xref: /freebsd/lib/libc/db/hash/hash.c (revision 8fc257994d0ce2396196d7a06d50d20c8015f4b7)
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 		if (hashp->BSIZE > MAX_BSIZE)
297 			hashp->BSIZE = MAX_BSIZE;
298 		hashp->BSHIFT = __log2(hashp->BSIZE);
299 	}
300 
301 	if (info) {
302 		if (info->bsize) {
303 			/* Round pagesize up to power of 2 */
304 			hashp->BSHIFT = __log2(info->bsize);
305 			hashp->BSIZE = 1 << hashp->BSHIFT;
306 			if (hashp->BSIZE > MAX_BSIZE) {
307 				errno = EINVAL;
308 				return (NULL);
309 			}
310 		}
311 		if (info->ffactor)
312 			hashp->FFACTOR = info->ffactor;
313 		if (info->hash)
314 			hashp->hash = info->hash;
315 		if (info->nelem)
316 			nelem = info->nelem;
317 		if (info->lorder) {
318 			if (info->lorder != BIG_ENDIAN &&
319 			    info->lorder != LITTLE_ENDIAN) {
320 				errno = EINVAL;
321 				return (NULL);
322 			}
323 			hashp->LORDER = info->lorder;
324 		}
325 	}
326 	/* init_htab should destroy the table and set errno if it fails */
327 	if (init_htab(hashp, nelem))
328 		return (NULL);
329 	else
330 		return (hashp);
331 }
332 /*
333  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
334  * the table and set errno, so we just pass the error information along.
335  *
336  * Returns 0 on No Error
337  */
338 static int
339 init_htab(HTAB *hashp, int nelem)
340 {
341 	int nbuckets, nsegs, l2;
342 
343 	/*
344 	 * Divide number of elements by the fill factor and determine a
345 	 * desired number of buckets.  Allocate space for the next greater
346 	 * power of two number of buckets.
347 	 */
348 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
349 
350 	l2 = __log2(MAX(nelem, 2));
351 	nbuckets = 1 << l2;
352 
353 	hashp->SPARES[l2] = l2 + 1;
354 	hashp->SPARES[l2 + 1] = l2 + 1;
355 	hashp->OVFL_POINT = l2;
356 	hashp->LAST_FREED = 2;
357 
358 	/* First bitmap page is at: splitpoint l2 page offset 1 */
359 	if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
360 		return (-1);
361 
362 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
363 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
364 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
365 	    hashp->BSHIFT) + 1;
366 
367 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
368 	nsegs = 1 << __log2(nsegs);
369 
370 	if (nsegs > hashp->DSIZE)
371 		hashp->DSIZE = nsegs;
372 	return (alloc_segs(hashp, nsegs));
373 }
374 
375 /********************** DESTROY/CLOSE ROUTINES ************************/
376 
377 /*
378  * Flushes any changes to the file if necessary and destroys the hashp
379  * structure, freeing all allocated space.
380  */
381 static int
382 hdestroy(HTAB *hashp)
383 {
384 	int i, save_errno;
385 
386 	save_errno = 0;
387 
388 #ifdef HASH_STATISTICS
389 	(void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
390 	    hash_accesses, hash_collisions);
391 	(void)fprintf(stderr, "hdestroy: expansions %ld\n",
392 	    hash_expansions);
393 	(void)fprintf(stderr, "hdestroy: overflows %ld\n",
394 	    hash_overflows);
395 	(void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
396 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
397 
398 	for (i = 0; i < NCACHED; i++)
399 		(void)fprintf(stderr,
400 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
401 #endif
402 	/*
403 	 * Call on buffer manager to free buffers, and if required,
404 	 * write them to disk.
