xref: /original-bsd/lib/libc/db/hash/hash_bigkey.c (revision b601de3b)
1fb91d121Sbostic /*-
2c0ca25c5Sbostic  * Copyright (c) 1990, 1993
3c0ca25c5Sbostic  *	The Regents of the University of California.  All rights reserved.
4fb91d121Sbostic  *
5fb91d121Sbostic  * This code is derived from software contributed to Berkeley by
6fb91d121Sbostic  * Margo Seltzer.
7fb91d121Sbostic  *
8fb91d121Sbostic  * %sccs.include.redist.c%
9fb91d121Sbostic  */
10fb91d121Sbostic 
11fb91d121Sbostic #if defined(LIBC_SCCS) && !defined(lint)
12*b601de3bSbostic static char sccsid[] = "@(#)hash_bigkey.c	8.2 (Berkeley) 02/21/94";
13fb91d121Sbostic #endif /* LIBC_SCCS and not lint */
14fb91d121Sbostic 
15c62a4a66Sbostic /*
16c62a4a66Sbostic  * PACKAGE: hash
17c62a4a66Sbostic  * DESCRIPTION:
18c62a4a66Sbostic  *	Big key/data handling for the hashing package.
19c62a4a66Sbostic  *
20c62a4a66Sbostic  * ROUTINES:
21c62a4a66Sbostic  * External
22c62a4a66Sbostic  *	__big_keydata
23c62a4a66Sbostic  *	__big_split
24c62a4a66Sbostic  *	__big_insert
25c62a4a66Sbostic  *	__big_return
26c62a4a66Sbostic  *	__big_delete
27c62a4a66Sbostic  *	__find_last_page
28c62a4a66Sbostic  * Internal
29c62a4a66Sbostic  *	collect_key
30c62a4a66Sbostic  *	collect_data
31c62a4a66Sbostic  */
32fb91d121Sbostic 
3343ccde4eSbostic #include <sys/param.h>
349d0f053fSbostic 
359d0f053fSbostic #include <errno.h>
36fb91d121Sbostic #include <stdio.h>
3787570ce5Sbostic #include <stdlib.h>
3887570ce5Sbostic #include <string.h>
399d0f053fSbostic 
40c62a4a66Sbostic #ifdef DEBUG
41c62a4a66Sbostic #include <assert.h>
42c62a4a66Sbostic #endif
439d0f053fSbostic 
44cbf11d4dSbostic #include <db.h>
45fb91d121Sbostic #include "hash.h"
46fb91d121Sbostic #include "page.h"
47c62a4a66Sbostic #include "extern.h"
48fb91d121Sbostic 
499d0f053fSbostic static int collect_key __P((HTAB *, BUFHEAD *, int, DBT *, int));
509d0f053fSbostic static int collect_data __P((HTAB *, BUFHEAD *, int, int));
51fb91d121Sbostic 
52fb91d121Sbostic /*
53c62a4a66Sbostic  * Big_insert
54c62a4a66Sbostic  *
55c62a4a66Sbostic  * You need to do an insert and the key/data pair is too big
56c62a4a66Sbostic  *
57c62a4a66Sbostic  * Returns:
58c62a4a66Sbostic  * 0 ==> OK
59c62a4a66Sbostic  *-1 ==> ERROR
60fb91d121Sbostic  */
61fb91d121Sbostic extern int
629d0f053fSbostic __big_insert(hashp, bufp, key, val)
639d0f053fSbostic 	HTAB *hashp;
64fb91d121Sbostic 	BUFHEAD *bufp;
65c62a4a66Sbostic 	const DBT *key, *val;
66fb91d121Sbostic {
67c62a4a66Sbostic 	register u_short *p;
68c62a4a66Sbostic 	int key_size, n, val_size;
69fb91d121Sbostic 	u_short space, move_bytes, off;
70c62a4a66Sbostic 	char *cp, *key_data, *val_data;
71c62a4a66Sbostic 
72c62a4a66Sbostic 	cp = bufp->page;		/* Character pointer of p. */
73c62a4a66Sbostic 	p = (u_short *)cp;
74fb91d121Sbostic 
750a9054daSbostic 	key_data = (char *)key->data;
76fb91d121Sbostic 	key_size = key->size;
770a9054daSbostic 	val_data = (char *)val->data;
78fb91d121Sbostic 	val_size = val->size;
79fb91d121Sbostic 
80fb91d121Sbostic 	/* First move the Key */
81c62a4a66Sbostic 	for (space = FREESPACE(p) - BIGOVERHEAD; key_size;
82fb91d121Sbostic 	    space = FREESPACE(p) - BIGOVERHEAD) {
