michael@0: /*-
michael@0:  * Copyright (c) 1990, 1993, 1994
michael@0:  *	The Regents of the University of California.  All rights reserved.
michael@0:  *
michael@0:  * This code is derived from software contributed to Berkeley by
michael@0:  * Margo Seltzer.
michael@0:  *
michael@0:  * Redistribution and use in source and binary forms, with or without
michael@0:  * modification, are permitted provided that the following conditions
michael@0:  * are met:
michael@0:  * 1. Redistributions of source code must retain the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer.
michael@0:  * 2. Redistributions in binary form must reproduce the above copyright
michael@0:  *    notice, this list of conditions and the following disclaimer in the
michael@0:  *    documentation and/or other materials provided with the distribution.
michael@0:  * 3. ***REMOVED*** - see 
michael@0:  *    ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change
michael@0:  * 4. Neither the name of the University nor the names of its contributors
michael@0:  *    may be used to endorse or promote products derived from this software
michael@0:  *    without specific prior written permission.
michael@0:  *
michael@0:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
michael@0:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
michael@0:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
michael@0:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
michael@0:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
michael@0:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
michael@0:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
michael@0:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
michael@0:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
michael@0:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
michael@0:  * SUCH DAMAGE.
michael@0:  */
michael@0: 
michael@0: #if defined(LIBC_SCCS) && !defined(lint)
michael@0: static char sccsid[] = "@(#)hash_bigkey.c	8.3 (Berkeley) 5/31/94";
michael@0: #endif /* LIBC_SCCS and not lint */
michael@0: 
michael@0: /*
michael@0:  * PACKAGE: hash
michael@0:  * DESCRIPTION:
michael@0:  *	Big key/data handling for the hashing package.
michael@0:  *
michael@0:  * ROUTINES:
michael@0:  * External
michael@0:  *	__big_keydata
michael@0:  *	__big_split
michael@0:  *	__big_insert
michael@0:  *	__big_return
michael@0:  *	__big_delete
michael@0:  *	__find_last_page
michael@0:  * Internal
michael@0:  *	collect_key
michael@0:  *	collect_data
michael@0:  */
michael@0: 
michael@0: #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh)
michael@0: #include <sys/param.h>
michael@0: #endif
michael@0: 
michael@0: #include <errno.h>
michael@0: #include <stdio.h>
michael@0: #include <stdlib.h>
michael@0: #include <string.h>
michael@0: 
michael@0: #ifdef DEBUG
michael@0: #include <assert.h>
michael@0: #endif
michael@0: 
michael@0: #include "mcom_db.h"
michael@0: #include "hash.h"
michael@0: #include "page.h"
michael@0: /* #include "extern.h" */
michael@0: 
michael@0: static int collect_key __P((HTAB *, BUFHEAD *, int, DBT *, int));
michael@0: static int collect_data __P((HTAB *, BUFHEAD *, int, int));
michael@0: 
michael@0: /*
michael@0:  * Big_insert
michael@0:  *
michael@0:  * You need to do an insert and the key/data pair is too big
michael@0:  *
michael@0:  * Returns:
michael@0:  * 0 ==> OK
michael@0:  *-1 ==> ERROR
michael@0:  */
michael@0: extern int
michael@0: __big_insert(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
michael@0: {
michael@0: 	register uint16 *p;
michael@0: 	uint key_size, n, val_size;
michael@0: 	uint16 space, move_bytes, off;
michael@0: 	char *cp, *key_data, *val_data;
michael@0: 
michael@0: 	cp = bufp->page;		/* Character pointer of p. */
michael@0: 	p = (uint16 *)cp;
michael@0: 
