp);
 627 #endif
 628         if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
 629             ((wsize = write(fp, (char*)whdrp, sizeof(HASHHDR))) == -1))
 630                 return (-1);
 631         else
 632                 if (wsize != sizeof(HASHHDR)) {
 633                         errno = EFTYPE;
 634                         hashp->dbmerrno = errno;
 635                         return (-1);
 636                 }
 637         for (i = 0; i < NCACHED; i++)
 638                 if (hashp->mapp[i])
 639                         if (__put_page(hashp, (char *)hashp->mapp[i],
 640                                 hashp->BITMAPS[i], 0, 1))
 641                                 return (-1);
 642         return (0);
 643 }
 644
 645 /*******************************SEARCH ROUTINES *****************************/
 646 /*
 647  * All the access routines return
 648  *
 649  * Returns:
 650  *       0 on SUCCESS
 651  *       1 to indicate an external DBM_ERROR (i.e. key not found, etc)
 652  *      -1 to indicate an internal DBM_ERROR (i.e. out of memory, etc)
 653  */
 654 static int
 655 hash_get(
 656         const DB *dbp,
 657         const DBT *key,
 658         DBT *data,
 659         uint flag)
 660 {
 661         HTAB *hashp;
 662         int rv;
 663
 664         hashp = (HTAB *)dbp->internal;
 665         if (!hashp)
 666                 return (DBM_ERROR);
 667
 668         if (flag) {
 669                 hashp->dbmerrno = errno = EINVAL;
 670                 return (DBM_ERROR);
 671         }
 672
 673         rv = hash_access(hashp, HASH_GET, (DBT *)key, data);
 674
 675         if(rv == DATABASE_CORRUPTED_ERROR)
 676           {
 677 #if defined(unix) && defined(DEBUG)
 678                 printf("\n\nDBM Database has been corrupted, tell Lou...\n\n");
 679 #endif
 680                 __remove_database((DB *)dbp);
 681           }
 682
 683         return(rv);
 684 }
 685
 686 static int
 687 hash_put(
 688         const DB *dbp,
 689         DBT *key,
 690         const DBT *data,
 691         uint flag)
 692 {
 693         HTAB *hashp;
 694         int rv;
 695
 696         hashp = (HTAB *)dbp->internal;
 697         if (!hashp)
 698                 return (DBM_ERROR);
 699
 700         if (flag && flag != R_NOOVERWRITE) {
 701                 hashp->dbmerrno = errno = EINVAL;
 702                 return (DBM_ERROR);
 703         }
 704         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 705                 hashp->dbmerrno = errno = EPERM;
 706                 return (DBM_ERROR);
 707         }
 708
 709         rv =  hash_access(hashp, flag == R_NOOVERWRITE ?
 710             HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data);
 711
 712         if(rv == DATABASE_CORRUPTED_ERROR)
 713           {
 714 #if defined(unix) && defined(DEBUG)
 715                 printf("\n\nDBM Database has been corrupted, tell Lou...\n\n");
 716 #endif
 717                 __remove_database((DB *)dbp);
 718           }
 719
 720         return(rv);
 721 }
 722
 723 static int
 724 hash_delete(
 725         const DB *dbp,
 726         const DBT *key,
 727         uint flag)              /* Ignored */
 728 {
 729         HTAB *hashp;
 730         int rv;
 731
 732         hashp = (HTAB *)dbp->internal;
 733         if (!hashp)
 734                 return (DBM_ERROR);
 735
 736         if (flag && flag != R_CURSOR) {
 737                 hashp->dbmerrno = errno = EINVAL;
 738                 return (DBM_ERROR);
 739         }
 740         if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
 741                 hashp->dbmerrno = errno = EPERM;
 742                 return (DBM_ERROR);
 743         }
 744         rv = hash_access(hashp, HASH_DELETE, (DBT *)key, NULL);
 745
 746         if(rv == DATABASE_CORRUPTED_ERROR)
 747           {
 748 #if defined(unix) && defined(DEBUG)
 749                 printf("\n\nDBM Database has been corrupted, tell Lou...\n\n");
 750 #endif
 751                 __remove_database((DB *)dbp);
 752           }
 753
 754         return(rv);
 755 }
 756
 757 #define MAX_OVERFLOW_HASH_ACCESS_LOOPS 2000
 758 /*
 759  * Assume that hashp has been set in wrapper routine.
