1 /* $NetBSD: mpool.c,v 1.16 2007/02/03 23:46:09 christos 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. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 #if defined(LIBC_SCCS) && !defined(lint) 34 #if 0 35 static char sccsid[] = "@(#)mpool.c 8.5 (Berkeley) 7/26/94"; 36 #else 37 __RCSID("$NetBSD: mpool.c,v 1.16 2007/02/03 23:46:09 christos Exp $"); 38 #endif 39 #endif /* LIBC_SCCS and not lint */ 40 41 #include "namespace.h" 42 #include <sys/queue.h> 43 #include <sys/stat.h> 44 45 #include <errno.h> 46 #include <stdio.h> 47 #include <stdlib.h> 48 #include <string.h> 49 #include <unistd.h> 50 51 #include <db.h> 52 53 #define __MPOOLINTERFACE_PRIVATE 54 #include <mpool.h> 55 56 #ifdef __weak_alias 57 __weak_alias(mpool_close,_mpool_close) 58 __weak_alias(mpool_filter,_mpool_filter) 59 __weak_alias(mpool_get,_mpool_get) 60 __weak_alias(mpool_new,_mpool_new) 61 __weak_alias(mpool_open,_mpool_open) 62 __weak_alias(mpool_put,_mpool_put) 63 __weak_alias(mpool_sync,_mpool_sync) 64 #endif 65 66 static BKT *mpool_bkt(MPOOL *); 67 static BKT *mpool_look(MPOOL *, pgno_t); 68 static int mpool_write(MPOOL *, BKT *); 69 70 /* 71 * mpool_open -- 72 * Initialize a memory pool. 73 */ 74 /*ARGSUSED*/ 75 MPOOL * 76 mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache) 77 { 78 struct stat sb; 79 MPOOL *mp; 80 int entry; 81 82 /* 83 * Get information about the file. 84 * 85 * XXX 86 * We don't currently handle pipes, although we should. 87 */ 88 if (fstat(fd, &sb)) 89 return (NULL); 90 if (!S_ISREG(sb.st_mode)) { 91 errno = ESPIPE; 92 return (NULL); 93 } 94 95 /* Allocate and initialize the MPOOL cookie. */ 96 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 97 return (NULL); 98 CIRCLEQ_INIT(&mp->lqh); 99 for (entry = 0; entry < HASHSIZE; ++entry) 100 CIRCLEQ_INIT(&mp->hqh[entry]); 101 mp->maxcache = maxcache; 102 mp->npages = (pgno_t)(sb.st_size / pagesize); 103 mp->pagesize = pagesize; 104 mp->fd = fd; 105 return (mp); 106 } 107 108 /* 109 * mpool_filter -- 110 * Initialize input/output filters. 111 */ 112 void 113 mpool_filter(MPOOL *mp, void (*pgin)(void *, pgno_t, void *), 114 void (*pgout)(void *, pgno_t, void *), void *pgcookie) 115 { 116 mp->pgin = pgin; 117 mp->pgout = pgout; 118 mp->pgcookie = pgcookie; 119 } 120 121 /* 122 * mpool_new -- 123 * Get a new page of memory. 124 */ 125 void * 126 mpool_new( MPOOL *mp, pgno_t *pgnoaddr) 127 { 128 struct _hqh *head; 129 BKT *bp; 130 131 if (mp->npages == MAX_PAGE_NUMBER) { 132 (void)fprintf(stderr, "mpool_new: page allocation overflow.\n"); 133 abort(); 134 } 135 #ifdef STATISTICS 136 ++mp->pagenew; 137 #endif 138 /* 139 * Get a BKT from the cache. Assign a new page number, attach 140 * it to the head of the hash chain, the tail of the lru chain, 141 * and return. 142 */ 143 if ((bp = mpool_bkt(mp)) == NULL) 144 return (NULL); 145 *pgnoaddr = bp->pgno = mp->npages++; 146 bp->flags = MPOOL_PINNED; 147 148 head = &mp->hqh[HASHKEY(bp->pgno)]; 149 CIRCLEQ_INSERT_HEAD(head, bp, hq); 150 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 151 return (bp->page); 152 } 153 154 /* 155 * mpool_get 156 * Get a page. 157 */ 158 /*ARGSUSED*/ 159 void * 160 mpool_get(MPOOL *mp, pgno_t pgno, u_int flags) 161 { 162 struct _hqh *head; 163 BKT *bp; 164 off_t off; 165 ssize_t nr; 166 167 /* Check for attempt to retrieve a non-existent page. */ 168 if (pgno >= mp->npages) { 169 errno = EINVAL; 170 return (NULL); 171 } 172 173 #ifdef STATISTICS 174 ++mp->pageget; 175 #endif 176 177 /* Check for a page that is cached. */ 178 if ((bp = mpool_look(mp, pgno)) != NULL) { 179 #ifdef DEBUG 180 if (bp->flags & MPOOL_PINNED) { 181 (void)fprintf(stderr, 182 "mpool_get: page %d already pinned\n", bp->pgno); 183 abort(); 184 } 185 #endif 186 /* 187 * Move the page to the head of the hash chain and the tail 188 * of the lru chain. 