1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
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 "namespace.h"
36#include <sys/param.h>
37#include <sys/stat.h>
38
39#include <errno.h>
40#include <fcntl.h>
41#include <stdio.h>
42#include <stdlib.h>
43#include <string.h>
44#include <unistd.h>
45#ifdef DEBUG
46#include <assert.h>
47#endif
48#include "un-namespace.h"
49
50#include <db.h>
51#include "hash.h"
52#include "page.h"
53#include "extern.h"
54
55static int   alloc_segs(HTAB *, int);
56static int   flush_meta(HTAB *);
57static int   hash_access(HTAB *, ACTION, DBT *, DBT *);
58static int   hash_close(DB *);
59static int   hash_delete(const DB *, const DBT *, u_int32_t);
60static int   hash_fd(const DB *);
61static int   hash_get(const DB *, const DBT *, DBT *, u_int32_t);
62static int   hash_put(const DB *, DBT *, const DBT *, u_int32_t);
63static void *hash_realloc(SEGMENT **, int, int);
64static int   hash_seq(const DB *, DBT *, DBT *, u_int32_t);
65static int   hash_sync(const DB *, u_int32_t);
66static int   hdestroy(HTAB *);
67static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
68static int   init_htab(HTAB *, int);
69#if BYTE_ORDER == LITTLE_ENDIAN
70static void  swap_header(HTAB *);
71static void  swap_header_copy(HASHHDR *, HASHHDR *);
72#endif
73
74/* Fast arithmetic, relying on powers of 2, */
75#define MOD(x, y)		((x) & ((y) - 1))
76
77#define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
78
79/* Return values */
80#define	SUCCESS	 (0)
81#define	ERROR	(-1)
82#define	ABNORMAL (1)
83
84#ifdef HASH_STATISTICS
85int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
86#endif
87
88/************************** INTERFACE ROUTINES ***************************/
89/* OPEN/CLOSE */
90
91/* ARGSUSED */
92DB *
93__hash_open(const char *file, int flags, int mode,
94    const HASHINFO *info,	/* Special directives for create */
95    int dflags)
96{
97	HTAB *hashp;
98	struct stat statbuf;
99	DB *dbp;
100	int bpages, hdrsize, new_table, nsegs, save_errno;
101
102	if ((flags & O_ACCMODE) == O_WRONLY) {
103		errno = EINVAL;
104		return (NULL);
105	}
106
107	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
108		return (NULL);
109	hashp->fp = -1;
110
111	/*
112	 * Even if user wants write only, we need to be able to read
113	 * the actual file, so we need to open it read/write. But, the
114	 * field in the hashp structure needs to be accurate so that
115	 * we can check accesses.
116	 */
117	hashp->flags = flags;
118
119	if (file) {
120		if ((hashp->fp = _open(file, flags | O_CLOEXEC, mode)) == -1)
121			RETURN_ERROR(errno, error0);
122		new_table = _fstat(hashp->fp, &statbuf) == 0 &&
123		    statbuf.st_size == 0 && (flags & O_ACCMODE) != O_RDONLY;
124	} else
125		new_table = 1;
126
127	if (new_table) {
128		if (!(hashp = init_hash(hashp, file, info)))
129			RETURN_ERROR(errno, error1);
130	} else {
131		/* Table already exists */
132		if (info && info->hash)
133			hashp->hash = info->hash;
134		else
135			hashp->hash = __default_hash;
136
137		hdrsize = _read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
138#if BYTE_ORDER == LITTLE_ENDIAN
139		swap_header(hashp);
140#endif
141		if (hdrsize == -1)
142			RETURN_ERROR(errno, error1);
143		if (hdrsize != sizeof(HASHHDR))
144			RETURN_ERROR(EFTYPE, error1);
145		/* Verify file type, versions and hash function */
146		if (hashp->MAGIC != HASHMAGIC)
147			RETURN_ERROR(EFTYPE, error1);
148#define	OLDHASHVERSION	1
149		if (hashp->VERSION != HASHVERSION &&
150		    hashp->VERSION != OLDHASHVERSION)
151			RETURN_ERROR(EFTYPE, error1);
152		if ((int32_t)hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
153			RETURN_ERROR(EFTYPE, error1);
154		/*
155		 * Figure out how many segments we need.  Max_Bucket is the
156		 * maximum bucket number, so the number of buckets is
157		 * max_bucket + 1.
