lib/libc/db/hash/hash_page.c

   1 /*      $NetBSD: hash_page.c,v 1.23 2008/09/11 12:58:00 joerg Exp $     */
   2
   3 /*-
   4  * Copyright (c) 1990, 1993, 1994
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * This code is derived from software contributed to Berkeley by
   8  * Margo Seltzer.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  * 3. Neither the name of the University nor the names of its contributors
  19  *    may be used to endorse or promote products derived from this software
  20  *    without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34
  35 #if HAVE_NBTOOL_CONFIG_H
  36 #include "nbtool_config.h"
  37 #endif
  38
  39 #include <sys/cdefs.h>
  40 __RCSID("$NetBSD: hash_page.c,v 1.23 2008/09/11 12:58:00 joerg Exp $");
  41
  42 /*
  43  * PACKAGE:  hashing
  44  *
  45  * DESCRIPTION:
  46  *      Page manipulation for hashing package.
  47  *
  48  * ROUTINES:
  49  *
  50  * External
  51  *      __get_page
  52  *      __add_ovflpage
  53  * Internal
  54  *      overflow_page
  55  *      open_temp
  56  */
  57
  58 #include "namespace.h"
  59
  60 #include <sys/types.h>
  61
  62 #include <errno.h>
  63 #include <fcntl.h>
  64 #include <signal.h>
  65 #include <stdio.h>
  66 #include <stdlib.h>
  67 #include <string.h>
  68 #include <unistd.h>
  69 #include <paths.h>
  70 #include <assert.h>
  71
  72 #include <db.h>
  73 #include "hash.h"
  74 #include "page.h"
  75 #include "extern.h"
  76
  77 static uint32_t *fetch_bitmap(HTAB *, int);
  78 static uint32_t  first_free(uint32_t);
  79 static int       open_temp(HTAB *);
  80 static uint16_t  overflow_page(HTAB *);
  81 static void      putpair(char *, const DBT *, const DBT *);
  82 static void      squeeze_key(uint16_t *, const DBT *, const DBT *);
  83 static int       ugly_split(HTAB *, uint32_t, BUFHEAD *, BUFHEAD *, int, int);
  84
  85 #define PAGE_INIT(P) { \
  86         ((uint16_t *)(void *)(P))[0] = 0; \
  87         temp = 3 * sizeof(uint16_t); \
  88         _DIAGASSERT(hashp->BSIZE >= temp); \
  89         ((uint16_t *)(void *)(P))[1] = (uint16_t)(hashp->BSIZE - temp); \
  90         ((uint16_t *)(void *)(P))[2] = hashp->BSIZE; \
  91 }
  92
  93 /*
  94  * This is called AFTER we have verified that there is room on the page for
  95  * the pair (PAIRFITS has returned true) so we go right ahead and start moving
  96  * stuff on.
  97  */
  98 static void
  99 putpair(char *p, const DBT *key, const DBT *val)
 100 {
 101         uint16_t *bp, n, off;
 102         size_t temp;
 103
 104         bp = (uint16_t *)(void *)p;
 105
 106         /* Enter the key first. */
 107         n = bp[0];
 108
 109         temp = OFFSET(bp);
 110         _DIAGASSERT(temp >= key->size);
 111         off = (uint16_t)(temp - key->size);
 112         memmove(p + off, key->data, key->size);
 113         bp[++n] = off;
 114
 115         /* Now the data. */
 116         _DIAGASSERT(off >= val->size);
 117         off -= (uint16_t)val->size;
 118         memmove(p + off, val->data, val->size);
 119         bp[++n] = off;
 120
 121         /* Adjust page info. */
 122         bp[0] = n;
 123         temp = (n + 3) * sizeof(uint16_t);
 124         _DIAGASSERT(off >= temp);
 125         bp[n + 1] = (uint16_t)(off - temp);
 126         bp[n + 2] = off;
 127 }
 128
 129 /*
 130  * Returns:
 131  *       0 OK
 132  *      -1 error
 133  */
 134 int
 135 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
 136 {
 137         uint16_t *bp, newoff;
 138         int n;
 139         uint16_t pairlen;
 140         size_t temp;
 141
 142         bp = (uint16_t *)(void *)bufp->page;
 143         n = bp[0];
 144
 145         if (bp[ndx + 1] < REAL_KEY)
 146                 return (__big_delete(hashp, bufp));
 147         if (ndx != 1)
 148                 newoff = bp[ndx - 1];
 149         else
 150                 newoff = hashp->BSIZE;
 151         pairlen = newoff - bp[ndx + 1];
 152
 153         if (ndx != (n - 1)) {
 154                 /* Hard Case -- need to shuffle keys */
 155                 int i;
 156                 char *src = bufp->page + (int)OFFSET(bp);
 157                 char *dst = src + (int)pairlen;
 158                 memmove(dst, src, (size_t)(bp[ndx + 1] - OFFSET(bp)));
 159
 160                 /* Now adjust the pointers */
 161                 for (i = ndx + 2; i <= n; i += 2) {
