xref: /freebsd/lib/libc/db/hash/hash.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)hash.c	8.9 (Berkeley) 6/16/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "namespace.h"
40 #include <sys/param.h>
41 #include <sys/stat.h>
42 
43 #include <errno.h>
44 #include <fcntl.h>
45 #include <stdio.h>
46 #include <stdlib.h>
47 #include <string.h>
48 #include <unistd.h>
49 #ifdef DEBUG
50 #include <assert.h>
51 #endif
52 #include "un-namespace.h"
53 
54 #include <db.h>
55 #include "hash.h"
56 #include "page.h"
57 #include "extern.h"
58 
59 static int   alloc_segs(HTAB *, int);
60 static int   flush_meta(HTAB *);
61 static int   hash_access(HTAB *, ACTION, DBT *, DBT *);
62 static int   hash_close(DB *);
63 static int   hash_delete(const DB *, const DBT *, u_int32_t);
64 static int   hash_fd(const DB *);
65 static int   hash_get(const DB *, const DBT *, DBT *, u_int32_t);
66 static int   hash_put(const DB *, DBT *, const DBT *, u_int32_t);
67 static void *hash_realloc(SEGMENT **, int, int);
68 static int   hash_seq(const DB *, DBT *, DBT *, u_int32_t);
69 static int   hash_sync(const DB *, u_int32_t);
70 static int   hdestroy(HTAB *);
71 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
72 static int   init_htab(HTAB *, int);
73 #if BYTE_ORDER == LITTLE_ENDIAN
74 static void  swap_header(HTAB *);
75 static void  swap_header_copy(HASHHDR *, HASHHDR *);
76 #endif
77 
78 /* Fast arithmetic, relying on powers of 2, */
79 #define MOD(x, y)		((x) & ((y) - 1))
80 
81 #define RETURN_ERROR(ERR, LOC)	{ save_errno = ERR; goto LOC; }
82 
83 /* Return values */
84 #define	SUCCESS	 (0)
85 #define	ERROR	(-1)
86 #define	ABNORMAL (1)
87 
88 #ifdef HASH_STATISTICS
89 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
90 #endif
91 
92 /************************** INTERFACE ROUTINES ***************************/
93 /* OPEN/CLOSE */
94 
95 /* ARGSUSED */
96 DB *
97 __hash_open(const char *file, int flags, int mode,
98     const HASHINFO *info,	/* Special directives for create */
99     int dflags)
100 {
101 	HTAB *hashp;
102 	struct stat statbuf;
103 	DB *dbp;
104 	int bpages, hdrsize, new_table, nsegs, save_errno;
105 
106 	if ((flags & O_ACCMODE) == O_WRONLY) {
107 		errno = EINVAL;
108 		return (NULL);
109 	}
110 
111 	if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
112 		return (NULL);
113 	hashp->fp = -1;
114 
115 	/*
116 	 * Even if user wants write only, we need to be able to read
117 	 * the actual file, so we need to open it read/write. But, the
118 	 * field in the hashp structure needs to be accurate so that
119 	 * we can check accesses.
120 	 */
121 	hashp->flags = flags;
122 
123 	if (file) {
124 		if ((hashp->fp = _open(file, flags, mode)) == -1)
125 			RETURN_ERROR(errno, error0);
126 		(void)_fcntl(hashp->fp, F_SETFD, 1);
127 		new_table = _fstat(hashp->fp, &statbuf) == 0 &&
128 		    statbuf.st_size == 0 && (flags & O_ACCMODE) != O_RDONLY;
129 	} else
130 		new_table = 1;
131 
132 	if (new_table) {
133 		if (!(hashp = init_hash(hashp, file, info)))
134 			RETURN_ERROR(errno, error1);
135 	} else {
136 		/* Table already exists */
137 		if (info && info->hash)
138 			hashp->hash = info->hash;
139 		else
140 			hashp->hash = __default_hash;
141 
142 		hdrsize = _read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
143 #if BYTE_ORDER == LITTLE_ENDIAN
144 		swap_header(hashp);
145 #endif
146 		if (hdrsize == -1)
147 			RETURN_ERROR(errno, error1);
148 		if (hdrsize != sizeof(HASHHDR))
149 			RETURN_ERROR(EFTYPE, error1);
150 		/* Verify file type, versions and hash function */
151 		if (hashp->MAGIC != HASHMAGIC)
152 			RETURN_ERROR(EFTYPE, error1);
153 #define	OLDHASHVERSION	1
154 		if (hashp->VERSION != HASHVERSION &&
155 		    hashp->VERSION != OLDHASHVERSION)
156 			RETURN_ERROR(EFTYPE, error1);
157 		if ((int32_t)hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
158 			RETURN_ERROR(EFTYPE, error1);
159 		/*
160 		 * Figure out how many segments we need.  Max_Bucket is the
161 		 * maximum bucket number, so the number of buckets is
162 		 * max_bucket + 1.
