1 /* $NetBSD: hash.c,v 1.40 2026/04/19 19:31:50 rillig 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.c,v 1.40 2026/04/19 19:31:50 rillig Exp $"); 41 42 #include "namespace.h" 43 #include <sys/param.h> 44 #include <sys/endian.h> 45 #include <sys/stat.h> 46 47 #include <errno.h> 48 #include <fcntl.h> 49 #include <stdio.h> 50 #include <stdlib.h> 51 #include <string.h> 52 #include <unistd.h> 53 #include <assert.h> 54 55 #include <db.h> 56 #include "hash.h" 57 #include "page.h" 58 #include "extern.h" 59 60 static int alloc_segs(HTAB *, int); 61 static int flush_meta(HTAB *); 62 static int hash_access(HTAB *, ACTION, DBT *, DBT *); 63 static int hash_close(DB *); 64 static int hash_delete(const DB *, const DBT *, uint32_t); 65 static int hash_fd(const DB *); 66 static int hash_get(const DB *, const DBT *, DBT *, uint32_t); 67 static int hash_put(const DB *, DBT *, const DBT *, uint32_t); 68 static void *hash_realloc(SEGMENT **, size_t, size_t); 69 static int hash_seq(const DB *, DBT *, DBT *, uint32_t); 70 static int hash_sync(const DB *, uint32_t); 71 static int hdestroy(HTAB *); 72 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *); 73 static int init_htab(HTAB *, size_t); 74 #if BYTE_ORDER == LITTLE_ENDIAN 75 static void swap_header(HTAB *); 76 static void swap_header_copy(HASHHDR *, HASHHDR *); 77 #endif 78 79 /* Fast arithmetic, relying on powers of 2, */ 80 #define MOD(x, y) ((x) & ((y) - 1)) 81 82 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; } 83 84 /* Return values */ 85 #define SUCCESS (0) 86 #define ERROR (-1) 87 #define ABNORMAL (1) 88 89 #ifdef HASH_STATISTICS 90 int hash_accesses, hash_collisions, hash_expansions, hash_overflows; 91 #endif 92 93 /************************** INTERFACE ROUTINES ***************************/ 94 /* OPEN/CLOSE */ 95 96 /* ARGSUSED */ 97 DB * 98 __hash_open(const char *file, int flags, mode_t mode, const HASHINFO *info, 99 int dflags) 100 { 101 HTAB *hashp; 102 struct stat statbuf; 103 DB *dbp; 104 int bpages, new_table, nsegs, save_errno; 105 ssize_t hdrsize; 106 107 if ((flags & O_ACCMODE) == O_WRONLY) { 108 errno = EINVAL; 109 return (NULL); 110 } 111 112 if (!(hashp = calloc(1, sizeof(HTAB)))) 113 return (NULL); 114 hashp->fp = -1; 115 116 /* 117 * Even if user wants write only, we need to be able to read 118 * the actual file, so we need to open it read/write. But, the 119 * field in the hashp structure needs to be accurate so that 120 * we can check accesses. 121 */ 122 hashp->flags = flags; 123 124 new_table = 0; 125 if (!file || (flags & O_TRUNC) || 126 (stat(file, &statbuf) && (errno == ENOENT))) { 127 if (errno == ENOENT) 128 errno = 0; /* Just in case someone looks at errno */ 129 new_table = 1; 130 } 131 if (file) { 132 if ((hashp->fp = __dbopen(file, flags, mode, &statbuf)) == -1) 133 RETURN_ERROR(errno, error0); 134 new_table |= statbuf.st_size == 0; 135 } 136 if (new_table) { 137 if (!