Home | History | Annotate | Line # | Download | only in hash
hash.c revision 1.26
      1 /*	$NetBSD: hash.c,v 1.26 2006/03/19 03:36:28 rtr 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 #include <sys/cdefs.h>
     36 #if defined(LIBC_SCCS) && !defined(lint)
     37 #if 0
     38 static char sccsid[] = "@(#)hash.c	8.9 (Berkeley) 6/16/94";
     39 #else
     40 __RCSID("$NetBSD: hash.c,v 1.26 2006/03/19 03:36:28 rtr Exp $");
     41 #endif
     42 #endif /* LIBC_SCCS and not lint */
     43 
     44 #include "namespace.h"
     45 #include <sys/param.h>
     46 #include <sys/stat.h>
     47 
     48 #include <errno.h>
     49 #include <fcntl.h>
     50 #include <stdio.h>
     51 #include <stdlib.h>
     52 #include <string.h>
     53 #include <unistd.h>
     54 #ifdef DEBUG
     55 #include <assert.h>
     56 #endif
     57 
     58 #include <db.h>
     59 #include "hash.h"
     60 #include "page.h"
     61 #include "extern.h"
     62 
     63 static int   alloc_segs __P((HTAB *, int));
     64 static int   flush_meta __P((HTAB *));
     65 static int   hash_access __P((HTAB *, ACTION, DBT *, DBT *));
     66 static int   hash_close __P((DB *));
     67 static int   hash_delete __P((const DB *, const DBT *, u_int32_t));
     68 static int   hash_fd __P((const DB *));
     69 static int   hash_get __P((const DB *, const DBT *, DBT *, u_int32_t));
     70 static int   hash_put __P((const DB *, DBT *, const DBT *, u_int32_t));
     71 static void *hash_realloc __P((SEGMENT **, int, int));
     72 static int   hash_seq __P((const DB *, DBT *, DBT *, u_int32_t));
     73 static int   hash_sync __P((const DB *, u_int32_t));
     74 static int   hdestroy __P((HTAB *));
     75 static HTAB *init_hash __P((HTAB *, const char *, const HASHINFO *));
     76 static int   init_htab __P((HTAB *, size_t));
     77 #if BYTE_ORDER == LITTLE_ENDIAN
     78 static void  swap_header __P((HTAB *));
     79 static void  swap_header_copy __P((HASHHDR *, HASHHDR *));
     80 #endif
     81 
     82 /* Fast arithmetic, relying on powers of 2, */
     83 #define MOD(x, y)		((x) & ((y) - 1))
     84 
     85 #define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
     86 
     87 /* Return values */
     88 #define	SUCCESS	 (0)
     89 #define	ERROR	(-1)
     90 #define	ABNORMAL (1)
     91 
     92 #ifdef HASH_STATISTICS
     93 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
     94 #endif
     95 
     96 /************************** INTERFACE ROUTINES ***************************/
     97 /* OPEN/CLOSE */
     98 
     99 /* ARGSUSED */
    100 extern DB *
    101 __hash_open(file, flags, mode, info, dflags)
    102 	const char *file;
    103 	int flags;
    104 	mode_t mode;
    105 	const HASHINFO *info;	/* Special directives for create */
    106 	int dflags;
    107 {
    108 	HTAB *hashp;
    109 	struct stat statbuf;
    110 	DB *dbp;
    111 	int bpages, hdrsize, new_table, nsegs, save_errno;
    112 
    113 	if ((flags & O_ACCMODE) == O_WRONLY) {
    114 		errno = EINVAL;
    115 		return (NULL);
    116 	}
    117 
    118 	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
    119 		return (NULL);
    120 	hashp->fp = -1;
    121 
    122 	/*
    123 	 * Even if user wants write only, we need to be able to read
    124 	 * the actual file, so we need to open it read/write. But, the
    125 	 * field in the hashp structure needs to be accurate so that
    126 	 * we can check accesses.
