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 * 3. 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 * @(#)hash_buf.c 8.5 (Berkeley) 7/15/94 33 * $FreeBSD: head/lib/libc/db/hash/hash_buf.c 190493 2009-03-28 06:40:48Z delphij $ 34 */ 35 36 /* 37 * PACKAGE: hash 38 * 39 * DESCRIPTION: 40 * Contains buffer management 41 * 42 * ROUTINES: 43 * External 44 * __buf_init 45 * __get_buf 46 * __buf_free 47 * __reclaim_buf 48 * Internal 49 * newbuf 50 */ 51 52 #include <sys/param.h> 53 54 #include <stddef.h> 55 #include <stdio.h> 56 #include <stdlib.h> 57 #include <string.h> 58 59 #ifdef DEBUG 60 #include <assert.h> 61 #endif 62 63 #include <db.h> 64 #include "hash.h" 65 #include "page.h" 66 #include "extern.h" 67 68 static BUFHEAD *newbuf(HTAB *, uint32_t, BUFHEAD *); 69 70 /* Unlink B from its place in the lru */ 71 #define BUF_REMOVE(B) { \ 72 (B)->prev->next = (B)->next; \ 73 (B)->next->prev = (B)->prev; \ 74 } 75 76 /* Insert B after P */ 77 #define BUF_INSERT(B, P) { \ 78 (B)->next = (P)->next; \ 79 (B)->prev = (P); \ 80 (P)->next = (B); \ 81 (B)->next->prev = (B); \ 82 } 83 84 #define MRU hashp->bufhead.next 85 #define LRU hashp->bufhead.prev 86 87 #define MRU_INSERT(B) BUF_INSERT((B), &hashp->bufhead) 88 #define LRU_INSERT(B) BUF_INSERT((B), LRU) 89 90 /* 91 * We are looking for a buffer with address "addr". If prev_bp is NULL, then 92 * address is a bucket index. If prev_bp is not NULL, then it points to the 93 * page previous to an overflow page that we are trying to find. 94 * 95 * CAVEAT: The buffer header accessed via prev_bp's ovfl field may no longer 96 * be valid. Therefore, you must always verify that its address matches the 97 * address you are seeking. 98 */ 99 BUFHEAD * 100 __get_buf(HTAB *hashp, uint32_t addr, 101 BUFHEAD *prev_bp, /* If prev_bp set, indicates a new overflow page. */ 102 int newpage) 103 { 104 BUFHEAD *bp; 105 uint32_t is_disk_mask; 106 int is_disk, segment_ndx; 107 SEGMENT segp; 108 109 is_disk = 0; 110 is_disk_mask = 0; 111 if (prev_bp) { 112 bp = prev_bp->ovfl; 113 if (!bp || (bp->addr != addr)) 114 bp = NULL; 115 if (!newpage) 116 is_disk = BUF_DISK; 117 } else { 118 /* Grab buffer out of directory */ 119 segment_ndx = addr & (hashp->SGSIZE - 1); 120 121 /* valid segment ensured by __call_hash() */ 122 segp = hashp->dir[addr >> hashp->SSHIFT]; 123 #ifdef DEBUG 124 assert(segp != NULL); 125 #endif 126 bp = PTROF(segp[segment_ndx]); 127 is_disk_mask = ISDISK(segp[segment_ndx]); 128 is_disk = is_disk_mask || !hashp->new_file; 129 } 130 131 if (!bp) { 132 bp = newbuf(hashp, addr, prev_bp); 133 if (!bp || 134 __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0)) 135 return (NULL); 136 if (!prev_bp) 137 segp[segment_ndx] = 138 (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask); 139 } else { 140 BUF_REMOVE(bp); 141 MRU_INSERT(bp); 142 } 143 return (bp); 144 } 145 146 /* 147 * We need a buffer for this page. Either allocate one, or evict a resident 148 * one (if we have as many buffers as we're allowed) and put this one in. 149 * 150 * If newbuf finds an error (returning NULL), it also sets errno. 151 */ 152 static BUFHEAD * 153 newbuf(HTAB *hashp, uint32_t addr, BUFHEAD *prev_bp) 154 { 155 BUFHEAD *bp; /* The buffer we're going to use */ 156 BUFHEAD *xbp; /* Temp pointer */ 157 BUFHEAD *next_xbp; 158 SEGMENT segp; 159 int segment_ndx; 160 uint16_t oaddr, *shortp; 161 162 oaddr = 0; 163 bp = LRU; 164 165 /* It is bad to overwrite the page under the cursor. */ 166 if (bp == hashp->cpage) { 167 BUF_REMOVE(bp); 168 MRU_INSERT(bp); 169 bp = LRU; 170 } 171 172 /* If prev_bp is part of bp overflow, create a new buffer. */ 173 if (hashp->nbufs == 0 && prev_bp && bp->ovfl) { 174 BUFHEAD *ovfl; 175 176 for (ovfl = bp->ovfl; ovfl ; ovfl = ovfl->ovfl) { 177 if (ovfl == prev_bp) { 178 hashp->nbufs++; 179 break; 180 } 181 } 182 } 183 184 /* 185 * If LRU buffer is pinned, the buffer pool is too small. We need to 186 * allocate more buffers. 