xref: /dragonfly/lib/libc/db/hash/hash.c (revision 611395e5)
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. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * $FreeBSD: src/lib/libc/db/hash/hash.c,v 1.8 2000/01/27 23:06:08 jasone Exp $
37  * $DragonFly: src/lib/libc/db/hash/hash.c,v 1.4 2004/10/25 19:38:01 drhodus Exp $
38  *
39  * @(#)hash.c	8.9 (Berkeley) 6/16/94
40  */
41 
42 #include <sys/param.h>
43 #include <sys/stat.h>
44 
45 #include <errno.h>
46 #include <fcntl.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <string.h>
50 #include <unistd.h>
51 #ifdef DEBUG
52 #include <assert.h>
53 #endif
54 
55 #include <db.h>
56 #include "hash.h"
57 #include "page.h"
58 #include "extern.h"
59 
60 static int   alloc_segs (HTAB *, int);
61 static int   flush_meta (HTAB *);
62 static int   hash_access (HTAB *, ACTION, DBT *, DBT *);
63 static int   hash_close (DB *);
64 static int   hash_delete (const DB *, const DBT *, u_int32_t);
65 static int   hash_fd (const DB *);
66 static int   hash_get (const DB *, const DBT *, DBT *, u_int32_t);
67 static int   hash_put (const DB *, DBT *, const DBT *, u_int32_t);
68 static void *hash_realloc (SEGMENT **, int, int);
69 static int   hash_seq (const DB *, DBT *, DBT *, u_int32_t);
70 static int   hash_sync (const DB *, u_int32_t);
71 static int   hdestroy (HTAB *);
72 static HTAB *init_hash (HTAB *, const char *, HASHINFO *);
73 static int   init_htab (HTAB *, int);
74 #if BYTE_ORDER == LITTLE_ENDIAN
75 static void  swap_header (HTAB *);
76 static void  swap_header_copy (HASHHDR *, HASHHDR *);
77 #endif
78 
79 /* Fast arithmetic, relying on powers of 2, */
80 #define MOD(x, y)		((x) & ((y) - 1))
81 
82 #define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
83 
84 /* Return values */
85 #define	SUCCESS	 (0)
86 #define	ERROR	(-1)
87 #define	ABNORMAL (1)
88 
89 #ifdef HASH_STATISTICS
90 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
91 #endif
92 
93 /************************** INTERFACE ROUTINES ***************************/
94 /* OPEN/CLOSE */
95 
96 extern DB *
97 __hash_open(file, flags, mode, info, dflags)
98 	const char *file;
99 	int flags, mode, dflags;
100 	const HASHINFO *info;	/* Special directives for create */
101 {
102 	HTAB *hashp;
103 	struct stat statbuf;
104 	DB *dbp;
105 	int bpages, hdrsize, new_table, nsegs, save_errno;
106 
107 	if ((flags & O_ACCMODE) == O_WRONLY) {
108 		errno = EINVAL;
109 		return (NULL);
110 	}
111 
112 	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
113 		return (NULL);
114 	hashp->fp = -1;
115 
116 	/*
117 	 * Even if user wants write only, we need to be able to read
118 	 * the actual file, so we need to open it read/write. But, the
119 	 * field in the hashp structure needs to be accurate so that
120 	 * we can check accesses.
