Home | History | Annotate | Line # | Download | only in hash
hash.c revision 1.8
      1 /*	$NetBSD: hash.c,v 1.8 1995/02/27 13:21:59 cgd 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. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  */
     38 
     39 #if defined(LIBC_SCCS) && !defined(lint)
     40 #if 0
     41 static char sccsid[] = "@(#)hash.c	8.9 (Berkeley) 6/16/94";
     42 #else
     43 static char rcsid[] = "$NetBSD: hash.c,v 1.8 1995/02/27 13:21:59 cgd Exp $";
     44 #endif
     45 #endif /* LIBC_SCCS and not lint */
     46 
     47 #include <sys/param.h>
     48 #include <sys/stat.h>
     49 
     50 #include <errno.h>
     51 #include <fcntl.h>
     52 #include <stdio.h>
     53 #include <stdlib.h>
     54 #include <string.h>
     55 #include <unistd.h>
     56 #ifdef DEBUG
     57 #include <assert.h>
     58 #endif
     59 
     60 #include <db.h>
     61 #include "hash.h"
     62 #include "page.h"
     63 #include "extern.h"
     64 
     65 static int   alloc_segs __P((HTAB *, int));
     66 static int   flush_meta __P((HTAB *));
     67 static int   hash_access __P((HTAB *, ACTION, DBT *, DBT *));
     68 static int   hash_close __P((DB *));
     69 static int   hash_delete __P((const DB *, const DBT *, u_int32_t));
     70 static int   hash_fd __P((const DB *));
     71 static int   hash_get __P((const DB *, const DBT *, DBT *, u_int32_t));
     72 static int   hash_put __P((const DB *, DBT *, const DBT *, u_int32_t));
     73 static void *hash_realloc __P((SEGMENT **, int, int));
     74 static int   hash_seq __P((const DB *, DBT *, DBT *, u_int32_t));
     75 static int   hash_sync __P((const DB *, u_int32_t));
     76 static int   hdestroy __P((HTAB *));
     77 static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *));
     78 static int   init_htab __P((HTAB *, int));
     79 #if BYTE_ORDER == LITTLE_ENDIAN
     80 static void  swap_header __P((HTAB *));
     81 static void  swap_header_copy __P((HASHHDR *, HASHHDR *));
     82 #endif
     83 
     84 /* Fast arithmetic, relying on powers of 2, */
     85 #define MOD(x, y)		((x) & ((y) - 1))
     86 
     87 #define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
     88 
     89 /* Return values */
     90 #define	SUCCESS	 (0)
     91 #define	ERROR	(-1)
     92 #define	ABNORMAL (1)
     93 
     94 #ifdef HASH_STATISTICS
     95 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
     96 #endif
     97 
     98 /************************** INTERFACE ROUTINES ***************************/
     99 /* OPEN/CLOSE */
    100 
    101 extern DB *
    102 __hash_open(file, flags, mode, info, dflags)
    103 	const char *file;
    104 	int flags, mode, dflags;
    105 	const HASHINFO *info;	/* Special directives for create */
    106 {
    107 	HTAB *hashp;
    108 	struct stat statbuf;
    109 	DB *dbp;
    110 	int bpages, hdrsize, new_table, nsegs, save_errno;
    111 
    112 	if ((flags & O_ACCMODE) == O_WRONLY) {
    113 		errno = EINVAL;
    114 		return (NULL);
    115 	}
    116 
    117 	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
    118 		return (NULL);
    119 	hashp->fp = -1;
    120 
    121 	/*
    122 	 * Even if user wants write only, we need to be able to read
    123 	 * the actual file, so we need to open it read/write. But, the
    124 	 * field in the hashp structure needs to be accurate so that
    125 	 * we can check accesses.
