1 /*-
2 * Copyright (c) 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Margo Seltzer.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. 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 #define _DEFAULT_SOURCE
34 #include <sys/param.h>
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)hash.c 8.9 (Berkeley) 6/16/94";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39 #include <sys/types.h>
40
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
53 #define __DBINTERFACE_PRIVATE /* activate prototypes from db_local.h */
54 #include "db_local.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 *, HASH_ACTION, DBT *, DBT *);
62 static int hash_close(DB *);
63 static int hash_delete(const DB *, const DBT *, u_int);
64 static int hash_fd(const DB *);
65 static int hash_get(const DB *, const DBT *, DBT *, u_int);
66 static int hash_put(const DB *, DBT *, const DBT *, u_int);
67 static void *hash_realloc(SEGMENT **, int, int);
68 static int hash_seq(const DB *, DBT *, DBT *, u_int);
69 static int hash_sync(const DB *, u_int);
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 /* Macros for min/max. */
79 #ifndef MIN
80 #define MIN(a,b) (((a)<(b))?(a):(b))
81 #endif
82 #ifndef MAX
83 #define MAX(a,b) (((a)>(b))?(a):(b))
84 #endif
85
86 /* Fast arithmetic, relying on powers of 2, */
87 #define MOD(x, y) ((x) & ((y) - 1))
88
89 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; }
90
91 /* Return values */
92 #define SUCCESS (0)
93 #define ERROR (-1)
94 #define ABNORMAL (1)
95
96 #ifdef HASH_STATISTICS
97 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
98 #endif
99
100 /************************** INTERFACE ROUTINES ***************************/
101 /* OPEN/CLOSE */
102
103 extern DB *
__hash_open(const char * file,int flags,int mode,int dflags,const HASHINFO * info)104 __hash_open (const char *file,
105 int flags,
106 int mode,
107 int dflags,
108 const HASHINFO *info) /* Special directives for create */
109 {
110 HTAB *hashp;
111
112 (void) dflags;
113 #ifdef __USE_INTERNAL_STAT64
114 struct stat64 statbuf;
115 #else
116 struct stat statbuf;
117 #endif
118 DB *dbp;
119 int bpages, hdrsize, new_table, nsegs, save_errno;
120
121 if ((flags & O_ACCMODE) == O_WRONLY) {
122 errno = EINVAL;
123 return (NULL);
124 }
125
126 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
127 return (NULL);
128 hashp->fp = -1;
129
130 /*
131 * Even if user wants write only, we need to be able to read
132 * the actual file, so we need to open it read/write. But, the
133 * field in the hashp structure needs to be accurate so that
134 * we can check accesses.
135 */
136 hashp->flags = flags;
137
138 new_table = 0;
139 if (!file || (flags & O_TRUNC) ||
140 #ifdef __USE_INTERNAL_STAT64
141 (stat64(file, &statbuf) && (errno == ENOENT))) {
142 #else
143 (stat(file, &statbuf) && (errno == ENOENT))) {
144 #endif
145 if (errno == ENOENT)
146 errno = 0; /* Just in case someone looks at errno */
147 new_table = 1;
148 }
149 if (file) {
150 if ((hashp->fp = open(file, flags, mode)) == -1)
151 RETURN_ERROR(errno, error0);
152
153 /* if the .db file is empty, and we had permission to create
154 a new .db file, then reinitialize the database */
155 if ((flags & O_CREAT) &&
156 #ifdef __USE_INTERNAL_STAT64
157 fstat64(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
158 #else
159 fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
160 #endif
161 new_table = 1;
162
163 #ifdef _HAVE_FCNTL
164 (void)fcntl(hashp->fp, F_SETFD, 1);
165 #endif
166 }
167 if (new_table) {
168 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
169 RETURN_ERROR(errno, error1);
170 } else {
171 /* Table already exists */
172 if (info && info->hash)
173 hashp->hash = info->hash;
174 else
175 hashp->hash = __default_hash;
176
177 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
178 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
179 swap_header(hashp);
180 #endif
181 if (hdrsize == -1)
182 RETURN_ERROR(errno, error1);
183 if (hdrsize != sizeof(HASHHDR))
184 RETURN_ERROR(EFTYPE, error1);
185 /* Verify file type, versions and hash function */
186 if (hashp->MAGIC != HASHMAGIC)
187 RETURN_ERROR(EFTYPE, error1);
188 #define OLDHASHVERSION 1
189 if (hashp->HASH_VERSION != HASHVERSION &&
190 hashp->HASH_VERSION != OLDHASHVERSION)
191 RETURN_ERROR(EFTYPE, error1);
192 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
193 RETURN_ERROR(EFTYPE, error1);
194 /* Check bucket size isn't too big for target int. */
195 if (hashp->BSIZE > INT_MAX)
196 RETURN_ERROR(EFTYPE, error1);
197 /*
198 * Figure out how many segments we need. Max_Bucket is the
199 * maximum bucket number, so the number of buckets is
200 * max_bucket + 1.
