1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Margo Seltzer. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /* 36 * PACKAGE: hash 37 * 38 * DESCRIPTION: 39 * Contains buffer management 40 * 41 * ROUTINES: 42 * External 43 * __buf_init 44 * __get_buf 45 * __buf_free 46 * __reclaim_buf 47 * Internal 48 * newbuf 49 */ 50 51 #include <sys/param.h> 52 53 #include <stddef.h> 54 #include <stdio.h> 55 #include <stdlib.h> 56 #include <string.h> 57 58 #ifdef DEBUG 59 #include <assert.h> 60 #endif 61 62 #include <db.h> 63 #include "hash.h" 64 #include "page.h" 65 #include "extern.h" 66 67 static BUFHEAD *newbuf(HTAB *, u_int32_t, BUFHEAD *); 68 69 /* Unlink B from its place in the lru */ 70 #define BUF_REMOVE(B) { \ 71 (B)->prev->next = (B)->next; \ 72 (B)->next->prev = (B)->prev; \ 73 } 74 75 /* Insert B after P */ 76 #define BUF_INSERT(B, P) { \ 77 (B)->next = (P)->next; \ 78 (B)->prev = (P); \ 79 (P)->next = (B); \ 80 (B)->next->prev = (B); \ 81 } 82 83 #define MRU hashp->bufhead.next 84 #define LRU hashp->bufhead.prev 85 86 #define MRU_INSERT(B) BUF_INSERT((B), &hashp->bufhead) 87 #define LRU_INSERT(B) BUF_INSERT((B), LRU) 88 89 /* 90 * We are looking for a buffer with address "addr". If prev_bp is NULL, then 91 * address is a bucket index. If prev_bp is not NULL, then it points to the 92 * page previous to an overflow page that we are trying to find. 93 * 94 * CAVEAT: The buffer header accessed via prev_bp's ovfl field may no longer 95 * be valid. Therefore, you must always verify that its address matches the 96 * address you are seeking. 97 */ 98 BUFHEAD * 99 __get_buf(HTAB *hashp, u_int32_t addr, 100 BUFHEAD *prev_bp, /* If prev_bp set, indicates a new overflow page. */ 101 int newpage) 102 { 103 int is_disk, segment_ndx, dir_ndx; 104 u_int32_t is_disk_mask; 105 SEGMENT segp; 106 BUFHEAD *bp; 107 108 is_disk = 0; 109 is_disk_mask = 0; 110 if (prev_bp) { 111 bp = prev_bp->ovfl; 112 if (!bp || (bp->addr != addr)) 113 bp = NULL; 114 if (!newpage) 115 is_disk = BUF_DISK; 116 } else { 117 /* Grab buffer out of directory */ 118 segment_ndx = addr & (hashp->SGSIZE - 1); 119 dir_ndx = addr >> hashp->SSHIFT; 120 if (dir_ndx >= hashp->nsegs) { 121 /* 122 * A bucket address could theoretically have been 123 * generated using maliciously crafted header values 124 * aimed at __call_hash. 125 */ 126 return (NULL); 127 } 128 /* valid segment ensured by __call_hash() */ 129 segp = hashp->dir[dir_ndx]; 130 #ifdef DEBUG 131 assert(segp != NULL); 132 #endif 133 bp = PTROF(segp[segment_ndx]); 134 is_disk_mask = ISDISK(segp[segment_ndx]); 135 is_disk = is_disk_mask || !hashp->new_file; 136 } 137 138 if (!bp) { 139 bp = newbuf(hashp, addr, prev_bp); 140 if (!bp || 141 __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0)) 142 return (NULL); 143 if (!prev_bp) 144 segp[segment_ndx] = 145 (BUFHEAD *)((intptr_t)bp | is_disk_mask); 146 } else { 147 BUF_REMOVE(bp); 148 MRU_INSERT(bp); 149 } 150 return (bp); 151 } 152 153 /* 154 * We need a buffer for this page. Either allocate one, or evict a resident 155 * one (if we have as many buffers as we're allowed) and put this one in. 156 * 157 * If newbuf finds an error (returning NULL), it also sets errno. 158 */ 159 static BUFHEAD * 160 newbuf(HTAB *hashp, u_int32_t addr, BUFHEAD *prev_bp) 161 { 162 BUFHEAD *bp; /* The buffer we're going to use */ 163 BUFHEAD *xbp; /* Temp pointer */ 164 BUFHEAD *next_xbp; 165 SEGMENT segp; 166 int segment_ndx; 167 u_int16_t oaddr, *shortp; 168 169 oaddr = 0; 170 bp = LRU; 171 172 /* It is bad to overwrite the page under the cursor. */ 173 if (bp == hashp->cpage) { 174 BUF_REMOVE(bp); 175 MRU_INSERT(bp); 176 bp = LRU; 177 } 178 179 /* If prev_bp is part of bp overflow, create a new buffer. */ 180 if (hashp->nbufs == 0 && prev_bp && bp->ovfl) { 181 BUFHEAD *ovfl; 182 183 for (ovfl = bp->ovfl; ovfl ; ovfl = ovfl->ovfl) { 184 if (ovfl == prev_bp) { 185 hashp->nbufs++; 186 break; 187 } 188 } 189 } 190 191 /* 192 * If LRU buffer is pinned, the buffer pool is too small. We need to 193 * allocate more buffers. 