(hashp = init_hash(hashp, file, info))) 138 RETURN_ERROR(errno, error1); 139 } else { 140 /* Table already exists */ 141 if (info && info->hash) 142 hashp->hash = info->hash; 143 else 144 hashp->hash = __default_hash; 145 146 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR)); 147 #if BYTE_ORDER == LITTLE_ENDIAN 148 swap_header(hashp); 149 #endif 150 if (hdrsize == -1) 151 RETURN_ERROR(errno, error1); 152 if (hdrsize != sizeof(HASHHDR)) 153 RETURN_ERROR(EFTYPE, error1); 154 /* Verify file type, versions and hash function */ 155 if (hashp->MAGIC != HASHMAGIC) 156 RETURN_ERROR(EFTYPE, error1); 157 #define OLDHASHVERSION 1 158 if (hashp->VERSION != HASHVERSION && 159 hashp->VERSION != OLDHASHVERSION) 160 RETURN_ERROR(EFTYPE, error1); 161 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != 162 (uint32_t)hashp->H_CHARKEY) 163 RETURN_ERROR(EFTYPE, error1); 164 /* 165 * Figure out how many segments we need. Max_Bucket is the 166 * maximum bucket number, so the number of buckets is 167 * max_bucket + 1. 168 */ 169 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) / 170 hashp->SGSIZE; 171 hashp->nsegs = 0; 172 if (alloc_segs(hashp, nsegs)) 173 /* 174 * If alloc_segs fails, table will have been destroyed 175 * and errno will have been set. 176 */ 177 return (NULL); 178 /* Read in bitmaps */ 179 bpages = (hashp->SPARES[hashp->OVFL_POINT] + 180 (unsigned int)(hashp->BSIZE << BYTE_SHIFT) - 1) >> 181 (hashp->BSHIFT + BYTE_SHIFT); 182 183 hashp->nmaps = bpages; 184 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(uint32_t *)); 185 } 186 187 /* Initialize Buffer Manager */ 188 if (info && info->cachesize) 189 __buf_init(hashp, info->cachesize); 190 else 191 __buf_init(hashp, DEF_BUFSIZE); 192 193 hashp->new_file = new_table; 194 hashp->save_file = file && (hashp->flags & O_RDWR); 195 hashp->cbucket = -1; 196 if (!(dbp = malloc(sizeof(*dbp)))) { 197 save_errno = errno; 198 hdestroy(hashp); 199 errno = save_errno; 200 return (NULL); 201 } 202 dbp->internal = hashp; 203 dbp->close = hash_close; 204 dbp->del = hash_delete; 205 dbp->fd = hash_fd; 206 dbp->get = hash_get; 207 dbp->put = hash_put; 208 dbp->seq = hash_seq; 209 dbp->sync = hash_sync; 210 dbp->type = DB_HASH; 211 212 #ifdef DEBUG1 213 (void)fprintf(stderr, 214 "%s\n%s%p\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%x\n%s%x\n%s%d\n%s%d\n", 215 "init_htab:", 216 "TABLE POINTER ", hashp, 217 "BUCKET SIZE ", hashp->BSIZE, 218 "BUCKET SHIFT ", hashp->BSHIFT, 219 "DIRECTORY SIZE ", hashp->DSIZE, 220 "SEGMENT SIZE ", hashp->SGSIZE, 221 "SEGMENT SHIFT ", hashp->SSHIFT, 222 "FILL FACTOR ", hashp->FFACTOR, 223 "MAX BUCKET ", hashp->MAX_BUCKET, 224 "OVFL POINT ", hashp->OVFL_POINT, 225 "LAST FREED ", hashp->LAST_FREED, 226 "HIGH MASK ", hashp->HIGH_MASK, 227 "LOW MASK ", hashp->LOW_MASK, 228 "NSEGS ", hashp->nsegs, 229 "NKEYS ", hashp->NKEYS); 230 #endif 231 #ifdef HASH_STATISTICS 232 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0; 233 #endif 234 return (dbp); 235 236 error1: 237 if (hashp != NULL) 238 (void)close(hashp->fp); 239 240 error0: 241 free(hashp); 242 errno = save_errno; 243 return (NULL); 244 } 245 246 static int 247 hash_close(DB *dbp) 248 { 249 HTAB *hashp; 250 int retval; 251 252 if (!dbp) 253 return (ERROR); 254 255 hashp = dbp->internal; 256 retval = hdestroy(hashp); 257 free(dbp); 258 return (retval); 259 } 260 261 static int 262 hash_fd(const DB *dbp) 263 { 