hash.c revision 1.30
1 /* $NetBSD: hash.c,v 1.30 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.c,v 1.30 2008/09/11 12:58:00 joerg Exp $"); 41 42 #include "namespace.h" 43 #include <sys/param.h> 44 #include <sys/stat.h> 45 46 #include <errno.h> 47 #include <fcntl.h> 48 #include <stdio.h> 49 #include <stdlib.h> 50 #include <string.h> 51 #include <unistd.h> 52 #include <assert.h> 53 54 #include <db.h> 55 #include "hash.h" 56 #include "page.h" 57 #include "extern.h" 58 59 static int alloc_segs(HTAB *, int); 60 static int flush_meta(HTAB *); 61 static int hash_access(HTAB *, ACTION, DBT *, DBT *); 62 static int hash_close(DB *); 63 static int hash_delete(const DB *, const DBT *, uint32_t); 64 static int hash_fd(const DB *); 65 static int hash_get(const DB *, const DBT *, DBT *, uint32_t); 66 static int hash_put(const DB *, DBT *, const DBT *, uint32_t); 67 static void *hash_realloc(SEGMENT **, size_t, size_t); 68 static int hash_seq(const DB *, DBT *, DBT *, uint32_t); 69 static int hash_sync(const DB *, uint32_t); 70 static int hdestroy(HTAB *); 71 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *); 72 static int init_htab(HTAB *, size_t); 73 #if BYTE_ORDER == LITTLE_ENDIAN 74 static void swap_header(HTAB *); 75 static void swap_header_copy(HASHHDR *, HASHHDR *); 76 #endif 77 78 /* Fast arithmetic, relying on powers of 2, */ 79 #define MOD(x, y) ((x) & ((y) - 1)) 80 81 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; } 82 83 /* Return values */ 84 #define SUCCESS (0) 85 #define ERROR (-1) 86 #define ABNORMAL (1) 87 88 #ifdef HASH_STATISTICS 89 int hash_accesses, hash_collisions, hash_expansions, hash_overflows; 90 #endif 91 92 /************************** INTERFACE ROUTINES ***************************/ 93 /* OPEN/CLOSE */ 94 95 /* ARGSUSED */ 96 DB * 97 __hash_open(const char *file, int flags, mode_t mode, const HASHINFO *info, 98 int dflags) 99 { 100 HTAB *hashp; 101 struct stat statbuf; 102 DB *dbp; 103 int bpages, new_table, nsegs, save_errno; 104 ssize_t hdrsize; 105 106 if ((flags & O_ACCMODE) == O_WRONLY) { 107 errno = EINVAL; 108 return (NULL); 109 } 110 111 if (!(hashp = calloc(1, sizeof(HTAB)))) 112 return (NULL); 113 hashp->fp = -1; 114 115 /* 116 * Even if user wants write only, we need to be able to read 117 * the actual file, so we need to open it read/write. But, the 118 * field in the hashp structure needs to be accurate so that 119 * we can check accesses. 120 */ 121 hashp->flags = flags; 122 123 new_table = 0; 124 if (!file || (flags & O_TRUNC) || 125 (stat(file, &statbuf) && (errno == ENOENT))) { 126 if (errno == ENOENT) 127 errno = 0; /* Just in case someone looks at errno */ 128 new_table = 1; 129 } 130 if (file) { 131 if ((hashp->fp = open(file, flags, mode)) == -1) 132 RETURN_ERROR(errno, error0); 133 if (fcntl(hashp->fp, F_SETFD, FD_CLOEXEC) == -1) 134 RETURN_ERROR(errno, error1); 135 if (fstat(hashp->fp, &statbuf) == -1) 136 RETURN_ERROR(errno, error1); 137 new_table |= statbuf.st_size == 0; 138 } 139 if (new_table) { 140 if (!