1 /*-
2 * Copyright (c) 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Margo Seltzer.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 #define _DEFAULT_SOURCE
34 #include <sys/param.h>
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)hash_page.c 8.7 (Berkeley) 8/16/94";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39
40 /*
41 * PACKAGE: hashing
42 *
43 * DESCRIPTION:
44 * Page manipulation for hashing package.
45 *
46 * ROUTINES:
47 *
48 * External
49 * __get_page
50 * __add_ovflpage
51 * Internal
52 * overflow_page
53 * open_temp
54 */
55
56 #include <sys/types.h>
57
58 #include <errno.h>
59 #include <fcntl.h>
60 #include <signal.h>
61 #include <stdio.h>
62 #include <stdlib.h>
63 #include <string.h>
64 #include <unistd.h>
65 #ifdef DEBUG
66 #include <assert.h>
67 #endif
68
69 #include "db_local.h"
70 #include "hash.h"
71 #include "page.h"
72 #include "extern.h"
73
74 static __uint32_t *fetch_bitmap(HTAB *, int);
75 static __uint32_t first_free(__uint32_t);
76 static int open_temp(HTAB *);
77 static __uint16_t overflow_page(HTAB *);
78 static void putpair(char *, const DBT *, const DBT *);
79 static void squeeze_key(__uint16_t *, const DBT *, const DBT *);
80 static int ugly_split
81 (HTAB *, __uint32_t, BUFHEAD *, BUFHEAD *, int, int);
82
83 #define PAGE_INIT(P) { \
84 ((__uint16_t *)(P))[0] = 0; \
85 ((__uint16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(__uint16_t); \
86 ((__uint16_t *)(P))[2] = hashp->BSIZE; \
87 }
88
89 /*
90 * This is called AFTER we have verified that there is room on the page for
91 * the pair (PAIRFITS has returned true) so we go right ahead and start moving
92 * stuff on.
93 */
94 static void
putpair(char * p,const DBT * key,const DBT * val)95 putpair(char *p, const DBT *key, const DBT *val)
96 {
97 __uint16_t *bp, n, off;
98
99 bp = (__uint16_t *)p;
100
101 /* Enter the key first. */
102 n = bp[0];
103
104 off = OFFSET(bp) - key->size;
105 memmove(p + off, key->data, key->size);
106 bp[++n] = off;
107
108 /* Now the data. */
109 off -= val->size;
110 memmove(p + off, val->data, val->size);
111 bp[++n] = off;
112
113 /* Adjust page info. */
114 bp[0] = n;
115 bp[n + 1] = off - ((n + 3) * sizeof(__uint16_t));
116 bp[n + 2] = off;
117 }
118
119 /*
120 * Returns:
121 * 0 OK
122 * -1 error
123 */
124 extern int
__delpair(HTAB * hashp,BUFHEAD * bufp,int ndx)125 __delpair(HTAB *hashp,
126 BUFHEAD *bufp,
127 int ndx)
128 {
129 __uint16_t *bp, newoff;
130 int n;
131 __uint16_t pairlen;
132
133 bp = (__uint16_t *)bufp->page;
134 n = bp[0];
135
136 if (bp[ndx + 1] < REAL_KEY)
137 return (__big_delete(hashp, bufp));
138 if (ndx != 1)
139 newoff = bp[ndx - 1];
140 else
141 newoff = hashp->BSIZE;
142 pairlen = newoff - bp[ndx + 1];
143
144 if (ndx != (n - 1)) {
145 /* Hard Case -- need to shuffle keys */
146 int i;
147 char *src = bufp->page + (int)OFFSET(bp);
148 char *dst = src + (int)pairlen;
149 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
150
151 /* Now adjust the pointers */
152 for (i = ndx + 2; i <= n; i += 2) {
153 if (bp[i + 1] == OVFLPAGE) {
154 bp[i - 2] = bp[i];