michael@0: 	key_data = (char *)key->data;
michael@0: 	key_size = key->size;
michael@0: 	val_data = (char *)val->data;
michael@0: 	val_size = val->size;
michael@0: 
michael@0: 	/* First move the Key */
michael@0: 	for (space = FREESPACE(p) - BIGOVERHEAD; key_size;
michael@0: 	    space = FREESPACE(p) - BIGOVERHEAD) {
michael@0: 		move_bytes = PR_MIN(space, key_size);
michael@0: 		off = OFFSET(p) - move_bytes;
michael@0: 		memmove(cp + off, key_data, move_bytes);
michael@0: 		key_size -= move_bytes;
michael@0: 		key_data += move_bytes;
michael@0: 		n = p[0];
michael@0: 		p[++n] = off;
michael@0: 		p[0] = ++n;
michael@0: 		FREESPACE(p) = off - PAGE_META(n);
michael@0: 		OFFSET(p) = off;
michael@0: 		p[n] = PARTIAL_KEY;
michael@0: 		bufp = __add_ovflpage(hashp, bufp);
michael@0: 		if (!bufp)
michael@0: 			return (-1);
michael@0: 		n = p[0];
michael@0: 		if (!key_size) {
michael@0: 			if (FREESPACE(p)) {
michael@0: 				move_bytes = PR_MIN(FREESPACE(p), val_size);
michael@0: 				off = OFFSET(p) - move_bytes;
michael@0: 				p[n] = off;
michael@0: 				memmove(cp + off, val_data, move_bytes);
michael@0: 				val_data += move_bytes;
michael@0: 				val_size -= move_bytes;
michael@0: 				p[n - 2] = FULL_KEY_DATA;
michael@0: 				FREESPACE(p) = FREESPACE(p) - move_bytes;
michael@0: 				OFFSET(p) = off;
michael@0: 			} else
michael@0: 				p[n - 2] = FULL_KEY;
michael@0: 		}
michael@0: 		p = (uint16 *)bufp->page;
michael@0: 		cp = bufp->page;
michael@0: 		bufp->flags |= BUF_MOD;
michael@0: 	}
michael@0: 
michael@0: 	/* Now move the data */
michael@0: 	for (space = FREESPACE(p) - BIGOVERHEAD; val_size;
michael@0: 	    space = FREESPACE(p) - BIGOVERHEAD) {
michael@0: 		move_bytes = PR_MIN(space, val_size);
michael@0: 		/*
michael@0: 		 * Here's the hack to make sure that if the data ends on the
michael@0: 		 * same page as the key ends, FREESPACE is at least one.
michael@0: 		 */
michael@0: 		if (space == val_size && val_size == val->size)
michael@0: 			move_bytes--;
michael@0: 		off = OFFSET(p) - move_bytes;
michael@0: 		memmove(cp + off, val_data, move_bytes);
michael@0: 		val_size -= move_bytes;
michael@0: 		val_data += move_bytes;
michael@0: 		n = p[0];
michael@0: 		p[++n] = off;
michael@0: 		p[0] = ++n;
michael@0: 		FREESPACE(p) = off - PAGE_META(n);
michael@0: 		OFFSET(p) = off;
michael@0: 		if (val_size) {
michael@0: 			p[n] = FULL_KEY;
michael@0: 			bufp = __add_ovflpage(hashp, bufp);
michael@0: 			if (!bufp)
michael@0: 				return (-1);
michael@0: 			cp = bufp->page;
michael@0: 			p = (uint16 *)cp;
michael@0: 		} else
michael@0: 			p[n] = FULL_KEY_DATA;
michael@0: 		bufp->flags |= BUF_MOD;
michael@0: 	}
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Called when bufp's page  contains a partial key (index should be 1)
michael@0:  *
michael@0:  * All pages in the big key/data pair except bufp are freed.  We cannot
michael@0:  * free bufp because the page pointing to it is lost and we can't get rid
michael@0:  * of its pointer.
michael@0:  *
michael@0:  * Returns:
michael@0:  * 0 => OK
michael@0:  *-1 => ERROR
michael@0:  */
michael@0: extern int
michael@0: __big_delete(HTAB *hashp, BUFHEAD *bufp)
michael@0: {
michael@0: 	register BUFHEAD *last_bfp, *rbufp;
michael@0: 	uint16 *bp, pageno;
michael@0: 	int key_done, n;
michael@0: 
michael@0: 	rbufp = bufp;
michael@0: 	last_bfp = NULL;
michael@0: 	bp = (uint16 *)bufp->page;
michael@0: 	pageno = 0;
michael@0: 	key_done = 0;
michael@0: 
michael@0: 	while (!key_done || (bp[2] != FULL_KEY_DATA)) {
michael@0: 		if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA)
michael@0: 			key_done = 1;
michael@0: 
michael@0: 		/*
michael@0: 		 * If there is freespace left on a FULL_KEY_DATA page, then
michael@0: 		 * the data is short and fits entirely on this page, and this
michael@0: 		 * is the last page.