201 */
202 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
203 hashp->SGSIZE;
204 hashp->nsegs = 0;
205 if (alloc_segs(hashp, nsegs))
206 /*
207 * If alloc_segs fails, table will have been destroyed
208 * and errno will have been set.
209 */
210 return (NULL);
211 /* Read in bitmaps */
212 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
213 (hashp->BSIZE << BYTE_SHIFT) - 1) >>
214 (hashp->BSHIFT + BYTE_SHIFT);
215
216 hashp->nmaps = bpages;
217 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(__uint32_t *));
218 }
219
220 /* Initialize Buffer Manager */
221 if (info && info->cachesize)
222 __buf_init(hashp, info->cachesize);
223 else
224 __buf_init(hashp, DEF_BUFSIZE);
225
226 hashp->new_file = new_table;
227 hashp->save_file = file && (hashp->flags & O_RDWR);
228 hashp->cbucket = -1;
229 if (!(dbp = (DB *)malloc(sizeof(DB)))) {
230 save_errno = errno;
231 hdestroy(hashp);
232 errno = save_errno;
233 return (NULL);
234 }
235 dbp->internal = hashp;
236 dbp->close = hash_close;
237 dbp->del = hash_delete;
238 dbp->fd = hash_fd;
239 dbp->get = hash_get;
240 dbp->put = hash_put;
241 dbp->seq = hash_seq;
242 dbp->sync = hash_sync;
243 dbp->type = DB_HASH;
244
245 #ifdef DEBUG
246 (void)fprintf(stderr,
247 "%s\n%s%x\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%x\n%s%x\n%s%d\n%s%d\n",
248 "init_htab:",
249 "TABLE POINTER ", hashp,
250 "BUCKET SIZE ", hashp->BSIZE,
251 "BUCKET SHIFT ", hashp->BSHIFT,
252 "DIRECTORY SIZE ", hashp->DSIZE,
253 "SEGMENT SIZE ", hashp->SGSIZE,
254 "SEGMENT SHIFT ", hashp->SSHIFT,
255 "FILL FACTOR ", hashp->FFACTOR,
256 "MAX BUCKET ", hashp->MAX_BUCKET,
257 "OVFL POINT ", hashp->OVFL_POINT,
258 "LAST FREED ", hashp->LAST_FREED,
259 "HIGH MASK ", hashp->HIGH_MASK,
260 "LOW MASK ", hashp->LOW_MASK,
261 "NSEGS ", hashp->nsegs,
262 "NKEYS ", hashp->NKEYS);
263 #endif
264 #ifdef HASH_STATISTICS
265 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
266 #endif
267 return (dbp);
268
269 error1:
270 if (hashp != NULL)
271 (void)close(hashp->fp);
272
273 error0:
274 free(hashp);
275 errno = save_errno;
276 return (NULL);
277 }
278
279 static int
280 hash_close(DB *dbp)
281 {
282 HTAB *hashp;
283 int retval;
284
285 if (!dbp)
286 return (ERROR);
287
288 hashp = (HTAB *)dbp->internal;
289 retval = hdestroy(hashp);
290 free(dbp);
291 return (retval);
292 }
293
294 static int
295 hash_fd(const DB *dbp)
296 {
297 HTAB *hashp;
298
299 if (!dbp)
300 return (ERROR);
301
302 hashp = (HTAB *)dbp->internal;
303 if (hashp->fp == -1) {
304 errno = ENOENT;
305 return (-1);
306 }
307 return (hashp->fp);
308 }
309
310 /************************** LOCAL CREATION ROUTINES **********************/
311 static HTAB *
312 init_hash(HTAB *hashp,
313 const char *file,
314 const HASHINFO *info)
315 {
316 #ifdef __USE_INTERNAL_STAT64
317 struct stat64 statbuf;
318 #else
319 struct stat statbuf;
320 #endif
321 int nelem;
322
323 nelem = 1;
324 hashp->NKEYS = 0;
325 hashp->LORDER = DB_BYTE_ORDER;
326 hashp->BSIZE = DEF_BUCKET_SIZE;
327 hashp->BSHIFT = DEF_BUCKET_SHIFT;
328 hashp->SGSIZE = DEF_SEGSIZE;
329 hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
330 hashp->DSIZE = DEF_DIRSIZE;
331 hashp->FFACTOR = DEF_FFACTOR;
332 hashp->hash = __default_hash;
333 memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