155 bp[i - 1] = bp[i + 1];
156 } else {
157 bp[i - 2] = bp[i] + pairlen;
158 bp[i - 1] = bp[i + 1] + pairlen;
159 }
160 }
161 }
162 /* Finally adjust the page data */
163 bp[n] = OFFSET(bp) + pairlen;
164 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(__uint16_t);
165 bp[0] = n - 2;
166 hashp->NKEYS--;
167
168 bufp->flags |= BUF_MOD;
169 return (0);
170 }
171 /*
172 * Returns:
173 * 0 ==> OK
174 * -1 ==> Error
175 */
176 extern int
__split_page(HTAB * hashp,__uint32_t obucket,__uint32_t nbucket)177 __split_page(HTAB *hashp,
178 __uint32_t obucket,
179 __uint32_t nbucket)
180 {
181 BUFHEAD *new_bufp, *old_bufp;
182 __uint16_t *ino;
183 char *np;
184 DBT key, val;
185 int ndx, retval;
186 __uint16_t n, copyto, diff, off, moved;
187 char *op;
188
189 copyto = (__uint16_t)hashp->BSIZE;
190 off = (__uint16_t)hashp->BSIZE;
191 old_bufp = __get_buf(hashp, obucket, NULL, 0);
192 if (old_bufp == NULL)
193 return (-1);
194 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
195 if (new_bufp == NULL)
196 return (-1);
197
198 old_bufp->flags |= (BUF_MOD | BUF_PIN);
199 new_bufp->flags |= (BUF_MOD | BUF_PIN);
200
201 ino = (__uint16_t *)(op = old_bufp->page);
202 np = new_bufp->page;
203
204 moved = 0;
205
206 for (n = 1, ndx = 1; n < ino[0]; n += 2) {
207 if (ino[n + 1] < REAL_KEY) {
208 retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
209 (int)copyto, (int)moved);
210 old_bufp->flags &= ~BUF_PIN;
211 new_bufp->flags &= ~BUF_PIN;
212 return (retval);
213
214 }
215 key.data = (u_char *)op + ino[n];
216 key.size = off - ino[n];
217
218 if (__call_hash(hashp, key.data, key.size) == obucket) {
219 /* Don't switch page */
220 diff = copyto - off;
221 if (diff) {
222 copyto = ino[n + 1] + diff;
223 memmove(op + copyto, op + ino[n + 1],
224 off - ino[n + 1]);
225 ino[ndx] = copyto + ino[n] - ino[n + 1];
226 ino[ndx + 1] = copyto;
227 } else
228 copyto = ino[n + 1];
229 ndx += 2;
230 } else {
231 /* Switch page */
232 val.data = (u_char *)op + ino[n + 1];
233 val.size = ino[n] - ino[n + 1];
234 putpair(np, &key, &val);
235 moved += 2;
236 }
237
238 off = ino[n + 1];
239 }
240
241 /* Now clean up the page */
242 ino[0] -= moved;
243 FREESPACE(ino) = copyto - sizeof(__uint16_t) * (ino[0] + 3);
244 OFFSET(ino) = copyto;
245
246 #ifdef DEBUG3
247 (void)fprintf(stderr, "split %d/%d\n",
248 ((__uint16_t *)np)[0] / 2,
249 ((__uint16_t *)op)[0] / 2);
250 #endif
251 /* unpin both pages */
252 old_bufp->flags &= ~BUF_PIN;
253 new_bufp->flags &= ~BUF_PIN;
254 return (0);
255 }
256
257 /*
258 * Called when we encounter an overflow or big key/data page during split
259 * handling. This is special cased since we have to begin checking whether
260 * the key/data pairs fit on their respective pages and because we may need
261 * overflow pages for both the old and new pages.
262 *
263 * The first page might be a page with regular key/data pairs in which case
264 * we have a regular overflow condition and just need to go on to the next
265 * page or it might be a big key/data pair in which case we need to fix the
266 * big key/data pair.
