xref: /minix/lib/libc/db/hash/hash.c (revision 0a6a1f1d)
1 /*	$NetBSD: hash.c,v 1.35 2015/06/22 21:16:02 christos Exp $	*/
2 
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 #if HAVE_NBTOOL_CONFIG_H
36 #include "nbtool_config.h"
37 #endif
38 
39 #include <sys/cdefs.h>
40 __RCSID("$NetBSD: hash.c,v 1.35 2015/06/22 21:16:02 christos Exp $");
41 
42 #include "namespace.h"
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 #include <assert.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 *, uint32_t);
64 static int   hash_fd(const DB *);
65 static int   hash_get(const DB *, const DBT *, DBT *, uint32_t);
66 static int   hash_put(const DB *, DBT *, const DBT *, uint32_t);
67 static void *hash_realloc(SEGMENT **, size_t, size_t);
68 static int   hash_seq(const DB *, DBT *, DBT *, uint32_t);
69 static int   hash_sync(const DB *, uint32_t);
70 static int   hdestroy(HTAB *);
71 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
72 static int   init_htab(HTAB *, size_t);
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 *
__hash_open(const char * file,int flags,mode_t mode,const HASHINFO * info,int dflags)97 __hash_open(const char *file, int flags, mode_t mode, const HASHINFO *info,
98     int dflags)
99 {
100 	HTAB *hashp;
101 	struct stat statbuf;
102 	DB *dbp;
103 	int bpages, new_table, nsegs, save_errno;
104 	ssize_t hdrsize;
105 
106 	if ((flags & O_ACCMODE) == O_WRONLY) {
107 		errno = EINVAL;
108 		return (NULL);
109 	}
110 
111 	if (!(hashp = 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 	new_table = 0;
124 	if (!file || (flags & O_TRUNC) ||
125 	    (stat(file, &statbuf) && (errno == ENOENT))) {
126 		if (errno == ENOENT)
127 			errno = 0; /* Just in case someone looks at errno */
128 		new_table = 1;
129 	}
130 	if (file) {
131 		if ((hashp->fp = __dbopen(file, flags, mode, &statbuf)) == -1)
132 			RETURN_ERROR(errno, error0);
133 		new_table |= statbuf.st_size == 0;
134 	}
135 	if (new_table) {
136 		if (!(hashp = init_hash(hashp, file, info)))
137 			RETURN_ERROR(errno, error1);
138 	} else {
139 		/* Table already exists */
140 		if (info && info->hash)
141 			hashp->hash = info->hash;
142 		else
143 			hashp->hash = __default_hash;
144 
145 		hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
146 #if BYTE_ORDER == LITTLE_ENDIAN
147 		swap_header(hashp);
148 #endif
149 		if (hdrsize == -1)
150 			RETURN_ERROR(errno, error1);
151 		if (hdrsize != sizeof(HASHHDR))
152 			RETURN_ERROR(EFTYPE, error1);
153 		/* Verify file type, versions and hash function */
154 		if (hashp->MAGIC != HASHMAGIC)
155 			RETURN_ERROR(EFTYPE, error1);
156 #define	OLDHASHVERSION	1
157 		if (hashp->VERSION != HASHVERSION &&
158 		    hashp->VERSION != OLDHASHVERSION)
159 			RETURN_ERROR(EFTYPE, error1);
160 		if (hashp->hash(CHARKEY, sizeof(CHARKEY)) !=
161 		    (uint32_t)hashp->H_CHARKEY)
162 			RETURN_ERROR(EFTYPE, error1);
163 		/*
164 		 * Figure out how many segments we need.  Max_Bucket is the
165 		 * maximum bucket number, so the number of buckets is
166 		 * max_bucket + 1.
167 		 */
168 		nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
169 			 hashp->SGSIZE;
170 		hashp->nsegs = 0;
171 		if (alloc_segs(hashp, nsegs))
172 			/*
173 			 * If alloc_segs fails, table will have been destroyed
174 			 * and errno will have been set.