 760  */
 761 static int
 762 hash_access(
 763         HTAB *hashp,
 764         ACTION action,
 765         DBT *key, DBT *val)
 766 {
 767         register BUFHEAD *rbufp;
 768         BUFHEAD *bufp, *save_bufp;
 769         register uint16 *bp;
 770         register long n, ndx, off;
 771         register size_t size;
 772         register char *kp;
 773         uint16 pageno;
 774         uint32 ovfl_loop_count=0;
 775     int32 last_overflow_page_no = -1;
 776
 777 #ifdef HASH_STATISTICS
 778         hash_accesses++;
 779 #endif
 780
 781         off = hashp->BSIZE;
 782         size = key->size;
 783         kp = (char *)key->data;
 784         rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
 785         if (!rbufp)
 786                 return (DATABASE_CORRUPTED_ERROR);
 787         save_bufp = rbufp;
 788
 789         /* Pin the bucket chain */
 790         rbufp->flags |= BUF_PIN;
 791         for (bp = (uint16 *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
 792         {
 793
 794                 if (bp[1] >= REAL_KEY) {
 795                         /* Real key/data pair */
 796                         if (size == (unsigned long)(off - *bp) &&
 797                             memcmp(kp, rbufp->page + *bp, size) == 0)
 798                                 goto found;
 799                         off = bp[1];
 800 #ifdef HASH_STATISTICS
 801                         hash_collisions++;
 802 #endif
 803                         bp += 2;
 804                         ndx += 2;
 805                 } else if (bp[1] == OVFLPAGE) {
 806
 807             /* database corruption: overflow loop detection */
 808             if(last_overflow_page_no == (int32)*bp)
 809                         return (DATABASE_CORRUPTED_ERROR);
 810
 811             last_overflow_page_no = *bp;
 812
 813                         rbufp = __get_buf(hashp, *bp, rbufp, 0);
 814                         if (!rbufp) {
 815                                 save_bufp->flags &= ~BUF_PIN;
 816                                 return (DBM_ERROR);
 817                         }
 818
 819             ovfl_loop_count++;
 820             if(ovfl_loop_count > MAX_OVERFLOW_HASH_ACCESS_LOOPS)
 821                         return (DATABASE_CORRUPTED_ERROR);
 822
 823                         /* FOR LOOP INIT */
 824                         bp = (uint16 *)rbufp->page;
 825                         n = *bp++;
 826                         ndx = 1;
 827                         off = hashp->BSIZE;
 828                 } else if (bp[1] < REAL_KEY) {
 829                         if ((ndx =
 830                             __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0)
 831                                 goto found;
 832                         if (ndx == -2) {
 833                                 bufp = rbufp;
 834                                 if (!(pageno =
 835                                     __find_last_page(hashp, &bufp))) {
 836                                         ndx = 0;
 837                                         rbufp = bufp;
 838                                         break;  /* FOR */
 839                                 }
 840                                 rbufp = __get_buf(hashp, pageno, bufp, 0);
 841                                 if (!rbufp) {
 842                                         save_bufp->flags &= ~BUF_PIN;
 843                                         return (DBM_ERROR);
 844                                 }
 845                                 /* FOR LOOP INIT */
 846                                 bp = (uint16 *)rbufp->page;
 847                                 n = *bp++;
 848                                 ndx = 1;
 849                                 off = hashp->BSIZE;
 850                         } else {
 851                                 save_bufp->flags &= ~BUF_PIN;
 852                                 return (DBM_ERROR);
 853
 854                         }
 855                 }
 856         }
 857
 858         /* Not found */
 859         switch (action) {
 860         case HASH_PUT:
 861         case HASH_PUTNEW:
 862                 if (__addel(hashp, rbufp, key, val)) {
 863                         save_bufp->flags &= ~BUF_PIN;
