db/hash/FreeBSD/hash_buf.c

   1 /*-
   2  * Copyright (c) 1990, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Margo Seltzer.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 4. Neither the name of the University nor the names of its contributors
  17  *    may be used to endorse or promote products derived from this software
  18  *    without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  30  * SUCH DAMAGE.
  31  */
  32
  33 #if defined(LIBC_SCCS) && !defined(lint)
  34 static char sccsid[] = "@(#)hash_buf.c  8.5 (Berkeley) 7/15/94";
  35 #endif /* LIBC_SCCS and not lint */
  36 #include <sys/cdefs.h>
  37 __FBSDID("$FreeBSD: src/lib/libc/db/hash/hash_buf.c,v 1.12 2009/03/28 06:40:48 delphij Exp $");
  38
  39 /*
  40  * PACKAGE: hash
  41  *
  42  * DESCRIPTION:
  43  *      Contains buffer management
  44  *
  45  * ROUTINES:
  46  * External
  47  *      __buf_init
  48  *      __get_buf
  49  *      __buf_free
  50  *      __reclaim_buf
  51  * Internal
  52  *      newbuf
  53  */
  54
  55 #include <sys/param.h>
  56
  57 #include <stddef.h>
  58 #include <stdio.h>
  59 #include <stdlib.h>
  60 #include <string.h>
  61
  62 #ifdef DEBUG
  63 #include <assert.h>
  64 #endif
  65
  66 #include <db.h>
  67 #include "hash.h"
  68 #include "page.h"
  69 #include "hash_extern.h"
  70
  71 static BUFHEAD *newbuf(HTAB *, u_int32_t, BUFHEAD *);
  72
  73 /* Unlink B from its place in the lru */
  74 #define BUF_REMOVE(B) { \
  75         (B)->prev->next = (B)->next; \
  76         (B)->next->prev = (B)->prev; \
  77 }
  78
  79 /* Insert B after P */
  80 #define BUF_INSERT(B, P) { \
  81         (B)->next = (P)->next; \
  82         (B)->prev = (P); \
  83         (P)->next = (B); \
  84         (B)->next->prev = (B); \
  85 }
  86
  87 #define MRU     hashp->bufhead.next
  88 #define LRU     hashp->bufhead.prev
  89
  90 #define MRU_INSERT(B)   BUF_INSERT((B), &hashp->bufhead)
  91 #define LRU_INSERT(B)   BUF_INSERT((B), LRU)
  92
  93 /*
  94  * We are looking for a buffer with address "addr".  If prev_bp is NULL, then
  95  * address is a bucket index.  If prev_bp is not NULL, then it points to the
  96  * page previous to an overflow page that we are trying to find.
  97  *
  98  * CAVEAT:  The buffer header accessed via prev_bp's ovfl field may no longer
  99  * be valid.  Therefore, you must always verify that its address matches the
 100  * address you are seeking.
 101  */
 102 BUFHEAD *
 103 __get_buf(HTAB *hashp, u_int32_t addr,
 104     BUFHEAD *prev_bp,   /* If prev_bp set, indicates a new overflow page. */
 105     int newpage)
 106 {
 107         BUFHEAD *bp;
 108         u_int32_t is_disk_mask;
 109         int is_disk, segment_ndx;
 110         SEGMENT segp;
 111
 112         is_disk = 0;
 113         is_disk_mask = 0;
 114         if (prev_bp) {
 115                 bp = prev_bp->ovfl;
 116                 if (!bp || (bp->addr != addr))
 117                         bp = NULL;
 118                 if (!newpage)
 119                         is_disk = BUF_DISK;
 120         } else {
 121                 /* Grab buffer out of directory */
 122                 segment_ndx = addr & (hashp->SGSIZE - 1);
 123
 124                 /* valid segment ensured by __call_hash() */
 125                 segp = hashp->dir[addr >> hashp->SSHIFT];
 126 #ifdef DEBUG
 127                 assert(segp != NULL);
 128 #endif
 129                 bp = PTROF(segp[segment_ndx]);
 130                 is_disk_mask = ISDISK(segp[segment_ndx]);
 131                 is_disk = is_disk_mask || !hashp->new_file;
 132         }
 133
 134         if (!bp) {
 135                 bp = newbuf(hashp, addr, prev_bp);
 136                 if (!bp ||
 137                     __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0))
 138                         return (NULL);
 139                 if (!prev_bp)
 140                         segp[segment_ndx] =
 141                             (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask);
 142         } else {
 143                 BUF_REMOVE(bp);
 144                 MRU_INSERT(bp);
 145         }
 146         return (bp);
 147 }
 148
 149 /*
 150  * We need a buffer for this page. Either allocate one, or evict a resident
 151  * one (if we have as many buffers as we're allowed) and put this one in.