121 	 */
122 	hashp->flags = flags;
123 
124 	new_table = 0;
125 	if (!file || (flags & O_TRUNC) ||
126 	    (stat(file, &statbuf) && (errno == ENOENT))) {
127 		if (errno == ENOENT)
128 			errno = 0; /* Just in case someone looks at errno */
129 		new_table = 1;
130 	}
131 	if (file) {
132 		if ((hashp->fp = _open(file, flags, mode)) == -1)
133 			RETURN_ERROR(errno, error0);
134 
135 		/* if the .db file is empty, and we had permission to create
136 		   a new .db file, then reinitialize the database */
137 		if ((flags & O_CREAT) &&
138 		     fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
139 			new_table = 1;
140 
141 		(void)_fcntl(hashp->fp, F_SETFD, 1);
142 	}
143 	if (new_table) {
144 		if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
145 			RETURN_ERROR(errno, error1);
146 	} else {
147 		/* Table already exists */
148 		if (info && info->hash)
149 			hashp->hash = info->hash;
150 		else
151 			hashp->hash = __default_hash;
152 
153 		hdrsize = _read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
154 #if BYTE_ORDER == LITTLE_ENDIAN
155 		swap_header(hashp);
156 #endif
157 		if (hdrsize == -1)
158 			RETURN_ERROR(errno, error1);
159 		if (hdrsize != sizeof(HASHHDR))
160 			RETURN_ERROR(EFTYPE, error1);
161 		/* Verify file type, versions and hash function */
162 		if (hashp->MAGIC != HASHMAGIC)
163 			RETURN_ERROR(EFTYPE, error1);
164 #define	OLDHASHVERSION	1
165 		if (hashp->VERSION != HASHVERSION &&
166 		    hashp->VERSION != OLDHASHVERSION)
167 			RETURN_ERROR(EFTYPE, error1);
168 		if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
169 			RETURN_ERROR(EFTYPE, error1);
170 		/*
171 		 * Figure out how many segments we need.  Max_Bucket is the
172 		 * maximum bucket number, so the number of buckets is
173 		 * max_bucket + 1.
174 		 */
175 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
176 			 hashp->SGSIZE;
177 		hashp->nsegs = 0;
178 		if (alloc_segs(hashp, nsegs))
179 			/*
180 			 * If alloc_segs fails, table will have been destroyed
181 			 * and errno will have been set.
182 			 */
183 			return (NULL);
184 		/* Read in bitmaps */
185 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
186 		    (hashp->BSIZE << BYTE_SHIFT) - 1) >>
187 		    (hashp->BSHIFT + BYTE_SHIFT);
188 
189 		hashp->nmaps = bpages;
190 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
191 	}
192 
193 	/* Initialize Buffer Manager */
194 	if (info && info->cachesize)
195 		__buf_init(hashp, info->cachesize);
196 	else
197 		__buf_init(hashp, DEF_BUFSIZE);
198 
199 	hashp->new_file = new_table;
200 	hashp->save_file = file && (hashp->flags & O_RDWR);
201 	hashp->cbucket = -1;
202 	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
203 		save_errno = errno;
204 		hdestroy(hashp);
205 		errno = save_errno;
206 		return (NULL);
207 	}
208 	dbp->internal = hashp;
209 	dbp->close = hash_close;
210 	dbp->del = hash_delete;
211 	dbp->fd = hash_fd;
212 	dbp->get = hash_get;
213 	dbp->put = hash_put;
214 	dbp->seq = hash_seq;
215 	dbp->sync = hash_sync;
216 	dbp->type = DB_HASH;
217 
218 #ifdef DEBUG
219 	(void)fprintf(stderr,
220 "%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",
221 	    "init_htab:",
222 	    "TABLE POINTER   ", hashp,
223 	    "BUCKET SIZE     ", hashp->BSIZE,
224 	    "BUCKET SHIFT    ", hashp->BSHIFT,
225 	    "DIRECTORY SIZE  ", hashp->DSIZE,
226 	    "SEGMENT SIZE    ", hashp->SGSIZE,
227 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
