1 /*-
2 * Copyright (c) 1990, 1993
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 * %sccs.include.redist.c%
9 */
10
11 #if defined(LIBC_SCCS) && !defined(lint)
12 static char sccsid[] = "@(#)hash.c 8.7 (Berkeley) 02/21/94";
13 #endif /* LIBC_SCCS and not lint */
14
15 #include <sys/param.h>
16 #include <sys/stat.h>
17
18 #include <errno.h>
19 #include <fcntl.h>
20 #include <stdio.h>
21 #include <stdlib.h>
22 #include <string.h>
23 #include <unistd.h>
24 #ifdef DEBUG
25 #include <assert.h>
26 #endif
27
28 #include <db.h>
29 #include "hash.h"
30 #include "page.h"
31 #include "extern.h"
32
33 static int alloc_segs __P((HTAB *, int));
34 static int flush_meta __P((HTAB *));
35 static int hash_access __P((HTAB *, ACTION, DBT *, DBT *));
36 static int hash_close __P((DB *));
37 static int hash_delete __P((const DB *, const DBT *, u_int));
38 static int hash_fd __P((const DB *));
39 static int hash_get __P((const DB *, const DBT *, DBT *, u_int));
40 static int hash_put __P((const DB *, DBT *, const DBT *, u_int));
41 static void *hash_realloc __P((SEGMENT **, int, int));
42 static int hash_seq __P((const DB *, DBT *, DBT *, u_int));
43 static int hash_sync __P((const DB *, u_int));
44 static int hdestroy __P((HTAB *));
45 static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *));
46 static int init_htab __P((HTAB *, int));
47 #if BYTE_ORDER == LITTLE_ENDIAN
48 static void swap_header __P((HTAB *));
49 static void swap_header_copy __P((HASHHDR *, HASHHDR *));
50 #endif
51
52 /* Fast arithmetic, relying on powers of 2, */
53 #define MOD(x, y) ((x) & ((y) - 1))
54
55 #define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; }
56
57 /* Return values */
58 #define SUCCESS (0)
59 #define ERROR (-1)
60 #define ABNORMAL (1)
61
62 #ifdef HASH_STATISTICS
63 long hash_accesses, hash_collisions, hash_expansions, hash_overflows;
64 #endif
65
66 /************************** INTERFACE ROUTINES ***************************/
67 /* OPEN/CLOSE */
68
69 extern DB *
70 __hash_open(file, flags, mode, info, dflags)
71 const char *file;
72 int flags, mode, dflags;
73 const HASHINFO *info; /* Special directives for create */
74 {
75 HTAB *hashp;
76 struct stat statbuf;
77 DB *dbp;
78 int bpages, hdrsize, new_table, nsegs, save_errno;
79
80 if ((flags & O_ACCMODE) == O_WRONLY) {
81 errno = EINVAL;
82 return (NULL);
83 }
84
85 if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB))))
86 return (NULL);
87 hashp->fp = -1;
88
89 /*
90 * Even if user wants write only, we need to be able to read
91 * the actual file, so we need to open it read/write. But, the
92 * field in the hashp structure needs to be accurate so that
93 * we can check accesses.