 162                         if (bp[i + 1] == OVFLPAGE) {
 163                                 bp[i - 2] = bp[i];
 164                                 bp[i - 1] = bp[i + 1];
 165                         } else {
 166                                 bp[i - 2] = bp[i] + pairlen;
 167                                 bp[i - 1] = bp[i + 1] + pairlen;
 168                         }
 169                 }
 170         }
 171         /* Finally adjust the page data */
 172         bp[n] = OFFSET(bp) + pairlen;
 173         temp = bp[n + 1] + pairlen + 2 * sizeof(uint16_t);
 174         _DIAGASSERT(temp <= 0xffff);
 175         bp[n - 1] = (uint16_t)temp;
 176         bp[0] = n - 2;
 177         hashp->NKEYS--;
 178
 179         bufp->flags |= BUF_MOD;
 180         return (0);
 181 }
 182 /*
 183  * Returns:
 184  *       0 ==> OK
 185  *      -1 ==> Error
 186  */
 187 int
 188 __split_page(HTAB *hashp, uint32_t obucket, uint32_t nbucket)
 189 {
 190         BUFHEAD *new_bufp, *old_bufp;
 191         uint16_t *ino;
 192         char *np;
 193         DBT key, val;
 194         int n, ndx, retval;
 195         uint16_t copyto, diff, off, moved;
 196         char *op;
 197         size_t temp;
 198
 199         copyto = (uint16_t)hashp->BSIZE;
 200         off = (uint16_t)hashp->BSIZE;
 201         old_bufp = __get_buf(hashp, obucket, NULL, 0);
 202         if (old_bufp == NULL)
 203                 return (-1);
 204         new_bufp = __get_buf(hashp, nbucket, NULL, 0);
 205         if (new_bufp == NULL)
 206                 return (-1);
 207
 208         old_bufp->flags |= (BUF_MOD | BUF_PIN);
 209         new_bufp->flags |= (BUF_MOD | BUF_PIN);
 210
 211         ino = (uint16_t *)(void *)(op = old_bufp->page);
 212         np = new_bufp->page;
 213
 214         moved = 0;
 215
 216         for (n = 1, ndx = 1; n < ino[0]; n += 2) {
 217                 if (ino[n + 1] < REAL_KEY) {
 218                         retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
 219                             (int)copyto, (int)moved);
 220                         old_bufp->flags &= ~BUF_PIN;
 221                         new_bufp->flags &= ~BUF_PIN;
 222                         return (retval);
 223
 224                 }
 225                 key.data = (uint8_t *)op + ino[n];
 226                 key.size = off - ino[n];
 227
 228                 if (__call_hash(hashp, key.data, (int)key.size) == obucket) {
 229                         /* Don't switch page */
 230                         diff = copyto - off;
 231                         if (diff) {
 232                                 copyto = ino[n + 1] + diff;
 233                                 memmove(op + copyto, op + ino[n + 1],
 234                                     (size_t)(off - ino[n + 1]));
 235                                 ino[ndx] = copyto + ino[n] - ino[n + 1];
 236                                 ino[ndx + 1] = copyto;
 237                         } else
 238                                 copyto = ino[n + 1];
 239                         ndx += 2;
 240                 } else {
 241                         /* Switch page */
 242                         val.data = (uint8_t *)op + ino[n + 1];
 243                         val.size = ino[n] - ino[n + 1];
 244                         putpair(np, &key, &val);
 245                         moved += 2;
 246                 }
 247
 248                 off = ino[n + 1];
 249         }
 250
 251         /* Now clean up the page */
 252         ino[0] -= moved;
 253         temp = sizeof(uint16_t) * (ino[0] + 3);
 254         _DIAGASSERT(copyto >= temp);
 255         FREESPACE(ino) = (uint16_t)(copyto - temp);
 256         OFFSET(ino) = copyto;
 257
 258 #ifdef DEBUG3
 259         (void)fprintf(stderr, "split %d/%d\n",
 260             ((uint16_t *)np)[0] / 2,
 261             ((uint16_t *)op)[0] / 2);
 262 #endif
 263         /* unpin both pages */
 264         old_bufp->flags &= ~BUF_PIN;
 265         new_bufp->flags &= ~BUF_PIN;
 266         return (0);
 267 }
 268
 269 /*
 270  * Called when we encounter an overflow or big key/data page during split
 271  * handling.  This is special cased since we have to begin checking whether
 272  * the key/data pairs fit on their respective pages and because we may need
 273  * overflow pages for both the old and new pages.