158		 */
159		nsegs = howmany(hashp->MAX_BUCKET + 1, hashp->SGSIZE);
160		if (alloc_segs(hashp, nsegs))
161			/*
162			 * If alloc_segs fails, table will have been destroyed
163			 * and errno will have been set.
164			 */
165			return (NULL);
166		/* Read in bitmaps */
167		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
168		    (hashp->BSIZE << BYTE_SHIFT) - 1) >>
169		    (hashp->BSHIFT + BYTE_SHIFT);
170
171		hashp->nmaps = bpages;
172		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
173	}
174
175	/* Initialize Buffer Manager */
176	if (info && info->cachesize)
177		__buf_init(hashp, info->cachesize);
178	else
179		__buf_init(hashp, DEF_BUFSIZE);
180
181	hashp->new_file = new_table;
182	hashp->save_file = file && (hashp->flags & O_RDWR);
183	hashp->cbucket = -1;
184	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
185		save_errno = errno;
186		hdestroy(hashp);
187		errno = save_errno;
188		return (NULL);
189	}
190	dbp->internal = hashp;
191	dbp->close = hash_close;
192	dbp->del = hash_delete;
193	dbp->fd = hash_fd;
194	dbp->get = hash_get;
195	dbp->put = hash_put;
196	dbp->seq = hash_seq;
197	dbp->sync = hash_sync;
198	dbp->type = DB_HASH;
199
200#ifdef DEBUG
201	(void)fprintf(stderr,
202"%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",
203	    "init_htab:",
204	    "TABLE POINTER   ", hashp,
205	    "BUCKET SIZE     ", hashp->BSIZE,
206	    "BUCKET SHIFT    ", hashp->BSHIFT,
207	    "DIRECTORY SIZE  ", hashp->DSIZE,
208	    "SEGMENT SIZE    ", hashp->SGSIZE,
209	    "SEGMENT SHIFT   ", hashp->SSHIFT,
210	    "FILL FACTOR     ", hashp->FFACTOR,
211	    "MAX BUCKET      ", hashp->MAX_BUCKET,
212	    "OVFL POINT	     ", hashp->OVFL_POINT,
213	    "LAST FREED      ", hashp->LAST_FREED,
214	    "HIGH MASK       ", hashp->HIGH_MASK,
215	    "LOW  MASK       ", hashp->LOW_MASK,
216	    "NSEGS           ", hashp->nsegs,
217	    "NKEYS           ", hashp->NKEYS);
218#endif
219#ifdef HASH_STATISTICS
220	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
221#endif
222	return (dbp);
223
224error1:
225	if (hashp != NULL)
226		(void)_close(hashp->fp);
227
228error0:
229	free(hashp);
230	errno = save_errno;
231	return (NULL);
232}
233
234static int
235hash_close(DB *dbp)
236{
237	HTAB *hashp;
238	int retval;
239
240	if (!dbp)
241		return (ERROR);
242
243	hashp = (HTAB *)dbp->internal;
244	retval = hdestroy(hashp);
245	free(dbp);
246	return (retval);
247}
248
249static int
250hash_fd(const DB *dbp)
251{
252	HTAB *hashp;
253
254	if (!dbp)
255		return (ERROR);
256
257	hashp = (HTAB *)dbp->internal;
258	if (hashp->fp == -1) {
259		errno = ENOENT;
260		return (-1);
261	}
262	return (hashp->fp);
263}
264
265/************************** LOCAL CREATION ROUTINES **********************/
266static HTAB *
267init_hash(HTAB *hashp, const char *file, const HASHINFO *info)
268{
269	struct stat statbuf;
270	int nelem;
271
272	nelem = 1;
273	hashp->NKEYS = 0;
274	hashp->LORDER = BYTE_ORDER;
275	hashp->BSIZE = DEF_BUCKET_SIZE;
276	hashp->BSHIFT = DEF_BUCKET_SHIFT;
277	hashp->SGSIZE = DEF_SEGSIZE;
278	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
279	hashp->DSIZE = DEF_DIRSIZE;
280	hashp->FFACTOR = DEF_FFACTOR;
281	hashp->hash = __default_hash;
282	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
283	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