(hashp = init_hash(hashp, file, info))) 141 RETURN_ERROR(errno, error1); 142 } else { 143 /* Table already exists */ 144 if (info && info->hash) 145 hashp->hash = info->hash; 146 else 147 hashp->hash = __default_hash; 148 149 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR)); 150 #if BYTE_ORDER == LITTLE_ENDIAN 151 swap_header(hashp); 152 #endif 153 if (hdrsize == -1) 154 RETURN_ERROR(errno, error1); 155 if (hdrsize != sizeof(HASHHDR)) 156 RETURN_ERROR(EFTYPE, error1); 157 /* Verify file type, versions and hash function */ 158 if (hashp->MAGIC != HASHMAGIC) 159 RETURN_ERROR(EFTYPE, error1); 160 #define OLDHASHVERSION 1 161 if (hashp->VERSION != HASHVERSION && 162 hashp->VERSION != OLDHASHVERSION) 163 RETURN_ERROR(EFTYPE, error1); 164 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY) 165 RETURN_ERROR(EFTYPE, error1); 166 /* 167 * Figure out how many segments we need. Max_Bucket is the 168 * maximum bucket number, so the number of buckets is 169 * max_bucket + 1. 170 */ 171 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) / 172 hashp->SGSIZE; 173 hashp->nsegs = 0; 174 if (alloc_segs(hashp, nsegs)) 175 /* 176 * If alloc_segs fails, table will have been destroyed 177 * and errno will have been set. 178 */ 179 return (NULL); 180 /* Read in bitmaps */ 181 bpages = (hashp->SPARES[hashp->OVFL_POINT] + 182 (unsigned int)(hashp->BSIZE << BYTE_SHIFT) - 1) >> 183 (hashp->BSHIFT + BYTE_SHIFT); 184 185 hashp->nmaps = bpages; 186 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(uint32_t *)); 187 } 188 189 /* Initialize Buffer Manager */ 190 if (info && info->cachesize) 191 __buf_init(hashp, info->cachesize); 192 else 193 __buf_init(hashp, DEF_BUFSIZE); 194 195 hashp->new_file = new_table; 196 hashp->save_file = file && (hashp->flags & O_RDWR); 197 hashp->cbucket = -1; 198 if (!(dbp = malloc(sizeof(DB)))) { 199 save_errno = errno; 200 hdestroy(hashp); 201 errno = save_errno; 202 return (NULL); 203 } 204 dbp->internal = hashp; 205 dbp->close = hash_close; 206 dbp->del = hash_delete; 207 dbp->fd = hash_fd; 208 dbp->get = hash_get; 209 dbp->put = hash_put; 210 dbp->seq = hash_seq; 211 dbp->sync = hash_sync; 212 dbp->type = DB_HASH; 213 214 #ifdef DEBUG 215 (void)fprintf(stderr, 216 "%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", 217 "init_htab:", 218 "TABLE POINTER ", hashp, 219 "BUCKET SIZE ", hashp->BSIZE, 220 "BUCKET SHIFT ", hashp->BSHIFT, 221 "DIRECTORY SIZE ", hashp->DSIZE, 222 "SEGMENT SIZE ", hashp->SGSIZE, 223 "SEGMENT SHIFT ", hashp->SSHIFT, 224 "FILL FACTOR ", hashp->FFACTOR, 225 "MAX BUCKET ", hashp->MAX_BUCKET, 226 "OVFL POINT ", hashp->OVFL_POINT, 227 "LAST FREED ", hashp->LAST_FREED, 228 "HIGH MASK ", hashp->HIGH_MASK, 229 "LOW MASK ", hashp->LOW_MASK, 230 "NSEGS ", hashp->nsegs, 231 "NKEYS ", hashp->NKEYS); 232 #endif 233 #ifdef HASH_STATISTICS 234 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0; 235 #endif 236 return (dbp); 237 238 error1: 239 if (hashp != NULL) 240 (void)close(hashp->fp); 241 242 error0: 243 free(hashp); 244 errno = save_errno; 245 return (NULL); 246 } 247 248 static int 249 hash_close(DB *dbp) 250 { 251 HTAB *hashp; 252 int retval; 253 254 if (!dbp) 255 return (ERROR); 256 257 hashp = dbp->internal; 258 retval = hdestroy(hashp); 259 free(dbp); 260 return (retval); 261 } 262 263 static int 264 hash_fd(const DB *dbp) 265 { 266 HTAB *hashp; 267 268 if (!dbp) 269 return (ERROR); 270 271 hashp = dbp->internal; 272 if (hashp->fp == -1) { 273 errno = ENOENT; 274 return (-1); 275 } 276 return (hashp->fp); 277 } 278 279 /************************** LOCAL CREATION ROUTINES **********************/ 280 static HTAB * 281 init_hash(HTAB *hashp, const char *file, const HASHINFO *info) 282 { 283 struct stat statbuf; 284 int nelem; 285 286 nelem = 1; 287 hashp->NKEYS = 0; 288 hashp->LORDER = BYTE_ORDER; 289 hashp->BSIZE = DEF_BUCKET_SIZE; 290 hashp->BSHIFT = DEF_BUCKET_SHIFT; 291 hashp->SGSIZE = DEF_SEGSIZE; 292 hashp->SSHIFT = DEF_SEGSIZE_SHIFT; 293 hashp->DSIZE = DEF_DIRSIZE; 294 hashp->FFACTOR = DEF_FFACTOR; 295 hashp->hash = __default_hash; 296 memset(hashp->SPARES, 0, sizeof(hashp->SPARES)); 297 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS)); 298 299 /* Fix bucket size to be optimal for file system */ 300 if (file != NULL) { 301 if (stat(file, &statbuf)) 302 return (NULL); 303 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE); 304 hashp->BSHIFT = __log2((uint32_t)hashp->BSIZE); 305 } 306 307 if (info) { 308 if (info->bsize) { 309 /* Round pagesize up to power of 2 */ 310 hashp->BSHIFT = __log2(info->bsize); 311 hashp->BSIZE = 1 << hashp->BSHIFT; 312 if (hashp->BSIZE > MAX_BSIZE) { 313 errno = EINVAL; 314 return (NULL); 315 } 316 } 317 if (info->ffactor) 318 hashp->FFACTOR = info->ffactor; 319 if (info->hash) 320 hashp->hash = info->hash; 321 if (info->nelem) 322 nelem = info->nelem; 323 if (info->lorder) { 324 if (info->lorder != BIG_ENDIAN && 325 info->lorder != LITTLE_ENDIAN) { 326 errno = EINVAL; 327 return (NULL); 328 } 329 hashp->LORDER = info->lorder; 330 } 331 } 332 /* init_htab should destroy the table and set errno if it fails */ 333 if (init_htab(hashp, (size_t)nelem)) 334 return (NULL); 335 else 336 return (hashp); 337 } 338 /* 339 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy 340 * the table and set errno, so we just pass the error information along. 341 * 342 * Returns 0 on No Error 343 */ 344 static int 345 init_htab(HTAB *hashp, size_t nelem) 346 { 347 int nbuckets; 348 uint32_t nsegs; 349 int l2; 350 351 /* 352 * Divide number of elements by the fill factor and determine a 353 * desired number of buckets. Allocate space for the next greater 354 * power of two number of buckets. 