405 	 */
406 	if (__buf_free(hashp, 1, hashp->save_file))
407 		save_errno = errno;
408 	if (hashp->dir) {
409 		free(*hashp->dir);	/* Free initial segments */
410 		/* Free extra segments */
411 		while (hashp->exsegs--)
412 			free(hashp->dir[--hashp->nsegs]);
413 		free(hashp->dir);
414 	}
415 	if (flush_meta(hashp) && !save_errno)
416 		save_errno = errno;
417 	/* Free Bigmaps */
418 	for (i = 0; i < hashp->nmaps; i++)
419 		if (hashp->mapp[i])
420 			free(hashp->mapp[i]);
421 	if (hashp->tmp_key)
422 		free(hashp->tmp_key);
423 	if (hashp->tmp_buf)
424 		free(hashp->tmp_buf);
425 
426 	if (hashp->fp != -1)
427 		(void)_close(hashp->fp);
428 
429 	free(hashp);
430 
431 	if (save_errno) {
432 		errno = save_errno;
433 		return (ERROR);
434 	}
435 	return (SUCCESS);
436 }
437 /*
438  * Write modified pages to disk
439  *
440  * Returns:
441  *	 0 == OK
442  *	-1 ERROR
443  */
444 static int
445 hash_sync(const DB *dbp, u_int32_t flags)
446 {
447 	HTAB *hashp;
448 
449 	if (flags != 0) {
450 		errno = EINVAL;
451 		return (ERROR);
452 	}
453 
454 	if (!dbp)
455 		return (ERROR);
456 
457 	hashp = (HTAB *)dbp->internal;
458 	if (!hashp->save_file)
459 		return (0);
460 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
461 		return (ERROR);
462 	hashp->new_file = 0;
463 	return (0);
464 }
465 
466 /*
467  * Returns:
468  *	 0 == OK
469  *	-1 indicates that errno should be set
470  */
471 static int
472 flush_meta(HTAB *hashp)
473 {
474 	HASHHDR *whdrp;
475 #if BYTE_ORDER == LITTLE_ENDIAN
476 	HASHHDR whdr;
477 #endif
478 	int fp, i, wsize;
479 
480 	if (!hashp->save_file)
481 		return (0);
482 	hashp->MAGIC = HASHMAGIC;
483 	hashp->VERSION = HASHVERSION;
484 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
485 
486 	fp = hashp->fp;
487 	whdrp = &hashp->hdr;
488 #if BYTE_ORDER == LITTLE_ENDIAN
489 	whdrp = &whdr;
490 	swap_header_copy(&hashp->hdr, whdrp);
491 #endif
492 	if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
493 		return (-1);
494 	else
495 		if (wsize != sizeof(HASHHDR)) {
496 			errno = EFTYPE;
497 			hashp->error = errno;
498 			return (-1);
499 		}
500 	for (i = 0; i < NCACHED; i++)
501 		if (hashp->mapp[i])
502 			if (__put_page(hashp, (char *)hashp->mapp[i],
503 				hashp->BITMAPS[i], 0, 1))
504 				return (-1);
505 	return (0);
506 }
507 
508 /*******************************SEARCH ROUTINES *****************************/
509 /*
510  * All the access routines return
511  *
512  * Returns:
513  *	 0 on SUCCESS
514  *	 1 to indicate an external ERROR (i.e. key not found, etc)
515  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
516  */
517 static int
518 hash_get(const DB *dbp, const DBT *key, DBT *data, u_int32_t flag)
519 {
520 	HTAB *hashp;
521 
522 	hashp = (HTAB *)dbp->internal;
523 	if (flag) {
524 		hashp->error = errno = EINVAL;
525 		return (ERROR);
526 	}
527 	return (hash_access(hashp, HASH_GET, (DBT *)key, data));
528 }
529 
530 static int
531 hash_put(const DB *dbp, DBT *key, const DBT *data, u_int32_t flag)
532 {
533 	HTAB *hashp;
534 
535 	hashp = (HTAB *)dbp->internal;
536 	if (flag && flag != R_NOOVERWRITE) {
537 		hashp->error = errno = EINVAL;
538 		return (ERROR);
539 	}
540 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
541 		hashp->error = errno = EPERM;
542 		return (ERROR);
543 	}
544 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
545 	    HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
546 }
547 
548 static int
549 hash_delete(const DB *dbp, const DBT *key,
550     u_int32_t flag)		/* Ignored */
551 {
552 	HTAB *hashp;
553 
554 	hashp = (HTAB *)dbp->internal;
555 	if (flag && flag != R_CURSOR) {
556 		hashp->error = errno = EINVAL;
557 		return (ERROR);
558 	}
559 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
560 		hashp->error = errno = EPERM;
561 		return (ERROR);
562 	}
563 	return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
564 }
565 
566 /*
567  * Assume that hashp has been set in wrapper routine.
