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 #define __LINUX_ERRNO_EXTENSIONS__
35 #include <sys/param.h>
36 #if defined(LIBC_SCCS) && !defined(lint)
37 static char sccsid[] = "@(#)hash.c 8.9 (Berkeley) 6/16/94";
38 #endif /* LIBC_SCCS and not lint */
39 #include <sys/cdefs.h>
40 #include <sys/types.h>
41
42 #include <sys/stat.h>
43
44 #include <errno.h>
45 #include <fcntl.h>
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <string.h>
49 #include <unistd.h>
50 #ifdef DEBUG
51 #include <assert.h>
52 #endif
53
54 #define __DBINTERFACE_PRIVATE /* activate prototypes from db_local.h */
55 #include "db_local.h"
56 #include "hash.h"
57 #include "page.h"
58 #include "extern.h"
59
60 static int alloc_segs(HTAB *, int);
61 static int flush_meta(HTAB *);
62 static int hash_access(HTAB *, HASH_ACTION, DBT *, DBT *);
63 static int hash_close(DB *);
64 static int hash_delete(const DB *, const DBT *, u_int);
65 static int hash_fd(const DB *);
66 static int hash_get(const DB *, const DBT *, DBT *, u_int);
67 static int hash_put(const DB *, DBT *, const DBT *, u_int);
68 static void *hash_realloc(SEGMENT **, int, int);
69 static int hash_seq(const DB *, DBT *, DBT *, u_int);
70 static int hash_sync(const DB *, u_int);
71 static int hdestroy(HTAB *);
72 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
73 static int init_htab(HTAB *, int);
74 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
75 static void swap_header(HTAB *);
76 static void swap_header_copy(HASHHDR *, HASHHDR *);
77 #endif
78
79 /* Macros for min/max. */
80 #ifndef MIN
81 #define MIN(a,b) (((a)<(b))?(a):(b))
82 #endif
83 #ifndef MAX
84 #define MAX(a,b) (((a)>(b))?(a):(b))
85 #endif
86
87 /* Fast arithmetic, relying on powers of 2, */
88 #define MOD(x, y) ((x) & ((y) - 1))
89
90 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; }
91
92 /* Return values */
93 #define SUCCESS (0)
94 #define ERROR (-1)
95 #define ABNORMAL (1)
96
97 #ifdef HASH_STATISTICS
98 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
99 #endif
100
101 /************************** INTERFACE ROUTINES ***************************/
102 /* OPEN/CLOSE */
103
104 extern DB *
__hash_open(const char * file,int flags,int mode,int dflags,const HASHINFO * info)105 __hash_open (const char *file,
106 int flags,
107 int mode,
108 int dflags,
109 const HASHINFO *info) /* Special directives for create */
110 {
111 HTAB *hashp;
112
113 (void) dflags;
114 #ifdef __USE_INTERNAL_STAT64
115 struct stat64 statbuf;
116 #else
117 struct stat statbuf;
118 #endif
119 DB *dbp;
120 int bpages, hdrsize, new_table, nsegs, save_errno;
121
122 if ((flags & O_ACCMODE) == O_WRONLY) {
123 errno = EINVAL;
124 return (NULL);
125 }
126
127 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
128 return (NULL);
129 hashp->fp = -1;
130
131 /*
132 * Even if user wants write only, we need to be able to read
133 * the actual file, so we need to open it read/write. But, the
134 * field in the hashp structure needs to be accurate so that
135 * we can check accesses.