 274  *
 275  * The first page might be a page with regular key/data pairs in which case
 276  * we have a regular overflow condition and just need to go on to the next
 277  * page or it might be a big key/data pair in which case we need to fix the
 278  * big key/data pair.
 279  *
 280  * Returns:
 281  *       0 ==> success
 282  *      -1 ==> failure
 283  */
 284 static int
 285 ugly_split(
 286         HTAB *hashp,
 287         uint32_t obucket,       /* Same as __split_page. */
 288         BUFHEAD *old_bufp,
 289         BUFHEAD *new_bufp,
 290         int copyto,     /* First byte on page which contains key/data values. */
 291         int moved       /* Number of pairs moved to new page. */
 292 )
 293 {
 294         BUFHEAD *bufp;  /* Buffer header for ino */
 295         uint16_t *ino;  /* Page keys come off of */
 296         uint16_t *np;   /* New page */
 297         uint16_t *op;   /* Page keys go on to if they aren't moving */
 298         size_t temp;
 299
 300         BUFHEAD *last_bfp;      /* Last buf header OVFL needing to be freed */
 301         DBT key, val;
 302         SPLIT_RETURN ret;
 303         uint16_t n, off, ov_addr, scopyto;
 304         char *cino;             /* Character value of ino */
 305
 306         bufp = old_bufp;
 307         ino = (uint16_t *)(void *)old_bufp->page;
 308         np = (uint16_t *)(void *)new_bufp->page;
 309         op = (uint16_t *)(void *)old_bufp->page;
 310         last_bfp = NULL;
 311         scopyto = (uint16_t)copyto;     /* ANSI */
 312
 313         n = ino[0] - 1;
 314         while (n < ino[0]) {
 315                 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
 316                         if (__big_split(hashp, old_bufp,
 317                             new_bufp, bufp, (int)bufp->addr, obucket, &ret))
 318                                 return (-1);
 319                         old_bufp = ret.oldp;
 320                         if (!old_bufp)
 321                                 return (-1);
 322                         op = (uint16_t *)(void *)old_bufp->page;
 323                         new_bufp = ret.newp;
 324                         if (!new_bufp)
 325                                 return (-1);
 326                         np = (uint16_t *)(void *)new_bufp->page;
 327                         bufp = ret.nextp;
 328                         if (!bufp)
 329                                 return (0);
 330                         cino = (char *)bufp->page;
 331                         ino = (uint16_t *)(void *)cino;
 332                         last_bfp = ret.nextp;
 333                 } else if (ino[n + 1] == OVFLPAGE) {
 334                         ov_addr = ino[n];
 335                         /*
 336                          * Fix up the old page -- the extra 2 are the fields
 337                          * which contained the overflow information.
