db/hash/FreeBSD/hash_page.c

   1 /*-
   2  * Copyright (c) 1990, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Margo Seltzer.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36
  37 #if defined(LIBC_SCCS) && !defined(lint)
  38 static char sccsid[] = "@(#)hash_page.c 8.7 (Berkeley) 8/16/94";
  39 #endif /* LIBC_SCCS and not lint */
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD: src/lib/libc/db/hash/hash_page.c,v 1.8 2002/03/21 22:46:26 obrien Exp $");
  42
  43 /*
  44  * PACKAGE:  hashing
  45  *
  46  * DESCRIPTION:
  47  *      Page manipulation for hashing package.
  48  *
  49  * ROUTINES:
  50  *
  51  * External
  52  *      __get_page
  53  *      __add_ovflpage
  54  * Internal
  55  *      overflow_page
  56  *      open_temp
  57  */
  58
  59 #include "namespace.h"
  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 #ifdef DEBUG
  70 #include <assert.h>
  71 #endif
  72 #include "un-namespace.h"
  73
  74 #include <db.h>
  75 #include "hash.h"
  76 #include "page.h"
  77 #include "extern.h"
  78
  79 static u_int32_t        *fetch_bitmap(HTAB *, int);
  80 static u_int32_t         first_free(u_int32_t);
  81 static int       open_temp(HTAB *);
  82 static u_int16_t         overflow_page(HTAB *);
  83 static void      putpair(char *, const DBT *, const DBT *);
  84 static void      squeeze_key(u_int16_t *, const DBT *, const DBT *);
  85 static int       ugly_split
  86 (HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
  87
  88 #define PAGE_INIT(P) { \
  89         ((u_int16_t *)(P))[0] = 0; \
  90         ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
  91         ((u_int16_t *)(P))[2] = hashp->BSIZE; \
  92 }
  93
  94 /*
  95  * This is called AFTER we have verified that there is room on the page for
  96  * the pair (PAIRFITS has returned true) so we go right ahead and start moving
  97  * stuff on.
  98  */
  99 static void
 100 putpair(p, key, val)
 101         char *p;
 102         const DBT *key, *val;
 103 {
 104         u_int16_t *bp, n, off;
 105
 106         bp = (u_int16_t *)p;
 107
 108         /* Enter the key first. */
 109         n = bp[0];
 110
 111         off = OFFSET(bp) - key->size;
 112         memmove(p + off, key->data, key->size);
 113         bp[++n] = off;
 114
 115         /* Now the data. */
 116         off -= val->size;
 117         memmove(p + off, val->data, val->size);
 118         bp[++n] = off;
 119
 120         /* Adjust page info. */
 121         bp[0] = n;
 122         bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
 123         bp[n + 2] = off;
 124 }
 125
 126 /*
 127  * Returns:
 128  *       0 OK
 129  *      -1 error
 130  */
 131 extern int
 132 __delpair(hashp, bufp, ndx)
 133         HTAB *hashp;
 134         BUFHEAD *bufp;
 135         int ndx;
 136 {
 137         u_int16_t *bp, newoff;
 138         int n;
 139         u_int16_t pairlen;
 140
 141         bp = (u_int16_t *)bufp->page;
 142         n = bp[0];
 143
 144         if (bp[ndx + 1] < REAL_KEY)
 145                 return (__big_delete(hashp, bufp));
 146         if (ndx != 1)
 147                 newoff = bp[ndx - 1];
 148         else
 149                 newoff = hashp->BSIZE;
 150         pairlen = newoff - bp[ndx + 1];
 151
 152         if (ndx != (n - 1)) {
 153                 /* Hard Case -- need to shuffle keys */
 154                 int i;
 155                 char *src = bufp->page + (int)OFFSET(bp);
 156                 char *dst = src + (int)pairlen;
 157                 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
 158
 159                 /* Now adjust the pointers */
 160                 for (i = ndx + 2; i <= n; i += 2) {
 161                         if (bp[i + 1] == OVFLPAGE) {
 162                                 bp[i - 2] = bp[i];
 163                                 bp[i - 1] = bp[i + 1];
 164                         } else {
 165                                 bp[i - 2] = bp[i] + pairlen;