michael@0: 		 */
michael@0: 		if (bp[2] == FULL_KEY_DATA && FREESPACE(bp))
michael@0: 			break;
michael@0: 		pageno = bp[bp[0] - 1];
michael@0: 		rbufp->flags |= BUF_MOD;
michael@0: 		rbufp = __get_buf(hashp, pageno, rbufp, 0);
michael@0: 		if (last_bfp)
michael@0: 			__free_ovflpage(hashp, last_bfp);
michael@0: 		last_bfp = rbufp;
michael@0: 		if (!rbufp)
michael@0: 			return (-1);		/* Error. */
michael@0: 		bp = (uint16 *)rbufp->page;
michael@0: 	}
michael@0: 
michael@0: 	/*
michael@0: 	 * If we get here then rbufp points to the last page of the big
michael@0: 	 * key/data pair.  Bufp points to the first one -- it should now be
michael@0: 	 * empty pointing to the next page after this pair.  Can't free it
michael@0: 	 * because we don't have the page pointing to it.
michael@0: 	 */
michael@0: 
michael@0: 	/* This is information from the last page of the pair. */
michael@0: 	n = bp[0];
michael@0: 	pageno = bp[n - 1];
michael@0: 
michael@0: 	/* Now, bp is the first page of the pair. */
michael@0: 	bp = (uint16 *)bufp->page;
michael@0: 	if (n > 2) {
michael@0: 		/* There is an overflow page. */
michael@0: 		bp[1] = pageno;
michael@0: 		bp[2] = OVFLPAGE;
michael@0: 		bufp->ovfl = rbufp->ovfl;
michael@0: 	} else
michael@0: 		/* This is the last page. */
michael@0: 		bufp->ovfl = NULL;
michael@0: 	n -= 2;
michael@0: 	bp[0] = n;
michael@0: 	FREESPACE(bp) = hashp->BSIZE - PAGE_META(n);
michael@0: 	OFFSET(bp) = hashp->BSIZE - 1;
michael@0: 
michael@0: 	bufp->flags |= BUF_MOD;
michael@0: 	if (rbufp)
michael@0: 		__free_ovflpage(hashp, rbufp);
michael@0: 	if (last_bfp != rbufp)
michael@0: 		__free_ovflpage(hashp, last_bfp);
michael@0: 
michael@0: 	hashp->NKEYS--;
michael@0: 	return (0);
michael@0: }
michael@0: /*
michael@0:  * Returns:
michael@0:  *  0 = key not found
michael@0:  * -1 = get next overflow page
michael@0:  * -2 means key not found and this is big key/data
michael@0:  * -3 error
michael@0:  */
michael@0: extern int
michael@0: __find_bigpair(HTAB *hashp, BUFHEAD *bufp, int ndx, char *key, int size)
michael@0: {
michael@0: 	register uint16 *bp;
michael@0: 	register char *p;
michael@0: 	int ksize;
michael@0: 	uint16 bytes;
michael@0: 	char *kkey;
michael@0: 
michael@0: 	bp = (uint16 *)bufp->page;
michael@0: 	p = bufp->page;
michael@0: 	ksize = size;
michael@0: 	kkey = key;
michael@0: 
michael@0: 	for (bytes = hashp->BSIZE - bp[ndx];
michael@0: 	    bytes <= size && bp[ndx + 1] == PARTIAL_KEY;
michael@0: 	    bytes = hashp->BSIZE - bp[ndx]) {
michael@0: 		if (memcmp(p + bp[ndx], kkey, bytes))
michael@0: 			return (-2);
michael@0: 		kkey += bytes;
michael@0: 		ksize -= bytes;
michael@0: 		bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0);
michael@0: 		if (!bufp)
michael@0: 			return (-3);
michael@0: 		p = bufp->page;
michael@0: 		bp = (uint16 *)p;
michael@0: 		ndx = 1;
michael@0: 	}
michael@0: 
michael@0: 	if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) {
michael@0: #ifdef HASH_STATISTICS
michael@0: 		++hash_collisions;
michael@0: #endif
michael@0: 		return (-2);
michael@0: 	} else
michael@0: 		return (ndx);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Given the buffer pointer of the first overflow page of a big pair,
michael@0:  * find the end of the big pair
michael@0:  *
michael@0:  * This will set bpp to the buffer header of the last page of the big pair.