 338                          */
 339                         ino[0] -= (moved + 2);
 340                         temp = sizeof(uint16_t) * (ino[0] + 3);
 341                         _DIAGASSERT(scopyto >= temp);
 342                         FREESPACE(ino) = (uint16_t)(scopyto - temp);
 343                         OFFSET(ino) = scopyto;
 344
 345                         bufp = __get_buf(hashp, (uint32_t)ov_addr, bufp, 0);
 346                         if (!bufp)
 347                                 return (-1);
 348
 349                         ino = (uint16_t *)(void *)bufp->page;
 350                         n = 1;
 351                         scopyto = hashp->BSIZE;
 352                         moved = 0;
 353
 354                         if (last_bfp)
 355                                 __free_ovflpage(hashp, last_bfp);
 356                         last_bfp = bufp;
 357                 }
 358                 /* Move regular sized pairs of there are any */
 359                 off = hashp->BSIZE;
 360                 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
 361                         cino = (char *)(void *)ino;
 362                         key.data = (uint8_t *)cino + ino[n];
 363                         key.size = off - ino[n];
 364                         val.data = (uint8_t *)cino + ino[n + 1];
 365                         val.size = ino[n] - ino[n + 1];
 366                         off = ino[n + 1];
 367
 368                         if (__call_hash(hashp, key.data, (int)key.size) == obucket) {
 369                                 /* Keep on old page */
 370                                 if (PAIRFITS(op, (&key), (&val)))
 371                                         putpair((char *)(void *)op, &key, &val);
 372                                 else {
 373                                         old_bufp =
 374                                             __add_ovflpage(hashp, old_bufp);
 375                                         if (!old_bufp)
 376                                                 return (-1);
 377                                         op = (uint16_t *)(void *)old_bufp->page;
 378                                         putpair((char *)(void *)op, &key, &val);
 379                                 }
 380                                 old_bufp->flags |= BUF_MOD;
 381                         } else {
 382                                 /* Move to new page */
 383                                 if (PAIRFITS(np, (&key), (&val)))
 384                                         putpair((char *)(void *)np, &key, &val);
 385                                 else {
 386                                         new_bufp =
 387                                             __add_ovflpage(hashp, new_bufp);
 388                                         if (!new_bufp)
 389                                                 return (-1);
 390                                         np = (uint16_t *)(void *)new_bufp->page;
 391                                         putpair((char *)(void *)np, &key, &val);
 392                                 }
 393                                 new_bufp->flags |= BUF_MOD;
 394                         }
 395                 }
 396         }
 397         if (last_bfp)
 398                 __free_ovflpage(hashp, last_bfp);
 399         return (0);
 400 }
 401
 402 /*
 403  * Add the given pair to the page
 404  *
 405  * Returns:
 406  *      0 ==> OK
 407  *      1 ==> failure
 408  */
 409 int
 410 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
 411 {
 412         uint16_t *bp, *sop;
 413         int do_expand;
 414
 415         bp = (uint16_t *)(void *)bufp->page;
 416         do_expand = 0;
 417         while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
 418                 /* Exception case */
 419                 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
 420                         /* This is the last page of a big key/data pair
 421                            and we need to add another page */
 422                         break;
 423                 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
 424                         bufp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp,
 425                             0);
 426                         if (!bufp)
 427                                 return (-1);
 428                         bp = (uint16_t *)(void *)bufp->page;
 429                 } else if (bp[bp[0]] != OVFLPAGE) {
 430                         /* Short key/data pairs, no more pages */
 431                         break;
 432                 } else {
 433                         /* Try to squeeze key on this page */
 434                         if (bp[2] >= REAL_KEY &&
 435                             FREESPACE(bp) >= PAIRSIZE(key, val)) {
 436                                 squeeze_key(bp, key, val);
 437                                 goto stats;
 438                         } else {
 439                                 bufp = __get_buf(hashp,
 440                                     (uint32_t)bp[bp[0] - 1], bufp, 0);
 441                                 if (!bufp)
 442                                         return (-1);
 443                                 bp = (uint16_t *)(void *)bufp->page;
 444                         }
 445                 }
 446
 447         if (PAIRFITS(bp, key, val))
 448                 putpair(bufp->page, key, val);
 449         else {
 450                 do_expand = 1;
 451                 bufp = __add_ovflpage(hashp, bufp);
 452                 if (!bufp)
 453                         return (-1);
 454                 sop = (uint16_t *)(void *)bufp->page;
 455
 456                 if (PAIRFITS(sop, key, val))
 457                         putpair((char *)(void *)sop, key, val);
 458                 else
 459                         if (__big_insert(hashp, bufp, key, val))
 460                                 return (-1);
 461         }
 462 stats:
 463         bufp->flags |= BUF_MOD;
 464         /*
 465          * If the average number of keys per bucket exceeds the fill factor,
 466          * expand the table.