136 */
137 hashp->flags = flags;
138
139 new_table = 0;
140 if (!file || (flags & O_TRUNC) ||
141 #ifdef __USE_INTERNAL_STAT64
142 (stat64(file, &statbuf) && (errno == ENOENT))) {
143 #else
144 (stat(file, &statbuf) && (errno == ENOENT))) {
145 #endif
146 if (errno == ENOENT)
147 errno = 0; /* Just in case someone looks at errno */
148 new_table = 1;
149 }
150 if (file) {
151 if ((hashp->fp = open(file, flags, mode)) == -1)
152 RETURN_ERROR(errno, error0);
153
154 /* if the .db file is empty, and we had permission to create
155 a new .db file, then reinitialize the database */
156 if ((flags & O_CREAT) &&
157 #ifdef __USE_INTERNAL_STAT64
158 fstat64(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
159 #else
160 fstat(hashp->fp, &statbuf) == 0 && statbuf.st_size == 0)
161 #endif
162 new_table = 1;
163
164 #ifdef _HAVE_FCNTL
165 (void)fcntl(hashp->fp, F_SETFD, 1);
166 #endif
167 }
168 if (new_table) {
169 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
170 RETURN_ERROR(errno, error1);
171 } else {
172 /* Table already exists */
173 if (info && info->hash)
174 hashp->hash = info->hash;
175 else
176 hashp->hash = __default_hash;
177
178 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
179 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
180 swap_header(hashp);
181 #endif
182 if (hdrsize == -1)
183 RETURN_ERROR(errno, error1);
184 if (hdrsize != sizeof(HASHHDR))
185 RETURN_ERROR(EFTYPE, error1);
186 /* Verify file type, versions and hash function */
187 if (hashp->MAGIC != HASHMAGIC)
188 RETURN_ERROR(EFTYPE, error1);
189 #define OLDHASHVERSION 1
190 if (hashp->HASH_VERSION != HASHVERSION &&
191 hashp->HASH_VERSION != OLDHASHVERSION)
192 RETURN_ERROR(EFTYPE, error1);
193 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
194 RETURN_ERROR(EFTYPE, error1);
195 /* Check bucket size isn't too big for target int. */
196 if (hashp->BSIZE > INT_MAX)
197 RETURN_ERROR(EFTYPE, error1);
198 /*
199 * Figure out how many segments we need. Max_Bucket is the
200 * maximum bucket number, so the number of buckets is
201 * max_bucket + 1.
202 */
203 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
204 hashp->SGSIZE;
205 hashp->nsegs = 0;
206 if (alloc_segs(hashp, nsegs))
207 /*
208 * If alloc_segs fails, table will have been destroyed
209 * and errno will have been set.
210 */
211 return (NULL);
212 /* Read in bitmaps */
213 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
214 (hashp->BSIZE << BYTE_SHIFT) - 1) >>
215 (hashp->BSHIFT + BYTE_SHIFT);
216
217 hashp->nmaps = bpages;
218 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(__uint32_t *));
219 }
220
221 /* Initialize Buffer Manager */
222 if (info && info->cachesize)
223 __buf_init(hashp, info->cachesize);
224 else
225 __buf_init(hashp, DEF_BUFSIZE);
226
227 hashp->new_file = new_table;
228 hashp->save_file = file && (hashp->flags & O_RDWR);
229 hashp->cbucket = -1;
230 if (!(dbp = (DB *)malloc(sizeof(DB)))) {
231 save_errno = errno;
232 hdestroy(hashp);
233 errno = save_errno;
234 return (NULL);
235 }
236 dbp->internal = hashp;
237 dbp->close = hash_close;
238 dbp->del = hash_delete;
239 dbp->fd = hash_fd;
240 dbp->get = hash_get;
241 dbp->put = hash_put;
242 dbp->seq = hash_seq;
243 dbp->sync = hash_sync;
244 dbp->type = DB_HASH;
245
246 #ifdef DEBUG
247 (void)fprintf(stderr,
