xref: /csrg-svn/lib/libc/db/hash/hash_page.c (revision 66215)
146372Sbostic /*-
262489Sbostic  * Copyright (c) 1990, 1993
362489Sbostic  *	The Regents of the University of California.  All rights reserved.
446372Sbostic  *
546372Sbostic  * This code is derived from software contributed to Berkeley by
646372Sbostic  * Margo Seltzer.
746372Sbostic  *
846372Sbostic  * %sccs.include.redist.c%
946372Sbostic  */
1046372Sbostic 
1146372Sbostic #if defined(LIBC_SCCS) && !defined(lint)
12*66215Sbostic static char sccsid[] = "@(#)hash_page.c	8.4 (Berkeley) 02/21/94";
1346372Sbostic #endif /* LIBC_SCCS and not lint */
1446372Sbostic 
1550997Sbostic /*
1650997Sbostic  * PACKAGE:  hashing
1750997Sbostic  *
1850997Sbostic  * DESCRIPTION:
1950997Sbostic  *	Page manipulation for hashing package.
2050997Sbostic  *
2150997Sbostic  * ROUTINES:
2250997Sbostic  *
2350997Sbostic  * External
2450997Sbostic  *	__get_page
2550997Sbostic  *	__add_ovflpage
2650997Sbostic  * Internal
2750997Sbostic  *	overflow_page
2850997Sbostic  *	open_temp
2950997Sbostic  */
3046372Sbostic 
3157586Sbostic #include <sys/types.h>
3257586Sbostic 
3357586Sbostic #include <errno.h>
3446562Sbostic #include <fcntl.h>
3546416Sbostic #include <signal.h>
3646372Sbostic #include <stdio.h>
3746562Sbostic #include <stdlib.h>
3846562Sbostic #include <string.h>
3946500Sbostic #include <unistd.h>
4050997Sbostic #ifdef DEBUG
4150997Sbostic #include <assert.h>
4250997Sbostic #endif
4357586Sbostic 
4457932Sbostic #include <db.h>
4546372Sbostic #include "hash.h"
4646372Sbostic #include "page.h"
4750997Sbostic #include "extern.h"
4846372Sbostic 
4957586Sbostic static u_long	*fetch_bitmap __P((HTAB *, int));
5051055Sbostic static u_long	 first_free __P((u_long));
5157586Sbostic static int	 open_temp __P((HTAB *));
5257586Sbostic static u_short	 overflow_page __P((HTAB *));
5351055Sbostic static void	 putpair __P((char *, const DBT *, const DBT *));
5451055Sbostic static void	 squeeze_key __P((u_short *, const DBT *, const DBT *));
5557586Sbostic static int	 ugly_split
5657586Sbostic 		    __P((HTAB *, u_int, BUFHEAD *, BUFHEAD *, int, int));
5746372Sbostic 
5850997Sbostic #define	PAGE_INIT(P) { \
5950997Sbostic 	((u_short *)(P))[0] = 0; \
6050997Sbostic 	((u_short *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_short); \
6150997Sbostic 	((u_short *)(P))[2] = hashp->BSIZE; \
6246372Sbostic }
6346372Sbostic 
6446372Sbostic /*
6550997Sbostic  * This is called AFTER we have verified that there is room on the page for
6650997Sbostic  * the pair (PAIRFITS has returned true) so we go right ahead and start moving
6750997Sbostic  * stuff on.
6850997Sbostic  */
6946372Sbostic static void
putpair(p,key,val)7046372Sbostic putpair(p, key, val)
7150997Sbostic 	char *p;
7250997Sbostic 	const DBT *key, *val;
7346372Sbostic {
7450997Sbostic 	register u_short *bp, n, off;
7546372Sbostic 
7650997Sbostic 	bp = (u_short *)p;
7750997Sbostic 
7850997Sbostic 	/* Enter the key first. */
7946372Sbostic 	n = bp[0];
8046372Sbostic 
8146372Sbostic 	off = OFFSET(bp) - key->size;
8258016Sbostic 	memmove(p + off, key->data, key->size);
8346372Sbostic 	bp[++n] = off;
8446372Sbostic 
8550997Sbostic 	/* Now the data. */
8646372Sbostic 	off -= val->size;
8758016Sbostic 	memmove(p + off, val->data, val->size);
8846372Sbostic 	bp[++n] = off;
8946372Sbostic 
9050997Sbostic 	/* Adjust page info. */
9146372Sbostic 	bp[0] = n;
9250997Sbostic 	bp[n + 1] = off - ((n + 3) * sizeof(u_short));
9350997Sbostic 	bp[n + 2] = off;
9446372Sbostic }
9550997Sbostic 
9646372Sbostic /*
9750997Sbostic  * Returns:
9850997Sbostic  *	 0 OK
9950997Sbostic  *	-1 error
10050997Sbostic  */
10146372Sbostic extern int
10257586Sbostic __delpair(hashp, bufp, ndx)
10357586Sbostic 	HTAB *hashp;
10450997Sbostic 	BUFHEAD *bufp;
10550997Sbostic 	register int ndx;
10646372Sbostic {
10750997Sbostic 	register u_short *bp, newoff;
10850997Sbostic 	register int n;
10946372Sbostic 	u_short pairlen;
11046372Sbostic 
11150997Sbostic 	bp = (u_short *)bufp->page;
11250997Sbostic 	n = bp[0];
11346372Sbostic 
11450997Sbostic 	if (bp[ndx + 1] < REAL_KEY)
11557586Sbostic 		return (__big_delete(hashp, bufp));
11650997Sbostic 	if (ndx != 1)
11750997Sbostic 		newoff = bp[ndx - 1];
11850997Sbostic 	else
11950997Sbostic 		newoff = hashp->BSIZE;
12050997Sbostic 	pairlen = newoff - bp[ndx + 1];
12150997Sbostic 
12250997Sbostic 	if (ndx != (n - 1)) {
12346372Sbostic 		/* Hard Case -- need to shuffle keys */
12446372Sbostic 		register int i;
12546372Sbostic 		register char *src = bufp->page + (int)OFFSET(bp);