 467          */
 468         hashp->NKEYS++;
 469         if (do_expand ||
 470             (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
 471                 return (__expand_table(hashp));
 472         return (0);
 473 }
 474
 475 /*
 476  *
 477  * Returns:
 478  *      pointer on success
 479  *      NULL on error
 480  */
 481 BUFHEAD *
 482 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
 483 {
 484         uint16_t *sp;
 485         uint16_t ndx, ovfl_num;
 486         size_t temp;
 487 #ifdef DEBUG1
 488         int tmp1, tmp2;
 489 #endif
 490         sp = (uint16_t *)(void *)bufp->page;
 491
 492         /* Check if we are dynamically determining the fill factor */
 493         if (hashp->FFACTOR == DEF_FFACTOR) {
 494                 hashp->FFACTOR = (uint32_t)sp[0] >> 1;
 495                 if (hashp->FFACTOR < MIN_FFACTOR)
 496                         hashp->FFACTOR = MIN_FFACTOR;
 497         }
 498         bufp->flags |= BUF_MOD;
 499         ovfl_num = overflow_page(hashp);
 500 #ifdef DEBUG1
 501         tmp1 = bufp->addr;
 502         tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
 503 #endif
 504         if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, (uint32_t)ovfl_num,
 505             bufp, 1)))
 506                 return (NULL);
 507         bufp->ovfl->flags |= BUF_MOD;
 508 #ifdef DEBUG1
 509         (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
 510             tmp1, tmp2, bufp->ovfl->addr);
 511 #endif
 512         ndx = sp[0];
 513         /*
 514          * Since a pair is allocated on a page only if there's room to add
 515          * an overflow page, we know that the OVFL information will fit on
 516          * the page.
 517          */
 518         sp[ndx + 4] = OFFSET(sp);
 519         temp = FREESPACE(sp);
 520         _DIAGASSERT(temp >= OVFLSIZE);
 521         sp[ndx + 3] = (uint16_t)(temp - OVFLSIZE);
 522         sp[ndx + 1] = ovfl_num;
 523         sp[ndx + 2] = OVFLPAGE;
 524         sp[0] = ndx + 2;
 525 #ifdef HASH_STATISTICS
 526         hash_overflows++;
 527 #endif
 528         return (bufp->ovfl);
 529 }
 530
 531 /*
 532  * Returns:
 533  *       0 indicates SUCCESS
 534  *      -1 indicates FAILURE
 535  */
 536 int
 537 __get_page(HTAB *hashp, char *p, uint32_t bucket, int is_bucket, int is_disk,
 538     int is_bitmap)
 539 {
 540         int fd, page, size;
 541         ssize_t rsize;
 542         uint16_t *bp;
 543         size_t temp;
 544
 545         fd = hashp->fp;
 546         size = hashp->BSIZE;
 547
 548         if ((fd == -1) || !is_disk) {
 549                 PAGE_INIT(p);
 550                 return (0);
 551         }
 552         if (is_bucket)
 553                 page = BUCKET_TO_PAGE(bucket);
 554         else
 555                 page = OADDR_TO_PAGE(bucket);
 556         if ((rsize = pread(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1)
 557                 return (-1);
 558         bp = (uint16_t *)(void *)p;
 559         if (!rsize)
 560                 bp[0] = 0;      /* We hit the EOF, so initialize a new page */
 561         else
 562                 if (rsize != size) {
 563                         errno = EFTYPE;
 564                         return (-1);
 565                 }
 566         if (!is_bitmap && !bp[0]) {
 567                 PAGE_INIT(p);
 568         } else
 569                 if (hashp->LORDER != BYTE_ORDER) {
 570                         int i, max;
 571
 572                         if (is_bitmap) {
 573                                 max = (uint32_t)hashp->BSIZE >> 2; /* divide by 4 */
 574                                 for (i = 0; i < max; i++)
 575                                         M_32_SWAP(((int *)(void *)p)[i]);
 576                         } else {
 577                                 M_16_SWAP(bp[0]);
 578                                 max = bp[0] + 2;
 579                                 for (i = 1; i <= max; i++)
 580                                         M_16_SWAP(bp[i]);
 581                         }
 582                 }
 583         return (0);
 584 }
 585
 586 /*
 587  * Write page p to disk
 588  *
 589  * Returns:
 590  *       0 ==> OK
 591  *      -1 ==>failure
 592  */
 593 int
 594 __put_page(HTAB *hashp, char *p, uint32_t bucket, int is_bucket, int is_bitmap)
 595 {
 596         int fd, page, size;
 597         ssize_t wsize;
 598
 599         size = hashp->BSIZE;
 600         if ((hashp->fp == -1) && open_temp(hashp))
 601                 return (-1);
 602         fd = hashp->fp;
 603
 604         if (hashp->LORDER != BYTE_ORDER) {
 605                 int i;
 606                 int max;
 607
 608                 if (is_bitmap) {
 609                         max = (uint32_t)hashp->BSIZE >> 2;      /* divide by 4 */
 610                         for (i = 0; i < max; i++)
 611                                 M_32_SWAP(((int *)(void *)p)[i]);
 612                 } else {
 613                         max = ((uint16_t *)(void *)p)[0] + 2;
 614                         for (i = 0; i <= max; i++)
 615                                 M_16_SWAP(((uint16_t *)(void *)p)[i]);
 616                 }
 617         }
 618         if (is_bucket)
 619                 page = BUCKET_TO_PAGE(bucket);
 620         else
 621                 page = OADDR_TO_PAGE(bucket);
 622         if ((wsize = pwrite(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1)
 623                 /* Errno is set */
 624                 return (-1);
 625         if (wsize != size) {
 626                 errno = EFTYPE;
 627                 return (-1);
 628         }
 629         return (0);
 630 }
 631
 632 #define BYTE_MASK       ((1 << INT_BYTE_SHIFT) -1)
 633 /*
 634  * Initialize a new bitmap page.  Bitmap pages are left in memory
 635  * once they are read in.