175 			 */
176 			return (NULL);
177 		/* Read in bitmaps */
178 		bpages = (hashp->SPARES[hashp->OVFL_POINT] +
179 		    (unsigned int)(hashp->BSIZE << BYTE_SHIFT) - 1) >>
180 		    (hashp->BSHIFT + BYTE_SHIFT);
181 
182 		hashp->nmaps = bpages;
183 		(void)memset(&hashp->mapp[0], 0, bpages * sizeof(uint32_t *));
184 	}
185 
186 	/* Initialize Buffer Manager */
187 	if (info && info->cachesize)
188 		__buf_init(hashp, info->cachesize);
189 	else
190 		__buf_init(hashp, DEF_BUFSIZE);
191 
192 	hashp->new_file = new_table;
193 	hashp->save_file = file && (hashp->flags & O_RDWR);
194 	hashp->cbucket = -1;
195 	if (!(dbp = malloc(sizeof(*dbp)))) {
196 		save_errno = errno;
197 		hdestroy(hashp);
198 		errno = save_errno;
199 		return (NULL);
200 	}
201 	dbp->internal = hashp;
202 	dbp->close = hash_close;
203 	dbp->del = hash_delete;
204 	dbp->fd = hash_fd;
205 	dbp->get = hash_get;
206 	dbp->put = hash_put;
207 	dbp->seq = hash_seq;
208 	dbp->sync = hash_sync;
209 	dbp->type = DB_HASH;
210 
211 #ifdef DEBUG1
212 	(void)fprintf(stderr,
213 "%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",
214 	    "init_htab:",
215 	    "TABLE POINTER   ", hashp,
216 	    "BUCKET SIZE     ", hashp->BSIZE,
217 	    "BUCKET SHIFT    ", hashp->BSHIFT,
218 	    "DIRECTORY SIZE  ", hashp->DSIZE,
219 	    "SEGMENT SIZE    ", hashp->SGSIZE,
220 	    "SEGMENT SHIFT   ", hashp->SSHIFT,
221 	    "FILL FACTOR     ", hashp->FFACTOR,
222 	    "MAX BUCKET      ", hashp->MAX_BUCKET,
223 	    "OVFL POINT	     ", hashp->OVFL_POINT,
224 	    "LAST FREED      ", hashp->LAST_FREED,
225 	    "HIGH MASK       ", hashp->HIGH_MASK,
226 	    "LOW  MASK       ", hashp->LOW_MASK,
227 	    "NSEGS           ", hashp->nsegs,
228 	    "NKEYS           ", hashp->NKEYS);
229 #endif
230 #ifdef HASH_STATISTICS
231 	hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
232 #endif
233 	return (dbp);
234 
235 error1:
236 	if (hashp != NULL)
237 		(void)close(hashp->fp);
238 
239 error0:
240 	free(hashp);
241 	errno = save_errno;
242 	return (NULL);
243 }
244 
245 static int
hash_close(DB * dbp)246 hash_close(DB *dbp)
247 {
248 	HTAB *hashp;
249 	int retval;
250 
251 	if (!dbp)
252 		return (ERROR);
253 
254 	hashp = dbp->internal;
255 	retval = hdestroy(hashp);
256 	free(dbp);
257 	return (retval);
258 }
259 
260 static int
hash_fd(const DB * dbp)261 hash_fd(const DB *dbp)
262 {
263 	HTAB *hashp;
264 
265 	if (!dbp)
266 		return (ERROR);
267 
268 	hashp = dbp->internal;
269 	if (hashp->fp == -1) {
270 		errno = ENOENT;
271 		return (-1);
272 	}
273 	return (hashp->fp);
274 }
275 
276 /************************** LOCAL CREATION ROUTINES **********************/
277 static HTAB *
init_hash(HTAB * hashp,const char * file,const HASHINFO * info)278 init_hash(HTAB *hashp, const char *file, const HASHINFO *info)
279 {
280 	struct stat statbuf;
281 	int nelem;
282 
283 	nelem = 1;
284 	hashp->NKEYS = 0;
285 	hashp->LORDER = BYTE_ORDER;
286 	hashp->BSIZE = DEF_BUCKET_SIZE;
287 	hashp->BSHIFT = DEF_BUCKET_SHIFT;
288 	hashp->SGSIZE = DEF_SEGSIZE;
289 	hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
290 	hashp->DSIZE = DEF_DIRSIZE;
291 	hashp->FFACTOR = DEF_FFACTOR;
292 	hashp->hash = __default_hash;
293 	memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
294 	memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
295 
296 	/* Fix bucket size to be optimal for file system */
297 	if (file != NULL) {
298 		if (stat(file, &statbuf))
