xref: /freebsd/lib/libc/db/hash/hash_page.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1990, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Margo Seltzer.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
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 __FBSDID("$FreeBSD$");
40 
41 /*
42  * PACKAGE:  hashing
43  *
44  * DESCRIPTION:
45  *	Page manipulation for hashing package.
46  *
47  * ROUTINES:
48  *
49  * External
50  *	__get_page
51  *	__add_ovflpage
52  * Internal
53  *	overflow_page
54  *	open_temp
55  */
56 
57 #include "namespace.h"
58 #include <sys/param.h>
59 
60 #include <errno.h>
61 #include <fcntl.h>
62 #include <signal.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <string.h>
66 #include <unistd.h>
67 #ifdef DEBUG
68 #include <assert.h>
69 #endif
70 #include "un-namespace.h"
71 #include "libc_private.h"
72 
73 #include <db.h>
74 #include "hash.h"
75 #include "page.h"
76 #include "extern.h"
77 
78 static u_int32_t *fetch_bitmap(HTAB *, int);
79 static u_int32_t  first_free(u_int32_t);
80 static int	  open_temp(HTAB *);
81 static u_int16_t  overflow_page(HTAB *);
82 static void	  putpair(char *, const DBT *, const DBT *);
83 static void	  squeeze_key(u_int16_t *, const DBT *, const DBT *);
84 static int	  ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
85 
86 #define	PAGE_INIT(P) { \
87 	((u_int16_t *)(P))[0] = 0; \
88 	((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
89 	((u_int16_t *)(P))[2] = hashp->BSIZE; \
90 }
91 
92 /*
93  * This is called AFTER we have verified that there is room on the page for
94  * the pair (PAIRFITS has returned true) so we go right ahead and start moving
95  * stuff on.
96  */
97 static void
98 putpair(char *p, const DBT *key, const DBT *val)
99 {
100 	u_int16_t *bp, n, off;
101 
102 	bp = (u_int16_t *)p;
103 
104 	/* Enter the key first. */
105 	n = bp[0];
106 
107 	off = OFFSET(bp) - key->size;
108 	memmove(p + off, key->data, key->size);
109 	bp[++n] = off;
110 
111 	/* Now the data. */
112 	off -= val->size;
113 	memmove(p + off, val->data, val->size);
114 	bp[++n] = off;
115 
116 	/* Adjust page info. */
117 	bp[0] = n;
118 	bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
119 	bp[n + 2] = off;
120 }
121 
122 /*
123  * Returns:
124  *	 0 OK
125  *	-1 error
126  */
127 int
128 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
129 {
130 	u_int16_t *bp, newoff, pairlen;
131 	int n;
132 
133 	bp = (u_int16_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 		if (ndx == hashp->cndx) {
162 			/*
163 			 * We just removed pair we were "pointing" to.
164 			 * By moving back the cndx we ensure subsequent
165 			 * hash_seq() calls won't skip over any entries.