264 HTAB *hashp; 265 266 if (!dbp) 267 return (ERROR); 268 269 hashp = dbp->internal; 270 if (hashp->fp == -1) { 271 errno = ENOENT; 272 return (-1); 273 } 274 return (hashp->fp); 275 } 276 277 /************************** LOCAL CREATION ROUTINES **********************/ 278 static HTAB * 279 init_hash(HTAB *hashp, const char *file, const HASHINFO *info) 280 { 281 struct stat statbuf; 282 int nelem; 283 284 nelem = 1; 285 hashp->NKEYS = 0; 286 hashp->LORDER = BYTE_ORDER; 287 hashp->BSIZE = DEF_BUCKET_SIZE; 288 hashp->BSHIFT = DEF_BUCKET_SHIFT; 289 hashp->SGSIZE = DEF_SEGSIZE; 290 hashp->SSHIFT = DEF_SEGSIZE_SHIFT; 291 hashp->DSIZE = DEF_DIRSIZE; 292 hashp->FFACTOR = DEF_FFACTOR; 293 hashp->hash = __default_hash; 294 memset(hashp->SPARES, 0, sizeof(hashp->SPARES)); 295 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS)); 296 297 /* Fix bucket size to be optimal for file system */ 298 if (file != NULL) { 299 if (stat(file, &statbuf)) 300 return (NULL); 301 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE); 302 hashp->BSHIFT = __log2((uint32_t)hashp->BSIZE); 303 } 304 305 if (info) { 306 if (info->bsize) { 307 /* Round pagesize up to power of 2 */ 308 hashp->BSHIFT = __log2(info->bsize); 309 hashp->BSIZE = 1 << hashp->BSHIFT; 310 if (hashp->BSIZE > MAX_BSIZE) { 311 errno = EINVAL; 312 return (NULL); 313 } 314 } 315 if (info->ffactor) 316 hashp->FFACTOR = info->ffactor; 317 if (info->hash) 318 hashp->hash = info->hash; 319 if (info->nelem) 320 nelem = info->nelem; 321 if (info->lorder) { 322 if (info->lorder != BIG_ENDIAN && 323 info->lorder != LITTLE_ENDIAN) { 324 errno = EINVAL; 325 return (NULL); 326 } 327 hashp->LORDER = info->lorder; 328 } 329 } 330 /* init_htab should destroy the table and set errno if it fails */ 331 if (init_htab(hashp, (size_t)nelem)) 332 return (NULL); 333 else 334 return (hashp); 335 } 336 /* 337 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy 338 * the table and set errno, so we just pass the error information along. 339 * 340 * Returns 0 on No Error 341 */ 342 static int 343 init_htab(HTAB *hashp, size_t nelem) 344 { 345 int nbuckets; 346 uint32_t nsegs; 347 int l2; 348 349 /* 350 * Divide number of elements by the fill factor and determine a 351 * desired number of buckets. Allocate space for the next greater 352 * power of two number of buckets. 353 */ 354 nelem = (nelem - 1) / hashp->FFACTOR + 1; 355 356 _DBFIT(nelem, uint32_t); 357 l2 = __log2(MAX((uint32_t)nelem, 2)); 358 nbuckets = 1 << l2; 359 360 hashp->SPARES[l2] = l2 + 1; 361 hashp->SPARES[l2 + 1] = l2 + 1; 362 hashp->OVFL_POINT = l2; 363 hashp->LAST_FREED = 2; 364 365 /* First bitmap page is at: splitpoint l2 page offset 1 */ 366 if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0)) 367 return (-1); 368 369 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1; 370 hashp->HIGH_MASK = (nbuckets << 1) - 1; 371 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >> 372 hashp->BSHIFT) + 1; 373 374 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1; 375 nsegs = 1 << __log2(nsegs); 376 377 if (nsegs > (uint32_t)hashp->DSIZE) 378 hashp->DSIZE = nsegs; 379 return (alloc_segs(hashp, (int)nsegs)); 380 } 381 382 /********************** DESTROY/CLOSE ROUTINES ************************/ 383 384 /* 385 * Flushes any changes to the file if necessary and destroys the hashp 386 * structure, freeing all allocated space. 