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/types.h>
35 #include <sys/stat.h>
36 #include <errno.h>
37 #include <fcntl.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <unistd.h>
42 #if !defined(DEBUG) && !defined(NDEBUG)
43 #define NDEBUG
44 #endif
45 #include <assert.h>
46
47 #define __DBINTERFACE_PRIVATE /* activate prototypes from db_local.h */
48 #include "db_local.h"
49 #include "hash.h"
50 #include "page.h"
51 #include "extern.h"
52
53 static int alloc_segs(HTAB *, int);
54 static int flush_meta(HTAB *);
55 static int hash_access(HTAB *, HASH_ACTION, DBT *, DBT *);
56 static int hash_close(DB *);
57 static int hash_delete(const DB *, const DBT *, u_int);
58 static int hash_fd(const DB *);
59 static int hash_get(const DB *, const DBT *, DBT *, u_int);
60 static int hash_put(const DB *, DBT *, const DBT *, u_int);
61 static void *hash_realloc(SEGMENT **, int, int);
62 static int hash_seq(const DB *, DBT *, DBT *, u_int);
63 static int hash_sync(const DB *, u_int);
64 static int hdestroy(HTAB *);
65 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
66 static int init_htab(HTAB *, int);
67 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
68 static void swap_header(HTAB *);
69 static void swap_header_copy(HASHHDR *, HASHHDR *);
70 #endif
71
72 /* Macros for min/max. */
73 #ifndef MIN
74 #define MIN(a,b) (((a)<(b))?(a):(b))
75 #endif
76 #ifndef MAX
77 #define MAX(a,b) (((a)>(b))?(a):(b))
78 #endif
79
80 /* Fast arithmetic, relying on powers of 2, */
81 #define MOD(x, y) ((x) & ((y) - 1))
82
83 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; }
84
85 /* Return values */
86 #define SUCCESS (0)
87 #define ERROR (-1)
88 #define ABNORMAL (1)
89
90 #ifdef HASH_STATISTICS
91 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
92 #endif
93
94 /************************** INTERFACE ROUTINES ***************************/
95 /* OPEN/CLOSE */
96
97 extern DB *
__hash_open(const char * file,int flags,int mode,int dflags,const HASHINFO * info)98 __hash_open (const char *file,
99 int flags,
100 int mode,
101 int dflags,
102 const HASHINFO *info) /* Special directives for create */
103 {
104 HTAB *hashp;
105
106 (void) dflags;
107 #ifdef __USE_INTERNAL_STAT64
108 struct stat64 statbuf;
109 #else
110 struct stat statbuf;
111 #endif
112 DB *dbp;
113 int bpages, hdrsize, new_table, nsegs, save_errno;
114
115 if ((flags & O_ACCMODE) == O_WRONLY) {
116 errno = EINVAL;
117 return (NULL);
118 }
119
120 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
121 return (NULL);
122 hashp->fp = -1;
123
124 /*
125 * Even if user wants write only, we need to be able to read
126 * the actual file, so we need to open it read/write. But, the
127 * field in the hashp structure needs to be accurate so that
128 * we can check accesses.