    126 	 */
    127 	hashp->flags = flags;
    128 
    129 	new_table = 0;
    130 	if (!file || (flags & O_TRUNC) ||
    131 	    (stat(file, &statbuf) && (errno == ENOENT))) {
    132 		if (errno == ENOENT)
    133 			errno = 0; /* Just in case someone looks at errno */
    134 		new_table = 1;
    135 	}
    136 	if (file) {
    137 		if ((hashp->fp = open(file, flags, mode)) == -1)
    138 			RETURN_ERROR(errno, error0);
    139 		(void)fcntl(hashp->fp, F_SETFD, 1);
    140 	}
    141 	if (new_table) {
    142 		if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
    143 			RETURN_ERROR(errno, error1);
    144 	} else {
    145 		/* Table already exists */
    146 		if (info && info->hash)
    147 			hashp->hash = info->hash;
    148 		else
    149 			hashp->hash = __default_hash;
    150 
    151 		hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
    152 #if BYTE_ORDER == LITTLE_ENDIAN
    153 		swap_header(hashp);
    154 #endif
    155 		if (hdrsize == -1)
    156 			RETURN_ERROR(errno, error1);
    157 		if (hdrsize != sizeof(HASHHDR))
    158 			RETURN_ERROR(EFTYPE, error1);
    159 		/* Verify file type, versions and hash function */
    160 		if (hashp->MAGIC != HASHMAGIC)
    161 			RETURN_ERROR(EFTYPE, error1);
    162 #define	OLDHASHVERSION	1
    163 		if (hashp->VERSION != HASHVERSION &&
    164 		    hashp->VERSION != OLDHASHVERSION)
    165 			RETURN_ERROR(EFTYPE, error1);
    166 		if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
    167 			RETURN_ERROR(EFTYPE, error1);
    168 		/*
    169 		 * Figure out how many segments we need.  Max_Bucket is the
    170 		 * maximum bucket number, so the number of buckets is
    171 		 * max_bucket + 1.
    172 		 */
    173 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
    174 			 hashp->SGSIZE;
    175 		hashp->nsegs = 0;
    176 		if (alloc_segs(hashp, nsegs))
    177 			/*
    178 			 * If alloc_segs fails, table will have been destroyed
    179 			 * and errno will have been set.
    180 			 */
    181 			return (NULL);
    182 		/* Read in bitmaps */
    183 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
    184 		    (hashp->BSIZE << BYTE_SHIFT) - 1) >>
    185 		    (hashp->BSHIFT + BYTE_SHIFT);
    186 
    187 		hashp->nmaps = bpages;
    188 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
    189 	}
    190 
    191 	/* Initialize Buffer Manager */
    192 	if (info && info->cachesize)
    193 		__buf_init(hashp, info->cachesize);
    194 	else
    195 		__buf_init(hashp, DEF_BUFSIZE);
    196 
    197 	hashp->new_file = new_table;
    198 	hashp->save_file = file && (hashp->flags & O_RDWR);
    199 	hashp->cbucket = -1;
    200 	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
    201 		save_errno = errno;
    202 		hdestroy(hashp);
    203 		errno = save_errno;
    204 		return (NULL);
    205 	}
    206 	dbp->internal = hashp;
    207 	dbp->close = hash_close;
    208 	dbp->del = hash_delete;
    209 	dbp->fd = hash_fd;
    210 	dbp->get = hash_get;
    211 	dbp->put = hash_put;
    212 	dbp->seq = hash_seq;
    213 	dbp->sync = hash_sync;
    214 	dbp->type = DB_HASH;
    215 
    216 #ifdef DEBUG
    217 	(void)fprintf(stderr,
    218 "%s\n%s%x\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",
    219 	    "init_htab:",
    220 	    "TABLE POINTER   ", hashp,
    221 	    "BUCKET SIZE     ", hashp->BSIZE,
    222 	    "BUCKET SHIFT    ", hashp->BSHIFT,
    223 	    "DIRECTORY SIZE  ", hashp->DSIZE,
    224 	    "SEGMENT SIZE    ", hashp->SGSIZE,
    225 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
    226 	    "FILL FACTOR     ", hashp->FFACTOR,
    227 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
    228 	    "OVFL POINT	     ", hashp->OVFL_POINT,
    229 	    "LAST FREED      ", hashp->LAST_FREED,