12646372Sbostic 		register char *dst = src + (int)pairlen;
12758016Sbostic 		memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
12846372Sbostic 
12946372Sbostic 		/* Now adjust the pointers */
13050997Sbostic 		for (i = ndx + 2; i <= n; i += 2) {
13150997Sbostic 			if (bp[i + 1] == OVFLPAGE) {
13250997Sbostic 				bp[i - 2] = bp[i];
13350997Sbostic 				bp[i - 1] = bp[i + 1];
13450997Sbostic 			} else {
13550997Sbostic 				bp[i - 2] = bp[i] + pairlen;
13650997Sbostic 				bp[i - 1] = bp[i + 1] + pairlen;
13750997Sbostic 			}
13846372Sbostic 		}
13946372Sbostic 	}
14046372Sbostic 	/* Finally adjust the page data */
14146372Sbostic 	bp[n] = OFFSET(bp) + pairlen;
14250997Sbostic 	bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_short);
14350997Sbostic 	bp[0] = n - 2;
14446372Sbostic 	hashp->NKEYS--;
14546372Sbostic 
14646372Sbostic 	bufp->flags |= BUF_MOD;
14746372Sbostic 	return (0);
14846372Sbostic }
14946372Sbostic /*
15050997Sbostic  * Returns:
15150997Sbostic  *	 0 ==> OK
15250997Sbostic  *	-1 ==> Error
15350997Sbostic  */
15446372Sbostic extern int
15557586Sbostic __split_page(hashp, obucket, nbucket)
15657586Sbostic 	HTAB *hashp;
15750997Sbostic 	u_int obucket, nbucket;
15846372Sbostic {
15950997Sbostic 	register BUFHEAD *new_bufp, *old_bufp;
16046372Sbostic 	register u_short *ino;
16150997Sbostic 	register char *np;
16250997Sbostic 	DBT key, val;
16350997Sbostic 	int n, ndx, retval;
16450997Sbostic 	u_short copyto, diff, off, moved;
16550997Sbostic 	char *op;
16646372Sbostic 
16750997Sbostic 	copyto = (u_short)hashp->BSIZE;
16850997Sbostic 	off = (u_short)hashp->BSIZE;
16957586Sbostic 	old_bufp = __get_buf(hashp, obucket, NULL, 0);
17053870Sbostic 	if (old_bufp == NULL)
17153870Sbostic 		return (-1);
17257586Sbostic 	new_bufp = __get_buf(hashp, nbucket, NULL, 0);
17353870Sbostic 	if (new_bufp == NULL)
17453870Sbostic 		return (-1);
17546372Sbostic 
17650997Sbostic 	old_bufp->flags |= (BUF_MOD | BUF_PIN);
17750997Sbostic 	new_bufp->flags |= (BUF_MOD | BUF_PIN);
17846372Sbostic 
17946372Sbostic 	ino = (u_short *)(op = old_bufp->page);
18046372Sbostic 	np = new_bufp->page;
18146372Sbostic 
18246372Sbostic 	moved = 0;
18346372Sbostic 
18450997Sbostic 	for (n = 1, ndx = 1; n < ino[0]; n += 2) {
18550997Sbostic 		if (ino[n + 1] < REAL_KEY) {
18657586Sbostic 			retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
18751055Sbostic 			    (int)copyto, (int)moved);
18850997Sbostic 			old_bufp->flags &= ~BUF_PIN;
18950997Sbostic 			new_bufp->flags &= ~BUF_PIN;
19050997Sbostic 			return (retval);
19150997Sbostic 
19246372Sbostic 		}
19350997Sbostic 		key.data = (u_char *)op + ino[n];
19446372Sbostic 		key.size = off - ino[n];
19546372Sbostic 
19657586Sbostic 		if (__call_hash(hashp, key.data, key.size) == obucket) {
19750997Sbostic 			/* Don't switch page */
19850997Sbostic 			diff = copyto - off;
19950997Sbostic 			if (diff) {
20050997Sbostic 				copyto = ino[n + 1] + diff;
20158016Sbostic 				memmove(op + copyto, op + ino[n + 1],
20250997Sbostic 				    off - ino[n + 1]);
20350997Sbostic 				ino[ndx] = copyto + ino[n] - ino[n + 1];
20450997Sbostic 				ino[ndx + 1] = copyto;
20550997Sbostic 			} else
20650997Sbostic 				copyto = ino[n + 1];
20750997Sbostic 			ndx += 2;
20846372Sbostic 		} else {
20950997Sbostic 			/* Switch page */
21050997Sbostic 			val.data = (u_char *)op + ino[n + 1];
21150997Sbostic 			val.size = ino[n] - ino[n + 1];
21250997Sbostic 			putpair(np, &key, &val);
21350997Sbostic 			moved += 2;
21446372Sbostic 		}
21546372Sbostic 
21650997Sbostic 		off = ino[n + 1];
21746372Sbostic 	}
21846372Sbostic 
21946372Sbostic 	/* Now clean up the page */
22046372Sbostic 	ino[0] -= moved;
22150997Sbostic 	FREESPACE(ino) = copyto - sizeof(u_short) * (ino[0] + 3);
22246372Sbostic 	OFFSET(ino) = copyto;
22346372Sbostic 
22446372Sbostic #ifdef DEBUG3
22550997Sbostic 	(void)fprintf(stderr, "split %d/%d\n",
22650997Sbostic 	    ((u_short *)np)[0] / 2,
22750997Sbostic 	    ((u_short *)op)[0] / 2);
22846372Sbostic #endif
22946460Sbostic 	/* unpin both pages */
23046460Sbostic 	old_bufp->flags &= ~BUF_PIN;
23146460Sbostic 	new_bufp->flags &= ~BUF_PIN;
23250997Sbostic 	return (0);
23346372Sbostic }
23450997Sbostic 
23546372Sbostic /*
23650997Sbostic  * Called when we encounter an overflow or big key/data page during split
23750997Sbostic  * handling.  This is special cased since we have to begin checking whether
23850997Sbostic  * the key/data pairs fit on their respective pages and because we may need
23950997Sbostic  * overflow pages for both the old and new pages.
