1 #pragma ident "%Z%%M% %I% %E% SMI" 2 3 /*- 4 * Copyright (c) 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #if defined(LIBC_SCCS) && !defined(lint) 37 static char sccsid[] = "@(#)mpool.c 8.7 (Berkeley) 11/2/95"; 38 #endif /* LIBC_SCCS and not lint */ 39 40 #include <sys/param.h> 41 #include <sys/stat.h> 42 43 #include <errno.h> 44 #include <stdio.h> 45 #include <stdlib.h> 46 #include <string.h> 47 #include <unistd.h> 48 49 #include "db-int.h" 50 #include "mpool.h" 51 52 static BKT *mpool_bkt __P((MPOOL *)); 53 static BKT *mpool_look __P((MPOOL *, db_pgno_t)); 54 static int mpool_write __P((MPOOL *, BKT *)); 55 56 /* 57 * mpool_open -- 58 * Initialize a memory pool. 59 */ 60 MPOOL * 61 mpool_open(key, fd, pagesize, maxcache) 62 void *key; 63 int fd; 64 db_pgno_t pagesize, maxcache; 65 { 66 struct stat sb; 67 MPOOL *mp; 68 int entry; 69 70 /* 71 * Get information about the file. 72 * 73 * XXX 74 * We don't currently handle pipes, although we should. 75 */ 76 if (fstat(fd, &sb)) 77 return (NULL); 78 if (!S_ISREG(sb.st_mode)) { 79 errno = ESPIPE; 80 return (NULL); 81 } 82 83 /* Allocate and initialize the MPOOL cookie. */ 84 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 85 return (NULL); 86 CIRCLEQ_INIT(&mp->lqh); 87 for (entry = 0; entry < HASHSIZE; ++entry) 88 CIRCLEQ_INIT(&mp->hqh[entry]); 89 mp->maxcache = maxcache; 90 mp->npages = sb.st_size / pagesize; 91 mp->pagesize = pagesize; 92 mp->fd = fd; 93 return (mp); 94 } 95 96 /* 97 * mpool_filter -- 98 * Initialize input/output filters. 99 */ 100 void 101 mpool_filter(mp, pgin, pgout, pgcookie) 102 MPOOL *mp; 103 void (*pgin) __P((void *, db_pgno_t, void *)); 104 void (*pgout) __P((void *, db_pgno_t, void *)); 105 void *pgcookie; 106 { 107 mp->pgin = pgin; 108 mp->pgout = pgout; 109 mp->pgcookie = pgcookie; 110 } 111 112 /* 113 * mpool_new -- 114 * Get a new page of memory. 115 */ 116 void * 117 mpool_new(mp, pgnoaddr, flags) 118 MPOOL *mp; 119 db_pgno_t *pgnoaddr; 120 u_int flags; 121 { 122 struct _hqh *head; 123 BKT *bp; 124 125 if (mp->npages == MAX_PAGE_NUMBER) { 126 (void)fprintf(stderr, "mpool_new: page allocation overflow.\n"); 127 abort(); 128 } 129 #ifdef STATISTICS 130 ++mp->pagenew; 131 #endif 132 /* 133 * Get a BKT from the cache. Assign a new page number, attach 134 * it to the head of the hash chain, the tail of the lru chain, 135 * and return. 136 */ 137 if ((bp = mpool_bkt(mp)) == NULL) 138 return (NULL); 139 if (flags == MPOOL_PAGE_REQUEST) { 140 mp->npages++; 141 bp->pgno = *pgnoaddr; 142 } else 143 bp->pgno = *pgnoaddr = mp->npages++; 144 145 bp->flags = MPOOL_PINNED | MPOOL_INUSE; 146 147 head = &mp->hqh[HASHKEY(bp->pgno)]; 148 CIRCLEQ_INSERT_HEAD(head, bp, hq); 149 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 150 return (bp->page); 151 } 152 153 int 154 mpool_delete(mp, page) 155 MPOOL *mp; 156 void *page; 157 { 158 struct _hqh *head; 159 BKT *bp; 160 161 bp = (BKT *)((char *)page - sizeof(BKT)); 162 163 #ifdef DEBUG 164 if (!(bp->flags & MPOOL_PINNED)) { 165 (void)fprintf(stderr, 166 "mpool_delete: page %d not pinned\n", bp->pgno); 167 abort(); 168 } 169 #endif 170 171 /* Remove from the hash and lru queues. */ 172 head = &mp->hqh[HASHKEY(bp->pgno)]; 173 CIRCLEQ_REMOVE(head, bp, hq); 174 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 175 176 free(bp); 177 return (RET_SUCCESS); 178 } 179 180 /* 181 * mpool_get 182 * Get a page. 183 */ 184 void * 185 mpool_get(mp, pgno, flags) 186 MPOOL *mp; 187 db_pgno_t pgno; 188 u_int flags; /* XXX not used? */ 189 { 190 struct _hqh *head; 191 BKT *bp; 192 off_t off; 193 int nr; 194 195 #ifdef STATISTICS 196 ++mp->pageget; 197 #endif 198 199 /* Check for a page that is cached. */ 200 if ((bp = mpool_look(mp, pgno)) != NULL) { 201 #ifdef DEBUG 202 if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) { 203 (void)fprintf(stderr, 204 "mpool_get: page %d already pinned\n", bp->pgno); 205 abort(); 206 } 207 #endif 208 /* 209 * Move the page to the head of the hash chain and the tail 210 * of the lru chain. 