163 		 */
164 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
165 			 hashp->SGSIZE;
166 		if (alloc_segs(hashp, nsegs))
167 			/*
168 			 * If alloc_segs fails, table will have been destroyed
169 			 * and errno will have been set.
170 			 */
171 			return (NULL);
172 		/* Read in bitmaps */
173 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
174 		    (hashp->BSIZE << BYTE_SHIFT) - 1) >>
175 		    (hashp->BSHIFT + BYTE_SHIFT);
176 
177 		hashp->nmaps = bpages;
178 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *));
179 	}
180 
181 	/* Initialize Buffer Manager */
182 	if (info && info->cachesize)
183 		__buf_init(hashp, info->cachesize);
184 	else
185 		__buf_init(hashp, DEF_BUFSIZE);
186 
187 	hashp->new_file = new_table;
188 	hashp->save_file = file && (hashp->flags & O_RDWR);
189 	hashp->cbucket = -1;
190 	if (!(dbp = (DB *)malloc(sizeof(DB)))) {
191 		save_errno = errno;
192 		hdestroy(hashp);
193 		errno = save_errno;
194 		return (NULL);
195 	}
196 	dbp->internal = hashp;
197 	dbp->close = hash_close;
198 	dbp->del = hash_delete;
199 	dbp->fd = hash_fd;
200 	dbp->get = hash_get;
201 	dbp->put = hash_put;
202 	dbp->seq = hash_seq;
203 	dbp->sync = hash_sync;
204 	dbp->type = DB_HASH;
205 
206 #ifdef DEBUG
207 	(void)fprintf(stderr,
208 "%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",
209 	    "init_htab:",
210 	    "TABLE POINTER   ", hashp,
211 	    "BUCKET SIZE     ", hashp->BSIZE,
212 	    "BUCKET SHIFT    ", hashp->BSHIFT,
213 	    "DIRECTORY SIZE  ", hashp->DSIZE,
214 	    "SEGMENT SIZE    ", hashp->SGSIZE,
215 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
216 	    "FILL FACTOR     ", hashp->FFACTOR,
217 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
218 	    "OVFL POINT	     ", hashp->OVFL_POINT,
219 	    "LAST FREED      ", hashp->LAST_FREED,
220 	    "HIGH MASK       ", hashp->HIGH_MASK,
221 	    "LOW  MASK       ", hashp->LOW_MASK,
222 	    "NSEGS           ", hashp->nsegs,
223 	    "NKEYS           ", hashp->NKEYS);
224 #endif
225 #ifdef HASH_STATISTICS
226 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
227 #endif
228 	return (dbp);
229 
230 error1:
231 	if (hashp != NULL)
232 		(void)_close(hashp->fp);
233 
234 error0:
235 	free(hashp);
236 	errno = save_errno;
237 	return (NULL);
238 }
239 
240 static int
241 hash_close(DB *dbp)
242 {
243 	HTAB *hashp;
244 	int retval;
245 
246 	if (!dbp)
247 		return (ERROR);
248 
249 	hashp = (HTAB *)dbp->internal;
250 	retval = hdestroy(hashp);
251 	free(dbp);
252 	return (retval);
253 }
254 
255 static int
256 hash_fd(const DB *dbp)
257 {
258 	HTAB *hashp;
259 
260 	if (!dbp)
261 		return (ERROR);
262 
263 	hashp = (HTAB *)dbp->internal;
264 	if (hashp->fp == -1) {
265 		errno = ENOENT;
266 		return (-1);
267 	}
268 	return (hashp->fp);
269 }
270 
271 /************************** LOCAL CREATION ROUTINES **********************/
272 static HTAB *
273 init_hash(HTAB *hashp, const char *file, const HASHINFO *info)
274 {
275 	struct stat statbuf;
276 	int nelem;
277 
278 	nelem = 1;