 636  */
 637 int
 638 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
 639 {
 640         uint32_t *ip;
 641         int clearbytes, clearints;
 642
 643         if ((ip = malloc((size_t)hashp->BSIZE)) == NULL)
 644                 return (1);
 645         hashp->nmaps++;
 646         clearints = ((uint32_t)(nbits - 1) >> INT_BYTE_SHIFT) + 1;
 647         clearbytes = clearints << INT_TO_BYTE;
 648         (void)memset(ip, 0, (size_t)clearbytes);
 649         (void)memset(((char *)(void *)ip) + clearbytes, 0xFF,
 650             (size_t)(hashp->BSIZE - clearbytes));
 651         ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
 652         SETBIT(ip, 0);
 653         hashp->BITMAPS[ndx] = (uint16_t)pnum;
 654         hashp->mapp[ndx] = ip;
 655         return (0);
 656 }
 657
 658 static uint32_t
 659 first_free(uint32_t map)
 660 {
 661         uint32_t i, mask;
 662
 663         mask = 0x1;
 664         for (i = 0; i < BITS_PER_MAP; i++) {
 665                 if (!(mask & map))
 666                         return (i);
 667                 mask = mask << 1;
 668         }
 669         return (i);
 670 }
 671
 672 static uint16_t
 673 overflow_page(HTAB *hashp)
 674 {
 675         uint32_t *freep = NULL;
 676         int max_free, offset, splitnum;
 677         uint16_t addr;
 678         int bit, first_page, free_bit, free_page, i, in_use_bits, j;
 679 #ifdef DEBUG2
 680         int tmp1, tmp2;
 681 #endif
 682         splitnum = hashp->OVFL_POINT;
 683         max_free = hashp->SPARES[splitnum];
 684
 685         free_page = (uint32_t)(max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
 686         free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
 687
 688         /* Look through all the free maps to find the first free block */
 689         first_page = (uint32_t)hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
 690         for ( i = first_page; i <= free_page; i++ ) {
 691                 if (!(freep = (uint32_t *)hashp->mapp[i]) &&
 692                     !(freep = fetch_bitmap(hashp, i)))
 693                         return (0);
 694                 if (i == free_page)
 695                         in_use_bits = free_bit;
 696                 else
 697                         in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
 698
 699                 if (i == first_page) {
 700                         bit = hashp->LAST_FREED &
 701                             ((hashp->BSIZE << BYTE_SHIFT) - 1);
 702                         j = bit / BITS_PER_MAP;
 703                         bit = bit & ~(BITS_PER_MAP - 1);
 704                 } else {
 705                         bit = 0;
 706                         j = 0;
 707                 }
 708                 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
 709                         if (freep[j] != ALL_SET)
 710                                 goto found;
 711         }
 712
 713         /* No Free Page Found */
 714         hashp->LAST_FREED = hashp->SPARES[splitnum];
 715         hashp->SPARES[splitnum]++;
 716         offset = hashp->SPARES[splitnum] -
 717             (splitnum ? hashp->SPARES[splitnum - 1] : 0);
 718
 719 #define OVMSG   "HASH: Out of overflow pages.  Increase page size\n"
 720         if (offset > SPLITMASK) {
 721                 if (++splitnum >= NCACHED) {
 722                         (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
 723                         errno = EFBIG;
 724                         return (0);
 725                 }
 726                 hashp->OVFL_POINT = splitnum;
 727                 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
 728                 hashp->SPARES[splitnum-1]--;
 729                 offset = 1;
 730         }
 731
 732         /* Check if we need to allocate a new bitmap page */
 733         if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
 734                 free_page++;
 735                 if (free_page >= NCACHED) {
 736                         (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
 737                         errno = EFBIG;
 738                         return (0);
 739                 }
 740                 /*
 741                  * This is tricky.  The 1 indicates that you want the new page
 742                  * allocated with 1 clear bit.  Actually, you are going to
 743                  * allocate 2 pages from this map.  The first is going to be
 744                  * the map page, the second is the overflow page we were
 745                  * looking for.  The init_bitmap routine automatically, sets
 746                  * the first bit of itself to indicate that the bitmap itself
 747                  * is in use.  We would explicitly set the second bit, but
 748                  * don't have to if we tell init_bitmap not to leave it clear
 749                  * in the first place.
