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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: src/lib/libc/db/hash/hash.c,v 1.8 2000/01/27 23:06:08 jasone Exp $ 33 * $DragonFly: src/lib/libc/db/hash/hash.c,v 1.9 2005/11/12 23:01:55 swildner Exp $ 34 * 35 * @(#)hash.c 8.9 (Berkeley) 6/16/94 36 */ 37 38 #include "namespace.h" 39 #include <sys/param.h> 40 #include <sys/stat.h> 41 42 #include <errno.h> 43 #include <fcntl.h> 44 #include <stdio.h> 45 #include <stdlib.h> 46 #include <string.h> 47 #include <unistd.h> 48 #ifdef DEBUG 49 #include <assert.h> 50 #endif 51 #include "un-namespace.h" 52 53 #include <db.h> 54 #include "hash.h" 55 #include "page.h" 56 #include "extern.h" 57 58 static int alloc_segs(HTAB *, int); 59 static int flush_meta(HTAB *); 60 static int hash_access(HTAB *, ACTION, const DBT *, DBT *); 61 static int hash_close(DB *); 62 static int hash_delete(const DB *, const DBT *, u_int32_t); 63 static int hash_fd(const DB *); 64 static int hash_get(const DB *, const DBT *, DBT *, u_int32_t); 65 static int hash_put(const DB *, DBT *, const DBT *, u_int32_t); 66 static void *hash_realloc(SEGMENT **, int, int); 67 static int hash_seq(const DB *, DBT *, DBT *, u_int32_t); 68 static int hash_sync(const DB *, u_int32_t); 69 static int hdestroy(HTAB *); 70 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *); 71 static int init_htab(HTAB *, int); 72 #if BYTE_ORDER == LITTLE_ENDIAN 73 static void swap_header(HTAB *); 74 static void swap_header_copy(HASHHDR *, HASHHDR *); 75 #endif 76 77 /* Fast arithmetic, relying on powers of 2, */ 78 #define MOD(x, y) ((x) & ((y) - 1)) 79 80 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; } 81 82 /* Return values */ 83 #define SUCCESS (0) 84 #define ERROR (-1) 85 #define ABNORMAL (1) 86 87 #ifdef HASH_STATISTICS 88 int hash_accesses, hash_collisions, hash_expansions, hash_overflows; 89 #endif 90 91 /************************** INTERFACE ROUTINES ***************************/ 92 /* OPEN/CLOSE */ 93 94 extern DB * 95 __hash_open(const char *file, int flags, int mode, const HASHINFO * info, 96 int dflags __unused) 97 { 98 HTAB *hashp; 99 struct stat statbuf; 100 DB *dbp; 101 int bpages, hdrsize, new_table, nsegs, save_errno; 102 103 if ((flags & O_ACCMODE) == O_WRONLY) { 104 errno = EINVAL; 105 return (NULL); 106 } 107 108 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB)))) 109 return (NULL); 110 hashp->fp = -1; 111 112 /* 113 * Even if user wants write only, we need to be able to read 114 * the actual file, so we need to open it read/write. But, the 115 * field in the hashp structure needs to be accurate so that 116 * we can check accesses. 117 */ 118 hashp->flags = flags; 119 120 new_table = 0; 121 if (!file || (flags & O_TRUNC) || 122 (stat(file, &statbuf) && (errno == ENOENT))) { 123 if (errno == ENOENT) 124 errno = 0; /* Just in case someone looks at errno */ 125 new_table = 1; 126 } 127 if (file) { 128 if ((hashp->fp = _open(file, flags, mode)) == -1) 129 RETURN_ERROR(errno, error0); 130 131 /* if the .db file is empty, and we had permission to create 132 a new .db file, then reinitialize the database */ 133 if ((flags & O_CREAT) && 134 _fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0) 135 new_table = 1; 136 137 _fcntl(hashp->fp, F_SETFD, 1); 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 (hashp->BSIZE << BYTE_SHIFT) - 1) >> 183 (hashp->BSHIFT + BYTE_SHIFT); 184 185 hashp->nmaps = bpages; 186 memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_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 = (DB *)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 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 _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 = (HTAB *)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 = (HTAB *)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 = statbuf.st_blksize; 304 hashp->BSHIFT = __log2(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, 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, int nelem) 346 { 347 int nbuckets, nsegs; 348 int l2; 349 350 /* 351 * Divide number of elements by the fill factor and determine a 352 * desired number of buckets. Allocate space for the next greater 353 * power of two number of buckets. 354 */ 355 nelem = (nelem - 1) / hashp->FFACTOR + 1; 356 357 l2 = __log2(MAX(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, 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 > hashp->DSIZE) 378 hashp->DSIZE = nsegs; 379 return (alloc_segs(hashp, 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 fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n", 397 hash_accesses, hash_collisions); 398 fprintf(stderr, "hdestroy: expansions %ld\n", 399 hash_expansions); 400 fprintf(stderr, "hdestroy: overflows %ld\n", 401 hash_overflows); 402 fprintf(stderr, "keys %ld maxp %d segmentcount %d\n", 403 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs); 404 405 for (i = 0; i < NCACHED; i++) 406 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 _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, u_int32_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 = (HTAB *)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, wsize; 482 483 if (!hashp->save_file) 484 return (0); 485 hashp->MAGIC = HASHMAGIC; 486 hashp->VERSION = HASHVERSION; 487 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY)); 488 489 fp = hashp->fp; 490 whdrp = &hashp->hdr; 491 #if BYTE_ORDER == LITTLE_ENDIAN 492 whdrp = &whdr; 493 swap_header_copy(&hashp->hdr, whdrp); 494 #endif 495 if ((lseek(fp, (off_t)0, SEEK_SET) == -1) || 496 ((wsize = _write(fp, whdrp, sizeof(HASHHDR))) == -1)) 497 return (-1); 498 else 499 if (wsize != sizeof(HASHHDR)) { 500 errno = EFTYPE; 501 hashp->error = errno; 502 return (-1); 503 } 504 for (i = 0; i < NCACHED; i++) 505 if (hashp->mapp[i]) 506 if (__put_page(hashp, (char *)hashp->mapp[i], 507 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, u_int32_t flag) 523 { 524 HTAB *hashp; 525 526 hashp = (HTAB *)dbp->internal; 527 if (flag) { 528 hashp->error = EINVAL; 529 errno = EINVAL; 530 return (ERROR); 531 } 532 return (hash_access(hashp, HASH_GET, key, data)); 533 } 534 535 static int 536 hash_put(const DB *dbp, DBT *key, const DBT *data, u_int32_t flag) 537 { 538 HTAB *hashp; 539 540 hashp = (HTAB *)dbp->internal; 541 if (flag && flag != R_NOOVERWRITE) { 542 hashp->error = errno = EINVAL; 543 return (ERROR); 544 } 545 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 546 hashp->error = errno = EPERM; 547 return (ERROR); 548 } 549 return (hash_access(hashp, flag == R_NOOVERWRITE ? 