299 			return (NULL);
300 		hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
301 		hashp->BSHIFT = __log2((uint32_t)hashp->BSIZE);
302 	}
303 
304 	if (info) {
305 		if (info->bsize) {
306 			/* Round pagesize up to power of 2 */
307 			hashp->BSHIFT = __log2(info->bsize);
308 			hashp->BSIZE = 1 << hashp->BSHIFT;
309 			if (hashp->BSIZE > MAX_BSIZE) {
310 				errno = EINVAL;
311 				return (NULL);
312 			}
313 		}
314 		if (info->ffactor)
315 			hashp->FFACTOR = info->ffactor;
316 		if (info->hash)
317 			hashp->hash = info->hash;
318 		if (info->nelem)
319 			nelem = info->nelem;
320 		if (info->lorder) {
321 			if (info->lorder != BIG_ENDIAN &&
322 			    info->lorder != LITTLE_ENDIAN) {
323 				errno = EINVAL;
324 				return (NULL);
325 			}
326 			hashp->LORDER = info->lorder;
327 		}
328 	}
329 	/* init_htab should destroy the table and set errno if it fails */
330 	if (init_htab(hashp, (size_t)nelem))
331 		return (NULL);
332 	else
333 		return (hashp);
334 }
335 /*
336  * This calls alloc_segs which may run out of memory.  Alloc_segs will destroy
337  * the table and set errno, so we just pass the error information along.
338  *
339  * Returns 0 on No Error
340  */
341 static int
init_htab(HTAB * hashp,size_t nelem)342 init_htab(HTAB *hashp, size_t nelem)
343 {
344 	int nbuckets;
345 	uint32_t nsegs;
346 	int l2;
347 
348 	/*
349 	 * Divide number of elements by the fill factor and determine a
350 	 * desired number of buckets.  Allocate space for the next greater
351 	 * power of two number of buckets.
352 	 */
353 	nelem = (nelem - 1) / hashp->FFACTOR + 1;
354 
355 	_DBFIT(nelem, uint32_t);
356 	l2 = __log2(MAX((uint32_t)nelem, 2));
357 	nbuckets = 1 << l2;
358 
359 	hashp->SPARES[l2] = l2 + 1;
360 	hashp->SPARES[l2 + 1] = l2 + 1;
361 	hashp->OVFL_POINT = l2;
362 	hashp->LAST_FREED = 2;
363 
364 	/* First bitmap page is at: splitpoint l2 page offset 1 */
365 	if (__ibitmap(hashp, (int)OADDR_OF(l2, 1), l2 + 1, 0))
366 		return (-1);
367 
368 	hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
369 	hashp->HIGH_MASK = (nbuckets << 1) - 1;
370 	/* LINTED constant in conditional context */
371 	hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
372 	    hashp->BSHIFT) + 1;
373 
374 	nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
375 	nsegs = 1 << __log2(nsegs);
376 
377 	if (nsegs > (uint32_t)hashp->DSIZE)
378 		hashp->DSIZE = nsegs;
379 	return (alloc_segs(hashp, (int)nsegs));
380 }
381 
382 /********************** DESTROY/CLOSE ROUTINES ************************/
383 
384 /*
385  * Flushes any changes to the file if necessary and destroys the hashp
386  * structure, freeing all allocated space.
387  */
388 static int
hdestroy(HTAB * hashp)389 hdestroy(HTAB *hashp)
390 {
391 	int i, save_errno;
392 
393 	save_errno = 0;
394 
395 #ifdef HASH_STATISTICS
396 	(void)fprintf(stderr, "hdestroy: accesses %d collisions %d\n",
397 	    hash_accesses, hash_collisions);
398 	(void)fprintf(stderr, "hdestroy: expansions %d\n",
399 	    hash_expansions);
400 	(void)fprintf(stderr, "hdestroy: overflows %d\n",
401 	    hash_overflows);
402 	(void)fprintf(stderr, "keys %d maxp %d segmentcount %d\n",
403 	    hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
404 
405 	for (i = 0; i < NCACHED; i++)
406 		(void)fprintf(stderr,
407 		    "spares[%d] = %d\n", i, hashp->SPARES[i]);
408 #endif
409 	/*
410 	 * Call on buffer manager to free buffers, and if required,
411 	 * write them to disk.
