1 /* $OpenBSD: mpool.c,v 1.6 1999/02/15 05:11:25 millert Exp $ */ 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 #if 0 38 static char sccsid[] = "@(#)mpool.c 8.7 (Berkeley) 11/2/95"; 39 #else 40 static char rcsid[] = "$OpenBSD: mpool.c,v 1.6 1999/02/15 05:11:25 millert Exp $"; 41 #endif 42 #endif /* LIBC_SCCS and not lint */ 43 44 #include <sys/param.h> 45 #include <sys/queue.h> 46 #include <sys/stat.h> 47 48 #include <errno.h> 49 #include <stdio.h> 50 #include <stdlib.h> 51 #include <string.h> 52 #include <unistd.h> 53 54 #include <db.h> 55 56 #define __MPOOLINTERFACE_PRIVATE 57 #include <mpool.h> 58 59 static BKT *mpool_bkt __P((MPOOL *)); 60 static BKT *mpool_look __P((MPOOL *, pgno_t)); 61 static int mpool_write __P((MPOOL *, BKT *)); 62 63 /* 64 * mpool_open -- 65 * Initialize a memory pool. 66 */ 67 /* ARGSUSED */ 68 MPOOL * 69 mpool_open(key, fd, pagesize, maxcache) 70 void *key; 71 int fd; 72 pgno_t pagesize, maxcache; 73 { 74 struct stat sb; 75 MPOOL *mp; 76 int entry; 77 78 /* 79 * Get information about the file. 80 * 81 * XXX 82 * We don't currently handle pipes, although we should. 83 */ 84 if (fstat(fd, &sb)) 85 return (NULL); 86 if (!S_ISREG(sb.st_mode)) { 87 errno = ESPIPE; 88 return (NULL); 89 } 90 91 /* Allocate and initialize the MPOOL cookie. */ 92 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 93 return (NULL); 94 CIRCLEQ_INIT(&mp->lqh); 95 for (entry = 0; entry < HASHSIZE; ++entry) 96 CIRCLEQ_INIT(&mp->hqh[entry]); 97 mp->maxcache = maxcache; 98 mp->npages = sb.st_size / pagesize; 99 mp->pagesize = pagesize; 100 mp->fd = fd; 101 return (mp); 102 } 103 104 /* 105 * mpool_filter -- 106 * Initialize input/output filters. 107 */ 108 void 109 mpool_filter(mp, pgin, pgout, pgcookie) 110 MPOOL *mp; 111 void (*pgin) __P((void *, pgno_t, void *)); 112 void (*pgout) __P((void *, pgno_t, void *)); 113 void *pgcookie; 114 { 115 mp->pgin = pgin; 116 mp->pgout = pgout; 117 mp->pgcookie = pgcookie; 118 } 119 120 /* 121 * mpool_new -- 122 * Get a new page of memory. 123 */ 124 void * 125 mpool_new(mp, pgnoaddr, flags) 126 MPOOL *mp; 127 pgno_t *pgnoaddr; 128 u_int flags; 129 { 130 struct _hqh *head; 131 BKT *bp; 132 133 if (mp->npages == MAX_PAGE_NUMBER) { 134 (void)fprintf(stderr, "mpool_new: page allocation overflow.\n"); 135 abort(); 136 } 137 #ifdef STATISTICS 138 ++mp->pagenew; 139 #endif 140 /* 141 * Get a BKT from the cache. Assign a new page number, attach 142 * it to the head of the hash chain, the tail of the lru chain, 143 * and return. 144 */ 145 if ((bp = mpool_bkt(mp)) == NULL) 146 return (NULL); 147 if (flags == MPOOL_PAGE_REQUEST) { 148 mp->npages++; 149 bp->pgno = *pgnoaddr; 150 } else 151 bp->pgno = *pgnoaddr = mp->npages++; 152 153 bp->flags = MPOOL_PINNED | MPOOL_INUSE; 154 155 head = &mp->hqh[HASHKEY(bp->pgno)]; 156 CIRCLEQ_INSERT_HEAD(head, bp, hq); 157 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 158 return (bp->page); 159 } 160 161 int 162 mpool_delete(mp, page) 163 MPOOL *mp; 164 void *page; 165 { 166 struct _hqh *head; 167 BKT *bp; 168 169 bp = (BKT *)((char *)page - sizeof(BKT)); 170 171 #ifdef DEBUG 172 if (!(bp->flags & MPOOL_PINNED)) { 173 (void)fprintf(stderr, 174 "mpool_delete: page %d not pinned\n", bp->pgno); 175 abort(); 176 } 177 #endif 178 179 /* Remove from the hash and lru queues. */ 180 head = &mp->hqh[HASHKEY(bp->pgno)]; 181 CIRCLEQ_REMOVE(head, bp, hq); 182 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 183 184 free(bp); 185 return (RET_SUCCESS); 186 } 187 188 /* 189 * mpool_get 190 * Get a page. 191 */ 192 /* ARGSUSED */ 193 void * 194 mpool_get(mp, pgno, flags) 195 MPOOL *mp; 196 pgno_t pgno; 197 u_int flags; /* XXX not used? */ 198 { 199 struct _hqh *head; 200 BKT *bp; 201 off_t off; 202 int nr; 203 204 #ifdef STATISTICS 205 ++mp->pageget; 206 #endif 207 208 /* Check for a page that is cached. */ 209 if ((bp = mpool_look(mp, pgno)) != NULL) { 210 #ifdef DEBUG 211 if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) { 212 (void)fprintf(stderr, 213 "mpool_get: page %d already pinned\n", bp->pgno); 214 abort(); 215 } 216 #endif 217 /* 218 * Move the page to the head of the hash chain and the tail 219 * of the lru chain. 