550 HASH_PUTNEW : HASH_PUT, key, __DECONST(DBT *, data))); 551 } 552 553 static int 554 hash_delete(const DB *dbp, const DBT *key, u_int32_t flag) 555 { 556 HTAB *hashp; 557 558 hashp = (HTAB *)dbp->internal; 559 if (flag && flag != R_CURSOR) { 560 hashp->error = errno = EINVAL; 561 return (ERROR); 562 } 563 if ((hashp->flags & O_ACCMODE) == O_RDONLY) { 564 hashp->error = errno = EPERM; 565 return (ERROR); 566 } 567 return (hash_access(hashp, HASH_DELETE, 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, const DBT *key, DBT *val) 575 { 576 BUFHEAD *rbufp; 577 BUFHEAD *bufp, *save_bufp; 578 u_int16_t *bp; 579 int n, ndx, off, size; 580 const char *kp; 581 u_int16_t pageno; 582 583 #ifdef HASH_STATISTICS 584 hash_accesses++; 585 #endif 586 587 off = hashp->BSIZE; 588 size = key->size; 589 kp = (char *)key->data; 590 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0); 591 if (!rbufp) 592 return (ERROR); 593 save_bufp = rbufp; 594 595 /* Pin the bucket chain */ 596 rbufp->flags |= BUF_PIN; 597 for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;) 598 if (bp[1] >= REAL_KEY) { 599 /* Real key/data pair */ 600 if (size == off - *bp && 601 memcmp(kp, rbufp->page + *bp, size) == 0) 602 goto found; 603 off = bp[1]; 604 #ifdef HASH_STATISTICS 605 hash_collisions++; 606 #endif 607 bp += 2; 608 ndx += 2; 609 } else if (bp[1] == OVFLPAGE) { 610 rbufp = __get_buf(hashp, *bp, rbufp, 0); 611 if (!rbufp) { 612 save_bufp->flags &= ~BUF_PIN; 613 return (ERROR); 614 } 615 /* FOR LOOP INIT */ 616 bp = (u_int16_t *)rbufp->page; 617 n = *bp++; 618 ndx = 1; 619 off = hashp->BSIZE; 620 } else if (bp[1] < REAL_KEY) { 621 if ((ndx = 622 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0) 623 goto found; 624 if (ndx == -2) { 625 bufp = rbufp; 626 if (!(pageno = 627 __find_last_page(hashp, &bufp))) { 628 ndx = 0; 629 rbufp = bufp; 630 break; /* FOR */ 631 } 632 rbufp = __get_buf(hashp, pageno, bufp, 0); 633 if (!rbufp) { 634 save_bufp->flags &= ~BUF_PIN; 635 return (ERROR); 636 } 637 /* FOR LOOP INIT */ 638 bp = (u_int16_t *)rbufp->page; 639 n = *bp++; 640 ndx = 1; 641 off = hashp->BSIZE; 642 } else { 643 save_bufp->flags &= ~BUF_PIN; 644 return (ERROR); 645 } 646 } 647 648 /* Not found */ 649 switch (action) { 650 case HASH_PUT: 651 case HASH_PUTNEW: 652 if (__addel(hashp, rbufp, key, val)) { 653 save_bufp->flags &= ~BUF_PIN; 654 return (ERROR); 655 } else { 656 save_bufp->flags &= ~BUF_PIN; 657 return (SUCCESS); 658 } 659 case HASH_GET: 660 case HASH_DELETE: 661 default: 662 save_bufp->flags &= ~BUF_PIN; 663 return (ABNORMAL); 664 } 665 666 found: 667 switch (action) { 668 case HASH_PUTNEW: 669 save_bufp->flags &= ~BUF_PIN; 670 return (ABNORMAL); 671 case HASH_GET: 672 bp = (u_int16_t *)rbufp->page; 673 if (bp[ndx + 1] < REAL_KEY) { 674 if (__big_return(hashp, rbufp, ndx, val, 0)) 675 return (ERROR); 676 } else { 677 val->data = (u_char *)rbufp->page + (int)bp[ndx + 1]; 678 val->size = bp[ndx] - bp[ndx + 1]; 679 } 680 break; 681 case HASH_PUT: 682 if ((__delpair(hashp, rbufp, ndx)) || 683 (__addel(hashp, rbufp, key, val))) { 684 save_bufp->flags &= ~BUF_PIN; 685 return (ERROR); 686 } 687 break; 688 case HASH_DELETE: 689 if (__delpair(hashp, rbufp, ndx)) 690 return (ERROR); 691 break; 692 default: 693 abort(); 694 } 695 save_bufp->flags &= ~BUF_PIN; 696 return (SUCCESS); 697 } 698 699 static int 700 hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t flag) 701 { 702 u_int32_t bucket; 703 BUFHEAD *bufp; 704 HTAB *hashp; 705 u_int16_t *bp, ndx; 706 707 hashp = (HTAB *)dbp->internal; 708 if (flag && flag != R_FIRST && flag != R_NEXT) { 709 hashp->error = errno = EINVAL; 710 return (ERROR); 711 } 712 #ifdef HASH_STATISTICS 713 hash_accesses++; 714 #endif 715 if ((hashp->cbucket < 0) || (flag == R_FIRST)) { 716 hashp->cbucket = 0; 717 hashp->cndx = 1; 718 hashp->cpage = NULL; 719 } 720 721 for (bp = NULL; !bp || !bp[0]; ) { 722 if (!