xref: /freebsd/lib/libc/db/hash/hash.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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  * $FreeBSD$
37  */
38 
39 #if defined(LIBC_SCCS) && !defined(lint)
40 static char sccsid[] = "@(#)hash.c	8.9 (Berkeley) 6/16/94";
41 #endif /* LIBC_SCCS and not lint */
42 
43 #include <sys/param.h>
44 #include <sys/stat.h>
45 
46 #include <errno.h>
47 #include <fcntl.h>
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <unistd.h>
52 #ifdef DEBUG
53 #include <assert.h>
54 #endif
55 
56 #include <db.h>
57 #include "hash.h"
58 #include "page.h"
59 #include "extern.h"
60 
61 static int   alloc_segs __P((HTAB *, int));
62 static int   flush_meta __P((HTAB *));
63 static int   hash_access __P((HTAB *, ACTION, DBT *, DBT *));
64 static int   hash_close __P((DB *));
65 static int   hash_delete __P((const DB *, const DBT *, u_int32_t));
66 static int   hash_fd __P((const DB *));
67 static int   hash_get __P((const DB *, const DBT *, DBT *, u_int32_t));
68 static int   hash_put __P((const DB *, DBT *, const DBT *, u_int32_t));
69 static void *hash_realloc __P((SEGMENT **, int, int));
70 static int   hash_seq __P((const DB *, DBT *, DBT *, u_int32_t));
71 static int   hash_sync __P((const DB *, u_int32_t));
72 static int   hdestroy __P((HTAB *));
73 static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *));
74 static int   init_htab __P((HTAB *, int));
75 #if BYTE_ORDER == LITTLE_ENDIAN
76 static void  swap_header __P((HTAB *));
77 static void  swap_header_copy __P((HASHHDR *, HASHHDR *));
78 #endif
79 
80 /* Fast arithmetic, relying on powers of 2, */
81 #define MOD(x, y)		((x) & ((y) - 1))
82 
83 #define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
84 
85 /* Return values */
86 #define	SUCCESS	 (0)
87 #define	ERROR	(-1)
88 #define	ABNORMAL (1)
89 
90 #ifdef HASH_STATISTICS
91 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
92 #endif
93 
94 /************************** INTERFACE ROUTINES ***************************/
95 /* OPEN/CLOSE */
96 
97 extern DB *
98 __hash_open(file, flags, mode, info, dflags)
99 	const char *file;
100 	int flags, mode, dflags;
101 	const HASHINFO *info;	/* Special directives for create */
102 {
103 	HTAB *hashp;
104 	struct stat statbuf;
105 	DB *dbp;
106 	int bpages, hdrsize, new_table, nsegs, save_errno;
107 
108 	if ((flags & O_ACCMODE) == O_WRONLY) {
109 		errno = EINVAL;
110 		return (NULL);
111 	}
112 
113 	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
114 		return (NULL);
115 	hashp->fp = -1;
116 
117 	/*
118 	 * Even if user wants write only, we need to be able to read
119 	 * the actual file, so we need to open it read/write. But, the
120 	 * field in the hashp structure needs to be accurate so that
121 	 * we can check accesses.
