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 * @(#)hash.c 8.9 (Berkeley) 6/16/94
33 * $FreeBSD: head/lib/libc/db/hash/hash.c 206178 2010-04-05 10:12:21Z avg $
34 */
35
36 #include "namespace.h"
37 #include <sys/param.h>
38 #include <sys/stat.h>
39
40 #include <errno.h>
41 #include <fcntl.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <string.h>
45 #include <unistd.h>
46 #ifdef DEBUG
47 #include <assert.h>
48 #endif
49 #include "un-namespace.h"
50
51 #include <db.h>
52 #include "hash.h"
53 #include "page.h"
54 #include "extern.h"
55
56 static int alloc_segs(HTAB *, int);
57 static int flush_meta(HTAB *);
58 static int hash_access(HTAB *, ACTION, DBT *, DBT *);
59 static int hash_close(DB *);
60 static int hash_delete(const DB *, const DBT *, uint32_t);
61 static int hash_fd(const DB *);
62 static int hash_get(const DB *, const DBT *, DBT *, uint32_t);
63 static int hash_put(const DB *, DBT *, const DBT *, uint32_t);
64 static void *hash_realloc(SEGMENT **, int, int);
65 static int hash_seq(const DB *, DBT *, DBT *, uint32_t);
66 static int hash_sync(const DB *, uint32_t);
67 static int hdestroy(HTAB *);
68 static HTAB *init_hash(HTAB *, const char *, const HASHINFO *);
69 static int init_htab(HTAB *, int);
70 #if BYTE_ORDER == LITTLE_ENDIAN
71 static void swap_header(HTAB *);
72 static void swap_header_copy(HASHHDR *, HASHHDR *);
73 #endif
74
75 /* Fast arithmetic, relying on powers of 2, */
76 #define MOD(x, y) ((x) & ((y) - 1))
77
78 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; }
79
80 /* Return values */
81 #define SUCCESS (0)
82 #define ERROR (-1)
83 #define ABNORMAL (1)
84
85 #ifdef HASH_STATISTICS
86 int hash_accesses, hash_collisions, hash_expansions, hash_overflows;
87 #endif
88
89 /************************** INTERFACE ROUTINES ***************************/
90 /* OPEN/CLOSE */
91
92 /* ARGSUSED */
93 DB *
__hash_open(const char * file,int flags,mode_t mode,const HASHINFO * info,__unused int dflags)94 __hash_open(const char *file, int flags, mode_t mode,
95 const HASHINFO *info, /* Special directives for create */
96 __unused int dflags)
97 {
98 HTAB *hashp;
99 struct stat statbuf;
100 DB *dbp;
101 int bpages, hdrsize, new_table, nsegs, save_errno;
102
103 if ((flags & O_ACCMODE) == O_WRONLY) {
104 errno = EINVAL;
105 return (NULL);
106 }
107
108 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
109 return (NULL);
110 hashp->fp = -1;
111
112 /*
113 * Even if user wants write only, we need to be able to read
114 * the actual file, so we need to open it read/write. But, the
115 * field in the hashp structure needs to be accurate so that
116 * we can check accesses.
117 */
118 hashp->flags = flags;
119
120 if (file) {
121 if ((hashp->fp = _open(file, flags | O_CLOEXEC, mode)) == -1)
122 RETURN_ERROR(errno, error0);
123 _fcntl(hashp->fp, F_SETFD, 1);
124 new_table = _fstat(hashp->fp, &statbuf) == 0 &&
125 statbuf.st_size == 0 && (flags & O_ACCMODE) != O_RDONLY;
126 } else
127 new_table = 1;
128
129 if (new_table) {
130 if (!(hashp = init_hash(hashp, file, info)))
131 RETURN_ERROR(errno, error1);
132 } else {
133 /* Table already exists */
134 if (info && info->hash)
135 hashp->hash = info->hash;
136 else
137 hashp->hash = __default_hash;
138
139 hdrsize = _read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
140 #if BYTE_ORDER == LITTLE_ENDIAN
141 swap_header(hashp);
142 #endif
143 if (hdrsize == -1)
144 RETURN_ERROR(errno, error1);
145 if (hdrsize != sizeof(HASHHDR))
146 RETURN_ERROR(EFTYPE, error1);
147 /* Verify file type, versions and hash function */
148 if (hashp->MAGIC != HASHMAGIC)
149 RETURN_ERROR(EFTYPE, error1);
150 #define OLDHASHVERSION 1
151 if (hashp->VERSION != HASHVERSION &&
152 hashp->VERSION != OLDHASHVERSION)
153 RETURN_ERROR(EFTYPE, error1);
154 if ((int32_t)hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
155 RETURN_ERROR(EFTYPE, error1);
156 /*
157 * Figure out how many segments we need. Max_Bucket is the
158 * maximum bucket number, so the number of buckets is
159 * max_bucket + 1.