248 "%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",
249 "init_htab:",
250 "TABLE POINTER ", hashp,
251 "BUCKET SIZE ", hashp->BSIZE,
252 "BUCKET SHIFT ", hashp->BSHIFT,
253 "DIRECTORY SIZE ", hashp->DSIZE,
254 "SEGMENT SIZE ", hashp->SGSIZE,
255 "SEGMENT SHIFT ", hashp->SSHIFT,
256 "FILL FACTOR ", hashp->FFACTOR,
257 "MAX BUCKET ", hashp->MAX_BUCKET,
258 "OVFL POINT ", hashp->OVFL_POINT,
259 "LAST FREED ", hashp->LAST_FREED,
260 "HIGH MASK ", hashp->HIGH_MASK,
261 "LOW MASK ", hashp->LOW_MASK,
262 "NSEGS ", hashp->nsegs,
263 "NKEYS ", hashp->NKEYS);
264 #endif
265 #ifdef HASH_STATISTICS
266 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
267 #endif
268 return (dbp);
269
270 error1:
271 if (hashp != NULL)
272 (void)close(hashp->fp);
273
274 error0:
275 free(hashp);
276 errno = save_errno;
277 return (NULL);
278 }
279
280 static int
281 hash_close(DB *dbp)
282 {
283 HTAB *hashp;
284 int retval;
285
286 if (!dbp)
287 return (ERROR);
288
289 hashp = (HTAB *)dbp->internal;
290 retval = hdestroy(hashp);
291 free(dbp);
292 return (retval);
293 }
294
295 static int
296 hash_fd(const DB *dbp)
297 {
298 HTAB *hashp;
299
300 if (!dbp)
301 return (ERROR);
302
303 hashp = (HTAB *)dbp->internal;
304 if (hashp->fp == -1) {
305 errno = ENOENT;
306 return (-1);
307 }
308 return (hashp->fp);
309 }
310
311 /************************** LOCAL CREATION ROUTINES **********************/
312 static HTAB *
313 init_hash(HTAB *hashp,
314 const char *file,
315 const HASHINFO *info)
316 {
317 #ifdef __USE_INTERNAL_STAT64
318 struct stat64 statbuf;
319 #else
320 struct stat statbuf;
321 #endif
322 int nelem;
323
324 nelem = 1;
325 hashp->NKEYS = 0;
326 hashp->LORDER = DB_BYTE_ORDER;
327 hashp->BSIZE = DEF_BUCKET_SIZE;
328 hashp->BSHIFT = DEF_BUCKET_SHIFT;
329 hashp->SGSIZE = DEF_SEGSIZE;
330 hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
331 hashp->DSIZE = DEF_DIRSIZE;
332 hashp->FFACTOR = DEF_FFACTOR;
333 hashp->hash = __default_hash;
334 memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
335 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
336
337 /* Fix bucket size to be optimal for file system */
338 if (file != NULL) {
339 #ifdef __USE_INTERNAL_STAT64
340 if (stat64(file, &statbuf))
341 #else
342 if (stat(file, &statbuf))
343 #endif
344 return (NULL);
345 hashp->BSIZE = MIN(statbuf.st_blksize, MAX_BSIZE);
346 hashp->BSHIFT = __log2(hashp->BSIZE);
347 }
348
349 if (info) {
350 if (info->bsize) {
351 /* Round pagesize up to power of 2 */
352 hashp->BSHIFT = __log2(info->bsize);
353 hashp->BSIZE = 1 << hashp->BSHIFT;
354 if (hashp->BSIZE > MAX_BSIZE) {
355 errno = EINVAL;
356 return (NULL);
357 }
358 }
359 if (info->ffactor)
360 hashp->FFACTOR = info->ffactor;
361 if (info->hash)
362 hashp->hash = info->hash;
363 if (info->nelem)
364 nelem = info->nelem;
365 if (info->lorder) {
366 if (info->lorder != DB_BIG_ENDIAN &&
367 info->lorder != DB_LITTLE_ENDIAN) {
368 errno = EINVAL;
369 return (NULL);
370 }
371 hashp->LORDER = info->lorder;
372 }
373 }
374 /* init_htab should destroy the table and set errno if it fails */
375 if (init_htab(hashp, nelem))
376 return (NULL);
377 else
378 return (hashp);
379 }
380 /*
381 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy
382 * the table and set errno, so we just pass the error information along.