334 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
335
336 /* Fix bucket size to be optimal for file system */
337 if (file != NULL) {
338 #ifdef __USE_INTERNAL_STAT64
339 if (stat64(file, &statbuf))
340 #else
341 if (stat(file, &statbuf))
342 #endif
343 return (NULL);
344 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
345 hashp->BSHIFT = __log2(hashp->BSIZE);
346 }
347
348 if (info) {
349 if (info->bsize) {
350 /* Round pagesize up to power of 2 */
351 hashp->BSHIFT = __log2(info->bsize);
352 hashp->BSIZE = 1 << hashp->BSHIFT;
353 if (hashp->BSIZE > MAX_BSIZE) {
354 errno = EINVAL;
355 return (NULL);
356 }
357 }
358 if (info->ffactor)
359 hashp->FFACTOR = info->ffactor;
360 if (info->hash)
361 hashp->hash = info->hash;
362 if (info->nelem)
363 nelem = info->nelem;
364 if (info->lorder) {
365 if (info->lorder != DB_BIG_ENDIAN &&
366 info->lorder != DB_LITTLE_ENDIAN) {
367 errno = EINVAL;
368 return (NULL);
369 }
370 hashp->LORDER = info->lorder;
371 }
372 }
373 /* init_htab should destroy the table and set errno if it fails */
374 if (init_htab(hashp, nelem))
375 return (NULL);
376 else
377 return (hashp);
378 }
379 /*
380 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy
381 * the table and set errno, so we just pass the error information along.
382 *
383 * Returns 0 on No Error
384 */
385 static int
386 init_htab(HTAB *hashp, int nelem)
387 {
388 int nbuckets, nsegs;
389 int l2;
390
391 /*
392 * Divide number of elements by the fill factor and determine a
393 * desired number of buckets. Allocate space for the next greater
394 * power of two number of buckets.
395 */
396 nelem = (nelem - 1) / hashp->FFACTOR + 1;
397
398 l2 = __log2(MAX(nelem, 2));
399 nbuckets = 1 << l2;
400
401 hashp->SPARES[l2] = l2 + 1;
402 hashp->SPARES[l2 + 1] = l2 + 1;
403 hashp->OVFL_POINT = l2;
404 hashp->LAST_FREED = 2;
405
406 /* First bitmap page is at: splitpoint l2 page offset 1 */
407 if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
408 return (-1);
409
410 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
411 hashp->HIGH_MASK = (nbuckets << 1) - 1;
412 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
413 hashp->BSHIFT) + 1;
414
415 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
416 nsegs = 1 << __log2(nsegs);
417
418 if (nsegs > hashp->DSIZE)
419 hashp->DSIZE = nsegs;
420 return (alloc_segs(hashp, nsegs));
421 }
422
423 /********************** DESTROY/CLOSE ROUTINES ************************/
424
425 /*
426 * Flushes any changes to the file if necessary and destroys the hashp
427 * structure, freeing all allocated space.
428 */
429 static int
430 hdestroy(HTAB *hashp)
431 {
432 int i, save_errno;
433
434 save_errno = 0;
435
436 #ifdef HASH_STATISTICS
437 (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
438 hash_accesses, hash_collisions);
439 (void)fprintf(stderr, "hdestroy: expansions %ld\n",
440 hash_expansions);
441 (void)fprintf(stderr, "hdestroy: overflows %ld\n",
442 hash_overflows);
443 (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
444 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
445
446 for (i = 0; i < NCACHED; i++)
447 (void)fprintf(stderr,
448 "spares[%d] = %d\n", i, hashp->SPARES[i]);
449 #endif
450 /*
451 * Call on buffer manager to free buffers, and if required,
452 * write them to disk.