    230 	    "HIGH MASK       ", hashp->HIGH_MASK,
    231 	    "LOW  MASK       ", hashp->LOW_MASK,
    232 	    "NSEGS           ", hashp->nsegs,
    233 	    "NKEYS           ", hashp->NKEYS);
    234 #endif
    235 #ifdef HASH_STATISTICS
    236 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
    237 #endif
    238 	return (dbp);
    239 
    240 error1:
    241 	if (hashp != NULL)
    242 		(void)close(hashp->fp);
    243 
    244 error0:
    245 	free(hashp);
    246 	errno = save_errno;
    247 	return (NULL);
    248 }
    249 
    250 static int
    251 hash_close(dbp)
    252 	DB *dbp;
    253 {
    254 	HTAB *hashp;
    255 	int retval;
    256 
    257 	if (!dbp)
    258 		return (ERROR);
    259 
    260 	hashp = (HTAB *)dbp->internal;
    261 	retval = hdestroy(hashp);
    262 	free(dbp);
    263 	return (retval);
    264 }
    265 
    266 static int
    267 hash_fd(dbp)
    268 	const DB *dbp;
    269 {
    270 	HTAB *hashp;
    271 
    272 	if (!dbp)
    273 		return (ERROR);
    274 
    275 	hashp = (HTAB *)dbp->internal;
    276 	if (hashp->fp == -1) {
    277 		errno = ENOENT;
    278 		return (-1);
    279 	}
    280 	return (hashp->fp);
    281 }
    282 
    283 /************************** LOCAL CREATION ROUTINES **********************/
    284 static HTAB *
    285 init_hash(hashp, file, info)
    286 	HTAB *hashp;
    287 	const char *file;
    288 	HASHINFO *info;
    289 {
    290 	struct stat statbuf;
    291 	int nelem;
    292 
    293 	nelem = 1;
    294 	hashp->NKEYS = 0;
    295 	hashp->LORDER = BYTE_ORDER;
    296 	hashp->BSIZE = DEF_BUCKET_SIZE;
    297 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
    298 	hashp->SGSIZE = DEF_SEGSIZE;
    299 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
    300 	hashp->DSIZE = DEF_DIRSIZE;
    301 	hashp->FFACTOR = DEF_FFACTOR;
    302 	hashp->hash = __default_hash;
    303 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
    304 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
    305 
    306 	/* Fix bucket size to be optimal for file system */
    307 	if (file != NULL) {
    308 		if (stat(file, &statbuf))
    309 			return (NULL);
    310 		hashp->BSIZE = statbuf.st_blksize;
    311 		hashp->BSHIFT = __log2(hashp->BSIZE);
    312 	}
    313 
    314 	if (info) {
    315 		if (info->bsize) {
    316 			/* Round pagesize up to power of 2 */
    317 			hashp->BSHIFT = __log2(info->bsize);
    318 			hashp->BSIZE = 1 << hashp->BSHIFT;
    319 			if (hashp->BSIZE > MAX_BSIZE) {
    320 				errno = EINVAL;
    321 				return (NULL);
    322 			}
    323 		}
    324 		if (info->ffactor)
    325 			hashp->FFACTOR = info->ffactor;
    326 		if (info->hash)
    327 			hashp->hash = info->hash;
    328 		if (info->nelem)
    329 			nelem = info->nelem;
    330 		if (info->lorder) {
    331 			if (info->lorder != BIG_ENDIAN &&
    332 			    info->lorder != LITTLE_ENDIAN) {
    333 				errno = EINVAL;
    334 				return (NULL);
    335 			}
    336 			hashp->LORDER = info->lorder;
    337 		}
    338 	}
    339 	/* init_htab should destroy the table and set errno if it fails */
    340 	if (init_htab(hashp, nelem))
    341 		return (NULL);
    342 	else
    343 		return (hashp);
    344 }
    345 /*
    346  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
    347  * the table and set errno, so we just pass the error information along.
    348  *
    349  * Returns 0 on No Error
    350  */
    351 static int
    352 init_htab(hashp, nelem)
    353 	HTAB *hashp;
    354 	int nelem;
    355 {
    356 	register int nbuckets, nsegs;
    357 	int l2;
    358 
    359 	/*
    360 	 * Divide number of elements by the fill factor and determine a
    361 	 * desired number of buckets.  Allocate space for the next greater
    362 	 * power of two number of buckets.