24050997Sbostic  *
24150997Sbostic  * The first page might be a page with regular key/data pairs in which case
24250997Sbostic  * we have a regular overflow condition and just need to go on to the next
24350997Sbostic  * page or it might be a big key/data pair in which case we need to fix the
24450997Sbostic  * big key/data pair.
24550997Sbostic  *
24650997Sbostic  * Returns:
24750997Sbostic  *	 0 ==> success
24850997Sbostic  *	-1 ==> failure
24950997Sbostic  */
25046372Sbostic static int
ugly_split(hashp,obucket,old_bufp,new_bufp,copyto,moved)25157586Sbostic ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved)
25257586Sbostic 	HTAB *hashp;
25350997Sbostic 	u_int obucket;	/* Same as __split_page. */
25450997Sbostic 	BUFHEAD *old_bufp, *new_bufp;
25550997Sbostic 	int copyto;	/* First byte on page which contains key/data values. */
25650997Sbostic 	int moved;	/* Number of pairs moved to new page. */
25746372Sbostic {
25850997Sbostic 	register BUFHEAD *bufp;	/* Buffer header for ino */
25950997Sbostic 	register u_short *ino;	/* Page keys come off of */
26050997Sbostic 	register u_short *np;	/* New page */
26150997Sbostic 	register u_short *op;	/* Page keys go on to if they aren't moving */
26246372Sbostic 
26350997Sbostic 	BUFHEAD *last_bfp;	/* Last buf header OVFL needing to be freed */
26450997Sbostic 	DBT key, val;
26550997Sbostic 	SPLIT_RETURN ret;
26651055Sbostic 	u_short n, off, ov_addr, scopyto;
26750997Sbostic 	char *cino;		/* Character value of ino */
26846372Sbostic 
26950997Sbostic 	bufp = old_bufp;
27050997Sbostic 	ino = (u_short *)old_bufp->page;
27150997Sbostic 	np = (u_short *)new_bufp->page;
27250997Sbostic 	op = (u_short *)old_bufp->page;
27350997Sbostic 	last_bfp = NULL;
27450997Sbostic 	scopyto = (u_short)copyto;	/* ANSI */
27546372Sbostic 
27650997Sbostic 	n = ino[0] - 1;
27750997Sbostic 	while (n < ino[0]) {
27850997Sbostic 		if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
27957586Sbostic 			if (__big_split(hashp, old_bufp,
28064445Sbostic 			    new_bufp, bufp, bufp->addr, obucket, &ret))
28150997Sbostic 				return (-1);
28250997Sbostic 			old_bufp = ret.oldp;
28350997Sbostic 			if (!old_bufp)
28450997Sbostic 				return (-1);
28550997Sbostic 			op = (u_short *)old_bufp->page;
28650997Sbostic 			new_bufp = ret.newp;
28750997Sbostic 			if (!new_bufp)
28850997Sbostic 				return (-1);
28950997Sbostic 			np = (u_short *)new_bufp->page;
29050997Sbostic 			bufp = ret.nextp;
29150997Sbostic 			if (!bufp)
29250997Sbostic 				return (0);
29350997Sbostic 			cino = (char *)bufp->page;
29450997Sbostic 			ino = (u_short *)cino;
29550997Sbostic 			last_bfp = ret.nextp;
29650997Sbostic 		} else if (ino[n + 1] == OVFLPAGE) {
29750997Sbostic 			ov_addr = ino[n];
29850997Sbostic 			/*
29950997Sbostic 			 * Fix up the old page -- the extra 2 are the fields
30050997Sbostic 			 * which contained the overflow information.
30150997Sbostic 			 */
30250997Sbostic 			ino[0] -= (moved + 2);
30350997Sbostic 			FREESPACE(ino) =
30450997Sbostic 			    scopyto - sizeof(u_short) * (ino[0] + 3);
30550997Sbostic 			OFFSET(ino) = scopyto;
30646372Sbostic 
30757586Sbostic 			bufp = __get_buf(hashp, ov_addr, bufp, 0);
30850997Sbostic 			if (!bufp)
30950997Sbostic 				return (-1);
31046372Sbostic 
31150997Sbostic 			ino = (u_short *)bufp->page;
31250997Sbostic 			n = 1;
31350997Sbostic 			scopyto = hashp->BSIZE;
31450997Sbostic 			moved = 0;
31546372Sbostic 
31650997Sbostic 			if (last_bfp)
31757586Sbostic 				__free_ovflpage(hashp, last_bfp);
31850997Sbostic 			last_bfp = bufp;
31950997Sbostic 		}
32050997Sbostic 		/* Move regular sized pairs of there are any */
32150997Sbostic 		off = hashp->BSIZE;
32250997Sbostic 		for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
32350997Sbostic 			cino = (char *)ino;
32450997Sbostic 			key.data = (u_char *)cino + ino[n];
32550997Sbostic 			key.size = off - ino[n];
32650997Sbostic 			val.data = (u_char *)cino + ino[n + 1];
32750997Sbostic 			val.size = ino[n] - ino[n + 1];
32850997Sbostic 			off = ino[n + 1];
32946372Sbostic 
33057586Sbostic 			if (__call_hash(hashp, key.data, key.size) == obucket) {
33150997Sbostic 				/* Keep on old page */
33250997Sbostic 				if (PAIRFITS(op, (&key), (&val)))
33350997Sbostic 					putpair((char *)op, &key, &val);
33450997Sbostic 				else {
33557586Sbostic 					old_bufp =
33657586Sbostic 					    __add_ovflpage(hashp, old_bufp);
33750997Sbostic 					if (!old_bufp)
33850997Sbostic 						return (-1);
33950997Sbostic 					op = (u_short *)old_bufp->page;
34050997Sbostic 					putpair((char *)op, &key, &val);
34150997Sbostic 				}
34250997Sbostic 				old_bufp->flags |= BUF_MOD;
34350997Sbostic 			} else {
34450997Sbostic 				/* Move to new page */
34550997Sbostic 				if (PAIRFITS(np, (&key), (&val)))
34650997Sbostic 					putpair((char *)np, &key, &val);
34750997Sbostic 				else {
34857586Sbostic 					new_bufp =