michael@0:  * It will return the pageno of the overflow page following the last page
michael@0:  * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
michael@0:  * bucket)
michael@0:  */
michael@0: extern uint16
michael@0: __find_last_page(HTAB *hashp, BUFHEAD **bpp)
michael@0: {
michael@0: 	BUFHEAD *bufp;
michael@0: 	uint16 *bp, pageno;
michael@0: 	uint n;
michael@0: 
michael@0: 	bufp = *bpp;
michael@0: 	bp = (uint16 *)bufp->page;
michael@0: 	for (;;) {
michael@0: 		n = bp[0];
michael@0: 
michael@0: 		/*
michael@0: 		 * This is the last page if: the tag is FULL_KEY_DATA and
michael@0: 		 * either only 2 entries OVFLPAGE marker is explicit there
michael@0: 		 * is freespace on the page.
michael@0: 		 */
michael@0: 		if (bp[2] == FULL_KEY_DATA &&
michael@0: 		    ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp))))
michael@0: 			break;
michael@0: 
michael@0: 		/* LJM bound the size of n to reasonable limits
michael@0: 		 */
michael@0: 		if(n > hashp->BSIZE/sizeof(uint16))
michael@0: 			return(0);
michael@0: 
michael@0: 		pageno = bp[n - 1];
michael@0: 		bufp = __get_buf(hashp, pageno, bufp, 0);
michael@0: 		if (!bufp)
michael@0: 			return (0);	/* Need to indicate an error! */
michael@0: 		bp = (uint16 *)bufp->page;
michael@0: 	}
michael@0: 
michael@0: 	*bpp = bufp;
michael@0: 	if (bp[0] > 2)
michael@0: 		return (bp[3]);
michael@0: 	else
michael@0: 		return (0);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Return the data for the key/data pair that begins on this page at this
michael@0:  * index (index should always be 1).
michael@0:  */
michael@0: extern int
michael@0: __big_return(
michael@0: 	HTAB *hashp,
michael@0: 	BUFHEAD *bufp,
michael@0: 	int ndx,
michael@0: 	DBT *val,
michael@0: 	int set_current)
michael@0: {
michael@0: 	BUFHEAD *save_p;
michael@0: 	uint16 *bp, len, off, save_addr;
michael@0: 	char *tp;
michael@0: 	int save_flags;
michael@0: 
michael@0: 	bp = (uint16 *)bufp->page;
michael@0: 	while (bp[ndx + 1] == PARTIAL_KEY) {
michael@0: 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
michael@0: 		if (!bufp)
michael@0: 			return (-1);
michael@0: 		bp = (uint16 *)bufp->page;
michael@0: 		ndx = 1;
michael@0: 	}
michael@0: 
michael@0: 	if (bp[ndx + 1] == FULL_KEY) {
michael@0: 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
michael@0: 		if (!bufp)
michael@0: 			return (-1);
michael@0: 		bp = (uint16 *)bufp->page;
michael@0: 		save_p = bufp;
michael@0: 		save_addr = save_p->addr;
michael@0: 		off = bp[1];
michael@0: 		len = 0;
michael@0: 	} else
michael@0: 		if (!FREESPACE(bp)) {
michael@0: 			/*
michael@0: 			 * This is a hack.  We can't distinguish between
michael@0: 			 * FULL_KEY_DATA that contains complete data or
michael@0: 			 * incomplete data, so we require that if the data
michael@0: 			 * is complete, there is at least 1 byte of free
michael@0: 			 * space left.