94 */
95 hashp->flags = flags;
96
97 new_table = 0;
98 if (!file || (flags & O_TRUNC) ||
99 (stat(file, &statbuf) && (errno == ENOENT))) {
100 if (errno == ENOENT)
101 errno = 0; /* Just in case someone looks at errno */
102 new_table = 1;
103 }
104 if (file) {
105 if ((hashp->fp = open(file, flags, mode)) == -1)
106 RETURN_ERROR(errno, error0);
107 (void)fcntl(hashp->fp, F_SETFD, 1);
108 }
109 if (new_table) {
110 if (!(hashp = init_hash(hashp, file, (HASHINFO *)info)))
111 RETURN_ERROR(errno, error1);
112 } else {
113 /* Table already exists */
114 if (info && info->hash)
115 hashp->hash = info->hash;
116 else
117 hashp->hash = __default_hash;
118
119 hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR));
120 #if BYTE_ORDER == LITTLE_ENDIAN
121 swap_header(hashp);
122 #endif
123 if (hdrsize == -1)
124 RETURN_ERROR(errno, error1);
125 if (hdrsize != sizeof(HASHHDR))
126 RETURN_ERROR(EFTYPE, error1);
127 /* Verify file type, versions and hash function */
128 if (hashp->MAGIC != HASHMAGIC)
129 RETURN_ERROR(EFTYPE, error1);
130 #define OLDHASHVERSION 1
131 if (hashp->VERSION != HASHVERSION &&
132 hashp->VERSION != OLDHASHVERSION)
133 RETURN_ERROR(EFTYPE, error1);
134 if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY)
135 RETURN_ERROR(EFTYPE, error1);
136 /*
137 * Figure out how many segments we need. Max_Bucket is the
138 * maximum bucket number, so the number of buckets is
139 * max_bucket + 1.
140 */
141 nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) /
142 hashp->SGSIZE;
143 hashp->nsegs = 0;
144 if (alloc_segs(hashp, nsegs))
145 /*
146 * If alloc_segs fails, table will have been destroyed
147 * and errno will have been set.
148 */
149 return (NULL);
150 /* Read in bitmaps */
151 bpages = (hashp->SPARES[hashp->OVFL_POINT] +
152 (hashp->BSIZE << BYTE_SHIFT) - 1) >>
153 (hashp->BSHIFT + BYTE_SHIFT);
154
155 hashp->nmaps = bpages;
156 (void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_long *));
157 }
158
159 /* Initialize Buffer Manager */
160 if (info && info->cachesize)
161 __buf_init(hashp, info->cachesize);
162 else
163 __buf_init(hashp, DEF_BUFSIZE);
164
165 hashp->new_file = new_table;
166 hashp->save_file = file && (hashp->flags & O_RDWR);
167 hashp->cbucket = -1;
168 if (!(dbp = (DB *)malloc(sizeof(DB)))) {
169 save_errno = errno;
170 hdestroy(hashp);
171 errno = save_errno;
172 return (NULL);
173 }
174 dbp->internal = hashp;
175 dbp->close = hash_close;
176 dbp->del = hash_delete;
177 dbp->fd = hash_fd;
178 dbp->get = hash_get;
179 dbp->put = hash_put;
180 dbp->seq = hash_seq;
181 dbp->sync = hash_sync;
182 dbp->type = DB_HASH;
183
184 #ifdef DEBUG
185 (void)fprintf(stderr,
186 "%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",
187 "init_htab:",
188 "TABLE POINTER ", hashp,
189 "BUCKET SIZE ", hashp->BSIZE,
190 "BUCKET SHIFT ", hashp->BSHIFT,
191 "DIRECTORY SIZE ", hashp->DSIZE,
192 "SEGMENT SIZE ", hashp->SGSIZE,
193 "SEGMENT SHIFT ", hashp->SSHIFT,
194 "FILL FACTOR ", hashp->FFACTOR,
195 "MAX BUCKET ", hashp->MAX_BUCKET,
196 "OVFL POINT ", hashp->OVFL_POINT,
197 "LAST FREED ", hashp->LAST_FREED,
198 "HIGH MASK ", hashp->HIGH_MASK,
