1 /*- 2 * Copyright (c) 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 #if defined(LIBC_SCCS) && !defined(lint) 35 static char sccsid[] = "@(#)mpool.c 8.7 (Berkeley) 11/2/95"; 36 #endif /* LIBC_SCCS and not lint */ 37 38 #include <sys/param.h> 39 #include <sys/stat.h> 40 41 #include <errno.h> 42 #include <stdio.h> 43 #include <stdlib.h> 44 #include <string.h> 45 #include <unistd.h> 46 47 #include "db-int.h" 48 #include "mpool.h" 49 50 static BKT *mpool_bkt __P((MPOOL *)); 51 static BKT *mpool_look __P((MPOOL *, db_pgno_t)); 52 static int mpool_write __P((MPOOL *, BKT *)); 53 54 /* 55 * mpool_open -- 56 * Initialize a memory pool. 57 */ 58 MPOOL * 59 mpool_open(void *key, int fd, db_pgno_t pagesize, db_pgno_t maxcache) 60 { 61 struct stat sb; 62 MPOOL *mp; 63 int entry; 64 65 /* 66 * Get information about the file. 67 * 68 * XXX 69 * We don't currently handle pipes, although we should. 70 */ 71 if (fstat(fd, &sb)) 72 return (NULL); 73 if (!S_ISREG(sb.st_mode)) { 74 errno = ESPIPE; 75 return (NULL); 76 } 77 78 /* Allocate and initialize the MPOOL cookie. */ 79 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 80 return (NULL); 81 TAILQ_INIT(&mp->lqh); 82 for (entry = 0; entry < HASHSIZE; ++entry) 83 TAILQ_INIT(&mp->hqh[entry]); 84 mp->maxcache = maxcache; 85 mp->npages = sb.st_size / pagesize; 86 mp->pagesize = pagesize; 87 mp->fd = fd; 88 return (mp); 89 } 90 91 /* 92 * mpool_filter -- 93 * Initialize input/output filters. 94 */ 95 void 96 mpool_filter(MPOOL *mp, void (*pgin) __P((void *, db_pgno_t, void *)), 97 void (*pgout) __P((void *, db_pgno_t, void *)), void *pgcookie) 98 { 99 mp->pgin = pgin; 100 mp->pgout = pgout; 101 mp->pgcookie = pgcookie; 102 } 103 104 /* 105 * mpool_new -- 106 * Get a new page of memory. 107 */ 108 void * 109 mpool_new(MPOOL *mp, db_pgno_t *pgnoaddr, u_int flags) 110 { 111 struct _hqh *head; 112 BKT *bp; 113 114 if (mp->npages == MAX_PAGE_NUMBER) { 115 (void)fprintf(stderr, "mpool_new: page allocation overflow.\n"); 116 abort(); 117 } 118 #ifdef STATISTICS 119 ++mp->pagenew; 120 #endif 121 /* 122 * Get a BKT from the cache. Assign a new page number, attach 123 * it to the head of the hash chain, the tail of the lru chain, 124 * and return. 125 */ 126 if ((bp = mpool_bkt(mp)) == NULL) 127 return (NULL); 128 if (flags == MPOOL_PAGE_REQUEST) { 129 mp->npages++; 130 bp->pgno = *pgnoaddr; 131 } else 132 bp->pgno = *pgnoaddr = mp->npages++; 133 134 bp->flags = MPOOL_PINNED | MPOOL_INUSE; 135 136 head = &mp->hqh[HASHKEY(bp->pgno)]; 137 TAILQ_INSERT_HEAD(head, bp, hq); 138 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 139 return (bp->page); 140 } 141 142 int 143 mpool_delete(MPOOL *mp, void *page) 144 { 145 struct _hqh *head; 146 BKT *bp; 147 148 bp = (void *)((char *)page - sizeof(BKT)); 149 150 #ifdef DEBUG 151 if (!(bp->flags & MPOOL_PINNED)) { 152 (void)fprintf(stderr, 153 "mpool_delete: page %d not pinned\n", bp->pgno); 154 abort(); 155 } 156 #endif 157 158 /* Remove from the hash and lru queues. */ 159 head = &mp->hqh[HASHKEY(bp->pgno)]; 160 TAILQ_REMOVE(head, bp, hq); 161 TAILQ_REMOVE(&mp->lqh, bp, q); 162 163 free(bp); 164 return (RET_SUCCESS); 165 } 166 167 /* 168 * mpool_get 169 * Get a page. 170 */ 171 void * 172 mpool_get(MPOOL *mp, db_pgno_t pgno, u_int flags) 173 { 174 struct _hqh *head; 175 BKT *bp; 176 off_t off; 177 int nr; 178 179 #ifdef STATISTICS 180 ++mp->pageget; 181 #endif 182 183 /* Check for a page that is cached. */ 184 if ((bp = mpool_look(mp, pgno)) != NULL) { 185 #ifdef DEBUG 186 if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) { 187 (void)fprintf(stderr, 188 "mpool_get: page %d already pinned\n", bp->pgno); 189 abort(); 190 } 191 #endif 192 /* 193 * Move the page to the head of the hash chain and the tail 194 * of the lru chain. 