129 */
130 hashp->flags = flags;
131
132 new_table = 0;
133 if (!file || (flags & O_TRUNC) ||
134 #ifdef __USE_INTERNAL_STAT64
135 (stat64(file, &statbuf) && (errno == ENOENT))) {
136 #else
137 (stat(file, &statbuf) && (errno == ENOENT))) {
138 #endif
139 if (errno == ENOENT)
140 errno = 0; /* Just in case someone looks at errno */
141 new_table = 1;
142 }
143 if (file) {
144 if ((hashp->fp = open(file, flags, mode)) == -1)
145 RETURN_ERROR(errno, error0);
146
147 /* if the .db file is empty, and we had permission to create
148 a new .db file, then reinitialize the database */
149 if ((flags & O_CREAT) &&
150 #ifdef __USE_INTERNAL_STAT64
151 fstat64(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
152 #else
153 fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
154 #endif
155 new_table = 1;
156
157 #ifdef _HAVE_FCNTL
158 (void)fcntl(hashp->fp, F_SETFD, 1);
159 #endif
160 }
161 if (new_table) {
162 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
163 RETURN_ERROR(errno, error1);
164 } else {
165 /* Table already exists */
166 if (info && info->hash)
167 hashp->hash = info->hash;
168 else
169 hashp->hash = __default_hash;
170
171 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
172 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
173 swap_header(hashp);
174 #endif
175 if (hdrsize == -1)
176 RETURN_ERROR(errno, error1);
177 if (hdrsize != sizeof(HASHHDR))
178 RETURN_ERROR(EFTYPE, error1);
179 /* Verify file type, versions and hash function */
180 if (hashp->MAGIC != HASHMAGIC)
181 RETURN_ERROR(EFTYPE, error1);
182 #define OLDHASHVERSION 1
183 if (hashp->HASH_VERSION != HASHVERSION &&
184 hashp->HASH_VERSION != OLDHASHVERSION)
185 RETURN_ERROR(EFTYPE, error1);
186 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
187 RETURN_ERROR(EFTYPE, error1);
188 /* Check bucket size isn't too big for target int. */
189 if (hashp->BSIZE > INT_MAX)
190 RETURN_ERROR(EFTYPE, error1);
191 /*
192 * Figure out how many segments we need. Max_Bucket is the
193 * maximum bucket number, so the number of buckets is
194 * max_bucket + 1.
195 */
196 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
197 hashp->SGSIZE;
198 hashp->nsegs = 0;
199 if (alloc_segs(hashp, nsegs))
200 /*
201 * If alloc_segs fails, table will have been destroyed
202 * and errno will have been set.
203 */
204 return (NULL);
205 /* Read in bitmaps */
206 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
207 (hashp->BSIZE << BYTE_SHIFT) - 1) >>
208 (hashp->BSHIFT + BYTE_SHIFT);
209
210 hashp->nmaps = bpages;
211 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(__uint32_t *));
212 }
213
214 /* Initialize Buffer Manager */
215 if (info && info->cachesize)
216 __buf_init(hashp, info->cachesize);
217 else
218 __buf_init(hashp, DEF_BUFSIZE);
219
220 hashp->new_file = new_table;
221 hashp->save_file = file && (hashp->flags & O_RDWR);
222 hashp->cbucket = -1;
223 if (!(dbp = (DB *)malloc(sizeof(DB)))) {
224 save_errno = errno;
225 hdestroy(hashp);
226 errno = save_errno;
227 return (NULL);
228 }
229 dbp->internal = hashp;
230 dbp->close = hash_close;
231 dbp->del = hash_delete;
232 dbp->fd = hash_fd;
233 dbp->get = hash_get;
234 dbp->put = hash_put;
235 dbp->seq = hash_seq;
236 dbp->sync = hash_sync;
237 dbp->type = DB_HASH;
238
239 #ifdef DEBUG
240 (void)fprintf(stderr,
241 "%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",
242 "init_htab:",
243 "TABLE POINTER ", hashp,
244 "BUCKET SIZE ", hashp->BSIZE,
245 "BUCKET SHIFT ", hashp->BSHIFT,