(bufp = hashp->cpage)) { 723 for (bucket = hashp->cbucket; 724 bucket <= hashp->MAX_BUCKET; 725 bucket++, hashp->cndx = 1) { 726 bufp = __get_buf(hashp, bucket, NULL, 0); 727 if (!bufp) 728 return (ERROR); 729 hashp->cpage = bufp; 730 bp = (u_int16_t *)bufp->page; 731 if (bp[0]) 732 break; 733 } 734 hashp->cbucket = bucket; 735 if (hashp->cbucket > hashp->MAX_BUCKET) { 736 hashp->cbucket = -1; 737 return (ABNORMAL); 738 } 739 } else 740 bp = (u_int16_t *)hashp->cpage->page; 741 742 #ifdef DEBUG 743 assert(bp); 744 assert(bufp); 745 #endif 746 while (bp[hashp->cndx + 1] == OVFLPAGE) { 747 bufp = hashp->cpage = 748 __get_buf(hashp, bp[hashp->cndx], bufp, 0); 749 if (!bufp) 750 return (ERROR); 751 bp = (u_int16_t *)(bufp->page); 752 hashp->cndx = 1; 753 } 754 if (!bp[0]) { 755 hashp->cpage = NULL; 756 ++hashp->cbucket; 757 } 758 } 759 ndx = hashp->cndx; 760 if (bp[ndx + 1] < REAL_KEY) { 761 if (__big_keydata(hashp, bufp, key, data, 1)) 762 return (ERROR); 763 } else { 764 key->data = (u_char *)hashp->cpage->page + bp[ndx]; 765 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx]; 766 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1]; 767 data->size = bp[ndx] - bp[ndx + 1]; 768 ndx += 2; 769 if (ndx > bp[0]) { 770 hashp->cpage = NULL; 771 hashp->cbucket++; 772 hashp->cndx = 1; 773 } else 774 hashp->cndx = ndx; 775 } 776 return (SUCCESS); 777 } 778 779 /********************************* UTILITIES ************************/ 780 781 /* 782 * Returns: 783 * 0 ==> OK 784 * -1 ==> Error 785 */ 786 extern int 787 __expand_table(HTAB *hashp) 788 { 789 u_int32_t old_bucket, new_bucket; 790 int dirsize, new_segnum, spare_ndx; 791 792 #ifdef HASH_STATISTICS 793 hash_expansions++; 794 #endif 795 new_bucket = ++hashp->MAX_BUCKET; 796 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK); 797 798 new_segnum = new_bucket >> hashp->SSHIFT; 799 800 /* Check if we need a new segment */ 801 if (new_segnum >= hashp->nsegs) { 802 /* Check if we need to expand directory */ 803 if (new_segnum >= hashp->DSIZE) { 804 /* Reallocate directory */ 805 dirsize = hashp->DSIZE * sizeof(SEGMENT *); 806 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1)) 807 return (-1); 808 hashp->DSIZE = dirsize << 1; 809 } 810 if ((hashp->dir[new_segnum] = 811 (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL) 812 return (-1); 813 hashp->exsegs++; 814 hashp->nsegs++; 815 } 816 /* 817 * If the split point is increasing (MAX_BUCKET's log base 2 818 * * increases), we need to copy the current contents of the spare 819 * split bucket to the next bucket. 