412 	 */
413 	if (__buf_free(hashp, 1, hashp->save_file))
414 		save_errno = errno;
415 	if (hashp->dir) {
416 		free(*hashp->dir);	/* Free initial segments */
417 		/* Free extra segments */
418 		while (hashp->exsegs--)
419 			free(hashp->dir[--hashp->nsegs]);
420 		free(hashp->dir);
421 	}
422 	if (flush_meta(hashp) && !save_errno)
423 		save_errno = errno;
424 	/* Free Bigmaps */
425 	for (i = 0; i < hashp->nmaps; i++)
426 		if (hashp->mapp[i])
427 			free(hashp->mapp[i]);
428 
429 	if (hashp->fp != -1)
430 		(void)close(hashp->fp);
431 
432 	free(hashp);
433 
434 	if (save_errno) {
435 		errno = save_errno;
436 		return (ERROR);
437 	}
438 	return (SUCCESS);
439 }
440 /*
441  * Write modified pages to disk
442  *
443  * Returns:
444  *	 0 == OK
445  *	-1 ERROR
446  */
447 static int
hash_sync(const DB * dbp,uint32_t flags)448 hash_sync(const DB *dbp, uint32_t flags)
449 {
450 	HTAB *hashp;
451 
452 	if (flags != 0) {
453 		errno = EINVAL;
454 		return (ERROR);
455 	}
456 
457 	if (!dbp)
458 		return (ERROR);
459 
460 	hashp = dbp->internal;
461 	if (!hashp->save_file)
462 		return (0);
463 	if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
464 		return (ERROR);
465 	hashp->new_file = 0;
466 	return (0);
467 }
468 
469 /*
470  * Returns:
471  *	 0 == OK
472  *	-1 indicates that errno should be set
473  */
474 static int
flush_meta(HTAB * hashp)475 flush_meta(HTAB *hashp)
476 {
477 	HASHHDR *whdrp;
478 #if BYTE_ORDER == LITTLE_ENDIAN
479 	HASHHDR whdr;
480 #endif
481 	int fp, i;
482 	ssize_t wsize;
483 
484 	if (!hashp->save_file)
485 		return (0);
486 	hashp->MAGIC = HASHMAGIC;
487 	hashp->VERSION = HASHVERSION;
488 	hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
489 
490 	fp = hashp->fp;
491 	whdrp = &hashp->hdr;
492 #if BYTE_ORDER == LITTLE_ENDIAN
493 	whdrp = &whdr;
494 	swap_header_copy(&hashp->hdr, whdrp);
495 #endif
496 	if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
497 		return (-1);
498 	else
499 		if (wsize != sizeof(HASHHDR)) {
500 			errno = EFTYPE;
501 			hashp->err = errno;
502 			return (-1);
503 		}
504 	for (i = 0; i < NCACHED; i++)
505 		if (hashp->mapp[i])
506 			if (__put_page(hashp, (char *)(void *)hashp->mapp[i],
507 				(u_int)hashp->BITMAPS[i], 0, 1))
508 				return (-1);
509 	return (0);
510 }
511 
512 /*******************************SEARCH ROUTINES *****************************/
513 /*
514  * All the access routines return
515  *
516  * Returns:
517  *	 0 on SUCCESS
518  *	 1 to indicate an external ERROR (i.e. key not found, etc)
519  *	-1 to indicate an internal ERROR (i.e. out of memory, etc)
520  */
521 static int
hash_get(const DB * dbp,const DBT * key,DBT * data,uint32_t flag)522 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag)
523 {
524 	HTAB *hashp;
525 
526 	hashp = dbp->internal;
527 	if (flag) {
528 		hashp->err = errno = EINVAL;
529 		return (ERROR);
530 	}
531 	return (hash_access(hashp, HASH_GET, __UNCONST(key), data));
532 }
533 
534 static int
hash_put(const DB * dbp,DBT * key,const DBT * data,uint32_t flag)535 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag)
536 {
537 	HTAB *hashp;
538 
539 	hashp = dbp->internal;
540 	if (flag && flag != R_NOOVERWRITE) {
541 		hashp->err = errno = EINVAL;
542 		return (ERROR);
543 	}
544 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
545 		hashp->err = errno = EPERM;
546 		return (ERROR);
547 	}
548 	/* LINTED const castaway */
549 	return (hash_access(hashp, flag == R_NOOVERWRITE ?