199 "LOW MASK ", hashp->LOW_MASK,
200 "NSEGS ", hashp->nsegs,
201 "NKEYS ", hashp->NKEYS);
202 #endif
203 #ifdef HASH_STATISTICS
204 hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0;
205 #endif
206 return (dbp);
207
208 error1:
209 if (hashp != NULL)
210 (void)close(hashp->fp);
211
212 error0:
213 free(hashp);
214 errno = save_errno;
215 return (NULL);
216 }
217
218 static int
hash_close(dbp)219 hash_close(dbp)
220 DB *dbp;
221 {
222 HTAB *hashp;
223 int retval;
224
225 if (!dbp)
226 return (ERROR);
227
228 hashp = (HTAB *)dbp->internal;
229 retval = hdestroy(hashp);
230 free(dbp);
231 return (retval);
232 }
233
234 static int
hash_fd(dbp)235 hash_fd(dbp)
236 const DB *dbp;
237 {
238 HTAB *hashp;
239
240 if (!dbp)
241 return (ERROR);
242
243 hashp = (HTAB *)dbp->internal;
244 if (hashp->fp == -1) {
245 errno = ENOENT;
246 return (-1);
247 }
248 return (hashp->fp);
249 }
250
251 /************************** LOCAL CREATION ROUTINES **********************/
252 static HTAB *
init_hash(hashp,file,info)253 init_hash(hashp, file, info)
254 HTAB *hashp;
255 const char *file;
256 HASHINFO *info;
257 {
258 struct stat statbuf;
259 int nelem;
260
261 nelem = 1;
262 hashp->NKEYS = 0;
263 hashp->LORDER = BYTE_ORDER;
264 hashp->BSIZE = DEF_BUCKET_SIZE;
265 hashp->BSHIFT = DEF_BUCKET_SHIFT;
266 hashp->SGSIZE = DEF_SEGSIZE;
267 hashp->SSHIFT = DEF_SEGSIZE_SHIFT;
268 hashp->DSIZE = DEF_DIRSIZE;
269 hashp->FFACTOR = DEF_FFACTOR;
270 hashp->hash = __default_hash;
271 memset(hashp->SPARES, 0, sizeof(hashp->SPARES));
272 memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS));
273
274 /* Fix bucket size to be optimal for file system */
275 if (file != NULL) {
276 if (stat(file, &statbuf))
277 return (NULL);
278 hashp->BSIZE = statbuf.st_blksize;
279 hashp->BSHIFT = __log2(hashp->BSIZE);
280 }
281
282 if (info) {
283 if (info->bsize) {
284 /* Round pagesize up to power of 2 */
285 hashp->BSHIFT = __log2(info->bsize);
286 hashp->BSIZE = 1 << hashp->BSHIFT;
287 if (hashp->BSIZE > MAX_BSIZE) {
288 errno = EINVAL;
289 return (NULL);
290 }
291 }
292 if (info->ffactor)
293 hashp->FFACTOR = info->ffactor;
294 if (info->hash)
295 hashp->hash = info->hash;
296 if (info->nelem)
297 nelem = info->nelem;
298 if (info->lorder) {
299 if (info->lorder != BIG_ENDIAN &&
300 info->lorder != LITTLE_ENDIAN) {
301 errno = EINVAL;
302 return (NULL);
303 }
304 hashp->LORDER = info->lorder;
305 }
306 }
307 /* init_htab should destroy the table and set errno if it fails */
308 if (init_htab(hashp, nelem))
309 return (NULL);
310 else
311 return (hashp);
312 }
313 /*
314 * This calls alloc_segs which may run out of memory. Alloc_segs will destroy
315 * the table and set errno, so we just pass the error information along.
316 *
317 * Returns 0 on No Error
318 */
319 static int
init_htab(hashp,nelem)320 init_htab(hashp, nelem)
321 HTAB *hashp;
322 int nelem;
323 {
324 register int nbuckets, nsegs;
325 int l2;
326
327 /*
328 * Divide number of elements by the fill factor and determine a
329 * desired number of buckets. Allocate space for the next greater
330 * power of two number of buckets.