267 *
268 * Returns:
269 * 0 ==> success
270 * -1 ==> failure
271 */
272 static int
ugly_split(HTAB * hashp,__uint32_t obucket,BUFHEAD * old_bufp,BUFHEAD * new_bufp,int copyto,int moved)273 ugly_split(HTAB *hashp,
274 __uint32_t obucket, /* Same as __split_page. */
275 BUFHEAD *old_bufp,
276 BUFHEAD *new_bufp,
277 int copyto, /* First byte on page which contains key/data values. */
278 int moved) /* Number of pairs moved to new page. */
279 {
280 BUFHEAD *bufp; /* Buffer header for ino */
281 __uint16_t *ino; /* Page keys come off of */
282 __uint16_t *np; /* New page */
283 __uint16_t *op; /* Page keys go on to if they aren't moving */
284
285 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
286 DBT key, val;
287 SPLIT_RETURN ret;
288 __uint16_t n, off, ov_addr, scopyto;
289 char *cino; /* Character value of ino */
290
291 bufp = old_bufp;
292 ino = (__uint16_t *)old_bufp->page;
293 np = (__uint16_t *)new_bufp->page;
294 op = (__uint16_t *)old_bufp->page;
295 last_bfp = NULL;
296 scopyto = (__uint16_t)copyto; /* ANSI */
297
298 n = ino[0] - 1;
299 while (n < ino[0]) {
300 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
301 if (__big_split(hashp, old_bufp,
302 new_bufp, bufp, bufp->addr, obucket, &ret))
303 return (-1);
304 old_bufp = ret.oldp;
305 if (!old_bufp)
306 return (-1);
307 op = (__uint16_t *)old_bufp->page;
308 new_bufp = ret.newp;
309 if (!new_bufp)
310 return (-1);
311 np = (__uint16_t *)new_bufp->page;
312 bufp = ret.nextp;
313 if (!bufp)
314 return (0);
315 cino = (char *)bufp->page;
316 ino = (__uint16_t *)cino;
317 last_bfp = ret.nextp;
318 } else if (ino[n + 1] == OVFLPAGE) {
319 ov_addr = ino[n];
320 /*
321 * Fix up the old page -- the extra 2 are the fields
322 * which contained the overflow information.
323 */
324 ino[0] -= (moved + 2);
325 FREESPACE(ino) =
326 scopyto - sizeof(__uint16_t) * (ino[0] + 3);
327 OFFSET(ino) = scopyto;
328
329 bufp = __get_buf(hashp, ov_addr, bufp, 0);
330 if (!bufp)
331 return (-1);
332
333 ino = (__uint16_t *)bufp->page;
334 n = 1;
335 scopyto = hashp->BSIZE;
336 moved = 0;
337
338 if (last_bfp)
339 __free_ovflpage(hashp, last_bfp);
340 last_bfp = bufp;
341 }
342 /* Move regular sized pairs of there are any */
343 off = hashp->BSIZE;
344 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
345 cino = (char *)ino;
346 key.data = (u_char *)cino + ino[n];
347 key.size = off - ino[n];
348 val.data = (u_char *)cino + ino[n + 1];
349 val.size = ino[n] - ino[n + 1];
350 off = ino[n + 1];
351
352 if (__call_hash(hashp, key.data, key.size) == obucket) {
353 /* Keep on old page */
354 if (PAIRFITS(op, (&key), (&val)))
355 putpair((char *)op, &key, &val);
356 else {
357 old_bufp =
358 __add_ovflpage(hashp, old_bufp);
359 if (!old_bufp)
360 return (-1);
361 op = (__uint16_t *)old_bufp->page;
362 putpair((char *)op, &key, &val);
363 }
364 old_bufp->flags |= BUF_MOD;
365 } else {
366 /* Move to new page */
367 if (PAIRFITS(np, (&key), (&val)))
368 putpair((char *)np, &key, &val);
369 else {
370 new_bufp =
371 __add_ovflpage(hashp, new_bufp);
372 if (!new_bufp)
373 return (-1);
374 np = (__uint16_t *)new_bufp->page;
375 putpair((char *)np, &key, &val);
376 }
377 new_bufp->flags |= BUF_MOD;
378 }
379 }
380 }
381 if (last_bfp)
382 __free_ovflpage(hashp, last_bfp);
383 return (0);
384 }
385
386 /*
387 * Add the given pair to the page
388 *
389 * Returns:
390 * 0 ==> OK
391 * 1 ==> failure
392 */
393 extern int
__addel(HTAB * hashp,BUFHEAD * bufp,const DBT * key,const DBT * val)394 __addel(HTAB *hashp,
395 BUFHEAD *bufp,
396 const DBT *key,
397 const DBT *val)
398 {
399 __uint16_t *bp, *sop;
400 int do_expand;
401
402 bp = (__uint16_t *)bufp->page;
403 do_expand = 0;
404 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
405 /* Exception case */
406 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
407 /* This is the last page of a big key/data pair
408 and we need to add another page */
409 break;
410 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
411 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
412 if (!bufp)
413 return (-1);
414 bp = (__uint16_t *)bufp->page;
415 } else
416 /* Try to squeeze key on this page */
417 if (FREESPACE(bp) > PAIRSIZE(key, val)) {
418 squeeze_key(bp, key, val);
419 return (0);
420 } else {
421 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
422 if (!bufp)
423 return (-1);
424 bp = (__uint16_t *)bufp->page;
425 }
426
427 if (PAIRFITS(bp, key, val))
428 putpair(bufp->page, key, val);
429 else {
430 do_expand = 1;
431 bufp = __add_ovflpage(hashp, bufp);
432 if (!bufp)
433 return (-1);
434 sop = (__uint16_t *)bufp->page;
435
436 if (PAIRFITS(sop, key, val))
437 putpair((char *)sop, key, val);
438 else
439 if (__big_insert(hashp, bufp, key, val))
440 return (-1);
441 }
442 bufp->flags |= BUF_MOD;
443 /*
444 * If the average number of keys per bucket exceeds the fill factor,
445 * expand the table.