228 	    "FILL FACTOR     ", hashp->FFACTOR,
229 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
230 	    "OVFL POINT	     ", hashp->OVFL_POINT,
231 	    "LAST FREED      ", hashp->LAST_FREED,
232 	    "HIGH MASK       ", hashp->HIGH_MASK,
233 	    "LOW  MASK       ", hashp->LOW_MASK,
234 	    "NSEGS           ", hashp->nsegs,
235 	    "NKEYS           ", hashp->NKEYS);
236 #endif
237 #ifdef HASH_STATISTICS
238 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
239 #endif
240 	return (dbp);
241 
242 error1:
243 	if (hashp != NULL)
244 		(void)_close(hashp->fp);
245 
246 error0:
247 	free(hashp);
248 	errno = save_errno;
249 	return (NULL);
250 }
251 
252 static int
253 hash_close(dbp)
254 	DB *dbp;
255 {
256 	HTAB *hashp;
257 	int retval;
258 
259 	if (!dbp)
260 		return (ERROR);
261 
262 	hashp = (HTAB *)dbp->internal;
263 	retval = hdestroy(hashp);
264 	free(dbp);
265 	return (retval);
266 }
267 
268 static int
269 hash_fd(dbp)
270 	const DB *dbp;
271 {
272 	HTAB *hashp;
273 
274 	if (!dbp)
275 		return (ERROR);
276 
277 	hashp = (HTAB *)dbp->internal;
278 	if (hashp->fp == -1) {
279 		errno = ENOENT;
280 		return (-1);
281 	}
282 	return (hashp->fp);
283 }
284 
285 /************************** LOCAL CREATION ROUTINES **********************/
286 static HTAB *
287 init_hash(hashp, file, info)
288 	HTAB *hashp;
289 	const char *file;
290 	HASHINFO *info;
291 {
292 	struct stat statbuf;
293 	int nelem;
294 
295 	nelem = 1;
296 	hashp->NKEYS = 0;
297 	hashp->LORDER = BYTE_ORDER;
298 	hashp->BSIZE = DEF_BUCKET_SIZE;
299 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
300 	hashp->SGSIZE = DEF_SEGSIZE;
301 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
302 	hashp->DSIZE = DEF_DIRSIZE;
303 	hashp->FFACTOR = DEF_FFACTOR;
304 	hashp->hash = __default_hash;
305 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
306 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
307 
308 	/* Fix bucket size to be optimal for file system */
309 	if (file != NULL) {
310 		if (stat(file, &statbuf))
311 			return (NULL);
312 		hashp->BSIZE = statbuf.st_blksize;
313 		hashp->BSHIFT = __log2(hashp->BSIZE);
314 	}
315 
316 	if (info) {
317 		if (info->bsize) {
318 			/* Round pagesize up to power of 2 */
319 			hashp->BSHIFT = __log2(info->bsize);
320 			hashp->BSIZE = 1 << hashp->BSHIFT;
321 			if (hashp->BSIZE > MAX_BSIZE) {
322 				errno = EINVAL;
323 				return (NULL);
324 			}
325 		}
326 		if (info->ffactor)
327 			hashp->FFACTOR = info->ffactor;
328 		if (info->hash)
329 			hashp->hash = info->hash;
330 		if (info->nelem)
331 			nelem = info->nelem;
332 		if (info->lorder) {
333 			if (info->lorder != BIG_ENDIAN &&
334 			    info->lorder != LITTLE_ENDIAN) {
335 				errno = EINVAL;
336 				return (NULL);
337 			}
338 			hashp->LORDER = info->lorder;
339 		}
340 	}
341 	/* init_htab should destroy the table and set errno if it fails */
342 	if (init_htab(hashp, nelem))
343 		return (NULL);
344 	else
345 		return (hashp);
346 }
347 /*
348  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
349  * the table and set errno, so we just pass the error information along.
350  *
351  * Returns 0 on No Error
352  */
353 static int
354 init_htab(hashp, nelem)
355 	HTAB *hashp;
356 	int nelem;
357 {
358 	int nbuckets, nsegs;
359 	int l2;
360 
361 	/*
362 	 * Divide number of elements by the fill factor and determine a
363 	 * desired number of buckets.  Allocate space for the next greater
364 	 * power of two number of buckets.