122 	 */
123 	hashp->flags = flags;
124 
125 	new_table = 0;
126 	if (!file || (flags & O_TRUNC) ||
127 	    (stat(file, &statbuf) && (errno == ENOENT))) {
128 		if (errno == ENOENT)
129 			errno = 0; /* Just in case someone looks at errno */
130 		new_table = 1;
131 	}
132 	if (file) {
133 		if ((hashp->fp = _open(file, flags, mode)) == -1)
134 			RETURN_ERROR(errno, error0);
135 
136 		/* if the .db file is empty, and we had permission to create
137 		   a new .db file, then reinitialize the database */
138 		if ((flags & O_CREAT) &&
139 		     fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
140 			new_table = 1;
141 
142 		(void)_fcntl(hashp->fp, F_SETFD, 1);
143 	}
144 	if (new_table) {
145 		if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
146 			RETURN_ERROR(errno, error1);
147 	} else {
148 		/* Table already exists */
149 		if (info && info->hash)
150 			hashp->hash = info->hash;
151 		else
152 			hashp->hash = __default_hash;
153 
154 		hdrsize = _read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
155 #if BYTE_ORDER == LITTLE_ENDIAN
156 		swap_header(hashp);
157 #endif
158 		if (hdrsize == -1)
159 			RETURN_ERROR(errno, error1);
160 		if (hdrsize != sizeof(HASHHDR))
161 			RETURN_ERROR(EFTYPE, error1);
162 		/* Verify file type, versions and hash function */
163 		if (hashp->MAGIC != HASHMAGIC)
164 			RETURN_ERROR(EFTYPE, error1);
165 #define	OLDHASHVERSION	1
166 		if (hashp->VERSION != HASHVERSION &&
167 		    hashp->VERSION != OLDHASHVERSION)
168 			RETURN_ERROR(EFTYPE, error1);
169 		if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
170 			RETURN_ERROR(EFTYPE, error1);
171 		/*
172 		 * Figure out how many segments we need.  Max_Bucket is the
173 		 * maximum bucket number, so the number of buckets is
174 		 * max_bucket + 1.
175 		 */
176 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
177 			 hashp->SGSIZE;
178 		hashp->nsegs = 0;
179 		if (alloc_segs(hashp, nsegs))
180 			/*
181 			 * If alloc_segs fails, table will have been destroyed
182 			 * and errno will have been set.
183 			 */
184 			return (NULL);
185 		/* Read in bitmaps */
186 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
187 		    (hashp->BSIZE << BYTE_SHIFT) - 1) >>
188 		    (hashp->BSHIFT + BYTE_SHIFT);
189 
190 		hashp->nmaps = bpages;
191 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
192 	}
193 
194 	/* Initialize Buffer Manager */
195 	if (info && info->cachesize)
196 		__buf_init(hashp, info->cachesize);
197 	else
198 		__buf_init(hashp, DEF_BUFSIZE);
199 
200 	hashp->new_file = new_table;
201 	hashp->save_file = file && (hashp->flags & O_RDWR);
202 	hashp->cbucket = -1;
203 	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
204 		save_errno = errno;
205 		hdestroy(hashp);
206 		errno = save_errno;
207 		return (NULL);
208 	}
209 	dbp->internal = hashp;
210 	dbp->close = hash_close;
211 	dbp->del = hash_delete;
212 	dbp->fd = hash_fd;
213 	dbp->get = hash_get;
214 	dbp->put = hash_put;
215 	dbp->seq = hash_seq;
216 	dbp->sync = hash_sync;
217 	dbp->type = DB_HASH;
218 
219 #ifdef DEBUG
220 	(void)fprintf(stderr,
221 "%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",
222 	    "init_htab:",
223 	    "TABLE POINTER   ", hashp,
224 	    "BUCKET SIZE     ", hashp->BSIZE,
225 	    "BUCKET SHIFT    ", hashp->BSHIFT,
226 	    "DIRECTORY SIZE  ", hashp->DSIZE,
227 	    "SEGMENT SIZE    ", hashp->SGSIZE,
228 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