453 */
454 if (__buf_free(hashp, 1, hashp->save_file))
455 save_errno = errno;
456 if (hashp->dir) {
457 free(*hashp->dir); /* Free initial segments */
458 /* Free extra segments */
459 while (hashp->exsegs--)
460 free(hashp->dir[--hashp->nsegs]);
461 free(hashp->dir);
462 }
463 if (flush_meta(hashp) && !save_errno)
464 save_errno = errno;
465 /* Free Bigmaps */
466 for (i = 0; i < hashp->nmaps; i++)
467 if (hashp->mapp[i])
468 free(hashp->mapp[i]);
469
470 if (hashp->fp != -1)
471 (void)close(hashp->fp);
472
473 free(hashp);
474
475 if (save_errno) {
476 errno = save_errno;
477 return (ERROR);
478 }
479 return (SUCCESS);
480 }
481 /*
482 * Write modified pages to disk
483 *
484 * Returns:
485 * 0 == OK
486 * -1 ERROR
487 */
488 static int
489 hash_sync(const DB *dbp, u_int flags)
490 {
491 HTAB *hashp;
492
493 if (flags != 0) {
494 errno = EINVAL;
495 return (ERROR);
496 }
497
498 if (!dbp)
499 return (ERROR);
500
501 hashp = (HTAB *)dbp->internal;
502 if (!hashp->save_file)
503 return (0);
504 if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
505 return (ERROR);
506 hashp->new_file = 0;
507 return (0);
508 }
509
510 /*
511 * Returns:
512 * 0 == OK
513 * -1 indicates that errno should be set
514 */
515 static int
516 flush_meta(HTAB *hashp)
517 {
518 HASHHDR *whdrp;
519 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
520 HASHHDR whdr;
521 #endif
522 int fp, i, wsize;
523
524 if (!hashp->save_file)
525 return (0);
526 hashp->MAGIC = HASHMAGIC;
527 hashp->HASH_VERSION = HASHVERSION;
528 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
529
530 fp = hashp->fp;
531 whdrp = &hashp->hdr;
532 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
533 whdrp = &whdr;
534 swap_header_copy(&hashp->hdr, whdrp);
535 #endif
536 if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
537 ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
538 return (-1);
539 else
540 if (wsize != sizeof(HASHHDR)) {
541 errno = EFTYPE;
542 hashp->error = errno;
543 return (-1);
544 }
545 for (i = 0; i < NCACHED; i++)
546 if (hashp->mapp[i])
547 if (__put_page(hashp, (char *)hashp->mapp[i],
548 hashp->BITMAPS[i], 0, 1))
549 return (-1);
550 return (0);
551 }
552
553 /*******************************SEARCH ROUTINES *****************************/
554 /*
555 * All the access routines return
556 *
557 * Returns:
558 * 0 on SUCCESS
559 * 1 to indicate an external ERROR (i.e. key not found, etc)
560 * -1 to indicate an internal ERROR (i.e. out of memory, etc)
561 */
562 static int
563 hash_get(const DB *dbp,
564 const DBT *key,
565 DBT *data,
566 u_int flag)
567 {
568 HTAB *hashp;
569
570 hashp = (HTAB *)dbp->internal;
571 if (flag) {
572 hashp->error = errno = EINVAL;
573 return (ERROR);
574 }
575 return (hash_access(hashp, HASH_GET, (DBT *)key, data));
576 }
577
578 static int
579 hash_put(const DB *dbp,
580 DBT *key,
581 const DBT *data,
582 u_int flag)
583 {
584 HTAB *hashp;
585
586 hashp = (HTAB *)dbp->internal;
587 if (flag && flag != R_NOOVERWRITE) {
588 hashp->error = EINVAL;
589 errno = EINVAL;
590 return (ERROR);
591 }
592 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
593 hashp->error = errno = EPERM;
594 return (ERROR);
595 }
596 return (hash_access(hashp, flag == R_NOOVERWRITE ?
597 HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
598 }
599
600 static int
601 hash_delete(const DB *dbp,
602 const DBT *key,
603 u_int flag) /* Ignored */
604 {
605 HTAB *hashp;
606
607 hashp = (HTAB *)dbp->internal;
608 if (flag && flag != R_CURSOR) {
609 hashp->error = errno = EINVAL;
610 return (ERROR);
611 }
612 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
613 hashp->error = errno = EPERM;
614 return (ERROR);
615 }
616 return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
617 }
618
619 /*
620 * Assume that hashp has been set in wrapper routine.