34957586Sbostic 					    __add_ovflpage(hashp, new_bufp);
35050997Sbostic 					if (!new_bufp)
35150997Sbostic 						return (-1);
35250997Sbostic 					np = (u_short *)new_bufp->page;
35350997Sbostic 					putpair((char *)np, &key, &val);
35450997Sbostic 				}
35550997Sbostic 				new_bufp->flags |= BUF_MOD;
35650997Sbostic 			}
35746372Sbostic 		}
35846372Sbostic 	}
35950997Sbostic 	if (last_bfp)
36057586Sbostic 		__free_ovflpage(hashp, last_bfp);
36150997Sbostic 	return (0);
36250997Sbostic }
36346372Sbostic 
36446372Sbostic /*
36550997Sbostic  * Add the given pair to the page
36650997Sbostic  *
36750997Sbostic  * Returns:
36850997Sbostic  *	0 ==> OK
36950997Sbostic  *	1 ==> failure
37050997Sbostic  */
37146372Sbostic extern int
37257586Sbostic __addel(hashp, bufp, key, val)
37357586Sbostic 	HTAB *hashp;
37450997Sbostic 	BUFHEAD *bufp;
37550997Sbostic 	const DBT *key, *val;
37646372Sbostic {
37750997Sbostic 	register u_short *bp, *sop;
37850997Sbostic 	int do_expand;
37946372Sbostic 
38050997Sbostic 	bp = (u_short *)bufp->page;
38150997Sbostic 	do_expand = 0;
38264445Sbostic 	while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
38350997Sbostic 		/* Exception case */
38464445Sbostic 		if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
38564445Sbostic 			/* This is the last page of a big key/data pair
38664445Sbostic 			   and we need to add another page */
38764445Sbostic 			break;
38864445Sbostic 		else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
38964445Sbostic 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
39050997Sbostic 			if (!bufp)
39150997Sbostic 				return (-1);
39250997Sbostic 			bp = (u_short *)bufp->page;
39350997Sbostic 		} else
39450997Sbostic 			/* Try to squeeze key on this page */
39550997Sbostic 			if (FREESPACE(bp) > PAIRSIZE(key, val)) {
39650997Sbostic 				squeeze_key(bp, key, val);
39750997Sbostic 				return (0);
39850997Sbostic 			} else {
39957586Sbostic 				bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
40050997Sbostic 				if (!bufp)
40150997Sbostic 					return (-1);
40250997Sbostic 				bp = (u_short *)bufp->page;
40350997Sbostic 			}
40446372Sbostic 
40550997Sbostic 	if (PAIRFITS(bp, key, val))
40650997Sbostic 		putpair(bufp->page, key, val);
40750997Sbostic 	else {
40850997Sbostic 		do_expand = 1;
40957586Sbostic 		bufp = __add_ovflpage(hashp, bufp);
41050997Sbostic 		if (!bufp)
41150997Sbostic 			return (-1);
41250997Sbostic 		sop = (u_short *)bufp->page;
41346372Sbostic 
41450997Sbostic 		if (PAIRFITS(sop, key, val))
41550997Sbostic 			putpair((char *)sop, key, val);
41650997Sbostic 		else
41757586Sbostic 			if (__big_insert(hashp, bufp, key, val))
41850997Sbostic 				return (-1);
41946372Sbostic 	}
42050997Sbostic 	bufp->flags |= BUF_MOD;
42150997Sbostic 	/*
42250997Sbostic 	 * If the average number of keys per bucket exceeds the fill factor,
42350997Sbostic 	 * expand the table.
42450997Sbostic 	 */
42550997Sbostic 	hashp->NKEYS++;
42650997Sbostic 	if (do_expand ||
42750997Sbostic 	    (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
42857586Sbostic 		return (__expand_table(hashp));
42950997Sbostic 	return (0);
43046372Sbostic }
43146372Sbostic 
43246372Sbostic /*
43350997Sbostic  *
43450997Sbostic  * Returns:
43550997Sbostic  *	pointer on success
43650997Sbostic  *	NULL on error
43750997Sbostic  */
43846372Sbostic extern BUFHEAD *
43957586Sbostic __add_ovflpage(hashp, bufp)
44057586Sbostic 	HTAB *hashp;
44150997Sbostic 	BUFHEAD *bufp;
44246372Sbostic {
44350997Sbostic 	register u_short *sp;
44450997Sbostic 	u_short ndx, ovfl_num;
44546372Sbostic #ifdef DEBUG1
44650997Sbostic 	int tmp1, tmp2;
44746372Sbostic #endif
44850997Sbostic 	sp = (u_short *)bufp->page;
44960246Sbostic 
45060246Sbostic 	/* Check if we are dynamically determining the fill factor */
45160246Sbostic 	if (hashp->FFACTOR == DEF_FFACTOR) {
45260246Sbostic 		hashp->FFACTOR = sp[0] >> 1;
45360246Sbostic 		if (hashp->FFACTOR < MIN_FFACTOR)
45460246Sbostic 			hashp->FFACTOR = MIN_FFACTOR;
45560246Sbostic 	}
45650997Sbostic 	bufp->flags |= BUF_MOD;
45757586Sbostic 	ovfl_num = overflow_page(hashp);
45846372Sbostic #ifdef DEBUG1
45950997Sbostic 	tmp1 = bufp->addr;
46050997Sbostic 	tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
46146372Sbostic #endif
46257586Sbostic 	if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
46350997Sbostic 		return (NULL);
46450997Sbostic 	bufp->ovfl->flags |= BUF_MOD;
46546372Sbostic #ifdef DEBUG1
46650997Sbostic 	(void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
46750997Sbostic 	    tmp1, tmp2, bufp->ovfl->addr);
46846372Sbostic #endif
46950997Sbostic 	ndx = sp[0];
47050997Sbostic 	/*
47150997Sbostic 	 * Since a pair is allocated on a page only if there's room to add
47250997Sbostic 	 * an overflow page, we know that the OVFL information will fit on
47350997Sbostic 	 * the page.
