andre@3: /*-
andre@3:  * Copyright (c) 1990, 1993, 1994
andre@3:  *	The Regents of the University of California.  All rights reserved.
andre@3:  *
andre@3:  * This code is derived from software contributed to Berkeley by
andre@3:  * Margo Seltzer.
andre@3:  *
andre@3:  * Redistribution and use in source and binary forms, with or without
andre@3:  * modification, are permitted provided that the following conditions
andre@3:  * are met:
andre@3:  * 1. Redistributions of source code must retain the above copyright
andre@3:  *    notice, this list of conditions and the following disclaimer.
andre@3:  * 2. Redistributions in binary form must reproduce the above copyright
andre@3:  *    notice, this list of conditions and the following disclaimer in the
andre@3:  *    documentation and/or other materials provided with the distribution.
andre@3:  * 3. ***REMOVED*** - see 
andre@3:  *    ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change
andre@3:  * 4. Neither the name of the University nor the names of its contributors
andre@3:  *    may be used to endorse or promote products derived from this software
andre@3:  *    without specific prior written permission.
andre@3:  *
andre@3:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
andre@3:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
andre@3:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
andre@3:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
andre@3:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
andre@3:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
andre@3:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
andre@3:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
andre@3:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
andre@3:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
andre@3:  * SUCH DAMAGE.
andre@3:  */
andre@3: 
andre@3: #if defined(LIBC_SCCS) && !defined(lint)
andre@3: static char sccsid[] = "@(#)hash_buf.c	8.5 (Berkeley) 7/15/94";
andre@3: #endif /* LIBC_SCCS and not lint */
andre@3: 
andre@3: /*
andre@3:  * PACKAGE: hash
andre@3:  *
andre@3:  * DESCRIPTION:
andre@3:  *	Contains buffer management
andre@3:  *
andre@3:  * ROUTINES:
andre@3:  * External
andre@3:  *	__buf_init
andre@3:  *	__get_buf
andre@3:  *	__buf_free
andre@3:  *	__reclaim_buf
andre@3:  * Internal
andre@3:  *	newbuf
andre@3:  */
andre@3: #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh)
andre@3: #include <sys/param.h>
andre@3: #endif
andre@3: 
andre@3: #include <errno.h>
andre@3: #include <stddef.h>
andre@3: #include <stdio.h>
andre@3: #include <stdlib.h>
andre@3: #include <string.h>
andre@3: 
andre@3: #ifdef DEBUG
andre@3: #include <assert.h>
andre@3: #endif
andre@3: 
andre@3: #include "mcom_db.h"
andre@3: #include "hash.h"
andre@3: #include "page.h"
andre@3: /* #include "extern.h" */
andre@3: 
andre@3: static BUFHEAD *newbuf __P((HTAB *, uint32, BUFHEAD *));
andre@3: 
andre@3: /* Unlink B from its place in the lru */
andre@3: #define BUF_REMOVE(B) { \
andre@3: 	(B)->prev->next = (B)->next; \
andre@3: 	(B)->next->prev = (B)->prev; \
andre@3: }
andre@3: 
andre@3: /* Insert B after P */
andre@3: #define BUF_INSERT(B, P) { \
andre@3: 	(B)->next = (P)->next; \
andre@3: 	(B)->prev = (P); \
andre@3: 	(P)->next = (B); \
andre@3: 	(B)->next->prev = (B); \
andre@3: }
andre@3: 
andre@3: #define	MRU	hashp->bufhead.next
andre@3: #define	LRU	hashp->bufhead.prev
andre@3: 
andre@3: #define MRU_INSERT(B)	BUF_INSERT((B), &hashp->bufhead)
andre@3: #define LRU_INSERT(B)	BUF_INSERT((B), LRU)
andre@3: 
andre@3: /*
andre@3:  * We are looking for a buffer with address "addr".  If prev_bp is NULL, then
andre@3:  * address is a bucket index.  If prev_bp is not NULL, then it points to the
andre@3:  * page previous to an overflow page that we are trying to find.
andre@3:  *
andre@3:  * CAVEAT:  The buffer header accessed via prev_bp's ovfl field may no longer
andre@3:  * be valid.  Therefore, you must always verify that its address matches the
andre@3:  * address you are seeking.