550 	    HASH_PUTNEW : HASH_PUT, __UNCONST(key), __UNCONST(data)));
551 }
552 
553 static int
hash_delete(const DB * dbp,const DBT * key,uint32_t flag)554 hash_delete(const DB *dbp, const DBT *key, uint32_t flag)
555 {
556 	HTAB *hashp;
557 
558 	hashp = dbp->internal;
559 	if (flag && flag != R_CURSOR) {
560 		hashp->err = errno = EINVAL;
561 		return (ERROR);
562 	}
563 	if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
564 		hashp->err = errno = EPERM;
565 		return (ERROR);
566 	}
567 	return hash_access(hashp, HASH_DELETE, __UNCONST(key), NULL);
568 }
569 
570 /*
571  * Assume that hashp has been set in wrapper routine.
572  */
573 static int
hash_access(HTAB * hashp,ACTION action,DBT * key,DBT * val)574 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
575 {
576 	BUFHEAD *rbufp;
577 	BUFHEAD *bufp, *save_bufp;
578 	uint16_t *bp;
579 	int n, ndx, off;
580 	size_t size;
581 	char *kp;
582 	uint16_t pageno;
583 
584 #ifdef HASH_STATISTICS
585 	hash_accesses++;
586 #endif
587 
588 	off = hashp->BSIZE;
589 	size = key->size;
590 	kp = (char *)key->data;
591 	rbufp = __get_buf(hashp, __call_hash(hashp, kp, (int)size), NULL, 0);
592 	if (!rbufp)
593 		return (ERROR);
594 	save_bufp = rbufp;
595 
596 	/* Pin the bucket chain */
597 	rbufp->flags |= BUF_PIN;
598 	for (bp = (uint16_t *)(void *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
599 		if (bp[1] >= REAL_KEY) {
600 			/* Real key/data pair */
601 			if (size == (size_t)(off - *bp) &&
602 			    memcmp(kp, rbufp->page + *bp, size) == 0)
603 				goto found;
604 			off = bp[1];
605 #ifdef HASH_STATISTICS
606 			hash_collisions++;
607 #endif
608 			bp += 2;
609 			ndx += 2;
610 		} else if (bp[1] == OVFLPAGE) {
611 			rbufp = __get_buf(hashp, (uint32_t)*bp, rbufp, 0);
612 			if (!rbufp) {
613 				save_bufp->flags &= ~BUF_PIN;
614 				return (ERROR);
615 			}
616 			/* FOR LOOP INIT */
617 			bp = (uint16_t *)(void *)rbufp->page;
618 			n = *bp++;
619 			ndx = 1;
620 			off = hashp->BSIZE;
621 		} else if (bp[1] < REAL_KEY) {
622 			if ((ndx =
623 			    __find_bigpair(hashp, rbufp, ndx, kp, (int)size)) > 0)
624 				goto found;
625 			if (ndx == -2) {
626 				bufp = rbufp;
627 				if (!(pageno =
628 				    __find_last_page(hashp, &bufp))) {
629 					ndx = 0;
630 					rbufp = bufp;
631 					break;	/* FOR */
632 				}
633 				rbufp = __get_buf(hashp, (uint32_t)pageno,
634 				    bufp, 0);
635 				if (!rbufp) {
636 					save_bufp->flags &= ~BUF_PIN;
637 					return (ERROR);
638 				}
639 				/* FOR LOOP INIT */
640 				bp = (uint16_t *)(void *)rbufp->page;
641 				n = *bp++;
642 				ndx = 1;
643 				off = hashp->BSIZE;
644 			} else {
645 				save_bufp->flags &= ~BUF_PIN;
646 				return (ERROR);
647 			}
648 		}
649 
650 	/* Not found */
651 	switch (action) {
652 	case HASH_PUT:
653 	case HASH_PUTNEW:
654 		if (__addel(hashp, rbufp, key, val)) {
655 			save_bufp->flags &= ~BUF_PIN;
656 			return (ERROR);
657 		} else {
658 			save_bufp->flags &= ~BUF_PIN;
659 			return (SUCCESS);
660 		}
661 	case HASH_GET:
662 	case HASH_DELETE:
663 	default:
664 		save_bufp->flags &= ~BUF_PIN;
665 		return (ABNORMAL);
666 	}
667 
668 found:
669 	switch (action) {
670 	case HASH_PUTNEW:
671 		save_bufp->flags &= ~BUF_PIN;
672 		return (ABNORMAL);
673 	case HASH_GET:
674 		bp = (uint16_t *)(void *)rbufp->page;
675 		if (bp[ndx + 1] < REAL_KEY) {
676 			if (__big_return(hashp, rbufp, ndx, val, 0))
677 				return (ERROR);
678 		} else {
679 			val->data = (uint8_t *)rbufp->page + (int)bp[ndx + 1];
680 			val->size = bp[ndx] - bp[ndx + 1];
681 		}
682 		break;
683 	case HASH_PUT:
684 		if ((__delpair(hashp, rbufp, ndx)) ||
685 		    (__addel(hashp, rbufp, key, val))) {
686 			save_bufp->flags &= ~BUF_PIN;
687 			return (ERROR);
688 		}
689 		break;
690 	case HASH_DELETE:
691 		if (__delpair(hashp, rbufp, ndx))
692 			return (ERROR);
693 		/*
694 		 * Our index lags 2 behind on the same page when we are
695 		 * deleting the element pointed to by the index; otherwise
696 		 * deleting randomly from an iterated hash produces undefined
697 		 * results.