47450997Sbostic 	 */
47550997Sbostic 	sp[ndx + 4] = OFFSET(sp);
47650997Sbostic 	sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
47750997Sbostic 	sp[ndx + 1] = ovfl_num;
47850997Sbostic 	sp[ndx + 2] = OVFLPAGE;
47950997Sbostic 	sp[0] = ndx + 2;
48046372Sbostic #ifdef HASH_STATISTICS
48150997Sbostic 	hash_overflows++;
48246372Sbostic #endif
48350997Sbostic 	return (bufp->ovfl);
48446372Sbostic }
48546372Sbostic 
48646372Sbostic /*
48750997Sbostic  * Returns:
48850997Sbostic  *	 0 indicates SUCCESS
48950997Sbostic  *	-1 indicates FAILURE
49050997Sbostic  */
49150997Sbostic extern int
49257586Sbostic __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap)
49357586Sbostic 	HTAB *hashp;
49450997Sbostic 	char *p;
49550997Sbostic 	u_int bucket;
49650997Sbostic 	int is_bucket, is_disk, is_bitmap;
49746372Sbostic {
49850997Sbostic 	register int fd, page, size;
49950997Sbostic 	int rsize;
50050997Sbostic 	u_short *bp;
50146372Sbostic 
50250997Sbostic 	fd = hashp->fp;
50350997Sbostic 	size = hashp->BSIZE;
50446372Sbostic 
50550997Sbostic 	if ((fd == -1) || !is_disk) {
50650997Sbostic 		PAGE_INIT(p);
50750997Sbostic 		return (0);
50850997Sbostic 	}
50950997Sbostic 	if (is_bucket)
51050997Sbostic 		page = BUCKET_TO_PAGE(bucket);
51150997Sbostic 	else
51250997Sbostic 		page = OADDR_TO_PAGE(bucket);
51355314Sbostic 	if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
51450997Sbostic 	    ((rsize = read(fd, p, size)) == -1))
51550997Sbostic 		return (-1);
51650997Sbostic 	bp = (u_short *)p;
51750997Sbostic 	if (!rsize)
51850997Sbostic 		bp[0] = 0;	/* We hit the EOF, so initialize a new page */
51950997Sbostic 	else
52050997Sbostic 		if (rsize != size) {
52150997Sbostic 			errno = EFTYPE;
52250997Sbostic 			return (-1);
52350997Sbostic 		}
52453403Sbostic 	if (!is_bitmap && !bp[0]) {
52550997Sbostic 		PAGE_INIT(p);
52650997Sbostic 	} else
52750997Sbostic 		if (hashp->LORDER != BYTE_ORDER) {
52850997Sbostic 			register int i, max;
52946372Sbostic 
53050997Sbostic 			if (is_bitmap) {
53150997Sbostic 				max = hashp->BSIZE >> 2; /* divide by 4 */
53250997Sbostic 				for (i = 0; i < max; i++)
53366193Sbostic 					M_32_SWAP(((long *)p)[i]);
53450997Sbostic 			} else {
53566193Sbostic 				M_16_SWAP(bp[0]);
53650997Sbostic 				max = bp[0] + 2;
53750997Sbostic 				for (i = 1; i <= max; i++)
53866193Sbostic 					M_16_SWAP(bp[i]);
53950997Sbostic 			}
54050997Sbostic 		}
54150997Sbostic 	return (0);
54246372Sbostic }
54346372Sbostic 
54450997Sbostic /*
54550997Sbostic  * Write page p to disk
54650997Sbostic  *
54750997Sbostic  * Returns:
54850997Sbostic  *	 0 ==> OK
54950997Sbostic  *	-1 ==>failure
55050997Sbostic  */
55146372Sbostic extern int
55257586Sbostic __put_page(hashp, p, bucket, is_bucket, is_bitmap)
55357586Sbostic 	HTAB *hashp;
55450997Sbostic 	char *p;
55550997Sbostic 	u_int bucket;
55650997Sbostic 	int is_bucket, is_bitmap;
55746372Sbostic {
55850997Sbostic 	register int fd, page, size;
55950997Sbostic 	int wsize;
56046372Sbostic 
56150997Sbostic 	size = hashp->BSIZE;
56257586Sbostic 	if ((hashp->fp == -1) && open_temp(hashp))
56351073Sbostic 		return (-1);
56450997Sbostic 	fd = hashp->fp;
56546372Sbostic 
56650997Sbostic 	if (hashp->LORDER != BYTE_ORDER) {
56750997Sbostic 		register int i;
56850997Sbostic 		register int max;
56946372Sbostic 
57050997Sbostic 		if (is_bitmap) {
57150997Sbostic 			max = hashp->BSIZE >> 2;	/* divide by 4 */
57250997Sbostic 			for (i = 0; i < max; i++)
57366193Sbostic 				M_32_SWAP(((long *)p)[i]);
57450997Sbostic 		} else {
57550997Sbostic 			max = ((u_short *)p)[0] + 2;
57650997Sbostic 			for (i = 0; i <= max; i++)
57766193Sbostic 				M_16_SWAP(((u_short *)p)[i]);
57850997Sbostic 		}
57946372Sbostic 	}
58050997Sbostic 	if (is_bucket)
58150997Sbostic 		page = BUCKET_TO_PAGE(bucket);
58250997Sbostic 	else
58350997Sbostic 		page = OADDR_TO_PAGE(bucket);
58455314Sbostic 	if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
58550997Sbostic 	    ((wsize = write(fd, p, size)) == -1))
58650997Sbostic 		/* Errno is set */
58750997Sbostic 		return (-1);
58850997Sbostic 	if (wsize != size) {
58950997Sbostic 		errno = EFTYPE;
59050997Sbostic 		return (-1);
59150997Sbostic 	}
59250997Sbostic 	return (0);
59346372Sbostic }
59450997Sbostic 
59546372Sbostic #define BYTE_MASK	((1 << INT_BYTE_SHIFT) -1)
59646372Sbostic /*
59750997Sbostic  * Initialize a new bitmap page.  Bitmap pages are left in memory
59850997Sbostic  * once they are read in.
