11573Srgrimes/*- 219029Speter * Copyright (c) 1990, 1993, 1994 31573Srgrimes * The Regents of the University of California. All rights reserved. 41573Srgrimes * 51573Srgrimes * This code is derived from software contributed to Berkeley by 61573Srgrimes * Margo Seltzer. 71573Srgrimes * 81573Srgrimes * Redistribution and use in source and binary forms, with or without 91573Srgrimes * modification, are permitted provided that the following conditions 101573Srgrimes * are met: 111573Srgrimes * 1. Redistributions of source code must retain the above copyright 121573Srgrimes * notice, this list of conditions and the following disclaimer. 131573Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 141573Srgrimes * notice, this list of conditions and the following disclaimer in the 151573Srgrimes * documentation and/or other materials provided with the distribution. 161573Srgrimes * 4. Neither the name of the University nor the names of its contributors 171573Srgrimes * may be used to endorse or promote products derived from this software 181573Srgrimes * without specific prior written permission. 191573Srgrimes * 201573Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 211573Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 221573Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 231573Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 241573Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 251573Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 261573Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 271573Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 281573Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 291573Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 301573Srgrimes * SUCH DAMAGE. 311573Srgrimes */ 321573Srgrimes 331573Srgrimes#if defined(LIBC_SCCS) && !defined(lint) 3419029Speterstatic char sccsid[] = "@(#)hash_buf.c 8.5 (Berkeley) 7/15/94"; 351573Srgrimes#endif /* LIBC_SCCS and not lint */ 3692889Sobrien#include <sys/cdefs.h> 3792889Sobrien__FBSDID("$FreeBSD$"); 381573Srgrimes 391573Srgrimes/* 401573Srgrimes * PACKAGE: hash 411573Srgrimes * 421573Srgrimes * DESCRIPTION: 431573Srgrimes * Contains buffer management 441573Srgrimes * 451573Srgrimes * ROUTINES: 461573Srgrimes * External 471573Srgrimes * __buf_init 481573Srgrimes * __get_buf 491573Srgrimes * __buf_free 501573Srgrimes * __reclaim_buf 511573Srgrimes * Internal 521573Srgrimes * newbuf 531573Srgrimes */ 541573Srgrimes 551573Srgrimes#include <sys/param.h> 561573Srgrimes 5719029Speter#include <stddef.h> 581573Srgrimes#include <stdio.h> 591573Srgrimes#include <stdlib.h> 60190482Sdelphij#include <string.h> 6119029Speter 621573Srgrimes#ifdef DEBUG 631573Srgrimes#include <assert.h> 641573Srgrimes#endif 651573Srgrimes 661573Srgrimes#include <db.h> 671573Srgrimes#include "hash.h" 681573Srgrimes#include "page.h" 691573Srgrimes#include "extern.h" 701573Srgrimes 7192905Sobrienstatic BUFHEAD *newbuf(HTAB *, u_int32_t, BUFHEAD *); 721573Srgrimes 731573Srgrimes/* Unlink B from its place in the lru */ 741573Srgrimes#define BUF_REMOVE(B) { \ 751573Srgrimes (B)->prev->next = (B)->next; \ 761573Srgrimes (B)->next->prev = (B)->prev; \ 771573Srgrimes} 781573Srgrimes 791573Srgrimes/* Insert B after P */ 801573Srgrimes#define BUF_INSERT(B, P) { \ 811573Srgrimes (B)->next = (P)->next; \ 821573Srgrimes (B)->prev = (P); \ 831573Srgrimes (P)->next = (B); \ 841573Srgrimes (B)->next->prev = (B); \ 851573Srgrimes} 861573Srgrimes 871573Srgrimes#define