160 */
161 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
162 hashp->SGSIZE;
163 if (alloc_segs(hashp, nsegs))
164 /*
165 * If alloc_segs fails, table will have been destroyed
166 * and errno will have been set.
167 */
168 return (NULL);
169 /* Read in bitmaps */
170 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
171 (hashp->BSIZE << BYTE_SHIFT) - 1) >>
172 (hashp->BSHIFT + BYTE_SHIFT);
173
174 hashp->nmaps = bpages;
175 memset(&hashp->mapp[0], 0, bpages * sizeof(uint32_t *));
176 }
177
178 /* Initialize Buffer Manager */
179 if (info && info->cachesize)
180 __buf_init(hashp, info->cachesize);
181 else
182 __buf_init(hashp, DEF_BUFSIZE);
183
184 hashp->new_file = new_table;
185 hashp->save_file = file && (hashp->flags & O_RDWR);
186 hashp->cbucket = -1;
187 if (!(dbp = (DB *)malloc(sizeof(DB)))) {
188 save_errno = errno;
189 hdestroy(hashp);
190 errno = save_errno;
191 return (NULL);
192 }
193 dbp->internal = hashp;
194 dbp->close = hash_close;
195 dbp->del = hash_delete;
196 dbp->fd = hash_fd;
197 dbp->get = hash_get;
198 dbp->put = hash_put;
199 dbp->seq = hash_seq;
200 dbp->sync = hash_sync;
201 dbp->type = DB_HASH;
202
203 #ifdef DEBUG
204 fprintf(stderr,
205 "%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",
206 "init_htab:",
207 "TABLE POINTER ", hashp,
208 "BUCKET SIZE ", hashp->BSIZE,
209 "BUCKET SHIFT ", hashp->BSHIFT,
210 "DIRECTORY SIZE ", hashp->DSIZE,
211 "SEGMENT SIZE ", hashp->SGSIZE,
212 "SEGMENT SHIFT ", hashp->SSHIFT,
213 "FILL FACTOR ", hashp->FFACTOR,
214 "MAX BUCKET ", hashp->MAX_BUCKET,
215 "OVFL POINT ", hashp->OVFL_POINT,
216 "LAST FREED ", hashp->LAST_FREED,
217 "HIGH MASK ", hashp->HIGH_MASK,
218 "LOW MASK ", hashp->LOW_MASK,
219 "NSEGS ", hashp->nsegs,
220 "NKEYS ", hashp->NKEYS);
221 #endif
222 #ifdef HASH_STATISTICS
223 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
224 #endif
225 return (dbp);
226
227 error1:
228 if (hashp != NULL)
229 _close(hashp->fp);
230
231 error0:
232 free(hashp);
233 errno = save_errno;
234 return (NULL);
235 }
236
237 static int
hash_close(DB * dbp)238 hash_close(DB *dbp)
239 {
240 HTAB *hashp;
241 int retval;
242
243 if (!dbp)
244 return (ERROR);
245
246 hashp = (HTAB *)dbp->internal;
247 retval = hdestroy(hashp);
248 free(dbp);
249 return (retval);
250 }
251
252 static int
hash_fd(const DB * dbp)253 hash_fd(const DB *dbp)
254 {
255 HTAB *hashp;
256
257 if (!dbp)
258 return (ERROR);
259
260 hashp = (HTAB *)dbp->internal;
261 if (hashp->fp == -1) {
262 errno = ENOENT;
263 return (-1);
264 }
265 return (hashp->fp);
266 }
267
268 /************************** LOCAL CREATION ROUTINES **********************/
269 static HTAB *
init_hash(HTAB * hashp,const char * file,const HASHINFO * info)270 init_hash(HTAB *hashp, const char *file, const HASHINFO *info)
271 {
272 struct stat statbuf;
273 int nelem;
274
275 nelem = 1;
276 hashp->NKEYS = 0;
277 hashp->LORDER = BYTE_ORDER;