446 */
447 hashp->NKEYS++;
448 if (do_expand ||
449 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
450 return (__expand_table(hashp));
451 return (0);
452 }
453
454 /*
455 *
456 * Returns:
457 * pointer on success
458 * NULL on error
459 */
460 extern BUFHEAD *
__add_ovflpage(HTAB * hashp,BUFHEAD * bufp)461 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
462 {
463 __uint16_t *sp;
464 __uint16_t ndx, ovfl_num;
465 #ifdef DEBUG1
466 int tmp1, tmp2;
467 #endif
468 sp = (__uint16_t *)bufp->page;
469
470 /* Check if we are dynamically determining the fill factor */
471 if (hashp->FFACTOR == DEF_FFACTOR) {
472 hashp->FFACTOR = sp[0] >> 1;
473 if (hashp->FFACTOR < MIN_FFACTOR)
474 hashp->FFACTOR = MIN_FFACTOR;
475 }
476 bufp->flags |= BUF_MOD;
477 ovfl_num = overflow_page(hashp);
478 #ifdef DEBUG1
479 tmp1 = bufp->addr;
480 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
481 #endif
482 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
483 return (NULL);
484 bufp->ovfl->flags |= BUF_MOD;
485 #ifdef DEBUG1
486 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
487 tmp1, tmp2, bufp->ovfl->addr);
488 #endif
489 ndx = sp[0];
490 /*
491 * Since a pair is allocated on a page only if there's room to add
492 * an overflow page, we know that the OVFL information will fit on
493 * the page.
494 */
495 sp[ndx + 4] = OFFSET(sp);
496 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
497 sp[ndx + 1] = ovfl_num;
498 sp[ndx + 2] = OVFLPAGE;
499 sp[0] = ndx + 2;
500 #ifdef HASH_STATISTICS
501 hash_overflows++;
502 #endif
503 return (bufp->ovfl);
504 }
505
506 /*
507 * Returns:
508 * 0 indicates SUCCESS
509 * -1 indicates FAILURE
510 */
511 extern int
__get_page(HTAB * hashp,char * p,__uint32_t bucket,int is_bucket,int is_disk,int is_bitmap)512 __get_page(HTAB *hashp,
513 char *p,
514 __uint32_t bucket,
515 int is_bucket,
516 int is_disk,
517 int is_bitmap)
518 {
519 int fd, page, size;
520 int rsize;
521 __uint16_t *bp;
522
523 fd = hashp->fp;
524 size = hashp->BSIZE;
525
526 if ((fd == -1) || !is_disk) {
527 PAGE_INIT(p);
528 return (0);
529 }
530 if (is_bucket)
531 page = BUCKET_TO_PAGE(bucket);
532 else
533 page = OADDR_TO_PAGE(bucket);
534 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
535 ((rsize = read(fd, p, size)) == -1))
536 return (-1);
537 bp = (__uint16_t *)p;
538 if (!rsize)
539 bp[0] = 0; /* We hit the EOF, so initialize a new page */
540 else
541 if (rsize != size) {
542 errno = EFTYPE;
543 return (-1);
544 }
545 if (!is_bitmap && !bp[0]) {
546 PAGE_INIT(p);
547 } else
548 if (hashp->LORDER != DB_BYTE_ORDER) {
549 int i, max;
550
551 if (is_bitmap) {
552 max = hashp->BSIZE >> 2; /* divide by 4 */
553 for (i = 0; i < max; i++)
554 M_32_SWAP(((int *)p)[i]);
555 } else {
556 M_16_SWAP(bp[0]);
557 max = bp[0] + 2;
558 for (i = 1; i <= max; i++)
559 M_16_SWAP(bp[i]);
560 }
561 }
562 return (0);
563 }
564
565 /*
566 * Write page p to disk
567 *
568 * Returns:
569 * 0 ==> OK
570 * -1 ==>failure
571 */
572 extern int
__put_page(HTAB * hashp,char * p,__uint32_t bucket,int is_bucket,int is_bitmap)573 __put_page(HTAB *hashp,
574 char *p,
575 __uint32_t bucket,
576 int is_bucket,
577 int is_bitmap)
578 {