229 	    "FILL FACTOR     ", hashp->FFACTOR,
230 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
231 	    "OVFL POINT	     ", hashp->OVFL_POINT,
232 	    "LAST FREED      ", hashp->LAST_FREED,
233 	    "HIGH MASK       ", hashp->HIGH_MASK,
234 	    "LOW  MASK       ", hashp->LOW_MASK,
235 	    "NSEGS           ", hashp->nsegs,
236 	    "NKEYS           ", hashp->NKEYS);
237 #endif
238 #ifdef HASH_STATISTICS
239 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
240 #endif
241 	return (dbp);
242 
243 error1:
244 	if (hashp != NULL)
245 		(void)_close(hashp->fp);
246 
247 error0:
248 	free(hashp);
249 	errno = save_errno;
250 	return (NULL);
251 }
252 
253 static int
254 hash_close(dbp)
255 	DB *dbp;
256 {
257 	HTAB *hashp;
258 	int retval;
259 
260 	if (!dbp)
261 		return (ERROR);
262 
263 	hashp = (HTAB *)dbp->internal;
264 	retval = hdestroy(hashp);
265 	free(dbp);
266 	return (retval);
267 }
268 
269 static int
270 hash_fd(dbp)
271 	const DB *dbp;
272 {
273 	HTAB *hashp;
274 
275 	if (!dbp)
276 		return (ERROR);
277 
278 	hashp = (HTAB *)dbp->internal;
279 	if (hashp->fp == -1) {
280 		errno = ENOENT;
281 		return (-1);
282 	}
283 	return (hashp->fp);
284 }
285 
286 /************************** LOCAL CREATION ROUTINES **********************/
287 static HTAB *
288 init_hash(hashp, file, info)
289 	HTAB *hashp;
290 	const char *file;
291 	HASHINFO *info;
292 {
293 	struct stat statbuf;
294 	int nelem;
295 
296 	nelem = 1;
297 	hashp->NKEYS = 0;
298 	hashp->LORDER = BYTE_ORDER;
299 	hashp->BSIZE = DEF_BUCKET_SIZE;
300 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
301 	hashp->SGSIZE = DEF_SEGSIZE;
302 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
303 	hashp->DSIZE = DEF_DIRSIZE;
304 	hashp->FFACTOR = DEF_FFACTOR;
305 	hashp->hash = __default_hash;
306 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
307 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
308 
309 	/* Fix bucket size to be optimal for file system */
310 	if (file != NULL) {
311 		if (stat(file, &statbuf))
312 			return (NULL);
313 		hashp->BSIZE = statbuf.st_blksize;
314 		hashp->BSHIFT = __log2(hashp->BSIZE);
315 	}
316 
317 	if (info) {
318 		if (info->bsize) {
319 			/* Round pagesize up to power of 2 */
320 			hashp->BSHIFT = __log2(info->bsize);
321 			hashp->BSIZE = 1 << hashp->BSHIFT;
322 			if (hashp->BSIZE > MAX_BSIZE) {
323 				errno = EINVAL;
324 				return (NULL);
325 			}
326 		}
327 		if (info->ffactor)
328 			hashp->FFACTOR = info->ffactor;
329 		if (info->hash)
330 			hashp->hash = info->hash;
331 		if (info->nelem)
332 			nelem = info->nelem;
333 		if (info->lorder) {
334 			if (info->lorder != BIG_ENDIAN &&
335 			    info->lorder != LITTLE_ENDIAN) {
336 				errno = EINVAL;
337 				return (NULL);
338 			}
339 			hashp->LORDER = info->lorder;
340 		}
341 	}
342 	/* init_htab should destroy the table and set errno if it fails */
343 	if (init_htab(hashp, nelem))
344 		return (NULL);
345 	else
346 		return (hashp);
347 }
348 /*
349  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
350  * the table and set errno, so we just pass the error information along.
351  *
352  * Returns 0 on No Error
353  */
354 static int
355 init_htab(hashp, nelem)
356 	HTAB *hashp;
357 	int nelem;
358 {
359 	register int nbuckets, nsegs;
360 	int l2;
361 
362 	/*
363 	 * Divide number of elements by the fill factor and determine a
364 	 * desired number of buckets.  Allocate space for the next greater
365 	 * power of two number of buckets.