    127 	 */
    128 	hashp->flags = flags;
    129 
    130 	new_table = 0;
    131 	if (!file || (flags & O_TRUNC) ||
    132 	    (stat(file, &statbuf) && (errno == ENOENT))) {
    133 		if (errno == ENOENT)
    134 			errno = 0; /* Just in case someone looks at errno */
    135 		new_table = 1;
    136 	}
    137 	if (file) {
    138 		if ((hashp->fp = open(file, flags, mode)) == -1)
    139 			RETURN_ERROR(errno, error0);
    140 		if (fcntl(hashp->fp, F_SETFD, FD_CLOEXEC) == -1)
    141 			RETURN_ERROR(errno, error1);
    142 		if (fstat(hashp->fp, &statbuf) == -1)
    143 			RETURN_ERROR(errno, error1);
    144 		new_table |= statbuf.st_size == 0;
    145 	}
    146 	if (new_table) {
    147 		if (!(hashp = init_hash(hashp, file, info)))
    148 			RETURN_ERROR(errno, error1);
    149 	} else {
    150 		/* Table already exists */
    151 		if (info && info->hash)
    152 			hashp->hash = info->hash;
    153 		else
    154 			hashp->hash = __default_hash;
    155 
    156 		hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
    157 #if BYTE_ORDER == LITTLE_ENDIAN
    158 		swap_header(hashp);
    159 #endif
    160 		if (hdrsize == -1)
    161 			RETURN_ERROR(errno, error1);
    162 		if (hdrsize != sizeof(HASHHDR))
    163 			RETURN_ERROR(EFTYPE, error1);
    164 		/* Verify file type, versions and hash function */
    165 		if (hashp->MAGIC != HASHMAGIC)
    166 			RETURN_ERROR(EFTYPE, error1);
    167 #define	OLDHASHVERSION	1
    168 		if (hashp->VERSION != HASHVERSION &&
    169 		    hashp->VERSION != OLDHASHVERSION)
    170 			RETURN_ERROR(EFTYPE, error1);
    171 		if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
    172 			RETURN_ERROR(EFTYPE, error1);
    173 		/*
    174 		 * Figure out how many segments we need.  Max_Bucket is the
    175 		 * maximum bucket number, so the number of buckets is
    176 		 * max_bucket + 1.
    177 		 */
    178 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
    179 			 hashp->SGSIZE;
    180 		hashp->nsegs = 0;
    181 		if (alloc_segs(hashp, nsegs))
    182 			/*
    183 			 * If alloc_segs fails, table will have been destroyed
    184 			 * and errno will have been set.
    185 			 */
    186 			return (NULL);
    187 		/* Read in bitmaps */
    188 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
    189 		    (unsigned int)(hashp->BSIZE << BYTE_SHIFT) - 1) >>
    190 		    (hashp->BSHIFT + BYTE_SHIFT);
    191 
    192 		hashp->nmaps = bpages;
    193 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
    194 	}
    195 
    196 	/* Initialize Buffer Manager */
    197 	if (info && info->cachesize)
    198 		__buf_init(hashp, info->cachesize);
    199 	else
    200 		__buf_init(hashp, DEF_BUFSIZE);
    201 
    202 	hashp->new_file = new_table;
    203 	hashp->save_file = file && (hashp->flags & O_RDWR);
    204 	hashp->cbucket = -1;
    205 	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
    206 		save_errno = errno;
    207 		hdestroy(hashp);
    208 		errno = save_errno;
    209 		return (NULL);
    210 	}
    211 	dbp->internal = hashp;
    212 	dbp->close = hash_close;
    213 	dbp->del = hash_delete;
    214 	dbp->fd = hash_fd;
    215 	dbp->get = hash_get;
    216 	dbp->put = hash_put;
    217 	dbp->seq = hash_seq;
    218 	dbp->sync = hash_sync;
    219 	dbp->type = DB_HASH;
    220 
    221 #ifdef DEBUG
    222 	(void)fprintf(stderr,
    223 "%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",
    224 	    "init_htab:",
    225 	    "TABLE POINTER   ", hashp,
    226 	    "BUCKET SIZE     ", hashp->BSIZE,
    227 	    "BUCKET SHIFT    ", hashp->BSHIFT,
    228 	    "DIRECTORY SIZE  ", hashp->DSIZE,
    229 	    "SEGMENT SIZE    ", hashp->SGSIZE,
    230 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
    231 	    "FILL FACTOR     ", hashp->FFACTOR,
    232 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
    233 	    "OVFL POINT	     ", hashp->OVFL_POINT,