278 hashp->BSIZE = DEF_BUCKET_SIZE;
279 hashp->BSHIFT = DEF_BUCKET_SHIFT;
280 hashp->SGSIZE = DEF_SEGSIZE;
281 hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
282 hashp->DSIZE = DEF_DIRSIZE;
283 hashp->FFACTOR = DEF_FFACTOR;
284 hashp->hash = __default_hash;
285 memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
286 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
287
288 /* Fix bucket size to be optimal for file system */
289 if (file != NULL) {
290 if (stat(file, &statbuf))
291 return (NULL);
292 hashp->BSIZE = NOM_BSIZE;
293 if (hashp->BSIZE > MAX_BSIZE)
294 hashp->BSIZE = MAX_BSIZE;
295 hashp->BSHIFT = __log2(hashp->BSIZE);
296 }
297
298 if (info) {
299 if (info->bsize) {
300 /* Round pagesize up to power of 2 */
301 hashp->BSHIFT = __log2(info->bsize);
302 hashp->BSIZE = 1 << hashp->BSHIFT;
303 if (hashp->BSIZE > MAX_BSIZE) {
304 errno = EINVAL;
305 return (NULL);
306 }
307 }
308 if (info->ffactor)
309 hashp->FFACTOR = info->ffactor;
310 if (info->hash)
311 hashp->hash = info->hash;
312 if (info->nelem)
313 nelem = info->nelem;
314 if (info->lorder) {
315 if (info->lorder != BIG_ENDIAN &&
316 info->lorder != LITTLE_ENDIAN) {
317 errno = EINVAL;
318 return (NULL);
319 }
320 hashp->LORDER = info->lorder;
321 }
322 }
323 /* init_htab should destroy the table and set errno if it fails */
324 if (init_htab(hashp, nelem))
325 return (NULL);
326 else
327 return (hashp);
328 }
329 /*
330 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy
331 * the table and set errno, so we just pass the error information along.
332 *
333 * Returns 0 on No Error
334 */
335 static int
init_htab(HTAB * hashp,int nelem)336 init_htab(HTAB *hashp, int nelem)
337 {
338 int nbuckets, nsegs, l2;
339
340 /*
341 * Divide number of elements by the fill factor and determine a
342 * desired number of buckets. Allocate space for the next greater
343 * power of two number of buckets.
344 */
345 nelem = (nelem - 1) / hashp->FFACTOR + 1;
346
347 l2 = __log2(MAX(nelem, 2));
348 nbuckets = 1 << l2;
349
350 hashp->SPARES[l2] = l2 + 1;
351 hashp->SPARES[l2 + 1] = l2 + 1;
352 hashp->OVFL_POINT = l2;
353 hashp->LAST_FREED = 2;
354
355 /* First bitmap page is at: splitpoint l2 page offset 1 */
356 if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
357 return (-1);
358
359 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
360 hashp->HIGH_MASK = (nbuckets << 1) - 1;
361 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
362 hashp->BSHIFT) + 1;
363
364 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
365 nsegs = 1 << __log2(nsegs);
366
367 if (nsegs > hashp->DSIZE)
368 hashp->DSIZE = nsegs;
369 return (alloc_segs(hashp, nsegs));
370 }
371
372 /********************** DESTROY/CLOSE ROUTINES ************************/
373
374 /*
375 * Flushes any changes to the file if necessary and destroys the hashp
376 * structure, freeing all allocated space.