355 */ 356 nelem = (nelem - 1) / hashp->FFACTOR + 1; 357 358 _DBFIT(nelem, uint32_t); 359 l2 = __log2(MAX((uint32_t)nelem, 2)); 360 nbuckets = 1 << l2; 361 362 hashp->SPARES[l2] = l2 + 1; 363 hashp->SPARES[l2 + 1] = l2 + 1; 364 hashp->OVFL_POINT = l2; 365 hashp->LAST_FREED = 2; 366 367 /* First bitmap page is at: splitpoint l2 page offset 1 */ 368 if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0)) 369 return (-1); 370 371 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1; 372 hashp->HIGH_MASK = (nbuckets << 1) - 1; 373 /* LINTED constant in conditional context */ 374 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >> 375 hashp->BSHIFT) + 1; 376 377 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1; 378 nsegs = 1 << __log2(nsegs); 379 380 if (nsegs > hashp->DSIZE) 381 hashp->DSIZE = nsegs; 382 return (alloc_segs(hashp, (int)nsegs)); 383 } 384 385 /********************** DESTROY/CLOSE ROUTINES ************************/ 386 387 /* 388 * Flushes any changes to the file if necessary and destroys the hashp 389 * structure, freeing all allocated space. 390 */ 391 static int 392 hdestroy(HTAB *hashp) 393 { 394 int i, save_errno; 395 396 save_errno = 0; 397 398 #ifdef HASH_STATISTICS 399 (void)fprintf(stderr, "hdestroy: accesses %d collisions %d\n", 400 hash_accesses, hash_collisions); 401 (void)fprintf(stderr, "hdestroy: expansions %d\n", 402 hash_expansions); 403 (void)fprintf(stderr, "hdestroy: overflows %d\n", 404 hash_overflows); 405 (void)fprintf(stderr, "keys %d maxp %d segmentcount %d\n", 406 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs); 407 408 for (i = 0; i < NCACHED; i++) 409 (void)fprintf(stderr, 410 "spares[%d] = %d\n", i, hashp->SPARES[i]); 411 #endif 412 /* 413 * Call on buffer manager to free buffers, and if required, 414 * write them to disk. 415 */ 416 if (__buf_free(hashp, 1, hashp->save_file)) 417 save_errno = errno; 418 if (hashp->dir) { 419 free(*hashp->dir); /* Free initial segments */ 420 /* Free extra segments */ 421 while (hashp->exsegs--) 422 free(hashp->dir[--hashp->nsegs]); 423 free(hashp->dir); 424 } 425 if (flush_meta(hashp) && !save_errno) 426 save_errno = errno; 427 /* Free Bigmaps */ 428 for (i = 0; i < hashp->nmaps; i++) 429 if (hashp->mapp[i]) 430 free(hashp->mapp[i]); 431 432 if (hashp->fp != -1) 433 (void)close(hashp->fp); 434 435 free(hashp); 436 437 if (save_errno) { 438 errno = save_errno; 439 return (ERROR); 440 } 441 return (SUCCESS); 442 } 443 /* 444 * Write modified pages to disk 445 * 446 * Returns: 447 * 0 == OK 448 * -1 ERROR 449 */ 450 static int 451 hash_sync(const DB *dbp, uint32_t flags) 452 { 453 HTAB *hashp; 454 455 if (flags != 0) { 456 errno = EINVAL; 457 return (ERROR); 458 } 459 460 if (!dbp) 461 return (ERROR); 462 463 hashp = dbp->internal; 464 if (!hashp->save_file) 465 return (0); 466 if (__buf_free(hashp, 0, 1) || flush_meta(hashp)) 467 return (ERROR); 468 hashp->new_file = 0; 469 return (0); 470 } 471 472 /* 473 * Returns: 474 * 0 == OK 475 * -1 indicates that errno should be set 476 */ 477 static int 478 flush_meta(HTAB *hashp) 479 { 480 HASHHDR *whdrp; 481 #if BYTE_ORDER == LITTLE_ENDIAN 482 HASHHDR whdr; 483 #endif 484 int fp, i; 485 ssize_t wsize; 486 487 if (!hashp->save_file) 488 return (0); 489 hashp->MAGIC = HASHMAGIC; 490 hashp->VERSION = HASHVERSION; 491 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY)); 492 493 fp = hashp->fp; 494 whdrp = &hashp->hdr; 495 #if BYTE_ORDER == LITTLE_ENDIAN 496 whdrp = &whdr; 