    363 	 */
    364 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
    365 
    366 	l2 = __log2(MAX(nelem, 2));
    367 	nbuckets = 1 << l2;
    368 
    369 	hashp->SPARES[l2] = l2 + 1;
    370 	hashp->SPARES[l2 + 1] = l2 + 1;
    371 	hashp->OVFL_POINT = l2;
    372 	hashp->LAST_FREED = 2;
    373 
    374 	/* First bitmap page is at: splitpoint l2 page offset 1 */
    375 	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
    376 		return (-1);
    377 
    378 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
    379 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
    380 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
    381 	    hashp->BSHIFT) + 1;
    382 
    383 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
    384 	nsegs = 1 << __log2(nsegs);
    385 
    386 	if (nsegs > hashp->DSIZE)
    387 		hashp->DSIZE = nsegs;
    388 	return (alloc_segs(hashp, nsegs));
    389 }
    390 
    391 /********************** DESTROY/CLOSE ROUTINES ************************/
    392 
    393 /*
    394  * Flushes any changes to the file if necessary and destroys the hashp
    395  * structure, freeing all allocated space.
    396  */
    397 static int
    398 hdestroy(hashp)
    399 	HTAB *hashp;
    400 {
    401 	int i, save_errno;
    402 
    403 	save_errno = 0;
    404 
    405 #ifdef HASH_STATISTICS
    406 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
    407 	    hash_accesses, hash_collisions);
    408 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
    409 	    hash_expansions);
    410 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
    411 	    hash_overflows);
    412 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
    413 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
    414 
    415 	for (i = 0; i < NCACHED; i++)
    416 		(void)fprintf(stderr,
    417 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
    418 #endif
    419 	/*
    420 	 * Call on buffer manager to free buffers, and if required,
    421 	 * write them to disk.
    422 	 */
    423 	if (__buf_free(hashp, 1, hashp->save_file))
    424 		save_errno = errno;
    425 	if (hashp->dir) {
    426 		free(*hashp->dir);	/* Free initial segments */
    427 		/* Free extra segments */
    428 		while (hashp->exsegs--)
    429 			free(hashp->dir[--hashp->nsegs]);
    430 		free(hashp->dir);
    431 	}
    432 	if (flush_meta(hashp) && !save_errno)
    433 		save_errno = errno;
    434 	/* Free Bigmaps */
    435 	for (i = 0; i < hashp->nmaps; i++)
    436 		if (hashp->mapp[i])
    437 			free(hashp->mapp[i]);
    438 
    439 	if (hashp->fp != -1)
    440 		(void)close(hashp->fp);
    441 
    442 	free(hashp);
    443 
    444 	if (save_errno) {
    445 		errno = save_errno;
    446 		return (ERROR);
    447 	}
    448 	return (SUCCESS);
    449 }
    450 /*
    451  * Write modified pages to disk
    452  *
    453  * Returns:
    454  *	 0 == OK
    455  *	-1 ERROR
    456  */
    457 static int
    458 hash_sync(dbp, flags)
    459 	const DB *dbp;
    460 	u_int32_t flags;
    461 {
    462 	HTAB *hashp;
    463 
    464 	if (flags != 0) {
    465 		errno = EINVAL;
    466 		return (ERROR);
    467 	}
    468 
    469 	if (!dbp)
    470 		return (ERROR);
    471 
    472 	hashp = (HTAB *)dbp->internal;
    473 	if (!hashp->save_file)
    474 		return (0);
    475 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
    476 		return (ERROR);
    477 	hashp->new_file = 0;
    478 	return (0);
    479 }
    480 
    481 /*
    482  * Returns:
    483  *	 0 == OK
    484  *	-1 indicates that errno should be set
    485  */
    486 static int
    487 flush_meta(hashp)
    488 	HTAB *hashp;
    489 {
    490 	HASHHDR *whdrp;
    491 #if BYTE_ORDER == LITTLE_ENDIAN
    492 	HASHHDR whdr;
    493 #endif
    494 	int fp, i, wsize;
    495 
    496 	if (!hashp->save_file)
    497 		return (0);
    498 	hashp->MAGIC = HASHMAGIC;
    499 	hashp->VERSION = HASHVERSION;
    500 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