383 *
384 * Returns 0 on No Error
385 */
386 static int
387 init_htab(HTAB *hashp, int nelem)
388 {
389 int nbuckets, nsegs;
390 int l2;
391
392 /*
393 * Divide number of elements by the fill factor and determine a
394 * desired number of buckets. Allocate space for the next greater
395 * power of two number of buckets.
396 */
397 nelem = (nelem - 1) / hashp->FFACTOR + 1;
398
399 l2 = __log2(MAX(nelem, 2));
400 nbuckets = 1 << l2;
401
402 hashp->SPARES[l2] = l2 + 1;
403 hashp->SPARES[l2 + 1] = l2 + 1;
404 hashp->OVFL_POINT = l2;
405 hashp->LAST_FREED = 2;
406
407 /* First bitmap page is at: splitpoint l2 page offset 1 */
408 if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
409 return (-1);
410
411 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
412 hashp->HIGH_MASK = (nbuckets << 1) - 1;
413 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
414 hashp->BSHIFT) + 1;
415
416 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
417 nsegs = 1 << __log2(nsegs);
418
419 if (nsegs > hashp->DSIZE)
420 hashp->DSIZE = nsegs;
421 return (alloc_segs(hashp, nsegs));
422 }
423
424 /********************** DESTROY/CLOSE ROUTINES ************************/
425
426 /*
427 * Flushes any changes to the file if necessary and destroys the hashp
428 * structure, freeing all allocated space.
429 */
430 static int
431 hdestroy(HTAB *hashp)
432 {
433 int i, save_errno;
434
435 save_errno = 0;
436
437 #ifdef HASH_STATISTICS
438 (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
439 hash_accesses, hash_collisions);
440 (void)fprintf(stderr, "hdestroy: expansions %ld\n",
441 hash_expansions);
442 (void)fprintf(stderr, "hdestroy: overflows %ld\n",
443 hash_overflows);
444 (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
445 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
446
447 for (i = 0; i < NCACHED; i++)
448 (void)fprintf(stderr,
449 "spares[%d] = %d\n", i, hashp->SPARES[i]);
450 #endif
451 /*
452 * Call on buffer manager to free buffers, and if required,
453 * write them to disk.
454 */
455 if (__buf_free(hashp, 1, hashp->save_file))
456 save_errno = errno;
457 if (hashp->dir) {
458 free(*hashp->dir); /* Free initial segments */
459 /* Free extra segments */
460 while (hashp->exsegs--)
461 free(hashp->dir[--hashp->nsegs]);
462 free(hashp->dir);
463 }
464 if (flush_meta(hashp) && !save_errno)
465 save_errno = errno;
466 /* Free Bigmaps */
467 for (i = 0; i < hashp->nmaps; i++)
468 if (hashp->mapp[i])
469 free(hashp->mapp[i]);
470
471 if (hashp->fp != -1)
472 (void)close(hashp->fp);
473
474 free(hashp);
475
476 if (save_errno) {
477 errno = save_errno;
478 return (ERROR);
479 }
480 return (SUCCESS);
481 }
482 /*
483 * Write modified pages to disk
484 *
485 * Returns:
486 * 0 == OK
487 * -1 ERROR
488 */
489 static int
490 hash_sync(const DB *dbp, u_int flags)
491 {
492 HTAB *hashp;
493
494 if (flags != 0) {
495 errno = EINVAL;
496 return (ERROR);
497 }
498
499 if (!dbp)
500 return (ERROR);
501
502 hashp = (HTAB *)dbp->internal;
503 if (!hashp->save_file)
504 return (0);
505 if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
506 return (ERROR);
507 hashp->new_file = 0;
508 return (0);
509 }
510
511 /*
512 * Returns:
513 * 0 == OK
514 * -1 indicates that errno should be set
515 */
516 static int
517 flush_meta(HTAB *hashp)