387 */ 388 static int 389 hdestroy(HTAB *hashp) 390 { 391 int i, save_errno; 392 393 save_errno = 0; 394 395 #ifdef HASH_STATISTICS 396 (void)fprintf(stderr, "hdestroy: accesses %d collisions %d\n", 397 hash_accesses, hash_collisions); 398 (void)fprintf(stderr, "hdestroy: expansions %d\n", 399 hash_expansions); 400 (void)fprintf(stderr, "hdestroy: overflows %d\n", 401 hash_overflows); 402 (void)fprintf(stderr, "keys %d maxp %d segmentcount %d\n", 403 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs); 404 405 for (i = 0; i < NCACHED; i++) 406 (void)fprintf(stderr, 407 "spares[%d] = %d\n", i, hashp->SPARES[i]); 408 #endif 409 /* 410 * Call on buffer manager to free buffers, and if required, 411 * write them to disk. 412 */ 413 if (__buf_free(hashp, 1, hashp->save_file)) 414 save_errno = errno; 415 if (hashp->dir) { 416 free(*hashp->dir); /* Free initial segments */ 417 /* Free extra segments */ 418 while (hashp->exsegs--) 419 free(hashp->dir[--hashp->nsegs]); 420 free(hashp->dir); 421 } 422 if (flush_meta(hashp) && !save_errno) 423 save_errno = errno; 424 /* Free Bigmaps */ 425 for (i = 0; i < hashp->nmaps; i++) 426 if (hashp->mapp[i]) 427 free(hashp->mapp[i]); 428 429 if (hashp->fp != -1) 430 (void)close(hashp->fp); 431 432 free(hashp); 433 434 if (save_errno) { 435 errno = save_errno; 436 return (ERROR); 437 } 438 return (SUCCESS); 439 } 440 /* 441 * Write modified pages to disk 442 * 443 * Returns: 444 * 0 == OK 445 * -1 ERROR 446 */ 447 static int 448 hash_sync(const DB *dbp, uint32_t flags) 449 { 450 HTAB *hashp; 451 452 if (flags != 0) { 453 errno = EINVAL; 454 return (ERROR); 455 } 456 457 if (!dbp) 458 return (ERROR); 459 460 hashp = dbp->internal; 461 if (!hashp->save_file) 462 return (0); 463 if (__buf_free(hashp, 0, 1) || flush_meta(hashp)) 464 return (ERROR); 465 hashp->new_file = 0; 466 return (0); 467 } 468 469 /* 470 * Returns: 471 * 0 == OK 472 * -1 indicates that errno should be set 473 */ 474 static int 475 flush_meta(HTAB *hashp) 476 { 477 HASHHDR *whdrp; 478 #if BYTE_ORDER == LITTLE_ENDIAN 479 HASHHDR whdr; 480 #endif 481 int fp, i; 482 ssize_t wsize; 483 484 if (!hashp->save_file) 485 return (0); 486 hashp->MAGIC = HASHMAGIC; 487 hashp->VERSION = HASHVERSION; 488 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY)); 489 490 fp = hashp->fp; 491 whdrp = &hashp->hdr; 492 #if BYTE_ORDER == LITTLE_ENDIAN 493 whdrp = &whdr; 494 swap_header_copy(&hashp->hdr, whdrp); 495 #endif 496 if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1) 497 return (-1); 498 else 499 if (wsize != sizeof(HASHHDR)) { 500 errno = EFTYPE; 501 hashp->err = errno; 502 return (-1); 503 } 504 for (i = 0; i < NCACHED; i++) 505 if (hashp->mapp[i]) 506 if (__put_page(hashp, (char *)(void *)hashp->mapp[i], 507 (u_int)hashp->BITMAPS[i], 0, 1)) 508 return (-1); 509 return (0); 510 } 511 512 /*******************************SEARCH ROUTINES *****************************/ 513 /* 514 * All the