365 	 */
366 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
367 
368 	l2 = __log2(MAX(nelem, 2));
369 	nbuckets = 1 << l2;
370 
371 	hashp->SPARES[l2] = l2 + 1;
372 	hashp->SPARES[l2 + 1] = l2 + 1;
373 	hashp->OVFL_POINT = l2;
374 	hashp->LAST_FREED = 2;
375 
376 	/* First bitmap page is at: splitpoint l2 page offset 1 */
377 	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
378 		return (-1);
379 
380 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
381 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
382 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
383 	    hashp->BSHIFT) + 1;
384 
385 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
386 	nsegs = 1 << __log2(nsegs);
387 
388 	if (nsegs > hashp->DSIZE)
389 		hashp->DSIZE = nsegs;
390 	return (alloc_segs(hashp, nsegs));
391 }
392 
393 /********************** DESTROY/CLOSE ROUTINES ************************/
394 
395 /*
396  * Flushes any changes to the file if necessary and destroys the hashp
397  * structure, freeing all allocated space.
398  */
399 static int
400 hdestroy(hashp)
401 	HTAB *hashp;
402 {
403 	int i, save_errno;
404 
405 	save_errno = 0;
406 
407 #ifdef HASH_STATISTICS
408 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
409 	    hash_accesses, hash_collisions);
410 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
411 	    hash_expansions);
412 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
413 	    hash_overflows);
414 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
415 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
416 
417 	for (i = 0; i < NCACHED; i++)
418 		(void)fprintf(stderr,
419 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
420 #endif
421 	/*
422 	 * Call on buffer manager to free buffers, and if required,
423 	 * write them to disk.
424 	 */
425 	if (__buf_free(hashp, 1, hashp->save_file))
426 		save_errno = errno;
427 	if (hashp->dir) {
428 		free(*hashp->dir);	/* Free initial segments */
429 		/* Free extra segments */
430 		while (hashp->exsegs--)
431 			free(hashp->dir[--hashp->nsegs]);
432 		free(hashp->dir);
433 	}
434 	if (flush_meta(hashp) && !save_errno)
435 		save_errno = errno;
436 	/* Free Bigmaps */
437 	for (i = 0; i < hashp->nmaps; i++)
438 		if (hashp->mapp[i])
439 			free(hashp->mapp[i]);
440 
441 	if (hashp->fp != -1)
442 		(void)_close(hashp->fp);
443 
444 	free(hashp);
445 
446 	if (save_errno) {
447 		errno = save_errno;
448 		return (ERROR);
449 	}
450 	return (SUCCESS);
451 }
452 /*
453  * Write modified pages to disk
454  *
455  * Returns:
456  *	 0 == OK
457  *	-1 ERROR
458  */
459 static int
460 hash_sync(dbp, flags)
461 	const DB *dbp;
462 	u_int32_t flags;
463 {
464 	HTAB *hashp;
465 
466 	if (flags != 0) {
467 		errno = EINVAL;
468 		return (ERROR);
469 	}
470 
471 	if (!dbp)
472 		return (ERROR);
473 
474 	hashp = (HTAB *)dbp->internal;
475 	if (!hashp->save_file)
476 		return (0);
477 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
478 		return (ERROR);
479 	hashp->new_file = 0;
480 	return (0);
481 }
482 
483 /*
484  * Returns:
485  *	 0 == OK
486  *	-1 indicates that errno should be set
487  */
488 static int
489 flush_meta(hashp)
490 	HTAB *hashp;
491 {
492 	HASHHDR *whdrp;
493 #if BYTE_ORDER == LITTLE_ENDIAN
494 	HASHHDR whdr;
495 #endif
496 	int fp, i, wsize;
497 
498 	if (!hashp->save_file)
499 		return (0);
500 	hashp->MAGIC = HASHMAGIC;
501 	hashp->VERSION = HASHVERSION;
502 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