518 {
519 HASHHDR *whdrp;
520 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
521 HASHHDR whdr;
522 #endif
523 int fp, i, wsize;
524
525 if (!hashp->save_file)
526 return (0);
527 hashp->MAGIC = HASHMAGIC;
528 hashp->HASH_VERSION = HASHVERSION;
529 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
530
531 fp = hashp->fp;
532 whdrp = &hashp->hdr;
533 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
534 whdrp = &whdr;
535 swap_header_copy(&hashp->hdr, whdrp);
536 #endif
537 if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
538 ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
539 return (-1);
540 else
541 if (wsize != sizeof(HASHHDR)) {
542 errno = EFTYPE;
543 hashp->error = errno;
544 return (-1);
545 }
546 for (i = 0; i < NCACHED; i++)
547 if (hashp->mapp[i])
548 if (__put_page(hashp, (char *)hashp->mapp[i],
549 hashp->BITMAPS[i], 0, 1))
550 return (-1);
551 return (0);
552 }
553
554 /*******************************SEARCH ROUTINES *****************************/
555 /*
556 * All the access routines return
557 *
558 * Returns:
559 * 0 on SUCCESS
560 * 1 to indicate an external ERROR (i.e. key not found, etc)
561 * -1 to indicate an internal ERROR (i.e. out of memory, etc)
562 */
563 static int
564 hash_get(const DB *dbp,
565 const DBT *key,
566 DBT *data,
567 u_int flag)
568 {
569 HTAB *hashp;
570
571 hashp = (HTAB *)dbp->internal;
572 if (flag) {
573 hashp->error = errno = EINVAL;
574 return (ERROR);
575 }
576 return (hash_access(hashp, HASH_GET, (DBT *)key, data));
577 }
578
579 static int
580 hash_put(const DB *dbp,
581 DBT *key,
582 const DBT *data,
583 u_int flag)
584 {
585 HTAB *hashp;
586
587 hashp = (HTAB *)dbp->internal;
588 if (flag && flag != R_NOOVERWRITE) {
589 hashp->error = EINVAL;
590 errno = EINVAL;
591 return (ERROR);
592 }
593 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
594 hashp->error = errno = EPERM;
595 return (ERROR);
596 }
597 return (hash_access(hashp, flag == R_NOOVERWRITE ?
598 HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
599 }
600
601 static int
602 hash_delete(const DB *dbp,
603 const DBT *key,
604 u_int flag) /* Ignored */
605 {
606 HTAB *hashp;
607
608 hashp = (HTAB *)dbp->internal;
609 if (flag && flag != R_CURSOR) {
610 hashp->error = errno = EINVAL;
611 return (ERROR);
612 }
613 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
614 hashp->error = errno = EPERM;
615 return (ERROR);
616 }
617 return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
618 }
619
620 /*
621 * Assume that hashp has been set in wrapper routine.
622 */
623 static int
624 hash_access(HTAB *hashp,
625 HASH_ACTION action,
626 DBT *key,
627 DBT *val)
628 {
629 BUFHEAD *rbufp;
630 BUFHEAD *bufp, *save_bufp;
631 __uint16_t *bp;
632 int n, ndx, off, size;
633 char *kp;
634 __uint16_t pageno;
635
636 #ifdef HASH_STATISTICS
637 hash_accesses++;
638 #endif
639
640 off = hashp->BSIZE;
641 size = key->size;
642 kp = (char *)key->data;
643 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
644 if (!rbufp)
645 return (ERROR);
646 save_bufp = rbufp;
647
648 /* Pin the bucket chain */
649 rbufp->flags |= BUF_PIN;
650 for (bp = (__uint16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
651 if (bp[1] >= REAL_KEY) {
652 /* Real key/data pair */
653 if (size == off - (int) *bp &&