698 		 */
699 		if (ndx != hashp->cndx - 2 || rbufp != hashp->cpage)
700 			break;
701 
702 		if (hashp->cndx > 1) {
703 			/* Move back one element */
704 			hashp->cndx -= 2;
705 		} else {
706 			/*
707 			 * Move back one page, and indicate to go to the last
708 			 * element of the previous page by setting cndx to -1
709 			 */
710 			hashp->cbucket--;
711 			hashp->cpage = NULL;
712 			hashp->cndx = -1;
713 		}
714 		break;
715 	default:
716 		abort();
717 	}
718 	save_bufp->flags &= ~BUF_PIN;
719 	return (SUCCESS);
720 }
721 
722 static int
hash_seq(const DB * dbp,DBT * key,DBT * data,uint32_t flag)723 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag)
724 {
725 	uint32_t bucket;
726 	BUFHEAD *bufp = NULL; /* XXX: gcc */
727 	HTAB *hashp;
728 	uint16_t *bp, ndx;
729 
730 	hashp = dbp->internal;
731 	if (flag && flag != R_FIRST && flag != R_NEXT) {
732 		hashp->err = errno = EINVAL;
733 		return (ERROR);
734 	}
735 #ifdef HASH_STATISTICS
736 	hash_accesses++;
737 #endif
738 	if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
739 		hashp->cbucket = 0;
740 		hashp->cndx = 1;
741 		hashp->cpage = NULL;
742 	}
743 
744 next_bucket:
745 	for (bp = NULL; !bp || !bp[0]; ) {
746 		if (!(bufp = hashp->cpage)) {
747 			for (bucket = hashp->cbucket;
748 			    bucket <= (uint32_t)hashp->MAX_BUCKET;
749 			    bucket++) {
750 				bufp = __get_buf(hashp, bucket, NULL, 0);
751 				if (!bufp)
752 					return (ERROR);
753 				hashp->cpage = bufp;
754 				bp = (uint16_t *)(void *)bufp->page;
755 				if (bp[0])
756 					break;
757 			}
758 			hashp->cbucket = bucket;
759 			if (hashp->cbucket > hashp->MAX_BUCKET) {
760 				hashp->cbucket = -1;
761 				return (ABNORMAL);
762 			}
763 			if (hashp->cndx == -1) {
764 				/* move to the last element of the page */
765 				hashp->cndx = 1;
766 				while (bp[hashp->cndx - 1] != 0)
767 					hashp->cndx += 2;
768 			} else {
769 				/* start on the first element */
770 				hashp->cndx = 1;
771 			}
772 		} else {
773 			bp = (uint16_t *)(void *)hashp->cpage->page;
774 			if (flag == R_NEXT || flag == 0) {
775 				if (hashp->cndx > bp[0]) {
776 					hashp->cpage = NULL;
777 					hashp->cbucket++;
778 					hashp->cndx = 1;
779 					goto next_bucket;
780 				}
781 			}
782 		}
783 
784 
785 		_DIAGASSERT(bp != NULL);
786 		_DIAGASSERT(bufp != NULL);
787 		while (bp[hashp->cndx + 1] == OVFLPAGE) {
788 			bufp = hashp->cpage =
789 			    __get_buf(hashp, (uint32_t)bp[hashp->cndx], bufp,
790 				0);
791 			if (!bufp)
792 				return (ERROR);
793 			bp = (uint16_t *)(void *)(bufp->page);
794 			hashp->cndx = 1;
795 		}
796 		if (!bp[0]) {
797 			hashp->cpage = NULL;
798 			++hashp->cbucket;
799 		}
800 	}
801 	ndx = hashp->cndx;
802 	if (bp[ndx + 1] < REAL_KEY) {
803 		if (__big_keydata(hashp, bufp, key, data, 1))
804 			return (ERROR);
805 	} else {
806 		if (hashp->cpage == NULL)
807 			return (ERROR);
808 		key->data = (uint8_t *)hashp->cpage->page + bp[ndx];
809 		key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