195 */ 196 head = &mp->hqh[HASHKEY(bp->pgno)]; 197 TAILQ_REMOVE(head, bp, hq); 198 TAILQ_INSERT_HEAD(head, bp, hq); 199 TAILQ_REMOVE(&mp->lqh, bp, q); 200 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 201 202 /* Return a pinned page. */ 203 if (!(flags & MPOOL_IGNOREPIN)) 204 bp->flags |= MPOOL_PINNED; 205 return (bp->page); 206 } 207 208 /* Get a page from the cache. */ 209 if ((bp = mpool_bkt(mp)) == NULL) 210 return (NULL); 211 212 /* Read in the contents. */ 213 #ifdef STATISTICS 214 ++mp->pageread; 215 #endif 216 off = mp->pagesize * pgno; 217 if (off / mp->pagesize != pgno) { 218 /* Run past the end of the file, or at least the part we 219 can address without large-file support? */ 220 errno = E2BIG; 221 return NULL; 222 } 223 if (lseek(mp->fd, off, SEEK_SET) != off) 224 return (NULL); 225 226 if ((nr = read(mp->fd, bp->page, mp->pagesize)) != 227 (ssize_t)mp->pagesize) { 228 if (nr > 0) { 229 /* A partial read is definitely bad. */ 230 errno = EINVAL; 231 return (NULL); 232 } else { 233 /* 234 * A zero-length reads, means you need to create a 235 * new page. 236 */ 237 memset(bp->page, 0, mp->pagesize); 238 } 239 } 240 241 /* Set the page number, pin the page. */ 242 bp->pgno = pgno; 243 if (!(flags & MPOOL_IGNOREPIN)) 244 bp->flags = MPOOL_PINNED; 245 bp->flags |= MPOOL_INUSE; 246 247 /* 248 * Add the page to the head of the hash chain and the tail 249 * of the lru chain. 250 */ 251 head = &mp->hqh[HASHKEY(bp->pgno)]; 252 TAILQ_INSERT_HEAD(head, bp, hq); 253 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 254 255 /* Run through the user's filter. */ 256 if (mp->pgin != NULL) 257 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 258 259 return (bp->page); 260 } 261 262 /* 263 * mpool_put 264 * Return a page. 265 */ 266 int 267 mpool_put(MPOOL *mp, void *page, u_int flags) 268 { 269 BKT *bp; 270 271 #ifdef STATISTICS 272 ++mp->pageput; 273 #endif 274 bp = (void *)((char *)page - sizeof(BKT)); 275 #ifdef DEBUG 276 if (!(bp->flags & MPOOL_PINNED)) { 277 (void)fprintf(stderr, 278 "mpool_put: page %d not pinned\n", bp->pgno); 279 abort(); 280 } 281 #endif 282 bp->flags &= ~MPOOL_PINNED; 283 if (flags & MPOOL_DIRTY) 284 bp->flags |= flags & MPOOL_DIRTY; 285 return (RET_SUCCESS); 286 } 287 288 /* 289 * mpool_close 290 * Close the buffer pool. 291 */ 292 int 293 mpool_close(MPOOL *mp) 294 { 295 BKT *bp; 296 297 /* Free up any space allocated to the lru pages. */ 298 while ((bp = mp->lqh.tqh_first) != NULL) { 299 TAILQ_REMOVE(&mp->lqh, mp->lqh.tqh_first, q); 300 free(bp); 301 } 302 303 /* Free the MPOOL cookie. */ 304 free(mp); 305 return (RET_SUCCESS); 306 } 307 308 /* 309 * mpool_sync 310 * Sync the pool to disk. 311 */ 312 int 313 mpool_sync(MPOOL *mp) 314 { 315 BKT *bp; 316 317 /* Walk the lru chain, flushing any dirty pages to disk. */ 318 for (bp = mp->lqh.tqh_first; bp != NULL; bp = bp->q.tqe_next) 319 if (bp->flags & MPOOL_DIRTY && 320 mpool_write(mp, bp) == RET_ERROR) 321 return (RET_ERROR); 322 323 /* Sync the file descriptor. */ 324 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 325 } 326 327 /* 328 * mpool_bkt 329 * Get a page from the cache (or create one). 330 */ 331 static BKT * 332 mpool_bkt(MPOOL *mp) 333 { 334 struct _hqh *head; 335 BKT *bp; 336 337 /* If under the max cached, always create a new page. */ 338 if (mp->curcache < mp->maxcache) 339 goto new; 340 341 /* 342 * If the cache is max'd out, walk the lru list for a buffer we 343 * can flush. If we find one, write it (if necessary) and take it 344 * off any lists. If we don't find anything we grow the cache anyway. 345 * The cache never shrinks. 346 */ 347 for (bp = mp->lqh.tqh_first; bp != NULL; bp = bp->q.tqe_next) 348 if (!