83fb91d121Sbostic 		move_bytes = MIN(space, key_size);
84fb91d121Sbostic 		off = OFFSET(p) - move_bytes;
8589be7141Sbostic 		memmove(cp + off, key_data, move_bytes);
86fb91d121Sbostic 		key_size -= move_bytes;
87fb91d121Sbostic 		key_data += move_bytes;
88fb91d121Sbostic 		n = p[0];
89fb91d121Sbostic 		p[++n] = off;
90fb91d121Sbostic 		p[0] = ++n;
91fb91d121Sbostic 		FREESPACE(p) = off - PAGE_META(n);
92fb91d121Sbostic 		OFFSET(p) = off;
93fb91d121Sbostic 		p[n] = PARTIAL_KEY;
949d0f053fSbostic 		bufp = __add_ovflpage(hashp, bufp);
95c62a4a66Sbostic 		if (!bufp)
96fb91d121Sbostic 			return (-1);
97fb91d121Sbostic 		n = p[0];
98c62a4a66Sbostic 		if (!key_size)
99fb91d121Sbostic 			if (FREESPACE(p)) {
100fb91d121Sbostic 				move_bytes = MIN(FREESPACE(p), val_size);
101fb91d121Sbostic 				off = OFFSET(p) - move_bytes;
102fb91d121Sbostic 				p[n] = off;
10389be7141Sbostic 				memmove(cp + off, val_data, move_bytes);
104fb91d121Sbostic 				val_data += move_bytes;
105fb91d121Sbostic 				val_size -= move_bytes;
106fb91d121Sbostic 				p[n - 2] = FULL_KEY_DATA;
107fb91d121Sbostic 				FREESPACE(p) = FREESPACE(p) - move_bytes;
108fb91d121Sbostic 				OFFSET(p) = off;
109c62a4a66Sbostic 			} else
110c62a4a66Sbostic 				p[n - 2] = FULL_KEY;
111fb91d121Sbostic 		p = (u_short *)bufp->page;
112fb91d121Sbostic 		cp = bufp->page;
113fb91d121Sbostic 		bufp->flags |= BUF_MOD;
114fb91d121Sbostic 	}
115fb91d121Sbostic 
116fb91d121Sbostic 	/* Now move the data */
117c62a4a66Sbostic 	for (space = FREESPACE(p) - BIGOVERHEAD; val_size;
118fb91d121Sbostic 	    space = FREESPACE(p) - BIGOVERHEAD) {
119fb91d121Sbostic 		move_bytes = MIN(space, val_size);
120fb91d121Sbostic 		/*
121c62a4a66Sbostic 		 * Here's the hack to make sure that if the data ends on the
122c62a4a66Sbostic 		 * same page as the key ends, FREESPACE is at least one.
123fb91d121Sbostic 		 */
124c62a4a66Sbostic 		if (space == val_size && val_size == val->size)
125fb91d121Sbostic 			move_bytes--;
126fb91d121Sbostic 		off = OFFSET(p) - move_bytes;
12789be7141Sbostic 		memmove(cp + off, val_data, move_bytes);
128fb91d121Sbostic 		val_size -= move_bytes;
129fb91d121Sbostic 		val_data += move_bytes;
130fb91d121Sbostic 		n = p[0];
131fb91d121Sbostic 		p[++n] = off;
132fb91d121Sbostic 		p[0] = ++n;
133fb91d121Sbostic 		FREESPACE(p) = off - PAGE_META(n);
134fb91d121Sbostic 		OFFSET(p) = off;
135fb91d121Sbostic 		if (val_size) {
136fb91d121Sbostic 			p[n] = FULL_KEY;
1379d0f053fSbostic 			bufp = __add_ovflpage(hashp, bufp);
138c62a4a66Sbostic 			if (!bufp)
139fb91d121Sbostic 				return (-1);
140fb91d121Sbostic 			cp = bufp->page;
141fb91d121Sbostic 			p = (u_short *)cp;
142c62a4a66Sbostic 		} else
143fb91d121Sbostic 			p[n] = FULL_KEY_DATA;
144fb91d121Sbostic 		bufp->flags |= BUF_MOD;
145fb91d121Sbostic 	}
146fb91d121Sbostic 	return (0);
147fb91d121Sbostic }
148fb91d121Sbostic 
149fb91d121Sbostic /*
150c62a4a66Sbostic  * Called when bufp's page  contains a partial key (index should be 1)
151c62a4a66Sbostic  *
152c62a4a66Sbostic  * All pages in the big key/data pair except bufp are freed.  We cannot
153c62a4a66Sbostic  * free bufp because the page pointing to it is lost and we can't get rid
154c62a4a66Sbostic  * of its pointer.