 152  *
 153  * If newbuf finds an error (returning NULL), it also sets errno.
 154  */
 155 static BUFHEAD *
 156 newbuf(HTAB *hashp, u_int32_t addr, BUFHEAD *prev_bp)
 157 {
 158         BUFHEAD *bp;            /* The buffer we're going to use */
 159         BUFHEAD *xbp;           /* Temp pointer */
 160         BUFHEAD *next_xbp;
 161         SEGMENT segp;
 162         int segment_ndx;
 163         u_int16_t oaddr, *shortp;
 164
 165         oaddr = 0;
 166         bp = LRU;
 167
 168         /* It is bad to overwrite the page under the cursor. */
 169         if (bp == hashp->cpage) {
 170                 BUF_REMOVE(bp);
 171                 MRU_INSERT(bp);
 172                 bp = LRU;
 173         }
 174
 175         /* If prev_bp is part of bp overflow, create a new buffer. */
 176         if (hashp->nbufs == 0 && prev_bp && bp->ovfl) {
 177                 BUFHEAD *ovfl;
 178
 179                 for (ovfl = bp->ovfl; ovfl ; ovfl = ovfl->ovfl) {
 180                         if (ovfl == prev_bp) {
 181                                 hashp->nbufs++;
 182                                 break;
 183                         }
 184                 }
 185         }
 186
 187         /*
 188          * If LRU buffer is pinned, the buffer pool is too small. We need to
 189          * allocate more buffers.
 190          */
 191         if (hashp->nbufs || (bp->flags & BUF_PIN) || bp == hashp->cpage) {
 192                 /* Allocate a new one */
 193                 if ((bp = (BUFHEAD *)calloc(1, sizeof(BUFHEAD))) == NULL)
 194                         return (NULL);
 195                 if ((bp->page = (char *)calloc(1, hashp->BSIZE)) == NULL) {
 196                         free(bp);
 197                         return (NULL);
 198                 }
 199                 if (hashp->nbufs)
 200                         hashp->nbufs--;
 201         } else {
 202                 /* Kick someone out */
 203                 BUF_REMOVE(bp);
 204                 /*
 205                  * If this is an overflow page with addr 0, it's already been
 206                  * flushed back in an overflow chain and initialized.
 207                  */
 208                 if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) {
 209                         /*
 210                          * Set oaddr before __put_page so that you get it
 211                          * before bytes are swapped.
 212                          */
 213                         shortp = (u_int16_t *)bp->page;
 214                         if (shortp[0])
 215                                 oaddr = shortp[shortp[0] - 1];
 216                         if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page,
 217                             bp->addr, (int)IS_BUCKET(bp->flags), 0))
 218                                 return (NULL);
 219                         /*
 220                          * Update the pointer to this page (i.e. invalidate it).
 221                          *
 222                          * If this is a new file (i.e. we created it at open
 223                          * time), make sure that we mark pages which have been
 224                          * written to disk so we retrieve them from disk later,
 225                          * rather than allocating new pages.
 226                          */
 227                         if (IS_BUCKET(bp->flags)) {
 228                                 segment_ndx = bp->addr & (hashp->SGSIZE - 1);
 229                                 segp = hashp->dir[bp->addr >> hashp->SSHIFT];
 230 #ifdef DEBUG
 231                                 assert(segp != NULL);
 232 #endif
 233
 234                                 if (hashp->new_file &&
 235                                     ((bp->flags & BUF_MOD) ||
 236                                     ISDISK(segp[segment_ndx])))
 237                                         segp[segment_ndx] = (BUFHEAD *)BUF_DISK;
 238                                 else
 239                                         segp[segment_ndx] = NULL;
 240                         }
 241                         /*
 242                          * Since overflow pages can only be access by means of
 243                          * their bucket, free overflow pages associated with
 244                          * this bucket.