377 */
378 static int
hdestroy(HTAB * hashp)379 hdestroy(HTAB *hashp)
380 {
381 int i, save_errno;
382
383 save_errno = 0;
384
385 #ifdef HASH_STATISTICS
386 fprintf(stderr, "hdestroy: accesses %d collisions %d\n",
387 hash_accesses, hash_collisions);
388 fprintf(stderr, "hdestroy: expansions %d\n",
389 hash_expansions);
390 fprintf(stderr, "hdestroy: overflows %d\n",
391 hash_overflows);
392 fprintf(stderr, "keys %d maxp %d segmentcount %d\n",
393 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
394
395 for (i = 0; i < NCACHED; i++)
396 fprintf(stderr,
397 "spares[%d] = %d\n", i, hashp->SPARES[i]);
398 #endif
399 /*
400 * Call on buffer manager to free buffers, and if required,
401 * write them to disk.
402 */
403 if (__buf_free(hashp, 1, hashp->save_file))
404 save_errno = errno;
405 if (hashp->dir) {
406 free(*hashp->dir); /* Free initial segments */
407 /* Free extra segments */
408 while (hashp->exsegs--)
409 free(hashp->dir[--hashp->nsegs]);
410 free(hashp->dir);
411 }
412 if (flush_meta(hashp) && !save_errno)
413 save_errno = errno;
414 /* Free Bigmaps */
415 for (i = 0; i < hashp->nmaps; i++)
416 if (hashp->mapp[i])
417 free(hashp->mapp[i]);
418 if (hashp->tmp_key)
419 free(hashp->tmp_key);
420 if (hashp->tmp_buf)
421 free(hashp->tmp_buf);
422
423 if (hashp->fp != -1)
424 _close(hashp->fp);
425
426 free(hashp);
427
428 if (save_errno) {
429 errno = save_errno;
430 return (ERROR);
431 }
432 return (SUCCESS);
433 }
434 /*
435 * Write modified pages to disk
436 *
437 * Returns:
438 * 0 == OK
439 * -1 ERROR
440 */
441 static int
hash_sync(const DB * dbp,uint32_t flags)442 hash_sync(const DB *dbp, uint32_t flags)
443 {
444 HTAB *hashp;
445
446 if (flags != 0) {
447 errno = EINVAL;
448 return (ERROR);
449 }
450
451 if (!dbp)
452 return (ERROR);
453
454 hashp = (HTAB *)dbp->internal;
455 if (!hashp->save_file)
456 return (0);
457 if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
458 return (ERROR);
459 hashp->new_file = 0;
460 return (0);
461 }
462
463 /*
464 * Returns:
465 * 0 == OK
466 * -1 indicates that errno should be set
467 */
468 static int
flush_meta(HTAB * hashp)469 flush_meta(HTAB *hashp)
470 {
471 HASHHDR *whdrp;
472 #if BYTE_ORDER == LITTLE_ENDIAN
473 HASHHDR whdr;
474 #endif
475 int fp, i, wsize;
476
477 if (!hashp->save_file)
478 return (0);
479 hashp->MAGIC = HASHMAGIC;
480 hashp->VERSION = HASHVERSION;
481 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
482
483 fp = hashp->fp;
484 whdrp = &hashp->hdr;
485 #if BYTE_ORDER == LITTLE_ENDIAN
486 whdrp = &whdr;
487 swap_header_copy(&hashp->hdr, whdrp);
488 #endif
489 if ((wsize = pwrite(fp, whdrp, sizeof(HASHHDR), (off_t)0)) == -1)
490 return (-1);
491 else
492 if (wsize != sizeof(HASHHDR)) {
493 errno = EFTYPE;
494 hashp->error = errno;
495 return (-1);
496 }
497 for (i = 0; i < NCACHED; i++)
498 if (hashp->mapp[i])
499 if (__put_page(hashp, (char *)hashp->mapp[i],
500 hashp->BITMAPS[i], 0, 1))
501 return (-1);
502 return (0);
503 }
504
505 /*******************************SEARCH ROUTINES *****************************/
506 /*
507 * All the access routines return