331 */
332 nelem = (nelem - 1) / hashp->FFACTOR + 1;
333
334 l2 = __log2(MAX(nelem, 2));
335 nbuckets = 1 << l2;
336
337 hashp->SPARES[l2] = l2 + 1;
338 hashp->SPARES[l2 + 1] = l2 + 1;
339 hashp->OVFL_POINT = l2;
340 hashp->LAST_FREED = 2;
341
342 /* First bitmap page is at: splitpoint l2 page offset 1 */
343 if (__init_bitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0))
344 return (-1);
345
346 hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1;
347 hashp->HIGH_MASK = (nbuckets << 1) - 1;
348 hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >>
349 hashp->BSHIFT) + 1;
350
351 nsegs = (nbuckets - 1) / hashp->SGSIZE + 1;
352 nsegs = 1 << __log2(nsegs);
353
354 if (nsegs > hashp->DSIZE)
355 hashp->DSIZE = nsegs;
356 return (alloc_segs(hashp, nsegs));
357 }
358
359 /********************** DESTROY/CLOSE ROUTINES ************************/
360
361 /*
362 * Flushes any changes to the file if necessary and destroys the hashp
363 * structure, freeing all allocated space.
364 */
365 static int
hdestroy(hashp)366 hdestroy(hashp)
367 HTAB *hashp;
368 {
369 int i, save_errno;
370
371 save_errno = 0;
372
373 #ifdef HASH_STATISTICS
374 (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n",
375 hash_accesses, hash_collisions);
376 (void)fprintf(stderr, "hdestroy: expansions %ld\n",
377 hash_expansions);
378 (void)fprintf(stderr, "hdestroy: overflows %ld\n",
379 hash_overflows);
380 (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n",
381 hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs);
382
383 for (i = 0; i < NCACHED; i++)
384 (void)fprintf(stderr,
385 "spares[%d] = %d\n", i, hashp->SPARES[i]);
386 #endif
387 /*
388 * Call on buffer manager to free buffers, and if required,
389 * write them to disk.
390 */
391 if (__buf_free(hashp, 1, hashp->save_file))
392 save_errno = errno;
393 if (hashp->dir) {
394 free(*hashp->dir); /* Free initial segments */
395 /* Free extra segments */
396 while (hashp->exsegs--)
397 free(hashp->dir[--hashp->nsegs]);
398 free(hashp->dir);
399 }
400 if (flush_meta(hashp) && !save_errno)
401 save_errno = errno;
402 /* Free Bigmaps */
403 for (i = 0; i < hashp->nmaps; i++)
404 if (hashp->mapp[i])
405 free(hashp->mapp[i]);
406
407 if (hashp->fp != -1)
408 (void)close(hashp->fp);
409
410 free(hashp);
411
412 if (save_errno) {
413 errno = save_errno;
414 return (ERROR);
415 }
416 return (SUCCESS);
417 }
418 /*
419 * Write modified pages to disk
420 *
421 * Returns:
422 * 0 == OK
423 * -1 ERROR
424 */
425 static int
hash_sync(dbp,flags)426 hash_sync(dbp, flags)
427 const DB *dbp;
428 u_int flags;
429 {
430 HTAB *hashp;
431
432 if (flags != 0) {
433 errno = EINVAL;
434 return (ERROR);
435 }
436
437 if (!dbp)
438 return (ERROR);
439
440 hashp = (HTAB *)dbp->internal;
441 if (!hashp->save_file)
442 return (0);
443 if (__buf_free(hashp, 0, 1) || flush_meta(hashp))
444 return (ERROR);
445 hashp->new_file = 0;
446 return (0);
447 }
448
449 /*
450 * Returns:
451 * 0 == OK
452 * -1 indicates that errno should be set
453 */
454 static int
flush_meta(hashp)455 flush_meta(hashp)
456 HTAB *hashp;
457 {
458 HASHHDR *whdrp;
459 #if BYTE_ORDER == LITTLE_ENDIAN
460 HASHHDR whdr;
461 #endif
462 int fp, i, wsize;
463
464 if (!hashp->save_file)