246 "DIRECTORY SIZE ", hashp->DSIZE,
247 "SEGMENT SIZE ", hashp->SGSIZE,
248 "SEGMENT SHIFT ", hashp->SSHIFT,
249 "FILL FACTOR ", hashp->FFACTOR,
250 "MAX BUCKET ", hashp->MAX_BUCKET,
251 "OVFL POINT ", hashp->OVFL_POINT,
252 "LAST FREED ", hashp->LAST_FREED,
253 "HIGH MASK ", hashp->HIGH_MASK,
254 "LOW MASK ", hashp->LOW_MASK,
255 "NSEGS ", hashp->nsegs,
256 "NKEYS ", hashp->NKEYS);
257 #endif
258 #ifdef HASH_STATISTICS
259 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
260 #endif
261 return (dbp);
262
263 error1:
264 if (hashp != NULL)
265 (void)close(hashp->fp);
266
267 error0:
268 free(hashp);
269 errno = save_errno;
270 return (NULL);
271 }
272
273 static int
274 hash_close(DB *dbp)
275 {
276 HTAB *hashp;
277 int retval;
278
279 if (!dbp)
280 return (ERROR);
281
282 hashp = (HTAB *)dbp->internal;
283 retval = hdestroy(hashp);
284 free(dbp);
285 return (retval);
286 }
287
288 static int
289 hash_fd(const DB *dbp)
290 {
291 HTAB *hashp;
292
293 if (!dbp)
294 return (ERROR);
295
296 hashp = (HTAB *)dbp->internal;
297 if (hashp->fp == -1) {
298 errno = ENOENT;
299 return (-1);
300 }
301 return (hashp->fp);
302 }
303
304 /************************** LOCAL CREATION ROUTINES **********************/
305 static HTAB *
306 init_hash(HTAB *hashp,
307 const char *file,
308 const HASHINFO *info)
309 {
310 #ifdef __USE_INTERNAL_STAT64
311 struct stat64 statbuf;
312 #else
313 struct stat statbuf;
314 #endif
315 int nelem;
316
317 nelem = 1;
318 hashp->NKEYS = 0;
319 hashp->LORDER = DB_BYTE_ORDER;
320 hashp->BSIZE = DEF_BUCKET_SIZE;
321 hashp->BSHIFT = DEF_BUCKET_SHIFT;
322 hashp->SGSIZE = DEF_SEGSIZE;
323 hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
324 hashp->DSIZE = DEF_DIRSIZE;
325 hashp->FFACTOR = DEF_FFACTOR;
326 hashp->hash = __default_hash;
327 memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
328 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
329
330 /* Fix bucket size to be optimal for file system */
331 if (file != NULL) {
332 #ifdef __USE_INTERNAL_STAT64
333 if (stat64(file, &statbuf))
334 #else
335 if (stat(file, &statbuf))
336 #endif
337 return (NULL);
338 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
339 hashp->BSHIFT = __log2(hashp->BSIZE);
340 }
341
342 if (info) {
343 if (info->bsize) {
344 /* Round pagesize up to power of 2 */
345 hashp->BSHIFT = __log2(info->bsize);
346 hashp->BSIZE = 1 << hashp->BSHIFT;
347 if (hashp->BSIZE > MAX_BSIZE) {
348 errno = EINVAL;
349 return (NULL);
350 }
351 }
352 if (info->ffactor)
353 hashp->FFACTOR = info->ffactor;
354 if (info->hash)
355 hashp->hash = info->hash;
356 if (info->nelem)
357 nelem = info->nelem;
358 if (info->lorder) {
359 if (info->lorder != DB_BIG_ENDIAN &&
360 info->lorder != DB_LITTLE_ENDIAN) {
361 errno = EINVAL;
362 return (NULL);
363 }
364 hashp->LORDER = info->lorder;
365 }
366 }
367 /* init_htab should destroy the table and set errno if it fails */
368 if (init_htab(hashp, nelem))
369 return (NULL);
370 else
371 return (hashp);
372 }
373 /*
374 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy
375 * the table and set errno, so we just pass the error information along.
376 *
377 * Returns 0 on No Error
378 */
379 static int
380 init_htab(HTAB *hashp, int nelem)
381 {
382 int nbuckets, nsegs;
383 int l2;
384
385 /*
386 * Divide number of elements by the fill factor and determine a
387 * desired number of buckets. Allocate space for the next greater
388 * power of two number of buckets.