284
285	/* Fix bucket size to be optimal for file system */
286	if (file != NULL) {
287		if (stat(file, &statbuf))
288			return (NULL);
289		hashp->BSIZE = statbuf.st_blksize;
290		if (hashp->BSIZE > MAX_BSIZE)
291			hashp->BSIZE = MAX_BSIZE;
292		hashp->BSHIFT = __log2(hashp->BSIZE);
293	}
294
295	if (info) {
296		if (info->bsize) {
297			/* Round pagesize up to power of 2 */
298			hashp->BSHIFT = __log2(info->bsize);
299			hashp->BSIZE = 1 << hashp->BSHIFT;
300			if (hashp->BSIZE > MAX_BSIZE) {
301				errno = EINVAL;
302				return (NULL);
303			}
304		}
305		if (info->ffactor)
306			hashp->FFACTOR = info->ffactor;
307		if (info->hash)
308			hashp->hash = info->hash;
309		if (info->nelem)
310			nelem = info->nelem;
311		if (info->lorder) {
312			if (info->lorder != BIG_ENDIAN &&
313			    info->lorder != LITTLE_ENDIAN) {
314				errno = EINVAL;
315				return (NULL);
316			}
317			hashp->LORDER = info->lorder;
318		}
319	}
320	/* init_htab should destroy the table and set errno if it fails */
321	if (init_htab(hashp, nelem))
322		return (NULL);
323	else
324		return (hashp);
325}
326/*
327 * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
328 * the table and set errno, so we just pass the error information along.
329 *
330 * Returns 0 on No Error
331 */
332static int
333init_htab(HTAB *hashp, int nelem)
334{
335	int nbuckets, nsegs, l2;
336
337	/*
338	 * Divide number of elements by the fill factor and determine a
339	 * desired number of buckets.  Allocate space for the next greater
340	 * power of two number of buckets.
341	 */
342	nelem = (nelem - 1) / hashp->FFACTOR + 1;
343
344	l2 = __log2(MAX(nelem, 2));
345	nbuckets = 1 << l2;
346
347	hashp->SPARES[l2] = l2 + 1;
348	hashp->SPARES[l2 + 1] = l2 + 1;
349	hashp->OVFL_POINT = l2;
350	hashp->LAST_FREED = 2;
351
352	/* First bitmap page is at: splitpoint l2 page offset 1 */
353	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
354		return (-1);
355
356	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
357	hashp->HIGH_MASK = (nbuckets << 1) - 1;
358	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
359	    hashp->BSHIFT) + 1;
360
361	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
362	nsegs = 1 << __log2(nsegs);
363
364	if (nsegs > hashp->DSIZE)
365		hashp->DSIZE = nsegs;
366	return (alloc_segs(hashp, nsegs));
367}
368
369/********************** DESTROY/CLOSE ROUTINES ************************/
370
371/*
372 * Flushes any changes to the file if necessary and destroys the hashp
373 * structure, freeing all allocated space.
374 */
375static int
376hdestroy(HTAB *hashp)
377{
378	int i, save_errno;
379
380	save_errno = 0;
381
382#ifdef HASH_STATISTICS
383	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
384	    hash_accesses, hash_collisions);
385	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
386	    hash_expansions);
387	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
388	    hash_overflows);
389	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
390	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
391
392	for (i = 0; i < NCACHED; i++)
393		(void)fprintf(stderr,
394		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
395#endif
396	/*
397	 * Call on buffer manager to free buffers, and if required,
398	 * write them to disk.
