xref: /freebsd/lib/libc/db/hash/hash_page.c (revision 783d3ff6d7fae619db8a7990b8a6387de0c677b5)
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 /*
36  * PACKAGE:  hashing
37  *
38  * DESCRIPTION:
39  *	Page manipulation for hashing package.
40  *
41  * ROUTINES:
42  *
43  * External
44  *	__get_page
45  *	__add_ovflpage
46  * Internal
47  *	overflow_page
48  *	open_temp
49  */
50 
51 #include "namespace.h"
52 #include <sys/param.h>
53 
54 #include <errno.h>
55 #include <fcntl.h>
56 #include <signal.h>
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <string.h>
60 #include <unistd.h>
61 #ifdef DEBUG
62 #include <assert.h>
63 #endif
64 #include "un-namespace.h"
65 #include "libc_private.h"
66 
67 #include <db.h>
68 #include "hash.h"
69 #include "page.h"
70 #include "extern.h"
71 
72 static u_int32_t *fetch_bitmap(HTAB *, int);
73 static u_int32_t  first_free(u_int32_t);
74 static int	  open_temp(HTAB *);
75 static u_int16_t  overflow_page(HTAB *);
76 static void	  putpair(char *, const DBT *, const DBT *);
77 static void	  squeeze_key(u_int16_t *, const DBT *, const DBT *);
78 static int	  ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
79 
80 #define	PAGE_INIT(P) { \
81 	((u_int16_t *)(P))[0] = 0; \
82 	((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
83 	((u_int16_t *)(P))[2] = hashp->BSIZE; \
84 }
85 
86 /*
87  * This is called AFTER we have verified that there is room on the page for
88  * the pair (PAIRFITS has returned true) so we go right ahead and start moving
89  * stuff on.
90  */
91 static void
92 putpair(char *p, const DBT *key, const DBT *val)
93 {
94 	u_int16_t *bp, n, off;
95 
96 	bp = (u_int16_t *)p;
97 
98 	/* Enter the key first. */
99 	n = bp[0];
100 
101 	off = OFFSET(bp) - key->size;
102 	memmove(p + off, key->data, key->size);
103 	bp[++n] = off;
104 
105 	/* Now the data. */
106 	off -= val->size;
107 	memmove(p + off, val->data, val->size);
108 	bp[++n] = off;
109 
110 	/* Adjust page info. */
111 	bp[0] = n;
112 	bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
113 	bp[n + 2] = off;
114 }
115 
116 /*
117  * Returns:
118  *	 0 OK
119  *	-1 error
120  */
121 int
122 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
123 {
124 	u_int16_t *bp, newoff, pairlen;
125 	int n;
126 
127 	bp = (u_int16_t *)bufp->page;
128 	n = bp[0];
129 
130 	if (bp[ndx + 1] < REAL_KEY)
131 		return (__big_delete(hashp, bufp));
132 	if (ndx != 1)
133 		newoff = bp[ndx - 1];
134 	else
135 		newoff = hashp->BSIZE;
136 	pairlen = newoff - bp[ndx + 1];
137 
138 	if (ndx != (n - 1)) {
139 		/* Hard Case -- need to shuffle keys */
140 		int i;
141 		char *src = bufp->page + (int)OFFSET(bp);
142 		char *dst = src + (int)pairlen;
143 		memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
144 
145 		/* Now adjust the pointers */
146 		for (i = ndx + 2; i <= n; i += 2) {
147 			if (bp[i + 1] == OVFLPAGE) {
148 				bp[i - 2] = bp[i];
149 				bp[i - 1] = bp[i + 1];
150 			} else {
151 				bp[i - 2] = bp[i] + pairlen;
152 				bp[i - 1] = bp[i + 1] + pairlen;
153 			}
154 		}
155 		if (ndx == hashp->cndx) {
156 			/*
157 			 * We just removed pair we were "pointing" to.
158 			 * By moving back the cndx we ensure subsequent
159 			 * hash_seq() calls won't skip over any entries.