 166                                 bp[i - 1] = bp[i + 1] + pairlen;
 167                         }
 168                 }
 169         }
 170         /* Finally adjust the page data */
 171         bp[n] = OFFSET(bp) + pairlen;
 172         bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
 173         bp[0] = n - 2;
 174         hashp->NKEYS--;
 175
 176         bufp->flags |= BUF_MOD;
 177         return (0);
 178 }
 179 /*
 180  * Returns:
 181  *       0 ==> OK
 182  *      -1 ==> Error
 183  */
 184 extern int
 185 __split_page(hashp, obucket, nbucket)
 186         HTAB *hashp;
 187         u_int32_t obucket, nbucket;
 188 {
 189         BUFHEAD *new_bufp, *old_bufp;
 190         u_int16_t *ino;
 191         char *np;
 192         DBT key, val;
 193         int n, ndx, retval;
 194         u_int16_t copyto, diff, off, moved;
 195         char *op;
 196
 197         copyto = (u_int16_t)hashp->BSIZE;
 198         off = (u_int16_t)hashp->BSIZE;
 199         old_bufp = __get_buf(hashp, obucket, NULL, 0);
 200         if (old_bufp == NULL)
 201                 return (-1);
 202         new_bufp = __get_buf(hashp, nbucket, NULL, 0);
 203         if (new_bufp == NULL)
 204                 return (-1);
 205
 206         old_bufp->flags |= (BUF_MOD | BUF_PIN);
 207         new_bufp->flags |= (BUF_MOD | BUF_PIN);
 208
 209         ino = (u_int16_t *)(op = old_bufp->page);
 210         np = new_bufp->page;
 211
 212         moved = 0;
 213
 214         for (n = 1, ndx = 1; n < ino[0]; n += 2) {
 215                 if (ino[n + 1] < REAL_KEY) {
 216                         retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
 217                             (int)copyto, (int)moved);
 218                         old_bufp->flags &= ~BUF_PIN;
 219                         new_bufp->flags &= ~BUF_PIN;
 220                         return (retval);
 221
 222                 }
 223                 key.data = (u_char *)op + ino[n];
 224                 key.size = off - ino[n];
 225
 226                 if (__call_hash(hashp, key.data, key.size) == obucket) {
 227                         /* Don't switch page */
 228                         diff = copyto - off;
 229                         if (diff) {
 230                                 copyto = ino[n + 1] + diff;
 231                                 memmove(op + copyto, op + ino[n + 1],
 232                                     off - ino[n + 1]);
 233                                 ino[ndx] = copyto + ino[n] - ino[n + 1];
 234                                 ino[ndx + 1] = copyto;
 235                         } else
 236                                 copyto = ino[n + 1];
 237                         ndx += 2;
 238                 } else {
 239                         /* Switch page */
 240                         val.data = (u_char *)op + ino[n + 1];
 241                         val.size = ino[n] - ino[n + 1];
 242                         putpair(np, &key, &val);
 243                         moved += 2;
 244                 }
 245
 246                 off = ino[n + 1];
 247         }
 248
 249         /* Now clean up the page */
 250         ino[0] -= moved;
 251         FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
 252         OFFSET(ino) = copyto;
 253
 254 #ifdef DEBUG3
 255         (void)fprintf(stderr, "split %d/%d\n",
 256             ((u_int16_t *)np)[0] / 2,
 257             ((u_int16_t *)op)[0] / 2);
 258 #endif
 259         /* unpin both pages */
 260         old_bufp->flags &= ~BUF_PIN;
 261         new_bufp->flags &= ~BUF_PIN;
 262         return (0);
 263 }
 264
 265 /*
 266  * Called when we encounter an overflow or big key/data page during split
 267  * handling.  This is special cased since we have to begin checking whether
 268  * the key/data pairs fit on their respective pages and because we may need
 269  * overflow pages for both the old and new pages.