568  */
569 static int
570 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
571 {
572 	BUFHEAD *rbufp;
573 	BUFHEAD *bufp, *save_bufp;
574 	u_int16_t *bp;
575 	int n, ndx, off, size;
576 	char *kp;
577 	u_int16_t pageno;
578 
579 #ifdef HASH_STATISTICS
580 	hash_accesses++;
581 #endif
582 
583 	off = hashp->BSIZE;
584 	size = key->size;
585 	kp = (char *)key->data;
586 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
587 	if (!rbufp)
588 		return (ERROR);
589 	save_bufp = rbufp;
590 
591 	/* Pin the bucket chain */
592 	rbufp->flags |= BUF_PIN;
593 	for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
594 		if (bp[1] >= REAL_KEY) {
595 			/* Real key/data pair */
596 			if (size == off - *bp &&
597 			    memcmp(kp, rbufp->page + *bp, size) == 0)
598 				goto found;
599 			off = bp[1];
600 #ifdef HASH_STATISTICS
601 			hash_collisions++;
602 #endif
603 			bp += 2;
604 			ndx += 2;
605 		} else if (bp[1] == OVFLPAGE) {
606 			rbufp = __get_buf(hashp, *bp, rbufp, 0);
607 			if (!rbufp) {
608 				save_bufp->flags &= ~BUF_PIN;
609 				return (ERROR);
610 			}
611 			/* FOR LOOP INIT */
612 			bp = (u_int16_t *)rbufp->page;
613 			n = *bp++;
614 			ndx = 1;
615 			off = hashp->BSIZE;
616 		} else if (bp[1] < REAL_KEY) {
617 			if ((ndx =
618 			    __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
619 				goto found;
620 			if (ndx == -2) {
621 				bufp = rbufp;
622 				if (!(pageno =
623 				    __find_last_page(hashp, &bufp))) {
624 					ndx = 0;
625 					rbufp = bufp;
626 					break;	/* FOR */
627 				}
628 				rbufp = __get_buf(hashp, pageno, bufp, 0);
629 				if (!rbufp) {
630 					save_bufp->flags &= ~BUF_PIN;
631 					return (ERROR);
632 				}
633 				/* FOR LOOP INIT */
634 				bp = (u_int16_t *)rbufp->page;
635 				n = *bp++;
636 				ndx = 1;
637 				off = hashp->BSIZE;
638 			} else {
639 				save_bufp->flags &= ~BUF_PIN;
640 				return (ERROR);
641 			}
642 		}
643 
644 	/* Not found */
645 	switch (action) {
646 	case HASH_PUT:
647 	case HASH_PUTNEW:
648 		if (__addel(hashp, rbufp, key, val)) {
649 			save_bufp->flags &= ~BUF_PIN;
650 			return (ERROR);
651 		} else {
652 			save_bufp->flags &= ~BUF_PIN;
653 			return (SUCCESS);
654 		}
655 	case HASH_GET:
656 	case HASH_DELETE:
657 	default:
658 		save_bufp->flags &= ~BUF_PIN;
659 		return (ABNORMAL);
660 	}
661 
662 found:
663 	switch (action) {
664 	case HASH_PUTNEW:
665 		save_bufp->flags &= ~BUF_PIN;
666 		return (ABNORMAL);
667 	case HASH_GET:
668 		bp = (u_int16_t *)rbufp->page;
669 		if (bp[ndx + 1] < REAL_KEY) {
670 			if (__big_return(hashp, rbufp, ndx, val, 0))
671 				return (ERROR);
672 		} else {
673 			val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
674 			val->size = bp[ndx] - bp[ndx + 1];
675 		}
676 		break;
677 	case HASH_PUT:
678 		if ((__delpair(hashp, rbufp, ndx)) ||
679 		    (__addel(hashp, rbufp, key, val))) {
680 			save_bufp->flags &= ~BUF_PIN;
681 			return (ERROR);
682 		}
683 		break;
684 	case HASH_DELETE:
685 		if (__delpair(hashp, rbufp, ndx))
686 			return (ERROR);
687 		break;
688 	default:
689 		abort();
690 	}
691 	save_bufp->flags &= ~BUF_PIN;
692 	return (SUCCESS);
693 }
694 
695 static int
696 hash_seq(const DB *dbp, DBT *key, DBT *data, u_int32_t flag)
697 {
698 	u_int32_t bucket;
699 	BUFHEAD *bufp;
700 	HTAB *hashp;
701 	u_int16_t *bp, ndx;
702 
703 	hashp = (HTAB *)dbp->internal;
704 	if (flag && flag != R_FIRST && flag != R_NEXT) {
705 		hashp->error = errno = EINVAL;
706 		return (ERROR);
707 	}
708 #ifdef HASH_STATISTICS
709 	hash_accesses++;
710 #endif
711 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
712 		hashp->cbucket = 0;
713 		hashp->cndx = 1;