189 */ 190 head = &mp->hqh[HASHKEY(bp->pgno)]; 191 CIRCLEQ_REMOVE(head, bp, hq); 192 CIRCLEQ_INSERT_HEAD(head, bp, hq); 193 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 194 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 195 196 /* Return a pinned page. */ 197 bp->flags |= MPOOL_PINNED; 198 return (bp->page); 199 } 200 201 /* Get a page from the cache. */ 202 if ((bp = mpool_bkt(mp)) == NULL) 203 return (NULL); 204 205 /* Read in the contents. */ 206 #ifdef STATISTICS 207 ++mp->pageread; 208 #endif 209 off = mp->pagesize * pgno; 210 if ((nr = pread(mp->fd, bp->page, (size_t)mp->pagesize, off)) != (int)mp->pagesize) { 211 if (nr >= 0) 212 errno = EFTYPE; 213 return (NULL); 214 } 215 216 /* Set the page number, pin the page. */ 217 bp->pgno = pgno; 218 bp->flags = MPOOL_PINNED; 219 220 /* 221 * Add the page to the head of the hash chain and the tail 222 * of the lru chain. 223 */ 224 head = &mp->hqh[HASHKEY(bp->pgno)]; 225 CIRCLEQ_INSERT_HEAD(head, bp, hq); 226 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 227 228 /* Run through the user's filter. */ 229 if (mp->pgin != NULL) 230 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 231 232 return (bp->page); 233 } 234 235 /* 236 * mpool_put 237 * Return a page. 238 */ 239 /*ARGSUSED*/ 240 int 241 mpool_put(MPOOL *mp, void *page, u_int flags) 242 { 243 BKT *bp; 244 245 #ifdef STATISTICS 246 ++mp->pageput; 247 #endif 248 bp = (BKT *)(void *)((char *)page - sizeof(BKT)); 249 #ifdef DEBUG 250 if (!(bp->flags & MPOOL_PINNED)) { 251 (void)fprintf(stderr, 252 "mpool_put: page %d not pinned\n", bp->pgno); 253 abort(); 254 } 255 #endif 256 bp->flags &= ~MPOOL_PINNED; 257 bp->flags |= flags & MPOOL_DIRTY; 258 return (RET_SUCCESS); 259 } 260 261 /* 262 * mpool_close 263 * Close the buffer pool. 264 */ 265 int 266 mpool_close(MPOOL *mp) 267 { 268 BKT *bp; 269 270 /* Free up any space allocated to the lru pages. */ 271 while ((bp = mp->lqh.cqh_first) != (void *)&mp->lqh) { 272 CIRCLEQ_REMOVE(&mp->lqh, mp->lqh.cqh_first, q); 273 free(bp); 274 } 275 276 /* Free the MPOOL cookie. */ 277 free(mp); 278 return (RET_SUCCESS); 279 } 280 281 /* 282 * mpool_sync 283 * Sync the pool to disk. 284 */ 285 int 286 mpool_sync(MPOOL *mp) 287 { 288 BKT *bp; 289 290 /* Walk the lru chain, flushing any dirty pages to disk. */ 291 for (bp = mp->lqh.cqh_first; 292 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 293 if (bp->flags & MPOOL_DIRTY && 294 mpool_write(mp, bp) == RET_ERROR) 295 return (RET_ERROR); 296 297 /* Sync the file descriptor. */ 298 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 299 } 300 301 /* 302 * mpool_bkt 303 * Get a page from the cache (or create one). 304 */ 305 static BKT * 306 mpool_bkt(MPOOL *mp) 307 { 308 struct _hqh *head; 309 BKT *bp; 310 311 /* If under the max cached, always create a new page. */ 312 if (mp->curcache < mp->maxcache) 313 goto new; 314 315 /* 316 * If the cache is max'd out, walk the lru list for a buffer we 317 * can flush. If we find one, write it (if necessary) and take it 318 * off any lists. If we don't find anything we grow the cache anyway. 319 * The cache never shrinks. 320 */ 321 for (bp = mp->lqh.cqh_first; 322 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 323 if (!