465 return (0);
466 hashp->MAGIC = HASHMAGIC;
467 hashp->VERSION = HASHVERSION;
468 hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY));
469
470 fp = hashp->fp;
471 whdrp = &hashp->hdr;
472 #if BYTE_ORDER == LITTLE_ENDIAN
473 whdrp = &whdr;
474 swap_header_copy(&hashp->hdr, whdrp);
475 #endif
476 if ((lseek(fp, (off_t)0, SEEK_SET) == -1) ||
477 ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1))
478 return (-1);
479 else
480 if (wsize != sizeof(HASHHDR)) {
481 errno = EFTYPE;
482 hashp->errno = errno;
483 return (-1);
484 }
485 for (i = 0; i < NCACHED; i++)
486 if (hashp->mapp[i])
487 if (__put_page(hashp, (char *)hashp->mapp[i],
488 hashp->BITMAPS[i], 0, 1))
489 return (-1);
490 return (0);
491 }
492
493 /*******************************SEARCH ROUTINES *****************************/
494 /*
495 * All the access routines return
496 *
497 * Returns:
498 * 0 on SUCCESS
499 * 1 to indicate an external ERROR (i.e. key not found, etc)
500 * -1 to indicate an internal ERROR (i.e. out of memory, etc)
501 */
502 static int
hash_get(dbp,key,data,flag)503 hash_get(dbp, key, data, flag)
504 const DB *dbp;
505 const DBT *key;
506 DBT *data;
507 u_int flag;
508 {
509 HTAB *hashp;
510
511 hashp = (HTAB *)dbp->internal;
512 if (flag) {
513 hashp->errno = errno = EINVAL;
514 return (ERROR);
515 }
516 return (hash_access(hashp, HASH_GET, (DBT *)key, data));
517 }
518
519 static int
hash_put(dbp,key,data,flag)520 hash_put(dbp, key, data, flag)
521 const DB *dbp;
522 DBT *key;
523 const DBT *data;
524 u_int flag;
525 {
526 HTAB *hashp;
527
528 hashp = (HTAB *)dbp->internal;
529 if (flag && flag != R_NOOVERWRITE) {
530 hashp->errno = errno = EINVAL;
531 return (ERROR);
532 }
533 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
534 hashp->errno = errno = EPERM;
535 return (ERROR);
536 }
537 return (hash_access(hashp, flag == R_NOOVERWRITE ?
538 HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data));
539 }
540
541 static int
hash_delete(dbp,key,flag)542 hash_delete(dbp, key, flag)
543 const DB *dbp;
544 const DBT *key;
545 u_int flag; /* Ignored */
546 {
547 HTAB *hashp;
548
549 hashp = (HTAB *)dbp->internal;
550 if (flag && flag != R_CURSOR) {
551 hashp->errno = errno = EINVAL;
552 return (ERROR);
553 }
554 if ((hashp->flags & O_ACCMODE) == O_RDONLY) {
555 hashp->errno = errno = EPERM;
556 return (ERROR);
557 }
558 return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL));
559 }
560
561 /*
562 * Assume that hashp has been set in wrapper routine.
563 */
564 static int
hash_access(hashp,action,key,val)565 hash_access(hashp, action, key, val)
566 HTAB *hashp;
567 ACTION action;
568 DBT *key, *val;
569 {
570 register BUFHEAD *rbufp;
571 BUFHEAD *bufp, *save_bufp;
572 register u_short *bp;
573 register int n, ndx, off, size;
574 register char *kp;
575 u_short pageno;
576
577 #ifdef HASH_STATISTICS
578 hash_accesses++;
579 #endif
580
581 off = hashp->BSIZE;
582 size = key->size;
583 kp = (char *)key->data;
584 rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0);
585 if (!rbufp)
586 return (ERROR);
587 save_bufp = rbufp;
588
589 /* Pin the bucket chain */
590 rbufp->flags |= BUF_PIN;
591 for (bp = (u_short *)rbufp->page, n = *bp++, ndx = 1; ndx < n;)