    234 	    "LAST FREED      ", hashp->LAST_FREED,
    235 	    "HIGH MASK       ", hashp->HIGH_MASK,
    236 	    "LOW  MASK       ", hashp->LOW_MASK,
    237 	    "NSEGS           ", hashp->nsegs,
    238 	    "NKEYS           ", hashp->NKEYS);
    239 #endif
    240 #ifdef HASH_STATISTICS
    241 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
    242 #endif
    243 	return (dbp);
    244 
    245 error1:
    246 	if (hashp != NULL)
    247 		(void)close(hashp->fp);
    248 
    249 error0:
    250 	free(hashp);
    251 	errno = save_errno;
    252 	return (NULL);
    253 }
    254 
    255 static int
    256 hash_close(dbp)
    257 	DB *dbp;
    258 {
    259 	HTAB *hashp;
    260 	int retval;
    261 
    262 	if (!dbp)
    263 		return (ERROR);
    264 
    265 	hashp = (HTAB *)dbp->internal;
    266 	retval = hdestroy(hashp);
    267 	free(dbp);
    268 	return (retval);
    269 }
    270 
    271 static int
    272 hash_fd(dbp)
    273 	const DB *dbp;
    274 {
    275 	HTAB *hashp;
    276 
    277 	if (!dbp)
    278 		return (ERROR);
    279 
    280 	hashp = (HTAB *)dbp->internal;
    281 	if (hashp->fp == -1) {
    282 		errno = ENOENT;
    283 		return (-1);
    284 	}
    285 	return (hashp->fp);
    286 }
    287 
    288 /************************** LOCAL CREATION ROUTINES **********************/
    289 static HTAB *
    290 init_hash(hashp, file, info)
    291 	HTAB *hashp;
    292 	const char *file;
    293 	const HASHINFO *info;
    294 {
    295 	struct stat statbuf;
    296 	int nelem;
    297 
    298 	nelem = 1;
    299 	hashp->NKEYS = 0;
    300 	hashp->LORDER = BYTE_ORDER;
    301 	hashp->BSIZE = DEF_BUCKET_SIZE;
    302 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
    303 	hashp->SGSIZE = DEF_SEGSIZE;
    304 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
    305 	hashp->DSIZE = DEF_DIRSIZE;
    306 	hashp->FFACTOR = DEF_FFACTOR;
    307 	hashp->hash = __default_hash;
    308 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
    309 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
    310 
    311 	/* Fix bucket size to be optimal for file system */
    312 	if (file != NULL) {
    313 		if (stat(file, &statbuf))
    314 			return (NULL);
    315 		hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
    316 		hashp->BSHIFT = __log2((u_int32_t)hashp->BSIZE);
    317 	}
    318 
    319 	if (info) {
    320 		if (info->bsize) {
    321 			/* Round pagesize up to power of 2 */
    322 			hashp->BSHIFT = __log2(info->bsize);
    323 			hashp->BSIZE = 1 << hashp->BSHIFT;
    324 			if (hashp->BSIZE > MAX_BSIZE) {
    325 				errno = EINVAL;
    326 				return (NULL);
    327 			}
    328 		}
    329 		if (info->ffactor)
    330 			hashp->FFACTOR = info->ffactor;
    331 		if (info->hash)
    332 			hashp->hash = info->hash;
    333 		if (info->nelem)
    334 			nelem = info->nelem;
    335 		if (info->lorder) {
    336 			if (info->lorder != BIG_ENDIAN &&
    337 			    info->lorder != LITTLE_ENDIAN) {
    338 				errno = EINVAL;
    339 				return (NULL);
    340 			}
    341 			hashp->LORDER = info->lorder;
    342 		}
    343 	}
    344 	/* init_htab should destroy the table and set errno if it fails */
    345 	if (init_htab(hashp, (size_t)nelem))
    346 		return (NULL);
    347 	else
    348 		return (hashp);
    349 }
    350 /*
    351  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
    352  * the table and set errno, so we just pass the error information along.
    353  *
    354  * Returns 0 on No Error
    355  */
    356 static int
    357 init_htab(hashp, nelem)
    358 	HTAB *hashp;
    359 	size_t nelem;
    360 {
    361 	register int nbuckets;
    362 	u_int32_t nsegs;
    363 	int l2;
    364 
    365 	/*
    366 	 * Divide number of elements by the fill factor and determine a
    367 	 * desired number of buckets.  Allocate space for the next greater
    368 	 * power of two number of buckets.