366 	 */
367 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
368 
369 	l2 = __log2(MAX(nelem, 2));
370 	nbuckets = 1 << l2;
371 
372 	hashp->SPARES[l2] = l2 + 1;
373 	hashp->SPARES[l2 + 1] = l2 + 1;
374 	hashp->OVFL_POINT = l2;
375 	hashp->LAST_FREED = 2;
376 
377 	/* First bitmap page is at: splitpoint l2 page offset 1 */
378 	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
379 		return (-1);
380 
381 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
382 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
383 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
384 	    hashp->BSHIFT) + 1;
385 
386 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
387 	nsegs = 1 << __log2(nsegs);
388 
389 	if (nsegs > hashp->DSIZE)
390 		hashp->DSIZE = nsegs;
391 	return (alloc_segs(hashp, nsegs));
392 }
393 
394 /********************** DESTROY/CLOSE ROUTINES ************************/
395 
396 /*
397  * Flushes any changes to the file if necessary and destroys the hashp
398  * structure, freeing all allocated space.
399  */
400 static int
401 hdestroy(hashp)
402 	HTAB *hashp;
403 {
404 	int i, save_errno;
405 
406 	save_errno = 0;
407 
408 #ifdef HASH_STATISTICS
409 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
410 	    hash_accesses, hash_collisions);
411 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
412 	    hash_expansions);
413 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
414 	    hash_overflows);
415 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
416 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
417 
418 	for (i = 0; i < NCACHED; i++)
419 		(void)fprintf(stderr,
420 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
421 #endif
422 	/*
423 	 * Call on buffer manager to free buffers, and if required,
424 	 * write them to disk.
425 	 */
426 	if (__buf_free(hashp, 1, hashp->save_file))
427 		save_errno = errno;
428 	if (hashp->dir) {
429 		free(*hashp->dir);	/* Free initial segments */
430 		/* Free extra segments */
431 		while (hashp->exsegs--)
432 			free(hashp->dir[--hashp->nsegs]);
433 		free(hashp->dir);
434 	}
435 	if (flush_meta(hashp) && !save_errno)
436 		save_errno = errno;
437 	/* Free Bigmaps */
438 	for (i = 0; i < hashp->nmaps; i++)
439 		if (hashp->mapp[i])
440 			free(hashp->mapp[i]);
441 
442 	if (hashp->fp != -1)
443 		(void)_close(hashp->fp);
444 
445 	free(hashp);
446 
447 	if (save_errno) {
448 		errno = save_errno;
449 		return (ERROR);
450 	}
451 	return (SUCCESS);
452 }
453 /*
454  * Write modified pages to disk
455  *
456  * Returns:
457  *	 0 == OK
458  *	-1 ERROR
459  */
460 static int
461 hash_sync(dbp, flags)
462 	const DB *dbp;
463 	u_int32_t flags;
464 {
465 	HTAB *hashp;
466 
467 	if (flags != 0) {
468 		errno = EINVAL;
469 		return (ERROR);
470 	}
471 
472 	if (!dbp)
473 		return (ERROR);
474 
475 	hashp = (HTAB *)dbp->internal;
476 	if (!hashp->save_file)
477 		return (0);
478 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
479 		return (ERROR);
480 	hashp->new_file = 0;
481 	return (0);
482 }
483 
484 /*
485  * Returns:
486  *	 0 == OK
487  *	-1 indicates that errno should be set
488  */
489 static int
490 flush_meta(hashp)
491 	HTAB *hashp;
492 {
493 	HASHHDR *whdrp;
494 #if BYTE_ORDER == LITTLE_ENDIAN
495 	HASHHDR whdr;
496 #endif
497 	int fp, i, wsize;
498 
499 	if (!hashp->save_file)
500 		return (0);
501 	hashp->MAGIC = HASHMAGIC;
502 	hashp->VERSION = HASHVERSION;
503 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