592 if (bp[1] >= REAL_KEY) {
593 /* Real key/data pair */
594 if (size == off - *bp &&
595 memcmp(kp, rbufp->page + *bp, size) == 0)
596 goto found;
597 off = bp[1];
598 #ifdef HASH_STATISTICS
599 hash_collisions++;
600 #endif
601 bp += 2;
602 ndx += 2;
603 } else if (bp[1] == OVFLPAGE) {
604 rbufp = __get_buf(hashp, *bp, rbufp, 0);
605 if (!rbufp) {
606 save_bufp->flags &= ~BUF_PIN;
607 return (ERROR);
608 }
609 /* FOR LOOP INIT */
610 bp = (u_short *)rbufp->page;
611 n = *bp++;
612 ndx = 1;
613 off = hashp->BSIZE;
614 } else if (bp[1] < REAL_KEY) {
615 if ((ndx =
616 __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0)
617 goto found;
618 if (ndx == -2) {
619 bufp = rbufp;
620 if (!(pageno =
621 __find_last_page(hashp, &bufp))) {
622 ndx = 0;
623 rbufp = bufp;
624 break; /* FOR */
625 }
626 rbufp = __get_buf(hashp, pageno, bufp, 0);
627 if (!rbufp) {
628 save_bufp->flags &= ~BUF_PIN;
629 return (ERROR);
630 }
631 /* FOR LOOP INIT */
632 bp = (u_short *)rbufp->page;
633 n = *bp++;
634 ndx = 1;
635 off = hashp->BSIZE;
636 } else {
637 save_bufp->flags &= ~BUF_PIN;
638 return (ERROR);
639 }
640 }
641
642 /* Not found */
643 switch (action) {
644 case HASH_PUT:
645 case HASH_PUTNEW:
646 if (__addel(hashp, rbufp, key, val)) {
647 save_bufp->flags &= ~BUF_PIN;
648 return (ERROR);
649 } else {
650 save_bufp->flags &= ~BUF_PIN;
651 return (SUCCESS);
652 }
653 case HASH_GET:
654 case HASH_DELETE:
655 default:
656 save_bufp->flags &= ~BUF_PIN;
657 return (ABNORMAL);
658 }
659
660 found:
661 switch (action) {
662 case HASH_PUTNEW:
663 save_bufp->flags &= ~BUF_PIN;
664 return (ABNORMAL);
665 case HASH_GET:
666 bp = (u_short *)rbufp->page;
667 if (bp[ndx + 1] < REAL_KEY) {
668 if (__big_return(hashp, rbufp, ndx, val, 0))
669 return (ERROR);
670 } else {
671 val->data = (u_char *)rbufp->page + (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(dbp,key,data,flag)694 hash_seq(dbp, key, data, flag)
695 const DB *dbp;
696 DBT *key, *data;
697 u_int flag;
698 {
699 register u_int bucket;
700 register BUFHEAD *bufp;
701 HTAB *hashp;
702 u_short *bp, ndx;
703
704 hashp = (HTAB *)dbp->internal;
705 if (flag && flag != R_FIRST && flag != R_NEXT) {
706 hashp->errno = errno = EINVAL;
707 return (ERROR);
708 }
709 #ifdef HASH_STATISTICS
710 hash_accesses++;
711 #endif
712 if ((hashp->cbucket < 0) || (flag == R_FIRST)) {
713 hashp->cbucket = 0;
714 hashp->cndx = 1;
715 hashp->cpage = NULL;
716 }
717
718 for (bp = NULL; !bp || !bp[0]; ) {
719 if (!(bufp = hashp->cpage)) {
720 for (bucket = hashp->cbucket;
721 bucket <= hashp->MAX_BUCKET;
722 bucket++, hashp->cndx = 1) {
723 bufp = __get_buf(hashp, bucket, NULL, 0);
724 if (!bufp)
725 return (ERROR);
726 hashp->cpage = bufp;
727 bp = (u_short *)bufp->page;
728 if (bp[0])
729 break;
730 }
731 hashp->cbucket = bucket;
732 if (hashp->cbucket > hashp->MAX_BUCKET) {
733 hashp->cbucket = -1;
734 return (ABNORMAL);
735 }
736 } else
737 bp = (u_short *)hashp->cpage->page;
738
739 #ifdef DEBUG
740 assert(bp);
741 assert(bufp);
742 #endif
743 while (bp[hashp->cndx + 1] == OVFLPAGE) {
744 bufp = hashp->cpage =
745 __get_buf(hashp, bp[hashp->cndx], bufp, 0);
746 if (!bufp)
747 return (ERROR);
748 bp = (u_short *)(bufp->page);