166 			 */
167 			hashp->cndx -= 2;
168 		}
169 	}
170 	/* Finally adjust the page data */
171 	bp[n] = OFFSET(bp) + pairlen;
172 	bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
173 	bp[0] = n - 2;
174 	hashp->NKEYS--;
175 
176 	bufp->flags |= BUF_MOD;
177 	return (0);
178 }
179 /*
180  * Returns:
181  *	 0 ==> OK
182  *	-1 ==> Error
183  */
184 int
185 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
186 {
187 	BUFHEAD *new_bufp, *old_bufp;
188 	u_int16_t *ino;
189 	char *np;
190 	DBT key, val;
191 	int n, ndx, retval;
192 	u_int16_t copyto, diff, off, moved;
193 	char *op;
194 
195 	copyto = (u_int16_t)hashp->BSIZE;
196 	off = (u_int16_t)hashp->BSIZE;
197 	old_bufp = __get_buf(hashp, obucket, NULL, 0);
198 	if (old_bufp == NULL)
199 		return (-1);
200 	new_bufp = __get_buf(hashp, nbucket, NULL, 0);
201 	if (new_bufp == NULL)
202 		return (-1);
203 
204 	old_bufp->flags |= (BUF_MOD | BUF_PIN);
205 	new_bufp->flags |= (BUF_MOD | BUF_PIN);
206 
207 	ino = (u_int16_t *)(op = old_bufp->page);
208 	np = new_bufp->page;
209 
210 	moved = 0;
211 
212 	for (n = 1, ndx = 1; n < ino[0]; n += 2) {
213 		if (ino[n + 1] < REAL_KEY) {
214 			retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
215 			    (int)copyto, (int)moved);
216 			old_bufp->flags &= ~BUF_PIN;
217 			new_bufp->flags &= ~BUF_PIN;
218 			return (retval);
219 
220 		}
221 		key.data = (u_char *)op + ino[n];
222 		key.size = off - ino[n];
223 
224 		if (__call_hash(hashp, key.data, key.size) == obucket) {
225 			/* Don't switch page */
226 			diff = copyto - off;
227 			if (diff) {
228 				copyto = ino[n + 1] + diff;
229 				memmove(op + copyto, op + ino[n + 1],
230 				    off - ino[n + 1]);
231 				ino[ndx] = copyto + ino[n] - ino[n + 1];
232 				ino[ndx + 1] = copyto;
233 			} else
234 				copyto = ino[n + 1];
235 			ndx += 2;
236 		} else {
237 			/* Switch page */
238 			val.data = (u_char *)op + ino[n + 1];
239 			val.size = ino[n] - ino[n + 1];
240 			putpair(np, &key, &val);
241 			moved += 2;
242 		}
243 
244 		off = ino[n + 1];
245 	}
246 
247 	/* Now clean up the page */
248 	ino[0] -= moved;
249 	FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
250 	OFFSET(ino) = copyto;
251 
252 #ifdef DEBUG3
253 	(void)fprintf(stderr, "split %d/%d\n",
254 	    ((u_int16_t *)np)[0] / 2,
255 	    ((u_int16_t *)op)[0] / 2);
256 #endif
257 	/* unpin both pages */
258 	old_bufp->flags &= ~BUF_PIN;
259 	new_bufp->flags &= ~BUF_PIN;
260 	return (0);
261 }
262 
263 /*
264  * Called when we encounter an overflow or big key/data page during split
265  * handling.  This is special cased since we have to begin checking whether
266  * the key/data pairs fit on their respective pages and because we may need
267  * overflow pages for both the old and new pages.
268  *
269  * The first page might be a page with regular key/data pairs in which case
270  * we have a regular overflow condition and just need to go on to the next
271  * page or it might be a big key/data pair in which case we need to fix the
272  * big key/data pair.