504 
505 	fp = hashp->fp;
506 	whdrp = &hashp->hdr;
507 #if BYTE_ORDER == LITTLE_ENDIAN
508 	whdrp = &whdr;
509 	swap_header_copy(&hashp->hdr, whdrp);
510 #endif
511 	if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
512 	    ((wsize = _write(fp, whdrp, sizeof(HASHHDR))) == -1))
513 		return (-1);
514 	else
515 		if (wsize != sizeof(HASHHDR)) {
516 			errno = EFTYPE;
517 			hashp->error = errno;
518 			return (-1);
519 		}
520 	for (i = 0; i < NCACHED; i++)
521 		if (hashp->mapp[i])
522 			if (__put_page(hashp, (char *)hashp->mapp[i],
523 				hashp->BITMAPS[i], 0, 1))
524 				return (-1);
525 	return (0);
526 }
527 
528 /*******************************SEARCH ROUTINES *****************************/
529 /*
530  * All the access routines return
531  *
532  * Returns:
533  *	 0 on SUCCESS
534  *	 1 to indicate an external ERROR (i.e. key not found, etc)
535  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
536  */
537 static int
538 hash_get(dbp, key, data, flag)
539 	const DB *dbp;
540 	const DBT *key;
541 	DBT *data;
542 	u_int32_t flag;
543 {
544 	HTAB *hashp;
545 
546 	hashp = (HTAB *)dbp->internal;
547 	if (flag) {
548 		hashp->error = errno = EINVAL;
549 		return (ERROR);
550 	}
551 	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
552 }
553 
554 static int
555 hash_put(dbp, key, data, flag)
556 	const DB *dbp;
557 	DBT *key;
558 	const DBT *data;
559 	u_int32_t flag;
560 {
561 	HTAB *hashp;
562 
563 	hashp = (HTAB *)dbp->internal;
564 	if (flag && flag != R_NOOVERWRITE) {
565 		hashp->error = errno = EINVAL;
566 		return (ERROR);
567 	}
568 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
569 		hashp->error = errno = EPERM;
570 		return (ERROR);
571 	}
572 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
573 	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
574 }
575 
576 static int
577 hash_delete(dbp, key, flag)
578 	const DB *dbp;
579 	const DBT *key;
580 	u_int32_t flag;		/* Ignored */
581 {
582 	HTAB *hashp;
583 
584 	hashp = (HTAB *)dbp->internal;
585 	if (flag && flag != R_CURSOR) {
586 		hashp->error = errno = EINVAL;
587 		return (ERROR);
588 	}
589 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
590 		hashp->error = errno = EPERM;
591 		return (ERROR);
592 	}
593 	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
594 }
595 
596 /*
597  * Assume that hashp has been set in wrapper routine.
598  */
599 static int
600 hash_access(hashp, action, key, val)
601 	HTAB *hashp;
602 	ACTION action;
603 	DBT *key, *val;
604 {
605 	register BUFHEAD *rbufp;
606 	BUFHEAD *bufp, *save_bufp;
607 	register u_int16_t *bp;
608 	register int n, ndx, off, size;
609 	register char *kp;
610 	u_int16_t pageno;
611 
612 #ifdef HASH_STATISTICS
613 	hash_accesses++;
614 #endif
615 
616 	off = hashp->BSIZE;
617 	size = key->size;
618 	kp = (char *)key->data;
619 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
620 	if (!rbufp)
621 		return (ERROR);
622 	save_bufp = rbufp;
623 
624 	/* Pin the bucket chain */
625 	rbufp->flags |= BUF_PIN;
626 	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
627 		if (bp[1] >= REAL_KEY) {
628 			/* Real key/data pair */
629 			if (size == off - *bp &&
630 			    memcmp(kp, rbufp->page + *bp, size) == 0)
631 				goto found;
632 			off = bp[1];
633 #ifdef HASH_STATISTICS
634 			hash_collisions++;
635 #endif
636 			bp += 2;
637 			ndx += 2;
638 		} else if (bp[1] == OVFLPAGE) {
639 			rbufp = __get_buf(hashp, *bp, rbufp, 0);
640 			if (!rbufp) {