155c62a4a66Sbostic  *
156c62a4a66Sbostic  * Returns:
157c62a4a66Sbostic  * 0 => OK
158c62a4a66Sbostic  *-1 => ERROR
159fb91d121Sbostic  */
160fb91d121Sbostic extern int
1619d0f053fSbostic __big_delete(hashp, bufp)
1629d0f053fSbostic 	HTAB *hashp;
163fb91d121Sbostic 	BUFHEAD *bufp;
164fb91d121Sbostic {
165c62a4a66Sbostic 	register BUFHEAD *last_bfp, *rbufp;
166c62a4a66Sbostic 	u_short *bp, pageno;
167c62a4a66Sbostic 	int key_done, n;
168c62a4a66Sbostic 
169c62a4a66Sbostic 	rbufp = bufp;
170c62a4a66Sbostic 	last_bfp = NULL;
171c62a4a66Sbostic 	bp = (u_short *)bufp->page;
172c62a4a66Sbostic 	pageno = 0;
173c62a4a66Sbostic 	key_done = 0;
174fb91d121Sbostic 
175fb91d121Sbostic 	while (!key_done || (bp[2] != FULL_KEY_DATA)) {
176c62a4a66Sbostic 		if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA)
177c62a4a66Sbostic 			key_done = 1;
178fb91d121Sbostic 
179fb91d121Sbostic 		/*
180c62a4a66Sbostic 		 * If there is freespace left on a FULL_KEY_DATA page, then
181c62a4a66Sbostic 		 * the data is short and fits entirely on this page, and this
182c62a4a66Sbostic 		 * is the last page.
183fb91d121Sbostic 		 */
184c62a4a66Sbostic 		if (bp[2] == FULL_KEY_DATA && FREESPACE(bp))
185c62a4a66Sbostic 			break;
186fb91d121Sbostic 		pageno = bp[bp[0] - 1];
187fb91d121Sbostic 		rbufp->flags |= BUF_MOD;
1889d0f053fSbostic 		rbufp = __get_buf(hashp, pageno, rbufp, 0);
189c62a4a66Sbostic 		if (last_bfp)
1909d0f053fSbostic 			__free_ovflpage(hashp, last_bfp);
191fb91d121Sbostic 		last_bfp = rbufp;
192c62a4a66Sbostic 		if (!rbufp)
193c62a4a66Sbostic 			return (-1);		/* Error. */
194fb91d121Sbostic 		bp = (u_short *)rbufp->page;
195fb91d121Sbostic 	}
196fb91d121Sbostic 
197fb91d121Sbostic 	/*
198c62a4a66Sbostic 	 * If we get here then rbufp points to the last page of the big
199c62a4a66Sbostic 	 * key/data pair.  Bufp points to the first one -- it should now be
200c62a4a66Sbostic 	 * empty pointing to the next page after this pair.  Can't free it
201c62a4a66Sbostic 	 * because we don't have the page pointing to it.
202fb91d121Sbostic 	 */
203fb91d121Sbostic 
204c62a4a66Sbostic 	/* This is information from the last page of the pair. */
205fb91d121Sbostic 	n = bp[0];
206fb91d121Sbostic 	pageno = bp[n - 1];
207fb91d121Sbostic 
208c62a4a66Sbostic 	/* Now, bp is the first page of the pair. */
209fb91d121Sbostic 	bp = (u_short *)bufp->page;
210fb91d121Sbostic 	if (n > 2) {
211c62a4a66Sbostic 		/* There is an overflow page. */
212fb91d121Sbostic 		bp[1] = pageno;
213fb91d121Sbostic 		bp[2] = OVFLPAGE;
214fb91d121Sbostic 		bufp->ovfl = rbufp->ovfl;
215c62a4a66Sbostic 	} else
216c62a4a66Sbostic 		/* This is the last page. */
217fb91d121Sbostic 		bufp->ovfl = NULL;
218fb91d121Sbostic 	n -= 2;
219fb91d121Sbostic 	bp[0] = n;
220fb91d121Sbostic 	FREESPACE(bp) = hashp->BSIZE - PAGE_META(n);
221fb91d121Sbostic 	OFFSET(bp) = hashp->BSIZE - 1;
222fb91d121Sbostic 
223fb91d121Sbostic 	bufp->flags |= BUF_MOD;
224c62a4a66Sbostic 	if (rbufp)
2259d0f053fSbostic 		__free_ovflpage(hashp, rbufp);
226c62a4a66Sbostic 	if (last_bfp != rbufp)
2279d0f053fSbostic 		__free_ovflpage(hashp, last_bfp);
228fb91d121Sbostic 
229fb91d121Sbostic 	hashp->NKEYS--;
230fb91d121Sbostic 	return (0);
231fb91d121Sbostic }
232fb91d121Sbostic /*
233c62a4a66Sbostic  * Returns:
234c62a4a66Sbostic  *  0 = key not found
235c62a4a66Sbostic  * -1 = get next overflow page
236c62a4a66Sbostic  * -2 means key not found and this is big key/data
237c62a4a66Sbostic  * -3 error
238fb91d121Sbostic  */
239fb91d121Sbostic extern int
2409d0f053fSbostic __find_bigpair(hashp, bufp, ndx, key, size)