273  *
274  * Returns:
275  *	 0 ==> success
276  *	-1 ==> failure
277  */
278 static int
279 ugly_split(HTAB *hashp,
280     u_int32_t obucket,	/* Same as __split_page. */
281     BUFHEAD *old_bufp,
282     BUFHEAD *new_bufp,
283     int copyto,		/* First byte on page which contains key/data values. */
284     int moved)		/* Number of pairs moved to new page. */
285 {
286 	BUFHEAD *bufp;	/* Buffer header for ino */
287 	u_int16_t *ino;	/* Page keys come off of */
288 	u_int16_t *np;	/* New page */
289 	u_int16_t *op;	/* Page keys go on to if they aren't moving */
290 
291 	BUFHEAD *last_bfp;	/* Last buf header OVFL needing to be freed */
292 	DBT key, val;
293 	SPLIT_RETURN ret;
294 	u_int16_t n, off, ov_addr, scopyto;
295 	char *cino;		/* Character value of ino */
296 
297 	bufp = old_bufp;
298 	ino = (u_int16_t *)old_bufp->page;
299 	np = (u_int16_t *)new_bufp->page;
300 	op = (u_int16_t *)old_bufp->page;
301 	last_bfp = NULL;
302 	scopyto = (u_int16_t)copyto;	/* ANSI */
303 
304 	n = ino[0] - 1;
305 	while (n < ino[0]) {
306 		if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
307 			if (__big_split(hashp, old_bufp,
308 			    new_bufp, bufp, bufp->addr, obucket, &ret))
309 				return (-1);
310 			old_bufp = ret.oldp;
311 			if (!old_bufp)
312 				return (-1);
313 			op = (u_int16_t *)old_bufp->page;
314 			new_bufp = ret.newp;
315 			if (!new_bufp)
316 				return (-1);
317 			np = (u_int16_t *)new_bufp->page;
318 			bufp = ret.nextp;
319 			if (!bufp)
320 				return (0);
321 			cino = (char *)bufp->page;
322 			ino = (u_int16_t *)cino;
323 			last_bfp = ret.nextp;
324 		} else if (ino[n + 1] == OVFLPAGE) {
325 			ov_addr = ino[n];
326 			/*
327 			 * Fix up the old page -- the extra 2 are the fields
328 			 * which contained the overflow information.
329 			 */
330 			ino[0] -= (moved + 2);
331 			FREESPACE(ino) =
332 			    scopyto - sizeof(u_int16_t) * (ino[0] + 3);
333 			OFFSET(ino) = scopyto;
334 
335 			bufp = __get_buf(hashp, ov_addr, bufp, 0);
336 			if (!bufp)
337 				return (-1);
338 
339 			ino = (u_int16_t *)bufp->page;
340 			n = 1;
341 			scopyto = hashp->BSIZE;
342 			moved = 0;
343 
344 			if (last_bfp)
345 				__free_ovflpage(hashp, last_bfp);
346 			last_bfp = bufp;
347 		}
348 		/* Move regular sized pairs of there are any */
349 		off = hashp->BSIZE;
350 		for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
351 			cino = (char *)ino;
352 			key.data = (u_char *)cino + ino[n];
353 			key.size = off - ino[n];
354 			val.data = (u_char *)cino + ino[n + 1];
355 			val.size = ino[n] - ino[n + 1];
356 			off = ino[n + 1];
357 
358 			if (__call_hash(hashp, key.data, key.size) == obucket) {
359 				/* Keep on old page */
360 				if (PAIRFITS(op, (&key), (&val)))
361 					putpair((char *)op, &key, &val);
362 				else {
363 					old_bufp =
364 					    __add_ovflpage(hashp, old_bufp);
365 					if (!old_bufp)
366 						return (-1);
367 					op = (u_int16_t *)old_bufp->page;
368 					putpair((char *)op, &key, &val);
369 				}
370 				old_bufp->flags |= BUF_MOD;
371 			} else {
372 				/* Move to new page */
373 				if (PAIRFITS(np, (&key), (&val)))
374 					putpair((char *)np, &key, &val);
375 				else {
376 					new_bufp =
377 					    __add_ovflpage(hashp, new_bufp);
378 					if (!new_bufp)
379 						return (-1);