641 				save_bufp->flags &= ~BUF_PIN;
642 				return (ERROR);
643 			}
644 			/* FOR LOOP INIT */
645 			bp = (u_int16_t *)rbufp->page;
646 			n = *bp++;
647 			ndx = 1;
648 			off = hashp->BSIZE;
649 		} else if (bp[1] < REAL_KEY) {
650 			if ((ndx =
651 			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
652 				goto found;
653 			if (ndx == -2) {
654 				bufp = rbufp;
655 				if (!(pageno =
656 				    __find_last_page(hashp, &bufp))) {
657 					ndx = 0;
658 					rbufp = bufp;
659 					break;	/* FOR */
660 				}
661 				rbufp = __get_buf(hashp, pageno, bufp, 0);
662 				if (!rbufp) {
663 					save_bufp->flags &= ~BUF_PIN;
664 					return (ERROR);
665 				}
666 				/* FOR LOOP INIT */
667 				bp = (u_int16_t *)rbufp->page;
668 				n = *bp++;
669 				ndx = 1;
670 				off = hashp->BSIZE;
671 			} else {
672 				save_bufp->flags &= ~BUF_PIN;
673 				return (ERROR);
674 			}
675 		}
676 
677 	/* Not found */
678 	switch (action) {
679 	case HASH_PUT:
680 	case HASH_PUTNEW:
681 		if (__addel(hashp, rbufp, key, val)) {
682 			save_bufp->flags &= ~BUF_PIN;
683 			return (ERROR);
684 		} else {
685 			save_bufp->flags &= ~BUF_PIN;
686 			return (SUCCESS);
687 		}
688 	case HASH_GET:
689 	case HASH_DELETE:
690 	default:
691 		save_bufp->flags &= ~BUF_PIN;
692 		return (ABNORMAL);
693 	}
694 
695 found:
696 	switch (action) {
697 	case HASH_PUTNEW:
698 		save_bufp->flags &= ~BUF_PIN;
699 		return (ABNORMAL);
700 	case HASH_GET:
701 		bp = (u_int16_t *)rbufp->page;
702 		if (bp[ndx + 1] < REAL_KEY) {
703 			if (__big_return(hashp, rbufp, ndx, val, 0))
704 				return (ERROR);
705 		} else {
706 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
707 			val->size = bp[ndx] - bp[ndx + 1];
708 		}
709 		break;
710 	case HASH_PUT:
711 		if ((__delpair(hashp, rbufp, ndx)) ||
712 		    (__addel(hashp, rbufp, key, val))) {
713 			save_bufp->flags &= ~BUF_PIN;
714 			return (ERROR);
715 		}
716 		break;
717 	case HASH_DELETE:
718 		if (__delpair(hashp, rbufp, ndx))
719 			return (ERROR);
720 		break;
721 	default:
722 		abort();
723 	}
724 	save_bufp->flags &= ~BUF_PIN;
725 	return (SUCCESS);
726 }
727 
728 static int
729 hash_seq(dbp, key, data, flag)
730 	const DB *dbp;
731 	DBT *key, *data;
732 	u_int32_t flag;
733 {
734 	register u_int32_t bucket;
735 	register BUFHEAD *bufp;
736 	HTAB *hashp;
737 	u_int16_t *bp, ndx;
738 
739 	hashp = (HTAB *)dbp->internal;
740 	if (flag && flag != R_FIRST && flag != R_NEXT) {
741 		hashp->error = errno = EINVAL;
742 		return (ERROR);
743 	}
744 #ifdef HASH_STATISTICS
745 	hash_accesses++;
746 #endif
747 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
748 		hashp->cbucket = 0;
749 		hashp->cndx = 1;
750 		hashp->cpage = NULL;
751 	}
752 
753 	for (bp = NULL; !bp || !bp[0]; ) {
754 		if (!(bufp = hashp->cpage)) {
755 			for (bucket = hashp->cbucket;
756 			    bucket <= hashp->MAX_BUCKET;
757 			    bucket++, hashp->cndx = 1) {
758 				bufp = __get_buf(hashp, bucket, NULL, 0);
759 				if (!bufp)
760 					return (ERROR);
761 				hashp->cpage = bufp;
762 				bp = (u_int16_t *)bufp->page;
763 				if (bp[0])
764 					break;
765 			}
766 			hashp->cbucket = bucket;
767 			if (hashp->cbucket > hashp->MAX_BUCKET) {
768 				hashp->cbucket = -1;
769 				return (ABNORMAL);
770 			}
771 		} else
772 			bp = (u_int16_t *)hashp->cpage->page;
773 
774 #ifdef DEBUG
775 		assert(bp);
776 		assert(bufp);
777 #endif
778 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
779 			bufp = hashp->cpage =