389 */
390 nelem = (nelem - 1) / hashp->FFACTOR + 1;
391
392 l2 = __log2(MAX(nelem, 2));
393 nbuckets = 1 << l2;
394
395 hashp->SPARES[l2] = l2 + 1;
396 hashp->SPARES[l2 + 1] = l2 + 1;
397 hashp->OVFL_POINT = l2;
398 hashp->LAST_FREED = 2;
399
400 /* First bitmap page is at: splitpoint l2 page offset 1 */
401 if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
402 return (-1);
403
404 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
405 hashp->HIGH_MASK = (nbuckets << 1) - 1;
406 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
407 hashp->BSHIFT) + 1;
408
409 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
410 nsegs = 1 << __log2(nsegs);
411
412 if (nsegs > hashp->DSIZE)
413 hashp->DSIZE = nsegs;
414 return (alloc_segs(hashp, nsegs));
415 }
416
417 /********************** DESTROY/CLOSE ROUTINES ************************/
418
419 /*
420 * Flushes any changes to the file if necessary and destroys the hashp
421 * structure, freeing all allocated space.
422 */
423 static int
424 hdestroy(HTAB *hashp)
425 {
426 int i, save_errno;
427
428 save_errno = 0;
429
430 #ifdef HASH_STATISTICS
431 (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
432 hash_accesses, hash_collisions);
433 (void)fprintf(stderr, "hdestroy: expansions %ld\n",
434 hash_expansions);
435 (void)fprintf(stderr, "hdestroy: overflows %ld\n",
436 hash_overflows);
437 (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
438 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
439
440 for (i = 0; i < NCACHED; i++)
441 (void)fprintf(stderr,
442 "spares[%d] = %d\n", i, hashp->SPARES[i]);
443 #endif
444 /*
445 * Call on buffer manager to free buffers, and if required,
446 * write them to disk.
447 */
448 if (__buf_free(hashp, 1, hashp->save_file))
449 save_errno = errno;
450 if (hashp->dir) {
451 free(*hashp->dir); /* Free initial segments */
452 /* Free extra segments */
453 while (hashp->exsegs--)
454 free(hashp->dir[--hashp->nsegs]);
455 free(hashp->dir);
456 }
457 if (flush_meta(hashp) && !save_errno)
458 save_errno = errno;
459 /* Free Bigmaps */
460 for (i = 0; i < hashp->nmaps; i++)
461 if (hashp->mapp[i])
462 free(hashp->mapp[i]);
463
464 if (hashp->fp != -1)
465 (void)close(hashp->fp);
466
467 free(hashp);
468
469 if (save_errno) {
470 errno = save_errno;
471 return (ERROR);
472 }
473 return (SUCCESS);
474 }
475 /*
476 * Write modified pages to disk
477 *
478 * Returns:
479 * 0 == OK
480 * -1 ERROR
481 */
482 static int
483 hash_sync(const DB *dbp, u_int flags)
484 {
485 HTAB *hashp;
486
487 if (flags != 0) {
488 errno = EINVAL;
489 return (ERROR);
490 }
491
492 if (!dbp)
493 return (ERROR);
494
495 hashp = (HTAB *)dbp->internal;
496 if (!hashp->save_file)
497 return (0);
498 if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
499 return (ERROR);
500 hashp->new_file = 0;
501 return (0);
502 }
503
504 /*
505 * Returns:
506 * 0 == OK
507 * -1 indicates that errno should be set
508 */
509 static int
510 flush_meta(HTAB *hashp)
511 {
512 HASHHDR *whdrp;
513 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
514 HASHHDR whdr;
515 #endif
516 int fp, i, wsize;
517
518 if (!hashp->save_file)
519 return (0);
520 hashp->MAGIC = HASHMAGIC;
521 hashp->HASH_VERSION = HASHVERSION;