andre@3:  */
andre@3: extern BUFHEAD *
andre@3: __get_buf(HTAB *hashp, uint32 addr, BUFHEAD *prev_bp, int newpage)
andre@3: /* If prev_bp set, indicates a new overflow page. */
andre@3: {
andre@3: 	register BUFHEAD *bp;
andre@3: 	register uint32 is_disk_mask;
andre@3: 	register int is_disk, segment_ndx = 0;
andre@3: 	SEGMENT segp = 0;
andre@3: 
andre@3: 	is_disk = 0;
andre@3: 	is_disk_mask = 0;
andre@3: 	if (prev_bp) {
andre@3: 		bp = prev_bp->ovfl;
andre@3: 		if (!bp || (bp->addr != addr))
andre@3: 			bp = NULL;
andre@3: 		if (!newpage)
andre@3: 			is_disk = BUF_DISK;
andre@3: 	} else {
andre@3: 		/* Grab buffer out of directory */
andre@3: 		segment_ndx = addr & (hashp->SGSIZE - 1);
andre@3: 
andre@3: 		/* valid segment ensured by __call_hash() */
andre@3: 		segp = hashp->dir[addr >> hashp->SSHIFT];
andre@3: #ifdef DEBUG
andre@3: 		assert(segp != NULL);
andre@3: #endif  
andre@3: 
andre@3: 		bp = PTROF(segp[segment_ndx]);
andre@3: 
andre@3: 		is_disk_mask = ISDISK(segp[segment_ndx]);
andre@3: 		is_disk = is_disk_mask || !hashp->new_file;
andre@3: 	}
andre@3: 
andre@3: 	if (!bp) {
andre@3: 		bp = newbuf(hashp, addr, prev_bp);
andre@3: 		if (!bp)
andre@3: 			return(NULL);
andre@3: 		if(__get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0))
andre@3: 		  {
andre@3: 			/* free bp and its page */
andre@3: 			if(prev_bp)
andre@3: 			  {
andre@3: 				/* if prev_bp is set then the new page that
andre@3: 				 * failed is hooked onto prev_bp as an overflow page.
andre@3: 				 * if we don't remove the pointer to the bad page
andre@3: 				 * we may try and access it later and we will die
andre@3: 				 * horribly because it will have already been
andre@3: 				 * free'd and overwritten with bogus data.
andre@3: 				 */
andre@3: 				prev_bp->ovfl = NULL;
andre@3: 			  }
andre@3: 			BUF_REMOVE(bp);
andre@3: 			free(bp->page);
andre@3: 			free(bp);
andre@3: 			return (NULL);
andre@3: 		  }
andre@3: 
andre@3: 		if (!prev_bp)			
andre@3: 		  {
andre@3: #if 0
andre@3: 			/* 16 bit windows and mac can't handle the
andre@3: 			 * oring of the is disk flag.
andre@3: 		 	 */			
andre@3: 			segp[segment_ndx] =
andre@3: 			    (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask);
andre@3: #else   
andre@3: 			/* set the is_disk thing inside the structure
andre@3: 			 */
andre@3: 			bp->is_disk = is_disk_mask;
andre@3: 			segp[segment_ndx] = bp;
andre@3: #endif
andre@3: 		  }
andre@3: 	} else {  
andre@3: 		BUF_REMOVE(bp);
andre@3: 		MRU_INSERT(bp);                 
andre@3: 	}
andre@3: 	return (bp);
andre@3: }
andre@3: 
andre@3: /*
andre@3:  * We need a buffer for this page. Either allocate one, or evict a resident
andre@3:  * one (if we have as many buffers as we're allowed) and put this one in.
andre@3:  *
andre@3:  * If newbuf finds an error (returning NULL), it also sets errno.
andre@3:  */
andre@3: static BUFHEAD *
andre@3: newbuf(HTAB *hashp, uint32 addr, BUFHEAD *prev_bp)
andre@3: {
andre@3: 	register BUFHEAD *bp;		/* The buffer we're going to use */
andre@3: 	register BUFHEAD *xbp;		/* Temp pointer */
andre@3: 	register BUFHEAD *next_xbp;
andre@3: 	SEGMENT segp;
andre@3: 	int segment_ndx;
andre@3: 	uint16 oaddr, *shortp;
andre@3: 
andre@3: 	oaddr = 0;
andre@3: 	bp = LRU;
andre@3: 	/*
andre@3: 	 * If LRU buffer is pinned, the buffer pool is too small. We need to
andre@3: 	 * allocate more buffers.