279 	hashp->NKEYS = 0;
280 	hashp->LORDER = BYTE_ORDER;
281 	hashp->BSIZE = DEF_BUCKET_SIZE;
282 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
283 	hashp->SGSIZE = DEF_SEGSIZE;
284 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
285 	hashp->DSIZE = DEF_DIRSIZE;
286 	hashp->FFACTOR = DEF_FFACTOR;
287 	hashp->hash = __default_hash;
288 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
289 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
290 
291 	/* Fix bucket size to be optimal for file system */
292 	if (file != NULL) {
293 		if (stat(file, &statbuf))
294 			return (NULL);
295 		hashp->BSIZE = statbuf.st_blksize;
296 		hashp->BSHIFT = __log2(hashp->BSIZE);
297 	}
298 
299 	if (info) {
300 		if (info->bsize) {
301 			/* Round pagesize up to power of 2 */
302 			hashp->BSHIFT = __log2(info->bsize);
303 			hashp->BSIZE = 1 << hashp->BSHIFT;
304 			if (hashp->BSIZE > MAX_BSIZE) {
305 				errno = EINVAL;
306 				return (NULL);
307 			}
308 		}
309 		if (info->ffactor)
310 			hashp->FFACTOR = info->ffactor;
311 		if (info->hash)
312 			hashp->hash = info->hash;
313 		if (info->nelem)
314 			nelem = info->nelem;
315 		if (info->lorder) {
316 			if (info->lorder != BIG_ENDIAN &&
317 			    info->lorder != LITTLE_ENDIAN) {
318 				errno = EINVAL;
319 				return (NULL);
320 			}
321 			hashp->LORDER = info->lorder;
322 		}
323 	}
324 	/* init_htab should destroy the table and set errno if it fails */
325 	if (init_htab(hashp, nelem))
326 		return (NULL);
327 	else
328 		return (hashp);
329 }
330 /*
331  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
332  * the table and set errno, so we just pass the error information along.
333  *
334  * Returns 0 on No Error
335  */
336 static int
337 init_htab(HTAB *hashp, int nelem)
338 {
339 	int nbuckets, nsegs, l2;
340 
341 	/*
342 	 * Divide number of elements by the fill factor and determine a
343 	 * desired number of buckets.  Allocate space for the next greater
344 	 * power of two number of buckets.
345 	 */
346 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
347 
348 	l2 = __log2(MAX(nelem, 2));
349 	nbuckets = 1 << l2;
350 
351 	hashp->SPARES[l2] = l2 + 1;
352 	hashp->SPARES[l2 + 1] = l2 + 1;
353 	hashp->OVFL_POINT = l2;
354 	hashp->LAST_FREED = 2;
355 
356 	/* First bitmap page is at: splitpoint l2 page offset 1 */
357 	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
358 		return (-1);
359 
360 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
361 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
362 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
363 	    hashp->BSHIFT) + 1;
364 
365 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
366 	nsegs = 1 << __log2(nsegs);
367 
368 	if (nsegs > hashp->DSIZE)
369 		hashp->DSIZE = nsegs;
370 	return (alloc_segs(hashp, nsegs));
371 }
372 
373 /********************** DESTROY/CLOSE ROUTINES ************************/
374 
375 /*
376  * Flushes any changes to the file if necessary and destroys the hashp
377  * structure, freeing all allocated space.