399	 */
400	if (__buf_free(hashp, 1, hashp->save_file))
401		save_errno = errno;
402	if (hashp->dir) {
403		free(*hashp->dir);	/* Free initial segments */
404		/* Free extra segments */
405		while (hashp->exsegs--)
406			free(hashp->dir[--hashp->nsegs]);
407		free(hashp->dir);
408	}
409	if (flush_meta(hashp) && !save_errno)
410		save_errno = errno;
411	/* Free Bigmaps */
412	for (i = 0; i < hashp->nmaps; i++)
413		if (hashp->mapp[i])
414			free(hashp->mapp[i]);
415	if (hashp->tmp_key)
416		free(hashp->tmp_key);
417	if (hashp->tmp_buf)
418		free(hashp->tmp_buf);
419
420	if (hashp->fp != -1) {
421		if (hashp->save_file)
422			(void)_fsync(hashp->fp);
423		(void)_close(hashp->fp);
424	}
425
426	free(hashp);
427
428	if (save_errno) {
429		errno = save_errno;
430		return (ERROR);
431	}
432	return (SUCCESS);
433}
434/*
435 * Write modified pages to disk
436 *
437 * Returns:
438 *	 0 == OK
439 *	-1 ERROR
440 */
441static int
442hash_sync(const DB *dbp, u_int32_t flags)
443{
444	HTAB *hashp;
445
446	if (flags != 0) {
447		errno = EINVAL;
448		return (ERROR);
449	}
450
451	if (!dbp)
452		return (ERROR);
453
454	hashp = (HTAB *)dbp->internal;
455	if (!hashp->save_file)
456		return (0);
457	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
458		return (ERROR);
459	if (hashp->fp != -1 && _fsync(hashp->fp) != 0)
460		return (ERROR);
461	hashp->new_file = 0;
462	return (0);
463}
464
465/*
466 * Returns:
467 *	 0 == OK
468 *	-1 indicates that errno should be set
469 */
470static int
471flush_meta(HTAB *hashp)
472{
473	HASHHDR *whdrp;
474#if BYTE_ORDER == LITTLE_ENDIAN
475	HASHHDR whdr;
476#endif
477	int fp, i, wsize;
478
479	if (!hashp->save_file)
480		return (0);
481	hashp->MAGIC = HASHMAGIC;
482	hashp->VERSION = HASHVERSION;
483	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
484
485	fp = hashp->fp;
486	whdrp = &hashp->hdr;
487#if BYTE_ORDER == LITTLE_ENDIAN
488	whdrp = &whdr;
489	swap_header_copy(&hashp->hdr, whdrp);
490#endif
491	if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
492		return (-1);
493	else
494		if (wsize != sizeof(HASHHDR)) {
495			errno = EFTYPE;
496			hashp->error = errno;
497			return (-1);
498		}
499	for (i = 0; i < NCACHED; i++)
500		if (hashp->mapp[i])
501			if (__put_page(hashp, (char *)hashp->mapp[i],
502				hashp->BITMAPS[i], 0, 1))
503				return (-1);
504	return (0);
505}
506
507/*******************************SEARCH ROUTINES *****************************/
508/*
509 * All the access routines return
510 *
511 * Returns:
512 *	 0 on SUCCESS
513 *	 1 to indicate an external ERROR (i.e. key not found, etc)
514 *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
515 */
516static int
517hash_get(const DB *dbp, const DBT *key, DBT *data, u_int32_t flag)
518{
519	HTAB *hashp;
520
521	hashp = (HTAB *)dbp->internal;
522	if (flag) {
523		hashp->error = errno = EINVAL;
524		return (ERROR);
525	}
526	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
527}
528
529static int
530hash_put(const DB *dbp, DBT *key, const DBT *data, u_int32_t flag)
531{
532	HTAB *hashp;
533
534	hashp = (HTAB *)dbp->internal;
535	if (flag && flag != R_NOOVERWRITE) {
536		hashp->error = errno = EINVAL;
537		return (ERROR);
538	}
539	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
540		hashp->error = errno = EPERM;
541		return (ERROR);
542	}
543	return (hash_access(hashp, flag == R_NOOVERWRITE ?
544	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
545}
546
547static int
548hash_delete(const DB *dbp, const DBT *key,
549    u_int32_t flag)		/* Ignored */
550{
551	HTAB *hashp;
552
553	hashp = (HTAB *)dbp->internal;
554	if (flag && flag != R_CURSOR) {
555		hashp->error = errno = EINVAL;
556		return (ERROR);
557	}
558	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
559		hashp->error = errno = EPERM;
560		return (ERROR);
561	}
562	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
563}
564
565/*
566 * Assume that hashp has been set in wrapper routine.