220 */ 221 head = &mp->hqh[HASHKEY(bp->pgno)]; 222 CIRCLEQ_REMOVE(head, bp, hq); 223 CIRCLEQ_INSERT_HEAD(head, bp, hq); 224 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 225 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 226 227 /* Return a pinned page. */ 228 bp->flags |= MPOOL_PINNED; 229 return (bp->page); 230 } 231 232 /* Get a page from the cache. */ 233 if ((bp = mpool_bkt(mp)) == NULL) 234 return (NULL); 235 236 /* Read in the contents. */ 237 #ifdef STATISTICS 238 ++mp->pageread; 239 #endif 240 off = mp->pagesize * pgno; 241 if (lseek(mp->fd, off, SEEK_SET) != off) 242 return (NULL); 243 244 if ((nr = read(mp->fd, bp->page, mp->pagesize)) != mp->pagesize) { 245 if (nr > 0) { 246 /* A partial read is definitely bad. */ 247 errno = EINVAL; 248 return (NULL); 249 } else { 250 /* 251 * A zero-length reads, means you need to create a 252 * new page. 253 */ 254 memset(bp->page, 0, mp->pagesize); 255 } 256 } 257 258 /* Set the page number, pin the page. */ 259 bp->pgno = pgno; 260 if (!(flags & MPOOL_IGNOREPIN)) 261 bp->flags = MPOOL_PINNED; 262 bp->flags |= MPOOL_INUSE; 263 264 /* 265 * Add the page to the head of the hash chain and the tail 266 * of the lru chain. 267 */ 268 head = &mp->hqh[HASHKEY(bp->pgno)]; 269 CIRCLEQ_INSERT_HEAD(head, bp, hq); 270 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 271 272 /* Run through the user's filter. */ 273 if (mp->pgin != NULL) 274 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 275 276 return (bp->page); 277 } 278 279 /* 280 * mpool_put 281 * Return a page. 282 */ 283 /* ARGSUSED */ 284 int 285 mpool_put(mp, page, flags) 286 MPOOL *mp; 287 void *page; 288 u_int flags; 289 { 290 BKT *bp; 291 292 #ifdef STATISTICS 293 ++mp->pageput; 294 #endif 295 bp = (BKT *)((char *)page - sizeof(BKT)); 296 #ifdef DEBUG 297 if (!(bp->flags & MPOOL_PINNED)) { 298 (void)fprintf(stderr, 299 "mpool_put: page %d not pinned\n", bp->pgno); 300 abort(); 301 } 302 #endif 303 bp->flags &= ~MPOOL_PINNED; 304 if (flags & MPOOL_DIRTY) 305 bp->flags |= flags & MPOOL_DIRTY; 306 return (RET_SUCCESS); 307 } 308 309 /* 310 * mpool_close 311 * Close the buffer pool. 312 */ 313 int 314 mpool_close(mp) 315 MPOOL *mp; 316 { 317 BKT *bp; 318 319 /* Free up any space allocated to the lru pages. */ 320 while ((bp = mp->lqh.cqh_first) != (void *)&mp->lqh) { 321 CIRCLEQ_REMOVE(&mp->lqh, mp->lqh.cqh_first, q); 322 free(bp); 323 } 324 325 /* Free the MPOOL cookie. */ 326 free(mp); 327 return (RET_SUCCESS); 328 } 329 330 /* 331 * mpool_sync 332 * Sync the pool to disk. 333 */ 334 int 335 mpool_sync(mp) 336 MPOOL *mp; 337 { 338 BKT *bp; 339 340 /* Walk the lru chain, flushing any dirty pages to disk. */ 341 for (bp = mp->lqh.cqh_first; 342 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 343 if (bp->flags & MPOOL_DIRTY && 344 mpool_write(mp, bp) == RET_ERROR) 345 return (RET_ERROR); 346 347 /* Sync the file descriptor. */ 348 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 349 } 350 351 /* 352 * mpool_bkt 353 * Get a page from the cache (or create one). 