MRU hashp->bufhead.next 881573Srgrimes#define LRU hashp->bufhead.prev 891573Srgrimes 901573Srgrimes#define MRU_INSERT(B) BUF_INSERT((B), &hashp->bufhead) 911573Srgrimes#define LRU_INSERT(B) BUF_INSERT((B), LRU) 921573Srgrimes 931573Srgrimes/* 941573Srgrimes * We are looking for a buffer with address "addr". If prev_bp is NULL, then 951573Srgrimes * address is a bucket index. If prev_bp is not NULL, then it points to the 961573Srgrimes * page previous to an overflow page that we are trying to find. 971573Srgrimes * 981573Srgrimes * CAVEAT: The buffer header accessed via prev_bp's ovfl field may no longer 991573Srgrimes * be valid. Therefore, you must always verify that its address matches the 1001573Srgrimes * address you are seeking. 1011573Srgrimes */ 102189291SdelphijBUFHEAD * 103189291Sdelphij__get_buf(HTAB *hashp, u_int32_t addr, 104189291Sdelphij BUFHEAD *prev_bp, /* If prev_bp set, indicates a new overflow page. */ 105189291Sdelphij int newpage) 1061573Srgrimes{ 10792889Sobrien BUFHEAD *bp; 10892889Sobrien u_int32_t is_disk_mask; 10992889Sobrien int is_disk, segment_ndx; 1101573Srgrimes SEGMENT segp; 1111573Srgrimes 1121573Srgrimes is_disk = 0; 1131573Srgrimes is_disk_mask = 0; 1141573Srgrimes if (prev_bp) { 1151573Srgrimes bp = prev_bp->ovfl; 1161573Srgrimes if (!bp || (bp->addr != addr)) 1171573Srgrimes bp = NULL; 1181573Srgrimes if (!newpage) 1191573Srgrimes is_disk = BUF_DISK; 1201573Srgrimes } else { 1211573Srgrimes /* Grab buffer out of directory */ 1221573Srgrimes segment_ndx = addr & (hashp->SGSIZE - 1); 1231573Srgrimes 1241573Srgrimes /* valid segment ensured by __call_hash() */ 1251573Srgrimes segp = hashp->dir[addr >> hashp->SSHIFT]; 1261573Srgrimes#ifdef DEBUG 1271573Srgrimes assert(segp != NULL); 1281573Srgrimes#endif 1291573Srgrimes bp = PTROF(segp[segment_ndx]); 1301573Srgrimes is_disk_mask = ISDISK(segp[segment_ndx]); 1311573Srgrimes is_disk = is_disk_mask || !hashp->new_file; 1321573Srgrimes } 1331573Srgrimes 1341573Srgrimes if (!bp) { 1351573Srgrimes bp = newbuf(hashp, addr, prev_bp); 1361573Srgrimes if (!bp || 1371573Srgrimes __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0)) 1381573Srgrimes return (NULL); 1391573Srgrimes if (!prev_bp) 1401573Srgrimes segp[segment_ndx] = 14119029Speter (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask); 1421573Srgrimes } else { 1431573Srgrimes BUF_REMOVE(bp); 1441573Srgrimes MRU_INSERT(bp); 1451573Srgrimes } 1461573Srgrimes return (bp); 1471573Srgrimes} 1481573Srgrimes 1491573Srgrimes/* 1501573Srgrimes * We need a buffer for this page. Either allocate one, or evict a resident 1511573Srgrimes * one (if we have as many buffers as we're allowed) and put this one in. 1521573Srgrimes * 1531573Srgrimes * If newbuf finds an error (returning NULL), it also sets errno. 1541573Srgrimes */ 1551573Srgrimesstatic BUFHEAD * 156189291Sdelphijnewbuf(HTAB *hashp, u_int32_t addr, BUFHEAD *prev_bp) 1571573Srgrimes{ 15892889Sobrien BUFHEAD *bp; /* The buffer we're going to use */ 15992889Sobrien BUFHEAD *xbp; /* Temp pointer */ 16092889Sobrien BUFHEAD *next_xbp; 1611573Srgrimes SEGMENT segp; 1621573Srgrimes int segment_ndx; 16319029Speter u_int16_t oaddr, *shortp; 1641573Srgrimes 1651573Srgrimes oaddr = 0; 1661573Srgrimes bp = LRU; 167190492Sdelphij 168190492Sdelphij /* It is bad to overwrite the page under the cursor. */ 169190492Sdelphij if (bp == hashp->cpage) { 170190492Sdelphij BUF_REMOVE(bp); 171190492Sdelphij MRU_INSERT(bp); 172190492Sdelphij bp = LRU; 173190492Sdelphij } 174190492Sdelphij 175190492Sdelphij /* If prev_bp is part of bp overflow, create a new buffer. */ 176190492Sdelphij if (hashp->nbufs == 0 && prev_bp && bp->ovfl) { 177190492Sdelphij BUFHEAD *ovfl; 178190492Sdelphij 179190492Sdelphij for (ovfl = bp->ovfl; ovfl ; ovfl = ovfl->ovfl) { 180190492Sdelphij if (ovfl == prev_bp) { 181190492Sdelphij hashp->nbufs++; 182190492Sdelphij break; 183190492Sdelphij } 184190492Sdelphij } 185190492Sdelphij } 186190492Sdelphij 1871573Srgrimes /* 1881573Srgrimes * If LRU buffer is pinned, the buffer pool is too small. We need to 1891573Srgrimes * allocate more buffers. 1901573Srgrimes */ 191190492Sdelphij if (hashp->nbufs || (bp->flags & BUF_PIN) || bp == hashp->cpage) { 1921573Srgrimes /* Allocate a new one */ 193190482Sdelphij if ((bp = (BUFHEAD *)calloc(1, sizeof(BUFHEAD))) == NULL) 1941573Srgrimes return (NULL); 195190482Sdelphij if ((bp->page = (char *)calloc(1, hashp->BSIZE)) == NULL) { 1961573Srgrimes free(bp); 1971573Srgrimes return (NULL); 1981573Srgrimes } 1991573Srgrimes if (hashp->nbufs) 2001573Srgrimes hashp->nbufs--; 2011573Srgrimes } else { 2021573Srgrimes /* Kick someone out */ 2031573Srgrimes BUF_REMOVE(bp); 2041573Srgrimes /* 2051573Srgrimes * If this is an overflow page with addr 0, it's already been 2061573Srgrimes * flushed back in an overflow chain and initialized. 2071573Srgrimes */ 2081573Srgrimes if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) { 2091573Srgrimes /* 2101573Srgrimes * Set oaddr before __put_page so that you get it 2111573Srgrimes * before bytes are swapped. 2121573Srgrimes */ 21319029Speter shortp = (u_int16_t *)bp->page; 2141573Srgrimes if (shortp[0]) 2151573Srgrimes oaddr = shortp[shortp[0] - 1]; 2161573Srgrimes if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page, 2171573Srgrimes bp->addr, (int)IS_BUCKET(bp->flags), 0)) 2181573Srgrimes return (NULL); 2191573Srgrimes /* 2201573Srgrimes * Update the pointer to this page (i.e. invalidate it). 2211573Srgrimes * 2221573Srgrimes * If this is a new file (i.e. we created it at open 2231573Srgrimes * time), make sure that we mark pages which have been 2241573Srgrimes * written to disk so we retrieve them from disk later, 2251573Srgrimes * rather than allocating new pages. 2261573Srgrimes */ 2271573Srgrimes if (IS_BUCKET(bp->flags)) { 2281573Srgrimes segment_ndx = bp->addr & (hashp->SGSIZE - 1); 2291573Srgrimes segp = hashp->dir[bp->addr >> hashp->SSHIFT]; 2301573Srgrimes#ifdef DEBUG 2311573Srgrimes assert(segp != NULL); 2321573Srgrimes#endif 2331573Srgrimes 2341573Srgrimes if (hashp->new_file && 2351573Srgrimes ((bp->flags & BUF_MOD) || 2361573Srgrimes ISDISK(segp[segment_ndx]))) 2371573Srgrimes segp[segment_ndx] = (BUFHEAD *)BUF_DISK; 2381573Srgrimes else 2391573Srgrimes segp[segment_ndx] = NULL; 2401573Srgrimes } 2411573Srgrimes /* 2421573Srgrimes * Since overflow pages can only be access by means of 2431573Srgrimes * their bucket, free overflow pages associated with 2441573Srgrimes * this bucket. 