567 */
568static int
569hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
570{
571	BUFHEAD *rbufp;
572	BUFHEAD *bufp, *save_bufp;
573	u_int16_t *bp;
574	int n, ndx, off, size;
575	char *kp;
576	u_int16_t pageno;
577
578#ifdef HASH_STATISTICS
579	hash_accesses++;
580#endif
581
582	off = hashp->BSIZE;
583	size = key->size;
584	kp = (char *)key->data;
585	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
586	if (!rbufp)
587		return (ERROR);
588	save_bufp = rbufp;
589
590	/* Pin the bucket chain */
591	rbufp->flags |= BUF_PIN;
592	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
593		if (bp[1] >= REAL_KEY) {
594			/* Real key/data pair */
595			if (size == off - *bp &&
596			    memcmp(kp, rbufp->page + *bp, size) == 0)
597				goto found;
598			off = bp[1];
599#ifdef HASH_STATISTICS
600			hash_collisions++;
601#endif
602			bp += 2;
603			ndx += 2;
604		} else if (bp[1] == OVFLPAGE) {
605			rbufp = __get_buf(hashp, *bp, rbufp, 0);
606			if (!rbufp) {
607				save_bufp->flags &= ~BUF_PIN;
608				return (ERROR);
609			}
610			/* FOR LOOP INIT */
611			bp = (u_int16_t *)rbufp->page;
612			n = *bp++;
613			ndx = 1;
614			off = hashp->BSIZE;
615		} else if (bp[1] < REAL_KEY) {
616			if ((ndx =
617			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
618				goto found;
619			if (ndx == -2) {
620				bufp = rbufp;
621				if (!(pageno =
622				    __find_last_page(hashp, &bufp))) {
623					ndx = 0;
624					rbufp = bufp;
625					break;	/* FOR */
626				}
627				rbufp = __get_buf(hashp, pageno, bufp, 0);
628				if (!rbufp) {
629					save_bufp->flags &= ~BUF_PIN;
630					return (ERROR);
631				}
632				/* FOR LOOP INIT */
633				bp = (u_int16_t *)rbufp->page;
634				n = *bp++;
635				ndx = 1;
636				off = hashp->BSIZE;
637			} else {
638				save_bufp->flags &= ~BUF_PIN;
639				return (ERROR);
640			}
641		}
642
643	/* Not found */
644	switch (action) {
645	case HASH_PUT:
646	case HASH_PUTNEW:
647		if (__addel(hashp, rbufp, key, val)) {
648			save_bufp->flags &= ~BUF_PIN;
649			return (ERROR);
650		} else {
651			save_bufp->flags &= ~BUF_PIN;
652			return (SUCCESS);
653		}
654	case HASH_GET:
655	case HASH_DELETE:
656	default:
657		save_bufp->flags &= ~BUF_PIN;
658		return (ABNORMAL);
659	}
660
661found:
662	switch (action) {
663	case HASH_PUTNEW:
664		save_bufp->flags &= ~BUF_PIN;
665		return (ABNORMAL);
666	case HASH_GET:
667		bp = (u_int16_t *)rbufp->page;
668		if (bp[ndx + 1] < REAL_KEY) {
669			if (__big_return(hashp, rbufp, ndx, val, 0))
670				return (ERROR);
671		} else {
672			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
673			val->size = bp[ndx] - bp[ndx + 1];
674		}
675		break;
676	case HASH_PUT:
677		if ((__delpair(hashp, rbufp, ndx)) ||
678		    (__addel(hashp, rbufp, key, val))) {
679			save_bufp->flags &= ~BUF_PIN;
680			return (ERROR);
681		}
682		break;
683	case HASH_DELETE:
684		if (__delpair(hashp, rbufp, ndx))
685			return (ERROR);
686		break;
687	default:
688		abort();
689	}
690	save_bufp->flags &= ~BUF_PIN;
691	return (SUCCESS);
692}
693
694static int
695hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t flag)
696{
697	u_int32_t bucket;
698	BUFHEAD *bufp;
699	HTAB *hashp;
700	u_int16_t *bp, ndx;
701
702	hashp = (HTAB *)dbp->internal;
703	if (flag && flag != R_FIRST && flag != R_NEXT) {
704		hashp->error = errno = EINVAL;
705		return (ERROR);
706	}
707#ifdef HASH_STATISTICS
708	hash_accesses++;
709#endif
710	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
711		hashp->cbucket = 0;
712		hashp->cndx = 1;
713		hashp->cpage = NULL;