497 swap_header_copy(&hashp->hdr, whdrp); 498 #endif 499 if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1) 500 return (-1); 501 else 502 if (wsize != sizeof(HASHHDR)) { 503 errno = EFTYPE; 504 hashp->err = errno; 505 return (-1); 506 } 507 for (i = 0; i < NCACHED; i++) 508 if (hashp->mapp[i]) 509 if (__put_page(hashp, (char *)(void *)hashp->mapp[i], 510 (u_int)hashp->BITMAPS[i], 0, 1)) 511 return (-1); 512 return (0); 513 } 514 515 /*******************************SEARCH ROUTINES *****************************/ 516 /* 517 * All the access routines return 518 * 519 * Returns: 520 * 0 on SUCCESS 521 * 1 to indicate an external ERROR (i.e. key not found, etc) 522 * -1 to indicate an internal ERROR (i.e. out of memory, etc) 523 */ 524 static int 525 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag) 526 { 527 HTAB *hashp; 528 529 hashp = dbp->internal; 530 if (flag) { 531 hashp->err = errno = EINVAL; 532 return (ERROR); 533 } 534 return (hash_access(hashp, HASH_GET, __UNCONST(key), data)); 535 } 536 537 static int 538 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag) 539 { 540 HTAB *hashp; 541 542 hashp = dbp->internal; 543 if (flag && flag != R_NOOVERWRITE) { 544 hashp->err = errno = EINVAL; 545 return (ERROR); 546 } 547 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 548 hashp->err = errno = EPERM; 549 return (ERROR); 550 } 551 /* LINTED const castaway */ 552 return (hash_access(hashp, flag == R_NOOVERWRITE ? 553 HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data))); 554 } 555 556 static int 557 hash_delete(const DB *dbp, const DBT *key, uint32_t flag) 558 { 559 HTAB *hashp; 560 561 hashp = dbp->internal; 562 if (flag && flag != R_CURSOR) { 563 hashp->err = errno = EINVAL; 564 return (ERROR); 565 } 566 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 567 hashp->err = errno = EPERM; 568 return (ERROR); 569 } 570 return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL); 571 } 572 573 /* 574 * Assume that hashp has been set in wrapper routine. 575 */ 576 static int 577 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val) 578 { 579 BUFHEAD *rbufp; 580 BUFHEAD *bufp, *save_bufp; 581 uint16_t *bp; 582 int n, ndx, off; 583 size_t size; 584 char *kp; 585 uint16_t pageno; 586 587 #ifdef HASH_STATISTICS 588 hash_accesses++; 589 #endif 590 591 off = hashp->BSIZE; 592 size = key->size; 593 kp = (char *)key->data; 594 rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0); 595 if (!rbufp) 596 return (ERROR); 597 save_bufp = rbufp; 598 599 /* Pin the bucket chain */ 600 rbufp->flags |= BUF_PIN; 601 for (bp = (uint16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;) 602 if (bp[1] >= REAL_KEY) { 603 /* Real key/data pair */ 604 if (size == off - *bp && 605 memcmp(kp, rbufp->page + *bp, size) == 0) 606 goto found; 607 off = bp[1]; 608 #ifdef HASH_STATISTICS 609 hash_collisions++; 610 #endif 611 bp += 2; 612 ndx += 2; 613 } else if (bp[1] == OVFLPAGE) { 614 rbufp = __get_buf(hashp, (uint32_t)*bp, rbufp, 0); 615 if (!rbufp) { 616 save_bufp->flags &= ~BUF_PIN; 617 return (ERROR); 618 } 619 /* FOR LOOP INIT */ 620 bp = (uint16_t *)(void *)rbufp->page; 621 n = *bp++; 622 ndx = 1; 623 off = hashp->BSIZE; 624 } else if (bp[1] < REAL_KEY) { 625 if ((ndx = 626 __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0) 627 goto found; 628 if (ndx == -2) { 629 bufp = rbufp; 630 if (!