749 hashp->cndx = 1;
750 }
751 if (!bp[0]) {
752 hashp->cpage = NULL;
753 ++hashp->cbucket;
754 }
755 }
756 ndx = hashp->cndx;
757 if (bp[ndx + 1] < REAL_KEY) {
758 if (__big_keydata(hashp, bufp, key, data, 1))
759 return (ERROR);
760 } else {
761 key->data = (u_char *)hashp->cpage->page + bp[ndx];
762 key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx];
763 data->data = (u_char *)hashp->cpage->page + bp[ndx + 1];
764 data->size = bp[ndx] - bp[ndx + 1];
765 ndx += 2;
766 if (ndx > bp[0]) {
767 hashp->cpage = NULL;
768 hashp->cbucket++;
769 hashp->cndx = 1;
770 } else
771 hashp->cndx = ndx;
772 }
773 return (SUCCESS);
774 }
775
776 /********************************* UTILITIES ************************/
777
778 /*
779 * Returns:
780 * 0 ==> OK
781 * -1 ==> Error
782 */
783 extern int
784 __expand_table(hashp)
785 HTAB *hashp;
786 {
787 u_int old_bucket, new_bucket;
788 int dirsize, new_segnum, spare_ndx;
789
790 #ifdef HASH_STATISTICS
791 hash_expansions++;
792 #endif
793 new_bucket = ++hashp->MAX_BUCKET;
794 old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK);
795
796 new_segnum = new_bucket >> hashp->SSHIFT;
797
798 /* Check if we need a new segment */
799 if (new_segnum >= hashp->nsegs) {
800 /* Check if we need to expand directory */
801 if (new_segnum >= hashp->DSIZE) {
802 /* Reallocate directory */
803 dirsize = hashp->DSIZE * sizeof(SEGMENT *);
804 if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1))
805 return (-1);
806 hashp->DSIZE = dirsize << 1;
807 }
808 if ((hashp->dir[new_segnum] =
809 (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL)
810 return (-1);
811 hashp->exsegs++;
812 hashp->nsegs++;
813 }
814 /*
815 * If the split point is increasing (MAX_BUCKET's log base 2
816 * * increases), we need to copy the current contents of the spare
817 * split bucket to the next bucket.
818 */
819 spare_ndx = __log2(hashp->MAX_BUCKET + 1);
820 if (spare_ndx > hashp->OVFL_POINT) {
821 hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT];
822 hashp->OVFL_POINT = spare_ndx;
823 }
824
825 if (new_bucket > hashp->HIGH_MASK) {
826 /* Starting a new doubling */
827 hashp->LOW_MASK = hashp->HIGH_MASK;
828 hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK;
829 }
830 /* Relocate records to the new bucket */
831 return (__split_page(hashp, old_bucket, new_bucket));
832 }
833
834 /*
835 * If realloc guarantees that the pointer is not destroyed if the realloc
836 * fails, then this routine can go away.
837 */
838 static void *
hash_realloc(p_ptr,oldsize,newsize)839 hash_realloc(p_ptr, oldsize, newsize)
840 SEGMENT **p_ptr;
841 int oldsize, newsize;
842 {
843 register void *p;
844
845 if (p = malloc(newsize)) {
846 memmove(p, *p_ptr, oldsize);
847 memset((char *)p + oldsize, 0, newsize - oldsize);
848 free(*p_ptr);
849 *p_ptr = p;
850 }
851 return (p);
852 }
853
854 extern u_int
855 __call_hash(hashp, k, len)
856 HTAB *hashp;
857 char *k;
858 int len;
859 {
860 int n, bucket;
861
862 n = hashp->hash(k, len);
863 bucket = n & hashp->HIGH_MASK;
864 if (bucket > hashp->MAX_BUCKET)
865 bucket = bucket & hashp->LOW_MASK;
866 return (bucket);
867 }
868
869 /*
870 * Allocate segment table. On error, destroy the table and set errno.