160 			 */
161 			hashp->cndx -= 2;
162 		}
163 	}
164 	/* Finally adjust the page data */
165 	bp[n] = OFFSET(bp) + pairlen;
166 	bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
167 	bp[0] = n - 2;
168 	hashp->NKEYS--;
169 
170 	bufp->flags |= BUF_MOD;
171 	return (0);
172 }
173 /*
174  * Returns:
175  *	 0 ==> OK
176  *	-1 ==> Error
177  */
178 int
179 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
180 {
181 	BUFHEAD *new_bufp, *old_bufp;
182 	u_int16_t *ino;
183 	char *np;
184 	DBT key, val;
185 	int n, ndx, retval;
186 	u_int16_t copyto, diff, off, moved;
187 	char *op;
188 
189 	copyto = (u_int16_t)hashp->BSIZE;
190 	off = (u_int16_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 = (u_int16_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(u_int16_t) * (ino[0] + 3);
244 	OFFSET(ino) = copyto;
245 
246 #ifdef DEBUG3
247 	(void)fprintf(stderr, "split %d/%d\n",
248 	    ((u_int16_t *)np)[0] / 2,
249 	    ((u_int16_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
273 ugly_split(HTAB *hashp,
274     u_int32_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 	u_int16_t *ino;	/* Page keys come off of */
282 	u_int16_t *np;	/* New page */
283 	u_int16_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 	u_int16_t n, off, ov_addr, scopyto;
289 	char *cino;		/* Character value of ino */
290 
291 	bufp = old_bufp;
292 	ino = (u_int16_t *)old_bufp->page;
293 	np = (u_int16_t *)new_bufp->page;
294 	op = (u_int16_t *)old_bufp->page;
295 	last_bfp = NULL;
296 	scopyto = (u_int16_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 = (u_int16_t *)old_bufp->page;
308 			new_bufp = ret.newp;
309 			if (!new_bufp)
310 				return (-1);
311 			np = (u_int16_t *)new_bufp->page;
312 			bufp = ret.nextp;
313 			if (!bufp)
314 				return (0);
315 			cino = (char *)bufp->page;
316 			ino = (u_int16_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(u_int16_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 = (u_int16_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 = (u_int16_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 = (u_int16_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 int
394 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
395 {
396 	u_int16_t *bp, *sop;
397 	int do_expand;
398 
399 	bp = (u_int16_t *)bufp->page;
400 	do_expand = 0;
401 	while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
402 		/* Exception case */
403 		if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
404 			/* This is the last page of a big key/data pair
405 			   and we need to add another page */
406 			break;
407 		else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
408 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
409 			if (!bufp)
410 				return (-1);
411 			bp = (u_int16_t *)bufp->page;
412 		} else if (bp[bp[0]] != OVFLPAGE) {
413 			/* Short key/data pairs, no more pages */
414 			break;
415 		} else {
416 			/* Try to squeeze key on this page */
417 			if (bp[2] >= REAL_KEY &&
418 			    FREESPACE(bp) >= PAIRSIZE(key, val)) {
419 				squeeze_key(bp, key, val);
420 				goto stats;
421 			} else {
422 				bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
423 				if (!bufp)
424 					return (-1);
425 				bp = (u_int16_t *)bufp->page;
426 			}
427 		}
428 
429 	if (PAIRFITS(bp, key, val))
430 		putpair(bufp->page, key, val);
431 	else {
432 		do_expand = 1;
433 		bufp = __add_ovflpage(hashp, bufp);
434 		if (!bufp)
435 			return (-1);
436 		sop = (u_int16_t *)bufp->page;
437 
438 		if (PAIRFITS(sop, key, val))
439 			putpair((char *)sop, key, val);
440 		else
441 			if (__big_insert(hashp, bufp, key, val))
442 				return (-1);
443 	}
444 stats:
445 	bufp->flags |= BUF_MOD;
446 	/*
447 	 * If the average number of keys per bucket exceeds the fill factor,
448 	 * expand the table.
449 	 */
450 	hashp->NKEYS++;
451 	if (do_expand ||
452 	    (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
453 		return (__expand_table(hashp));
454 	return (0);
455 }
456 
457 /*
458  *
459  * Returns:
460  *	pointer on success
461  *	NULL on error
462  */
463 BUFHEAD *
464 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
465 {
466 	u_int16_t *sp, ndx, ovfl_num;
467 #ifdef DEBUG1
468 	int tmp1, tmp2;
469 #endif
470 	sp = (u_int16_t *)bufp->page;
471 
472 	/* Check if we are dynamically determining the fill factor */
473 	if (hashp->FFACTOR == DEF_FFACTOR) {
474 		hashp->FFACTOR = sp[0] >> 1;
475 		if (hashp->FFACTOR < MIN_FFACTOR)
476 			hashp->FFACTOR = MIN_FFACTOR;
477 	}
478 	bufp->flags |= BUF_MOD;
479 	ovfl_num = overflow_page(hashp);
480 #ifdef DEBUG1
481 	tmp1 = bufp->addr;
482 	tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
483 #endif
484 	if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
485 		return (NULL);
486 	bufp->ovfl->flags |= BUF_MOD;
487 #ifdef DEBUG1
488 	(void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
489 	    tmp1, tmp2, bufp->ovfl->addr);
490 #endif
491 	ndx = sp[0];
492 	/*
493 	 * Since a pair is allocated on a page only if there's room to add
494 	 * an overflow page, we know that the OVFL information will fit on
495 	 * the page.