378  */
379 static int
380 hdestroy(HTAB *hashp)
381 {
382 	int i, save_errno;
383 
384 	save_errno = 0;
385 
386 #ifdef HASH_STATISTICS
387 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
388 	    hash_accesses, hash_collisions);
389 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
390 	    hash_expansions);
391 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
392 	    hash_overflows);
393 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
394 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
395 
396 	for (i = 0; i < NCACHED; i++)
397 		(void)fprintf(stderr,
398 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
399 #endif
400 	/*
401 	 * Call on buffer manager to free buffers, and if required,
402 	 * write them to disk.
403 	 */
404 	if (__buf_free(hashp, 1, hashp->save_file))
405 		save_errno = errno;
406 	if (hashp->dir) {
407 		free(*hashp->dir);	/* Free initial segments */
408 		/* Free extra segments */
409 		while (hashp->exsegs--)
410 			free(hashp->dir[--hashp->nsegs]);
411 		free(hashp->dir);
412 	}
413 	if (flush_meta(hashp) && !save_errno)
414 		save_errno = errno;
415 	/* Free Bigmaps */
416 	for (i = 0; i < hashp->nmaps; i++)
417 		if (hashp->mapp[i])
418 			free(hashp->mapp[i]);
419 	if (hashp->tmp_key)
420 		free(hashp->tmp_key);
421 	if (hashp->tmp_buf)
422 		free(hashp->tmp_buf);
423 
424 	if (hashp->fp != -1)
425 		(void)_close(hashp->fp);
426 
427 	free(hashp);
428 
429 	if (save_errno) {
430 		errno = save_errno;
431 		return (ERROR);
432 	}
433 	return (SUCCESS);
434 }
435 /*
436  * Write modified pages to disk
437  *
438  * Returns:
439  *	 0 == OK
440  *	-1 ERROR
441  */
442 static int
443 hash_sync(const DB *dbp, u_int32_t flags)
444 {
445 	HTAB *hashp;
446 
447 	if (flags != 0) {
448 		errno = EINVAL;
449 		return (ERROR);
450 	}
451 
452 	if (!dbp)
453 		return (ERROR);
454 
455 	hashp = (HTAB *)dbp->internal;
456 	if (!hashp->save_file)
457 		return (0);
458 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
459 		return (ERROR);
460 	hashp->new_file = 0;
461 	return (0);
462 }
463 
464 /*
465  * Returns:
466  *	 0 == OK
467  *	-1 indicates that errno should be set
468  */
469 static int
470 flush_meta(HTAB *hashp)
471 {
472 	HASHHDR *whdrp;
473 #if BYTE_ORDER == LITTLE_ENDIAN
474 	HASHHDR whdr;
475 #endif
476 	int fp, i, wsize;
477 
478 	if (!hashp->save_file)
479 		return (0);
480 	hashp->MAGIC = HASHMAGIC;
481 	hashp->VERSION = HASHVERSION;
482 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
483 
484 	fp = hashp->fp;
485 	whdrp = &hashp->hdr;
486 #if BYTE_ORDER == LITTLE_ENDIAN
487 	whdrp = &whdr;
488 	swap_header_copy(&hashp->hdr, whdrp);
489 #endif
490 	if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
491 		return (-1);
492 	else
493 		if (wsize != sizeof(HASHHDR)) {
494 			errno = EFTYPE;
495 			hashp->error = errno;
496 			return (-1);
497 		}
498 	for (i = 0; i < NCACHED; i++)
499 		if (hashp->mapp[i])
500 			if (__put_page(hashp, (char *)hashp->mapp[i],
501 				hashp->BITMAPS[i], 0, 1))
502 				return (-1);
503 	return (0);
504 }
505 
506 /*******************************SEARCH ROUTINES *****************************/
507 /*
508  * All the access routines return
509  *
510  * Returns:
511  *	 0 on SUCCESS
512  *	 1 to indicate an external ERROR (i.e. key not found, etc)
513  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