    501 
    502 	fp = hashp->fp;
    503 	whdrp = &hashp->hdr;
    504 #if BYTE_ORDER == LITTLE_ENDIAN
    505 	whdrp = &whdr;
    506 	swap_header_copy(&hashp->hdr, whdrp);
    507 #endif
    508 	if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
    509 	    ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
    510 		return (-1);
    511 	else
    512 		if (wsize != sizeof(HASHHDR)) {
    513 			errno = EFTYPE;
    514 			hashp->errno = errno;
    515 			return (-1);
    516 		}
    517 	for (i = 0; i < NCACHED; i++)
    518 		if (hashp->mapp[i])
    519 			if (__put_page(hashp, (char *)hashp->mapp[i],
    520 				hashp->BITMAPS[i], 0, 1))
    521 				return (-1);
    522 	return (0);
    523 }
    524 
    525 /*******************************SEARCH ROUTINES *****************************/
    526 /*
    527  * All the access routines return
    528  *
    529  * Returns:
    530  *	 0 on SUCCESS
    531  *	 1 to indicate an external ERROR (i.e. key not found, etc)
    532  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
    533  */
    534 static int
    535 hash_get(dbp, key, data, flag)
    536 	const DB *dbp;
    537 	const DBT *key;
    538 	DBT *data;
    539 	u_int32_t flag;
    540 {
    541 	HTAB *hashp;
    542 
    543 	hashp = (HTAB *)dbp->internal;
    544 	if (flag) {
    545 		hashp->errno = errno = EINVAL;
    546 		return (ERROR);
    547 	}
    548 	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
    549 }
    550 
    551 static int
    552 hash_put(dbp, key, data, flag)
    553 	const DB *dbp;
    554 	DBT *key;
    555 	const DBT *data;
    556 	u_int32_t flag;
    557 {
    558 	HTAB *hashp;
    559 
    560 	hashp = (HTAB *)dbp->internal;
    561 	if (flag && flag != R_NOOVERWRITE) {
    562 		hashp->errno = errno = EINVAL;
    563 		return (ERROR);
    564 	}
    565 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
    566 		hashp->errno = errno = EPERM;
    567 		return (ERROR);
    568 	}
    569 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
    570 	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
    571 }
    572 
    573 static int
    574 hash_delete(dbp, key, flag)
    575 	const DB *dbp;
    576 	const DBT *key;
    577 	u_int32_t flag;		/* Ignored */
    578 {
    579 	HTAB *hashp;
    580 
    581 	hashp = (HTAB *)dbp->internal;
    582 	if (flag && flag != R_CURSOR) {
    583 		hashp->errno = errno = EINVAL;
    584 		return (ERROR);
    585 	}
    586 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
    587 		hashp->errno = errno = EPERM;
    588 		return (ERROR);
    589 	}
    590 	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
    591 }
    592 
    593 /*
    594  * Assume that hashp has been set in wrapper routine.
    595  */
    596 static int
    597 hash_access(hashp, action, key, val)
    598 	HTAB *hashp;
    599 	ACTION action;
    600 	DBT *key, *val;
    601 {
    602 	register BUFHEAD *rbufp;
    603 	BUFHEAD *bufp, *save_bufp;
    604 	register u_int16_t *bp;
    605 	register int n, ndx, off, size;
    606 	register char *kp;
    607 	u_int16_t pageno;
    608 
    609 #ifdef HASH_STATISTICS
    610 	hash_accesses++;
    611 #endif
    612 
    613 	off = hashp->BSIZE;
    614 	size = key->size;
    615 	kp = (char *)key->data;
    616 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
    617 	if (!rbufp)
    618 		return (ERROR);
    619 	save_bufp = rbufp;
    620 
    621 	/* Pin the bucket chain */
    622 	rbufp->flags |= BUF_PIN;
    623 	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
    624 		if (bp[1] >= REAL_KEY) {
    625 			/* Real key/data pair */
    626 			if (size == off - *bp &&
    627 			    memcmp(kp, rbufp->page + *bp, size) == 0)
    628 				goto found;
    629 			off = bp[1];
    630 #ifdef HASH_STATISTICS
    631 			hash_collisions++;
    632 #endif
    633 			bp += 2;
    634 			ndx += 2;
    635 		} else if (bp[1] == OVFLPAGE) {
    636 			rbufp = __get_buf(hashp, *bp, rbufp, 0);