780 			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
781 			if (!bufp)
782 				return (ERROR);
783 			bp = (u_int16_t *)(bufp->page);
784 			hashp->cndx = 1;
785 		}
786 		if (!bp[0]) {
787 			hashp->cpage = NULL;
788 			++hashp->cbucket;
789 		}
790 	}
791 	ndx = hashp->cndx;
792 	if (bp[ndx + 1] < REAL_KEY) {
793 		if (__big_keydata(hashp, bufp, key, data, 1))
794 			return (ERROR);
795 	} else {
796 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
797 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
798 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
799 		data->size = bp[ndx] - bp[ndx + 1];
800 		ndx += 2;
801 		if (ndx > bp[0]) {
802 			hashp->cpage = NULL;
803 			hashp->cbucket++;
804 			hashp->cndx = 1;
805 		} else
806 			hashp->cndx = ndx;
807 	}
808 	return (SUCCESS);
809 }
810 
811 /********************************* UTILITIES ************************/
812 
813 /*
814  * Returns:
815  *	 0 ==> OK
816  *	-1 ==> Error
817  */
818 extern int
819 __expand_table(hashp)
820 	HTAB *hashp;
821 {
822 	u_int32_t old_bucket, new_bucket;
823 	int dirsize, new_segnum, spare_ndx;
824 
825 #ifdef HASH_STATISTICS
826 	hash_expansions++;
827 #endif
828 	new_bucket = ++hashp->MAX_BUCKET;
829 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
830 
831 	new_segnum = new_bucket >> hashp->SSHIFT;
832 
833 	/* Check if we need a new segment */
834 	if (new_segnum >= hashp->nsegs) {
835 		/* Check if we need to expand directory */
836 		if (new_segnum >= hashp->DSIZE) {
837 			/* Reallocate directory */
838 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
839 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
840 				return (-1);
841 			hashp->DSIZE = dirsize << 1;
842 		}
843 		if ((hashp->dir[new_segnum] =
844 		    (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
845 			return (-1);
846 		hashp->exsegs++;
847 		hashp->nsegs++;
848 	}
849 	/*
850 	 * If the split point is increasing (MAX_BUCKET's log base 2
851 	 * * increases), we need to copy the current contents of the spare
852 	 * split bucket to the next bucket.
853 	 */
854 	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
855 	if (spare_ndx > hashp->OVFL_POINT) {
856 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
857 		hashp->OVFL_POINT = spare_ndx;
858 	}
859 
860 	if (new_bucket > hashp->HIGH_MASK) {
861 		/* Starting a new doubling */
862 		hashp->LOW_MASK = hashp->HIGH_MASK;
863 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
864 	}
865 	/* Relocate records to the new bucket */
866 	return (__split_page(hashp, old_bucket, new_bucket));
867 }
868 
869 /*
870  * If realloc guarantees that the pointer is not destroyed if the realloc
871  * fails, then this routine can go away.
872  */
873 static void *
874 hash_realloc(p_ptr, oldsize, newsize)
875 	SEGMENT **p_ptr;
876 	int oldsize, newsize;
877 {
878 	register void *p;
879 
880 	if ( (p = malloc(newsize)) ) {
881 		memmove(p, *p_ptr, oldsize);
882 		memset((char *)p + oldsize, 0, newsize - oldsize);
883 		free(*p_ptr);
884 		*p_ptr = p;
885 	}
886 	return (p);
887 }
888 
889 extern u_int32_t
890 __call_hash(hashp, k, len)
891 	HTAB *hashp;
892 	char *k;
893 	int len;
894 {
895 	int n, bucket;
896 
897 	n = hashp->hash(k, len);
898 	bucket = n & hashp->HIGH_MASK;
899 	if (bucket > hashp->MAX_BUCKET)
900 		bucket = bucket & hashp->LOW_MASK;
901 	return (bucket);
902 }
903 
904 /*
905  * Allocate segment table.  On error, destroy the table and set errno.