579 int fd, page, size;
580 int wsize;
581
582 size = hashp->BSIZE;
583 if ((hashp->fp == -1) && open_temp(hashp))
584 return (-1);
585 fd = hashp->fp;
586
587 if (hashp->LORDER != DB_BYTE_ORDER) {
588 int i;
589 int max;
590
591 if (is_bitmap) {
592 max = hashp->BSIZE >> 2; /* divide by 4 */
593 for (i = 0; i < max; i++)
594 M_32_SWAP(((int *)p)[i]);
595 } else {
596 max = ((__uint16_t *)p)[0] + 2;
597 for (i = 0; i <= max; i++)
598 M_16_SWAP(((__uint16_t *)p)[i]);
599 }
600 }
601 if (is_bucket)
602 page = BUCKET_TO_PAGE(bucket);
603 else
604 page = OADDR_TO_PAGE(bucket);
605 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
606 ((wsize = write(fd, p, size)) == -1))
607 /* Errno is set */
608 return (-1);
609 if (wsize != size) {
610 errno = EFTYPE;
611 return (-1);
612 }
613 return (0);
614 }
615
616 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
617 /*
618 * Initialize a new bitmap page. Bitmap pages are left in memory
619 * once they are read in.
620 */
621 extern int
__ibitmap(HTAB * hashp,int pnum,int nbits,int ndx)622 __ibitmap(HTAB *hashp,
623 int pnum,
624 int nbits,
625 int ndx)
626 {
627 __uint32_t *ip;
628 int clearbytes, clearints;
629
630 if ((ip = (__uint32_t *)malloc(hashp->BSIZE)) == NULL)
631 return (1);
632 hashp->nmaps++;
633 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
634 clearbytes = clearints << INT_TO_BYTE;
635 (void)memset((char *)ip, 0, clearbytes);
636 (void)memset(((char *)ip) + clearbytes, 0xFF,
637 hashp->BSIZE - clearbytes);
638 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
639 SETBIT(ip, 0);
640 hashp->BITMAPS[ndx] = (__uint16_t)pnum;
641 hashp->mapp[ndx] = ip;
642 return (0);
643 }
644
645 static __uint32_t
first_free(__uint32_t map)646 first_free(__uint32_t map)
647 {
648 __uint32_t i, mask;
649
650 mask = 0x1;
651 for (i = 0; i < BITS_PER_MAP; i++) {
652 if (!(mask & map))
653 return (i);
654 mask = mask << 1;
655 }
656 return (i);
657 }
658
659 static __uint16_t
overflow_page(HTAB * hashp)660 overflow_page(HTAB *hashp)
661 {
662 __uint32_t *freep = NULL;
663 int max_free, offset, splitnum;
664 __uint16_t addr;
665 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
666 #ifdef DEBUG2
667 int tmp1, tmp2;
668 #endif
669 splitnum = hashp->OVFL_POINT;
670 max_free = hashp->SPARES[splitnum];
671
672 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
673 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
674
675 /* Look through all the free maps to find the first free block */
676 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
677 for ( i = first_page; i <= free_page; i++ ) {
678 if (!(freep = (__uint32_t *)hashp->mapp[i]) &&
679 !(freep = fetch_bitmap(hashp, i)))
680 return (0);
681 if (i == free_page)
682 in_use_bits = free_bit;
683 else
684 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
685
686 if (i == first_page) {
687 bit = hashp->LAST_FREED &
688 ((hashp->BSIZE << BYTE_SHIFT) - 1);
689 j = bit / BITS_PER_MAP;
690 bit = bit & ~(BITS_PER_MAP - 1);
691 } else {
692 bit = 0;
693 j = 0;
694 }
695 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
696 if (freep[j] != ALL_SET)
697 goto found;
698 }
699
700 /* No Free Page Found */