 750                  */
 751                 if (__ibitmap(hashp,
 752                     (int)OADDR_OF(splitnum, offset), 1, free_page))
 753                         return (0);
 754                 hashp->SPARES[splitnum]++;
 755 #ifdef DEBUG2
 756                 free_bit = 2;
 757 #endif
 758                 offset++;
 759                 if (offset > SPLITMASK) {
 760                         if (++splitnum >= NCACHED) {
 761                                 (void)write(STDERR_FILENO, OVMSG,
 762                                     sizeof(OVMSG) - 1);
 763                                 errno = EFBIG;
 764                                 return (0);
 765                         }
 766                         hashp->OVFL_POINT = splitnum;
 767                         hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
 768                         hashp->SPARES[splitnum-1]--;
 769                         offset = 0;
 770                 }
 771         } else {
 772                 /*
 773                  * Free_bit addresses the last used bit.  Bump it to address
 774                  * the first available bit.
 775                  */
 776                 free_bit++;
 777                 SETBIT(freep, free_bit);
 778         }
 779
 780         /* Calculate address of the new overflow page */
 781         addr = OADDR_OF(splitnum, offset);
 782 #ifdef DEBUG2
 783         (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
 784             addr, free_bit, free_page);
 785 #endif
 786         return (addr);
 787
 788 found:
 789         bit = bit + first_free(freep[j]);
 790         SETBIT(freep, bit);
 791 #ifdef DEBUG2
 792         tmp1 = bit;
 793         tmp2 = i;
 794 #endif
 795         /*
 796          * Bits are addressed starting with 0, but overflow pages are addressed
 797          * beginning at 1. Bit is a bit addressnumber, so we need to increment
 798          * it to convert it to a page number.
 799          */
 800         bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
 801         if (bit >= hashp->LAST_FREED)
 802                 hashp->LAST_FREED = bit - 1;
 803
 804         /* Calculate the split number for this page */
 805         for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
 806         offset = (i ? bit - hashp->SPARES[i - 1] : bit);
 807         if (offset >= SPLITMASK) {
 808                 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
 809                 errno = EFBIG;
 810                 return (0);     /* Out of overflow pages */
 811         }
 812         addr = OADDR_OF(i, offset);
 813 #ifdef DEBUG2
 814         (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
 815             addr, tmp1, tmp2);
 816 #endif
 817
 818         /* Allocate and return the overflow page */
 819         return (addr);
 820 }
 821
 822 /*
 823  * Mark this overflow page as free.
 824  */
 825 void
 826 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
 827 {
 828         uint16_t addr;
 829         uint32_t *freep;
 830         int bit_address, free_page, free_bit;
 831         uint16_t ndx;
 832
 833         addr = obufp->addr;
 834 #ifdef DEBUG1
 835         (void)fprintf(stderr, "Freeing %d\n", addr);
 836 #endif
 837         ndx = (((uint32_t)addr) >> SPLITSHIFT);
 838         bit_address =
 839             (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
 840          if (bit_address < hashp->LAST_FREED)
 841                 hashp->LAST_FREED = bit_address;
 842         free_page = ((uint32_t)bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
 843         free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
 844
 845         if (!(freep = hashp->mapp[free_page]))
 846                 freep = fetch_bitmap(hashp, free_page);
 847         /*
 848          * This had better never happen.  It means we tried to read a bitmap
 849          * that has already had overflow pages allocated off it, and we
 850          * failed to read it from the file.