access routines return 515 * 516 * Returns: 517 * 0 on SUCCESS 518 * 1 to indicate an external ERROR (i.e. key not found, etc) 519 * -1 to indicate an internal ERROR (i.e. out of memory, etc) 520 */ 521 static int 522 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag) 523 { 524 HTAB *hashp; 525 526 hashp = dbp->internal; 527 if (flag) { 528 hashp->err = errno = EINVAL; 529 return (ERROR); 530 } 531 return (hash_access(hashp, HASH_GET, __UNCONST(key), data)); 532 } 533 534 static int 535 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag) 536 { 537 HTAB *hashp; 538 539 hashp = dbp->internal; 540 if (flag && flag != R_NOOVERWRITE) { 541 hashp->err = errno = EINVAL; 542 return (ERROR); 543 } 544 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 545 hashp->err = errno = EPERM; 546 return (ERROR); 547 } 548 /* LINTED const castaway */ 549 return (hash_access(hashp, flag == R_NOOVERWRITE ? 550 HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data))); 551 } 552 553 static int 554 hash_delete(const DB *dbp, const DBT *key, uint32_t flag) 555 { 556 HTAB *hashp; 557 558 hashp = dbp->internal; 559 if (flag && flag != R_CURSOR) { 560 hashp->err = errno = EINVAL; 561 return (ERROR); 562 } 563 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 564 hashp->err = errno = EPERM; 565 return (ERROR); 566 } 567 return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL); 568 } 569 570 /* 571 * Assume that hashp has been set in wrapper routine. 572 */ 573 static int 574 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val) 575 { 576 BUFHEAD *rbufp; 577 BUFHEAD *bufp, *save_bufp; 578 uint16_t *bp; 579 int n, ndx, off; 580 size_t size; 581 char *kp; 582 uint16_t pageno; 583 584 #ifdef HASH_STATISTICS 585 hash_accesses++; 586 #endif 587 588 off = HASH_BSIZE(hashp); 589 size = key->size; 590 kp = (char *)key->data; 591 rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0); 592 if (!rbufp) 593 return (ERROR); 594 save_bufp = rbufp; 595 596 /* Pin the bucket chain */ 597 rbufp->flags |= BUF_PIN; 598 for (bp = (uint16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;) 599 if (bp[1] >= REAL_KEY) { 600 /* Real key/data pair */ 601 if (size == (size_t)(off - *bp) && 602 memcmp(kp, rbufp->page + *bp, size) == 0) 603 goto found; 604 off = bp[1]; 605 #ifdef HASH_STATISTICS 606 hash_collisions++; 607 #endif 608 bp += 2; 609 ndx += 2; 610 } else if (bp[1] == OVFLPAGE) { 611 rbufp = __get_buf(hashp, (uint32_t)*bp, rbufp, 0); 612 if (!rbufp) { 613 save_bufp->flags &= ~BUF_PIN; 614 return (ERROR); 615 } 616 /* FOR LOOP INIT */ 617 bp = (uint16_t *)(void *)rbufp->page; 618 n = *bp++; 619 ndx = 1; 620 off = HASH_BSIZE(hashp); 621 } else if (bp[1] < REAL_KEY) { 622 if ((ndx = 623 __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0) 624 goto found; 625 if (ndx == -2) { 626 bufp = rbufp; 627 if (!