 270  *
 271  * The first page might be a page with regular key/data pairs in which case
 272  * we have a regular overflow condition and just need to go on to the next
 273  * page or it might be a big key/data pair in which case we need to fix the
 274  * big key/data pair.
 275  *
 276  * Returns:
 277  *       0 ==> success
 278  *      -1 ==> failure
 279  */
 280 static int
 281 ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved)
 282         HTAB *hashp;
 283         u_int32_t obucket;      /* Same as __split_page. */
 284         BUFHEAD *old_bufp, *new_bufp;
 285         int copyto;     /* First byte on page which contains key/data values. */
 286         int moved;              /* Number of pairs moved to new page. */
 287 {
 288         BUFHEAD *bufp;          /* Buffer header for ino */
 289         u_int16_t *ino;         /* Page keys come off of */
 290         u_int16_t *np;          /* New page */
 291         u_int16_t *op;          /* Page keys go on to if they aren't moving */
 292
 293         BUFHEAD *last_bfp;      /* Last buf header OVFL needing to be freed */
 294         DBT key, val;
 295         SPLIT_RETURN ret;
 296         u_int16_t n, off, ov_addr, scopyto;
 297         char *cino;             /* Character value of ino */
 298
 299         bufp = old_bufp;
 300         ino = (u_int16_t *)old_bufp->page;
 301         np = (u_int16_t *)new_bufp->page;
 302         op = (u_int16_t *)old_bufp->page;
 303         last_bfp = NULL;
 304         scopyto = (u_int16_t)copyto;    /* ANSI */
 305
 306         n = ino[0] - 1;
 307         while (n < ino[0]) {
 308                 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
 309                         if (__big_split(hashp, old_bufp,
 310                             new_bufp, bufp, bufp->addr, obucket, &ret))
 311                                 return (-1);
 312                         old_bufp = ret.oldp;
 313                         if (!old_bufp)
 314                                 return (-1);
 315                         op = (u_int16_t *)old_bufp->page;
 316                         new_bufp = ret.newp;
 317                         if (!new_bufp)
 318                                 return (-1);
 319                         np = (u_int16_t *)new_bufp->page;
 320                         bufp = ret.nextp;
 321                         if (!bufp)
 322                                 return (0);
 323                         cino = (char *)bufp->page;
 324                         ino = (u_int16_t *)cino;
 325                         last_bfp = ret.nextp;
 326                 } else if (ino[n + 1] == OVFLPAGE) {
 327                         ov_addr = ino[n];
 328                         /*
 329                          * Fix up the old page -- the extra 2 are the fields
 330                          * which contained the overflow information.
 331                          */
 332                         ino[0] -= (moved + 2);
 333                         FREESPACE(ino) =
 334                             scopyto - sizeof(u_int16_t) * (ino[0] + 3);
 335                         OFFSET(ino) = scopyto;
 336
 337                         bufp = __get_buf(hashp, ov_addr, bufp, 0);
 338                         if (!bufp)
 339                                 return (-1);
 340
 341                         ino = (u_int16_t *)bufp->page;
 342                         n = 1;
 343                         scopyto = hashp->BSIZE;
 344                         moved = 0;
 345
 346                         if (last_bfp)
 347                                 __free_ovflpage(hashp, last_bfp);
 348                         last_bfp = bufp;
 349                 }
 350                 /* Move regular sized pairs of there are any */
 351                 off = hashp->BSIZE;
 352                 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
 353                         cino = (char *)ino;
 354                         key.data = (u_char *)cino + ino[n];
 355                         key.size = off - ino[n];
 356                         val.data = (u_char *)cino + ino[n + 1];
 357                         val.size = ino[n] - ino[n + 1];
 358                         off = ino[n + 1];
 359
 360                         if (__call_hash(hashp, key.data, key.size) == obucket) {
 361                                 /* Keep on old page */
 362                                 if (PAIRFITS(op, (&key), (&val)))
 363                                         putpair((char *)op, &key, &val);
 364                                 else {
 365                                         old_bufp =
 366                                             __add_ovflpage(hashp, old_bufp);
 367                                         if (!old_bufp)
 368                                                 return (-1);
 369                                         op = (u_int16_t *)old_bufp->page;
 370                                         putpair((char *)op, &key, &val);
 371                                 }
 372                                 old_bufp->flags |= BUF_MOD;
 373                         } else {