508 *
509 * Returns:
510 * 0 on SUCCESS
511 * 1 to indicate an external ERROR (i.e. key not found, etc)
512 * -1 to indicate an internal ERROR (i.e. out of memory, etc)
513 */
514 static int
hash_get(const DB * dbp,const DBT * key,DBT * data,uint32_t flag)515 hash_get(const DB *dbp, const DBT *key, DBT *data, uint32_t flag)
516 {
517 HTAB *hashp;
518
519 hashp = (HTAB *)dbp->internal;
520 if (flag) {
521 hashp->error = errno = EINVAL;
522 return (ERROR);
523 }
524 return (hash_access(hashp, HASH_GET, (DBT *)key, data));
525 }
526
527 static int
hash_put(const DB * dbp,DBT * key,const DBT * data,uint32_t flag)528 hash_put(const DB *dbp, DBT *key, const DBT *data, uint32_t flag)
529 {
530 HTAB *hashp;
531
532 hashp = (HTAB *)dbp->internal;
533 if (flag && flag != R_NOOVERWRITE) {
534 hashp->error = errno = EINVAL;
535 return (ERROR);
536 }
537 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
538 hashp->error = errno = EPERM;
539 return (ERROR);
540 }
541 return (hash_access(hashp, flag == R_NOOVERWRITE ?
542 HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
543 }
544
545 static int
hash_delete(const DB * dbp,const DBT * key,uint32_t flag)546 hash_delete(const DB *dbp, const DBT *key,
547 uint32_t flag) /* Ignored */
548 {
549 HTAB *hashp;
550
551 hashp = (HTAB *)dbp->internal;
552 if (flag && flag != R_CURSOR) {
553 hashp->error = errno = EINVAL;
554 return (ERROR);
555 }
556 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
557 hashp->error = errno = EPERM;
558 return (ERROR);
559 }
560 return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
561 }
562
563 /*
564 * Assume that hashp has been set in wrapper routine.
565 */
566 static int
hash_access(HTAB * hashp,ACTION action,DBT * key,DBT * val)567 hash_access(HTAB *hashp, ACTION action, DBT *key, DBT *val)
568 {
569 BUFHEAD *rbufp;
570 BUFHEAD *bufp, *save_bufp;
571 uint16_t *bp;
572 int n, ndx, off, size;
573 char *kp;
574 uint16_t pageno;
575
576 #ifdef HASH_STATISTICS
577 hash_accesses++;
578 #endif
579
580 off = hashp->BSIZE;
581 size = key->size;
582 kp = (char *)key->data;
583 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
584 if (!rbufp)
585 return (ERROR);
586 save_bufp = rbufp;
587
588 /* Pin the bucket chain */
589 rbufp->flags |= BUF_PIN;
590 for (bp = (uint16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
591 if (bp[1] >= REAL_KEY) {
592 /* Real key/data pair */
593 if (size == off - *bp &&
594 memcmp(kp, rbufp->page + *bp, size) == 0)
595 goto found;
596 off = bp[1];
597 #ifdef HASH_STATISTICS
598 hash_collisions++;
599 #endif
600 bp += 2;
601 ndx += 2;
602 } else if (bp[1] == OVFLPAGE) {
603 rbufp = __get_buf(hashp, *bp, rbufp, 0);
604 if (!rbufp) {
605 save_bufp->flags &= ~BUF_PIN;
606 return (ERROR);
607 }
608 /* FOR LOOP INIT */
609 bp = (uint16_t *)rbufp->page;
610 n = *bp++;
611 ndx = 1;
612 off = hashp->BSIZE;
613 } else if (bp[1] < REAL_KEY) {
614 if ((ndx =
615 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
616 goto found;
617 if (ndx == -2) {
618 bufp = rbufp;
619 if (!(pageno =
620 __find_last_page(hashp, &bufp))) {
621 ndx = 0;