2419d0f053fSbostic 	HTAB *hashp;
242fb91d121Sbostic 	BUFHEAD *bufp;
243fb91d121Sbostic 	int ndx;
244fb91d121Sbostic 	char *key;
245fb91d121Sbostic 	int size;
246fb91d121Sbostic {
247c62a4a66Sbostic 	register u_short *bp;
248c62a4a66Sbostic 	register char *p;
249c62a4a66Sbostic 	int ksize;
250fb91d121Sbostic 	u_short bytes;
251c62a4a66Sbostic 	char *kkey;
252fb91d121Sbostic 
253c62a4a66Sbostic 	bp = (u_short *)bufp->page;
254c62a4a66Sbostic 	p = bufp->page;
255c62a4a66Sbostic 	ksize = size;
256c62a4a66Sbostic 	kkey = key;
257fb91d121Sbostic 
258fb91d121Sbostic 	for (bytes = hashp->BSIZE - bp[ndx];
259fb91d121Sbostic 	    bytes <= size && bp[ndx + 1] == PARTIAL_KEY;
260fb91d121Sbostic 	    bytes = hashp->BSIZE - bp[ndx]) {
26189be7141Sbostic 		if (memcmp(p + bp[ndx], kkey, bytes))
262c62a4a66Sbostic 			return (-2);
263fb91d121Sbostic 		kkey += bytes;
264fb91d121Sbostic 		ksize -= bytes;
2659d0f053fSbostic 		bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0);
266c62a4a66Sbostic 		if (!bufp)
267fb91d121Sbostic 			return (-3);
268fb91d121Sbostic 		p = bufp->page;
269fb91d121Sbostic 		bp = (u_short *)p;
270fb91d121Sbostic 		ndx = 1;
271fb91d121Sbostic 	}
272fb91d121Sbostic 
27389be7141Sbostic 	if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) {
274fb91d121Sbostic #ifdef HASH_STATISTICS
275c62a4a66Sbostic 		++hash_collisions;
276fb91d121Sbostic #endif
277fb91d121Sbostic 		return (-2);
278c62a4a66Sbostic 	} else
279c62a4a66Sbostic 		return (ndx);
280fb91d121Sbostic }
281fb91d121Sbostic 
282fb91d121Sbostic /*
283c62a4a66Sbostic  * Given the buffer pointer of the first overflow page of a big pair,
284c62a4a66Sbostic  * find the end of the big pair
285c62a4a66Sbostic  *
286c62a4a66Sbostic  * This will set bpp to the buffer header of the last page of the big pair.
287c62a4a66Sbostic  * It will return the pageno of the overflow page following the last page
288c62a4a66Sbostic  * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
289c62a4a66Sbostic  * bucket)
290fb91d121Sbostic  */
291fb91d121Sbostic extern u_short
2929d0f053fSbostic __find_last_page(hashp, bpp)
2939d0f053fSbostic 	HTAB *hashp;
294fb91d121Sbostic 	BUFHEAD **bpp;
295fb91d121Sbostic {
296c62a4a66Sbostic 	BUFHEAD *bufp;
297c62a4a66Sbostic 	u_short *bp, pageno;
298fb91d121Sbostic 	int n;
299fb91d121Sbostic 
300c62a4a66Sbostic 	bufp = *bpp;
301c62a4a66Sbostic 	bp = (u_short *)bufp->page;
302c62a4a66Sbostic 	for (;;) {
303fb91d121Sbostic 		n = bp[0];
304fb91d121Sbostic 
305fb91d121Sbostic 		/*
306c62a4a66Sbostic 		 * This is the last page if: the tag is FULL_KEY_DATA and
307c62a4a66Sbostic 		 * either only 2 entries OVFLPAGE marker is explicit there
308c62a4a66Sbostic 		 * is freespace on the page.
309fb91d121Sbostic 		 */
310fb91d121Sbostic 		if (bp[2] == FULL_KEY_DATA &&
311c62a4a66Sbostic 		    ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp))))
312c62a4a66Sbostic 			break;
313fb91d121Sbostic 
314fb91d121Sbostic 		pageno = bp[n - 1];
3159d0f053fSbostic 		bufp = __get_buf(hashp, pageno, bufp, 0);
316c62a4a66Sbostic 		if (!bufp)
317c62a4a66Sbostic 			return (0);	/* Need to indicate an error! */
318fb91d121Sbostic 		bp = (u_short *)bufp->page;
319fb91d121Sbostic 	}
320fb91d121Sbostic 
321fb91d121Sbostic 	*bpp = bufp;
322c62a4a66Sbostic 	if (bp[0] > 2)
323c62a4a66Sbostic 		return (bp[3]);
324c62a4a66Sbostic 	else
325c62a4a66Sbostic 		return (0);
326fb91d121Sbostic }
327fb91d121Sbostic 
328fb91d121Sbostic /*
329c62a4a66Sbostic  * Return the data for the key/data pair that begins on this page at this
330c62a4a66Sbostic  * index (index should always be 1).