    637 			if (!rbufp) {
    638 				save_bufp->flags &= ~BUF_PIN;
    639 				return (ERROR);
    640 			}
    641 			/* FOR LOOP INIT */
    642 			bp = (u_int16_t *)rbufp->page;
    643 			n = *bp++;
    644 			ndx = 1;
    645 			off = hashp->BSIZE;
    646 		} else if (bp[1] < REAL_KEY) {
    647 			if ((ndx =
    648 			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
    649 				goto found;
    650 			if (ndx == -2) {
    651 				bufp = rbufp;
    652 				if (!(pageno =
    653 				    __find_last_page(hashp, &bufp))) {
    654 					ndx = 0;
    655 					rbufp = bufp;
    656 					break;	/* FOR */
    657 				}
    658 				rbufp = __get_buf(hashp, pageno, bufp, 0);
    659 				if (!rbufp) {
    660 					save_bufp->flags &= ~BUF_PIN;
    661 					return (ERROR);
    662 				}
    663 				/* FOR LOOP INIT */
    664 				bp = (u_int16_t *)rbufp->page;
    665 				n = *bp++;
    666 				ndx = 1;
    667 				off = hashp->BSIZE;
    668 			} else {
    669 				save_bufp->flags &= ~BUF_PIN;
    670 				return (ERROR);
    671 			}
    672 		}
    673 
    674 	/* Not found */
    675 	switch (action) {
    676 	case HASH_PUT:
    677 	case HASH_PUTNEW:
    678 		if (__addel(hashp, rbufp, key, val)) {
    679 			save_bufp->flags &= ~BUF_PIN;
    680 			return (ERROR);
    681 		} else {
    682 			save_bufp->flags &= ~BUF_PIN;
    683 			return (SUCCESS);
    684 		}
    685 	case HASH_GET:
    686 	case HASH_DELETE:
    687 	default:
    688 		save_bufp->flags &= ~BUF_PIN;
    689 		return (ABNORMAL);
    690 	}
    691 
    692 found:
    693 	switch (action) {
    694 	case HASH_PUTNEW:
    695 		save_bufp->flags &= ~BUF_PIN;
    696 		return (ABNORMAL);
    697 	case HASH_GET:
    698 		bp = (u_int16_t *)rbufp->page;
    699 		if (bp[ndx + 1] < REAL_KEY) {
    700 			if (__big_return(hashp, rbufp, ndx, val, 0))
    701 				return (ERROR);
    702 		} else {
    703 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
    704 			val->size = bp[ndx] - bp[ndx + 1];
    705 		}
    706 		break;
    707 	case HASH_PUT:
    708 		if ((__delpair(hashp, rbufp, ndx)) ||
    709 		    (__addel(hashp, rbufp, key, val))) {
    710 			save_bufp->flags &= ~BUF_PIN;
    711 			return (ERROR);
    712 		}
    713 		break;
    714 	case HASH_DELETE:
    715 		if (__delpair(hashp, rbufp, ndx))
    716 			return (ERROR);
    717 		break;
    718 	default:
    719 		abort();
    720 	}
    721 	save_bufp->flags &= ~BUF_PIN;
    722 	return (SUCCESS);
    723 }
    724 
    725 static int
    726 hash_seq(dbp, key, data, flag)
    727 	const DB *dbp;
    728 	DBT *key, *data;
    729 	u_int32_t flag;
    730 {
    731 	register u_int32_t bucket;
    732 	register BUFHEAD *bufp;
    733 	HTAB *hashp;
    734 	u_int16_t *bp, ndx;
    735 
    736 	hashp = (HTAB *)dbp->internal;
    737 	if (flag && flag != R_FIRST && flag != R_NEXT) {
    738 		hashp->errno = errno = EINVAL;
    739 		return (ERROR);
    740 	}
    741 #ifdef HASH_STATISTICS
    742 	hash_accesses++;
    743 #endif
    744 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
    745 		hashp->cbucket = 0;
    746 		hashp->cndx = 1;
    747 		hashp->cpage = NULL;
    748 	}
    749 
    750 	for (bp = NULL; !bp || !bp[0]; ) {
    751 		if (!(bufp = hashp->cpage)) {
    752 			for (bucket = hashp->cbucket;
    753 			    bucket <= hashp->MAX_BUCKET;
    754 			    bucket++, hashp->cndx = 1) {
    755 				bufp = __get_buf(hashp, bucket, NULL, 0);
    756 				if (!bufp)
    757 					return (ERROR);
    758 				hashp->cpage = bufp;
    759 				bp = (u_int16_t *)bufp->page;
    760 				if (bp[0])
    761 					break;
    762 			}
    763 			hashp->cbucket = bucket;
    764 			if (hashp->cbucket > hashp->MAX_BUCKET) {
    765 				hashp->cbucket = -1;
    766 				return (ABNORMAL);
    767 			}
    768 		} else
    769 			bp = (u_int16_t *)hashp->cpage->page;
    770 
    771 #ifdef DEBUG
    772 		assert(bp);
    773 		assert(bufp);
    774 #endif
    775 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