622 rbufp = bufp;
623 break; /* FOR */
624 }
625 rbufp = __get_buf(hashp, pageno, bufp, 0);
626 if (!rbufp) {
627 save_bufp->flags &= ~BUF_PIN;
628 return (ERROR);
629 }
630 /* FOR LOOP INIT */
631 bp = (uint16_t *)rbufp->page;
632 n = *bp++;
633 ndx = 1;
634 off = hashp->BSIZE;
635 } else {
636 save_bufp->flags &= ~BUF_PIN;
637 return (ERROR);
638 }
639 }
640
641 /* Not found */
642 switch (action) {
643 case HASH_PUT:
644 case HASH_PUTNEW:
645 if (__addel(hashp, rbufp, key, val)) {
646 save_bufp->flags &= ~BUF_PIN;
647 return (ERROR);
648 } else {
649 save_bufp->flags &= ~BUF_PIN;
650 return (SUCCESS);
651 }
652 case HASH_GET:
653 case HASH_DELETE:
654 default:
655 save_bufp->flags &= ~BUF_PIN;
656 return (ABNORMAL);
657 }
658
659 found:
660 switch (action) {
661 case HASH_PUTNEW:
662 save_bufp->flags &= ~BUF_PIN;
663 return (ABNORMAL);
664 case HASH_GET:
665 bp = (uint16_t *)rbufp->page;
666 if (bp[ndx + 1] < REAL_KEY) {
667 if (__big_return(hashp, rbufp, ndx, val, 0))
668 return (ERROR);
669 } else {
670 val->data = (unsigned char *)rbufp->page +
671 (int)bp[ndx + 1];
672 val->size = bp[ndx] - bp[ndx + 1];
673 }
674 break;
675 case HASH_PUT:
676 if ((__delpair(hashp, rbufp, ndx)) ||
677 (__addel(hashp, rbufp, key, val))) {
678 save_bufp->flags &= ~BUF_PIN;
679 return (ERROR);
680 }
681 break;
682 case HASH_DELETE:
683 if (__delpair(hashp, rbufp, ndx))
684 return (ERROR);
685 break;
686 default:
687 abort();
688 }
689 save_bufp->flags &= ~BUF_PIN;
690 return (SUCCESS);
691 }
692
693 static int
hash_seq(const DB * dbp,DBT * key,DBT * data,uint32_t flag)694 hash_seq(const DB *dbp, DBT *key, DBT *data, uint32_t flag)
695 {
696 uint32_t bucket;
697 BUFHEAD *bufp;
698 HTAB *hashp;
699 uint16_t *bp, ndx;
700
701 hashp = (HTAB *)dbp->internal;
702 if (flag && flag != R_FIRST && flag != R_NEXT) {
703 hashp->error = errno = EINVAL;
704 return (ERROR);
705 }
706 #ifdef HASH_STATISTICS
707 hash_accesses++;
708 #endif
709 if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
710 hashp->cbucket = 0;
711 hashp->cndx = 1;
712 hashp->cpage = NULL;
713 }
714 next_bucket:
715 for (bp = NULL; !bp || !bp[0]; ) {
716 if (!(bufp = hashp->cpage)) {
717 for (bucket = hashp->cbucket;
718 bucket <= hashp->MAX_BUCKET;
719 bucket++, hashp->cndx = 1) {
720 bufp = __get_buf(hashp, bucket, NULL, 0);
721 if (!bufp)
722 return (ERROR);
723 hashp->cpage = bufp;
724 bp = (uint16_t *)bufp->page;
725 if (bp[0])
726 break;
727 }
728 hashp->cbucket = bucket;
729 if ((uint32_t)hashp->cbucket > hashp->MAX_BUCKET) {
730 hashp->cbucket = -1;
731 return (ABNORMAL);
732 }
733 } else {
734 bp = (uint16_t *)hashp->cpage->page;
735 if (flag == R_NEXT || flag == 0) {
736 hashp->cndx += 2;
737 if (hashp->cndx > bp[0]) {
738 hashp->cpage = NULL;
739 hashp->cbucket++;
740 hashp->cndx = 1;
741 goto next_bucket;
742 }
743 }
744 }
745
746 #ifdef DEBUG
747 assert(bp);
748 assert(bufp);
749 #endif
750 while (bp[hashp->cndx + 1] == OVFLPAGE) {
751 bufp = hashp->cpage =
752 __get_buf(hashp, bp[hashp->cndx], bufp, 0);
753 if (!bufp)
754 return (ERROR);
755 bp = (uint16_t *)(bufp->page);
756 hashp->cndx = 1;