(pageno = 628 __find_last_page(hashp, &bufp))) { 629 ndx = 0; 630 rbufp = bufp; 631 break; /* FOR */ 632 } 633 rbufp = __get_buf(hashp, (uint32_t)pageno, 634 bufp, 0); 635 if (!rbufp) { 636 save_bufp->flags &= ~BUF_PIN; 637 return (ERROR); 638 } 639 /* FOR LOOP INIT */ 640 bp = (uint16_t *)(void *)rbufp->page; 641 n = *bp++; 642 ndx = 1; 643 off = HASH_BSIZE(hashp); 644 } else { 645 save_bufp->flags &= ~BUF_PIN; 646 return (ERROR); 647 } 648 } 649 650 /* Not found */ 651 switch (action) { 652 case HASH_PUT: 653 case HASH_PUTNEW: 654 if (__addel(hashp, rbufp, key, val)) { 655 save_bufp->flags &= ~BUF_PIN; 656 return (ERROR); 657 } else { 658 save_bufp->flags &= ~BUF_PIN; 659 return (SUCCESS); 660 } 661 case HASH_GET: 662 case HASH_DELETE: 663 default: 664 save_bufp->flags &= ~BUF_PIN; 665 return (ABNORMAL); 666 } 667 668 found: 669 switch (action) { 670 case HASH_PUTNEW: 671 save_bufp->flags &= ~BUF_PIN; 672 return (ABNORMAL); 673 case HASH_GET: 674 bp = (uint16_t *)(void *)rbufp->page; 675 if (bp[ndx + 1] < REAL_KEY) { 676 if (__big_return(hashp, rbufp, ndx, val, 0)) 677 return (ERROR); 678 } else { 679 val->data = (uint8_t *)rbufp->page + (int)bp[ndx + 1]; 680 val->size = bp[ndx] - bp[ndx + 1]; 681 } 682 break; 683 case HASH_PUT: 684 if ((__delpair(hashp, rbufp, ndx)) || 685 (__addel(hashp, rbufp, key, val))) { 686 save_bufp->flags &= ~BUF_PIN; 687 return (ERROR); 688 } 689 break; 690 case HASH_DELETE: 691 if (__delpair(hashp, rbufp, ndx)) 692 return (ERROR); 693 /* 694 * Our index lags 2 behind on the same page when we are 695 * deleting the element pointed to by the index; otherwise 696 * deleting randomly from an iterated hash produces undefined 697 * results. 698 */ 699 if (ndx != hashp->cndx - 2 || rbufp != hashp->cpage) 700 break; 701 702 if (hashp->cndx > 1) { 703 /* Move back one element */ 704 hashp->cndx -= 2; 705 } else { 706 /* 707 * Move back one page, and indicate to go to the last 708 * element of the previous page by setting cndx to -1 709 */ 710 hashp->cbucket--; 711 hashp->cpage = NULL; 712 hashp->cndx = -1; 713 } 714 break; 715 default: 716 abort(); 717 } 718 save_bufp->flags &= ~BUF_PIN; 719 return (SUCCESS); 720 } 721 722 static int 723 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag) 724 { 725 uint32_t bucket; 726 BUFHEAD *bufp = NULL; /* XXX: gcc */ 727 HTAB *hashp; 728 uint16_t *bp, ndx; 729 730 hashp = dbp->internal; 731 if (flag && flag != R_FIRST && flag != R_NEXT) { 732 hashp->err = errno = EINVAL; 733 return (ERROR); 734 } 735 #ifdef HASH_STATISTICS 736 hash_accesses++; 737 #endif 738 if ((hashp->cbucket < 0) || (flag == R_FIRST)) { 739 hashp->cbucket = 0; 740 hashp->cndx = 1; 741 hashp->cpage = NULL; 742 } 743 744 next_bucket: 745 for (bp = NULL; !bp || !bp[0]; ) { 746 if (!(bufp = hashp->cpage)) { 747 for (bucket = hashp->cbucket; 748 bucket <= (uint32_t)hashp->MAX_BUCKET; 749 bucket++) { 750 bufp = __get_buf(hashp, bucket, NULL, 0); 751 if (!bufp) 752 return (ERROR); 753 hashp->cpage = bufp; 754 bp = (uint16_t *)(void *)bufp->page; 755 if (bp[0]) 756 break; 757 } 758 hashp->cbucket = bucket; 759 if (hashp->cbucket > hashp->MAX_BUCKET) { 760 hashp->cbucket = -1; 761 return (ABNORMAL); 762 } 763 if (hashp->cndx == -1) { 764 /* move to the last element of the page */ 765 hashp->cndx = 1; 766 while (bp[hashp->cndx - 1] != 0) 767 hashp->cndx += 2; 768 } else { 769 /* start on the first element */ 770 hashp->cndx = 1; 771 } 772 } else { 773 bp = (uint16_t *)(void *)bufp->page; 