(bp->flags & MPOOL_PINNED)) { 324 /* Flush if dirty. */ 325 if (bp->flags & MPOOL_DIRTY && 326 mpool_write(mp, bp) == RET_ERROR) 327 return (NULL); 328 #ifdef STATISTICS 329 ++mp->pageflush; 330 #endif 331 /* Remove from the hash and lru queues. */ 332 head = &mp->hqh[HASHKEY(bp->pgno)]; 333 CIRCLEQ_REMOVE(head, bp, hq); 334 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 335 #ifdef DEBUG 336 { 337 void *spage = bp->page; 338 (void)memset(bp, 0xff, 339 (size_t)(sizeof(BKT) + mp->pagesize)); 340 bp->page = spage; 341 } 342 #endif 343 return (bp); 344 } 345 346 new: if ((bp = (BKT *)malloc((size_t)(sizeof(BKT) + mp->pagesize))) == NULL) 347 return (NULL); 348 #ifdef STATISTICS 349 ++mp->pagealloc; 350 #endif 351 #if defined(DEBUG) || defined(PURIFY) 352 (void)memset(bp, 0xff, (size_t)(sizeof(BKT) + mp->pagesize)); 353 #endif 354 bp->page = (char *)(void *)bp + sizeof(BKT); 355 ++mp->curcache; 356 return (bp); 357 } 358 359 /* 360 * mpool_write 361 * Write a page to disk. 362 */ 363 static int 364 mpool_write(MPOOL *mp, BKT *bp) 365 { 366 off_t off; 367 368 #ifdef STATISTICS 369 ++mp->pagewrite; 370 #endif 371 372 /* Run through the user's filter. */ 373 if (mp->pgout) 374 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 375 376 off = mp->pagesize * bp->pgno; 377 if (pwrite(mp->fd, bp->page, (size_t)mp->pagesize, off) != (int)mp->pagesize) 378 return (RET_ERROR); 379 380 /* 381 * Re-run through the input filter since this page may soon be 382 * accessed via the cache, and whatever the user's output filter 383 * did may screw things up if we don't let the input filter 384 * restore the in-core copy. 385 */ 386 if (mp->pgin) 387 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 388 389 bp->flags &= ~MPOOL_DIRTY; 390 return (RET_SUCCESS); 391 } 392 393 /* 394 * mpool_look 395 * Lookup a page in the cache. 396 */ 397 static BKT * 398 mpool_look(MPOOL *mp, pgno_t pgno) 399 { 400 struct _hqh *head; 401 BKT *bp; 402 403 head = &mp->hqh[HASHKEY(pgno)]; 404 for (bp = head->cqh_first; bp != (void *)head; bp = bp->hq.cqe_next) 405 if (bp->pgno == pgno) { 406 #ifdef STATISTICS 407 ++mp->cachehit; 408 #endif 409 return (bp); 410 } 411 #ifdef STATISTICS 412 ++mp->cachemiss; 413 #endif 414 return (NULL); 415 } 416 417 #ifdef STATISTICS 418 /* 419 * mpool_stat 420 * Print out cache statistics. 421 */ 422 void 423 mpool_stat(mp) 424 MPOOL *mp; 425 { 426 BKT *bp; 427 int cnt; 428 const char *sep; 429 430 (void)fprintf(stderr, "%lu pages in the file\n", (u_long)mp->npages); 431 (void)fprintf(stderr, 432 "page size %lu, cacheing %lu pages of %lu page max cache\n", 433 (u_long)mp->pagesize, (u_long)mp->curcache, (u_long)mp->maxcache); 434 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 435 mp->pageput, mp->pageget, mp->pagenew); 436 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 437 mp->pagealloc, mp->pageflush); 438 if (mp->cachehit + mp->cachemiss) 439 (void)fprintf(stderr, 440 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 441 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 442 * 100, mp->cachehit, mp->cachemiss); 443 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 444 mp->pageread, mp->pagewrite); 445 446 sep = ""; 447 cnt = 0; 448 for (bp = mp->lqh.cqh_first; 449 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) { 450 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 451 if (bp->flags & MPOOL_DIRTY) 452 (void)fprintf(stderr, "d"); 453 if (bp->flags & MPOOL_PINNED) 454 (void)fprintf(stderr, "P"); 455 if (++cnt == 10) { 456 sep = "\n"; 457 cnt = 0; 458 } else 459 sep = ", "; 460 461 } 462 (void)fprintf(stderr, "\n"); 463 } 464 #endif 465