514  */
515 static int
516 hash_get(const DB *dbp, const DBT *key, DBT *data, u_int32_t flag)
517 {
518 	HTAB *hashp;
519 
520 	hashp = (HTAB *)dbp->internal;
521 	if (flag) {
522 		hashp->error = errno = EINVAL;
523 		return (ERROR);
524 	}
525 	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
526 }
527 
528 static int
529 hash_put(const DB *dbp, DBT *key, const DBT *data, u_int32_t flag)
530 {
531 	HTAB *hashp;
532 
533 	hashp = (HTAB *)dbp->internal;
534 	if (flag && flag != R_NOOVERWRITE) {
535 		hashp->error = errno = EINVAL;
536 		return (ERROR);
537 	}
538 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
539 		hashp->error = errno = EPERM;
540 		return (ERROR);
541 	}
542 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
543 	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
544 }
545 
546 static int
547 hash_delete(const DB *dbp, const DBT *key,
548     u_int32_t flag)		/* Ignored */
549 {
550 	HTAB *hashp;
551 
552 	hashp = (HTAB *)dbp->internal;
553 	if (flag && flag != R_CURSOR) {
554 		hashp->error = errno = EINVAL;
555 		return (ERROR);
556 	}
557 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
558 		hashp->error = errno = EPERM;
559 		return (ERROR);
560 	}
561 	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
562 }
563 
564 /*
565  * Assume that hashp has been set in wrapper routine.
566  */
567 static int
568 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
569 {
570 	BUFHEAD *rbufp;
571 	BUFHEAD *bufp, *save_bufp;
572 	u_int16_t *bp;
573 	int n, ndx, off, size;
574 	char *kp;
575 	u_int16_t pageno;
576 
577 #ifdef HASH_STATISTICS
578 	hash_accesses++;
579 #endif
580 
581 	off = hashp->BSIZE;
582 	size = key->size;
583 	kp = (char *)key->data;
584 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
585 	if (!rbufp)
586 		return (ERROR);
587 	save_bufp = rbufp;
588 
589 	/* Pin the bucket chain */
590 	rbufp->flags |= BUF_PIN;
591 	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
592 		if (bp[1] >= REAL_KEY) {
593 			/* Real key/data pair */
594 			if (size == off - *bp &&
595 			    memcmp(kp, rbufp->page + *bp, size) == 0)
596 				goto found;
597 			off = bp[1];
598 #ifdef HASH_STATISTICS
599 			hash_collisions++;
600 #endif
601 			bp += 2;
602 			ndx += 2;
603 		} else if (bp[1] == OVFLPAGE) {
604 			rbufp = __get_buf(hashp, *bp, rbufp, 0);
605 			if (!rbufp) {
606 				save_bufp->flags &= ~BUF_PIN;
607 				return (ERROR);
608 			}
609 			/* FOR LOOP INIT */
610 			bp = (u_int16_t *)rbufp->page;
611 			n = *bp++;
612 			ndx = 1;
613 			off = hashp->BSIZE;
614 		} else if (bp[1] < REAL_KEY) {
615 			if ((ndx =
616 			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
617 				goto found;
618 			if (ndx == -2) {
619 				bufp = rbufp;
620 				if (!(pageno =
621 				    __find_last_page(hashp, &bufp))) {
622 					ndx = 0;
623 					rbufp = bufp;
624 					break;	/* FOR */
625 				}
626 				rbufp = __get_buf(hashp, pageno, bufp, 0);
627 				if (!rbufp) {
628 					save_bufp->flags &= ~BUF_PIN;
629 					return (ERROR);
630 				}
631 				/* FOR LOOP INIT */
632 				bp = (u_int16_t *)rbufp->page;
633 				n = *bp++;
634 				ndx = 1;
635 				off = hashp->BSIZE;
636 			} else {
637 				save_bufp->flags &= ~BUF_PIN;
638 				return (ERROR);
639 			}
640 		}
641 
642 	/* Not found */
643 	switch (action) {
644 	case HASH_PUT:
645 	case HASH_PUTNEW:
646 		if (__addel(hashp, rbufp, key, val)) {
647 			save_bufp->flags &= ~BUF_PIN;
648 			return (ERROR);
649 		} else {
650 			save_bufp->flags &= ~BUF_PIN;
651 			return (SUCCESS);
652 		}
653 	case HASH_GET:
654 	case HASH_DELETE:
655 	default:
656 		save_bufp->flags &= ~BUF_PIN;
657 		return (ABNORMAL);
658 	}
659 
660 found:
661 	switch (action) {
662 	case HASH_PUTNEW:
663 		save_bufp->flags &= ~BUF_PIN;
664 		return (ABNORMAL);
665 	case HASH_GET:
666 		bp = (u_int16_t *)rbufp->page;
667 		if (bp[ndx + 1] < REAL_KEY) {
668 			if (__big_return(hashp, rbufp, ndx, val, 0))
669 				return (ERROR);
670 		} else {
671 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
672 			val->size = bp[ndx] - bp[ndx + 1];
673 		}
674 		break;
675 	case HASH_PUT:
676 		if ((__delpair(hashp, rbufp, ndx)) ||
677 		    (__addel(hashp, rbufp, key, val))) {
678 			save_bufp->flags &= ~BUF_PIN;
679 			return (ERROR);
680 		}
681 		break;
682 	case HASH_DELETE:
683 		if (__delpair(hashp, rbufp, ndx))
684 			return (ERROR);
685 		break;
686 	default:
687 		abort();
688 	}
689 	save_bufp->flags &= ~BUF_PIN;
690 	return (SUCCESS);
691 }
692 
693 static int
694 hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t flag)
695 {
696 	u_int32_t bucket;
697 	BUFHEAD *bufp;
698 	HTAB *hashp;
699 	u_int16_t *bp, ndx;
700 
701 	hashp = (HTAB *)dbp->internal;
702 	if (flag && flag != R_FIRST && flag != R_NEXT) {
703 		hashp->error = errno = EINVAL;
704 		return (ERROR);
705 	}
706 #ifdef HASH_STATISTICS
707 	hash_accesses++;
708 #endif
709 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
710 		hashp->cbucket = 0;
711 		hashp->cndx = 1;
712 		hashp->cpage = NULL;
713 	}
714  next_bucket:
715 	for (bp = NULL; !bp || !bp[0]; ) {
716 		if (!(bufp = hashp->cpage)) {
717 			for (bucket = hashp->cbucket;
718 			    bucket <= hashp->MAX_BUCKET;
719 			    bucket++, hashp->cndx = 1) {
720 				bufp = __get_buf(hashp, bucket, NULL, 0);
721 				if (!bufp)
722 					return (ERROR);
723 				hashp->cpage = bufp;
724 				bp = (u_int16_t *)bufp->page;
725 				if (bp[0])
726 					break;
727 			}
728 			hashp->cbucket = bucket;
729 			if ((u_int32_t)hashp->cbucket > hashp->MAX_BUCKET) {
730 				hashp->cbucket = -1;
731 				return (ABNORMAL);
732 			}
733 		} else {
734 			bp = (u_int16_t *)hashp->cpage->page;
735 			if (flag == R_NEXT) {
736 				hashp->cndx += 2;
737 				if (hashp->cndx > bp[0]) {
738 					hashp->cpage = NULL;
739 					hashp->cbucket++;
740 					hashp->cndx = 1;
741 					goto next_bucket;
742 				}
743 			}
744 		}
745 
746 #ifdef DEBUG
747 		assert(bp);
748 		assert(bufp);
749 #endif
750 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
751 			bufp = hashp->cpage =
752 			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
753 			if (!bufp)
754 				return (ERROR);
755 			bp = (u_int16_t *)(bufp->page);
756 			hashp->cndx = 1;
757 		}
758 		if (!bp[0]) {
759 			hashp->cpage = NULL;
760 			++hashp->cbucket;
761 		}
762 	}
763 	ndx = hashp->cndx;
764 	if (bp[ndx + 1] < REAL_KEY) {
765 		if (__big_keydata(hashp, bufp, key, data, 1))
766 			return (ERROR);
767 	} else {
768 		if (hashp->cpage == 0)
769 			return (ERROR);
770 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
771 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
772 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
773 		data->size = bp[ndx] - bp[ndx + 1];