380 					np = (u_int16_t *)new_bufp->page;
381 					putpair((char *)np, &key, &val);
382 				}
383 				new_bufp->flags |= BUF_MOD;
384 			}
385 		}
386 	}
387 	if (last_bfp)
388 		__free_ovflpage(hashp, last_bfp);
389 	return (0);
390 }
391 
392 /*
393  * Add the given pair to the page
394  *
395  * Returns:
396  *	0 ==> OK
397  *	1 ==> failure
398  */
399 int
400 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
401 {
402 	u_int16_t *bp, *sop;
403 	int do_expand;
404 
405 	bp = (u_int16_t *)bufp->page;
406 	do_expand = 0;
407 	while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
408 		/* Exception case */
409 		if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
410 			/* This is the last page of a big key/data pair
411 			   and we need to add another page */
412 			break;
413 		else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
414 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
415 			if (!bufp)
416 				return (-1);
417 			bp = (u_int16_t *)bufp->page;
418 		} else if (bp[bp[0]] != OVFLPAGE) {
419 			/* Short key/data pairs, no more pages */
420 			break;
421 		} else {
422 			/* Try to squeeze key on this page */
423 			if (bp[2] >= REAL_KEY &&
424 			    FREESPACE(bp) >= PAIRSIZE(key, val)) {
425 				squeeze_key(bp, key, val);
426 				goto stats;
427 			} else {
428 				bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
429 				if (!bufp)
430 					return (-1);
431 				bp = (u_int16_t *)bufp->page;
432 			}
433 		}
434 
435 	if (PAIRFITS(bp, key, val))
436 		putpair(bufp->page, key, val);
437 	else {
438 		do_expand = 1;
439 		bufp = __add_ovflpage(hashp, bufp);
440 		if (!bufp)
441 			return (-1);
442 		sop = (u_int16_t *)bufp->page;
443 
444 		if (PAIRFITS(sop, key, val))
445 			putpair((char *)sop, key, val);
446 		else
447 			if (__big_insert(hashp, bufp, key, val))
448 				return (-1);
449 	}
450 stats:
451 	bufp->flags |= BUF_MOD;
452 	/*
453 	 * If the average number of keys per bucket exceeds the fill factor,
454 	 * expand the table.
455 	 */
456 	hashp->NKEYS++;
457 	if (do_expand ||
458 	    (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
459 		return (__expand_table(hashp));
460 	return (0);
461 }
462 
463 /*
464  *
465  * Returns:
466  *	pointer on success
467  *	NULL on error
468  */
469 BUFHEAD *
470 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
471 {
472 	u_int16_t *sp, ndx, ovfl_num;
473 #ifdef DEBUG1
474 	int tmp1, tmp2;
475 #endif
476 	sp = (u_int16_t *)bufp->page;
477 
478 	/* Check if we are dynamically determining the fill factor */
479 	if (hashp->FFACTOR == DEF_FFACTOR) {
480 		hashp->FFACTOR = sp[0] >> 1;
481 		if (hashp->FFACTOR < MIN_FFACTOR)
482 			hashp->FFACTOR = MIN_FFACTOR;
483 	}
484 	bufp->flags |= BUF_MOD;
485 	ovfl_num = overflow_page(hashp);
486 #ifdef DEBUG1
487 	tmp1 = bufp->addr;
488 	tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
489 #endif
490 	if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
491 		return (NULL);
492 	bufp->ovfl->flags |= BUF_MOD;
493 #ifdef DEBUG1
494 	(void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
495 	    tmp1, tmp2, bufp->ovfl->addr);
496 #endif
497 	ndx = sp[0];
498 	/*
499 	 * Since a pair is allocated on a page only if there's room to add
500 	 * an overflow page, we know that the OVFL information will fit on
501 	 * the page.