 374                                 /* Move to new page */
 375                                 if (PAIRFITS(np, (&key), (&val)))
 376                                         putpair((char *)np, &key, &val);
 377                                 else {
 378                                         new_bufp =
 379                                             __add_ovflpage(hashp, new_bufp);
 380                                         if (!new_bufp)
 381                                                 return (-1);
 382                                         np = (u_int16_t *)new_bufp->page;
 383                                         putpair((char *)np, &key, &val);
 384                                 }
 385                                 new_bufp->flags |= BUF_MOD;
 386                         }
 387                 }
 388         }
 389         if (last_bfp)
 390                 __free_ovflpage(hashp, last_bfp);
 391         return (0);
 392 }
 393
 394 /*
 395  * Add the given pair to the page
 396  *
 397  * Returns:
 398  *      0 ==> OK
 399  *      1 ==> failure
 400  */
 401 extern int
 402 __addel(hashp, bufp, key, val)
 403         HTAB *hashp;
 404         BUFHEAD *bufp;
 405         const DBT *key, *val;
 406 {
 407         u_int16_t *bp, *sop;
 408         int do_expand;
 409
 410         bp = (u_int16_t *)bufp->page;
 411         do_expand = 0;
 412         while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
 413                 /* Exception case */
 414                 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
 415                         /* This is the last page of a big key/data pair
 416                            and we need to add another page */
 417                         break;
 418                 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
 419                         bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
 420                         if (!bufp)
 421                                 return (-1);
 422                         bp = (u_int16_t *)bufp->page;
 423                 } else
 424                         /* Try to squeeze key on this page */
 425                         if (FREESPACE(bp) > PAIRSIZE(key, val)) {
 426                                 squeeze_key(bp, key, val);
 427                                 return (0);
 428                         } else {
 429                                 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
 430                                 if (!bufp)
 431                                         return (-1);
 432                                 bp = (u_int16_t *)bufp->page;
 433                         }
 434
 435         if (PAIRFITS(bp, key, val))
 436                 putpair(bufp->page, key, val);
 437         else {
 438                 do_expand = 1;
 439                 bufp = __add_ovflpage(hashp, bufp);
 440                 if (!bufp)
 441                         return (-1);
 442                 sop = (u_int16_t *)bufp->page;
 443
 444                 if (PAIRFITS(sop, key, val))
 445                         putpair((char *)sop, key, val);
 446                 else
 447                         if (__big_insert(hashp, bufp, key, val))
 448                                 return (-1);
 449         }
 450         bufp->flags |= BUF_MOD;
 451         /*
 452          * If the average number of keys per bucket exceeds the fill factor,
 453          * expand the table.
 454          */
 455         hashp->NKEYS++;
 456         if (do_expand ||
 457             (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
 458                 return (__expand_table(hashp));
 459         return (0);
 460 }
 461
 462 /*
 463  *
 464  * Returns:
 465  *      pointer on success
 466  *      NULL on error
 467  */
 468 extern BUFHEAD *
 469 __add_ovflpage(hashp, bufp)
 470         HTAB *hashp;
 471         BUFHEAD *bufp;
 472 {
 473         u_int16_t *sp;
 474         u_int16_t ndx, ovfl_num;
 475 #ifdef DEBUG1
 476         int tmp1, tmp2;
 477 #endif
 478         sp = (u_int16_t *)bufp->page;
 479
 480         /* Check if we are dynamically determining the fill factor */
 481         if (hashp->FFACTOR == DEF_FFACTOR) {
 482                 hashp->FFACTOR = sp[0] >> 1;
 483                 if (hashp->FFACTOR < MIN_FFACTOR)
 484                         hashp->FFACTOR = MIN_FFACTOR;
 485         }
 486         bufp->flags |= BUF_MOD;
 487         ovfl_num = overflow_page(hashp);
 488 #ifdef DEBUG1
 489         tmp1 = bufp->addr;
 490         tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
 491 #endif
 492         if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
 493                 return (NULL);
 494         bufp->ovfl->flags |= BUF_MOD;
 495 #ifdef DEBUG1
 496         (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
 497             tmp1, tmp2, bufp->ovfl->addr);
 498 #endif
 499         ndx = sp[0];
 500         /*
 501          * Since a pair is allocated on a page only if there's room to add
 502          * an overflow page, we know that the OVFL information will fit on
 503          * the page.