774 if (flag == R_NEXT || flag == 0) { 775 if (hashp->cndx > bp[0]) { 776 hashp->cpage = NULL; 777 hashp->cbucket++; 778 hashp->cndx = 1; 779 goto next_bucket; 780 } 781 } 782 } 783 784 785 _DIAGASSERT(bp != NULL); 786 _DIAGASSERT(bufp != NULL); 787 while (bp[hashp->cndx + 1] == OVFLPAGE) { 788 bufp = hashp->cpage = 789 __get_buf(hashp, (uint32_t)bp[hashp->cndx], bufp, 790 0); 791 if (!bufp) 792 return (ERROR); 793 bp = (uint16_t *)(void *)(bufp->page); 794 hashp->cndx = 1; 795 } 796 if (!bp[0]) { 797 hashp->cpage = NULL; 798 ++hashp->cbucket; 799 } 800 } 801 ndx = hashp->cndx; 802 if (bp[ndx + 1] < REAL_KEY) { 803 if (__big_keydata(hashp, bufp, key, data, 1)) 804 return (ERROR); 805 hashp->cndx = 1; 806 } else { 807 if (hashp->cpage == NULL) 808 return (ERROR); 809 key->data = (uint8_t *)hashp->cpage->page + bp[ndx]; 810 key->size = (ndx > 1 ? bp[ndx - 1] : HASH_BSIZE(hashp)) - bp[ndx]; 811 data->data = (uint8_t *)hashp->cpage->page + bp[ndx + 1]; 812 data->size = bp[ndx] - bp[ndx + 1]; 813 hashp->cndx += 2; 814 } 815 return (SUCCESS); 816 } 817 818 /********************************* UTILITIES ************************/ 819 820 /* 821 * Returns: 822 * 0 ==> OK 823 * -1 ==> Error 824 */ 825 int 826 __expand_table(HTAB *hashp) 827 { 828 uint32_t old_bucket, new_bucket; 829 int new_segnum, spare_ndx; 830 size_t dirsize; 831 832 #ifdef HASH_STATISTICS 833 hash_expansions++; 834 #endif 835 new_bucket = ++hashp->MAX_BUCKET; 836 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK); 837 838 new_segnum = new_bucket >> hashp->SSHIFT; 839 840 /* Check if we need a new segment */ 841 if (new_segnum >= hashp->nsegs) { 842 /* Check if we need to expand directory */ 843 if (new_segnum >= hashp->DSIZE) { 844 /* Reallocate directory */ 845 dirsize = hashp->DSIZE * sizeof(SEGMENT *); 846 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1)) 847 return (-1); 848 dirsize <<= 1; 849 _DBFIT(dirsize, uint32_t); 850 hashp->DSIZE = (uint32_t)dirsize; 851 } 852 if ((hashp->dir[new_segnum] = 853 calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL) 854 return (-1); 855 hashp->exsegs++; 856 hashp->nsegs++; 857 } 858 /* 859 * If the split point is increasing (MAX_BUCKET's log base 2 860 * * increases), we need to copy the current contents of the spare 861 * split bucket to the next bucket. 862 */ 863 spare_ndx = __log2((uint32_t)(hashp->MAX_BUCKET + 1)); 864 if (spare_ndx > hashp->OVFL_POINT) { 865 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT]; 866 hashp->OVFL_POINT = spare_ndx; 867 } 868 869 if (new_bucket > (uint32_t)hashp->HIGH_MASK) { 870 /* Starting a new doubling */ 871 hashp->LOW_MASK = hashp->HIGH_MASK; 872 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK; 873 } 874 /* Relocate records to the new bucket */ 875 return (__split_page(hashp, old_bucket, new_bucket)); 876 } 877 878 /* 879 * If realloc guarantees that the pointer is not destroyed if the realloc 880 * fails, then this routine can go away. 881 */ 882 static void * 883 hash_realloc(SEGMENT **p_ptr, size_t oldsize, size_t newsize) 884 { 885 void *p; 886 887 if ((p = malloc(newsize)) != NULL) { 888 memmove(p, *p_ptr, oldsize); 889 memset((char *)p + oldsize, 0, newsize - oldsize); 890 free(*p_ptr); 891 *p_ptr = p; 892 } 893 return (p); 894 } 895 896 uint32_t 897 __call_hash(HTAB *hashp, char *k, int len) 898 { 899 int n, bucket; 900 901 n = hashp->hash(k, (size_t)len); 902 bucket = n & hashp->HIGH_MASK; 903 if (bucket > hashp->MAX_BUCKET) 904 bucket = bucket & hashp->LOW_MASK; 905 return (bucket); 906 } 907 908 /* 909 * Allocate segment table. On error, destroy the table and set errno. 