 864                         return (DBM_ERROR);
 865                 } else {
 866                         save_bufp->flags &= ~BUF_PIN;
 867                         return (SUCCESS);
 868                 }
 869         case HASH_GET:
 870         case HASH_DELETE:
 871         default:
 872                 save_bufp->flags &= ~BUF_PIN;
 873                 return (ABNORMAL);
 874         }
 875
 876 found:
 877         switch (action) {
 878         case HASH_PUTNEW:
 879                 save_bufp->flags &= ~BUF_PIN;
 880                 return (ABNORMAL);
 881         case HASH_GET:
 882                 bp = (uint16 *)rbufp->page;
 883                 if (bp[ndx + 1] < REAL_KEY) {
 884                         if (__big_return(hashp, rbufp, ndx, val, 0))
 885                                 return (DBM_ERROR);
 886                 } else {
 887                         val->data = (uint8 *)rbufp->page + (int)bp[ndx + 1];
 888                         val->size = bp[ndx] - bp[ndx + 1];
 889                 }
 890                 break;
 891         case HASH_PUT:
 892                 if ((__delpair(hashp, rbufp, ndx)) ||
 893                     (__addel(hashp, rbufp, key, val))) {
 894                         save_bufp->flags &= ~BUF_PIN;
 895                         return (DBM_ERROR);
 896                 }
 897                 break;
 898         case HASH_DELETE:
 899                 if (__delpair(hashp, rbufp, ndx))
 900                         return (DBM_ERROR);
 901                 break;
 902         default:
 903                 abort();
 904         }
 905         save_bufp->flags &= ~BUF_PIN;
 906         return (SUCCESS);
 907 }
 908
 909 static int
 910 hash_seq(
 911         const DB *dbp,
 912         DBT *key, DBT *data,
 913         uint flag)
 914 {
 915         register uint32 bucket;
 916         register BUFHEAD *bufp;
 917         HTAB *hashp;
 918         uint16 *bp, ndx;
 919
 920         hashp = (HTAB *)dbp->internal;
 921         if (!hashp)
 922                 return (DBM_ERROR);
 923
 924         if (flag && flag != R_FIRST && flag != R_NEXT) {
 925                 hashp->dbmerrno = errno = EINVAL;
 926                 return (DBM_ERROR);
 927         }
 928 #ifdef HASH_STATISTICS
 929         hash_accesses++;
 930 #endif
 931         if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
 932                 hashp->cbucket = 0;
 933                 hashp->cndx = 1;
 934                 hashp->cpage = NULL;
 935         }
 936
 937         for (bp = NULL; !bp || !bp[0]; ) {
 938                 if (!(bufp = hashp->cpage)) {
 939                         for (bucket = hashp->cbucket;
 940                             bucket <= (uint32)hashp->MAX_BUCKET;
 941                             bucket++, hashp->cndx = 1) {
 942                                 bufp = __get_buf(hashp, bucket, NULL, 0);
 943                                 if (!bufp)
 944                                         return (DBM_ERROR);
 945                                 hashp->cpage = bufp;
 946                                 bp = (uint16 *)bufp->page;
 947                                 if (bp[0])
 948                                         break;
 949                         }
 950                         hashp->cbucket = bucket;
 951                         if (hashp->cbucket > hashp->MAX_BUCKET) {
 952                                 hashp->cbucket = -1;
 953                                 return (ABNORMAL);
 954                         }
 955                 } else
 956                         bp = (uint16 *)hashp->cpage->page;
 957
 958 #ifdef DEBUG
 959                 assert(bp);
 960                 assert(bufp);
 961 #endif
 962                 while (bp[hashp->cndx + 1] == OVFLPAGE) {
 963                         bufp = hashp->cpage =
 964                             __get_buf(hashp, bp[hashp->cndx], bufp, 0);
 965                         if (!bufp)
 966                                 return (DBM_ERROR);
 967                         bp = (uint16 *)(bufp->page);
 968                         hashp->cndx = 1;
 969                 }
 970                 if (!bp[0]) {
 971                         hashp->cpage = NULL;