59950997Sbostic  */
60051055Sbostic extern int
60157586Sbostic __init_bitmap(hashp, pnum, nbits, ndx)
60257586Sbostic 	HTAB *hashp;
60350997Sbostic 	int pnum, nbits, ndx;
60446372Sbostic {
60550997Sbostic 	u_long *ip;
60650997Sbostic 	int clearbytes, clearints;
60746372Sbostic 
608*66215Sbostic 	if ((ip = (u_long *)malloc(hashp->BSIZE)) == NULL)
60951055Sbostic 		return (1);
61050997Sbostic 	hashp->nmaps++;
61150997Sbostic 	clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
61250997Sbostic 	clearbytes = clearints << INT_TO_BYTE;
61351055Sbostic 	(void)memset((char *)ip, 0, clearbytes);
61451055Sbostic 	(void)memset(((char *)ip) + clearbytes, 0xFF,
61550997Sbostic 	    hashp->BSIZE - clearbytes);
61650997Sbostic 	ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
61750997Sbostic 	SETBIT(ip, 0);
61850997Sbostic 	hashp->BITMAPS[ndx] = (u_short)pnum;
61950997Sbostic 	hashp->mapp[ndx] = ip;
62051055Sbostic 	return (0);
62146372Sbostic }
62250997Sbostic 
62351055Sbostic static u_long
first_free(map)62450997Sbostic first_free(map)
62550997Sbostic 	u_long map;
62646372Sbostic {
62750997Sbostic 	register u_long i, mask;
62846372Sbostic 
62950997Sbostic 	mask = 0x1;
63050997Sbostic 	for (i = 0; i < BITS_PER_MAP; i++) {
63150997Sbostic 		if (!(mask & map))
63250997Sbostic 			return (i);
63350997Sbostic 		mask = mask << 1;
63450997Sbostic 	}
63550997Sbostic 	return (i);
63646372Sbostic }
63746372Sbostic 
63857586Sbostic static u_short
overflow_page(hashp)63957586Sbostic overflow_page(hashp)
64057586Sbostic 	HTAB *hashp;
64146372Sbostic {
64250997Sbostic 	register u_long *freep;
64350997Sbostic 	register int max_free, offset, splitnum;
64450997Sbostic 	u_short addr;
64551061Sbostic 	int bit, first_page, free_bit, free_page, i, in_use_bits, j;
64646372Sbostic #ifdef DEBUG2
64750997Sbostic 	int tmp1, tmp2;
64846372Sbostic #endif
64951061Sbostic 	splitnum = hashp->OVFL_POINT;
65050997Sbostic 	max_free = hashp->SPARES[splitnum];
65146372Sbostic 
65250997Sbostic 	free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
65350997Sbostic 	free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
65446372Sbostic 
65550997Sbostic 	/* Look through all the free maps to find the first free block */
65651061Sbostic 	first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
65751061Sbostic 	for ( i = first_page; i <= free_page; i++ ) {
65850997Sbostic 		if (!(freep = (u_long *)hashp->mapp[i]) &&
65957586Sbostic 		    !(freep = fetch_bitmap(hashp, i)))
66050997Sbostic 			return (NULL);
66150997Sbostic 		if (i == free_page)
66250997Sbostic 			in_use_bits = free_bit;
66350997Sbostic 		else
66450997Sbostic 			in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
66551061Sbostic 
66651061Sbostic 		if (i == first_page) {
66751061Sbostic 			bit = hashp->LAST_FREED &
66851061Sbostic 			    ((hashp->BSIZE << BYTE_SHIFT) - 1);
66951061Sbostic 			j = bit / BITS_PER_MAP;
67051061Sbostic 			bit = bit & ~(BITS_PER_MAP - 1);
67151061Sbostic 		} else {
67251061Sbostic 			bit = 0;
67351061Sbostic 			j = 0;
67451061Sbostic 		}
67551061Sbostic 		for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
67650997Sbostic 			if (freep[j] != ALL_SET)
67750997Sbostic 				goto found;
67847250Sbostic 	}
67946372Sbostic 
68050997Sbostic 	/* No Free Page Found */
68151061Sbostic 	hashp->LAST_FREED = hashp->SPARES[splitnum];
68250997Sbostic 	hashp->SPARES[splitnum]++;
68350997Sbostic 	offset = hashp->SPARES[splitnum] -
68450997Sbostic 	    (splitnum ? hashp->SPARES[splitnum - 1] : 0);
68546372Sbostic 
68651061Sbostic #define	OVMSG	"HASH: Out of overflow pages.  Increase page size\n"
68751061Sbostic 	if (offset > SPLITMASK) {
68851061Sbostic 		if (++splitnum >= NCACHED) {
68951061Sbostic 			(void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
69051061Sbostic 			return (NULL);
69151061Sbostic 		}
69251061Sbostic 		hashp->OVFL_POINT = splitnum;
69351061Sbostic 		hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
69451061Sbostic 		hashp->SPARES[splitnum-1]--;
69562488Sbostic 		offset = 1;
69651061Sbostic 	}
69751061Sbostic 
69850997Sbostic 	/* Check if we need to allocate a new bitmap page */
69950997Sbostic 	if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
70050997Sbostic 		free_page++;
70150997Sbostic 		if (free_page >= NCACHED) {
70250997Sbostic 			(void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
70350997Sbostic 			return (NULL);
70450997Sbostic 		}
70550997Sbostic 		/*
70650997Sbostic 		 * This is tricky.  The 1 indicates that you want the new page
70750997Sbostic 		 * allocated with 1 clear bit.  Actually, you are going to
70850997Sbostic 		 * allocate 2 pages from this map.  The first is going to be
70950997Sbostic 		 * the map page, the second is the overflow page we were
71050997Sbostic 		 * looking for.  The init_bitmap routine automatically, sets
71150997Sbostic 		 * the first bit of itself to indicate that the bitmap itself
71250997Sbostic 		 * is in use.  We would explicitly set the second bit, but
71350997Sbostic 		 * don't have to if we tell init_bitmap not to leave it clear
71450997Sbostic 		 * in the first place.