    369 	 */
    370 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
    371 
    372 	l2 = __log2(MAX(nelem, 2));
    373 	nbuckets = 1 << l2;
    374 
    375 	hashp->SPARES[l2] = l2 + 1;
    376 	hashp->SPARES[l2 + 1] = l2 + 1;
    377 	hashp->OVFL_POINT = l2;
    378 	hashp->LAST_FREED = 2;
    379 
    380 	/* First bitmap page is at: splitpoint l2 page offset 1 */
    381 	if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0))
    382 		return (-1);
    383 
    384 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
    385 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
    386 	/* LINTED constant in conditional context */
    387 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
    388 	    hashp->BSHIFT) + 1;
    389 
    390 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
    391 	nsegs = 1 << __log2(nsegs);
    392 
    393 	if (nsegs > hashp->DSIZE)
    394 		hashp->DSIZE = nsegs;
    395 	return (alloc_segs(hashp, (int)nsegs));
    396 }
    397 
    398 /********************** DESTROY/CLOSE ROUTINES ************************/
    399 
    400 /*
    401  * Flushes any changes to the file if necessary and destroys the hashp
    402  * structure, freeing all allocated space.
    403  */
    404 static int
    405 hdestroy(hashp)
    406 	HTAB *hashp;
    407 {
    408 	int i, save_errno;
    409 
    410 	save_errno = 0;
    411 
    412 #ifdef HASH_STATISTICS
    413 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
    414 	    hash_accesses, hash_collisions);
    415 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
    416 	    hash_expansions);
    417 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
    418 	    hash_overflows);
    419 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
    420 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
    421 
    422 	for (i = 0; i < NCACHED; i++)
    423 		(void)fprintf(stderr,
    424 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
    425 #endif
    426 	/*
    427 	 * Call on buffer manager to free buffers, and if required,
    428 	 * write them to disk.
    429 	 */
    430 	if (__buf_free(hashp, 1, hashp->save_file))
    431 		save_errno = errno;
    432 	if (hashp->dir) {
    433 		free(*hashp->dir);	/* Free initial segments */
    434 		/* Free extra segments */
    435 		while (hashp->exsegs--)
    436 			free(hashp->dir[--hashp->nsegs]);
    437 		free(hashp->dir);
    438 	}
    439 	if (flush_meta(hashp) && !save_errno)
    440 		save_errno = errno;
    441 	/* Free Bigmaps */
    442 	for (i = 0; i < hashp->nmaps; i++)
    443 		if (hashp->mapp[i])
    444 			free(hashp->mapp[i]);
    445 
    446 	if (hashp->fp != -1)
    447 		(void)close(hashp->fp);
    448 
    449 	free(hashp);
    450 
    451 	if (save_errno) {
    452 		errno = save_errno;
    453 		return (ERROR);
    454 	}
    455 	return (SUCCESS);
    456 }
    457 /*
    458  * Write modified pages to disk
    459  *
    460  * Returns:
    461  *	 0 == OK
    462  *	-1 ERROR
    463  */
    464 static int
    465 hash_sync(dbp, flags)
    466 	const DB *dbp;
    467 	u_int32_t flags;
    468 {
    469 	HTAB *hashp;
    470 
    471 	if (flags != 0) {
    472 		errno = EINVAL;
    473 		return (ERROR);
    474 	}
    475 
    476 	if (!dbp)
    477 		return (ERROR);
    478 
    479 	hashp = (HTAB *)dbp->internal;
    480 	if (!hashp->save_file)
    481 		return (0);
    482 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
    483 		return (ERROR);
    484 	hashp->new_file = 0;
    485 	return (0);
    486 }
    487 
    488 /*
    489  * Returns:
    490  *	 0 == OK
    491  *	-1 indicates that errno should be set
    492  */
    493 static int
    494 flush_meta(hashp)
    495 	HTAB *hashp;
    496 {
    497 	HASHHDR *whdrp;
    498 #if BYTE_ORDER == LITTLE_ENDIAN
    499 	HASHHDR whdr;
    500 #endif