 972                         ++hashp->cbucket;
 973                 }
 974         }
 975         ndx = hashp->cndx;
 976         if (bp[ndx + 1] < REAL_KEY) {
 977                 if (__big_keydata(hashp, bufp, key, data, 1))
 978                         return (DBM_ERROR);
 979         } else {
 980                 key->data = (uint8 *)hashp->cpage->page + bp[ndx];
 981                 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
 982                 data->data = (uint8 *)hashp->cpage->page + bp[ndx + 1];
 983                 data->size = bp[ndx] - bp[ndx + 1];
 984                 ndx += 2;
 985                 if (ndx > bp[0]) {
 986                         hashp->cpage = NULL;
 987                         hashp->cbucket++;
 988                         hashp->cndx = 1;
 989                 } else
 990                         hashp->cndx = ndx;
 991         }
 992         return (SUCCESS);
 993 }
 994
 995 /********************************* UTILITIES ************************/
 996
 997 /*
 998  * Returns:
 999  *       0 ==> OK
1000  *      -1 ==> Error
1001  */
1002 extern int
1003 __expand_table(HTAB *hashp)
1004 {
1005         uint32 old_bucket, new_bucket;
1006         int new_segnum, spare_ndx;
1007         size_t dirsize;
1008
1009 #ifdef HASH_STATISTICS
1010         hash_expansions++;
1011 #endif
1012         new_bucket = ++hashp->MAX_BUCKET;
1013         old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
1014
1015         new_segnum = new_bucket >> hashp->SSHIFT;
1016
1017         /* Check if we need a new segment */
1018         if (new_segnum >= hashp->nsegs) {
1019                 /* Check if we need to expand directory */
1020                 if (new_segnum >= hashp->DSIZE) {
1021                         /* Reallocate directory */
1022                         dirsize = hashp->DSIZE * sizeof(SEGMENT *);
1023                         if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
1024                                 return (-1);
1025                         hashp->DSIZE = dirsize << 1;
1026                 }
1027                 if ((hashp->dir[new_segnum] =
1028                     (SEGMENT)calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
1029                         return (-1);
1030                 hashp->exsegs++;
1031                 hashp->nsegs++;
1032         }
1033         /*
1034          * If the split point is increasing (MAX_BUCKET's log base 2
1035          * * increases), we need to copy the current contents of the spare
1036          * split bucket to the next bucket.
1037          */
1038         spare_ndx = __log2((uint32)(hashp->MAX_BUCKET + 1));
1039         if (spare_ndx > hashp->OVFL_POINT) {
1040                 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
1041                 hashp->OVFL_POINT = spare_ndx;
1042         }
1043
1044         if (new_bucket > (uint32)hashp->HIGH_MASK) {
1045                 /* Starting a new doubling */
1046                 hashp->LOW_MASK = hashp->HIGH_MASK;
1047                 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
1048         }
1049         /* Relocate records to the new bucket */
1050         return (__split_page(hashp, old_bucket, new_bucket));
1051 }
1052
1053 /*
1054  * If realloc guarantees that the pointer is not destroyed if the realloc
1055  * fails, then this routine can go away.
1056  */
1057 static void *
1058 hash_realloc(
1059         SEGMENT **p_ptr,
1060         size_t oldsize, size_t newsize)
1061 {
1062         register void *p;
1063
1064         if ((p = malloc(newsize))) {
1065                 memmove(p, *p_ptr, oldsize);
1066                 memset((char *)p + oldsize, 0, newsize - oldsize);
1067                 free(*p_ptr);
1068                 *p_ptr = (SEGMENT *)p;
1069         }
1070         return (p);
1071 }
1072
1073 extern uint32
1074 __call_hash(HTAB *hashp, char *k, size_t len)
1075 {
1076         uint32 n, bucket;
1077
1078         n = hashp->hash(k, len);
1079         bucket = n & hashp->HIGH_MASK;
1080         if (bucket > (uint32)hashp->MAX_BUCKET)
1081                 bucket = bucket & hashp->LOW_MASK;
1082         return (bucket);
1083 }
1084
1085 /*
1086  * Allocate segment table.  On error, set errno.