714 		hashp->cpage = NULL;
715 	}
716 next_bucket:
717 	for (bp = NULL; !bp || !bp[0]; ) {
718 		if (!(bufp = hashp->cpage)) {
719 			for (bucket = hashp->cbucket;
720 			    bucket <= hashp->MAX_BUCKET;
721 			    bucket++, hashp->cndx = 1) {
722 				bufp = __get_buf(hashp, bucket, NULL, 0);
723 				if (!bufp)
724 					return (ERROR);
725 				hashp->cpage = bufp;
726 				bp = (u_int16_t *)bufp->page;
727 				if (bp[0])
728 					break;
729 			}
730 			hashp->cbucket = bucket;
731 			if ((u_int32_t)hashp->cbucket > hashp->MAX_BUCKET) {
732 				hashp->cbucket = -1;
733 				return (ABNORMAL);
734 			}
735 		} else {
736 			bp = (u_int16_t *)hashp->cpage->page;
737 			if (flag == R_NEXT || flag == 0) {
738 				hashp->cndx += 2;
739 				if (hashp->cndx > bp[0]) {
740 					hashp->cpage = NULL;
741 					hashp->cbucket++;
742 					hashp->cndx = 1;
743 					goto next_bucket;
744 				}
745 			}
746 		}
747 
748 #ifdef DEBUG
749 		assert(bp);
750 		assert(bufp);
751 #endif
752 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
753 			bufp = hashp->cpage =
754 			    __get_buf(hashp, bp[hashp->cndx], bufp, 0);
755 			if (!bufp)
756 				return (ERROR);
757 			bp = (u_int16_t *)(bufp->page);
758 			hashp->cndx = 1;
759 		}
760 		if (!bp[0]) {
761 			hashp->cpage = NULL;
762 			++hashp->cbucket;
763 		}
764 	}
765 	ndx = hashp->cndx;
766 	if (bp[ndx + 1] < REAL_KEY) {
767 		if (__big_keydata(hashp, bufp, key, data, 1))
768 			return (ERROR);
769 	} else {
770 		if (hashp->cpage == 0)
771 			return (ERROR);
772 		key->data = (u_char *)hashp->cpage->page + bp[ndx];
773 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
774 		data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
775 		data->size = bp[ndx] - bp[ndx + 1];
776 	}
777 	return (SUCCESS);
778 }
779 
780 /********************************* UTILITIES ************************/
781 
782 /*
783  * Returns:
784  *	 0 ==> OK
785  *	-1 ==> Error
786  */
787 int
788 __expand_table(HTAB *hashp)
789 {
790 	u_int32_t old_bucket, new_bucket;
791 	int dirsize, new_segnum, spare_ndx;
792 
793 #ifdef HASH_STATISTICS
794 	hash_expansions++;
795 #endif
796 	new_bucket = ++hashp->MAX_BUCKET;
797 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
798 
799 	new_segnum = new_bucket >> hashp->SSHIFT;
800 
801 	/* Check if we need a new segment */
802 	if (new_segnum >= hashp->nsegs) {
803 		/* Check if we need to expand directory */
804 		if (new_segnum >= hashp->DSIZE) {
805 			/* Reallocate directory */
806 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
807 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
808 				return (-1);
809 			hashp->DSIZE = dirsize << 1;
810 		}
811 		if ((hashp->dir[new_segnum] =
812 		    (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
813 			return (-1);
814 		hashp->exsegs++;
815 		hashp->nsegs++;
816 	}
817 	/*
818 	 * If the split point is increasing (MAX_BUCKET's log base 2
819 	 * * increases), we need to copy the current contents of the spare
820 	 * split bucket to the next bucket.
821 	 */
822 	spare_ndx = __log2(hashp->MAX_BUCKET + 1);
823 	if (spare_ndx > hashp->OVFL_POINT) {
824 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
825 		hashp->OVFL_POINT = spare_ndx;
826 	}
827 
828 	if (new_bucket > hashp->HIGH_MASK) {
829 		/* Starting a new doubling */
830 		hashp->LOW_MASK = hashp->HIGH_MASK;
831 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
832 	}
833 	/* Relocate records to the new bucket */
834 	return (__split_page(hashp, old_bucket, new_bucket));
835 }
836 
837 /*
838  * If realloc guarantees that the pointer is not destroyed if the realloc
839  * fails, then this routine can go away.