757 }
758 if (!bp[0]) {
759 hashp->cpage = NULL;
760 ++hashp->cbucket;
761 }
762 }
763 ndx = hashp->cndx;
764 if (bp[ndx + 1] < REAL_KEY) {
765 if (__big_keydata(hashp, bufp, key, data, 1))
766 return (ERROR);
767 } else {
768 if (hashp->cpage == 0)
769 return (ERROR);
770 key->data = (unsigned char *)hashp->cpage->page + bp[ndx];
771 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
772 data->data = (unsigned char *)hashp->cpage->page + bp[ndx + 1];
773 data->size = bp[ndx] - bp[ndx + 1];
774 }
775 return (SUCCESS);
776 }
777
778 /********************************* UTILITIES ************************/
779
780 /*
781 * Returns:
782 * 0 ==> OK
783 * -1 ==> Error
784 */
785 int
__expand_table(HTAB * hashp)786 __expand_table(HTAB *hashp)
787 {
788 uint32_t old_bucket, new_bucket;
789 int dirsize, new_segnum, spare_ndx;
790
791 #ifdef HASH_STATISTICS
792 hash_expansions++;
793 #endif
794 new_bucket = ++hashp->MAX_BUCKET;
795 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
796
797 new_segnum = new_bucket >> hashp->SSHIFT;
798
799 /* Check if we need a new segment */
800 if (new_segnum >= hashp->nsegs) {
801 /* Check if we need to expand directory */
802 if (new_segnum >= hashp->DSIZE) {
803 /* Reallocate directory */
804 dirsize = hashp->DSIZE * sizeof(SEGMENT *);
805 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
806 return (-1);
807 hashp->DSIZE = dirsize << 1;
808 }
809 if ((hashp->dir[new_segnum] =
810 (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
811 return (-1);
812 hashp->exsegs++;
813 hashp->nsegs++;
814 }
815 /*
816 * If the split point is increasing (MAX_BUCKET's log base 2
817 * * increases), we need to copy the current contents of the spare
818 * split bucket to the next bucket.
819 */
820 spare_ndx = __log2(hashp->MAX_BUCKET + 1);
821 if (spare_ndx > hashp->OVFL_POINT) {
822 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
823 hashp->OVFL_POINT = spare_ndx;
824 }
825
826 if (new_bucket > hashp->HIGH_MASK) {
827 /* Starting a new doubling */
828 hashp->LOW_MASK = hashp->HIGH_MASK;
829 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
830 }
831 /* Relocate records to the new bucket */
832 return (__split_page(hashp, old_bucket, new_bucket));
833 }
834
835 /*
836 * If realloc guarantees that the pointer is not destroyed if the realloc
837 * fails, then this routine can go away.
838 */
839 static void *
hash_realloc(SEGMENT ** p_ptr,int oldsize,int newsize)840 hash_realloc(SEGMENT **p_ptr, int oldsize, int newsize)
841 {
842 void *p;
843
844 if ( (p = malloc(newsize)) ) {
845 memmove(p, *p_ptr, oldsize);
846 memset((char *)p + oldsize, 0, newsize - oldsize);
847 free(*p_ptr);
848 *p_ptr = p;
849 }
850 return (p);
851 }
852
853 uint32_t
__call_hash(HTAB * hashp,char * k,int len)854 __call_hash(HTAB *hashp, char *k, int len)
855 {
856 unsigned int n, bucket;
857
858 n = hashp->hash(k, len);
859 bucket = n & hashp->HIGH_MASK;
860 if (bucket > hashp->MAX_BUCKET)
861 bucket = bucket & hashp->LOW_MASK;
862 return (bucket);
863 }
864
865 /*
866 * Allocate segment table. On error, destroy the table and set errno.