654 memcmp(kp, rbufp->page + *bp, size) == 0)
655 goto found;
656 off = bp[1];
657 #ifdef HASH_STATISTICS
658 hash_collisions++;
659 #endif
660 bp += 2;
661 ndx += 2;
662 } else if (bp[1] == OVFLPAGE) {
663 rbufp = __get_buf(hashp, *bp, rbufp, 0);
664 if (!rbufp) {
665 save_bufp->flags &= ~BUF_PIN;
666 return (ERROR);
667 }
668 /* FOR LOOP INIT */
669 bp = (__uint16_t *)rbufp->page;
670 n = *bp++;
671 ndx = 1;
672 off = hashp->BSIZE;
673 } else if (bp[1] < REAL_KEY) {
674 if ((ndx =
675 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
676 goto found;
677 if (ndx == -2) {
678 bufp = rbufp;
679 if (!(pageno =
680 __find_last_page(hashp, &bufp))) {
681 ndx = 0;
682 rbufp = bufp;
683 break; /* FOR */
684 }
685 rbufp = __get_buf(hashp, pageno, bufp, 0);
686 if (!rbufp) {
687 save_bufp->flags &= ~BUF_PIN;
688 return (ERROR);
689 }
690 /* FOR LOOP INIT */
691 bp = (__uint16_t *)rbufp->page;
692 n = *bp++;
693 ndx = 1;
694 off = hashp->BSIZE;
695 } else {
696 save_bufp->flags &= ~BUF_PIN;
697 return (ERROR);
698 }
699 }
700
701 /* Not found */
702 switch (action) {
703 case HASH_PUT:
704 case HASH_PUTNEW:
705 if (__addel(hashp, rbufp, key, val)) {
706 save_bufp->flags &= ~BUF_PIN;
707 return (ERROR);
708 } else {
709 save_bufp->flags &= ~BUF_PIN;
710 return (SUCCESS);
711 }
712 case HASH_GET:
713 case HASH_DELETE:
714 default:
715 save_bufp->flags &= ~BUF_PIN;
716 return (ABNORMAL);
717 }
718
719 found:
720 switch (action) {
721 case HASH_PUTNEW:
722 save_bufp->flags &= ~BUF_PIN;
723 return (ABNORMAL);
724 case HASH_GET:
725 bp = (__uint16_t *)rbufp->page;
726 if (bp[ndx + 1] < REAL_KEY) {
727 if (__big_return(hashp, rbufp, ndx, val, 0))
728 return (ERROR);
729 } else {
730 val->data = (u_char *)rbufp->page + (int)bp[ndx + 1];
731 val->size = bp[ndx] - bp[ndx + 1];
732 }
733 break;
734 case HASH_PUT:
735 if ((__delpair(hashp, rbufp, ndx)) ||
736 (__addel(hashp, rbufp, key, val))) {
737 save_bufp->flags &= ~BUF_PIN;
738 return (ERROR);
739 }
740 break;
741 case HASH_DELETE:
742 if (__delpair(hashp, rbufp, ndx))
743 return (ERROR);
744 break;
745 default:
746 abort();
747 }
748 save_bufp->flags &= ~BUF_PIN;
749 return (SUCCESS);
750 }
751
752 static int
753 hash_seq(const DB *dbp,
754 DBT *key,
755 DBT *data,
756 u_int flag)
757 {
758 int32_t bucket;
759 BUFHEAD *bufp;
760 HTAB *hashp;
761 __uint16_t *bp, ndx;
762
763 hashp = (HTAB *)dbp->internal;
764 if (flag && flag != R_FIRST && flag != R_NEXT) {
765 hashp->error = errno = EINVAL;
766 return (ERROR);
767 }
768 #ifdef HASH_STATISTICS
769 hash_accesses++;
770 #endif
771 if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
772 hashp->cbucket = 0;
773 hashp->cndx = 1;
774 hashp->cpage = NULL;
775 }
776
777 for (bp = NULL; !bp || !bp[0]; ) {
778 if (!(bufp = hashp->cpage)) {
779 for (bucket = hashp->cbucket;
780 bucket <= hashp->MAX_BUCKET;
781 bucket++, hashp->cndx = 1) {
782 bufp = __get_buf(hashp, bucket, NULL, 0);
783 if (!bufp)
784 return (ERROR);
785 hashp->cpage = bufp;
786 bp = (__uint16_t *)bufp->page;
787 if (bp[0])
788 break;
789 }
790 hashp->cbucket = bucket;
791 if (hashp->cbucket > hashp->MAX_BUCKET) {
792 hashp->cbucket = -1;
793 return (ABNORMAL);
794 }
795 } else
796 bp = (__uint16_t *)hashp->cpage->page;
797
798 #ifdef DEBUG
799 assert(bp);
800 assert(bufp);
801 #endif