910 * 911 * Returns 0 on success 912 */ 913 static int 914 alloc_segs(HTAB *hashp, int nsegs) 915 { 916 int i; 917 SEGMENT store; 918 919 int save_errno; 920 921 hashp->dir = calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *)); 922 if (hashp->dir == NULL) { 923 save_errno = errno; 924 (void)hdestroy(hashp); 925 errno = save_errno; 926 return (-1); 927 } 928 hashp->nsegs = nsegs; 929 if (nsegs == 0) 930 return 0; 931 /* Allocate segments */ 932 store = calloc((size_t)(nsegs << hashp->SSHIFT), sizeof(SEGMENT)); 933 if (store == NULL) { 934 save_errno = errno; 935 (void)hdestroy(hashp); 936 errno = save_errno; 937 return (-1); 938 } 939 for (i = 0; i < nsegs; i++) 940 hashp->dir[i] = &store[i << hashp->SSHIFT]; 941 return (0); 942 } 943 944 #if BYTE_ORDER == LITTLE_ENDIAN 945 /* 946 * Hashp->hdr needs to be byteswapped. 947 */ 948 static void 949 swap_header_copy(HASHHDR *srcp, HASHHDR *destp) 950 { 951 size_t i; 952 953 P_32_COPY(srcp->magic, destp->magic); 954 P_32_COPY(srcp->version, destp->version); 955 P_32_COPY(srcp->lorder, destp->lorder); 956 P_32_COPY(srcp->bsize, destp->bsize); 957 P_32_COPY(srcp->bshift, destp->bshift); 958 P_32_COPY(srcp->dsize, destp->dsize); 959 P_32_COPY(srcp->ssize, destp->ssize); 960 P_32_COPY(srcp->sshift, destp->sshift); 961 P_32_COPY(srcp->ovfl_point, destp->ovfl_point); 962 P_32_COPY(srcp->last_freed, destp->last_freed); 963 P_32_COPY(srcp->max_bucket, destp->max_bucket); 964 P_32_COPY(srcp->high_mask, destp->high_mask); 965 P_32_COPY(srcp->low_mask, destp->low_mask); 966 P_32_COPY(srcp->ffactor, destp->ffactor); 967 P_32_COPY(srcp->nkeys, destp->nkeys); 968 P_32_COPY(srcp->hdrpages, destp->hdrpages); 969 P_32_COPY(srcp->h_charkey, destp->h_charkey); 970 for (i = 0; i < NCACHED; i++) { 971 P_32_COPY(srcp->spares[i], destp->spares[i]); 972 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]); 973 } 974 } 975 976 static void 977 swap_header(HTAB *hashp) 978 { 979 HASHHDR *hdrp; 980 size_t i; 981 982 hdrp = &hashp->hdr; 983 984 M_32_SWAP(hdrp->magic); 985 M_32_SWAP(hdrp->version); 986 M_32_SWAP(hdrp->lorder); 987 M_32_SWAP(hdrp->bsize); 988 M_32_SWAP(hdrp->bshift); 989 M_32_SWAP(hdrp->dsize); 990 M_32_SWAP(hdrp->ssize); 991 M_32_SWAP(hdrp->sshift); 992 M_32_SWAP(hdrp->ovfl_point); 993 M_32_SWAP(hdrp->last_freed); 994 M_32_SWAP(hdrp->max_bucket); 995 M_32_SWAP(hdrp->high_mask); 996 M_32_SWAP(hdrp->low_mask); 997 M_32_SWAP(hdrp->ffactor); 998 M_32_SWAP(hdrp->nkeys); 999 M_32_SWAP(hdrp->hdrpages); 1000 M_32_SWAP(hdrp->h_charkey); 1001 for (i = 0; i < NCACHED; i++) { 1002 M_32_SWAP(hdrp->spares[i]); 1003 M_16_SWAP(hdrp->bitmaps[i]); 1004 } 1005 } 1006 #endif 1007
Indexes created Mon Aug 10 00:24:53 UTC 2026