502 	 */
503 	sp[ndx + 4] = OFFSET(sp);
504 	sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
505 	sp[ndx + 1] = ovfl_num;
506 	sp[ndx + 2] = OVFLPAGE;
507 	sp[0] = ndx + 2;
508 #ifdef HASH_STATISTICS
509 	hash_overflows++;
510 #endif
511 	return (bufp->ovfl);
512 }
513 
514 /*
515  * Returns:
516  *	 0 indicates SUCCESS
517  *	-1 indicates FAILURE
518  */
519 int
520 __get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
521     int is_bitmap)
522 {
523 	int fd, page, size, rsize;
524 	u_int16_t *bp;
525 
526 	fd = hashp->fp;
527 	size = hashp->BSIZE;
528 
529 	if ((fd == -1) || !is_disk) {
530 		PAGE_INIT(p);
531 		return (0);
532 	}
533 	if (is_bucket)
534 		page = BUCKET_TO_PAGE(bucket);
535 	else
536 		page = OADDR_TO_PAGE(bucket);
537 	if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
538 		return (-1);
539 	bp = (u_int16_t *)p;
540 	if (!rsize)
541 		bp[0] = 0;	/* We hit the EOF, so initialize a new page */
542 	else
543 		if (rsize != size) {
544 			errno = EFTYPE;
545 			return (-1);
546 		}
547 	if (!is_bitmap && !bp[0]) {
548 		PAGE_INIT(p);
549 	} else
550 		if (hashp->LORDER != BYTE_ORDER) {
551 			int i, max;
552 
553 			if (is_bitmap) {
554 				max = hashp->BSIZE >> 2; /* divide by 4 */
555 				for (i = 0; i < max; i++)
556 					M_32_SWAP(((int *)p)[i]);
557 			} else {
558 				M_16_SWAP(bp[0]);
559 				max = bp[0] + 2;
560 				for (i = 1; i <= max; i++)
561 					M_16_SWAP(bp[i]);
562 			}
563 		}
564 	return (0);
565 }
566 
567 /*
568  * Write page p to disk
569  *
570  * Returns:
571  *	 0 ==> OK
572  *	-1 ==>failure
573  */
574 int
575 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
576 {
577 	int fd, page, size;
578 	ssize_t wsize;
579 	char pbuf[MAX_BSIZE];
580 
581 	size = hashp->BSIZE;
582 	if ((hashp->fp == -1) && open_temp(hashp))
583 		return (-1);
584 	fd = hashp->fp;
585 
586 	if (hashp->LORDER != BYTE_ORDER) {
587 		int i, max;
588 
589 		memcpy(pbuf, p, size);
590 		if (is_bitmap) {
591 			max = hashp->BSIZE >> 2;	/* divide by 4 */
592 			for (i = 0; i < max; i++)
593 				M_32_SWAP(((int *)pbuf)[i]);
594 		} else {
595 			uint16_t *bp = (uint16_t *)(void *)pbuf;
596 			max = bp[0] + 2;
597 			for (i = 0; i <= max; i++)
598 				M_16_SWAP(bp[i]);
599 		}
600 		p = pbuf;
601 	}
602 	if (is_bucket)
603 		page = BUCKET_TO_PAGE(bucket);
604 	else
605 		page = OADDR_TO_PAGE(bucket);
606 	if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -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 int
622 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
623 {
624 	u_int32_t *ip;
625 	int clearbytes, clearints;
626 
627 	if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
628 		return (1);
629 	hashp->nmaps++;
630 	clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
631 	clearbytes = clearints << INT_TO_BYTE;
632 	(void)memset((char *)ip, 0, clearbytes);
633 	(void)memset(((char *)ip) + clearbytes, 0xFF,
634 	    hashp->BSIZE - clearbytes);
635 	ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
636 	SETBIT(ip, 0);
637 	hashp->BITMAPS[ndx] = (u_int16_t)pnum;
638 	hashp->mapp[ndx] = ip;
639 	return (0);
640 }
641 
642 static u_int32_t
643 first_free(u_int32_t map)
644 {
645 	u_int32_t i, mask;
646 
647 	mask = 0x1;
648 	for (i = 0; i < BITS_PER_MAP; i++) {
649 		if (!(mask & map))
650 			return (i);
651 		mask = mask << 1;
652 	}
653 	return (i);
654 }
655 
656 static u_int16_t
657 overflow_page(HTAB *hashp)
658 {
659 	u_int32_t *freep;
660 	int max_free, offset, splitnum;