621 */
622 static int
623 hash_access(HTAB *hashp,
624 HASH_ACTION action,
625 DBT *key,
626 DBT *val)
627 {
628 BUFHEAD *rbufp;
629 BUFHEAD *bufp, *save_bufp;
630 __uint16_t *bp;
631 int n, ndx, off, size;
632 char *kp;
633 __uint16_t pageno;
634
635 #ifdef HASH_STATISTICS
636 hash_accesses++;
637 #endif
638
639 off = hashp->BSIZE;
640 size = key->size;
641 kp = (char *)key->data;
642 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
643 if (!rbufp)
644 return (ERROR);
645 save_bufp = rbufp;
646
647 /* Pin the bucket chain */
648 rbufp->flags |= BUF_PIN;
649 for (bp = (__uint16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
650 if (bp[1] >= REAL_KEY) {
651 /* Real key/data pair */
652 if (size == off - (int) *bp &&
653 memcmp(kp, rbufp->page + *bp, size) == 0)
654 goto found;
655 off = bp[1];
656 #ifdef HASH_STATISTICS
657 hash_collisions++;
658 #endif
659 bp += 2;
660 ndx += 2;
661 } else if (bp[1] == OVFLPAGE) {
662 rbufp = __get_buf(hashp, *bp, rbufp, 0);
663 if (!rbufp) {
664 save_bufp->flags &= ~BUF_PIN;
665 return (ERROR);
666 }
667 /* FOR LOOP INIT */
668 bp = (__uint16_t *)rbufp->page;
669 n = *bp++;
670 ndx = 1;
671 off = hashp->BSIZE;
672 } else if (bp[1] < REAL_KEY) {
673 if ((ndx =
674 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
675 goto found;
676 if (ndx == -2) {
677 bufp = rbufp;
678 if (!(pageno =
679 __find_last_page(hashp, &bufp))) {
680 ndx = 0;
681 rbufp = bufp;
682 break; /* FOR */
683 }
684 rbufp = __get_buf(hashp, pageno, bufp, 0);
685 if (!rbufp) {
686 save_bufp->flags &= ~BUF_PIN;
687 return (ERROR);
688 }
689 /* FOR LOOP INIT */
690 bp = (__uint16_t *)rbufp->page;
691 n = *bp++;
692 ndx = 1;
693 off = hashp->BSIZE;
694 } else {
695 save_bufp->flags &= ~BUF_PIN;
696 return (ERROR);
697 }
698 }
699
700 /* Not found */
701 switch (action) {
702 case HASH_PUT:
703 case HASH_PUTNEW:
704 if (__addel(hashp, rbufp, key, val)) {
705 save_bufp->flags &= ~BUF_PIN;
706 return (ERROR);
707 } else {
708 save_bufp->flags &= ~BUF_PIN;
709 return (SUCCESS);
710 }
711 case HASH_GET:
712 case HASH_DELETE:
713 default:
714 save_bufp->flags &= ~BUF_PIN;
715 return (ABNORMAL);
716 }
717
718 found:
719 switch (action) {
720 case HASH_PUTNEW:
721 save_bufp->flags &= ~BUF_PIN;
722 return (ABNORMAL);
723 case HASH_GET:
724 bp = (__uint16_t *)rbufp->page;
725 if (bp[ndx + 1] < REAL_KEY) {
726 if (__big_return(hashp, rbufp, ndx, val, 0))
727 return (ERROR);
728 } else {
729 val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
730 val->size = bp[ndx] - bp[ndx + 1];
731 }
732 break;
733 case HASH_PUT:
734 if ((__delpair(hashp, rbufp, ndx)) ||
735 (__addel(hashp, rbufp, key, val))) {
736 save_bufp->flags &= ~BUF_PIN;
737 return (ERROR);
738 }
739 break;
740 case HASH_DELETE:
741 if (__delpair(hashp, rbufp, ndx))
742 return (ERROR);
743 break;
744 default:
745 abort();
746 }
747 save_bufp->flags &= ~BUF_PIN;
748 return (SUCCESS);
749 }
750
751 static int
752 hash_seq(const DB *dbp,
753 DBT *key,
754 DBT *data,
755 u_int flag)
756 {
757 int32_t bucket;
758 BUFHEAD *bufp;
759 HTAB *hashp;
760 __uint16_t *bp, ndx;
761
762 hashp = (HTAB *)dbp->internal;
763 if (flag && flag != R_FIRST && flag != R_NEXT) {
764 hashp->error = errno = EINVAL;
765 return (ERROR);
766 }
767 #ifdef HASH_STATISTICS
768 hash_accesses++;
769 #endif