2451573Srgrimes */ 2461573Srgrimes for (xbp = bp; xbp->ovfl;) { 2471573Srgrimes next_xbp = xbp->ovfl; 2481573Srgrimes xbp->ovfl = 0; 2491573Srgrimes xbp = next_xbp; 2501573Srgrimes 2511573Srgrimes /* Check that ovfl pointer is up date. */ 2521573Srgrimes if (IS_BUCKET(xbp->flags) || 2531573Srgrimes (oaddr != xbp->addr)) 2541573Srgrimes break; 2551573Srgrimes 25619029Speter shortp = (u_int16_t *)xbp->page; 2571573Srgrimes if (shortp[0]) 2581573Srgrimes /* set before __put_page */ 2591573Srgrimes oaddr = shortp[shortp[0] - 1]; 2601573Srgrimes if ((xbp->flags & BUF_MOD) && __put_page(hashp, 2611573Srgrimes xbp->page, xbp->addr, 0, 0)) 2621573Srgrimes return (NULL); 2631573Srgrimes xbp->addr = 0; 2641573Srgrimes xbp->flags = 0; 2651573Srgrimes BUF_REMOVE(xbp); 2661573Srgrimes LRU_INSERT(xbp); 2671573Srgrimes } 2681573Srgrimes } 2691573Srgrimes } 2701573Srgrimes 2711573Srgrimes /* Now assign this buffer */ 2721573Srgrimes bp->addr = addr; 2731573Srgrimes#ifdef DEBUG1 2741573Srgrimes (void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n", 2751573Srgrimes bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0); 2761573Srgrimes#endif 2771573Srgrimes bp->ovfl = NULL; 2781573Srgrimes if (prev_bp) { 2791573Srgrimes /* 2801573Srgrimes * If prev_bp is set, this is an overflow page, hook it in to 2811573Srgrimes * the buffer overflow links. 2821573Srgrimes */ 2831573Srgrimes#ifdef DEBUG1 2841573Srgrimes (void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n", 285190493Sdelphij prev_bp->addr, (prev_bp->ovfl ? prev_bp->ovfl->addr : 0), 2861573Srgrimes (bp ? bp->addr : 0)); 2871573Srgrimes#endif 2881573Srgrimes prev_bp->ovfl = bp; 2891573Srgrimes bp->flags = 0; 2901573Srgrimes } else 2911573Srgrimes bp->flags = BUF_BUCKET; 2921573Srgrimes MRU_INSERT(bp); 2931573Srgrimes return (bp); 2941573Srgrimes} 2951573Srgrimes 296189291Sdelphijvoid 297189291Sdelphij__buf_init(HTAB *hashp, int nbytes) 2981573Srgrimes{ 2991573Srgrimes BUFHEAD *bfp; 3001573Srgrimes int npages; 3011573Srgrimes 3021573Srgrimes bfp = &(hashp->bufhead); 3031573Srgrimes npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT; 3041573Srgrimes npages = MAX(npages, MIN_BUFFERS); 3051573Srgrimes 3061573Srgrimes hashp->nbufs = npages; 3071573Srgrimes bfp->next = bfp; 3081573Srgrimes bfp->prev = bfp; 3091573Srgrimes /* 3101573Srgrimes * This space is calloc'd so these are already null. 3111573Srgrimes * 3121573Srgrimes * bfp->ovfl = NULL; 3131573Srgrimes * bfp->flags = 0; 3141573Srgrimes * bfp->page = NULL; 3151573Srgrimes * bfp->addr = 0; 3161573Srgrimes */ 3171573Srgrimes} 3181573Srgrimes 319189291Sdelphijint 320189291Sdelphij__buf_free(HTAB *hashp, int do_free, int to_disk) 3211573Srgrimes{ 3221573Srgrimes BUFHEAD *bp; 3231573Srgrimes 3241573Srgrimes /* Need to make sure that buffer manager has been initialized */ 3251573Srgrimes if (!LRU) 3261573Srgrimes return (0); 3271573Srgrimes for (bp = LRU; bp != &hashp->bufhead;) { 3281573Srgrimes /* Check that the buffer is valid */ 3291573Srgrimes if (bp->addr || IS_BUCKET(bp->flags)) { 3301573Srgrimes if (to_disk && (bp->flags & BUF_MOD) && 3311573Srgrimes __put_page(hashp, bp->page, 3321573Srgrimes bp->addr, IS_BUCKET(bp->flags), 0)) 3331573Srgrimes return (-1); 3341573Srgrimes } 3351573Srgrimes /* Check if we are freeing stuff */ 3361573Srgrimes if (do_free) { 337190482Sdelphij if (bp->page) { 338190482Sdelphij (void)memset(bp->page, 0, hashp->BSIZE); 3391573Srgrimes free(bp->page); 340190482Sdelphij } 3411573Srgrimes BUF_REMOVE(bp); 3421573Srgrimes free(bp); 3431573Srgrimes bp = LRU; 3441573Srgrimes } else 3451573Srgrimes bp = bp->prev; 3461573Srgrimes } 3471573Srgrimes return (0); 3481573Srgrimes} 3491573Srgrimes 350189291Sdelphijvoid 351189291Sdelphij__reclaim_buf(HTAB *hashp, BUFHEAD *bp) 3521573Srgrimes{ 3531573Srgrimes bp->ovfl = 0; 3541573Srgrimes bp->addr = 0; 3551573Srgrimes bp->flags = 0; 3561573Srgrimes BUF_REMOVE(bp); 3571573Srgrimes LRU_INSERT(bp); 3581573Srgrimes} 359