496 	 */
497 	sp[ndx + 4] = OFFSET(sp);
498 	sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
499 	sp[ndx + 1] = ovfl_num;
500 	sp[ndx + 2] = OVFLPAGE;
501 	sp[0] = ndx + 2;
502 #ifdef HASH_STATISTICS
503 	hash_overflows++;
504 #endif
505 	return (bufp->ovfl);
506 }
507 
508 /*
509  * Returns:
510  *	 0 indicates SUCCESS
511  *	-1 indicates FAILURE
512  */
513 int
514 __get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
515     int is_bitmap)
516 {
517 	int fd, page, size, rsize;
518 	u_int16_t *bp;
519 
520 	fd = hashp->fp;
521 	size = hashp->BSIZE;
522 
523 	if ((fd == -1) || !is_disk) {
524 		PAGE_INIT(p);
525 		return (0);
526 	}
527 	if (is_bucket)
528 		page = BUCKET_TO_PAGE(bucket);
529 	else
530 		page = OADDR_TO_PAGE(bucket);
531 	if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
532 		return (-1);
533 	bp = (u_int16_t *)p;
534 	if (!rsize)
535 		bp[0] = 0;	/* We hit the EOF, so initialize a new page */
536 	else
537 		if (rsize != size) {
538 			errno = EFTYPE;
539 			return (-1);
540 		}
541 	if (!is_bitmap && !bp[0]) {
542 		PAGE_INIT(p);
543 	} else
544 		if (hashp->LORDER != BYTE_ORDER) {
545 			int i, max;
546 
547 			if (is_bitmap) {
548 				max = hashp->BSIZE >> 2; /* divide by 4 */
549 				for (i = 0; i < max; i++)
550 					M_32_SWAP(((int *)p)[i]);
551 			} else {
552 				M_16_SWAP(bp[0]);
553 				max = bp[0] + 2;
554 				for (i = 1; i <= max; i++)
555 					M_16_SWAP(bp[i]);
556 			}
557 		}
558 	return (0);
559 }
560 
561 /*
562  * Write page p to disk
563  *
564  * Returns:
565  *	 0 ==> OK
566  *	-1 ==>failure
567  */
568 int
569 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
570 {
571 	int fd, page, size;
572 	ssize_t wsize;
573 	char pbuf[MAX_BSIZE];
574 
575 	size = hashp->BSIZE;
576 	if ((hashp->fp == -1) && open_temp(hashp))
577 		return (-1);
578 	fd = hashp->fp;
579 
580 	if (hashp->LORDER != BYTE_ORDER) {
581 		int i, max;
582 
583 		memcpy(pbuf, p, size);
584 		if (is_bitmap) {
585 			max = hashp->BSIZE >> 2;	/* divide by 4 */
586 			for (i = 0; i < max; i++)
587 				M_32_SWAP(((int *)pbuf)[i]);
588 		} else {
589 			uint16_t *bp = (uint16_t *)(void *)pbuf;
590 			max = bp[0] + 2;
591 			for (i = 0; i <= max; i++)
592 				M_16_SWAP(bp[i]);
593 		}
594 		p = pbuf;
595 	}
596 	if (is_bucket)
597 		page = BUCKET_TO_PAGE(bucket);
598 	else
599 		page = OADDR_TO_PAGE(bucket);
600 	if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
601 		/* Errno is set */
602 		return (-1);
603 	if (wsize != size) {
604 		errno = EFTYPE;
605 		return (-1);
606 	}
607 	return (0);
608 }
609 
610 #define BYTE_MASK	((1 << INT_BYTE_SHIFT) -1)
611 /*
612  * Initialize a new bitmap page.  Bitmap pages are left in memory
613  * once they are read in.