522 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
523
524 fp = hashp->fp;
525 whdrp = &hashp->hdr;
526 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
527 whdrp = &whdr;
528 swap_header_copy(&hashp->hdr, whdrp);
529 #endif
530 if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
531 ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
532 return (-1);
533 else
534 if (wsize != sizeof(HASHHDR)) {
535 errno = EFTYPE;
536 hashp->error = errno;
537 return (-1);
538 }
539 for (i = 0; i < NCACHED; i++)
540 if (hashp->mapp[i])
541 if (__put_page(hashp, (char *)hashp->mapp[i],
542 hashp->BITMAPS[i], 0, 1))
543 return (-1);
544 return (0);
545 }
546
547 /*******************************SEARCH ROUTINES *****************************/
548 /*
549 * All the access routines return
550 *
551 * Returns:
552 * 0 on SUCCESS
553 * 1 to indicate an external ERROR (i.e. key not found, etc)
554 * -1 to indicate an internal ERROR (i.e. out of memory, etc)
555 */
556 static int
557 hash_get(const DB *dbp,
558 const DBT *key,
559 DBT *data,
560 u_int flag)
561 {
562 HTAB *hashp;
563
564 hashp = (HTAB *)dbp->internal;
565 if (flag) {
566 hashp->error = errno = EINVAL;
567 return (ERROR);
568 }
569 return (hash_access(hashp, HASH_GET, (DBT *)key, data));
570 }
571
572 static int
573 hash_put(const DB *dbp,
574 DBT *key,
575 const DBT *data,
576 u_int flag)
577 {
578 HTAB *hashp;
579
580 hashp = (HTAB *)dbp->internal;
581 if (flag && flag != R_NOOVERWRITE) {
582 hashp->error = EINVAL;
583 errno = EINVAL;
584 return (ERROR);
585 }
586 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
587 hashp->error = errno = EPERM;
588 return (ERROR);
589 }
590 return (hash_access(hashp, flag == R_NOOVERWRITE ?
591 HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
592 }
593
594 static int
595 hash_delete(const DB *dbp,
596 const DBT *key,
597 u_int flag) /* Ignored */
598 {
599 HTAB *hashp;
600
601 hashp = (HTAB *)dbp->internal;
602 if (flag && flag != R_CURSOR) {
603 hashp->error = errno = EINVAL;
604 return (ERROR);
605 }
606 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
607 hashp->error = errno = EPERM;
608 return (ERROR);
609 }
610 return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
611 }
612
613 /*
614 * Assume that hashp has been set in wrapper routine.
615 */
616 static int
617 hash_access(HTAB *hashp,
618 HASH_ACTION action,
619 DBT *key,
620 DBT *val)
621 {
622 BUFHEAD *rbufp;
623 BUFHEAD *bufp, *save_bufp;
624 __uint16_t *bp;
625 int n, ndx, off, size;
626 char *kp;
627 __uint16_t pageno;
628
629 #ifdef HASH_STATISTICS
630 hash_accesses++;
631 #endif
632
633 off = hashp->BSIZE;
634 size = key->size;
635 kp = (char *)key->data;
636 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
637 if (!rbufp)
638 return (ERROR);
639 save_bufp = rbufp;
640
641 /* Pin the bucket chain */
642 rbufp->flags |= BUF_PIN;
643 for (bp = (__uint16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
644 if (bp[1] >= REAL_KEY) {
645 /* Real key/data pair */
646 if (size == off - (int) *bp &&
647 memcmp(kp, rbufp->page + *bp, size) == 0)
648 goto found;
649 off = bp[1];
650 #ifdef HASH_STATISTICS
651 hash_collisions++;
652 #endif
653 bp += 2;