andre@3: 	 */
andre@3: 	if (hashp->nbufs || (bp->flags & BUF_PIN)) {
andre@3: 		/* Allocate a new one */
andre@3: 		if ((bp = (BUFHEAD *)malloc(sizeof(BUFHEAD))) == NULL)
andre@3: 			return (NULL);
andre@3: 
andre@3: 		/* this memset is supposedly unnecessary but lets add
andre@3: 		 * it anyways.
andre@3: 		 */
andre@3: 		memset(bp, 0xff, sizeof(BUFHEAD));
andre@3: 
andre@3: 		if ((bp->page = (char *)malloc((size_t)hashp->BSIZE)) == NULL) {
andre@3: 			free(bp);
andre@3: 			return (NULL);
andre@3: 		}
andre@3: 
andre@3: 		/* this memset is supposedly unnecessary but lets add
andre@3: 		 * it anyways.
andre@3: 		 */
andre@3: 		memset(bp->page, 0xff, (size_t)hashp->BSIZE);
andre@3: 
andre@3: 		if (hashp->nbufs)
andre@3: 			hashp->nbufs--;
andre@3: 	} else {
andre@3: 		/* Kick someone out */
andre@3: 		BUF_REMOVE(bp);
andre@3: 		/*
andre@3: 		 * If this is an overflow page with addr 0, it's already been
andre@3: 		 * flushed back in an overflow chain and initialized.
andre@3: 		 */
andre@3: 		if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) {
andre@3: 			/*
andre@3: 			 * Set oaddr before __put_page so that you get it
andre@3: 			 * before bytes are swapped.
andre@3: 			 */
andre@3: 			shortp = (uint16 *)bp->page;
andre@3: 			if (shortp[0])
andre@3: 			  {
andre@3: 				if(shortp[0] > (hashp->BSIZE / sizeof(uint16)))
andre@3: 				  {
andre@3: 					return(NULL);
andre@3: 				  }
andre@3: 				oaddr = shortp[shortp[0] - 1];
andre@3: 			  }
andre@3: 			if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page,
andre@3: 			    bp->addr, (int)IS_BUCKET(bp->flags), 0))
andre@3: 				return (NULL);
andre@3: 			/*
andre@3: 			 * Update the pointer to this page (i.e. invalidate it).
andre@3: 			 *
andre@3: 			 * If this is a new file (i.e. we created it at open
andre@3: 			 * time), make sure that we mark pages which have been
andre@3: 			 * written to disk so we retrieve them from disk later,
andre@3: 			 * rather than allocating new pages.
andre@3: 			 */
andre@3: 			if (IS_BUCKET(bp->flags)) {
andre@3: 				segment_ndx = bp->addr & (hashp->SGSIZE - 1);
andre@3: 				segp = hashp->dir[bp->addr >> hashp->SSHIFT];
andre@3: #ifdef DEBUG
andre@3: 				assert(segp != NULL);
andre@3: #endif
andre@3: 
andre@3: 				if (hashp->new_file &&
andre@3: 				    ((bp->flags & BUF_MOD) ||
andre@3: 				    ISDISK(segp[segment_ndx])))
andre@3: 					segp[segment_ndx] = (BUFHEAD *)BUF_DISK;
andre@3: 				else
andre@3: 					segp[segment_ndx] = NULL;
andre@3: 			}
andre@3: 			/*
andre@3: 			 * Since overflow pages can only be access by means of
andre@3: 			 * their bucket, free overflow pages associated with
andre@3: 			 * this bucket.
andre@3: 			 */
andre@3: 			for (xbp = bp; xbp->ovfl;) {
andre@3: 				next_xbp = xbp->ovfl;
andre@3: 				xbp->ovfl = 0;
andre@3: 				xbp = next_xbp;
andre@3: 
andre@3: 				/* leave pinned pages alone, we are still using
andre@3: 				 * them. */
andre@3: 				if (xbp->flags & BUF_PIN) {
andre@3: 					continue;
andre@3: 				}
andre@3: 
andre@3: 				/* Check that ovfl pointer is up date. */
andre@3: 				if (IS_BUCKET(xbp->flags) ||
andre@3: 				    (oaddr != xbp->addr))
andre@3: 					break;
andre@3: 
andre@3: 				shortp = (uint16 *)xbp->page;
andre@3: 				if (shortp[0])
andre@3: 				  {
andre@3: 					/* LJM is the number of reported
andre@3: 					 * pages way too much?