1087  *
1088  * Returns 0 on success
1089  */
1090 static int
1091 alloc_segs(
1092         HTAB *hashp,
1093         int nsegs)
1094 {
1095         register int i;
1096         register SEGMENT store;
1097
1098         if ((hashp->dir =
1099             (SEGMENT *)calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
1100                 errno = ENOMEM;
1101                 return (-1);
1102         }
1103         /* Allocate segments */
1104         if ((store =
1105             (SEGMENT)calloc((size_t)nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
1106                 errno = ENOMEM;
1107                 return (-1);
1108         }
1109         for (i = 0; i < nsegs; i++, hashp->nsegs++)
1110                 hashp->dir[i] = &store[i << hashp->SSHIFT];
1111         return (0);
1112 }
1113
1114 #if BYTE_ORDER == LITTLE_ENDIAN
1115 /*
1116  * Hashp->hdr needs to be byteswapped.
1117  */
1118 static void
1119 swap_header_copy(
1120         HASHHDR *srcp, HASHHDR *destp)
1121 {
1122         int i;
1123
1124         P_32_COPY(srcp->magic, destp->magic);
1125         P_32_COPY(srcp->version, destp->version);
1126         P_32_COPY(srcp->lorder, destp->lorder);
1127         P_32_COPY(srcp->bsize, destp->bsize);
1128         P_32_COPY(srcp->bshift, destp->bshift);
1129         P_32_COPY(srcp->dsize, destp->dsize);
1130         P_32_COPY(srcp->ssize, destp->ssize);
1131         P_32_COPY(srcp->sshift, destp->sshift);
1132         P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
1133         P_32_COPY(srcp->last_freed, destp->last_freed);
1134         P_32_COPY(srcp->max_bucket, destp->max_bucket);
1135         P_32_COPY(srcp->high_mask, destp->high_mask);
1136         P_32_COPY(srcp->low_mask, destp->low_mask);
1137         P_32_COPY(srcp->ffactor, destp->ffactor);
1138         P_32_COPY(srcp->nkeys, destp->nkeys);
1139         P_32_COPY(srcp->hdrpages, destp->hdrpages);
1140         P_32_COPY(srcp->h_charkey, destp->h_charkey);
1141         for (i = 0; i < NCACHED; i++) {
1142                 P_32_COPY(srcp->spares[i], destp->spares[i]);
1143                 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
1144         }
1145 }
1146
1147 static void
1148 swap_header(HTAB *hashp)
1149 {
1150         HASHHDR *hdrp;
1151         int i;
1152
1153         hdrp = &hashp->hdr;
1154
1155         M_32_SWAP(hdrp->magic);
1156         M_32_SWAP(hdrp->version);
1157         M_32_SWAP(hdrp->lorder);
1158         M_32_SWAP(hdrp->bsize);
1159         M_32_SWAP(hdrp->bshift);
1160         M_32_SWAP(hdrp->dsize);
1161         M_32_SWAP(hdrp->ssize);
1162         M_32_SWAP(hdrp->sshift);
1163         M_32_SWAP(hdrp->ovfl_point);
1164         M_32_SWAP(hdrp->last_freed);
1165         M_32_SWAP(hdrp->max_bucket);
1166         M_32_SWAP(hdrp->high_mask);
1167         M_32_SWAP(hdrp->low_mask);
1168         M_32_SWAP(hdrp->ffactor);
1169         M_32_SWAP(hdrp->nkeys);
1170         M_32_SWAP(hdrp->hdrpages);
1171         M_32_SWAP(hdrp->h_charkey);
1172         for (i = 0; i < NCACHED; i++) {
1173                 M_32_SWAP(hdrp->spares[i]);
1174                 M_16_SWAP(hdrp->bitmaps[i]);
1175         }
1176 }
1177 #endif