    501 	int fp, i, wsize;
    502 
    503 	if (!hashp->save_file)
    504 		return (0);
    505 	hashp->MAGIC = HASHMAGIC;
    506 	hashp->VERSION = HASHVERSION;
    507 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
    508 
    509 	fp = hashp->fp;
    510 	whdrp = &hashp->hdr;
    511 #if BYTE_ORDER == LITTLE_ENDIAN
    512 	whdrp = &whdr;
    513 	swap_header_copy(&hashp->hdr, whdrp);
    514 #endif
    515 	if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
    516 		return (-1);
    517 	else
    518 		if (wsize != sizeof(HASHHDR)) {
    519 			errno = EFTYPE;
    520 			hashp->err = errno;
    521 			return (-1);
    522 		}
    523 	for (i = 0; i < NCACHED; i++)
    524 		if (hashp->mapp[i])
    525 			if (__put_page(hashp, (char *)(void *)hashp->mapp[i],
    526 				(u_int)hashp->BITMAPS[i], 0, 1))
    527 				return (-1);
    528 	return (0);
    529 }
    530 
    531 /*******************************SEARCH ROUTINES *****************************/
    532 /*
    533  * All the access routines return
    534  *
    535  * Returns:
    536  *	 0 on SUCCESS
    537  *	 1 to indicate an external ERROR (i.e. key not found, etc)
    538  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
    539  */
    540 static int
    541 hash_get(dbp, key, data, flag)
    542 	const DB *dbp;
    543 	const DBT *key;
    544 	DBT *data;
    545 	u_int32_t flag;
    546 {
    547 	HTAB *hashp;
    548 
    549 	hashp = (HTAB *)dbp->internal;
    550 	if (flag) {
    551 		hashp->err = errno = EINVAL;
    552 		return (ERROR);
    553 	}
    554 	return (hash_access(hashp, HASH_GET, __UNCONST(key), data));
    555 }
    556 
    557 static int
    558 hash_put(dbp, key, data, flag)
    559 	const DB *dbp;
    560 	DBT *key;
    561 	const DBT *data;
    562 	u_int32_t flag;
    563 {
    564 	HTAB *hashp;
    565 
    566 	hashp = (HTAB *)dbp->internal;
    567 	if (flag && flag != R_NOOVERWRITE) {
    568 		hashp->err = errno = EINVAL;
    569 		return (ERROR);
    570 	}
    571 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
    572 		hashp->err = errno = EPERM;
    573 		return (ERROR);
    574 	}
    575 	/* LINTED const castaway */
    576 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
    577 	    HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data)));
    578 }
    579 
    580 static int
    581 hash_delete(dbp, key, flag)
    582 	const DB *dbp;
    583 	const DBT *key;
    584 	u_int32_t flag;		/* Ignored */
    585 {
    586 	HTAB *hashp;
    587 
    588 	hashp = (HTAB *)dbp->internal;
    589 	if (flag && flag != R_CURSOR) {
    590 		hashp->err = errno = EINVAL;
    591 		return (ERROR);
    592 	}
    593 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
    594 		hashp->err = errno = EPERM;
    595 		return (ERROR);
    596 	}
    597 	return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL);
    598 }
    599 
    600 /*
    601  * Assume that hashp has been set in wrapper routine.
    602  */
    603 static int
    604 hash_access(hashp, action, key, val)
    605 	HTAB *hashp;
    606 	ACTION action;
    607 	DBT *key, *val;
    608 {
    609 	register BUFHEAD *rbufp;
    610 	BUFHEAD *bufp, *save_bufp;
    611 	register u_int16_t *bp;
    612 	register int n, ndx, off;
    613 	size_t size;
    614 	register char *kp;
    615 	u_int16_t pageno;
    616 
    617 #ifdef HASH_STATISTICS
    618 	hash_accesses++;
    619 #endif
    620 
    621 	off = hashp->BSIZE;
    622 	size = key->size;
    623 	kp = (char *)key->data;
    624 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0);
    625 	if (!rbufp)
    626 		return (ERROR);
    627 	save_bufp = rbufp;
    628 
    629 	/* Pin the bucket chain */
    630 	rbufp->flags |= BUF_PIN;
    631 	for (bp = (u_int16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
    632 		if (bp[1] >= REAL_KEY) {
