1 /* $OpenBSD: mpool.c,v 1.10 2002/02/25 23:45:15 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 1.10 2002/02/25 23:45:15 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(MPOOL *); 60 static BKT *mpool_look(MPOOL *, pgno_t); 61 static int mpool_write(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)(void *, pgno_t, void *); 112 void (*pgout)(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 ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != mp->pagesize) { 242 switch (nr) { 243 case -1: 244 /* errno is set for us by pread(). */ 245 return (NULL); 246 case 0: 247 /* 248 * A zero-length read means you need to create a 249 * new page. 250 */ 251 memset(bp->page, 0, mp->pagesize); 252 default: 253 /* A partial read is definitely bad. */ 254 errno = EINVAL; 255 return (NULL); 256 } 257 } 258 259 /* Set the page number, pin the page. */ 260 bp->pgno = pgno; 261 if (!(flags & MPOOL_IGNOREPIN)) 262 bp->flags = MPOOL_PINNED; 263 bp->flags |= MPOOL_INUSE; 264 265 /* 266 * Add the page to the head of the hash chain and the tail 267 * of the lru chain. 268 */ 269 head = &mp->hqh[HASHKEY(bp->pgno)]; 270 CIRCLEQ_INSERT_HEAD(head, bp, hq); 271 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 272 273 /* Run through the user's filter. */ 274 if (mp->pgin != NULL) 275 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 276 277 return (bp->page); 278 } 279 280 /* 281 * mpool_put 282 * Return a page. 283 */ 284 /* ARGSUSED */ 285 int 286 mpool_put(mp, page, flags) 287 MPOOL *mp; 288 void *page; 289 u_int flags; 290 { 291 BKT *bp; 292 293 #ifdef STATISTICS 294 ++mp->pageput; 295 #endif 296 bp = (BKT *)((char *)page - sizeof(BKT)); 297 #ifdef DEBUG 298 if (!(bp->flags & MPOOL_PINNED)) { 299 (void)fprintf(stderr, 300 "mpool_put: page %d not pinned\n", bp->pgno); 301 abort(); 302 } 303 #endif 304 bp->flags &= ~MPOOL_PINNED; 305 if (flags & MPOOL_DIRTY) 306 bp->flags |= flags & MPOOL_DIRTY; 307 return (RET_SUCCESS); 308 } 309 310 /* 311 * mpool_close 312 * Close the buffer pool. 313 */ 314 int 315 mpool_close(mp) 316 MPOOL *mp; 317 { 318 BKT *bp; 319 320 /* Free up any space allocated to the lru pages. */ 321 while ((bp = mp->lqh.cqh_first) != (void *)&mp->lqh) { 322 CIRCLEQ_REMOVE(&mp->lqh, mp->lqh.cqh_first, q); 323 free(bp); 324 } 325 326 /* Free the MPOOL cookie. */ 327 free(mp); 328 return (RET_SUCCESS); 329 } 330 331 /* 332 * mpool_sync 333 * Sync the pool to disk. 334 */ 335 int 336 mpool_sync(mp) 337 MPOOL *mp; 338 { 339 BKT *bp; 340 341 /* Walk the lru chain, flushing any dirty pages to disk. */ 342 for (bp = mp->lqh.cqh_first; 343 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 344 if (bp->flags & MPOOL_DIRTY && 345 mpool_write(mp, bp) == RET_ERROR) 346 return (RET_ERROR); 347 348 /* Sync the file descriptor. */ 349 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 350 } 351 352 /* 353 * mpool_bkt 354 * Get a page from the cache (or create one). 355 */ 356 static BKT * 357 mpool_bkt(mp) 358 MPOOL *mp; 359 { 360 struct _hqh *head; 361 BKT *bp; 362 363 /* If under the max cached, always create a new page. */ 364 if (mp->curcache < mp->maxcache) 365 goto new; 366 367 /* 368 * If the cache is max'd out, walk the lru list for a buffer we 369 * can flush. If we find one, write it (if necessary) and take it 370 * off any lists. If we don't find anything we grow the cache anyway. 371 * The cache never shrinks. 372 */ 373 for (bp = mp->lqh.cqh_first; 374 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 375 if (!