 504          */
 505         sp[ndx + 4] = OFFSET(sp);
 506         sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
 507         sp[ndx + 1] = ovfl_num;
 508         sp[ndx + 2] = OVFLPAGE;
 509         sp[0] = ndx + 2;
 510 #ifdef HASH_STATISTICS
 511         hash_overflows++;
 512 #endif
 513         return (bufp->ovfl);
 514 }
 515
 516 /*
 517  * Returns:
 518  *       0 indicates SUCCESS
 519  *      -1 indicates FAILURE
 520  */
 521 extern int
 522 __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap)
 523         HTAB *hashp;
 524         char *p;
 525         u_int32_t bucket;
 526         int is_bucket, is_disk, is_bitmap;
 527 {
 528         int fd, page, size;
 529         int rsize;
 530         u_int16_t *bp;
 531
 532         fd = hashp->fp;
 533         size = hashp->BSIZE;
 534
 535         if ((fd == -1) || !is_disk) {
 536                 PAGE_INIT(p);
 537                 return (0);
 538         }
 539         if (is_bucket)
 540                 page = BUCKET_TO_PAGE(bucket);
 541         else
 542                 page = OADDR_TO_PAGE(bucket);
 543         if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
 544             ((rsize = _read(fd, p, size)) == -1))
 545                 return (-1);
 546         bp = (u_int16_t *)p;
 547         if (!rsize)
 548                 bp[0] = 0;      /* We hit the EOF, so initialize a new page */
 549         else
 550                 if (rsize != size) {
 551                         errno = EFTYPE;
 552                         return (-1);
 553                 }
 554         if (!is_bitmap && !bp[0]) {
 555                 PAGE_INIT(p);
 556         } else
 557                 if (hashp->LORDER != BYTE_ORDER) {
 558                         int i, max;
 559
 560                         if (is_bitmap) {
 561                                 max = hashp->BSIZE >> 2; /* divide by 4 */
 562                                 for (i = 0; i < max; i++)
 563                                         M_32_SWAP(((int *)p)[i]);
 564                         } else {
 565                                 M_16_SWAP(bp[0]);
 566                                 max = bp[0] + 2;
 567                                 for (i = 1; i <= max; i++)
 568                                         M_16_SWAP(bp[i]);
 569                         }
 570                 }
 571         return (0);
 572 }
 573
 574 /*
 575  * Write page p to disk
 576  *
 577  * Returns:
 578  *       0 ==> OK
 579  *      -1 ==>failure
 580  */
 581 extern int
 582 __put_page(hashp, p, bucket, is_bucket, is_bitmap)
 583         HTAB *hashp;
 584         char *p;
 585         u_int32_t bucket;
 586         int is_bucket, is_bitmap;
 587 {
 588         int fd, page, size;
 589         int wsize;
 590
 591         size = hashp->BSIZE;
 592         if ((hashp->fp == -1) && open_temp(hashp))
 593                 return (-1);
 594         fd = hashp->fp;
 595
 596         if (hashp->LORDER != BYTE_ORDER) {
 597                 int i;
 598                 int max;
 599
 600                 if (is_bitmap) {
 601                         max = hashp->BSIZE >> 2;        /* divide by 4 */
 602                         for (i = 0; i < max; i++)
 603                                 M_32_SWAP(((int *)p)[i]);
 604                 } else {
 605                         max = ((u_int16_t *)p)[0] + 2;
 606                         for (i = 0; i <= max; i++)
 607                                 M_16_SWAP(((u_int16_t *)p)[i]);
 608                 }
 609         }
 610         if (is_bucket)
 611                 page = BUCKET_TO_PAGE(bucket);
 612         else
 613                 page = OADDR_TO_PAGE(bucket);
 614         if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
 615             ((wsize = _write(fd, p, size)) == -1))
 616                 /* Errno is set */
 617                 return (-1);
 618         if (wsize != size) {
 619                 errno = EFTYPE;
 620                 return (-1);
 621         }
 622         return (0);
 623 }
 624
 625 #define BYTE_MASK       ((1 << INT_BYTE_SHIFT) -1)
 626 /*
 627  * Initialize a new bitmap page.  Bitmap pages are left in memory
 628  * once they are read in.