(pageno = 631 __find_last_page(hashp, &bufp))) { 632 ndx = 0; 633 rbufp = bufp; 634 break; /* FOR */ 635 } 636 rbufp = __get_buf(hashp, (uint32_t)pageno, 637 bufp, 0); 638 if (!rbufp) { 639 save_bufp->flags &= ~BUF_PIN; 640 return (ERROR); 641 } 642 /* FOR LOOP INIT */ 643 bp = (uint16_t *)(void *)rbufp->page; 644 n = *bp++; 645 ndx = 1; 646 off = hashp->BSIZE; 647 } else { 648 save_bufp->flags &= ~BUF_PIN; 649 return (ERROR); 650 } 651 } 652 653 /* Not found */ 654 switch (action) { 655 case HASH_PUT: 656 case HASH_PUTNEW: 657 if (__addel(hashp, rbufp, key, val)) { 658 save_bufp->flags &= ~BUF_PIN; 659 return (ERROR); 660 } else { 661 save_bufp->flags &= ~BUF_PIN; 662 return (SUCCESS); 663 } 664 case HASH_GET: 665 case HASH_DELETE: 666 default: 667 save_bufp->flags &= ~BUF_PIN; 668 return (ABNORMAL); 669 } 670 671 found: 672 switch (action) { 673 case HASH_PUTNEW: 674 save_bufp->flags &= ~BUF_PIN; 675 return (ABNORMAL); 676 case HASH_GET: 677 bp = (uint16_t *)(void *)rbufp->page; 678 if (bp[ndx + 1] < REAL_KEY) { 679 if (__big_return(hashp, rbufp, ndx, val, 0)) 680 return (ERROR); 681 } else { 682 val->data = (uint8_t *)rbufp->page + (int)bp[ndx + 1]; 683 val->size = bp[ndx] - bp[ndx + 1]; 684 } 685 break; 686 case HASH_PUT: 687 if ((__delpair(hashp, rbufp, ndx)) || 688 (__addel(hashp, rbufp, key, val))) { 689 save_bufp->flags &= ~BUF_PIN; 690 return (ERROR); 691 } 692 break; 693 case HASH_DELETE: 694 if (__delpair(hashp, rbufp, ndx)) 695 return (ERROR); 696 break; 697 default: 698 abort(); 699 } 700 save_bufp->flags &= ~BUF_PIN; 701 return (SUCCESS); 702 } 703 704 static int 705 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag) 706 { 707 uint32_t bucket; 708 BUFHEAD *bufp = NULL; /* XXX: gcc */ 709 HTAB *hashp; 710 uint16_t *bp, ndx; 711 712 hashp = dbp->internal; 713 if (flag && flag != R_FIRST && flag != R_NEXT) { 714 hashp->err = errno = EINVAL; 715 return (ERROR); 716 } 717 #ifdef HASH_STATISTICS 718 hash_accesses++; 719 #endif 720 if ((hashp->cbucket < 0) || (flag == R_FIRST)) { 721 hashp->cbucket = 0; 722 hashp->cndx = 1; 723 hashp->cpage = NULL; 724 } 725 726 for (bp = NULL; !bp || !bp[0]; ) { 727 if (!(bufp = hashp->cpage)) { 728 for (bucket = hashp->cbucket; 729 bucket <= hashp->MAX_BUCKET; 730 bucket++, hashp->cndx = 1) { 731 bufp = __get_buf(hashp, bucket, NULL, 0); 732 if (!bufp) 733 return (ERROR); 734 hashp->cpage = bufp; 735 bp = (uint16_t *)(void *)bufp->page; 736 if (bp[0]) 737 break; 738 } 739 hashp->cbucket = bucket; 740 if (hashp->cbucket > hashp->MAX_BUCKET) { 741 hashp->cbucket = -1; 742 return (ABNORMAL); 743 } 744 } else 745 bp = (uint16_t *)(void *)hashp->cpage->page; 746 747 _DIAGASSERT(bp != NULL); 748 _DIAGASSERT(bufp != NULL); 749 while (bp[hashp->cndx + 1] == OVFLPAGE) { 750 bufp = hashp->cpage = 751 __get_buf(hashp, (uint32_t)bp[hashp->cndx], bufp, 752 0); 753 if (!bufp) 754 return (ERROR); 755 bp = (uint16_t *)(void *)(bufp->page); 756 hashp->cndx = 1; 757 } 758 if (!bp[0]) { 759 hashp->cpage = NULL; 760 ++hashp->cbucket; 761 } 762 } 763 ndx = hashp->cndx; 764 if (bp[ndx + 1] < REAL_KEY) { 765 if (__big_keydata(hashp, bufp, key, data, 1)) 766 return (ERROR); 767 } else { 768 if (hashp->cpage == NULL) 769 return (ERROR); 770 key->data = (uint8_t *)hashp->cpage->page + bp[ndx]; 771 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx]; 772 data->data = (uint8_t *)hashp->cpage->page + bp[ndx + 1]; 773 data->size = bp[ndx] - bp[ndx + 1]; 774 ndx += 2; 775 if (ndx > bp[0]) { 776 hashp->cpage = NULL; 777 hashp->cbucket++; 778 hashp->cndx = 1; 779 } else 780 hashp->cndx = ndx; 781 } 782 return (SUCCESS); 783 } 784 785 /********************************* UTILITIES ************************/ 786 787 /* 788 * Returns: 789 * 0 ==> OK 790 * -1 ==> Error 791 */ 792 int 793 __expand_table(HTAB *hashp) 794 { 795 uint32_t old_bucket, new_bucket; 796 int new_segnum, spare_ndx; 797 size_t dirsize; 798 799 #ifdef HASH_STATISTICS 800 hash_expansions++; 801 #endif 802 new_bucket = ++hashp->MAX_BUCKET; 803 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK); 804 805 new_segnum = new_bucket >> hashp->SSHIFT; 806 807 /* Check if we need a new segment */ 808 if (new_segnum >= hashp->nsegs) { 809 /* Check if we need to expand directory */ 810 if (new_segnum >= hashp->DSIZE) { 811 /* Reallocate directory */ 812 dirsize = hashp->DSIZE * sizeof(SEGMENT *); 813 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1)) 814 return (-1); 815 hashp->DSIZE = dirsize << 1; 816 } 817 if ((hashp->dir[new_segnum] = 818 calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL) 819 return (-1); 820 hashp->exsegs++; 821 hashp->nsegs++; 822 } 823 /* 824 * If the split point is increasing (MAX_BUCKET's log base 2 825 * * increases), we need to copy the current contents of the spare 826 * split bucket to the next bucket. 827 */ 828 spare_ndx = __log2((uint32_t)(hashp->MAX_BUCKET + 1)); 829 if (spare_ndx > hashp->OVFL_POINT) { 830 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT]; 831 hashp->OVFL_POINT = spare_ndx; 832 } 833 834 if (new_bucket > hashp->HIGH_MASK) { 835 /* Starting a new doubling */ 836 hashp->LOW_MASK = hashp->HIGH_MASK; 837 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK; 838 } 839 /* Relocate records to the new bucket */ 840 return (__split_page(hashp, old_bucket, new_bucket)); 841 } 842 843 /* 844 * If realloc guarantees that the pointer is not destroyed if the realloc 845 * fails, then this routine can go away. 846 */ 847 static void * 848 hash_realloc(SEGMENT **p_ptr, size_t oldsize, size_t newsize) 849 { 850 void *p; 851 852 if ((p = malloc(newsize)) != NULL) { 853 memmove(p, *p_ptr, oldsize); 854 memset((char *)p + oldsize, 0, newsize - oldsize); 855 free(*p_ptr); 856 *p_ptr = p; 857 } 858 return (p); 859 } 860 861 uint32_t 862 __call_hash(HTAB *hashp, char *k, int len) 863 { 864 int n, bucket; 865 866 n = hashp->hash(k, (size_t)len); 867 bucket = n & hashp->HIGH_MASK; 868 if (bucket > hashp->MAX_BUCKET) 869 bucket = bucket & hashp->LOW_MASK; 870 return (bucket); 871 } 872 873 /* 874 * Allocate segment table. On error, destroy the table and set errno. 