614  */
615 int
616 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
617 {
618 	u_int32_t *ip;
619 	int clearbytes, clearints;
620 
621 	if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
622 		return (1);
623 	hashp->nmaps++;
624 	clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
625 	clearbytes = clearints << INT_TO_BYTE;
626 	(void)memset((char *)ip, 0, clearbytes);
627 	(void)memset(((char *)ip) + clearbytes, 0xFF,
628 	    hashp->BSIZE - clearbytes);
629 	ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
630 	SETBIT(ip, 0);
631 	hashp->BITMAPS[ndx] = (u_int16_t)pnum;
632 	hashp->mapp[ndx] = ip;
633 	return (0);
634 }
635 
636 static u_int32_t
637 first_free(u_int32_t map)
638 {
639 	u_int32_t i, mask;
640 
641 	mask = 0x1;
642 	for (i = 0; i < BITS_PER_MAP; i++) {
643 		if (!(mask & map))
644 			return (i);
645 		mask = mask << 1;
646 	}
647 	return (i);
648 }
649 
650 static u_int16_t
651 overflow_page(HTAB *hashp)
652 {
653 	u_int32_t *freep;
654 	int max_free, offset, splitnum;
655 	u_int16_t addr;
656 	int bit, first_page, free_bit, free_page, i, in_use_bits, j;
657 #ifdef DEBUG2
658 	int tmp1, tmp2;
659 #endif
660 	splitnum = hashp->OVFL_POINT;
661 	max_free = hashp->SPARES[splitnum];
662 
663 	free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
664 	free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
665 
666 	/* Look through all the free maps to find the first free block */
667 	first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
668 	for ( i = first_page; i <= free_page; i++ ) {
669 		if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
670 		    !(freep = fetch_bitmap(hashp, i)))
671 			return (0);
672 		if (i == free_page)
673 			in_use_bits = free_bit;
674 		else
675 			in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
676 
677 		if (i == first_page) {
678 			bit = hashp->LAST_FREED &
679 			    ((hashp->BSIZE << BYTE_SHIFT) - 1);
680 			j = bit / BITS_PER_MAP;
681 			bit = rounddown2(bit, BITS_PER_MAP);
682 		} else {
683 			bit = 0;
684 			j = 0;
685 		}
686 		for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
687 			if (freep[j] != ALL_SET)
688 				goto found;
689 	}
690 
691 	/* No Free Page Found */
692 	hashp->LAST_FREED = hashp->SPARES[splitnum];
693 	hashp->SPARES[splitnum]++;
694 	offset = hashp->SPARES[splitnum] -
695 	    (splitnum ? hashp->SPARES[splitnum - 1] : 0);
696 
697 #define	OVMSG	"HASH: Out of overflow pages.  Increase page size\n"
698 	if (offset > SPLITMASK) {
699 		if (++splitnum >= NCACHED) {
700 			(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
701 			errno = EFBIG;
702 			return (0);
703 		}
704 		hashp->OVFL_POINT = splitnum;
705 		hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
706 		hashp->SPARES[splitnum-1]--;
707 		offset = 1;
708 	}
709 
710 	/* Check if we need to allocate a new bitmap page */
711 	if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
712 		free_page++;
713 		if (free_page >= NCACHED) {
714 			(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
715 			errno = EFBIG;
716 			return (0);
717 		}
718 		/*
719 		 * This is tricky.  The 1 indicates that you want the new page
720 		 * allocated with 1 clear bit.  Actually, you are going to
721 		 * allocate 2 pages from this map.  The first is going to be
722 		 * the map page, the second is the overflow page we were
723 		 * looking for.  The init_bitmap routine automatically, sets
724 		 * the first bit of itself to indicate that the bitmap itself
725 		 * is in use.  We would explicitly set the second bit, but
726 		 * don't have to if we tell init_bitmap not to leave it clear
727 		 * in the first place.
728 		 */
729 		if (__ibitmap(hashp,
730 		    (int)OADDR_OF(splitnum, offset), 1, free_page))
731 			return (0);
732 		hashp->SPARES[splitnum]++;
733 #ifdef DEBUG2
734 		free_bit = 2;
735 #endif
736 		offset++;
737 		if (offset > SPLITMASK) {
738 			if (++splitnum >= NCACHED) {
739 				(void)_write(STDERR_FILENO, OVMSG,
740 				    sizeof(OVMSG) - 1);
741 				errno = EFBIG;
742 				return (0);
743 			}
744 			hashp->OVFL_POINT = splitnum;
745 			hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
746 			hashp->SPARES[splitnum-1]--;
747 			offset = 0;
748 		}
749 	} else {
750 		/*
751 		 * Free_bit addresses the last used bit.  Bump it to address
752 		 * the first available bit.
753 		 */
754 		free_bit++;
755 		SETBIT(freep, free_bit);
756 	}
757 
758 	/* Calculate address of the new overflow page */
759 	addr = OADDR_OF(splitnum, offset);
760 #ifdef DEBUG2
761 	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
762 	    addr, free_bit, free_page);
763 #endif
764 	return (addr);
765 
766 found:
767 	bit = bit + first_free(freep[j]);
768 	SETBIT(freep, bit);
769 #ifdef DEBUG2
770 	tmp1 = bit;
771 	tmp2 = i;
772 #endif
773 	/*
774 	 * Bits are addressed starting with 0, but overflow pages are addressed
775 	 * beginning at 1. Bit is a bit addressnumber, so we need to increment
776 	 * it to convert it to a page number.