187 */ 188 if (hashp->nbufs || (bp->flags & BUF_PIN) || bp == hashp->cpage) { 189 /* Allocate a new one */ 190 if ((bp = (BUFHEAD *)calloc(1, sizeof(BUFHEAD))) == NULL) 191 return (NULL); 192 if ((bp->page = (char *)calloc(1, hashp->BSIZE)) == NULL) { 193 free(bp); 194 return (NULL); 195 } 196 if (hashp->nbufs) 197 hashp->nbufs--; 198 } else { 199 /* Kick someone out */ 200 BUF_REMOVE(bp); 201 /* 202 * If this is an overflow page with addr 0, it's already been 203 * flushed back in an overflow chain and initialized. 204 */ 205 if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) { 206 /* 207 * Set oaddr before __put_page so that you get it 208 * before bytes are swapped. 209 */ 210 shortp = (uint16_t *)bp->page; 211 if (shortp[0]) 212 oaddr = shortp[shortp[0] - 1]; 213 if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page, 214 bp->addr, (int)IS_BUCKET(bp->flags), 0)) 215 return (NULL); 216 /* 217 * Update the pointer to this page (i.e. invalidate it). 218 * 219 * If this is a new file (i.e. we created it at open 220 * time), make sure that we mark pages which have been 221 * written to disk so we retrieve them from disk later, 222 * rather than allocating new pages. 223 */ 224 if (IS_BUCKET(bp->flags)) { 225 segment_ndx = bp->addr & (hashp->SGSIZE - 1); 226 segp = hashp->dir[bp->addr >> hashp->SSHIFT]; 227 #ifdef DEBUG 228 assert(segp != NULL); 229 #endif 230 231 if (hashp->new_file && 232 ((bp->flags & BUF_MOD) || 233 ISDISK(segp[segment_ndx]))) 234 segp[segment_ndx] = (BUFHEAD *)BUF_DISK; 235 else 236 segp[segment_ndx] = NULL; 237 } 238 /* 239 * Since overflow pages can only be access by means of 240 * their bucket, free overflow pages associated with 241 * this bucket. 242 */ 243 for (xbp = bp; xbp->ovfl;) { 244 next_xbp = xbp->ovfl; 245 xbp->ovfl = 0; 246 xbp = next_xbp; 247 248 /* Check that ovfl pointer is up date. */ 249 if (IS_BUCKET(xbp->flags) || 250 (oaddr != xbp->addr)) 251 break; 252 253 shortp = (uint16_t *)xbp->page; 254 if (shortp[0]) 255 /* set before __put_page */ 256 oaddr = shortp[shortp[0] - 1]; 257 if ((xbp->flags & BUF_MOD) && __put_page(hashp, 258 xbp->page, xbp->addr, 0, 0)) 259 return (NULL); 260 xbp->addr = 0; 261 xbp->flags = 0; 262 BUF_REMOVE(xbp); 263 LRU_INSERT(xbp); 264 } 265 } 266 } 267 268 /* Now assign this buffer */ 269 bp->addr = addr; 270 #ifdef DEBUG1 271 fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n", 272 bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0); 273 #endif 274 bp->ovfl = NULL; 275 if (prev_bp) { 276 /* 277 * If prev_bp is set, this is an overflow page, hook it in to 278 * the buffer overflow links. 279 */ 280 #ifdef DEBUG1 281 fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n", 282 prev_bp->addr, (prev_bp->ovfl ? prev_bp->ovfl->addr : 0), 283 (bp ? bp->addr : 0)); 284 #endif 285 prev_bp->ovfl = bp; 286 bp->flags = 0; 287 } else 288 bp->flags = BUF_BUCKET; 289 MRU_INSERT(bp); 290 return (bp); 291 } 292 293 void 294 __buf_init(HTAB *hashp, int nbytes) 295 { 296 BUFHEAD *bfp; 297 int npages; 298 299 bfp = &(hashp->bufhead); 300 npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT; 301 npages = MAX(npages, MIN_BUFFERS); 302 303 hashp->nbufs = npages; 304 bfp->next = bfp; 305 bfp->prev = bfp; 306 /* 307 * This space is calloc'd so these are already null. 308 * 309 * bfp->ovfl = NULL; 310 * bfp->flags = 0; 311 * bfp->page = NULL; 312 * bfp->addr = 0; 313 */ 314 } 315 316 int 317 __buf_free(HTAB *hashp, int do_free, int to_disk) 318 { 319 BUFHEAD *bp; 320 321 /* Need to make sure that buffer manager has been initialized */ 322 if (!LRU) 323 return (0); 324 for (bp = LRU; bp != &hashp->bufhead;) { 325 /* Check that the buffer is valid */ 326 if (bp->addr || IS_BUCKET(bp->flags)) { 327 if (to_disk && (bp->flags & BUF_MOD) && 328 __put_page(hashp, bp->page, 329 bp->addr, IS_BUCKET(bp->flags), 0)) 330 return (-1); 331 } 332 /* Check if we are freeing stuff */ 333 if (do_free) { 334 if (bp->page) { 335 memset(bp->page, 0, hashp->BSIZE); 336 free(bp->page); 337 } 338 BUF_REMOVE(bp); 339 free(bp); 340 bp = LRU; 341 } else 342 bp = bp->prev; 343 } 344 return (0); 345 } 346 347 void 348 __reclaim_buf(HTAB *hashp, BUFHEAD *bp) 349 { 350 bp->ovfl = 0; 351 bp->addr = 0; 352 bp->flags = 0; 353 BUF_REMOVE(bp); 354 LRU_INSERT(bp); 355 } 356