840  */
841 static void *
842 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
843 {
844 	void *p;
845 
846 	if ( (p = malloc(newsize)) ) {
847 		memmove(p, *p_ptr, oldsize);
848 		memset((char *)p + oldsize, 0, newsize - oldsize);
849 		free(*p_ptr);
850 		*p_ptr = p;
851 	}
852 	return (p);
853 }
854 
855 u_int32_t
856 __call_hash(HTAB *hashp, char *k, int len)
857 {
858 	unsigned int n, bucket;
859 
860 	n = hashp->hash(k, len);
861 	bucket = n & hashp->HIGH_MASK;
862 	if (bucket > hashp->MAX_BUCKET)
863 		bucket = bucket & hashp->LOW_MASK;
864 	return (bucket);
865 }
866 
867 /*
868  * Allocate segment table.  On error, destroy the table and set errno.
869  *
870  * Returns 0 on success
871  */
872 static int
873 alloc_segs(HTAB *hashp, int nsegs)
874 {
875 	int i;
876 	SEGMENT store;
877 
878 	int save_errno;
879 
880 	if ((hashp->dir =
881 	    (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
882 		save_errno = errno;
883 		(void)hdestroy(hashp);
884 		errno = save_errno;
885 		return (-1);
886 	}
887 	hashp->nsegs = nsegs;
888 	if (nsegs == 0)
889 		return (0);
890 	/* Allocate segments */
891 	if ((store = (SEGMENT)calloc(nsegs << hashp->SSHIFT,
892 	    sizeof(SEGMENT))) == NULL) {
893 		save_errno = errno;
894 		(void)hdestroy(hashp);
895 		errno = save_errno;
896 		return (-1);
897 	}
898 	for (i = 0; i < nsegs; i++)
899 		hashp->dir[i] = &store[i << hashp->SSHIFT];
900 	return (0);
901 }
902 
903 #if BYTE_ORDER == LITTLE_ENDIAN
904 /*
905  * Hashp->hdr needs to be byteswapped.
906  */
907 static void
908 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
909 {
910 	int i;
911 
912 	P_32_COPY(srcp->magic, destp->magic);
913 	P_32_COPY(srcp->version, destp->version);
914 	P_32_COPY(srcp->lorder, destp->lorder);
915 	P_32_COPY(srcp->bsize, destp->bsize);
916 	P_32_COPY(srcp->bshift, destp->bshift);
917 	P_32_COPY(srcp->dsize, destp->dsize);
918 	P_32_COPY(srcp->ssize, destp->ssize);
919 	P_32_COPY(srcp->sshift, destp->sshift);
920 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
921 	P_32_COPY(srcp->last_freed, destp->last_freed);
922 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
923 	P_32_COPY(srcp->high_mask, destp->high_mask);
924 	P_32_COPY(srcp->low_mask, destp->low_mask);
925 	P_32_COPY(srcp->ffactor, destp->ffactor);
926 	P_32_COPY(srcp->nkeys, destp->nkeys);
927 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
928 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
929 	for (i = 0; i < NCACHED; i++) {
930 		P_32_COPY(srcp->spares[i], destp->spares[i]);
931 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
932 	}
933 }
934 
935 static void
936 swap_header(HTAB *hashp)
937 {
938 	HASHHDR *hdrp;
939 	int i;
940 
941 	hdrp = &hashp->hdr;
942 
943 	M_32_SWAP(hdrp->magic);
944 	M_32_SWAP(hdrp->version);
945 	M_32_SWAP(hdrp->lorder);
946 	M_32_SWAP(hdrp->bsize);
947 	M_32_SWAP(hdrp->bshift);
948 	M_32_SWAP(hdrp->dsize);
949 	M_32_SWAP(hdrp->ssize);
950 	M_32_SWAP(hdrp->sshift);
951 	M_32_SWAP(hdrp->ovfl_point);
952 	M_32_SWAP(hdrp->last_freed);
953 	M_32_SWAP(hdrp->max_bucket);
954 	M_32_SWAP(hdrp->high_mask);
955 	M_32_SWAP(hdrp->low_mask);
956 	M_32_SWAP(hdrp->ffactor);
957 	M_32_SWAP(hdrp->nkeys);
958 	M_32_SWAP(hdrp->hdrpages);
959 	M_32_SWAP(hdrp->h_charkey);
960 	for (i = 0; i < NCACHED; i++) {
961 		M_32_SWAP(hdrp->spares[i]);
962 		M_16_SWAP(hdrp->bitmaps[i]);
963 	}
964 }
965 #endif
966