503 
504 	fp = hashp->fp;
505 	whdrp = &hashp->hdr;
506 #if BYTE_ORDER == LITTLE_ENDIAN
507 	whdrp = &whdr;
508 	swap_header_copy(&hashp->hdr, whdrp);
509 #endif
510 	if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
511 	    ((wsize = _write(fp, whdrp, sizeof(HASHHDR))) == -1))
512 		return (-1);
513 	else
514 		if (wsize != sizeof(HASHHDR)) {
515 			errno = EFTYPE;
516 			hashp->error = errno;
517 			return (-1);
518 		}
519 	for (i = 0; i < NCACHED; i++)
520 		if (hashp->mapp[i])
521 			if (__put_page(hashp, (char *)hashp->mapp[i],
522 				hashp->BITMAPS[i], 0, 1))
523 				return (-1);
524 	return (0);
525 }
526 
527 /*******************************SEARCH ROUTINES *****************************/
528 /*
529  * All the access routines return
530  *
531  * Returns:
532  *	 0 on SUCCESS
533  *	 1 to indicate an external ERROR (i.e. key not found, etc)
534  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
535  */
536 static int
537 hash_get(dbp, key, data, flag)
538 	const DB *dbp;
539 	const DBT *key;
540 	DBT *data;
541 	u_int32_t flag;
542 {
543 	HTAB *hashp;
544 
545 	hashp = (HTAB *)dbp->internal;
546 	if (flag) {
547 		hashp->error = errno = EINVAL;
548 		return (ERROR);
549 	}
550 	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
551 }
552 
553 static int
554 hash_put(dbp, key, data, flag)
555 	const DB *dbp;
556 	DBT *key;
557 	const DBT *data;
558 	u_int32_t flag;
559 {
560 	HTAB *hashp;
561 
562 	hashp = (HTAB *)dbp->internal;
563 	if (flag && flag != R_NOOVERWRITE) {
564 		hashp->error = errno = EINVAL;
565 		return (ERROR);
566 	}
567 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
568 		hashp->error = errno = EPERM;
569 		return (ERROR);
570 	}
571 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
572 	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
573 }
574 
575 static int
576 hash_delete(dbp, key, flag)
577 	const DB *dbp;
578 	const DBT *key;
579 	u_int32_t flag;		/* Ignored */
580 {
581 	HTAB *hashp;
582 
583 	hashp = (HTAB *)dbp->internal;
584 	if (flag && flag != R_CURSOR) {
585 		hashp->error = errno = EINVAL;
586 		return (ERROR);
587 	}
588 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
589 		hashp->error = errno = EPERM;
590 		return (ERROR);
591 	}
592 	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
593 }
594 
595 /*
596  * Assume that hashp has been set in wrapper routine.
597  */
598 static int
599 hash_access(hashp, action, key, val)
600 	HTAB *hashp;
601 	ACTION action;
602 	DBT *key, *val;
603 {
604 	BUFHEAD *rbufp;
605 	BUFHEAD *bufp, *save_bufp;
606 	u_int16_t *bp;
607 	int n, ndx, off, size;
608 	char *kp;
609 	u_int16_t pageno;
610 
611 #ifdef HASH_STATISTICS
612 	hash_accesses++;
613 #endif
614 
615 	off = hashp->BSIZE;
616 	size = key->size;
617 	kp = (char *)key->data;
618 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
619 	if (!rbufp)
620 		return (ERROR);
621 	save_bufp = rbufp;
622 
623 	/* Pin the bucket chain */
624 	rbufp->flags |= BUF_PIN;
625 	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
626 		if (bp[1] >= REAL_KEY) {
627 			/* Real key/data pair */
628 			if (size == off - *bp &&
629 			    memcmp(kp, rbufp->page + *bp, size) == 0)
630 				goto found;
631 			off = bp[1];
632 #ifdef HASH_STATISTICS
633 			hash_collisions++;
634 #endif
635 			bp += 2;
636 			ndx += 2;
637 		} else if (bp[1] == OVFLPAGE) {
638 			rbufp = __get_buf(hashp, *bp, rbufp, 0);
639 			if (!rbufp) {