714	}
715next_bucket:
716	for (bp = NULL; !bp || !bp[0]; ) {
717		if (!(bufp = hashp->cpage)) {
718			for (bucket = hashp->cbucket;
719			    bucket <= hashp->MAX_BUCKET;
720			    bucket++, hashp->cndx = 1) {
721				bufp = __get_buf(hashp, bucket, NULL, 0);
722				if (!bufp)
723					return (ERROR);
724				hashp->cpage = bufp;
725				bp = (u_int16_t *)bufp->page;
726				if (bp[0])
727					break;
728			}
729			hashp->cbucket = bucket;
730			if ((u_int32_t)hashp->cbucket > hashp->MAX_BUCKET) {
731				hashp->cbucket = -1;
732				return (ABNORMAL);
733			}
734		} else {
735			bp = (u_int16_t *)hashp->cpage->page;
736			if (flag == R_NEXT || flag == 0) {
737				hashp->cndx += 2;
738				if (hashp->cndx > bp[0]) {
739					hashp->cpage = NULL;
740					hashp->cbucket++;
741					hashp->cndx = 1;
742					goto next_bucket;
743				}
744			}
745		}
746
747#ifdef DEBUG
748		assert(bp);
749		assert(bufp);
750#endif
751		while (bp[hashp->cndx + 1] == OVFLPAGE) {
752			bufp = hashp->cpage =
753			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
754			if (!bufp)
755				return (ERROR);
756			bp = (u_int16_t *)(bufp->page);
757			hashp->cndx = 1;
758		}
759		if (!bp[0]) {
760			hashp->cpage = NULL;
761			++hashp->cbucket;
762		}
763	}
764	ndx = hashp->cndx;
765	if (bp[ndx + 1] < REAL_KEY) {
766		if (__big_keydata(hashp, bufp, key, data, 1))
767			return (ERROR);
768	} else {
769		if (hashp->cpage == NULL)
770			return (ERROR);
771		key->data = (u_char *)hashp->cpage->page + bp[ndx];
772		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
773		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
774		data->size = bp[ndx] - bp[ndx + 1];
775	}
776	return (SUCCESS);
777}
778
779/********************************* UTILITIES ************************/
780
781/*
782 * Returns:
783 *	 0 ==> OK
784 *	-1 ==> Error
785 */
786int
787__expand_table(HTAB *hashp)
788{
789	u_int32_t old_bucket, new_bucket;
790	int dirsize, new_segnum, spare_ndx;
791
792#ifdef HASH_STATISTICS
793	hash_expansions++;
794#endif
795	new_bucket = ++hashp->MAX_BUCKET;
796	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
797
798	new_segnum = new_bucket >> hashp->SSHIFT;
799
800	/* Check if we need a new segment */
801	if (new_segnum >= hashp->nsegs) {
802		/* Check if we need to expand directory */
803		if (new_segnum >= hashp->DSIZE) {
804			/* Reallocate directory */
805			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
806			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
807				return (-1);
808			hashp->DSIZE = dirsize << 1;
809		}
810		if ((hashp->dir[new_segnum] =
811		    calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
812			return (-1);
813		hashp->exsegs++;
814		hashp->nsegs++;
815	}
816	/*
817	 * If the split point is increasing (MAX_BUCKET's log base 2
818	 * * increases), we need to copy the current contents of the spare
819	 * split bucket to the next bucket.
820	 */
821	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
822	if (spare_ndx > hashp->OVFL_POINT) {
823		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
824		hashp->OVFL_POINT = spare_ndx;
825	}
826
827	if (new_bucket > hashp->HIGH_MASK) {
828		/* Starting a new doubling */
829		hashp->LOW_MASK = hashp->HIGH_MASK;
830		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
831	}
832	/* Relocate records to the new bucket */
833	return (__split_page(hashp, old_bucket, new_bucket));
834}
835
836/*
837 * If realloc guarantees that the pointer is not destroyed if the realloc
838 * fails, then this routine can go away.