(bp->flags & MPOOL_PINNED)) { 376 /* Flush if dirty. */ 377 if (bp->flags & MPOOL_DIRTY && 378 mpool_write(mp, bp) == RET_ERROR) 379 return (NULL); 380 #ifdef STATISTICS 381 ++mp->pageflush; 382 #endif 383 /* Remove from the hash and lru queues. */ 384 head = &mp->hqh[HASHKEY(bp->pgno)]; 385 CIRCLEQ_REMOVE(head, bp, hq); 386 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 387 #ifdef DEBUG 388 { void *spage; 389 spage = bp->page; 390 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 391 bp->page = spage; 392 } 393 #endif 394 bp->flags = 0; 395 return (bp); 396 } 397 398 new: if ((bp = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL) 399 return (NULL); 400 #ifdef STATISTICS 401 ++mp->pagealloc; 402 #endif 403 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 404 bp->page = (char *)bp + sizeof(BKT); 405 bp->flags = 0; 406 ++mp->curcache; 407 return (bp); 408 } 409 410 /* 411 * mpool_write 412 * Write a page to disk. 413 */ 414 static int 415 mpool_write(mp, bp) 416 MPOOL *mp; 417 BKT *bp; 418 { 419 off_t off; 420 421 #ifdef STATISTICS 422 ++mp->pagewrite; 423 #endif 424 425 /* Run through the user's filter. */ 426 if (mp->pgout) 427 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 428 429 off = mp->pagesize * bp->pgno; 430 if (pwrite(mp->fd, bp->page, mp->pagesize, off) != mp->pagesize) 431 return (RET_ERROR); 432 433 /* 434 * Re-run through the input filter since this page may soon be 435 * accessed via the cache, and whatever the user's output filter 436 * did may screw things up if we don't let the input filter 437 * restore the in-core copy. 438 */ 439 if (mp->pgin) 440 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 441 442 bp->flags &= ~MPOOL_DIRTY; 443 return (RET_SUCCESS); 444 } 445 446 /* 447 * mpool_look 448 * Lookup a page in the cache. 449 */ 450 static BKT * 451 mpool_look(mp, pgno) 452 MPOOL *mp; 453 pgno_t pgno; 454 { 455 struct _hqh *head; 456 BKT *bp; 457 458 head = &mp->hqh[HASHKEY(pgno)]; 459 for (bp = head->cqh_first; bp != (void *)head; bp = bp->hq.cqe_next) 460 if ((bp->pgno == pgno) && 461 ((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) { 462 #ifdef STATISTICS 463 ++mp->cachehit; 464 #endif 465 return (bp); 466 } 467 #ifdef STATISTICS 468 ++mp->cachemiss; 469 #endif 470 return (NULL); 471 } 472 473 #ifdef STATISTICS 474 /* 475 * mpool_stat 476 * Print out cache statistics. 477 */ 478 void 479 mpool_stat(mp) 480 MPOOL *mp; 481 { 482 BKT *bp; 483 int cnt; 484 char *sep; 485 486 (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); 487 (void)fprintf(stderr, 488 "page size %lu, cacheing %lu pages of %lu page max cache\n", 489 mp->pagesize, mp->curcache, mp->maxcache); 490 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 491 mp->pageput, mp->pageget, mp->pagenew); 492 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 493 mp->pagealloc, mp->pageflush); 494 if (mp->cachehit + mp->cachemiss) 495 (void)fprintf(stderr, 496 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 497 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 498 * 100, mp->cachehit, mp->cachemiss); 499 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 500 mp->pageread, mp->pagewrite); 501 502 sep = ""; 503 cnt = 0; 504 for (bp = mp->lqh.cqh_first; 505 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) { 506 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 507 if (bp->flags & MPOOL_DIRTY) 508 (void)fprintf(stderr, "d"); 509 if (bp->flags & MPOOL_PINNED) 510 (void)fprintf(stderr, "P"); 511 if (++cnt == 10) { 512 sep = "\n"; 513 cnt = 0; 514 } else 515 sep = ", "; 516 517 } 518 (void)fprintf(stderr, "\n"); 519 } 520 #endif 521