354 */ 355 static BKT * 356 mpool_bkt(mp) 357 MPOOL *mp; 358 { 359 struct _hqh *head; 360 BKT *bp; 361 362 /* If under the max cached, always create a new page. */ 363 if (mp->curcache < mp->maxcache) 364 goto new; 365 366 /* 367 * If the cache is max'd out, walk the lru list for a buffer we 368 * can flush. If we find one, write it (if necessary) and take it 369 * off any lists. If we don't find anything we grow the cache anyway. 370 * The cache never shrinks. 371 */ 372 for (bp = mp->lqh.cqh_first; 373 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 374 if (!(bp->flags & MPOOL_PINNED)) { 375 /* Flush if dirty. */ 376 if (bp->flags & MPOOL_DIRTY && 377 mpool_write(mp, bp) == RET_ERROR) 378 return (NULL); 379 #ifdef STATISTICS 380 ++mp->pageflush; 381 #endif 382 /* Remove from the hash and lru queues. */ 383 head = &mp->hqh[HASHKEY(bp->pgno)]; 384 CIRCLEQ_REMOVE(head, bp, hq); 385 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 386 #ifdef DEBUG 387 { void *spage; 388 spage = bp->page; 389 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 390 bp->page = spage; 391 } 392 #endif 393 bp->flags = 0; 394 return (bp); 395 } 396 397 new: if ((bp = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL) 398 return (NULL); 399 #ifdef STATISTICS 400 ++mp->pagealloc; 401 #endif 402 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 403 bp->page = (char *)bp + sizeof(BKT); 404 bp->flags = 0; 405 ++mp->curcache; 406 return (bp); 407 } 408 409 /* 410 * mpool_write 411 * Write a page to disk. 412 */ 413 static int 414 mpool_write(mp, bp) 415 MPOOL *mp; 416 BKT *bp; 417 { 418 off_t off; 419 420 #ifdef STATISTICS 421 ++mp->pagewrite; 422 #endif 423 424 /* Run through the user's filter. */ 425 if (mp->pgout) 426 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 427 428 off = mp->pagesize * bp->pgno; 429 if (lseek(mp->fd, off, SEEK_SET) != off) 430 return (RET_ERROR); 431 if (write(mp->fd, bp->page, mp->pagesize) != mp->pagesize) 432 return (RET_ERROR); 433 434 bp->flags &= ~MPOOL_DIRTY; 435 return (RET_SUCCESS); 436 } 437 438 /* 439 * mpool_look 440 * Lookup a page in the cache. 441 */ 442 static BKT * 443 mpool_look(mp, pgno) 444 MPOOL *mp; 445 pgno_t pgno; 446 { 447 struct _hqh *head; 448 BKT *bp; 449 450 head = &mp->hqh[HASHKEY(pgno)]; 451 for (bp = head->cqh_first; bp != (void *)head; bp = bp->hq.cqe_next) 452 if ((bp->pgno == pgno) && 453 (bp->flags & MPOOL_INUSE == MPOOL_INUSE)) { 454 #ifdef STATISTICS 455 ++mp->cachehit; 456 #endif 457 return (bp); 458 } 459 #ifdef STATISTICS 460 ++mp->cachemiss; 461 #endif 462 return (NULL); 463 } 464 465 #ifdef STATISTICS 466 /* 467 * mpool_stat 468 * Print out cache statistics. 469 */ 470 void 471 mpool_stat(mp) 472 MPOOL *mp; 473 { 474 BKT *bp; 475 int cnt; 476 char *sep; 477 478 (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); 479 (void)fprintf(stderr, 480 "page size %lu, cacheing %lu pages of %lu page max cache\n", 481 mp->pagesize, mp->curcache, mp->maxcache); 482 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 483 mp->pageput, mp->pageget, mp->pagenew); 484 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 485 mp->pagealloc, mp->pageflush); 486 if (mp->cachehit + mp->cachemiss) 487 (void)fprintf(stderr, 488 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 489 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 490 * 100, mp->cachehit, mp->cachemiss); 491 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 492 mp->pageread, mp->pagewrite); 493 494 sep = ""; 495 cnt = 0; 496 for (bp = mp->lqh.cqh_first; 497 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) { 498 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 499 if (bp->flags & MPOOL_DIRTY) 500 (void)fprintf(stderr, "d"); 501 if (bp->flags & MPOOL_PINNED) 502 (void)fprintf(stderr, "P"); 503 if (++cnt == 10) { 504 sep = "\n"; 505 cnt = 0; 506 } else 507 sep = ", "; 508 509 } 510 (void)fprintf(stderr, "\n"); 511 } 512 #endif 513