(bp->flags & MPOOL_PINNED)) { 349 /* Flush if dirty. */ 350 if (bp->flags & MPOOL_DIRTY && 351 mpool_write(mp, bp) == RET_ERROR) 352 return (NULL); 353 #ifdef STATISTICS 354 ++mp->pageflush; 355 #endif 356 /* Remove from the hash and lru queues. */ 357 head = &mp->hqh[HASHKEY(bp->pgno)]; 358 TAILQ_REMOVE(head, bp, hq); 359 TAILQ_REMOVE(&mp->lqh, bp, q); 360 #if defined(DEBUG) && !defined(DEBUG_IDX0SPLIT) 361 { void *spage; 362 spage = bp->page; 363 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 364 bp->page = spage; 365 } 366 #endif 367 bp->flags = 0; 368 return (bp); 369 } 370 371 new: if ((bp = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL) 372 return (NULL); 373 #ifdef STATISTICS 374 ++mp->pagealloc; 375 #endif 376 #if defined(DEBUG) || defined(PURIFY) || 1 377 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 378 #endif 379 bp->page = (char *)bp + sizeof(BKT); 380 bp->flags = 0; 381 ++mp->curcache; 382 return (bp); 383 } 384 385 /* 386 * mpool_write 387 * Write a page to disk. 388 */ 389 static int 390 mpool_write(MPOOL *mp, BKT *bp) 391 { 392 off_t off; 393 394 #ifdef STATISTICS 395 ++mp->pagewrite; 396 #endif 397 398 /* Run through the user's filter. */ 399 if (mp->pgout) 400 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 401 402 off = mp->pagesize * bp->pgno; 403 if (off / mp->pagesize != bp->pgno) { 404 /* Run past the end of the file, or at least the part we 405 can address without large-file support? */ 406 errno = E2BIG; 407 return RET_ERROR; 408 } 409 if (lseek(mp->fd, off, SEEK_SET) != off) 410 return (RET_ERROR); 411 if (write(mp->fd, bp->page, mp->pagesize) != 412 (ssize_t)mp->pagesize) 413 return (RET_ERROR); 414 415 /* 416 * Re-run through the input filter since this page may soon be 417 * accessed via the cache, and whatever the user's output filter 418 * did may screw things up if we don't let the input filter 419 * restore the in-core copy. 420 */ 421 if (mp->pgin) 422 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 423 bp->flags &= ~MPOOL_DIRTY; 424 return (RET_SUCCESS); 425 } 426 427 /* 428 * mpool_look 429 * Lookup a page in the cache. 430 */ 431 static BKT * 432 mpool_look(MPOOL *mp, db_pgno_t pgno) 433 { 434 struct _hqh *head; 435 BKT *bp; 436 437 head = &mp->hqh[HASHKEY(pgno)]; 438 for (bp = head->tqh_first; bp != NULL; bp = bp->hq.tqe_next) 439 if ((bp->pgno == pgno) && (bp->flags & MPOOL_INUSE)) { 440 #ifdef STATISTICS 441 ++mp->cachehit; 442 #endif 443 return (bp); 444 } 445 #ifdef STATISTICS 446 ++mp->cachemiss; 447 #endif 448 return (NULL); 449 } 450 451 #ifdef STATISTICS 452 /* 453 * mpool_stat 454 * Print out cache statistics. 455 */ 456 void 457 mpool_stat(MPOOL *mp) 458 { 459 BKT *bp; 460 int cnt; 461 char *sep; 462 463 (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); 464 (void)fprintf(stderr, 465 "page size %lu, cacheing %lu pages of %lu page max cache\n", 466 mp->pagesize, mp->curcache, mp->maxcache); 467 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 468 mp->pageput, mp->pageget, mp->pagenew); 469 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 470 mp->pagealloc, mp->pageflush); 471 if (mp->cachehit + mp->cachemiss) 472 (void)fprintf(stderr, 473 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 474 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 475 * 100, mp->cachehit, mp->cachemiss); 476 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 477 mp->pageread, mp->pagewrite); 478 479 sep = ""; 480 cnt = 0; 481 for (bp = mp->lqh.tqh_first; bp != NULL; bp = bp->q.tqe_next) { 482 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 483 if (bp->flags & MPOOL_DIRTY) 484 (void)fprintf(stderr, "d"); 485 if (bp->flags & MPOOL_PINNED) 486 (void)fprintf(stderr, "P"); 487 if (++cnt == 10) { 488 sep = "\n"; 489 cnt = 0; 490 } else 491 sep = ", "; 492 493 } 494 (void)fprintf(stderr, "\n"); 495 } 496 #else 497 void 498 mpool_stat(MPOOL *mp) 499 { 500 } 501 #endif 502