867 *
868 * Returns 0 on success
869 */
870 static int
alloc_segs(HTAB * hashp,int nsegs)871 alloc_segs(HTAB *hashp, int nsegs)
872 {
873 int i;
874 SEGMENT store;
875
876 int save_errno;
877
878 if ((hashp->dir =
879 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
880 save_errno = errno;
881 hdestroy(hashp);
882 errno = save_errno;
883 return (-1);
884 }
885 hashp->nsegs = nsegs;
886 if (nsegs == 0)
887 return (0);
888 /* Allocate segments */
889 if ((store = (SEGMENT)calloc(nsegs << hashp->SSHIFT,
890 sizeof(SEGMENT))) == NULL) {
891 save_errno = errno;
892 hdestroy(hashp);
893 errno = save_errno;
894 return (-1);
895 }
896 for (i = 0; i < nsegs; i++)
897 hashp->dir[i] = &store[i << hashp->SSHIFT];
898 return (0);
899 }
900
901 #if BYTE_ORDER == LITTLE_ENDIAN
902 /*
903 * Hashp->hdr needs to be byteswapped.
904 */
905 static void
swap_header_copy(HASHHDR * srcp,HASHHDR * destp)906 swap_header_copy(HASHHDR *srcp, HASHHDR *destp)
907 {
908 int i;
909
910 P_32_COPY(srcp->magic, destp->magic);
911 P_32_COPY(srcp->version, destp->version);
912 P_32_COPY(srcp->lorder, destp->lorder);
913 P_32_COPY(srcp->bsize, destp->bsize);
914 P_32_COPY(srcp->bshift, destp->bshift);
915 P_32_COPY(srcp->dsize, destp->dsize);
916 P_32_COPY(srcp->ssize, destp->ssize);
917 P_32_COPY(srcp->sshift, destp->sshift);
918 P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
919 P_32_COPY(srcp->last_freed, destp->last_freed);
920 P_32_COPY(srcp->max_bucket, destp->max_bucket);
921 P_32_COPY(srcp->high_mask, destp->high_mask);
922 P_32_COPY(srcp->low_mask, destp->low_mask);
923 P_32_COPY(srcp->ffactor, destp->ffactor);
924 P_32_COPY(srcp->nkeys, destp->nkeys);
925 P_32_COPY(srcp->hdrpages, destp->hdrpages);
926 P_32_COPY(srcp->h_charkey, destp->h_charkey);
927 for (i = 0; i < NCACHED; i++) {
928 P_32_COPY(srcp->spares[i], destp->spares[i]);
929 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
930 }
931 }
932
933 static void
swap_header(HTAB * hashp)934 swap_header(HTAB *hashp)
935 {
936 HASHHDR *hdrp;
937 int i;
938
939 hdrp = &hashp->hdr;
940
941 M_32_SWAP(hdrp->magic);
942 M_32_SWAP(hdrp->version);
943 M_32_SWAP(hdrp->lorder);
944 M_32_SWAP(hdrp->bsize);
945 M_32_SWAP(hdrp->bshift);
946 M_32_SWAP(hdrp->dsize);
947 M_32_SWAP(hdrp->ssize);
948 M_32_SWAP(hdrp->sshift);
949 M_32_SWAP(hdrp->ovfl_point);
950 M_32_SWAP(hdrp->last_freed);
951 M_32_SWAP(hdrp->max_bucket);
952 M_32_SWAP(hdrp->high_mask);
953 M_32_SWAP(hdrp->low_mask);
954 M_32_SWAP(hdrp->ffactor);
955 M_32_SWAP(hdrp->nkeys);
956 M_32_SWAP(hdrp->hdrpages);
957 M_32_SWAP(hdrp->h_charkey);
958 for (i = 0; i < NCACHED; i++) {
959 M_32_SWAP(hdrp->spares[i]);
960 M_16_SWAP(hdrp->bitmaps[i]);
961 }
962 }
963 #endif
964