331fb91d121Sbostic  */
332fb91d121Sbostic extern int
3339d0f053fSbostic __big_return(hashp, bufp, ndx, val, set_current)
3349d0f053fSbostic 	HTAB *hashp;
335fb91d121Sbostic 	BUFHEAD *bufp;
336fb91d121Sbostic 	int ndx;
337fb91d121Sbostic 	DBT *val;
338fb91d121Sbostic 	int set_current;
339fb91d121Sbostic {
340fb91d121Sbostic 	BUFHEAD *save_p;
341c62a4a66Sbostic 	u_short *bp, len, off, save_addr;
342c62a4a66Sbostic 	char *tp;
343fb91d121Sbostic 
344c62a4a66Sbostic 	bp = (u_short *)bufp->page;
345fb91d121Sbostic 	while (bp[ndx + 1] == PARTIAL_KEY) {
3469d0f053fSbostic 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
347c62a4a66Sbostic 		if (!bufp)
348c62a4a66Sbostic 			return (-1);
349fb91d121Sbostic 		bp = (u_short *)bufp->page;
350fb91d121Sbostic 		ndx = 1;
351fb91d121Sbostic 	}
352fb91d121Sbostic 
353fb91d121Sbostic 	if (bp[ndx + 1] == FULL_KEY) {
3549d0f053fSbostic 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
355c62a4a66Sbostic 		if (!bufp)
356c62a4a66Sbostic 			return (-1);
357fb91d121Sbostic 		bp = (u_short *)bufp->page;
358fb91d121Sbostic 		save_p = bufp;
359fb91d121Sbostic 		save_addr = save_p->addr;
360fb91d121Sbostic 		off = bp[1];
361fb91d121Sbostic 		len = 0;
362c62a4a66Sbostic 	} else
363c62a4a66Sbostic 		if (!FREESPACE(bp)) {
364fb91d121Sbostic 			/*
365c62a4a66Sbostic 			 * This is a hack.  We can't distinguish between
366c62a4a66Sbostic 			 * FULL_KEY_DATA that contains complete data or
367c62a4a66Sbostic 			 * incomplete data, so we require that if the data
368c62a4a66Sbostic 			 * is complete, there is at least 1 byte of free
369c62a4a66Sbostic 			 * space left.
370fb91d121Sbostic 			 */
371fb91d121Sbostic 			off = bp[bp[0]];
372fb91d121Sbostic 			len = bp[1] - off;
373fb91d121Sbostic 			save_p = bufp;
374fb91d121Sbostic 			save_addr = bufp->addr;
3759d0f053fSbostic 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
376c62a4a66Sbostic 			if (!bufp)
377c62a4a66Sbostic 				return (-1);
378fb91d121Sbostic 			bp = (u_short *)bufp->page;
379fb91d121Sbostic 		} else {
380c62a4a66Sbostic 			/* The data is all on one page. */
381fb91d121Sbostic 			tp = (char *)bp;
382fb91d121Sbostic 			off = bp[bp[0]];
3830a9054daSbostic 			val->data = (u_char *)tp + off;
384fb91d121Sbostic 			val->size = bp[1] - off;
385fb91d121Sbostic 			if (set_current) {
386c62a4a66Sbostic 				if (bp[0] == 2) {	/* No more buckets in
387c62a4a66Sbostic 							 * chain */
388fb91d121Sbostic 					hashp->cpage = NULL;
389fb91d121Sbostic 					hashp->cbucket++;
390fb91d121Sbostic 					hashp->cndx = 1;
391fb91d121Sbostic 				} else {
3929d0f053fSbostic 					hashp->cpage = __get_buf(hashp,
3939d0f053fSbostic 					    bp[bp[0] - 1], bufp, 0);
394c62a4a66Sbostic 					if (!hashp->cpage)
395c62a4a66Sbostic 						return (-1);
396fb91d121Sbostic 					hashp->cndx = 1;
397c62a4a66Sbostic 					if (!((u_short *)
398c62a4a66Sbostic 					    hashp->cpage->page)[0]) {
399fb91d121Sbostic 						hashp->cbucket++;
400fb91d121Sbostic 						hashp->cpage = NULL;
401fb91d121Sbostic 					}
402fb91d121Sbostic 				}
403fb91d121Sbostic 			}
404fb91d121Sbostic 			return (0);
405fb91d121Sbostic 		}
406fb91d121Sbostic 
4079d0f053fSbostic 	val->size = collect_data(hashp, bufp, (int)len, set_current);
408c62a4a66Sbostic 	if (val->size == -1)
409fb91d121Sbostic 		return (-1);
410fb91d121Sbostic 	if (save_p->addr != save_addr) {
411c62a4a66Sbostic 		/* We are pretty short on buffers. */
412fb91d121Sbostic 		errno = EINVAL;			/* OUT OF BUFFERS */
413fb91d121Sbostic 		return (-1);
414fb91d121Sbostic 	}
41589be7141Sbostic 	memmove(hashp->tmp_buf, (save_p->page) + off, len);
4160a9054daSbostic 	val->data = (u_char *)hashp->tmp_buf;
417fb91d121Sbostic 	return (0);
418fb91d121Sbostic }
419fb91d121Sbostic /*
420c62a4a66Sbostic  * Count how big the total datasize is by recursing through the pages.  Then
421c62a4a66Sbostic  * allocate a buffer and copy the data as you recurse up.