661 	u_int16_t addr;
662 	int bit, first_page, free_bit, free_page, i, in_use_bits, j;
663 #ifdef DEBUG2
664 	int tmp1, tmp2;
665 #endif
666 	splitnum = hashp->OVFL_POINT;
667 	max_free = hashp->SPARES[splitnum];
668 
669 	free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
670 	free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
671 
672 	/* Look through all the free maps to find the first free block */
673 	first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
674 	for ( i = first_page; i <= free_page; i++ ) {
675 		if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
676 		    !(freep = fetch_bitmap(hashp, i)))
677 			return (0);
678 		if (i == free_page)
679 			in_use_bits = free_bit;
680 		else
681 			in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
682 
683 		if (i == first_page) {
684 			bit = hashp->LAST_FREED &
685 			    ((hashp->BSIZE << BYTE_SHIFT) - 1);
686 			j = bit / BITS_PER_MAP;
687 			bit = rounddown2(bit, BITS_PER_MAP);
688 		} else {
689 			bit = 0;
690 			j = 0;
691 		}
692 		for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
693 			if (freep[j] != ALL_SET)
694 				goto found;
695 	}
696 
697 	/* No Free Page Found */
698 	hashp->LAST_FREED = hashp->SPARES[splitnum];
699 	hashp->SPARES[splitnum]++;
700 	offset = hashp->SPARES[splitnum] -
701 	    (splitnum ? hashp->SPARES[splitnum - 1] : 0);
702 
703 #define	OVMSG	"HASH: Out of overflow pages.  Increase page size\n"
704 	if (offset > SPLITMASK) {
705 		if (++splitnum >= NCACHED) {
706 			(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
707 			errno = EFBIG;
708 			return (0);
709 		}
710 		hashp->OVFL_POINT = splitnum;
711 		hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
712 		hashp->SPARES[splitnum-1]--;
713 		offset = 1;
714 	}
715 
716 	/* Check if we need to allocate a new bitmap page */
717 	if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
718 		free_page++;
719 		if (free_page >= NCACHED) {
720 			(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
721 			errno = EFBIG;
722 			return (0);
723 		}
724 		/*
725 		 * This is tricky.  The 1 indicates that you want the new page
726 		 * allocated with 1 clear bit.  Actually, you are going to
727 		 * allocate 2 pages from this map.  The first is going to be
728 		 * the map page, the second is the overflow page we were
729 		 * looking for.  The init_bitmap routine automatically, sets
730 		 * the first bit of itself to indicate that the bitmap itself
731 		 * is in use.  We would explicitly set the second bit, but
732 		 * don't have to if we tell init_bitmap not to leave it clear
733 		 * in the first place.
734 		 */
735 		if (__ibitmap(hashp,
736 		    (int)OADDR_OF(splitnum, offset), 1, free_page))
737 			return (0);
738 		hashp->SPARES[splitnum]++;
739 #ifdef DEBUG2
740 		free_bit = 2;
741 #endif
742 		offset++;
743 		if (offset > SPLITMASK) {
744 			if (++splitnum >= NCACHED) {
745 				(void)_write(STDERR_FILENO, OVMSG,
746 				    sizeof(OVMSG) - 1);
747 				errno = EFBIG;
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 		(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
793 		errno = EFBIG;
794 		return (0);	/* Out of overflow pages */
795 	}
796 	addr = OADDR_OF(i, offset);
797 #ifdef DEBUG2
798 	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
799 	    addr, tmp1, tmp2);
800 #endif
801 
802 	/* Allocate and return the overflow page */
803 	return (addr);
804 }
805 
806 /*
807  * Mark this overflow page as free.