 851          */
 852         _DIAGASSERT(freep != NULL);
 853         CLRBIT(freep, free_bit);
 854 #ifdef DEBUG2
 855         (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
 856             obufp->addr, free_bit, free_page);
 857 #endif
 858         __reclaim_buf(hashp, obufp);
 859 }
 860
 861 /*
 862  * Returns:
 863  *       0 success
 864  *      -1 failure
 865  */
 866 static int
 867 open_temp(HTAB *hashp)
 868 {
 869         sigset_t set, oset;
 870         char *envtmp;
 871         char namestr[PATH_MAX];
 872
 873         if (issetugid())
 874                 envtmp = NULL;
 875         else
 876                 envtmp = getenv("TMPDIR");
 877
 878         if (-1 == snprintf(namestr, sizeof(namestr), "%s/_hashXXXXXX",
 879             envtmp ? envtmp : _PATH_TMP))
 880                 return -1;
 881
 882         /* Block signals; make sure file goes away at process exit. */
 883         (void)sigfillset(&set);
 884         (void)sigprocmask(SIG_BLOCK, &set, &oset);
 885         if ((hashp->fp = mkstemp(namestr)) != -1) {
 886                 (void)unlink(namestr);
 887                 (void)fcntl(hashp->fp, F_SETFD, FD_CLOEXEC);
 888         }
 889         (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
 890         return (hashp->fp != -1 ? 0 : -1);
 891 }
 892
 893 /*
 894  * We have to know that the key will fit, but the last entry on the page is
 895  * an overflow pair, so we need to shift things.
 896  */
 897 static void
 898 squeeze_key(uint16_t *sp, const DBT *key, const DBT *val)
 899 {
 900         char *p;
 901         uint16_t free_space, n, off, pageno;
 902         size_t temp;
 903
 904         p = (char *)(void *)sp;
 905         n = sp[0];
 906         free_space = FREESPACE(sp);
 907         off = OFFSET(sp);
 908
 909         pageno = sp[n - 1];
 910         _DIAGASSERT(off >= key->size);
 911         off -= (uint16_t)key->size;
 912         sp[n - 1] = off;
 913         memmove(p + off, key->data, key->size);
 914         _DIAGASSERT(off >= val->size);
 915         off -= (uint16_t)val->size;
 916         sp[n] = off;
 917         memmove(p + off, val->data, val->size);
 918         sp[0] = n + 2;
 919         sp[n + 1] = pageno;
 920         sp[n + 2] = OVFLPAGE;
 921         temp = PAIRSIZE(key, val);
 922         _DIAGASSERT(free_space >= temp);
 923         FREESPACE(sp) = (uint16_t)(free_space - temp);
 924         OFFSET(sp) = off;
 925 }
 926
 927 static uint32_t *
 928 fetch_bitmap(HTAB *hashp, int ndx)
 929 {
 930         if (ndx >= hashp->nmaps)
 931                 return (NULL);
 932         if ((hashp->mapp[ndx] = malloc((size_t)hashp->BSIZE)) == NULL)
 933                 return (NULL);
 934         if (__get_page(hashp,
 935             (char *)(void *)hashp->mapp[ndx], (uint32_t)hashp->BITMAPS[ndx], 0, 1, 1)) {
 936                 free(hashp->mapp[ndx]);
 937                 return (NULL);
 938         }
 939         return (hashp->mapp[ndx]);
 940 }
 941
 942 #ifdef DEBUG4
 943 void print_chain(HTAB *, uint32_t);
 944 void
 945 print_chain(HTAB *hashp, uint32_t addr)
 946 {
 947         BUFHEAD *bufp;
 948         uint16_t *bp, oaddr;
 949
 950         (void)fprintf(stderr, "%d ", addr);
 951         bufp = __get_buf(hashp, addr, NULL, 0);
 952         bp = (uint16_t *)bufp->page;
 953         while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
 954                 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
 955                 oaddr = bp[bp[0] - 1];
 956                 (void)fprintf(stderr, "%d ", (int)oaddr);
 957                 bufp = __get_buf(hashp, (uint32_t)oaddr, bufp, 0);
 958                 bp = (uint16_t *)bufp->page;
 959         }
 960         (void)fprintf(stderr, "\n");
 961 }
 962 #endif