194 */ 195 if (hashp->nbufs || (bp->flags & BUF_PIN) || bp == hashp->cpage) { 196 /* Allocate a new one */ 197 if ((bp = (BUFHEAD *)calloc(1, sizeof(BUFHEAD))) == NULL) 198 return (NULL); 199 if ((bp->page = (char *)calloc(1, hashp->BSIZE)) == NULL) { 200 free(bp); 201 return (NULL); 202 } 203 if (hashp->nbufs) 204 hashp->nbufs--; 205 } else { 206 /* Kick someone out */ 207 BUF_REMOVE(bp); 208 /* 209 * If this is an overflow page with addr 0, it's already been 210 * flushed back in an overflow chain and initialized. 211 */ 212 if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) { 213 /* 214 * Set oaddr before __put_page so that you get it 215 * before bytes are swapped. 216 */ 217 shortp = (u_int16_t *)bp->page; 218 if (shortp[0]) 219 oaddr = shortp[shortp[0] - 1]; 220 if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page, 221 bp->addr, (int)IS_BUCKET(bp->flags), 0)) 222 return (NULL); 223 /* 224 * Update the pointer to this page (i.e. invalidate it). 225 * 226 * If this is a new file (i.e. we created it at open 227 * time), make sure that we mark pages which have been 228 * written to disk so we retrieve them from disk later, 229 * rather than allocating new pages. 230 */ 231 if (IS_BUCKET(bp->flags)) { 232 segment_ndx = bp->addr & (hashp->SGSIZE - 1); 233 segp = hashp->dir[bp->addr >> hashp->SSHIFT]; 234 #ifdef DEBUG 235 assert(segp != NULL); 236 #endif 237 238 if (hashp->new_file && 239 ((bp->flags & BUF_MOD) || 240 ISDISK(segp[segment_ndx]))) 241 segp[segment_ndx] = (BUFHEAD *)BUF_DISK; 242 else 243 segp[segment_ndx] = NULL; 244 } 245 /* 246 * Since overflow pages can only be access by means of 247 * their bucket, free overflow pages associated with 248 * this bucket. 249 */ 250 for (xbp = bp; xbp->ovfl;) { 251 next_xbp = xbp->ovfl; 252 xbp->ovfl = NULL; 253 xbp = next_xbp; 254 255 /* Check that ovfl pointer is up date. */ 256 if (IS_BUCKET(xbp->flags) || 257 (oaddr != xbp->addr)) 258 break; 259 260 shortp = (u_int16_t *)xbp->page; 261 if (shortp[0]) 262 /* set before __put_page */ 263 oaddr = shortp[shortp[0] - 1]; 264 if ((xbp->flags & BUF_MOD) && __put_page(hashp, 265 xbp->page, xbp->addr, 0, 0)) 266 return (NULL); 267 xbp->addr = 0; 268 xbp->flags = 0; 269 BUF_REMOVE(xbp); 270 LRU_INSERT(xbp); 271 } 272 } 273 } 274 275 /* Now assign this buffer */ 276 bp->addr = addr; 277 #ifdef DEBUG1 278 (void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n", 279 bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0); 280 #endif 281 bp->ovfl = NULL; 282 if (prev_bp) { 283 /* 284 * If prev_bp is set, this is an overflow page, hook it in to 285 * the buffer overflow links. 286 */ 287 #ifdef DEBUG1 288 (void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n", 289 prev_bp->addr, (prev_bp->ovfl ? prev_bp->ovfl->addr : 0), 290 (bp ? bp->addr : 0)); 291 #endif 292 prev_bp->ovfl = bp; 293 bp->flags = 0; 294 } else 295 bp->flags = BUF_BUCKET; 296 MRU_INSERT(bp); 297 return (bp); 298 } 299 300 void 301 __buf_init(HTAB *hashp, int nbytes) 302 { 303 BUFHEAD *bfp; 304 int npages; 305 306 bfp = &(hashp->bufhead); 307 npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT; 308 npages = MAX(npages, MIN_BUFFERS); 309 310 hashp->nbufs = npages; 311 bfp->next = bfp; 312 bfp->prev = bfp; 313 /* 314 * This space is calloc'd so these are already null. 315 * 316 * bfp->ovfl = NULL; 317 * bfp->flags = 0; 318 * bfp->page = NULL; 319 * bfp->addr = 0; 320 */ 321 } 322 323 int 324 __buf_free(HTAB *hashp, int do_free, int to_disk) 325 { 326 BUFHEAD *bp; 327 328 /* Need to make sure that buffer manager has been initialized */ 329 if (!LRU) 330 return (0); 331 for (bp = LRU; bp != &hashp->bufhead;) { 332 /* Check that the buffer is valid */ 333 if (bp->addr || IS_BUCKET(bp->flags)) { 334 if (to_disk && (bp->flags & BUF_MOD) && 335 __put_page(hashp, bp->page, 336 bp->addr, IS_BUCKET(bp->flags), 0)) 337 return (-1); 338 } 339 /* Check if we are freeing stuff */ 340 if (do_free) { 341 if (bp->page) { 342 (void)memset(bp->page, 0, hashp->BSIZE); 343 free(bp->page); 344 } 345 BUF_REMOVE(bp); 346 free(bp); 347 bp = LRU; 348 } else 349 bp = bp->prev; 350 } 351 return (0); 352 } 353 354 void 355 __reclaim_buf(HTAB *hashp, BUFHEAD *bp) 356 { 357 bp->ovfl = NULL; 358 bp->addr = 0; 359 bp->flags = 0; 360 BUF_REMOVE(bp); 361 LRU_INSERT(bp); 362 } 363