871 *
872 * Returns 0 on success
873 */
874 static int
alloc_segs(hashp,nsegs)875 alloc_segs(hashp, nsegs)
876 HTAB *hashp;
877 int nsegs;
878 {
879 register int i;
880 register SEGMENT store;
881
882 int save_errno;
883
884 if ((hashp->dir =
885 (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) {
886 save_errno = errno;
887 (void)hdestroy(hashp);
888 errno = save_errno;
889 return (-1);
890 }
891 /* Allocate segments */
892 if ((store =
893 (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) {
894 save_errno = errno;
895 (void)hdestroy(hashp);
896 errno = save_errno;
897 return (-1);
898 }
899 for (i = 0; i < nsegs; i++, hashp->nsegs++)
900 hashp->dir[i] = &store[i << hashp->SSHIFT];
901 return (0);
902 }
903
904 #if BYTE_ORDER == LITTLE_ENDIAN
905 /*
906 * Hashp->hdr needs to be byteswapped.
907 */
908 static void
swap_header_copy(srcp,destp)909 swap_header_copy(srcp, destp)
910 HASHHDR *srcp, *destp;
911 {
912 int i;
913
914 P_32_COPY(srcp->magic, destp->magic);
915 P_32_COPY(srcp->version, destp->version);
916 P_32_COPY(srcp->lorder, destp->lorder);
917 P_32_COPY(srcp->bsize, destp->bsize);
918 P_32_COPY(srcp->bshift, destp->bshift);
919 P_32_COPY(srcp->dsize, destp->dsize);
920 P_32_COPY(srcp->ssize, destp->ssize);
921 P_32_COPY(srcp->sshift, destp->sshift);
922 P_32_COPY(srcp->ovfl_point, destp->ovfl_point);
923 P_32_COPY(srcp->last_freed, destp->last_freed);
924 P_32_COPY(srcp->max_bucket, destp->max_bucket);
925 P_32_COPY(srcp->high_mask, destp->high_mask);
926 P_32_COPY(srcp->low_mask, destp->low_mask);
927 P_32_COPY(srcp->ffactor, destp->ffactor);
928 P_32_COPY(srcp->nkeys, destp->nkeys);
929 P_32_COPY(srcp->hdrpages, destp->hdrpages);
930 P_32_COPY(srcp->h_charkey, destp->h_charkey);
931 for (i = 0; i < NCACHED; i++) {
932 P_32_COPY(srcp->spares[i], destp->spares[i]);
933 P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]);
934 }
935 }
936
937 static void
swap_header(hashp)938 swap_header(hashp)
939 HTAB *hashp;
940 {
941 HASHHDR *hdrp;
942 int i;
943
944 hdrp = &hashp->hdr;
945
946 M_32_SWAP(hdrp->magic);
947 M_32_SWAP(hdrp->version);
948 M_32_SWAP(hdrp->lorder);
949 M_32_SWAP(hdrp->bsize);
950 M_32_SWAP(hdrp->bshift);
951 M_32_SWAP(hdrp->dsize);
952 M_32_SWAP(hdrp->ssize);
953 M_32_SWAP(hdrp->sshift);
954 M_32_SWAP(hdrp->ovfl_point);
955 M_32_SWAP(hdrp->last_freed);
956 M_32_SWAP(hdrp->max_bucket);
957 M_32_SWAP(hdrp->high_mask);
958 M_32_SWAP(hdrp->low_mask);
959 M_32_SWAP(hdrp->ffactor);
960 M_32_SWAP(hdrp->nkeys);
961 M_32_SWAP(hdrp->hdrpages);
962 M_32_SWAP(hdrp->h_charkey);
963 for (i = 0; i < NCACHED; i++) {
964 M_32_SWAP(hdrp->spares[i]);
965 M_16_SWAP(hdrp->bitmaps[i]);
966 }
967 }
968 #endif
969