770 if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
771 hashp->cbucket = 0;
772 hashp->cndx = 1;
773 hashp->cpage = NULL;
774 }
775
776 for (bp = NULL; !bp || !bp[0]; ) {
777 if (!(bufp = hashp->cpage)) {
778 for (bucket = hashp->cbucket;
779 bucket <= hashp->MAX_BUCKET;
780 bucket++, hashp->cndx = 1) {
781 bufp = __get_buf(hashp, bucket, NULL, 0);
782 if (!bufp)
783 return (ERROR);
784 hashp->cpage = bufp;
785 bp = (__uint16_t *)bufp->page;
786 if (bp[0])
787 break;
788 }
789 hashp->cbucket = bucket;
790 if (hashp->cbucket > hashp->MAX_BUCKET) {
791 hashp->cbucket = -1;
792 return (ABNORMAL);
793 }
794 } else
795 bp = (__uint16_t *)hashp->cpage->page;
796
797 #ifdef DEBUG
798 assert(bp);
799 assert(bufp);
800 #endif
801 while (bp[hashp->cndx + 1] == OVFLPAGE) {
802 bufp = hashp->cpage =
803 __get_buf(hashp, bp[hashp->cndx], bufp, 0);
804 if (!bufp)
805 return (ERROR);
806 bp = (__uint16_t *)(bufp->page);
807 hashp->cndx = 1;
808 }
809 if (!bp[0]) {
810 hashp->cpage = NULL;
811 ++hashp->cbucket;
812 }
813 }
814 ndx = hashp->cndx;
815 if (bp[ndx + 1] < REAL_KEY) {
816 if (__big_keydata(hashp, bufp, key, data, 1))
817 return (ERROR);
818 } else {
819 key->data = (u_char *)hashp->cpage->page + bp[ndx];
820 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
821 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
822 data->size = bp[ndx] - bp[ndx + 1];
823 ndx += 2;
824 if (ndx > bp[0]) {
825 hashp->cpage = NULL;
826 hashp->cbucket++;
827 hashp->cndx = 1;
828 } else
829 hashp->cndx = ndx;
830 }
831 return (SUCCESS);
832 }
833
834 /********************************* UTILITIES ************************/
835
836 /*
837 * Returns:
838 * 0 ==> OK
839 * -1 ==> Error
840 */
841 extern int
842 __expand_table(HTAB *hashp)
843 {
844 int32_t old_bucket, new_bucket;
845 int dirsize, new_segnum, spare_ndx;
846
847 #ifdef HASH_STATISTICS
848 hash_expansions++;
849 #endif
850 new_bucket = ++hashp->MAX_BUCKET;
851 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
852
853 new_segnum = new_bucket >> hashp->SSHIFT;
854
855 /* Check if we need a new segment */
856 if (new_segnum >= hashp->nsegs) {
857 /* Check if we need to expand directory */
858 if (new_segnum >= hashp->DSIZE) {
859 /* Reallocate directory */
860 dirsize = hashp->DSIZE * sizeof(SEGMENT *);
861 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
862 return (-1);
863 hashp->DSIZE = dirsize << 1;
864 }
865 if ((hashp->dir[new_segnum] =
866 (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
867 return (-1);
868 hashp->exsegs++;
869 hashp->nsegs++;
870 }
871 /*
872 * If the split point is increasing (MAX_BUCKET's log base 2
873 * * increases), we need to copy the current contents of the spare
874 * split bucket to the next bucket.
875 */
876 spare_ndx = __log2(hashp->MAX_BUCKET + 1);
877 if (spare_ndx > hashp->OVFL_POINT) {
878 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
879 hashp->OVFL_POINT = spare_ndx;
880 }
881
882 if (new_bucket > hashp->HIGH_MASK) {
883 /* Starting a new doubling */
884 hashp->LOW_MASK = hashp->HIGH_MASK;
885 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
886 }
887 /* Relocate records to the new bucket */
888 return (__split_page(hashp, old_bucket, new_bucket));
889 }
890
891 /*
892 * If realloc guarantees that the pointer is not destroyed if the realloc
893 * fails, then this routine can go away.