701 hashp->LAST_FREED = hashp->SPARES[splitnum];
702 hashp->SPARES[splitnum]++;
703 offset = hashp->SPARES[splitnum] -
704 (splitnum ? hashp->SPARES[splitnum - 1] : 0);
705
706 #define OVMSG "HASH: Out of overflow pages. Increase page size\n"
707 if (offset > SPLITMASK) {
708 if (++splitnum >= NCACHED) {
709 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
710 return (0);
711 }
712 hashp->OVFL_POINT = splitnum;
713 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
714 hashp->SPARES[splitnum-1]--;
715 offset = 1;
716 }
717
718 /* Check if we need to allocate a new bitmap page */
719 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
720 free_page++;
721 if (free_page >= NCACHED) {
722 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
723 return (0);
724 }
725 /*
726 * This is tricky. The 1 indicates that you want the new page
727 * allocated with 1 clear bit. Actually, you are going to
728 * allocate 2 pages from this map. The first is going to be
729 * the map page, the second is the overflow page we were
730 * looking for. The init_bitmap routine automatically, sets
731 * the first bit of itself to indicate that the bitmap itself
732 * is in use. We would explicitly set the second bit, but
733 * don't have to if we tell init_bitmap not to leave it clear
734 * in the first place.
735 */
736 if (__ibitmap(hashp,
737 (int)OADDR_OF(splitnum, offset), 1, free_page))
738 return (0);
739 hashp->SPARES[splitnum]++;
740 #ifdef DEBUG2
741 free_bit = 2;
742 #endif
743 offset++;
744 if (offset > SPLITMASK) {
745 if (++splitnum >= NCACHED) {
746 (void)write(STDERR_FILENO, OVMSG,
747 sizeof(OVMSG) - 1);
748 return (0);
749 }
750 hashp->OVFL_POINT = splitnum;
751 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
752 hashp->SPARES[splitnum-1]--;
753 offset = 0;
754 }
755 } else {
756 /*
757 * Free_bit addresses the last used bit. Bump it to address
758 * the first available bit.
759 */
760 free_bit++;
761 SETBIT(freep, free_bit);
762 }
763
764 /* Calculate address of the new overflow page */
765 addr = OADDR_OF(splitnum, offset);
766 #ifdef DEBUG2
767 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
768 addr, free_bit, free_page);
769 #endif
770 return (addr);
771
772 found:
773 bit = bit + first_free(freep[j]);
774 SETBIT(freep, bit);
775 #ifdef DEBUG2
776 tmp1 = bit;
777 tmp2 = i;
778 #endif
779 /*
780 * Bits are addressed starting with 0, but overflow pages are addressed
781 * beginning at 1. Bit is a bit addressnumber, so we need to increment
782 * it to convert it to a page number.
783 */
784 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
785 if (bit >= hashp->LAST_FREED)
786 hashp->LAST_FREED = bit - 1;
787
788 /* Calculate the split number for this page */
789 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
790 offset = (i ? bit - hashp->SPARES[i - 1] : bit);
791 if (offset >= SPLITMASK)
792 return (0); /* Out of overflow pages */
793 addr = OADDR_OF(i, offset);
794 #ifdef DEBUG2
795 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
796 addr, tmp1, tmp2);
797 #endif
798
799 /* Allocate and return the overflow page */
800 return (addr);
801 }
802
803 /*
804 * Mark this overflow page as free.