andre@3: 					 */
andre@3: 					if(shortp[0] > hashp->BSIZE/sizeof(uint16))
andre@3: 						return NULL;
andre@3: 					/* set before __put_page */
andre@3: 					oaddr = shortp[shortp[0] - 1];
andre@3: 				  }
andre@3: 				if ((xbp->flags & BUF_MOD) && __put_page(hashp,
andre@3: 				    xbp->page, xbp->addr, 0, 0))
andre@3: 					return (NULL);
andre@3: 				xbp->addr = 0;
andre@3: 				xbp->flags = 0;
andre@3: 				BUF_REMOVE(xbp);
andre@3: 				LRU_INSERT(xbp);
andre@3: 			}
andre@3: 		}
andre@3: 	}
andre@3: 
andre@3: 	/* Now assign this buffer */
andre@3: 	bp->addr = addr;
andre@3: #ifdef DEBUG1
andre@3: 	(void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n",
andre@3: 	    bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0);
andre@3: #endif
andre@3: 	bp->ovfl = NULL;
andre@3: 	if (prev_bp) {
andre@3: 		/*
andre@3: 		 * If prev_bp is set, this is an overflow page, hook it in to
andre@3: 		 * the buffer overflow links.
andre@3: 		 */
andre@3: #ifdef DEBUG1
andre@3: 		(void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n",
andre@3: 		    prev_bp->addr, (prev_bp->ovfl ? bp->ovfl->addr : 0),
andre@3: 		    (bp ? bp->addr : 0));
andre@3: #endif
andre@3: 		prev_bp->ovfl = bp;
andre@3: 		bp->flags = 0;
andre@3: 	} else
andre@3: 		bp->flags = BUF_BUCKET;
andre@3: 	MRU_INSERT(bp);
andre@3: 	return (bp);
andre@3: }
andre@3: 
andre@3: extern void __buf_init(HTAB *hashp, int32 nbytes)
andre@3: {
andre@3: 	BUFHEAD *bfp;
andre@3: 	int npages;
andre@3: 
andre@3: 	bfp = &(hashp->bufhead);
andre@3: 	npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT;
andre@3: 	npages = PR_MAX(npages, MIN_BUFFERS);
andre@3: 
andre@3: 	hashp->nbufs = npages;
andre@3: 	bfp->next = bfp;
andre@3: 	bfp->prev = bfp;
andre@3: 	/*
andre@3: 	 * This space is calloc'd so these are already null.
andre@3: 	 *
andre@3: 	 * bfp->ovfl = NULL;
andre@3: 	 * bfp->flags = 0;
andre@3: 	 * bfp->page = NULL;
andre@3: 	 * bfp->addr = 0;
andre@3: 	 */
andre@3: }
andre@3: 
andre@3: extern int
andre@3: __buf_free(HTAB *hashp, int do_free, int to_disk)
andre@3: {
andre@3: 	BUFHEAD *bp;
andre@3: 	int status = -1;
andre@3: 
andre@3: 	/* Need to make sure that buffer manager has been initialized */
andre@3: 	if (!LRU)
andre@3: 		return (0);
andre@3: 	for (bp = LRU; bp != &hashp->bufhead;) {
andre@3: 		/* Check that the buffer is valid */
andre@3: 		if (bp->addr || IS_BUCKET(bp->flags)) {
andre@3: 			if (to_disk && (bp->flags & BUF_MOD) &&
andre@3: 			    (status = __put_page(hashp, bp->page,
andre@3: 			    bp->addr, IS_BUCKET(bp->flags), 0))) {
andre@3: 			  
andre@3: 				if (do_free) {
andre@3: 					if (bp->page)
andre@3: 						free(bp->page);
andre@3: 					BUF_REMOVE(bp);
andre@3: 					free(bp);
andre@3: 				}
andre@3: 				
andre@3: 				return (status);
andre@3: 			}
andre@3: 		}
andre@3: 		/* Check if we are freeing stuff */
andre@3: 		if (do_free) {
andre@3: 			if (bp->page)
andre@3: 				free(bp->page);
andre@3: 			BUF_REMOVE(bp);
andre@3: 			free(bp);
andre@3: 			bp = LRU;
andre@3: 		} else
andre@3: 			bp = bp->prev;
andre@3: 	}
andre@3: 	return (0);
andre@3: }
andre@3: 
andre@3: extern void
andre@3: __reclaim_buf(HTAB *hashp, BUFHEAD *bp)
andre@3: {
andre@3: 	bp->ovfl = 0;
andre@3: 	bp->addr = 0;
andre@3: 	bp->flags = 0;
andre@3: 	BUF_REMOVE(bp);
andre@3: 	LRU_INSERT(bp);
andre@3: }