802 while (bp[hashp->cndx + 1] == OVFLPAGE) {
803 bufp = hashp->cpage =
804 __get_buf(hashp, bp[hashp->cndx], bufp, 0);
805 if (!bufp)
806 return (ERROR);
807 bp = (__uint16_t *)(bufp->page);
808 hashp->cndx = 1;
809 }
810 if (!bp[0]) {
811 hashp->cpage = NULL;
812 ++hashp->cbucket;
813 }
814 }
815 ndx = hashp->cndx;
816 if (bp[ndx + 1] < REAL_KEY) {
817 if (__big_keydata(hashp, bufp, key, data, 1))
818 return (ERROR);
819 } else {
820 key->data = (u_char *)hashp->cpage->page + bp[ndx];
821 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
822 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
823 data->size = bp[ndx] - bp[ndx + 1];
824 ndx += 2;
825 if (ndx > bp[0]) {
826 hashp->cpage = NULL;
827 hashp->cbucket++;
828 hashp->cndx = 1;
829 } else
830 hashp->cndx = ndx;
831 }
832 return (SUCCESS);
833 }
834
835 /********************************* UTILITIES ************************/
836
837 /*
838 * Returns:
839 * 0 ==> OK
840 * -1 ==> Error
841 */
842 extern int
843 __expand_table(HTAB *hashp)
844 {
845 int32_t old_bucket, new_bucket;
846 int dirsize, new_segnum, spare_ndx;
847
848 #ifdef HASH_STATISTICS
849 hash_expansions++;
850 #endif
851 new_bucket = ++hashp->MAX_BUCKET;
852 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
853
854 new_segnum = new_bucket >> hashp->SSHIFT;
855
856 /* Check if we need a new segment */
857 if (new_segnum >= hashp->nsegs) {
858 /* Check if we need to expand directory */
859 if (new_segnum >= hashp->DSIZE) {
860 /* Reallocate directory */
861 dirsize = hashp->DSIZE * sizeof(SEGMENT *);
862 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
863 return (-1);
864 hashp->DSIZE = dirsize << 1;
865 }
866 SEGMENT seg = calloc(hashp->SGSIZE, sizeof(SEGMENT));
867 if (seg == NULL)
868 return (-1);
869 hashp->dir[new_segnum] = seg;
870 hashp->exsegs++;
871 hashp->nsegs++;
872 }
873 /*
874 * If the split point is increasing (MAX_BUCKET's log base 2
875 * * increases), we need to copy the current contents of the spare
876 * split bucket to the next bucket.
877 */
878 spare_ndx = __log2(hashp->MAX_BUCKET + 1);
879 if (spare_ndx > hashp->OVFL_POINT) {
880 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
881 hashp->OVFL_POINT = spare_ndx;
882 }
883
884 if (new_bucket > hashp->HIGH_MASK) {
885 /* Starting a new doubling */
886 hashp->LOW_MASK = hashp->HIGH_MASK;
887 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
888 }
889 /* Relocate records to the new bucket */
890 return (__split_page(hashp, old_bucket, new_bucket));
891 }
892
893 /*
894 * If realloc guarantees that the pointer is not destroyed if the realloc
895 * fails, then this routine can go away.
896 */
897 static void *
898 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
899 {
900 void *p;
901
902 if ( (p = malloc(newsize)) ) {
903 memmove(p, *p_ptr, oldsize);
904 memset((char *)p + oldsize, 0, newsize - oldsize);
905 free(*p_ptr);
906 *p_ptr = p;
907 }
908 return (p);
909 }
910
911 extern __uint32_t
912 __call_hash(HTAB *hashp,
913 char *k,
914 int len)
915 {
916 int32_t n, bucket;
917
918 n = hashp->hash(k, len);
919 bucket = n & hashp->HIGH_MASK;
920 if (bucket > hashp->MAX_BUCKET)
921 bucket = bucket & hashp->LOW_MASK;
922 return (bucket);
923 }
924
925 /*
926 * Allocate segment table. On error, destroy the table and set errno.