906  *
907  * Returns 0 on success
908  */
909 static int
910 alloc_segs(hashp, nsegs)
911 	HTAB *hashp;
912 	int nsegs;
913 {
914 	register int i;
915 	register SEGMENT store;
916 
917 	int save_errno;
918 
919 	if ((hashp->dir =
920 	    (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
921 		save_errno = errno;
922 		(void)hdestroy(hashp);
923 		errno = save_errno;
924 		return (-1);
925 	}
926 	/* Allocate segments */
927 	if ((store =
928 	    (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
929 		save_errno = errno;
930 		(void)hdestroy(hashp);
931 		errno = save_errno;
932 		return (-1);
933 	}
934 	for (i = 0; i < nsegs; i++, hashp->nsegs++)
935 		hashp->dir[i] = &store[i << hashp->SSHIFT];
936 	return (0);
937 }
938 
939 #if BYTE_ORDER == LITTLE_ENDIAN
940 /*
941  * Hashp->hdr needs to be byteswapped.
942  */
943 static void
944 swap_header_copy(srcp, destp)
945 	HASHHDR *srcp, *destp;
946 {
947 	int i;
948 
949 	P_32_COPY(srcp->magic, destp->magic);
950 	P_32_COPY(srcp->version, destp->version);
951 	P_32_COPY(srcp->lorder, destp->lorder);
952 	P_32_COPY(srcp->bsize, destp->bsize);
953 	P_32_COPY(srcp->bshift, destp->bshift);
954 	P_32_COPY(srcp->dsize, destp->dsize);
955 	P_32_COPY(srcp->ssize, destp->ssize);
956 	P_32_COPY(srcp->sshift, destp->sshift);
957 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
958 	P_32_COPY(srcp->last_freed, destp->last_freed);
959 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
960 	P_32_COPY(srcp->high_mask, destp->high_mask);
961 	P_32_COPY(srcp->low_mask, destp->low_mask);
962 	P_32_COPY(srcp->ffactor, destp->ffactor);
963 	P_32_COPY(srcp->nkeys, destp->nkeys);
964 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
965 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
966 	for (i = 0; i < NCACHED; i++) {
967 		P_32_COPY(srcp->spares[i], destp->spares[i]);
968 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
969 	}
970 }
971 
972 static void
973 swap_header(hashp)
974 	HTAB *hashp;
975 {
976 	HASHHDR *hdrp;
977 	int i;
978 
979 	hdrp = &hashp->hdr;
980 
981 	M_32_SWAP(hdrp->magic);
982 	M_32_SWAP(hdrp->version);
983 	M_32_SWAP(hdrp->lorder);
984 	M_32_SWAP(hdrp->bsize);
985 	M_32_SWAP(hdrp->bshift);
986 	M_32_SWAP(hdrp->dsize);
987 	M_32_SWAP(hdrp->ssize);
988 	M_32_SWAP(hdrp->sshift);
989 	M_32_SWAP(hdrp->ovfl_point);
990 	M_32_SWAP(hdrp->last_freed);
991 	M_32_SWAP(hdrp->max_bucket);
992 	M_32_SWAP(hdrp->high_mask);
993 	M_32_SWAP(hdrp->low_mask);
994 	M_32_SWAP(hdrp->ffactor);
995 	M_32_SWAP(hdrp->nkeys);
996 	M_32_SWAP(hdrp->hdrpages);
997 	M_32_SWAP(hdrp->h_charkey);
998 	for (i = 0; i < NCACHED; i++) {
999 		M_32_SWAP(hdrp->spares[i]);
1000 		M_16_SWAP(hdrp->bitmaps[i]);
1001 	}
1002 }
1003 #endif
1004