654 ndx += 2;
655 } else if (bp[1] == OVFLPAGE) {
656 rbufp = __get_buf(hashp, *bp, rbufp, 0);
657 if (!rbufp) {
658 save_bufp->flags &= ~BUF_PIN;
659 return (ERROR);
660 }
661 /* FOR LOOP INIT */
662 bp = (__uint16_t *)rbufp->page;
663 n = *bp++;
664 ndx = 1;
665 off = hashp->BSIZE;
666 } else if (bp[1] < REAL_KEY) {
667 if ((ndx =
668 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
669 goto found;
670 if (ndx == -2) {
671 bufp = rbufp;
672 if (!(pageno =
673 __find_last_page(hashp, &bufp))) {
674 ndx = 0;
675 rbufp = bufp;
676 break; /* FOR */
677 }
678 rbufp = __get_buf(hashp, pageno, bufp, 0);
679 if (!rbufp) {
680 save_bufp->flags &= ~BUF_PIN;
681 return (ERROR);
682 }
683 /* FOR LOOP INIT */
684 bp = (__uint16_t *)rbufp->page;
685 n = *bp++;
686 ndx = 1;
687 off = hashp->BSIZE;
688 } else {
689 save_bufp->flags &= ~BUF_PIN;
690 return (ERROR);
691 }
692 }
693
694 /* Not found */
695 switch (action) {
696 case HASH_PUT:
697 case HASH_PUTNEW:
698 if (__addel(hashp, rbufp, key, val)) {
699 save_bufp->flags &= ~BUF_PIN;
700 return (ERROR);
701 } else {
702 save_bufp->flags &= ~BUF_PIN;
703 return (SUCCESS);
704 }
705 case HASH_GET:
706 case HASH_DELETE:
707 default:
708 save_bufp->flags &= ~BUF_PIN;
709 return (ABNORMAL);
710 }
711
712 found:
713 switch (action) {
714 case HASH_PUTNEW:
715 save_bufp->flags &= ~BUF_PIN;
716 return (ABNORMAL);
717 case HASH_GET:
718 bp = (__uint16_t *)rbufp->page;
719 if (bp[ndx + 1] < REAL_KEY) {
720 if (__big_return(hashp, rbufp, ndx, val, 0))
721 return (ERROR);
722 } else {
723 val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
724 val->size = bp[ndx] - bp[ndx + 1];
725 }
726 break;
727 case HASH_PUT:
728 if ((__delpair(hashp, rbufp, ndx)) ||
729 (__addel(hashp, rbufp, key, val))) {
730 save_bufp->flags &= ~BUF_PIN;
731 return (ERROR);
732 }
733 break;
734 case HASH_DELETE:
735 if (__delpair(hashp, rbufp, ndx))
736 return (ERROR);
737 break;
738 default:
739 abort();
740 }
741 save_bufp->flags &= ~BUF_PIN;
742 return (SUCCESS);
743 }
744
745 static int
746 hash_seq(const DB *dbp,
747 DBT *key,
748 DBT *data,
749 u_int flag)
750 {
751 int32_t bucket;
752 BUFHEAD *bufp;
753 HTAB *hashp;
754 __uint16_t *bp, ndx;
755
756 hashp = (HTAB *)dbp->internal;
757 if (flag && flag != R_FIRST && flag != R_NEXT) {
758 hashp->error = errno = EINVAL;
759 return (ERROR);
760 }
761 #ifdef HASH_STATISTICS
762 hash_accesses++;
763 #endif
764 if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
765 hashp->cbucket = 0;
766 hashp->cndx = 1;
767 hashp->cpage = NULL;
768 }
769
770 for (bp = NULL; !bp || !bp[0]; ) {
771 if (!(bufp = hashp->cpage)) {
772 for (bucket = hashp->cbucket;
773 bucket <= hashp->MAX_BUCKET;
774 bucket++, hashp->cndx = 1) {
775 bufp = __get_buf(hashp, bucket, NULL, 0);
776 if (!bufp)
777 return (ERROR);
778 hashp->cpage = bufp;
779 bp = (__uint16_t *)bufp->page;
780 if (bp[0])
781 break;
782 }
783 hashp->cbucket = bucket;
784 if (hashp->cbucket > hashp->MAX_BUCKET) {
785 hashp->cbucket = -1;
786 return (ABNORMAL);
787 }
788 } else
789 bp = (__uint16_t *)hashp->cpage->page;
790
791 assert(bp);
792 assert(bufp);
793
794 while (bp[hashp->cndx + 1] == OVFLPAGE) {
795 bufp = hashp->cpage =