839 */
840static void *
841hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
842{
843	void *p;
844
845	if ( (p = malloc(newsize)) ) {
846		memmove(p, *p_ptr, oldsize);
847		memset((char *)p + oldsize, 0, newsize - oldsize);
848		free(*p_ptr);
849		*p_ptr = p;
850	}
851	return (p);
852}
853
854u_int32_t
855__call_hash(HTAB *hashp, char *k, int len)
856{
857	unsigned int n, bucket;
858
859	n = hashp->hash(k, len);
860	bucket = n & hashp->HIGH_MASK;
861	if (bucket > hashp->MAX_BUCKET)
862		bucket = bucket & hashp->LOW_MASK;
863	return (bucket);
864}
865
866/*
867 * Allocate segment table.  On error, destroy the table and set errno.
868 *
869 * Returns 0 on success
870 */
871static int
872alloc_segs(HTAB *hashp, int nsegs)
873{
874	int i;
875	SEGMENT store;
876
877	int save_errno;
878
879	if ((hashp->dir =
880	    calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
881		save_errno = errno;
882		(void)hdestroy(hashp);
883		errno = save_errno;
884		return (-1);
885	}
886	hashp->nsegs = nsegs;
887	if (nsegs == 0)
888		return (0);
889	/* Allocate segments */
890	if ((store = calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
891		save_errno = errno;
892		(void)hdestroy(hashp);
893		errno = save_errno;
894		return (-1);
895	}
896	for (i = 0; i < nsegs; i++)
897		hashp->dir[i] = &store[i << hashp->SSHIFT];
898	return (0);
899}
900
901#if BYTE_ORDER == LITTLE_ENDIAN
902/*
903 * Hashp->hdr needs to be byteswapped.
904 */
905static void
906swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
907{
908	int i;
909
910	P_32_COPY(srcp->magic, destp->magic);
911	P_32_COPY(srcp->version, destp->version);
912	P_32_COPY(srcp->lorder, destp->lorder);
913	P_32_COPY(srcp->bsize, destp->bsize);
914	P_32_COPY(srcp->bshift, destp->bshift);
915	P_32_COPY(srcp->dsize, destp->dsize);
916	P_32_COPY(srcp->ssize, destp->ssize);
917	P_32_COPY(srcp->sshift, destp->sshift);
918	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
919	P_32_COPY(srcp->last_freed, destp->last_freed);
920	P_32_COPY(srcp->max_bucket, destp->max_bucket);
921	P_32_COPY(srcp->high_mask, destp->high_mask);
922	P_32_COPY(srcp->low_mask, destp->low_mask);
923	P_32_COPY(srcp->ffactor, destp->ffactor);
924	P_32_COPY(srcp->nkeys, destp->nkeys);
925	P_32_COPY(srcp->hdrpages, destp->hdrpages);
926	P_32_COPY(srcp->h_charkey, destp->h_charkey);
927	for (i = 0; i < NCACHED; i++) {
928		P_32_COPY(srcp->spares[i], destp->spares[i]);
929		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
930	}
931}
932
933static void
934swap_header(HTAB *hashp)
935{
936	HASHHDR *hdrp;
937	int i;
938
939	hdrp = &hashp->hdr;
940
941	M_32_SWAP(hdrp->magic);
942	M_32_SWAP(hdrp->version);
943	M_32_SWAP(hdrp->lorder);
944	M_32_SWAP(hdrp->bsize);
945	M_32_SWAP(hdrp->bshift);
946	M_32_SWAP(hdrp->dsize);
947	M_32_SWAP(hdrp->ssize);
948	M_32_SWAP(hdrp->sshift);
949	M_32_SWAP(hdrp->ovfl_point);
950	M_32_SWAP(hdrp->last_freed);
951	M_32_SWAP(hdrp->max_bucket);
952	M_32_SWAP(hdrp->high_mask);
953	M_32_SWAP(hdrp->low_mask);
954	M_32_SWAP(hdrp->ffactor);
955	M_32_SWAP(hdrp->nkeys);
956	M_32_SWAP(hdrp->hdrpages);
957	M_32_SWAP(hdrp->h_charkey);
958	for (i = 0; i < NCACHED; i++) {
959		M_32_SWAP(hdrp->spares[i]);
960		M_16_SWAP(hdrp->bitmaps[i]);
961	}
962}
963#endif
964