810 		data->data = (uint8_t *)hashp->cpage->page + bp[ndx + 1];
811 		data->size = bp[ndx] - bp[ndx + 1];
812 	}
813 	hashp->cndx += 2;
814 	return (SUCCESS);
815 }
816 
817 /********************************* UTILITIES ************************/
818 
819 /*
820  * Returns:
821  *	 0 ==> OK
822  *	-1 ==> Error
823  */
824 int
__expand_table(HTAB * hashp)825 __expand_table(HTAB *hashp)
826 {
827 	uint32_t old_bucket, new_bucket;
828 	int new_segnum, spare_ndx;
829 	size_t dirsize;
830 
831 #ifdef HASH_STATISTICS
832 	hash_expansions++;
833 #endif
834 	new_bucket = ++hashp->MAX_BUCKET;
835 	old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
836 
837 	new_segnum = new_bucket >> hashp->SSHIFT;
838 
839 	/* Check if we need a new segment */
840 	if (new_segnum >= hashp->nsegs) {
841 		/* Check if we need to expand directory */
842 		if (new_segnum >= hashp->DSIZE) {
843 			/* Reallocate directory */
844 			dirsize = hashp->DSIZE * sizeof(SEGMENT *);
845 			if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
846 				return (-1);
847 			dirsize <<= 1;
848 			_DBFIT(dirsize, uint32_t);
849 			hashp->DSIZE = (uint32_t)dirsize;
850 		}
851 		if ((hashp->dir[new_segnum] =
852 		    calloc((size_t)hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
853 			return (-1);
854 		hashp->exsegs++;
855 		hashp->nsegs++;
856 	}
857 	/*
858 	 * If the split point is increasing (MAX_BUCKET's log base 2
859 	 * * increases), we need to copy the current contents of the spare
860 	 * split bucket to the next bucket.
861 	 */
862 	spare_ndx = __log2((uint32_t)(hashp->MAX_BUCKET + 1));
863 	if (spare_ndx > hashp->OVFL_POINT) {
864 		hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
865 		hashp->OVFL_POINT = spare_ndx;
866 	}
867 
868 	if (new_bucket > (uint32_t)hashp->HIGH_MASK) {
869 		/* Starting a new doubling */
870 		hashp->LOW_MASK = hashp->HIGH_MASK;
871 		hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
872 	}
873 	/* Relocate records to the new bucket */
874 	return (__split_page(hashp, old_bucket, new_bucket));
875 }
876 
877 /*
878  * If realloc guarantees that the pointer is not destroyed if the realloc
879  * fails, then this routine can go away.
880  */
881 static void *
hash_realloc(SEGMENT ** p_ptr,size_t oldsize,size_t newsize)882 hash_realloc(SEGMENT **p_ptr, size_t oldsize, size_t newsize)
883 {
884 	void *p;
885 
886 	if ((p = malloc(newsize)) != NULL) {
887 		memmove(p, *p_ptr, oldsize);
888 		memset((char *)p + oldsize, 0, newsize - oldsize);
889 		free(*p_ptr);
890 		*p_ptr = p;
891 	}
892 	return (p);
893 }
894 
895 uint32_t
__call_hash(HTAB * hashp,char * k,int len)896 __call_hash(HTAB *hashp, char *k, int len)
897 {
898 	int n, bucket;
899 
900 	n = hashp->hash(k, (size_t)len);
901 	bucket = n & hashp->HIGH_MASK;
902 	if (bucket > hashp->MAX_BUCKET)
903 		bucket = bucket & hashp->LOW_MASK;
904 	return (bucket);
905 }
906 
907 /*
908  * Allocate segment table.  On error, destroy the table and set errno.