211 */ 212 head = &mp->hqh[HASHKEY(bp->pgno)]; 213 CIRCLEQ_REMOVE(head, bp, hq); 214 CIRCLEQ_INSERT_HEAD(head, bp, hq); 215 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 216 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 217 218 /* Return a pinned page. */ 219 bp->flags |= MPOOL_PINNED; 220 return (bp->page); 221 } 222 223 /* Get a page from the cache. */ 224 if ((bp = mpool_bkt(mp)) == NULL) 225 return (NULL); 226 227 /* Read in the contents. */ 228 #ifdef STATISTICS 229 ++mp->pageread; 230 #endif 231 off = mp->pagesize * pgno; 232 if (off / mp->pagesize != pgno) { 233 /* Run past the end of the file, or at least the part we 234 can address without large-file support? */ 235 errno = E2BIG; 236 return NULL; 237 } 238 if (lseek(mp->fd, off, SEEK_SET) != off) 239 return (NULL); 240 241 if ((nr = read(mp->fd, bp->page, mp->pagesize)) != mp->pagesize) { 242 if (nr > 0) { 243 /* A partial read is definitely bad. */ 244 errno = EINVAL; 245 return (NULL); 246 } else { 247 /* 248 * A zero-length reads, means you need to create a 249 * new page. 250 */ 251 memset(bp->page, 0, mp->pagesize); 252 } 253 } 254 255 /* Set the page number, pin the page. */ 256 bp->pgno = pgno; 257 if (!(flags & MPOOL_IGNOREPIN)) 258 bp->flags = MPOOL_PINNED; 259 bp->flags |= MPOOL_INUSE; 260 261 /* 262 * Add the page to the head of the hash chain and the tail 263 * of the lru chain. 264 */ 265 head = &mp->hqh[HASHKEY(bp->pgno)]; 266 CIRCLEQ_INSERT_HEAD(head, bp, hq); 267 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 268 269 /* Run through the user's filter. */ 270 if (mp->pgin != NULL) 271 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 272 273 return (bp->page); 274 } 275 276 /* 277 * mpool_put 278 * Return a page. 279 */ 280 int 281 mpool_put(mp, page, flags) 282 MPOOL *mp; 283 void *page; 284 u_int flags; 285 { 286 BKT *bp; 287 288 #ifdef STATISTICS 289 ++mp->pageput; 290 #endif 291 bp = (BKT *)((char *)page - sizeof(BKT)); 292 #ifdef DEBUG 293 if (!(bp->flags & MPOOL_PINNED)) { 294 (void)fprintf(stderr, 295 "mpool_put: page %d not pinned\n", bp->pgno); 296 abort(); 297 } 298 #endif 299 bp->flags &= ~MPOOL_PINNED; 300 if (flags & MPOOL_DIRTY) 301 bp->flags |= flags & MPOOL_DIRTY; 302 return (RET_SUCCESS); 303 } 304 305 /* 306 * mpool_close 307 * Close the buffer pool. 308 */ 309 int 310 mpool_close(mp) 311 MPOOL *mp; 312 { 313 BKT *bp; 314 315 /* Free up any space allocated to the lru pages. */ 316 while ((bp = mp->lqh.cqh_first) != (void *)&mp->lqh) { 317 CIRCLEQ_REMOVE(&mp->lqh, mp->lqh.cqh_first, q); 318 free(bp); 319 } 320 321 /* Free the MPOOL cookie. */ 322 free(mp); 323 return (RET_SUCCESS); 324 } 325 326 /* 327 * mpool_sync 328 * Sync the pool to disk. 329 */ 330 int 331 mpool_sync(mp) 332 MPOOL *mp; 333 { 334 BKT *bp; 335 336 /* Walk the lru chain, flushing any dirty pages to disk. */ 337 for (bp = mp->lqh.cqh_first; 338 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 339 if (bp->flags & MPOOL_DIRTY && 340 mpool_write(mp, bp) == RET_ERROR) 341 return (RET_ERROR); 342 343 /* Sync the file descriptor. */ 344 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 345 } 346 347 /* 348 * mpool_bkt 349 * Get a page from the cache (or create one). 350 */ 351 static BKT * 352 mpool_bkt(mp) 353 MPOOL *mp; 354 { 355 struct _hqh *head; 356 BKT *bp; 357 358 /* If under the max cached, always create a new page. */ 359 if (mp->curcache < mp->maxcache) 360 goto new; 361 362 /* 363 * If the cache is max'd out, walk the lru list for a buffer we 364 * can flush. If we find one, write it (if necessary) and take it 365 * off any lists. If we don't find anything we grow the cache anyway. 