71550997Sbostic 		 */
71657586Sbostic 		if (__init_bitmap(hashp, (int)OADDR_OF(splitnum, offset),
71751055Sbostic 		    1, free_page))
71851055Sbostic 			return (NULL);
71950997Sbostic 		hashp->SPARES[splitnum]++;
72046372Sbostic #ifdef DEBUG2
72150997Sbostic 		free_bit = 2;
72246372Sbostic #endif
72350997Sbostic 		offset++;
72451061Sbostic 		if (offset > SPLITMASK) {
72551061Sbostic 			if (++splitnum >= NCACHED) {
72651061Sbostic 				(void)write(STDERR_FILENO, OVMSG,
72751061Sbostic 				    sizeof(OVMSG) - 1);
72851061Sbostic 				return (NULL);
72951061Sbostic 			}
73051061Sbostic 			hashp->OVFL_POINT = splitnum;
73151061Sbostic 			hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
73251061Sbostic 			hashp->SPARES[splitnum-1]--;
73351061Sbostic 			offset = 0;
73451061Sbostic 		}
73550997Sbostic 	} else {
73650997Sbostic 		/*
73750997Sbostic 		 * Free_bit addresses the last used bit.  Bump it to address
73850997Sbostic 		 * the first available bit.
73950997Sbostic 		 */
74050997Sbostic 		free_bit++;
74150997Sbostic 		SETBIT(freep, free_bit);
74250997Sbostic 	}
74346372Sbostic 
74450997Sbostic 	/* Calculate address of the new overflow page */
74550997Sbostic 	addr = OADDR_OF(splitnum, offset);
74646372Sbostic #ifdef DEBUG2
74750997Sbostic 	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
74850997Sbostic 	    addr, free_bit, free_page);
74946372Sbostic #endif
75050997Sbostic 	return (addr);
75146372Sbostic 
75246372Sbostic found:
75350997Sbostic 	bit = bit + first_free(freep[j]);
75450997Sbostic 	SETBIT(freep, bit);
75546372Sbostic #ifdef DEBUG2
75650997Sbostic 	tmp1 = bit;
75750997Sbostic 	tmp2 = i;
75846372Sbostic #endif
75950997Sbostic 	/*
76050997Sbostic 	 * Bits are addressed starting with 0, but overflow pages are addressed
76150997Sbostic 	 * beginning at 1. Bit is a bit addressnumber, so we need to increment
76250997Sbostic 	 * it to convert it to a page number.
76350997Sbostic 	 */
76450997Sbostic 	bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
76551061Sbostic 	if (bit >= hashp->LAST_FREED)
76651061Sbostic 		hashp->LAST_FREED = bit - 1;
76746372Sbostic 
76850997Sbostic 	/* Calculate the split number for this page */
76950997Sbostic 	for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
77050997Sbostic 	offset = (i ? bit - hashp->SPARES[i - 1] : bit);
77150997Sbostic 	if (offset >= SPLITMASK)
77250997Sbostic 		return (NULL);	/* Out of overflow pages */
77350997Sbostic 	addr = OADDR_OF(i, offset);
77446372Sbostic #ifdef DEBUG2
77550997Sbostic 	(void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
77650997Sbostic 	    addr, tmp1, tmp2);
77746372Sbostic #endif
77846372Sbostic 
77950997Sbostic 	/* Allocate and return the overflow page */
78050997Sbostic 	return (addr);
78146372Sbostic }
78246372Sbostic 
78346372Sbostic /*
78450997Sbostic  * Mark this overflow page as free.
78550997Sbostic  */
78650997Sbostic extern void
78757586Sbostic __free_ovflpage(hashp, obufp)
78857586Sbostic 	HTAB *hashp;
78950997Sbostic 	BUFHEAD *obufp;
79046372Sbostic {
79150997Sbostic 	register u_short addr;
79250997Sbostic 	u_long *freep;
79350997Sbostic 	int bit_address, free_page, free_bit;
79450997Sbostic 	u_short ndx;
79546372Sbostic 
79650997Sbostic 	addr = obufp->addr;
79746372Sbostic #ifdef DEBUG1
79850997Sbostic 	(void)fprintf(stderr, "Freeing %d\n", addr);
79946372Sbostic #endif
80050997Sbostic 	ndx = (((u_short)addr) >> SPLITSHIFT);
80150997Sbostic 	bit_address =
80250997Sbostic 	    (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
80351061Sbostic 	 if (bit_address < hashp->LAST_FREED)
80451061Sbostic 		hashp->LAST_FREED = bit_address;
80550997Sbostic 	free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
80650997Sbostic 	free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
80746372Sbostic 
80850997Sbostic 	if (!(freep = hashp->mapp[free_page]))
80957586Sbostic 		freep = fetch_bitmap(hashp, free_page);
81050997Sbostic #ifdef DEBUG
81150997Sbostic 	/*
81250997Sbostic 	 * This had better never happen.  It means we tried to read a bitmap
81350997Sbostic 	 * that has already had overflow pages allocated off it, and we
81450997Sbostic 	 * failed to read it from the file.