michael@0: 			 */
michael@0: 			off = bp[bp[0]];
michael@0: 			len = bp[1] - off;
michael@0: 			save_p = bufp;
michael@0: 			save_addr = bufp->addr;
michael@0: 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
michael@0: 			if (!bufp)
michael@0: 				return (-1);
michael@0: 			bp = (uint16 *)bufp->page;
michael@0: 		} else {
michael@0: 			/* The data is all on one page. */
michael@0: 			tp = (char *)bp;
michael@0: 			off = bp[bp[0]];
michael@0: 			val->data = (uint8 *)tp + off;
michael@0: 			val->size = bp[1] - off;
michael@0: 			if (set_current) {
michael@0: 				if (bp[0] == 2) {	/* No more buckets in
michael@0: 							 * chain */
michael@0: 					hashp->cpage = NULL;
michael@0: 					hashp->cbucket++;
michael@0: 					hashp->cndx = 1;
michael@0: 				} else {
michael@0: 					hashp->cpage = __get_buf(hashp,
michael@0: 					    bp[bp[0] - 1], bufp, 0);
michael@0: 					if (!hashp->cpage)
michael@0: 						return (-1);
michael@0: 					hashp->cndx = 1;
michael@0: 					if (!((uint16 *)
michael@0: 					    hashp->cpage->page)[0]) {
michael@0: 						hashp->cbucket++;
michael@0: 						hashp->cpage = NULL;
michael@0: 					}
michael@0: 				}
michael@0: 			}
michael@0: 			return (0);
michael@0: 		}
michael@0: 
michael@0: 	/* pin our saved buf so that we don't lose if 
michael@0: 	 * we run out of buffers */
michael@0:  	save_flags = save_p->flags;
michael@0: 	save_p->flags |= BUF_PIN;
michael@0: 	val->size = collect_data(hashp, bufp, (int)len, set_current);
michael@0: 	save_p->flags = save_flags;
michael@0: 	if (val->size == (size_t)-1)
michael@0: 		return (-1);
michael@0: 	if (save_p->addr != save_addr) {
michael@0: 		/* We are pretty short on buffers. */
michael@0: 		errno = EINVAL;			/* OUT OF BUFFERS */
michael@0: 		return (-1);
michael@0: 	}
michael@0: 	memmove(hashp->tmp_buf, (save_p->page) + off, len);
michael@0: 	val->data = (uint8 *)hashp->tmp_buf;
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Count how big the total datasize is by looping through the pages.  Then
michael@0:  * allocate a buffer and copy the data in the second loop. NOTE: Our caller
michael@0:  * may already have a bp which it is holding onto. The caller is
michael@0:  * responsible for copying that bp into our temp buffer. 'len' is how much
michael@0:  * space to reserve for that buffer.
michael@0:  */
michael@0: static int
michael@0: collect_data(
michael@0: 	HTAB *hashp,
michael@0: 	BUFHEAD *bufp,
michael@0: 	int len, int set)
michael@0: {
michael@0: 	register uint16 *bp;
michael@0: 	BUFHEAD *save_bufp;
michael@0: 	int save_flags;
michael@0: 	int mylen, totlen;
michael@0: 
michael@0: 	/*
michael@0: 	 * save the input buf head because we need to walk the list twice.
michael@0: 	 * pin it to make sure it doesn't leave the buffer pool. 
michael@0: 	 * This has the effect of growing the buffer pool if necessary.
michael@0: 	 */
michael@0: 	save_bufp = bufp;
michael@0: 	save_flags = save_bufp->flags;
michael@0: 	save_bufp->flags |= BUF_PIN;
michael@0: 
michael@0: 	/* read the length of the buffer */
michael@0: 	for (totlen = len; bufp ; bufp = __get_buf(hashp, bp[bp[0]-1], bufp, 0)) {
michael@0: 		bp = (uint16 *)bufp->page;
michael@0: 		mylen = hashp->BSIZE - bp[1];
michael@0: 
michael@0: 		/* if mylen ever goes negative it means that the
michael@0: 		 * page is screwed up.