    776 			bufp = hashp->cpage =
    777 			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
    778 			if (!bufp)
    779 				return (ERROR);
    780 			bp = (u_int16_t *)(bufp->page);
    781 			hashp->cndx = 1;
    782 		}
    783 		if (!bp[0]) {
    784 			hashp->cpage = NULL;
    785 			++hashp->cbucket;
    786 		}
    787 	}
    788 	ndx = hashp->cndx;
    789 	if (bp[ndx + 1] < REAL_KEY) {
    790 		if (__big_keydata(hashp, bufp, key, data, 1))
    791 			return (ERROR);
    792 	} else {
    793 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
    794 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
    795 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
    796 		data->size = bp[ndx] - bp[ndx + 1];
    797 		ndx += 2;
    798 		if (ndx > bp[0]) {
    799 			hashp->cpage = NULL;
    800 			hashp->cbucket++;
    801 			hashp->cndx = 1;
    802 		} else
    803 			hashp->cndx = ndx;
    804 	}
    805 	return (SUCCESS);
    806 }
    807 
    808 /********************************* UTILITIES ************************/
    809 
    810 /*
    811  * Returns:
    812  *	 0 ==> OK
    813  *	-1 ==> Error
    814  */
    815 extern int
    816 __expand_table(hashp)
    817 	HTAB *hashp;
    818 {
    819 	u_int32_t old_bucket, new_bucket;
    820 	int dirsize, new_segnum, spare_ndx;
    821 
    822 #ifdef HASH_STATISTICS
    823 	hash_expansions++;
    824 #endif
    825 	new_bucket = ++hashp->MAX_BUCKET;
    826 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
    827 
    828 	new_segnum = new_bucket >> hashp->SSHIFT;
    829 
    830 	/* Check if we need a new segment */
    831 	if (new_segnum >= hashp->nsegs) {
    832 		/* Check if we need to expand directory */
    833 		if (new_segnum >= hashp->DSIZE) {
    834 			/* Reallocate directory */
    835 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
    836 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
    837 				return (-1);
    838 			hashp->DSIZE = dirsize << 1;
    839 		}
    840 		if ((hashp->dir[new_segnum] =
    841 		    (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
    842 			return (-1);
    843 		hashp->exsegs++;
    844 		hashp->nsegs++;
    845 	}
    846 	/*
    847 	 * If the split point is increasing (MAX_BUCKET's log base 2
    848 	 * * increases), we need to copy the current contents of the spare
    849 	 * split bucket to the next bucket.
    850 	 */
    851 	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
    852 	if (spare_ndx > hashp->OVFL_POINT) {
    853 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
    854 		hashp->OVFL_POINT = spare_ndx;
    855 	}
    856 
    857 	if (new_bucket > hashp->HIGH_MASK) {
    858 		/* Starting a new doubling */
    859 		hashp->LOW_MASK = hashp->HIGH_MASK;
    860 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
    861 	}
    862 	/* Relocate records to the new bucket */
    863 	return (__split_page(hashp, old_bucket, new_bucket));
    864 }
    865 
    866 /*
    867  * If realloc guarantees that the pointer is not destroyed if the realloc
    868  * fails, then this routine can go away.
    869  */
    870 static void *
    871 hash_realloc(p_ptr, oldsize, newsize)
    872 	SEGMENT **p_ptr;
    873 	int oldsize, newsize;
    874 {
    875 	register void *p;
    876 
    877 	if (p = malloc(newsize)) {
    878 		memmove(p, *p_ptr, oldsize);
    879 		memset((char *)p + oldsize, 0, newsize - oldsize);
    880 		free(*p_ptr);
    881 		*p_ptr = p;
    882 	}
    883 	return (p);
    884 }
    885 
    886 extern u_int32_t
    887 __call_hash(hashp, k, len)
    888 	HTAB *hashp;
    889 	char *k;
    890 	int len;
    891 {
    892 	int n, bucket;
    893 
    894 	n = hashp->hash(k, len);
    895 	bucket = n & hashp->HIGH_MASK;
    896 	if (bucket > hashp->MAX_BUCKET)
    897 		bucket = bucket & hashp->LOW_MASK;
    898 	return (bucket);
    899 }
    900 
    901 /*
    902  * Allocate segment table.  On error, destroy the table and set errno.