422fb91d121Sbostic  */
423fb91d121Sbostic static int
collect_data(hashp,bufp,len,set)4249d0f053fSbostic collect_data(hashp, bufp, len, set)
4259d0f053fSbostic 	HTAB *hashp;
426fb91d121Sbostic 	BUFHEAD *bufp;
427c62a4a66Sbostic 	int len, set;
428fb91d121Sbostic {
429c62a4a66Sbostic 	register u_short *bp;
430c62a4a66Sbostic 	register char *p;
431c62a4a66Sbostic 	BUFHEAD *xbp;
432fb91d121Sbostic 	u_short save_addr;
433fb91d121Sbostic 	int mylen, totlen;
434fb91d121Sbostic 
435c62a4a66Sbostic 	p = bufp->page;
436c62a4a66Sbostic 	bp = (u_short *)p;
437fb91d121Sbostic 	mylen = hashp->BSIZE - bp[1];
438fb91d121Sbostic 	save_addr = bufp->addr;
439fb91d121Sbostic 
440fb91d121Sbostic 	if (bp[2] == FULL_KEY_DATA) {		/* End of Data */
441fb91d121Sbostic 		totlen = len + mylen;
442c62a4a66Sbostic 		if (hashp->tmp_buf)
443c62a4a66Sbostic 			free(hashp->tmp_buf);
444*b601de3bSbostic 		if ((hashp->tmp_buf = (char *)malloc(totlen)) == NULL)
445fb91d121Sbostic 			return (-1);
446fb91d121Sbostic 		if (set) {
447fb91d121Sbostic 			hashp->cndx = 1;
448fb91d121Sbostic 			if (bp[0] == 2) {	/* No more buckets in chain */
449fb91d121Sbostic 				hashp->cpage = NULL;
450fb91d121Sbostic 				hashp->cbucket++;
451fb91d121Sbostic 			} else {
452c62a4a66Sbostic 				hashp->cpage =
4539d0f053fSbostic 				    __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
454c62a4a66Sbostic 				if (!hashp->cpage)
455fb91d121Sbostic 					return (-1);
456c62a4a66Sbostic 				else if (!((u_short *)hashp->cpage->page)[0]) {
457fb91d121Sbostic 					hashp->cbucket++;
458fb91d121Sbostic 					hashp->cpage = NULL;
459fb91d121Sbostic 				}
460fb91d121Sbostic 			}
461fb91d121Sbostic 		}
462fb91d121Sbostic 	} else {
4639d0f053fSbostic 		xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
4649d0f053fSbostic 		if (!xbp || ((totlen =
4659d0f053fSbostic 		    collect_data(hashp, xbp, len + mylen, set)) < 1))
466fb91d121Sbostic 			return (-1);
467fb91d121Sbostic 	}
468fb91d121Sbostic 	if (bufp->addr != save_addr) {
469c62a4a66Sbostic 		errno = EINVAL;			/* Out of buffers. */
470fb91d121Sbostic 		return (-1);
471fb91d121Sbostic 	}
47289be7141Sbostic 	memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], mylen);
473fb91d121Sbostic 	return (totlen);
474fb91d121Sbostic }
475fb91d121Sbostic 
476fb91d121Sbostic /*
477c62a4a66Sbostic  * Fill in the key and data for this big pair.