820 */ 821 spare_ndx = __log2(hashp->MAX_BUCKET + 1); 822 if (spare_ndx > hashp->OVFL_POINT) { 823 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT]; 824 hashp->OVFL_POINT = spare_ndx; 825 } 826 827 if (new_bucket > hashp->HIGH_MASK) { 828 /* Starting a new doubling */ 829 hashp->LOW_MASK = hashp->HIGH_MASK; 830 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK; 831 } 832 /* Relocate records to the new bucket */ 833 return (__split_page(hashp, old_bucket, new_bucket)); 834 } 835 836 /* 837 * If realloc guarantees that the pointer is not destroyed if the realloc 838 * fails, then this routine can go away. 839 */ 840 static void * 841 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize) 842 { 843 void *p; 844 845 if ( (p = malloc(newsize)) ) { 846 memmove(p, *p_ptr, oldsize); 847 memset((char *)p + oldsize, 0, newsize - oldsize); 848 free(*p_ptr); 849 *p_ptr = p; 850 } 851 return (p); 852 } 853 854 extern u_int32_t 855 __call_hash(HTAB *hashp, const char *k, int len) 856 { 857 int n, bucket; 858 859 n = hashp->hash(k, len); 860 bucket = n & hashp->HIGH_MASK; 861 if (bucket > hashp->MAX_BUCKET) 862 bucket = bucket & hashp->LOW_MASK; 863 return (bucket); 864 } 865 866 /* 867 * Allocate segment table. On error, destroy the table and set errno. 868 * 869 * Returns 0 on success 870 */ 871 static int 872 alloc_segs(HTAB *hashp, int nsegs) 873 { 874 int i; 875 SEGMENT store; 876 877 int save_errno; 878 879 if ((hashp->dir = 880 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) { 881 save_errno = errno; 882 hdestroy(hashp); 883 errno = save_errno; 884 return (-1); 885 } 886 /* Allocate segments */ 887 if ((store = 888 (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) { 889 save_errno = errno; 890 hdestroy(hashp); 891 errno = save_errno; 892 return (-1); 893 } 894 for (i = 0; i < nsegs; i++, hashp->nsegs++) 895 hashp->dir[i] = &store[i << hashp->SSHIFT]; 896 return (0); 897 } 898 899 #if BYTE_ORDER == LITTLE_ENDIAN 900 /* 901 * Hashp->hdr needs to be byteswapped. 902 */ 903 static void 904 swap_header_copy(HASHHDR *srcp, HASHHDR *destp) 905 { 906 int i; 907 908 P_32_COPY(srcp->magic, destp->magic); 909 P_32_COPY(srcp->version, destp->version); 910 P_32_COPY(srcp->lorder, destp->lorder); 911 P_32_COPY(srcp->bsize, destp->bsize); 912 P_32_COPY(srcp->bshift, destp->bshift); 913 P_32_COPY(srcp->dsize, destp->dsize); 914 P_32_COPY(srcp->ssize, destp->ssize); 915 P_32_COPY(srcp->sshift, destp->sshift); 916 P_32_COPY(srcp->ovfl_point, destp->ovfl_point); 917 P_32_COPY(srcp->last_freed, destp->last_freed); 918 P_32_COPY(srcp->max_bucket, destp->max_bucket); 919 P_32_COPY(srcp->high_mask, destp->high_mask); 920 P_32_COPY(srcp->low_mask, destp->low_mask); 921 P_32_COPY(srcp->ffactor, destp->ffactor); 922 P_32_COPY(srcp->nkeys, destp->nkeys); 923 P_32_COPY(srcp->hdrpages, destp->hdrpages); 924 P_32_COPY(srcp->h_charkey, destp->h_charkey); 925 for (i = 0; i < NCACHED; i++) { 926 P_32_COPY(srcp->spares[i], destp->spares[i]); 927 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]); 928 } 929 } 930 931 static void 932 swap_header(HTAB *hashp) 933 { 934 HASHHDR *hdrp; 935 int i; 936 937 hdrp = &hashp->hdr; 938 939 M_32_SWAP(hdrp->magic); 940 M_32_SWAP(hdrp->version); 941 M_32_SWAP(hdrp->lorder); 942 M_32_SWAP(hdrp->bsize); 943 M_32_SWAP(hdrp->bshift); 944 M_32_SWAP(hdrp->dsize); 945 M_32_SWAP(hdrp->ssize); 946 M_32_SWAP(hdrp->sshift); 947 M_32_SWAP(hdrp->ovfl_point); 948 M_32_SWAP(hdrp->last_freed); 949 M_32_SWAP(hdrp->max_bucket); 950 M_32_SWAP(hdrp->high_mask); 951 M_32_SWAP(hdrp->low_mask); 952 M_32_SWAP(hdrp->ffactor); 953 M_32_SWAP(hdrp->nkeys); 954 M_32_SWAP(hdrp->hdrpages); 955 M_32_SWAP(hdrp->h_charkey); 956 for (i = 0; i < NCACHED; i++) { 957 M_32_SWAP(hdrp->spares[i]); 958 M_16_SWAP(hdrp->bitmaps[i]); 959 } 960 } 961 #endif 962