michael@0: 		 */
michael@0: 		if (mylen < 0) {
michael@0: 			save_bufp->flags = save_flags;
michael@0: 			return (-1);
michael@0:  		}
michael@0: 		totlen += mylen;
michael@0: 		if (bp[2] == FULL_KEY_DATA) {		/* End of Data */
michael@0: 			break;
michael@0: 		}
michael@0: 	}
michael@0: 
michael@0:  	if (!bufp) {
michael@0: 		save_bufp->flags = save_flags;
michael@0: 		return (-1);
michael@0: 	}
michael@0: 
michael@0: 	/* allocate a temp buf */
michael@0: 	if (hashp->tmp_buf)
michael@0: 		free(hashp->tmp_buf);
michael@0: 	if ((hashp->tmp_buf = (char *)malloc((size_t)totlen)) == NULL) {
michael@0: 		save_bufp->flags = save_flags;
michael@0: 		return (-1);
michael@0:  	}
michael@0: 
michael@0: 	/* copy the buffers back into temp buf */
michael@0: 	for (bufp = save_bufp; bufp ;
michael@0: 				bufp = __get_buf(hashp, bp[bp[0]-1], bufp, 0)) {
michael@0: 		bp = (uint16 *)bufp->page;
michael@0: 		mylen = hashp->BSIZE - bp[1];
michael@0: 		memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], (size_t)mylen);
michael@0: 		len += mylen;
michael@0: 		if (bp[2] == FULL_KEY_DATA) {
michael@0: 			break;
michael@0: 		}
michael@0: 	}
michael@0: 
michael@0: 	/* 'clear' the pin flags */
michael@0: 	save_bufp->flags = save_flags;
michael@0: 
michael@0: 	/* update the database cursor */
michael@0: 	if (set) {
michael@0: 		hashp->cndx = 1;
michael@0: 		if (bp[0] == 2) {	/* No more buckets in chain */
michael@0: 			hashp->cpage = NULL;
michael@0: 			hashp->cbucket++;
michael@0: 		} else {
michael@0: 			hashp->cpage = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
michael@0: 			if (!hashp->cpage)
michael@0: 				return (-1);
michael@0: 			else if (!((uint16 *)hashp->cpage->page)[0]) {
michael@0: 				hashp->cbucket++;
michael@0: 				hashp->cpage = NULL;
michael@0: 			}
michael@0: 		}
michael@0: 	}
michael@0: 	return (totlen);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Fill in the key and data for this big pair.
michael@0:  */
michael@0: extern int
michael@0: __big_keydata(
michael@0: 	HTAB *hashp, 
michael@0: 	BUFHEAD *bufp, 
michael@0: 	DBT *key, DBT *val,
michael@0: 	int set)
michael@0: {
michael@0: 	key->size = collect_key(hashp, bufp, 0, val, set);
michael@0: 	if (key->size == (size_t)-1)
michael@0: 		return (-1);
michael@0: 	key->data = (uint8 *)hashp->tmp_key;
michael@0: 	return (0);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Count how big the total key size is by recursing through the pages.  Then
michael@0:  * collect the data, allocate a buffer and copy the key as you recurse up.
michael@0:  */
michael@0: static int
michael@0: collect_key(
michael@0: 	HTAB *hashp,
michael@0: 	BUFHEAD *bufp,
michael@0: 	int len,
michael@0: 	DBT *val,
michael@0: 	int set)
michael@0: {
michael@0: 	BUFHEAD *xbp;
michael@0: 	char *p;
michael@0: 	int mylen, totlen;
michael@0: 	uint16 *bp, save_addr;
michael@0: 
michael@0: 	p = bufp->page;
michael@0: 	bp = (uint16 *)p;
michael@0: 	mylen = hashp->BSIZE - bp[1];
michael@0: 
michael@0: 	save_addr = bufp->addr;
michael@0: 	totlen = len + mylen;
michael@0: 	if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) {    /* End of Key. */
michael@0: 		if (hashp->tmp_key != NULL)
michael@0: 			free(hashp->tmp_key);
michael@0: 		if ((hashp->tmp_key = (char *)malloc((size_t)totlen)) == NULL)
michael@0: 			return (-1);
michael@0: 		if (__big_return(hashp, bufp, 1, val, set))
michael@0: 			return (-1);
michael@0: 	} else {
michael@0: 		xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
michael@0: 		if (!xbp || ((totlen =
michael@0: 		    collect_key(hashp, xbp, totlen, val, set)) < 1))
michael@0: 			return (-1);
michael@0: 	}
michael@0: 	if (bufp->addr != save_addr) {
michael@0: 		errno = EINVAL;		/* MIS -- OUT OF BUFFERS */
michael@0: 		return (-1);
michael@0: 	}
michael@0: 	memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], (size_t)mylen);
michael@0: 	return (totlen);