640 				save_bufp->flags &= ~BUF_PIN;
641 				return (ERROR);
642 			}
643 			/* FOR LOOP INIT */
644 			bp = (u_int16_t *)rbufp->page;
645 			n = *bp++;
646 			ndx = 1;
647 			off = hashp->BSIZE;
648 		} else if (bp[1] < REAL_KEY) {
649 			if ((ndx =
650 			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
651 				goto found;
652 			if (ndx == -2) {
653 				bufp = rbufp;
654 				if (!(pageno =
655 				    __find_last_page(hashp, &bufp))) {
656 					ndx = 0;
657 					rbufp = bufp;
658 					break;	/* FOR */
659 				}
660 				rbufp = __get_buf(hashp, pageno, bufp, 0);
661 				if (!rbufp) {
662 					save_bufp->flags &= ~BUF_PIN;
663 					return (ERROR);
664 				}
665 				/* FOR LOOP INIT */
666 				bp = (u_int16_t *)rbufp->page;
667 				n = *bp++;
668 				ndx = 1;
669 				off = hashp->BSIZE;
670 			} else {
671 				save_bufp->flags &= ~BUF_PIN;
672 				return (ERROR);
673 			}
674 		}
675 
676 	/* Not found */
677 	switch (action) {
678 	case HASH_PUT:
679 	case HASH_PUTNEW:
680 		if (__addel(hashp, rbufp, key, val)) {
681 			save_bufp->flags &= ~BUF_PIN;
682 			return (ERROR);
683 		} else {
684 			save_bufp->flags &= ~BUF_PIN;
685 			return (SUCCESS);
686 		}
687 	case HASH_GET:
688 	case HASH_DELETE:
689 	default:
690 		save_bufp->flags &= ~BUF_PIN;
691 		return (ABNORMAL);
692 	}
693 
694 found:
695 	switch (action) {
696 	case HASH_PUTNEW:
697 		save_bufp->flags &= ~BUF_PIN;
698 		return (ABNORMAL);
699 	case HASH_GET:
700 		bp = (u_int16_t *)rbufp->page;
701 		if (bp[ndx + 1] < REAL_KEY) {
702 			if (__big_return(hashp, rbufp, ndx, val, 0))
703 				return (ERROR);
704 		} else {
705 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
706 			val->size = bp[ndx] - bp[ndx + 1];
707 		}
708 		break;
709 	case HASH_PUT:
710 		if ((__delpair(hashp, rbufp, ndx)) ||
711 		    (__addel(hashp, rbufp, key, val))) {
712 			save_bufp->flags &= ~BUF_PIN;
713 			return (ERROR);
714 		}
715 		break;
716 	case HASH_DELETE:
717 		if (__delpair(hashp, rbufp, ndx))
718 			return (ERROR);
719 		break;
720 	default:
721 		abort();
722 	}
723 	save_bufp->flags &= ~BUF_PIN;
724 	return (SUCCESS);
725 }
726 
727 static int
728 hash_seq(dbp, key, data, flag)
729 	const DB *dbp;
730 	DBT *key, *data;
731 	u_int32_t flag;
732 {
733 	u_int32_t bucket;
734 	BUFHEAD *bufp;
735 	HTAB *hashp;
736 	u_int16_t *bp, ndx;
737 
738 	hashp = (HTAB *)dbp->internal;
739 	if (flag && flag != R_FIRST && flag != R_NEXT) {
740 		hashp->error = errno = EINVAL;
741 		return (ERROR);
742 	}
743 #ifdef HASH_STATISTICS
744 	hash_accesses++;
745 #endif
746 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
747 		hashp->cbucket = 0;
748 		hashp->cndx = 1;
749 		hashp->cpage = NULL;
750 	}
751 
752 	for (bp = NULL; !bp || !bp[0]; ) {
753 		if (!(bufp = hashp->cpage)) {
754 			for (bucket = hashp->cbucket;
755 			    bucket <= hashp->MAX_BUCKET;
756 			    bucket++, hashp->cndx = 1) {
757 				bufp = __get_buf(hashp, bucket, NULL, 0);
758 				if (!bufp)
759 					return (ERROR);
760 				hashp->cpage = bufp;
761 				bp = (u_int16_t *)bufp->page;
762 				if (bp[0])
763 					break;
764 			}
765 			hashp->cbucket = bucket;
766 			if (hashp->cbucket > hashp->MAX_BUCKET) {
767 				hashp->cbucket = -1;
768 				return (ABNORMAL);
769 			}
770 		} else
771 			bp = (u_int16_t *)hashp->cpage->page;
772 
773 #ifdef DEBUG
774 		assert(bp);
775 		assert(bufp);
776 #endif
777 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
778 			bufp = hashp->cpage =