81550997Sbostic 	 */
81650997Sbostic 	if (!freep)
81750997Sbostic 		assert(0);
81850997Sbostic #endif
81950997Sbostic 	CLRBIT(freep, free_bit);
82046372Sbostic #ifdef DEBUG2
82150997Sbostic 	(void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
82250997Sbostic 	    obufp->addr, free_bit, free_page);
82346372Sbostic #endif
82457586Sbostic 	__reclaim_buf(hashp, obufp);
82546372Sbostic }
82646372Sbostic 
82746372Sbostic /*
82850997Sbostic  * Returns:
82950997Sbostic  *	 0 success
83050997Sbostic  *	-1 failure
83150997Sbostic  */
83246372Sbostic static int
open_temp(hashp)83357586Sbostic open_temp(hashp)
83457586Sbostic 	HTAB *hashp;
83546372Sbostic {
83650997Sbostic 	sigset_t set, oset;
83750997Sbostic 	static char namestr[] = "_hashXXXXXX";
83846372Sbostic 
83950997Sbostic 	/* Block signals; make sure file goes away at process exit. */
84055315Sbostic 	(void)sigfillset(&set);
84150997Sbostic 	(void)sigprocmask(SIG_BLOCK, &set, &oset);
84250997Sbostic 	if ((hashp->fp = mkstemp(namestr)) != -1) {
84350997Sbostic 		(void)unlink(namestr);
84450997Sbostic 		(void)fcntl(hashp->fp, F_SETFD, 1);
84550997Sbostic 	}
84650997Sbostic 	(void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
84750997Sbostic 	return (hashp->fp != -1 ? 0 : -1);
84846372Sbostic }
84946372Sbostic 
85050997Sbostic /*
85150997Sbostic  * We have to know that the key will fit, but the last entry on the page is
85250997Sbostic  * an overflow pair, so we need to shift things.
85350997Sbostic  */
85446372Sbostic static void
squeeze_key(sp,key,val)85550997Sbostic squeeze_key(sp, key, val)
85650997Sbostic 	u_short *sp;
85750997Sbostic 	const DBT *key, *val;
85846372Sbostic {
85950997Sbostic 	register char *p;
86050997Sbostic 	u_short free_space, n, off, pageno;
86146372Sbostic 
86250997Sbostic 	p = (char *)sp;
86350997Sbostic 	n = sp[0];
86450997Sbostic 	free_space = FREESPACE(sp);
86550997Sbostic 	off = OFFSET(sp);
86646372Sbostic 
86750997Sbostic 	pageno = sp[n - 1];
86850997Sbostic 	off -= key->size;
86950997Sbostic 	sp[n - 1] = off;
87058016Sbostic 	memmove(p + off, key->data, key->size);
87150997Sbostic 	off -= val->size;
87250997Sbostic 	sp[n] = off;
87358016Sbostic 	memmove(p + off, val->data, val->size);
87450997Sbostic 	sp[0] = n + 2;
87550997Sbostic 	sp[n + 1] = pageno;
87650997Sbostic 	sp[n + 2] = OVFLPAGE;
87750997Sbostic 	FREESPACE(sp) = free_space - PAIRSIZE(key, val);
87850997Sbostic 	OFFSET(sp) = off;
87946372Sbostic }
88046372Sbostic 
88147250Sbostic static u_long *
fetch_bitmap(hashp,ndx)88257586Sbostic fetch_bitmap(hashp, ndx)
88357586Sbostic 	HTAB *hashp;
88450997Sbostic 	int ndx;
88547250Sbostic {
886*66215Sbostic 	if (ndx >= hashp->nmaps)
88750997Sbostic 		return (NULL);
888*66215Sbostic 	if ((hashp->mapp[ndx] = (u_long *)malloc(hashp->BSIZE)) == NULL)
889*66215Sbostic 		return (NULL);
890*66215Sbostic 	if (__get_page(hashp,
891*66215Sbostic 	    (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
892*66215Sbostic 		free(hashp->mapp[ndx]);
893*66215Sbostic 		return (NULL);
894*66215Sbostic 	}
89550997Sbostic 	return (hashp->mapp[ndx]);
89650997Sbostic }
89747250Sbostic 
89846372Sbostic #ifdef DEBUG4
89950997Sbostic int
print_chain(addr)90050997Sbostic print_chain(addr)
90150997Sbostic 	int addr;
90246372Sbostic {
90350997Sbostic 	BUFHEAD *bufp;
90450997Sbostic 	short *bp, oaddr;
90546372Sbostic 
90650997Sbostic 	(void)fprintf(stderr, "%d ", addr);
90757586Sbostic 	bufp = __get_buf(hashp, addr, NULL, 0);
90846372Sbostic 	bp = (short *)bufp->page;
90950997Sbostic 	while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
91050997Sbostic 		((bp[0] > 2) && bp[2] < REAL_KEY))) {
91150997Sbostic 		oaddr = bp[bp[0] - 1];
91250997Sbostic 		(void)fprintf(stderr, "%d ", (int)oaddr);
91357586Sbostic 		bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
91446372Sbostic 		bp = (short *)bufp->page;
91546372Sbostic 	}
91650997Sbostic 	(void)fprintf(stderr, "\n");
91746372Sbostic }
91846372Sbostic #endif
919