478fb91d121Sbostic  */
479fb91d121Sbostic extern int
4809d0f053fSbostic __big_keydata(hashp, bufp, key, val, set)
4819d0f053fSbostic 	HTAB *hashp;
482fb91d121Sbostic 	BUFHEAD *bufp;
483fb91d121Sbostic 	DBT *key, *val;
484fb91d121Sbostic 	int set;
485fb91d121Sbostic {
4869d0f053fSbostic 	key->size = collect_key(hashp, bufp, 0, val, set);
487c62a4a66Sbostic 	if (key->size == -1)
488fb91d121Sbostic 		return (-1);
4890a9054daSbostic 	key->data = (u_char *)hashp->tmp_key;
490fb91d121Sbostic 	return (0);
491fb91d121Sbostic }
492fb91d121Sbostic 
493fb91d121Sbostic /*
494c62a4a66Sbostic  * Count how big the total key size is by recursing through the pages.  Then
495c62a4a66Sbostic  * collect the data, allocate a buffer and copy the key as you recurse up.
496fb91d121Sbostic  */
497fb91d121Sbostic static int
collect_key(hashp,bufp,len,val,set)4989d0f053fSbostic collect_key(hashp, bufp, len, val, set)
4999d0f053fSbostic 	HTAB *hashp;
500fb91d121Sbostic 	BUFHEAD *bufp;
501fb91d121Sbostic 	int len;
502fb91d121Sbostic 	DBT *val;
503fb91d121Sbostic 	int set;
504fb91d121Sbostic {
505fb91d121Sbostic 	BUFHEAD *xbp;
506c62a4a66Sbostic 	char *p;
507c62a4a66Sbostic 	int mylen, totlen;
508c62a4a66Sbostic 	u_short *bp, save_addr;
509fb91d121Sbostic 
510c62a4a66Sbostic 	p = bufp->page;
511c62a4a66Sbostic 	bp = (u_short *)p;
512fb91d121Sbostic 	mylen = hashp->BSIZE - bp[1];
513fb91d121Sbostic 
514fb91d121Sbostic 	save_addr = bufp->addr;
515fb91d121Sbostic 	totlen = len + mylen;
516c62a4a66Sbostic 	if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) {    /* End of Key. */
517*b601de3bSbostic 		if (hashp->tmp_key != NULL)
518c62a4a66Sbostic 			free(hashp->tmp_key);
519*b601de3bSbostic 		if ((hashp->tmp_key = (char *)malloc(totlen)) == NULL)
520fb91d121Sbostic 			return (-1);
5219d0f053fSbostic 		if (__big_return(hashp, bufp, 1, val, set))
522b0a3b35bSbostic 			return (-1);
523fb91d121Sbostic 	} else {
5249d0f053fSbostic 		xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
5259d0f053fSbostic 		if (!xbp || ((totlen =
5269d0f053fSbostic 		    collect_key(hashp, xbp, totlen, val, set)) < 1))
527fb91d121Sbostic 			return (-1);
528fb91d121Sbostic 	}
529fb91d121Sbostic 	if (bufp->addr != save_addr) {
530fb91d121Sbostic 		errno = EINVAL;		/* MIS -- OUT OF BUFFERS */
531fb91d121Sbostic 		return (-1);
532fb91d121Sbostic 	}
53389be7141Sbostic 	memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], mylen);
534fb91d121Sbostic 	return (totlen);
535fb91d121Sbostic }
536fb91d121Sbostic 
537fb91d121Sbostic /*
538c62a4a66Sbostic  * Returns:
539c62a4a66Sbostic  *  0 => OK
540c62a4a66Sbostic  * -1 => error
541fb91d121Sbostic  */
542fb91d121Sbostic extern int
5439d0f053fSbostic __big_split(hashp, op, np, big_keyp, addr, obucket, ret)
5449d0f053fSbostic 	HTAB *hashp;
545fb91d121Sbostic 	BUFHEAD *op;	/* Pointer to where to put keys that go in old bucket */
546fb91d121Sbostic 	BUFHEAD *np;	/* Pointer to new bucket page */
547c62a4a66Sbostic 			/* Pointer to first page containing the big key/data */
548c62a4a66Sbostic 	BUFHEAD *big_keyp;
549c62a4a66Sbostic 	int addr;	/* Address of big_keyp */
550cb6cacddSbostic 	u_int   obucket;/* Old Bucket */
551fb91d121Sbostic 	SPLIT_RETURN *ret;
552fb91d121Sbostic {
553fb91d121Sbostic 	register BUFHEAD *tmpp;
554fb91d121Sbostic 	register u_short *tp;
555c62a4a66Sbostic 	BUFHEAD *bp;
556fb91d121Sbostic 	DBT key, val;
557cb6cacddSbostic 	u_int change;
558c62a4a66Sbostic 	u_short free_space, n, off;
559c62a4a66Sbostic 
560c62a4a66Sbostic 	bp = big_keyp;
561fb91d121Sbostic 