 629  */
 630 extern int
 631 __ibitmap(hashp, pnum, nbits, ndx)
 632         HTAB *hashp;
 633         int pnum, nbits, ndx;
 634 {
 635         u_int32_t *ip;
 636         int clearbytes, clearints;
 637
 638         if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
 639                 return (1);
 640         hashp->nmaps++;
 641         clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
 642         clearbytes = clearints << INT_TO_BYTE;
 643         (void)memset((char *)ip, 0, clearbytes);
 644         (void)memset(((char *)ip) + clearbytes, 0xFF,
 645             hashp->BSIZE - clearbytes);
 646         ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
 647         SETBIT(ip, 0);
 648         hashp->BITMAPS[ndx] = (u_int16_t)pnum;
 649         hashp->mapp[ndx] = ip;
 650         return (0);
 651 }
 652
 653 static u_int32_t
 654 first_free(map)
 655         u_int32_t map;
 656 {
 657         u_int32_t i, mask;
 658
 659         mask = 0x1;
 660         for (i = 0; i < BITS_PER_MAP; i++) {
 661                 if (!(mask & map))
 662                         return (i);
 663                 mask = mask << 1;
 664         }
 665         return (i);
 666 }
 667
 668 static u_int16_t
 669 overflow_page(hashp)
 670         HTAB *hashp;
 671 {
 672         u_int32_t *freep;
 673         int max_free, offset, splitnum;
 674         u_int16_t addr;
 675         int bit, first_page, free_bit, free_page, i, in_use_bits, j;
 676 #ifdef DEBUG2
 677         int tmp1, tmp2;
 678 #endif
 679         splitnum = hashp->OVFL_POINT;
 680         max_free = hashp->SPARES[splitnum];
 681
 682         free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
 683         free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
 684
 685         /* Look through all the free maps to find the first free block */
 686         first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
 687         for ( i = first_page; i <= free_page; i++ ) {
 688                 if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
 689                     !(freep = fetch_bitmap(hashp, i)))
 690                         return (0);
 691                 if (i == free_page)
 692                         in_use_bits = free_bit;
 693                 else
 694                         in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
 695
 696                 if (i == first_page) {
 697                         bit = hashp->LAST_FREED &
 698                             ((hashp->BSIZE << BYTE_SHIFT) - 1);
 699                         j = bit / BITS_PER_MAP;
 700                         bit = bit & ~(BITS_PER_MAP - 1);
 701                 } else {
 702                         bit = 0;
 703                         j = 0;
 704                 }
 705                 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
 706                         if (freep[j] != ALL_SET)
 707                                 goto found;
 708         }
 709
 710         /* No Free Page Found */
 711         hashp->LAST_FREED = hashp->SPARES[splitnum];
 712         hashp->SPARES[splitnum]++;
 713         offset = hashp->SPARES[splitnum] -
 714             (splitnum ? hashp->SPARES[splitnum - 1] : 0);
 715
 716 #define OVMSG   "HASH: Out of overflow pages.  Increase page size\n"
 717         if (offset > SPLITMASK) {
 718                 if (++splitnum >= NCACHED) {
 719                         (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
 720                         return (0);
 721                 }
 722                 hashp->OVFL_POINT = splitnum;
 723                 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
 724                 hashp->SPARES[splitnum-1]--;
 725                 offset = 1;
 726         }
 727
 728         /* Check if we need to allocate a new bitmap page */
 729         if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
 730                 free_page++;
 731                 if (free_page >= NCACHED) {
 732                         (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
 733                         return (0);
 734                 }
 735                 /*
 736                  * This is tricky.  The 1 indicates that you want the new page
 737                  * allocated with 1 clear bit.  Actually, you are going to
 738                  * allocate 2 pages from this map.  The first is going to be
 739                  * the map page, the second is the overflow page we were
 740                  * looking for.  The init_bitmap routine automatically, sets
 741                  * the first bit of itself to indicate that the bitmap itself
 742                  * is in use.  We would explicitly set the second bit, but
 743                  * don't have to if we tell init_bitmap not to leave it clear
 744                  * in the first place.