    633 			/* Real key/data pair */
    634 			if (size == off - *bp &&
    635 			    memcmp(kp, rbufp->page + *bp, size) == 0)
    636 				goto found;
    637 			off = bp[1];
    638 #ifdef HASH_STATISTICS
    639 			hash_collisions++;
    640 #endif
    641 			bp += 2;
    642 			ndx += 2;
    643 		} else if (bp[1] == OVFLPAGE) {
    644 			rbufp = __get_buf(hashp, (u_int32_t)*bp, rbufp, 0);
    645 			if (!rbufp) {
    646 				save_bufp->flags &= ~BUF_PIN;
    647 				return (ERROR);
    648 			}
    649 			/* FOR LOOP INIT */
    650 			bp = (u_int16_t *)(void *)rbufp->page;
    651 			n = *bp++;
    652 			ndx = 1;
    653 			off = hashp->BSIZE;
    654 		} else if (bp[1] < REAL_KEY) {
    655 			if ((ndx =
    656 			    __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0)
    657 				goto found;
    658 			if (ndx == -2) {
    659 				bufp = rbufp;
    660 				if (!(pageno =
    661 				    __find_last_page(hashp, &bufp))) {
    662 					ndx = 0;
    663 					rbufp = bufp;
    664 					break;	/* FOR */
    665 				}
    666 				rbufp = __get_buf(hashp, (u_int32_t)pageno,
    667 				    bufp, 0);
    668 				if (!rbufp) {
    669 					save_bufp->flags &= ~BUF_PIN;
    670 					return (ERROR);
    671 				}
    672 				/* FOR LOOP INIT */
    673 				bp = (u_int16_t *)(void *)rbufp->page;
    674 				n = *bp++;
    675 				ndx = 1;
    676 				off = hashp->BSIZE;
    677 			} else {
    678 				save_bufp->flags &= ~BUF_PIN;
    679 				return (ERROR);
    680 			}
    681 		}
    682 
    683 	/* Not found */
    684 	switch (action) {
    685 	case HASH_PUT:
    686 	case HASH_PUTNEW:
    687 		if (__addel(hashp, rbufp, key, val)) {
    688 			save_bufp->flags &= ~BUF_PIN;
    689 			return (ERROR);
    690 		} else {
    691 			save_bufp->flags &= ~BUF_PIN;
    692 			return (SUCCESS);
    693 		}
    694 	case HASH_GET:
    695 	case HASH_DELETE:
    696 	default:
    697 		save_bufp->flags &= ~BUF_PIN;
    698 		return (ABNORMAL);
    699 	}
    700 
    701 found:
    702 	switch (action) {
    703 	case HASH_PUTNEW:
    704 		save_bufp->flags &= ~BUF_PIN;
    705 		return (ABNORMAL);
    706 	case HASH_GET:
    707 		bp = (u_int16_t *)(void *)rbufp->page;
    708 		if (bp[ndx + 1] < REAL_KEY) {
    709 			if (__big_return(hashp, rbufp, ndx, val, 0))
    710 				return (ERROR);
    711 		} else {
    712 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
    713 			val->size = bp[ndx] - bp[ndx + 1];
    714 		}
    715 		break;
    716 	case HASH_PUT:
    717 		if ((__delpair(hashp, rbufp, ndx)) ||
    718 		    (__addel(hashp, rbufp, key, val))) {
    719 			save_bufp->flags &= ~BUF_PIN;
    720 			return (ERROR);
    721 		}
    722 		break;
    723 	case HASH_DELETE:
    724 		if (__delpair(hashp, rbufp, ndx))
    725 			return (ERROR);
    726 		break;
    727 	default:
    728 		abort();
    729 	}
    730 	save_bufp->flags &= ~BUF_PIN;
    731 	return (SUCCESS);
    732 }
    733 
    734 static int
    735 hash_seq(dbp, key, data, flag)
    736 	const DB *dbp;
    737 	DBT *key, *data;
    738 	u_int32_t flag;
    739 {
    740 	register u_int32_t bucket;
    741 	register BUFHEAD *bufp = NULL; /* XXX: gcc */
    742 	HTAB *hashp;
    743 	u_int16_t *bp, ndx;
    744 
    745 	hashp = (HTAB *)dbp->internal;
    746 	if (flag && flag != R_FIRST && flag != R_NEXT) {
    747 		hashp->err = errno = EINVAL;
    748 		return (ERROR);
    749 	}
    750 #ifdef HASH_STATISTICS
    751 	hash_accesses++;
    752 #endif
    753 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
    754 		hashp->cbucket = 0;
    755 		hashp->cndx = 1;
    756 		hashp->cpage = NULL;
    757 	}
    758 
    759 	for (bp = NULL; !bp || !bp[0]; ) {
    760 		if (!(bufp = hashp->cpage)) {
    761 			for (bucket = hashp->cbucket;
    762 			    bucket <= hashp->MAX_BUCKET;