894 */
895 static void *
896 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
897 {
898 void *p;
899
900 if ( (p = malloc(newsize)) ) {
901 memmove(p, *p_ptr, oldsize);
902 memset((char *)p + oldsize, 0, newsize - oldsize);
903 free(*p_ptr);
904 *p_ptr = p;
905 }
906 return (p);
907 }
908
909 extern __uint32_t
910 __call_hash(HTAB *hashp,
911 char *k,
912 int len)
913 {
914 int32_t n, bucket;
915
916 n = hashp->hash(k, len);
917 bucket = n & hashp->HIGH_MASK;
918 if (bucket > hashp->MAX_BUCKET)
919 bucket = bucket & hashp->LOW_MASK;
920 return (bucket);
921 }
922
923 /*
924 * Allocate segment table. On error, destroy the table and set errno.
925 *
926 * Returns 0 on success
927 */
928 static int
929 alloc_segs(HTAB *hashp, int nsegs)
930 {
931 int i;
932 SEGMENT store;
933
934 int save_errno;
935
936 if ((hashp->dir =
937 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
938 save_errno = errno;
939 (void)hdestroy(hashp);
940 errno = save_errno;
941 return (-1);
942 }
943 /* Allocate segments */
944 if ((store =
945 (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
946 save_errno = errno;
947 (void)hdestroy(hashp);
948 errno = save_errno;
949 return (-1);
950 }
951 for (i = 0; i < nsegs; i++, hashp->nsegs++)
952 hashp->dir[i] = &store[i << hashp->SSHIFT];
953 return (0);
954 }
955
956 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
957 /*
958 * Hashp->hdr needs to be byteswapped.
959 */
960 static void
961 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
962 {
963 int i;
964
965 P_32_COPY(srcp->magic, destp->magic);
966 P_32_COPY(srcp->version, destp->version);
967 P_32_COPY(srcp->lorder, destp->lorder);
968 P_32_COPY(srcp->bsize, destp->bsize);
969 P_32_COPY(srcp->bshift, destp->bshift);
970 P_32_COPY(srcp->dsize, destp->dsize);
971 P_32_COPY(srcp->ssize, destp->ssize);
972 P_32_COPY(srcp->sshift, destp->sshift);
973 P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
974 P_32_COPY(srcp->last_freed, destp->last_freed);
975 P_32_COPY(srcp->max_bucket, destp->max_bucket);
976 P_32_COPY(srcp->high_mask, destp->high_mask);
977 P_32_COPY(srcp->low_mask, destp->low_mask);
978 P_32_COPY(srcp->ffactor, destp->ffactor);
979 P_32_COPY(srcp->nkeys, destp->nkeys);
980 P_32_COPY(srcp->hdrpages, destp->hdrpages);
981 P_32_COPY(srcp->h_charkey, destp->h_charkey);
982 for (i = 0; i < NCACHED; i++) {
983 P_32_COPY(srcp->spares[i], destp->spares[i]);
984 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
985 }
986 }
987
988 static void
989 swap_header(HTAB *hashp)
990 {
991 HASHHDR *hdrp;
992 int i;
993
994 hdrp = &hashp->hdr;
995
996 M_32_SWAP(hdrp->magic);
997 M_32_SWAP(hdrp->version);
998 M_32_SWAP(hdrp->lorder);
999 M_32_SWAP(hdrp->bsize);
1000 M_32_SWAP(hdrp->bshift);
1001 M_32_SWAP(hdrp->dsize);
1002 M_32_SWAP(hdrp->ssize);
1003 M_32_SWAP(hdrp->sshift);
1004 M_32_SWAP(hdrp->ovfl_point);
1005 M_32_SWAP(hdrp->last_freed);
1006 M_32_SWAP(hdrp->max_bucket);
1007 M_32_SWAP(hdrp->high_mask);
1008 M_32_SWAP(hdrp->low_mask);
1009 M_32_SWAP(hdrp->ffactor);
1010 M_32_SWAP(hdrp->nkeys);
1011 M_32_SWAP(hdrp->hdrpages);
1012 M_32_SWAP(hdrp->h_charkey);
1013 for (i = 0; i < NCACHED; i++) {
1014 M_32_SWAP(hdrp->spares[i]);
1015 M_16_SWAP(hdrp->bitmaps[i]);
1016 }
1017 }
1018 #endif
1019