796 __get_buf(hashp, bp[hashp->cndx], bufp, 0);
797 if (!bufp)
798 return (ERROR);
799 bp = (__uint16_t *)(bufp->page);
800 hashp->cndx = 1;
801 }
802 if (!bp[0]) {
803 hashp->cpage = NULL;
804 ++hashp->cbucket;
805 }
806 }
807 ndx = hashp->cndx;
808 if (bp[ndx + 1] < REAL_KEY) {
809 if (__big_keydata(hashp, bufp, key, data, 1))
810 return (ERROR);
811 } else {
812 key->data = (u_char *)hashp->cpage->page + bp[ndx];
813 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
814 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
815 data->size = bp[ndx] - bp[ndx + 1];
816 ndx += 2;
817 if (ndx > bp[0]) {
818 hashp->cpage = NULL;
819 hashp->cbucket++;
820 hashp->cndx = 1;
821 } else
822 hashp->cndx = ndx;
823 }
824 return (SUCCESS);
825 }
826
827 /********************************* UTILITIES ************************/
828
829 /*
830 * Returns:
831 * 0 ==> OK
832 * -1 ==> Error
833 */
834 extern int
835 __expand_table(HTAB *hashp)
836 {
837 int32_t old_bucket, new_bucket;
838 int dirsize, new_segnum, spare_ndx;
839
840 #ifdef HASH_STATISTICS
841 hash_expansions++;
842 #endif
843 new_bucket = ++hashp->MAX_BUCKET;
844 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
845
846 new_segnum = new_bucket >> hashp->SSHIFT;
847
848 /* Check if we need a new segment */
849 if (new_segnum >= hashp->nsegs) {
850 /* Check if we need to expand directory */
851 if (new_segnum >= hashp->DSIZE) {
852 /* Reallocate directory */
853 dirsize = hashp->DSIZE * sizeof(SEGMENT *);
854 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
855 return (-1);
856 hashp->DSIZE = dirsize << 1;
857 }
858 SEGMENT seg = calloc(hashp->SGSIZE, sizeof(SEGMENT));
859 if (seg == NULL)
860 return (-1);
861 hashp->dir[new_segnum] = seg;
862 hashp->exsegs++;
863 hashp->nsegs++;
864 }
865 /*
866 * If the split point is increasing (MAX_BUCKET's log base 2
867 * * increases), we need to copy the current contents of the spare
868 * split bucket to the next bucket.
869 */
870 spare_ndx = __log2(hashp->MAX_BUCKET + 1);
871 if (spare_ndx > hashp->OVFL_POINT) {
872 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
873 hashp->OVFL_POINT = spare_ndx;
874 }
875
876 if (new_bucket > hashp->HIGH_MASK) {
877 /* Starting a new doubling */
878 hashp->LOW_MASK = hashp->HIGH_MASK;
879 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
880 }
881 /* Relocate records to the new bucket */
882 return (__split_page(hashp, old_bucket, new_bucket));
883 }
884
885 /*
886 * If realloc guarantees that the pointer is not destroyed if the realloc
887 * fails, then this routine can go away.
888 */
889 static void *
890 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
891 {
892 void *p;
893
894 if ( (p = malloc(newsize)) ) {
895 memmove(p, *p_ptr, oldsize);
896 memset((char *)p + oldsize, 0, newsize - oldsize);
897 free(*p_ptr);
898 *p_ptr = p;
899 }
900 return (p);
901 }
902
903 extern __uint32_t
904 __call_hash(HTAB *hashp,
905 char *k,
906 int len)
907 {
908 int32_t n, bucket;
909
910 n = hashp->hash(k, len);
911 bucket = n & hashp->HIGH_MASK;
912 if (bucket > hashp->MAX_BUCKET)
913 bucket = bucket & hashp->LOW_MASK;
914 return (bucket);
915 }
916
917 /*
918 * Allocate segment table. On error, destroy the table and set errno.