927 *
928 * Returns 0 on success
929 */
930 static int
931 alloc_segs(HTAB *hashp, int nsegs)
932 {
933 int i;
934 SEGMENT store;
935
936 int save_errno;
937
938 if ((hashp->dir =
939 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
940 save_errno = errno;
941 (void)hdestroy(hashp);
942 errno = save_errno;
943 return (-1);
944 }
945 /* Allocate segments */
946 if ((store =
947 (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
948 save_errno = errno;
949 (void)hdestroy(hashp);
950 errno = save_errno;
951 return (-1);
952 }
953 hashp->dir[0] = store;
954 for (i = 1; i < nsegs; i++, hashp->nsegs++)
955 hashp->dir[i] = &store[i << hashp->SSHIFT];
956 return (0);
957 }
958
959 #if (_BYTE_ORDER == _LITTLE_ENDIAN)
960 /*
961 * Hashp->hdr needs to be byteswapped.
962 */
963 static void
964 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
965 {
966 int i;
967
968 P_32_COPY(srcp->magic, destp->magic);
969 P_32_COPY(srcp->version, destp->version);
970 P_32_COPY(srcp->lorder, destp->lorder);
971 P_32_COPY(srcp->bsize, destp->bsize);
972 P_32_COPY(srcp->bshift, destp->bshift);
973 P_32_COPY(srcp->dsize, destp->dsize);
974 P_32_COPY(srcp->ssize, destp->ssize);
975 P_32_COPY(srcp->sshift, destp->sshift);
976 P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
977 P_32_COPY(srcp->last_freed, destp->last_freed);
978 P_32_COPY(srcp->max_bucket, destp->max_bucket);
979 P_32_COPY(srcp->high_mask, destp->high_mask);
980 P_32_COPY(srcp->low_mask, destp->low_mask);
981 P_32_COPY(srcp->ffactor, destp->ffactor);
982 P_32_COPY(srcp->nkeys, destp->nkeys);
983 P_32_COPY(srcp->hdrpages, destp->hdrpages);
984 P_32_COPY(srcp->h_charkey, destp->h_charkey);
985 for (i = 0; i < NCACHED; i++) {
986 P_32_COPY(srcp->spares[i], destp->spares[i]);
987 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
988 }
989 }
990
991 static void
992 swap_header(HTAB *hashp)
993 {
994 HASHHDR *hdrp;
995 int i;
996
997 hdrp = &hashp->hdr;
998
999 M_32_SWAP(hdrp->magic);
1000 M_32_SWAP(hdrp->version);
1001 M_32_SWAP(hdrp->lorder);
1002 M_32_SWAP(hdrp->bsize);
1003 M_32_SWAP(hdrp->bshift);
1004 M_32_SWAP(hdrp->dsize);
1005 M_32_SWAP(hdrp->ssize);
1006 M_32_SWAP(hdrp->sshift);
1007 M_32_SWAP(hdrp->ovfl_point);
1008 M_32_SWAP(hdrp->last_freed);
1009 M_32_SWAP(hdrp->max_bucket);
1010 M_32_SWAP(hdrp->high_mask);
1011 M_32_SWAP(hdrp->low_mask);
1012 M_32_SWAP(hdrp->ffactor);
1013 M_32_SWAP(hdrp->nkeys);
1014 M_32_SWAP(hdrp->hdrpages);
1015 M_32_SWAP(hdrp->h_charkey);
1016 for (i = 0; i < NCACHED; i++) {
1017 M_32_SWAP(hdrp->spares[i]);
1018 M_16_SWAP(hdrp->bitmaps[i]);
1019 }
1020 }
1021 #endif
1022