562fb91d121Sbostic 	/* Now figure out where the big key/data goes */
5639d0f053fSbostic 	if (__big_keydata(hashp, big_keyp, &key, &val, 0))
564fb91d121Sbostic 		return (-1);
5659d0f053fSbostic 	change = (__call_hash(hashp, key.data, key.size) != obucket);
566fb91d121Sbostic 
5679d0f053fSbostic 	if (ret->next_addr = __find_last_page(hashp, &big_keyp)) {
5689d0f053fSbostic 		if (!(ret->nextp =
5699d0f053fSbostic 		    __get_buf(hashp, ret->next_addr, big_keyp, 0)))
570fb91d121Sbostic 			return (-1);;
571c62a4a66Sbostic 	} else
572fb91d121Sbostic 		ret->nextp = NULL;
573fb91d121Sbostic 
574fb91d121Sbostic 	/* Now make one of np/op point to the big key/data pair */
575c62a4a66Sbostic #ifdef DEBUG
576fb91d121Sbostic 	assert(np->ovfl == NULL);
577c62a4a66Sbostic #endif
578c62a4a66Sbostic 	if (change)
579c62a4a66Sbostic 		tmpp = np;
580c62a4a66Sbostic 	else
581c62a4a66Sbostic 		tmpp = op;
582fb91d121Sbostic 
583fb91d121Sbostic 	tmpp->flags |= BUF_MOD;
584fb91d121Sbostic #ifdef DEBUG1
585c62a4a66Sbostic 	(void)fprintf(stderr,
586c62a4a66Sbostic 	    "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr,
587c62a4a66Sbostic 	    (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0));
588fb91d121Sbostic #endif
589fb91d121Sbostic 	tmpp->ovfl = bp;	/* one of op/np point to big_keyp */
590fb91d121Sbostic 	tp = (u_short *)tmpp->page;
591c62a4a66Sbostic #ifdef DEBUG
592fb91d121Sbostic 	assert(FREESPACE(tp) >= OVFLSIZE);
593c62a4a66Sbostic #endif
594fb91d121Sbostic 	n = tp[0];
595fb91d121Sbostic 	off = OFFSET(tp);
596fb91d121Sbostic 	free_space = FREESPACE(tp);
597c62a4a66Sbostic 	tp[++n] = (u_short)addr;
598fb91d121Sbostic 	tp[++n] = OVFLPAGE;
599fb91d121Sbostic 	tp[0] = n;
600fb91d121Sbostic 	OFFSET(tp) = off;
601fb91d121Sbostic 	FREESPACE(tp) = free_space - OVFLSIZE;
602fb91d121Sbostic 
603fb91d121Sbostic 	/*
604c62a4a66Sbostic 	 * Finally, set the new and old return values. BIG_KEYP contains a
605c62a4a66Sbostic 	 * pointer to the last page of the big key_data pair. Make sure that
606c62a4a66Sbostic 	 * big_keyp has no following page (2 elements) or create an empty
607c62a4a66Sbostic 	 * following page.
608fb91d121Sbostic 	 */
609fb91d121Sbostic 
610fb91d121Sbostic 	ret->newp = np;
611fb91d121Sbostic 	ret->oldp = op;
612fb91d121Sbostic 
613fb91d121Sbostic 	tp = (u_short *)big_keyp->page;
614fb91d121Sbostic 	big_keyp->flags |= BUF_MOD;
615fb91d121Sbostic 	if (tp[0] > 2) {
616fb91d121Sbostic 		/*
617c62a4a66Sbostic 		 * There may be either one or two offsets on this page.  If
618c62a4a66Sbostic 		 * there is one, then the overflow page is linked on normally
619c62a4a66Sbostic 		 * and tp[4] is OVFLPAGE.  If there are two, tp[4] contains
620c62a4a66Sbostic 		 * the second offset and needs to get stuffed in after the
621c62a4a66Sbostic 		 * next overflow page is added.
622fb91d121Sbostic 		 */
623fb91d121Sbostic 		n = tp[4];
624fb91d121Sbostic 		free_space = FREESPACE(tp);
625fb91d121Sbostic 		off = OFFSET(tp);
626fb91d121Sbostic 		tp[0] -= 2;
627fb91d121Sbostic 		FREESPACE(tp) = free_space + OVFLSIZE;
628fb91d121Sbostic 		OFFSET(tp) = off;
6299d0f053fSbostic 		tmpp = __add_ovflpage(hashp, big_keyp);
630c62a4a66Sbostic 		if (!tmpp)
631fb91d121Sbostic 			return (-1);
632fb91d121Sbostic 		tp[4] = n;
633c62a4a66Sbostic 	} else
634fb91d121Sbostic 		tmpp = big_keyp;
635fb91d121Sbostic 
636c62a4a66Sbostic 	if (change)
637c62a4a66Sbostic 		ret->newp = tmpp;
638c62a4a66Sbostic 	else
639c62a4a66Sbostic 		ret->oldp = tmpp;
640fb91d121Sbostic 	return (0);
641fb91d121Sbostic }
642