805 */
806 extern void
__free_ovflpage(HTAB * hashp,BUFHEAD * obufp)807 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
808 {
809 __uint16_t addr;
810 __uint32_t *freep;
811 int bit_address, free_page, free_bit;
812 __uint16_t ndx;
813
814 addr = obufp->addr;
815 #ifdef DEBUG1
816 (void)fprintf(stderr, "Freeing %d\n", addr);
817 #endif
818 ndx = (((__uint16_t)addr) >> SPLITSHIFT);
819 bit_address =
820 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
821 if (bit_address < hashp->LAST_FREED)
822 hashp->LAST_FREED = bit_address;
823 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
824 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
825
826 if (!(freep = hashp->mapp[free_page]))
827 freep = fetch_bitmap(hashp, free_page);
828 #ifdef DEBUG
829 /*
830 * This had better never happen. It means we tried to read a bitmap
831 * that has already had overflow pages allocated off it, and we
832 * failed to read it from the file.
833 */
834 if (!freep)
835 assert(0);
836 #endif
837 CLRBIT(freep, free_bit);
838 #ifdef DEBUG2
839 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
840 obufp->addr, free_bit, free_page);
841 #endif
842 __reclaim_buf(hashp, obufp);
843 }
844
845 /*
846 * Returns:
847 * 0 success
848 * -1 failure
849 */
850 static int
open_temp(HTAB * hashp)851 open_temp(HTAB *hashp)
852 {
853 sigset_t set, oset;
854 static char namestr[] = "_hashXXXXXX";
855
856 /* Block signals; make sure file goes away at process exit. */
857 (void)sigfillset(&set);
858 (void)sigprocmask(SIG_BLOCK, &set, &oset);
859 if ((hashp->fp = mkstemp(namestr)) != -1) {
860 (void)unlink(namestr);
861 #ifdef _HAVE_FCNTL
862 (void)fcntl(hashp->fp, F_SETFD, 1);
863 #endif
864 }
865 (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
866 return (hashp->fp != -1 ? 0 : -1);
867 }
868
869 /*
870 * We have to know that the key will fit, but the last entry on the page is
871 * an overflow pair, so we need to shift things.
872 */
873 static void
squeeze_key(__uint16_t * sp,const DBT * key,const DBT * val)874 squeeze_key(__uint16_t *sp,
875 const DBT *key,
876 const DBT *val)
877 {
878 char *p;
879 __uint16_t free_space, n, off, pageno;
880
881 p = (char *)sp;
882 n = sp[0];
883 free_space = FREESPACE(sp);
884 off = OFFSET(sp);
885
886 pageno = sp[n - 1];
887 off -= key->size;
888 sp[n - 1] = off;
889 memmove(p + off, key->data, key->size);
890 off -= val->size;
891 sp[n] = off;
892 memmove(p + off, val->data, val->size);
893 sp[0] = n + 2;
894 sp[n + 1] = pageno;
895 sp[n + 2] = OVFLPAGE;
896 FREESPACE(sp) = free_space - PAIRSIZE(key, val);
897 OFFSET(sp) = off;
898 }
899
900 static __uint32_t *
fetch_bitmap(HTAB * hashp,int ndx)901 fetch_bitmap(HTAB *hashp, int ndx)
902 {
903 if (ndx >= hashp->nmaps)
904 return (NULL);
905 if ((hashp->mapp[ndx] = (__uint32_t *)malloc(hashp->BSIZE)) == NULL)
906 return (NULL);
907 if (__get_page(hashp,
908 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
909 free(hashp->mapp[ndx]);
910 return (NULL);
911 }
912 return (hashp->mapp[ndx]);
913 }
914
915 #ifdef DEBUG4
916 int
print_chain(addr)917 print_chain(addr)
918 int addr;
919 {
920 BUFHEAD *bufp;
921 short *bp, oaddr;
922
923 (void)fprintf(stderr, "%d ", addr);
924 bufp = __get_buf(hashp, addr, NULL, 0);
925 bp = (short *)bufp->page;
926 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
927 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
928 oaddr = bp[bp[0] - 1];
929 (void)fprintf(stderr, "%d ", (int)oaddr);
930 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
931 bp = (short *)bufp->page;
932 }
933 (void)fprintf(stderr, "\n");
934 }
935 #endif
936