909  *
910  * Returns 0 on success
911  */
912 static int
alloc_segs(HTAB * hashp,int nsegs)913 alloc_segs(HTAB *hashp, int nsegs)
914 {
915 	int i;
916 	SEGMENT store;
917 
918 	int save_errno;
919 
920 	hashp->dir = calloc((size_t)hashp->DSIZE, sizeof(SEGMENT *));
921 	if (hashp->dir == NULL) {
922 		save_errno = errno;
923 		(void)hdestroy(hashp);
924 		errno = save_errno;
925 		return (-1);
926 	}
927 	hashp->nsegs = nsegs;
928 	if (nsegs == 0)
929 		return 0;
930 	/* Allocate segments */
931 	store = calloc((size_t)(nsegs << hashp->SSHIFT), sizeof(SEGMENT));
932 	if (store == NULL) {
933 		save_errno = errno;
934 		(void)hdestroy(hashp);
935 		errno = save_errno;
936 		return (-1);
937 	}
938 	for (i = 0; i < nsegs; i++)
939 		hashp->dir[i] = &store[i << hashp->SSHIFT];
940 	return (0);
941 }
942 
943 #if BYTE_ORDER == LITTLE_ENDIAN
944 /*
945  * Hashp->hdr needs to be byteswapped.
946  */
947 static void
swap_header_copy(HASHHDR * srcp,HASHHDR * destp)948 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
949 {
950 	size_t i;
951 
952 	P_32_COPY(srcp->magic, destp->magic);
953 	P_32_COPY(srcp->version, destp->version);
954 	P_32_COPY(srcp->lorder, destp->lorder);
955 	P_32_COPY(srcp->bsize, destp->bsize);
956 	P_32_COPY(srcp->bshift, destp->bshift);
957 	P_32_COPY(srcp->dsize, destp->dsize);
958 	P_32_COPY(srcp->ssize, destp->ssize);
959 	P_32_COPY(srcp->sshift, destp->sshift);
960 	P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
961 	P_32_COPY(srcp->last_freed, destp->last_freed);
962 	P_32_COPY(srcp->max_bucket, destp->max_bucket);
963 	P_32_COPY(srcp->high_mask, destp->high_mask);
964 	P_32_COPY(srcp->low_mask, destp->low_mask);
965 	P_32_COPY(srcp->ffactor, destp->ffactor);
966 	P_32_COPY(srcp->nkeys, destp->nkeys);
967 	P_32_COPY(srcp->hdrpages, destp->hdrpages);
968 	P_32_COPY(srcp->h_charkey, destp->h_charkey);
969 	for (i = 0; i < NCACHED; i++) {
970 		P_32_COPY(srcp->spares[i], destp->spares[i]);
971 		P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
972 	}
973 }
974 
975 static void
swap_header(HTAB * hashp)976 swap_header(HTAB *hashp)
977 {
978 	HASHHDR *hdrp;
979 	size_t i;
980 
981 	hdrp = &hashp->hdr;
982 
983 	M_32_SWAP(hdrp->magic);
984 	M_32_SWAP(hdrp->version);
985 	M_32_SWAP(hdrp->lorder);
986 	M_32_SWAP(hdrp->bsize);
987 	M_32_SWAP(hdrp->bshift);
988 	M_32_SWAP(hdrp->dsize);
989 	M_32_SWAP(hdrp->ssize);
990 	M_32_SWAP(hdrp->sshift);
991 	M_32_SWAP(hdrp->ovfl_point);
992 	M_32_SWAP(hdrp->last_freed);
993 	M_32_SWAP(hdrp->max_bucket);
994 	M_32_SWAP(hdrp->high_mask);
995 	M_32_SWAP(hdrp->low_mask);
996 	M_32_SWAP(hdrp->ffactor);
997 	M_32_SWAP(hdrp->nkeys);
998 	M_32_SWAP(hdrp->hdrpages);
999 	M_32_SWAP(hdrp->h_charkey);
1000 	for (i = 0; i < NCACHED; i++) {
1001 		M_32_SWAP(hdrp->spares[i]);
1002 		M_16_SWAP(hdrp->bitmaps[i]);
1003 	}
1004 }
1005 #endif
1006