    763 			    bucket++, hashp->cndx = 1) {
    764 				bufp = __get_buf(hashp, bucket, NULL, 0);
    765 				if (!bufp)
    766 					return (ERROR);
    767 				hashp->cpage = bufp;
    768 				bp = (u_int16_t *)(void *)bufp->page;
    769 				if (bp[0])
    770 					break;
    771 			}
    772 			hashp->cbucket = bucket;
    773 			if (hashp->cbucket > hashp->MAX_BUCKET) {
    774 				hashp->cbucket = -1;
    775 				return (ABNORMAL);
    776 			}
    777 		} else
    778 			bp = (u_int16_t *)(void *)hashp->cpage->page;
    779 
    780 #ifdef DEBUG
    781 		assert(bp);
    782 		assert(bufp);
    783 #endif
    784 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
    785 			bufp = hashp->cpage =
    786 			    __get_buf(hashp, (u_int32_t)bp[hashp->cndx], bufp,
    787 				0);
    788 			if (!bufp)
    789 				return (ERROR);
    790 			bp = (u_int16_t *)(void *)(bufp->page);
    791 			hashp->cndx = 1;
    792 		}
    793 		if (!bp[0]) {
    794 			hashp->cpage = NULL;
    795 			++hashp->cbucket;
    796 		}
    797 	}
    798 	ndx = hashp->cndx;
    799 	if (bp[ndx + 1] < REAL_KEY) {
    800 		if (__big_keydata(hashp, bufp, key, data, 1))
    801 			return (ERROR);
    802 	} else {
    803 		if (hashp->cpage == NULL)
    804 			return (ERROR);
    805 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
    806 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
    807 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
    808 		data->size = bp[ndx] - bp[ndx + 1];
    809 		ndx += 2;
    810 		if (ndx > bp[0]) {
    811 			hashp->cpage = NULL;
    812 			hashp->cbucket++;
    813 			hashp->cndx = 1;
    814 		} else
    815 			hashp->cndx = ndx;
    816 	}
    817 	return (SUCCESS);
    818 }
    819 
    820 /********************************* UTILITIES ************************/
    821 
    822 /*
    823  * Returns:
    824  *	 0 ==> OK
    825  *	-1 ==> Error
    826  */
    827 extern int
    828 __expand_table(hashp)
    829 	HTAB *hashp;
    830 {
    831 	u_int32_t old_bucket, new_bucket;
    832 	int dirsize, new_segnum, spare_ndx;
    833 
    834 #ifdef HASH_STATISTICS
    835 	hash_expansions++;
    836 #endif
    837 	new_bucket = ++hashp->MAX_BUCKET;
    838 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
    839 
    840 	new_segnum = new_bucket >> hashp->SSHIFT;
    841 
    842 	/* Check if we need a new segment */
    843 	if (new_segnum >= hashp->nsegs) {
    844 		/* Check if we need to expand directory */
    845 		if (new_segnum >= hashp->DSIZE) {
    846 			/* Reallocate directory */
    847 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
    848 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
    849 				return (-1);
    850 			hashp->DSIZE = dirsize << 1;
    851 		}
    852 		if ((hashp->dir[new_segnum] =
    853 		    (SEGMENT)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((u_int32_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 > 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(p_ptr, oldsize, newsize)
    884 	SEGMENT **p_ptr;
    885 	int oldsize, newsize;
    886 {
    887 	register void *p;
    888 
    889 	if ((p = malloc((size_t)newsize)) != NULL) {
    890 		memmove(p, *p_ptr, (size_t)oldsize);
    891 		memset((char *)p + oldsize, 0, (size_t)(newsize - oldsize));
    892 		free(*p_ptr);
    893 		*p_ptr = p;
    894 	}
    895 	return (p);
    896 }
    897 
    898 extern u_int32_t
    899 __call_hash(hashp, k, len)
    900 	HTAB *hashp;
    901 	char *k;
    902 	int len;
    903 {
    904 	int n, bucket;
    905 
    906 	n = hashp->hash(k, (size_t)len);
    907 	bucket = n & hashp->HIGH_MASK;
    908 	if (bucket > hashp->MAX_BUCKET)
    909 		bucket = bucket & hashp->LOW_MASK;
    910 	return (bucket);
    911 }
    912 
    913 /*
    914  * Allocate segment table.  On error, destroy the table and set errno.