777 	 */
778 	bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
779 	if (bit >= hashp->LAST_FREED)
780 		hashp->LAST_FREED = bit - 1;
781 
782 	/* Calculate the split number for this page */
783 	for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
784 	offset = (i ? bit - hashp->SPARES[i - 1] : bit);
785 	if (offset >= SPLITMASK) {
786 		(void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
787 		errno = EFBIG;
788 		return (0);	/* Out of overflow pages */
789 	}
790 	addr = OADDR_OF(i, offset);
791 #ifdef DEBUG2
792 	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
793 	    addr, tmp1, tmp2);
794 #endif
795 
796 	/* Allocate and return the overflow page */
797 	return (addr);
798 }
799 
800 /*
801  * Mark this overflow page as free.
802  */
803 void
804 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
805 {
806 	u_int16_t addr;
807 	u_int32_t *freep;
808 	int bit_address, free_page, free_bit;
809 	u_int16_t ndx;
810 
811 	addr = obufp->addr;
812 #ifdef DEBUG1
813 	(void)fprintf(stderr, "Freeing %d\n", addr);
814 #endif
815 	ndx = (((u_int16_t)addr) >> SPLITSHIFT);
816 	bit_address =
817 	    (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
818 	 if (bit_address < hashp->LAST_FREED)
819 		hashp->LAST_FREED = bit_address;
820 	free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
821 	free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
822 
823 	if (!(freep = hashp->mapp[free_page]))
824 		freep = fetch_bitmap(hashp, free_page);
825 #ifdef DEBUG
826 	/*
827 	 * This had better never happen.  It means we tried to read a bitmap
828 	 * that has already had overflow pages allocated off it, and we
829 	 * failed to read it from the file.
830 	 */
831 	if (!freep)
832 		assert(0);
833 #endif
834 	CLRBIT(freep, free_bit);
835 #ifdef DEBUG2
836 	(void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
837 	    obufp->addr, free_bit, free_page);
838 #endif
839 	__reclaim_buf(hashp, obufp);
840 }
841 
842 /*
843  * Returns:
844  *	 0 success
845  *	-1 failure
846  */
847 static int
848 open_temp(HTAB *hashp)
849 {
850 	sigset_t set, oset;
851 	int len;
852 	char *envtmp;
853 	char path[MAXPATHLEN];
854 
855 	envtmp = secure_getenv("TMPDIR");
856 	len = snprintf(path,
857 	    sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
858 	if (len < 0 || len >= (int)sizeof(path)) {
859 		errno = ENAMETOOLONG;
860 		return (-1);
861 	}
862 
863 	/* Block signals; make sure file goes away at process exit. */
864 	(void)sigfillset(&set);
865 	(void)__libc_sigprocmask(SIG_BLOCK, &set, &oset);
866 	if ((hashp->fp = mkostemp(path, O_CLOEXEC)) != -1)
867 		(void)unlink(path);
868 	(void)__libc_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
869 	return (hashp->fp != -1 ? 0 : -1);
870 }
871 
872 /*
873  * We have to know that the key will fit, but the last entry on the page is
874  * an overflow pair, so we need to shift things.
875  */
876 static void
877 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val)
878 {
879 	char *p;
880 	u_int16_t free_space, n, off, pageno;
881 
882 	p = (char *)sp;
883 	n = sp[0];
884 	free_space = FREESPACE(sp);
885 	off = OFFSET(sp);
886 
887 	pageno = sp[n - 1];
888 	off -= key->size;
889 	sp[n - 1] = off;
890 	memmove(p + off, key->data, key->size);
891 	off -= val->size;
892 	sp[n] = off;
893 	memmove(p + off, val->data, val->size);
894 	sp[0] = n + 2;
895 	sp[n + 1] = pageno;
896 	sp[n + 2] = OVFLPAGE;
897 	FREESPACE(sp) = free_space - PAIRSIZE(key, val);
898 	OFFSET(sp) = off;
899 }
900 
901 static u_int32_t *
902 fetch_bitmap(HTAB *hashp, int ndx)
903 {
904 	if (ndx >= hashp->nmaps)
905 		return (NULL);
906 	if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
907 		return (NULL);
908 	if (__get_page(hashp,
909 	    (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
910 		free(hashp->mapp[ndx]);
911 		return (NULL);
912 	}
913 	return (hashp->mapp[ndx]);
914 }
915 
916 #ifdef DEBUG4
917 int
918 print_chain(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