 245                          */
 246                         for (xbp = bp; xbp->ovfl;) {
 247                                 next_xbp = xbp->ovfl;
 248                                 xbp->ovfl = 0;
 249                                 xbp = next_xbp;
 250
 251                                 /* Check that ovfl pointer is up date. */
 252                                 if (IS_BUCKET(xbp->flags) ||
 253                                     (oaddr != xbp->addr))
 254                                         break;
 255
 256                                 shortp = (u_int16_t *)xbp->page;
 257                                 if (shortp[0])
 258                                         /* set before __put_page */
 259                                         oaddr = shortp[shortp[0] - 1];
 260                                 if ((xbp->flags & BUF_MOD) && __put_page(hashp,
 261                                     xbp->page, xbp->addr, 0, 0))
 262                                         return (NULL);
 263                                 xbp->addr = 0;
 264                                 xbp->flags = 0;
 265                                 BUF_REMOVE(xbp);
 266                                 LRU_INSERT(xbp);
 267                         }
 268                 }
 269         }
 270
 271         /* Now assign this buffer */
 272         bp->addr = addr;
 273 #ifdef DEBUG1
 274         (void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n",
 275             bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0);
 276 #endif
 277         bp->ovfl = NULL;
 278         if (prev_bp) {
 279                 /*
 280                  * If prev_bp is set, this is an overflow page, hook it in to
 281                  * the buffer overflow links.
 282                  */
 283 #ifdef DEBUG1
 284                 (void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n",
 285                     prev_bp->addr, (prev_bp->ovfl ? prev_bp->ovfl->addr : 0),
 286                     (bp ? bp->addr : 0));
 287 #endif
 288                 prev_bp->ovfl = bp;
 289                 bp->flags = 0;
 290         } else
 291                 bp->flags = BUF_BUCKET;
 292         MRU_INSERT(bp);
 293         return (bp);
 294 }
 295
 296 __private_extern__ void
 297 __buf_init(HTAB *hashp, int nbytes)
 298 {
 299         BUFHEAD *bfp;
 300         int npages;
 301
 302         bfp = &(hashp->bufhead);
 303         npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT;
 304         npages = MAX(npages, MIN_BUFFERS);
 305
 306         hashp->nbufs = npages;
 307         bfp->next = bfp;
 308         bfp->prev = bfp;
 309         /*
 310          * This space is calloc'd so these are already null.
 311          *
 312          * bfp->ovfl = NULL;
 313          * bfp->flags = 0;
 314          * bfp->page = NULL;
 315          * bfp->addr = 0;
 316          */
 317 }
 318
 319 int
 320 __buf_free(HTAB *hashp, int do_free, int to_disk)
 321 {
 322         BUFHEAD *bp;
 323
 324         /* Need to make sure that buffer manager has been initialized */
 325         if (!LRU)
 326                 return (0);
 327         for (bp = LRU; bp != &hashp->bufhead;) {
 328                 /* Check that the buffer is valid */
 329                 if (bp->addr || IS_BUCKET(bp->flags)) {
 330                         if (to_disk && (bp->flags & BUF_MOD) &&
 331                             __put_page(hashp, bp->page,
 332                             bp->addr, IS_BUCKET(bp->flags), 0))
 333                                 return (-1);
 334                 }
 335                 /* Check if we are freeing stuff */
 336                 if (do_free) {
 337                         if (bp->page) {
 338                                 (void)memset(bp->page, 0, hashp->BSIZE);
 339                                 free(bp->page);
 340                         }
 341                         BUF_REMOVE(bp);
 342                         free(bp);
 343                         bp = LRU;
 344                 } else
 345                         bp = bp->prev;
 346         }
 347         return (0);
 348 }
 349
 350 void
 351 __reclaim_buf(HTAB *hashp, BUFHEAD *bp)
 352 {
 353         bp->ovfl = 0;
 354         bp->addr = 0;
 355         bp->flags = 0;
 356         BUF_REMOVE(bp);
 357         LRU_INSERT(bp);
 358 }