    915  *
    916  * Returns 0 on success
    917  */
    918 static int
    919 alloc_segs(hashp, nsegs)
    920 	HTAB *hashp;
    921 	int nsegs;
    922 {
    923 	register int i;
    924 	register SEGMENT store;
    925 
    926 	int save_errno;
    927 
    928 	if ((hashp->dir =
    929 	    (SEGMENT *)calloc((size_t)hashp->DSIZE,
    930 	    sizeof(SEGMENT *))) == NULL) {
    931 		save_errno = errno;
    932 		(void)hdestroy(hashp);
    933 		errno = save_errno;
    934 		return (-1);
    935 	}
    936 	hashp->nsegs = nsegs;
    937 	if (nsegs == 0)
    938 		return 0;
    939 	/* Allocate segments */
    940 	if ((store = (SEGMENT)calloc((size_t)(nsegs << hashp->SSHIFT),
    941 	    sizeof(SEGMENT))) == NULL) {
    942 		save_errno = errno;
    943 		(void)hdestroy(hashp);
    944 		errno = save_errno;
    945 		return (-1);
    946 	}
    947 	for (i = 0; i < nsegs; i++)
    948 		hashp->dir[i] = &store[i << hashp->SSHIFT];
    949 	return (0);
    950 }
    951 
    952 #if BYTE_ORDER == LITTLE_ENDIAN
    953 /*
    954  * Hashp->hdr needs to be byteswapped.
    955  */
    956 static void
    957 swap_header_copy(srcp, destp)
    958 	HASHHDR *srcp, *destp;
    959 {
    960 	int i;
    961 
    962 	P_32_COPY(srcp->magic, destp->magic);
    963 	P_32_COPY(srcp->version, destp->version);
    964 	P_32_COPY(srcp->lorder, destp->lorder);
    965 	P_32_COPY(srcp->bsize, destp->bsize);
    966 	P_32_COPY(srcp->bshift, destp->bshift);
    967 	P_32_COPY(srcp->dsize, destp->dsize);
    968 	P_32_COPY(srcp->ssize, destp->ssize);
    969 	P_32_COPY(srcp->sshift, destp->sshift);
    970 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
    971 	P_32_COPY(srcp->last_freed, destp->last_freed);
    972 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
    973 	P_32_COPY(srcp->high_mask, destp->high_mask);
    974 	P_32_COPY(srcp->low_mask, destp->low_mask);
    975 	P_32_COPY(srcp->ffactor, destp->ffactor);
    976 	P_32_COPY(srcp->nkeys, destp->nkeys);
    977 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
    978 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
    979 	for (i = 0; i < NCACHED; i++) {
    980 		P_32_COPY(srcp->spares[i], destp->spares[i]);
    981 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
    982 	}
    983 }
    984 
    985 static void
    986 swap_header(hashp)
    987 	HTAB *hashp;
    988 {
    989 	HASHHDR *hdrp;
    990 	int i;
    991 
    992 	hdrp = &hashp->hdr;
    993 
    994 	M_32_SWAP(hdrp->magic);
    995 	M_32_SWAP(hdrp->version);
    996 	M_32_SWAP(hdrp->lorder);
    997 	M_32_SWAP(hdrp->bsize);
    998 	M_32_SWAP(hdrp->bshift);
    999 	M_32_SWAP(hdrp->dsize);
   1000 	M_32_SWAP(hdrp->ssize);
   1001 	M_32_SWAP(hdrp->sshift);
   1002 	M_32_SWAP(hdrp->ovfl_point);
   1003 	M_32_SWAP(hdrp->last_freed);
   1004 	M_32_SWAP(hdrp->max_bucket);
   1005 	M_32_SWAP(hdrp->high_mask);
   1006 	M_32_SWAP(hdrp->low_mask);
   1007 	M_32_SWAP(hdrp->ffactor);
   1008 	M_32_SWAP(hdrp->nkeys);
   1009 	M_32_SWAP(hdrp->hdrpages);
   1010 	M_32_SWAP(hdrp->h_charkey);
   1011 	for (i = 0; i < NCACHED; i++) {
   1012 		M_32_SWAP(hdrp->spares[i]);
   1013 		M_16_SWAP(hdrp->bitmaps[i]);
   1014 	}
   1015 }
   1016 #endif
   1017