919 *
920 * Returns 0 on success
921 */
922 static int
923 alloc_segs(HTAB *hashp, int nsegs)
924 {
925 int i;
926 SEGMENT store;
927
928 int save_errno;
929
930 if ((hashp->dir =
931 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
932 save_errno = errno;
933 (void)hdestroy(hashp);
934 errno = save_errno;
935 return (-1);
936 }
937 /* Allocate segments */
938 if ((store =
939 (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
940 save_errno = errno;
941 (void)hdestroy(hashp);
942 errno = save_errno;
943 return (-1);
944 }
945 hashp->dir[0] = store;
946 for (i = 1; i < nsegs; i++, hashp->nsegs++)
947 hashp->dir[i] = &store[i << hashp->SSHIFT];
948 return (0);
949 }
950
951 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
952 /*
953 * Hashp->hdr needs to be byteswapped.
954 */
955 static void
956 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
957 {
958 int i;
959
960 P_32_COPY(srcp->magic, destp->magic);
961 P_32_COPY(srcp->version, destp->version);
962 P_32_COPY(srcp->lorder, destp->lorder);
963 P_32_COPY(srcp->bsize, destp->bsize);
964 P_32_COPY(srcp->bshift, destp->bshift);
965 P_32_COPY(srcp->dsize, destp->dsize);
966 P_32_COPY(srcp->ssize, destp->ssize);
967 P_32_COPY(srcp->sshift, destp->sshift);
968 P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
969 P_32_COPY(srcp->last_freed, destp->last_freed);
970 P_32_COPY(srcp->max_bucket, destp->max_bucket);
971 P_32_COPY(srcp->high_mask, destp->high_mask);
972 P_32_COPY(srcp->low_mask, destp->low_mask);
973 P_32_COPY(srcp->ffactor, destp->ffactor);
974 P_32_COPY(srcp->nkeys, destp->nkeys);
975 P_32_COPY(srcp->hdrpages, destp->hdrpages);
976 P_32_COPY(srcp->h_charkey, destp->h_charkey);
977 for (i = 0; i < NCACHED; i++) {
978 P_32_COPY(srcp->spares[i], destp->spares[i]);
979 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
980 }
981 }
982
983 static void
984 swap_header(HTAB *hashp)
985 {
986 HASHHDR *hdrp;
987 int i;
988
989 hdrp = &hashp->hdr;
990
991 M_32_SWAP(hdrp->magic);
992 M_32_SWAP(hdrp->version);
993 M_32_SWAP(hdrp->lorder);
994 M_32_SWAP(hdrp->bsize);
995 M_32_SWAP(hdrp->bshift);
996 M_32_SWAP(hdrp->dsize);
997 M_32_SWAP(hdrp->ssize);
998 M_32_SWAP(hdrp->sshift);
999 M_32_SWAP(hdrp->ovfl_point);
1000 M_32_SWAP(hdrp->last_freed);
1001 M_32_SWAP(hdrp->max_bucket);
1002 M_32_SWAP(hdrp->high_mask);
1003 M_32_SWAP(hdrp->low_mask);
1004 M_32_SWAP(hdrp->ffactor);
1005 M_32_SWAP(hdrp->nkeys);
1006 M_32_SWAP(hdrp->hdrpages);
1007 M_32_SWAP(hdrp->h_charkey);
1008 for (i = 0; i < NCACHED; i++) {
1009 M_32_SWAP(hdrp->spares[i]);
1010 M_16_SWAP(hdrp->bitmaps[i]);
1011 }
1012 }
1013 #endif
1014