808  */
809 void
810 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
811 {
812 	u_int16_t addr;
813 	u_int32_t *freep;
814 	int bit_address, free_page, free_bit;
815 	u_int16_t ndx;
816 
817 	addr = obufp->addr;
818 #ifdef DEBUG1
819 	(void)fprintf(stderr, "Freeing %d\n", addr);
820 #endif
821 	ndx = (((u_int16_t)addr) >> SPLITSHIFT);
822 	bit_address =
823 	    (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
824 	 if (bit_address < hashp->LAST_FREED)
825 		hashp->LAST_FREED = bit_address;
826 	free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
827 	free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
828 
829 	if (!(freep = hashp->mapp[free_page]))
830 		freep = fetch_bitmap(hashp, free_page);
831 #ifdef DEBUG
832 	/*
833 	 * This had better never happen.  It means we tried to read a bitmap
834 	 * that has already had overflow pages allocated off it, and we
835 	 * failed to read it from the file.
836 	 */
837 	if (!freep)
838 		assert(0);
839 #endif
840 	CLRBIT(freep, free_bit);
841 #ifdef DEBUG2
842 	(void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
843 	    obufp->addr, free_bit, free_page);
844 #endif
845 	__reclaim_buf(hashp, obufp);
846 }
847 
848 /*
849  * Returns:
850  *	 0 success
851  *	-1 failure
852  */
853 static int
854 open_temp(HTAB *hashp)
855 {
856 	sigset_t set, oset;
857 	int len;
858 	char *envtmp;
859 	char path[MAXPATHLEN];
860 
861 	envtmp = secure_getenv("TMPDIR");
862 	len = snprintf(path,
863 	    sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
864 	if (len < 0 || len >= (int)sizeof(path)) {
865 		errno = ENAMETOOLONG;
866 		return (-1);
867 	}
868 
869 	/* Block signals; make sure file goes away at process exit. */
870 	(void)sigfillset(&set);
871 	(void)__libc_sigprocmask(SIG_BLOCK, &set, &oset);
872 	if ((hashp->fp = mkostemp(path, O_CLOEXEC)) != -1)
873 		(void)unlink(path);
874 	(void)__libc_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
875 	return (hashp->fp != -1 ? 0 : -1);
876 }
877 
878 /*
879  * We have to know that the key will fit, but the last entry on the page is
880  * an overflow pair, so we need to shift things.
881  */
882 static void
883 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val)
884 {
885 	char *p;
886 	u_int16_t free_space, n, off, pageno;
887 
888 	p = (char *)sp;
889 	n = sp[0];
890 	free_space = FREESPACE(sp);
891 	off = OFFSET(sp);
892 
893 	pageno = sp[n - 1];
894 	off -= key->size;
895 	sp[n - 1] = off;
896 	memmove(p + off, key->data, key->size);
897 	off -= val->size;
898 	sp[n] = off;
899 	memmove(p + off, val->data, val->size);
900 	sp[0] = n + 2;
901 	sp[n + 1] = pageno;
902 	sp[n + 2] = OVFLPAGE;
903 	FREESPACE(sp) = free_space - PAIRSIZE(key, val);
904 	OFFSET(sp) = off;
905 }
906 
907 static u_int32_t *
908 fetch_bitmap(HTAB *hashp, int ndx)
909 {
910 	if (ndx >= hashp->nmaps)
911 		return (NULL);
912 	if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
913 		return (NULL);
914 	if (__get_page(hashp,
915 	    (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
916 		free(hashp->mapp[ndx]);
917 		return (NULL);
918 	}
919 	return (hashp->mapp[ndx]);
920 }
921 
922 #ifdef DEBUG4
923 int
924 print_chain(int addr)
925 {
926 	BUFHEAD *bufp;
927 	short *bp, oaddr;
928 
929 	(void)fprintf(stderr, "%d ", addr);
930 	bufp = __get_buf(hashp, addr, NULL, 0);
931 	bp = (short *)bufp->page;
932 	while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
933 		((bp[0] > 2) && bp[2] < REAL_KEY))) {
934 		oaddr = bp[bp[0] - 1];
935 		(void)fprintf(stderr, "%d ", (int)oaddr);
936 		bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
937 		bp = (short *)bufp->page;
938 	}
939 	(void)fprintf(stderr, "\n");
940 }
941 #endif
942