875 * 876 * Returns 0 on success 877 */ 878 static int 879 alloc_segs(HTAB *hashp, int nsegs) 880 { 881 int i; 882 SEGMENT store; 883 884 int save_errno; 885 886 hashp->dir = calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *)); 887 if (hashp->dir == NULL) { 888 save_errno = errno; 889 (void)hdestroy(hashp); 890 errno = save_errno; 891 return (-1); 892 } 893 hashp->nsegs = nsegs; 894 if (nsegs == 0) 895 return 0; 896 /* Allocate segments */ 897 store = calloc((size_t)(nsegs << hashp->SSHIFT), sizeof(SEGMENT)); 898 if (store == NULL) { 899 save_errno = errno; 900 (void)hdestroy(hashp); 901 errno = save_errno; 902 return (-1); 903 } 904 for (i = 0; i < nsegs; i++) 905 hashp->dir[i] = &store[i << hashp->SSHIFT]; 906 return (0); 907 } 908 909 #if BYTE_ORDER == LITTLE_ENDIAN 910 /* 911 * Hashp->hdr needs to be byteswapped. 912 */ 913 static void 914 swap_header_copy(HASHHDR *srcp, HASHHDR *destp) 915 { 916 size_t i; 917 918 P_32_COPY(srcp->magic, destp->magic); 919 P_32_COPY(srcp->version, destp->version); 920 P_32_COPY(srcp->lorder, destp->lorder); 921 P_32_COPY(srcp->bsize, destp->bsize); 922 P_32_COPY(srcp->bshift, destp->bshift); 923 P_32_COPY(srcp->dsize, destp->dsize); 924 P_32_COPY(srcp->ssize, destp->ssize); 925 P_32_COPY(srcp->sshift, destp->sshift); 926 P_32_COPY(srcp->ovfl_point, destp->ovfl_point); 927 P_32_COPY(srcp->last_freed, destp->last_freed); 928 P_32_COPY(srcp->max_bucket, destp->max_bucket); 929 P_32_COPY(srcp->high_mask, destp->high_mask); 930 P_32_COPY(srcp->low_mask, destp->low_mask); 931 P_32_COPY(srcp->ffactor, destp->ffactor); 932 P_32_COPY(srcp->nkeys, destp->nkeys); 933 P_32_COPY(srcp->hdrpages, destp->hdrpages); 934 P_32_COPY(srcp->h_charkey, destp->h_charkey); 935 for (i = 0; i < NCACHED; i++) { 936 P_32_COPY(srcp->spares[i], destp->spares[i]); 937 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]); 938 } 939 } 940 941 static void 942 swap_header(HTAB *hashp) 943 { 944 HASHHDR *hdrp; 945 size_t i; 946 947 hdrp = &hashp->hdr; 948 949 M_32_SWAP(hdrp->magic); 950 M_32_SWAP(hdrp->version); 951 M_32_SWAP(hdrp->lorder); 952 M_32_SWAP(hdrp->bsize); 953 M_32_SWAP(hdrp->bshift); 954 M_32_SWAP(hdrp->dsize); 955 M_32_SWAP(hdrp->ssize); 956 M_32_SWAP(hdrp->sshift); 957 M_32_SWAP(hdrp->ovfl_point); 958 M_32_SWAP(hdrp->last_freed); 959 M_32_SWAP(hdrp->max_bucket); 960 M_32_SWAP(hdrp->high_mask); 961 M_32_SWAP(hdrp->low_mask); 962 M_32_SWAP(hdrp->ffactor); 963 M_32_SWAP(hdrp->nkeys); 964 M_32_SWAP(hdrp->hdrpages); 965 M_32_SWAP(hdrp->h_charkey); 966 for (i = 0; i < NCACHED; i++) { 967 M_32_SWAP(hdrp->spares[i]); 968 M_16_SWAP(hdrp->bitmaps[i]); 969 } 970 } 971 #endif 972
Indexes created Sun Sep 28 16:09:52 GMT 2025