774 	}
775 	return (SUCCESS);
776 }
777 
778 /********************************* UTILITIES ************************/
779 
780 /*
781  * Returns:
782  *	 0 ==> OK
783  *	-1 ==> Error
784  */
785 int
786 __expand_table(HTAB *hashp)
787 {
788 	u_int32_t old_bucket, new_bucket;
789 	int dirsize, new_segnum, spare_ndx;
790 
791 #ifdef HASH_STATISTICS
792 	hash_expansions++;
793 #endif
794 	new_bucket = ++hashp->MAX_BUCKET;
795 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
796 
797 	new_segnum = new_bucket >> hashp->SSHIFT;
798 
799 	/* Check if we need a new segment */
800 	if (new_segnum >= hashp->nsegs) {
801 		/* Check if we need to expand directory */
802 		if (new_segnum >= hashp->DSIZE) {
803 			/* Reallocate directory */
804 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
805 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
806 				return (-1);
807 			hashp->DSIZE = dirsize << 1;
808 		}
809 		if ((hashp->dir[new_segnum] =
810 		    (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
811 			return (-1);
812 		hashp->exsegs++;
813 		hashp->nsegs++;
814 	}
815 	/*
816 	 * If the split point is increasing (MAX_BUCKET's log base 2
817 	 * * increases), we need to copy the current contents of the spare
818 	 * split bucket to the next bucket.
819 	 */
820 	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
821 	if (spare_ndx > hashp->OVFL_POINT) {
822 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
823 		hashp->OVFL_POINT = spare_ndx;
824 	}
825 
826 	if (new_bucket > hashp->HIGH_MASK) {
827 		/* Starting a new doubling */
828 		hashp->LOW_MASK = hashp->HIGH_MASK;
829 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
830 	}
831 	/* Relocate records to the new bucket */
832 	return (__split_page(hashp, old_bucket, new_bucket));
833 }
834 
835 /*
836  * If realloc guarantees that the pointer is not destroyed if the realloc
837  * fails, then this routine can go away.
838  */
839 static void *
840 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
841 {
842 	void *p;
843 
844 	if ( (p = malloc(newsize)) ) {
845 		memmove(p, *p_ptr, oldsize);
846 		memset((char *)p + oldsize, 0, newsize - oldsize);
847 		free(*p_ptr);
848 		*p_ptr = p;
849 	}
850 	return (p);
851 }
852 
853 u_int32_t
854 __call_hash(HTAB *hashp, char *k, int len)
855 {
856 	unsigned int n, bucket;
857 
858 	n = hashp->hash(k, len);
859 	bucket = n & hashp->HIGH_MASK;
860 	if (bucket > hashp->MAX_BUCKET)
861 		bucket = bucket & hashp->LOW_MASK;
862 	return (bucket);
863 }
864 
865 /*
866  * Allocate segment table.  On error, destroy the table and set errno.
867  *
868  * Returns 0 on success
869  */
870 static int
871 alloc_segs(HTAB *hashp, int nsegs)
872 {
873 	int i;
874 	SEGMENT store;
875 
876 	int save_errno;
877 
878 	if ((hashp->dir =
879 	    (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
880 		save_errno = errno;
881 		(void)hdestroy(hashp);
882 		errno = save_errno;
883 		return (-1);
884 	}
885 	hashp->nsegs = nsegs;
886 	if (nsegs == 0)
887 		return (0);
888 	/* Allocate segments */
889 	if ((store = (SEGMENT)calloc(nsegs << hashp->SSHIFT,
890 	    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  */
905 static void
906 swap_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 
933 static void
934 swap_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