366 * The cache never shrinks. 367 */ 368 for (bp = mp->lqh.cqh_first; 369 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 370 if (!(bp->flags & MPOOL_PINNED)) { 371 /* Flush if dirty. */ 372 if (bp->flags & MPOOL_DIRTY && 373 mpool_write(mp, bp) == RET_ERROR) 374 return (NULL); 375 #ifdef STATISTICS 376 ++mp->pageflush; 377 #endif 378 /* Remove from the hash and lru queues. */ 379 head = &mp->hqh[HASHKEY(bp->pgno)]; 380 CIRCLEQ_REMOVE(head, bp, hq); 381 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 382 #ifdef DEBUG 383 { void *spage; 384 spage = bp->page; 385 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 386 bp->page = spage; 387 } 388 #endif 389 bp->flags = 0; 390 return (bp); 391 } 392 393 new: if ((bp = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL) 394 return (NULL); 395 #ifdef STATISTICS 396 ++mp->pagealloc; 397 #endif 398 #if defined(DEBUG) || defined(PURIFY) 399 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 400 #endif 401 bp->page = (char *)bp + sizeof(BKT); 402 bp->flags = 0; 403 ++mp->curcache; 404 return (bp); 405 } 406 407 /* 408 * mpool_write 409 * Write a page to disk. 410 */ 411 static int 412 mpool_write(mp, bp) 413 MPOOL *mp; 414 BKT *bp; 415 { 416 off_t off; 417 418 #ifdef STATISTICS 419 ++mp->pagewrite; 420 #endif 421 422 /* Run through the user's filter. */ 423 if (mp->pgout) 424 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 425 426 off = mp->pagesize * bp->pgno; 427 if (off / mp->pagesize != bp->pgno) { 428 /* Run past the end of the file, or at least the part we 429 can address without large-file support? */ 430 errno = E2BIG; 431 return RET_ERROR; 432 } 433 if (lseek(mp->fd, off, SEEK_SET) != off) 434 return (RET_ERROR); 435 if (write(mp->fd, bp->page, mp->pagesize) != mp->pagesize) 436 return (RET_ERROR); 437 438 bp->flags &= ~MPOOL_DIRTY; 439 return (RET_SUCCESS); 440 } 441 442 /* 443 * mpool_look 444 * Lookup a page in the cache. 445 */ 446 static BKT * 447 mpool_look(mp, pgno) 448 MPOOL *mp; 449 db_pgno_t pgno; 450 { 451 struct _hqh *head; 452 BKT *bp; 453 454 head = &mp->hqh[HASHKEY(pgno)]; 455 for (bp = head->cqh_first; bp != (void *)head; bp = bp->hq.cqe_next) 456 if ((bp->pgno == pgno) && 457 (bp->flags & MPOOL_INUSE == MPOOL_INUSE)) { 458 #ifdef STATISTICS 459 ++mp->cachehit; 460 #endif 461 return (bp); 462 } 463 #ifdef STATISTICS 464 ++mp->cachemiss; 465 #endif 466 return (NULL); 467 } 468 469 #ifdef STATISTICS 470 /* 471 * mpool_stat 472 * Print out cache statistics. 473 */ 474 void 475 mpool_stat(mp) 476 MPOOL *mp; 477 { 478 BKT *bp; 479 int cnt; 480 char *sep; 481 482 (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); 483 (void)fprintf(stderr, 484 "page size %lu, cacheing %lu pages of %lu page max cache\n", 485 mp->pagesize, mp->curcache, mp->maxcache); 486 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 487 mp->pageput, mp->pageget, mp->pagenew); 488 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 489 mp->pagealloc, mp->pageflush); 490 if (mp->cachehit + mp->cachemiss) 491 (void)fprintf(stderr, 492 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 493 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 494 * 100, mp->cachehit, mp->cachemiss); 495 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 496 mp->pageread, mp->pagewrite); 497 498 sep = ""; 499 cnt = 0; 500 for (bp = mp->lqh.cqh_first; 501 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) { 502 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 503 if (bp->flags & MPOOL_DIRTY) 504 (void)fprintf(stderr, "d"); 505 if (bp->flags & MPOOL_PINNED) 506 (void)fprintf(stderr, "P"); 507 if (++cnt == 10) { 508 sep = "\n"; 509 cnt = 0; 510 } else 511 sep = ", "; 512 513 } 514 (void)fprintf(stderr, "\n"); 515 } 516 #endif 517