    903  *
    904  * Returns 0 on success
    905  */
    906 static int
    907 alloc_segs(hashp, nsegs)
    908 	HTAB *hashp;
    909 	int nsegs;
    910 {
    911 	register int i;
    912 	register SEGMENT store;
    913 
    914 	int save_errno;
    915 
    916 	if ((hashp->dir =
    917 	    (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
    918 		save_errno = errno;
    919 		(void)hdestroy(hashp);
    920 		errno = save_errno;
    921 		return (-1);
    922 	}
    923 	/* Allocate segments */
    924 	if ((store =
    925 	    (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
    926 		save_errno = errno;
    927 		(void)hdestroy(hashp);
    928 		errno = save_errno;
    929 		return (-1);
    930 	}
    931 	for (i = 0; i < nsegs; i++, hashp->nsegs++)
    932 		hashp->dir[i] = &store[i << hashp->SSHIFT];
    933 	return (0);
    934 }
    935 
    936 #if BYTE_ORDER == LITTLE_ENDIAN
    937 /*
    938  * Hashp->hdr needs to be byteswapped.
    939  */
    940 static void
    941 swap_header_copy(srcp, destp)
    942 	HASHHDR *srcp, *destp;
    943 {
    944 	int i;
    945 
    946 	P_32_COPY(srcp->magic, destp->magic);
    947 	P_32_COPY(srcp->version, destp->version);
    948 	P_32_COPY(srcp->lorder, destp->lorder);
    949 	P_32_COPY(srcp->bsize, destp->bsize);
    950 	P_32_COPY(srcp->bshift, destp->bshift);
    951 	P_32_COPY(srcp->dsize, destp->dsize);
    952 	P_32_COPY(srcp->ssize, destp->ssize);
    953 	P_32_COPY(srcp->sshift, destp->sshift);
    954 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
    955 	P_32_COPY(srcp->last_freed, destp->last_freed);
    956 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
    957 	P_32_COPY(srcp->high_mask, destp->high_mask);
    958 	P_32_COPY(srcp->low_mask, destp->low_mask);
    959 	P_32_COPY(srcp->ffactor, destp->ffactor);
    960 	P_32_COPY(srcp->nkeys, destp->nkeys);
    961 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
    962 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
    963 	for (i = 0; i < NCACHED; i++) {
    964 		P_32_COPY(srcp->spares[i], destp->spares[i]);
    965 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
    966 	}
    967 }
    968 
    969 static void
    970 swap_header(hashp)
    971 	HTAB *hashp;
    972 {
    973 	HASHHDR *hdrp;
    974 	int i;
    975 
    976 	hdrp = &hashp->hdr;
    977 
    978 	M_32_SWAP(hdrp->magic);
    979 	M_32_SWAP(hdrp->version);
    980 	M_32_SWAP(hdrp->lorder);
    981 	M_32_SWAP(hdrp->bsize);
    982 	M_32_SWAP(hdrp->bshift);
    983 	M_32_SWAP(hdrp->dsize);
    984 	M_32_SWAP(hdrp->ssize);
    985 	M_32_SWAP(hdrp->sshift);
    986 	M_32_SWAP(hdrp->ovfl_point);
    987 	M_32_SWAP(hdrp->last_freed);
    988 	M_32_SWAP(hdrp->max_bucket);
    989 	M_32_SWAP(hdrp->high_mask);
    990 	M_32_SWAP(hdrp->low_mask);
    991 	M_32_SWAP(hdrp->ffactor);
    992 	M_32_SWAP(hdrp->nkeys);
    993 	M_32_SWAP(hdrp->hdrpages);
    994 	M_32_SWAP(hdrp->h_charkey);
    995 	for (i = 0; i < NCACHED; i++) {
    996 		M_32_SWAP(hdrp->spares[i]);
    997 		M_16_SWAP(hdrp->bitmaps[i]);
    998 	}
    999 }
   1000 #endif
   1001