 745                  */
 746                 if (__ibitmap(hashp,
 747                     (int)OADDR_OF(splitnum, offset), 1, free_page))
 748                         return (0);
 749                 hashp->SPARES[splitnum]++;
 750 #ifdef DEBUG2
 751                 free_bit = 2;
 752 #endif
 753                 offset++;
 754                 if (offset > SPLITMASK) {
 755                         if (++splitnum >= NCACHED) {
 756                                 (void)_write(STDERR_FILENO, OVMSG,
 757                                     sizeof(OVMSG) - 1);
 758                                 return (0);
 759                         }
 760                         hashp->OVFL_POINT = splitnum;
 761                         hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
 762                         hashp->SPARES[splitnum-1]--;
 763                         offset = 0;
 764                 }
 765         } else {
 766                 /*
 767                  * Free_bit addresses the last used bit.  Bump it to address
 768                  * the first available bit.
 769                  */
 770                 free_bit++;
 771                 SETBIT(freep, free_bit);
 772         }
 773
 774         /* Calculate address of the new overflow page */
 775         addr = OADDR_OF(splitnum, offset);
 776 #ifdef DEBUG2
 777         (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
 778             addr, free_bit, free_page);
 779 #endif
 780         return (addr);
 781
 782 found:
 783         bit = bit + first_free(freep[j]);
 784         SETBIT(freep, bit);
 785 #ifdef DEBUG2
 786         tmp1 = bit;
 787         tmp2 = i;
 788 #endif
 789         /*
 790          * Bits are addressed starting with 0, but overflow pages are addressed
 791          * beginning at 1. Bit is a bit addressnumber, so we need to increment
 792          * it to convert it to a page number.
 793          */
 794         bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
 795         if (bit >= hashp->LAST_FREED)
 796                 hashp->LAST_FREED = bit - 1;
 797
 798         /* Calculate the split number for this page */
 799         for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
 800         offset = (i ? bit - hashp->SPARES[i - 1] : bit);
 801         if (offset >= SPLITMASK)
 802                 return (0);     /* Out of overflow pages */
 803         addr = OADDR_OF(i, offset);
 804 #ifdef DEBUG2
 805         (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
 806             addr, tmp1, tmp2);
 807 #endif
 808
 809         /* Allocate and return the overflow page */
 810         return (addr);
 811 }
 812
 813 /*
 814  * Mark this overflow page as free.