779 			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
780 			if (!bufp)
781 				return (ERROR);
782 			bp = (u_int16_t *)(bufp->page);
783 			hashp->cndx = 1;
784 		}
785 		if (!bp[0]) {
786 			hashp->cpage = NULL;
787 			++hashp->cbucket;
788 		}
789 	}
790 	ndx = hashp->cndx;
791 	if (bp[ndx + 1] < REAL_KEY) {
792 		if (__big_keydata(hashp, bufp, key, data, 1))
793 			return (ERROR);
794 	} else {
795 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
796 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
797 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
798 		data->size = bp[ndx] - bp[ndx + 1];
799 		ndx += 2;
800 		if (ndx > bp[0]) {
801 			hashp->cpage = NULL;
802 			hashp->cbucket++;
803 			hashp->cndx = 1;
804 		} else
805 			hashp->cndx = ndx;
806 	}
807 	return (SUCCESS);
808 }
809 
810 /********************************* UTILITIES ************************/
811 
812 /*
813  * Returns:
814  *	 0 ==> OK
815  *	-1 ==> Error
816  */
817 extern int
818 __expand_table(hashp)
819 	HTAB *hashp;
820 {
821 	u_int32_t old_bucket, new_bucket;
822 	int dirsize, new_segnum, spare_ndx;
823 
824 #ifdef HASH_STATISTICS
825 	hash_expansions++;
826 #endif
827 	new_bucket = ++hashp->MAX_BUCKET;
828 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
829 
830 	new_segnum = new_bucket >> hashp->SSHIFT;
831 
832 	/* Check if we need a new segment */
833 	if (new_segnum >= hashp->nsegs) {
834 		/* Check if we need to expand directory */
835 		if (new_segnum >= hashp->DSIZE) {
836 			/* Reallocate directory */
837 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
838 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
839 				return (-1);
840 			hashp->DSIZE = dirsize << 1;
841 		}
842 		if ((hashp->dir[new_segnum] =
843 		    (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
844 			return (-1);
845 		hashp->exsegs++;
846 		hashp->nsegs++;
847 	}
848 	/*
849 	 * If the split point is increasing (MAX_BUCKET's log base 2
850 	 * * increases), we need to copy the current contents of the spare
851 	 * split bucket to the next bucket.
852 	 */
853 	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
854 	if (spare_ndx > hashp->OVFL_POINT) {
855 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
856 		hashp->OVFL_POINT = spare_ndx;
857 	}
858 
859 	if (new_bucket > hashp->HIGH_MASK) {
860 		/* Starting a new doubling */
861 		hashp->LOW_MASK = hashp->HIGH_MASK;
862 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
863 	}
864 	/* Relocate records to the new bucket */
865 	return (__split_page(hashp, old_bucket, new_bucket));
866 }
867 
868 /*
869  * If realloc guarantees that the pointer is not destroyed if the realloc
870  * fails, then this routine can go away.
871  */
872 static void *
873 hash_realloc(p_ptr, oldsize, newsize)
874 	SEGMENT **p_ptr;
875 	int oldsize, newsize;
876 {
877 	void *p;
878 
879 	if ( (p = malloc(newsize)) ) {
880 		memmove(p, *p_ptr, oldsize);
881 		memset((char *)p + oldsize, 0, newsize - oldsize);
882 		free(*p_ptr);
883 		*p_ptr = p;
884 	}
885 	return (p);
886 }
887 
888 extern u_int32_t
889 __call_hash(hashp, k, len)
890 	HTAB *hashp;
891 	char *k;
892 	int len;
893 {
894 	int n, bucket;
895 
896 	n = hashp->hash(k, len);
897 	bucket = n & hashp->HIGH_MASK;
898 	if (bucket > hashp->MAX_BUCKET)
899 		bucket = bucket & hashp->LOW_MASK;
900 	return (bucket);
901 }
902 
903 /*
904  * Allocate segment table.  On error, destroy the table and set errno.