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Returns:
michael@0:  *  0 => OK
michael@0:  * -1 => error
michael@0:  */
michael@0: extern int
michael@0: __big_split(
michael@0: 	HTAB *hashp,
michael@0: 	BUFHEAD *op,	/* Pointer to where to put keys that go in old bucket */
michael@0: 	BUFHEAD *np,	/* Pointer to new bucket page */
michael@0: 			/* Pointer to first page containing the big key/data */
michael@0: 	BUFHEAD *big_keyp,
michael@0: 	uint32 addr,	/* Address of big_keyp */
michael@0: 	uint32   obucket,/* Old Bucket */
michael@0: 	SPLIT_RETURN *ret)
michael@0: {
michael@0: 	register BUFHEAD *tmpp;
michael@0: 	register uint16 *tp;
michael@0: 	BUFHEAD *bp;
michael@0: 	DBT key, val;
michael@0: 	uint32 change;
michael@0: 	uint16 free_space, n, off;
michael@0: 
michael@0: 	bp = big_keyp;
michael@0: 
michael@0: 	/* Now figure out where the big key/data goes */
michael@0: 	if (__big_keydata(hashp, big_keyp, &key, &val, 0))
michael@0: 		return (-1);
michael@0: 	change = (__call_hash(hashp,(char*) key.data, key.size) != obucket);
michael@0: 
michael@0: 	if ((ret->next_addr = __find_last_page(hashp, &big_keyp))) {
michael@0: 		if (!(ret->nextp =
michael@0: 		    __get_buf(hashp, ret->next_addr, big_keyp, 0)))
michael@0: 			return (-1);;
michael@0: 	} else
michael@0: 		ret->nextp = NULL;
michael@0: 
michael@0: 	/* Now make one of np/op point to the big key/data pair */
michael@0: #ifdef DEBUG
michael@0: 	assert(np->ovfl == NULL);
michael@0: #endif
michael@0: 	if (change)
michael@0: 		tmpp = np;
michael@0: 	else
michael@0: 		tmpp = op;
michael@0: 
michael@0: 	tmpp->flags |= BUF_MOD;
michael@0: #ifdef DEBUG1
michael@0: 	(void)fprintf(stderr,
michael@0: 	    "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr,
michael@0: 	    (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0));
michael@0: #endif
michael@0: 	tmpp->ovfl = bp;	/* one of op/np point to big_keyp */
michael@0: 	tp = (uint16 *)tmpp->page;
michael@0: 
michael@0: 
michael@0: #if 0  /* this get's tripped on database corrupted error */
michael@0: 	assert(FREESPACE(tp) >= OVFLSIZE);
michael@0: #endif
michael@0: 	if(FREESPACE(tp) < OVFLSIZE)
michael@0: 		return(DATABASE_CORRUPTED_ERROR);
michael@0: 
michael@0: 	n = tp[0];
michael@0: 	off = OFFSET(tp);
michael@0: 	free_space = FREESPACE(tp);
michael@0: 	tp[++n] = (uint16)addr;
michael@0: 	tp[++n] = OVFLPAGE;
michael@0: 	tp[0] = n;
michael@0: 	OFFSET(tp) = off;
michael@0: 	FREESPACE(tp) = free_space - OVFLSIZE;
michael@0: 
michael@0: 	/*
michael@0: 	 * Finally, set the new and old return values. BIG_KEYP contains a
michael@0: 	 * pointer to the last page of the big key_data pair. Make sure that
michael@0: 	 * big_keyp has no following page (2 elements) or create an empty
michael@0: 	 * following page.
michael@0: 	 */
michael@0: 
michael@0: 	ret->newp = np;
michael@0: 	ret->oldp = op;
michael@0: 
michael@0: 	tp = (uint16 *)big_keyp->page;
michael@0: 	big_keyp->flags |= BUF_MOD;
michael@0: 	if (tp[0] > 2) {
michael@0: 		/*
michael@0: 		 * There may be either one or two offsets on this page.  If
michael@0: 		 * there is one, then the overflow page is linked on normally
michael@0: 		 * and tp[4] is OVFLPAGE.  If there are two, tp[4] contains
michael@0: 		 * the second offset and needs to get stuffed in after the
michael@0: 		 * next overflow page is added.
michael@0: 		 */
michael@0: 		n = tp[4];
michael@0: 		free_space = FREESPACE(tp);
michael@0: 		off = OFFSET(tp);
michael@0: 		tp[0] -= 2;
michael@0: 		FREESPACE(tp) = free_space + OVFLSIZE;
michael@0: 		OFFSET(tp) = off;
michael@0: 		tmpp = __add_ovflpage(hashp, big_keyp);
michael@0: 		if (!tmpp)
michael@0: 			return (-1);
michael@0: 		tp[4] = n;
michael@0: 	} else
michael@0: 		tmpp = big_keyp;
michael@0: 
michael@0: 	if (change)
michael@0: 		ret->newp = tmpp;
michael@0: 	else
michael@0: 		ret->oldp = tmpp;
michael@0: 	return (0);
michael@0: }