 815  */
 816 extern void
 817 __free_ovflpage(hashp, obufp)
 818         HTAB *hashp;
 819         BUFHEAD *obufp;
 820 {
 821         u_int16_t addr;
 822         u_int32_t *freep;
 823         int bit_address, free_page, free_bit;
 824         u_int16_t ndx;
 825
 826         addr = obufp->addr;
 827 #ifdef DEBUG1
 828         (void)fprintf(stderr, "Freeing %d\n", addr);
 829 #endif
 830         ndx = (((u_int16_t)addr) >> SPLITSHIFT);
 831         bit_address =
 832             (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
 833          if (bit_address < hashp->LAST_FREED)
 834                 hashp->LAST_FREED = bit_address;
 835         free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
 836         free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
 837
 838         if (!(freep = hashp->mapp[free_page]))
 839                 freep = fetch_bitmap(hashp, free_page);
 840 #ifdef DEBUG
 841         /*
 842          * This had better never happen.  It means we tried to read a bitmap
 843          * that has already had overflow pages allocated off it, and we
 844          * failed to read it from the file.
 845          */
 846         if (!freep)
 847                 assert(0);
 848 #endif
 849         CLRBIT(freep, free_bit);
 850 #ifdef DEBUG2
 851         (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
 852             obufp->addr, free_bit, free_page);
 853 #endif
 854         __reclaim_buf(hashp, obufp);
 855 }
 856
 857 /*
 858  * Returns:
 859  *       0 success
 860  *      -1 failure
 861  */
 862 static int
 863 open_temp(hashp)
 864         HTAB *hashp;
 865 {
 866         sigset_t set, oset;
 867         static char namestr[] = "_hashXXXXXX";
 868
 869         /* Block signals; make sure file goes away at process exit. */
 870         (void)sigfillset(&set);
 871         (void)_sigprocmask(SIG_BLOCK, &set, &oset);
 872         if ((hashp->fp = mkstemp(namestr)) != -1) {
 873                 (void)unlink(namestr);
 874                 (void)_fcntl(hashp->fp, F_SETFD, 1);
 875         }
 876         (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
 877         return (hashp->fp != -1 ? 0 : -1);
 878 }
 879
 880 /*
 881  * We have to know that the key will fit, but the last entry on the page is
 882  * an overflow pair, so we need to shift things.
 883  */
 884 static void
 885 squeeze_key(sp, key, val)
 886         u_int16_t *sp;
 887         const DBT *key, *val;
 888 {
 889         char *p;
 890         u_int16_t free_space, n, off, pageno;
 891
 892         p = (char *)sp;
 893         n = sp[0];
 894         free_space = FREESPACE(sp);
 895         off = OFFSET(sp);
 896
 897         pageno = sp[n - 1];
 898         off -= key->size;
 899         sp[n - 1] = off;
 900         memmove(p + off, key->data, key->size);
 901         off -= val->size;
 902         sp[n] = off;
 903         memmove(p + off, val->data, val->size);
 904         sp[0] = n + 2;
 905         sp[n + 1] = pageno;
 906         sp[n + 2] = OVFLPAGE;
 907         FREESPACE(sp) = free_space - PAIRSIZE(key, val);
 908         OFFSET(sp) = off;
 909 }
 910
 911 static u_int32_t *
 912 fetch_bitmap(hashp, ndx)
 913         HTAB *hashp;
 914         int ndx;
 915 {
 916         if (ndx >= hashp->nmaps)
 917                 return (NULL);
 918         if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
 919                 return (NULL);
 920         if (__get_page(hashp,
 921             (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
 922                 free(hashp->mapp[ndx]);
 923                 return (NULL);
 924         }
 925         return (hashp->mapp[ndx]);
 926 }
 927
 928 #ifdef DEBUG4
 929 int
 930 print_chain(addr)
 931         int addr;
 932 {
 933         BUFHEAD *bufp;
 934         short *bp, oaddr;
 935
 936         (void)fprintf(stderr, "%d ", addr);
 937         bufp = __get_buf(hashp, addr, NULL, 0);
 938         bp = (short *)bufp->page;
 939         while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
 940                 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
 941                 oaddr = bp[bp[0] - 1];
 942                 (void)fprintf(stderr, "%d ", (int)oaddr);
 943                 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
 944                 bp = (short *)bufp->page;
 945         }
 946         (void)fprintf(stderr, "\n");
 947 }
 948 #endif