905  *
906  * Returns 0 on success
907  */
908 static int
909 alloc_segs(hashp, nsegs)
910 	HTAB *hashp;
911 	int nsegs;
912 {
913 	int i;
914 	SEGMENT store;
915 
916 	int save_errno;
917 
918 	if ((hashp->dir =
919 	    (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
920 		save_errno = errno;
921 		(void)hdestroy(hashp);
922 		errno = save_errno;
923 		return (-1);
924 	}
925 	/* Allocate segments */
926 	if ((store =
927 	    (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
928 		save_errno = errno;
929 		(void)hdestroy(hashp);
930 		errno = save_errno;
931 		return (-1);
932 	}
933 	for (i = 0; i < nsegs; i++, hashp->nsegs++)
934 		hashp->dir[i] = &store[i << hashp->SSHIFT];
935 	return (0);
936 }
937 
938 #if BYTE_ORDER == LITTLE_ENDIAN
939 /*
940  * Hashp->hdr needs to be byteswapped.
941  */
942 static void
943 swap_header_copy(srcp, destp)
944 	HASHHDR *srcp, *destp;
945 {
946 	int i;
947 
948 	P_32_COPY(srcp->magic, destp->magic);
949 	P_32_COPY(srcp->version, destp->version);
950 	P_32_COPY(srcp->lorder, destp->lorder);
951 	P_32_COPY(srcp->bsize, destp->bsize);
952 	P_32_COPY(srcp->bshift, destp->bshift);
953 	P_32_COPY(srcp->dsize, destp->dsize);
954 	P_32_COPY(srcp->ssize, destp->ssize);
955 	P_32_COPY(srcp->sshift, destp->sshift);
956 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
957 	P_32_COPY(srcp->last_freed, destp->last_freed);
958 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
959 	P_32_COPY(srcp->high_mask, destp->high_mask);
960 	P_32_COPY(srcp->low_mask, destp->low_mask);
961 	P_32_COPY(srcp->ffactor, destp->ffactor);
962 	P_32_COPY(srcp->nkeys, destp->nkeys);
963 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
964 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
965 	for (i = 0; i < NCACHED; i++) {
966 		P_32_COPY(srcp->spares[i], destp->spares[i]);
967 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
968 	}
969 }
970 
971 static void
972 swap_header(hashp)
973 	HTAB *hashp;
974 {
975 	HASHHDR *hdrp;
976 	int i;
977 
978 	hdrp = &hashp->hdr;
979 
980 	M_32_SWAP(hdrp->magic);
981 	M_32_SWAP(hdrp->version);
982 	M_32_SWAP(hdrp->lorder);
983 	M_32_SWAP(hdrp->bsize);
984 	M_32_SWAP(hdrp->bshift);
985 	M_32_SWAP(hdrp->dsize);
986 	M_32_SWAP(hdrp->ssize);
987 	M_32_SWAP(hdrp->sshift);
988 	M_32_SWAP(hdrp->ovfl_point);
989 	M_32_SWAP(hdrp->last_freed);
990 	M_32_SWAP(hdrp->max_bucket);
991 	M_32_SWAP(hdrp->high_mask);
992 	M_32_SWAP(hdrp->low_mask);
993 	M_32_SWAP(hdrp->ffactor);
994 	M_32_SWAP(hdrp->nkeys);
995 	M_32_SWAP(hdrp->hdrpages);
996 	M_32_SWAP(hdrp->h_charkey);
997 	for (i = 0; i < NCACHED; i++) {
998 		M_32_SWAP(hdrp->spares[i]);
999 		M_16_SWAP(hdrp->bitmaps[i]);
1000 	}
1001 }
1002 #endif
1003