1 /* $NetBSD: mpool.c,v 1.19 2009/04/22 18:44:06 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 #if HAVE_NBTOOL_CONFIG_H 33 #include "nbtool_config.h" 34 #endif 35 36 #include <sys/cdefs.h> 37 #ifndef __minix 38 __RCSID("$NetBSD: mpool.c,v 1.19 2009/04/22 18:44:06 christos Exp $"); 39 #endif 40 41 #ifndef __minix 42 #include "namespace.h" 43 #endif 44 #include <sys/queue.h> 45 #include <sys/stat.h> 46 47 #include <errno.h> 48 #include <stdio.h> 49 #include <stdlib.h> 50 #include <string.h> 51 #include <unistd.h> 52 53 #include <db.h> 54 55 #define __MPOOLINTERFACE_PRIVATE 56 #include <mpool.h> 57 58 #ifdef __weak_alias 59 __weak_alias(mpool_close,_mpool_close) 60 __weak_alias(mpool_filter,_mpool_filter) 61 __weak_alias(mpool_get,_mpool_get) 62 __weak_alias(mpool_new,_mpool_new) 63 __weak_alias(mpool_open,_mpool_open) 64 __weak_alias(mpool_put,_mpool_put) 65 __weak_alias(mpool_sync,_mpool_sync) 66 #endif 67 68 static BKT *mpool_bkt(MPOOL *); 69 static BKT *mpool_look(MPOOL *, pgno_t); 70 static int mpool_write(MPOOL *, BKT *); 71 72 /* 73 * mpool_open -- 74 * Initialize a memory pool. 75 */ 76 /*ARGSUSED*/ 77 MPOOL * 78 mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache) 79 { 80 struct stat sb; 81 MPOOL *mp; 82 int entry; 83 84 /* 85 * Get information about the file. 86 * 87 * XXX 88 * We don't currently handle pipes, although we should. 89 */ 90 if (fstat(fd, &sb)) 91 return (NULL); 92 if (!S_ISREG(sb.st_mode)) { 93 errno = ESPIPE; 94 return (NULL); 95 } 96 97 /* Allocate and initialize the MPOOL cookie. */ 98 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 99 return (NULL); 100 CIRCLEQ_INIT(&mp->lqh); 101 for (entry = 0; entry < HASHSIZE; ++entry) 102 CIRCLEQ_INIT(&mp->hqh[entry]); 103 mp->maxcache = maxcache; 104 mp->npages = (pgno_t)(sb.st_size / pagesize); 105 mp->pagesize = pagesize; 106 mp->fd = fd; 107 return (mp); 108 } 109 110 /* 111 * mpool_filter -- 112 * Initialize input/output filters. 113 */ 114 void 115 mpool_filter(MPOOL *mp, void (*pgin)(void *, pgno_t, void *), 116 void (*pgout)(void *, pgno_t, void *), void *pgcookie) 117 { 118 mp->pgin = pgin; 119 mp->pgout = pgout; 120 mp->pgcookie = pgcookie; 121 } 122 123 /* 124 * mpool_new -- 125 * Get a new page of memory. 126 */ 127 void * 128 mpool_new( MPOOL *mp, pgno_t *pgnoaddr) 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 *pgnoaddr = bp->pgno = mp->npages++; 148 bp->flags = MPOOL_PINNED; 149 150 head = &mp->hqh[HASHKEY(bp->pgno)]; 151 CIRCLEQ_INSERT_HEAD(head, bp, hq); 152 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 153 return (bp->page); 154 } 155 156 /* 157 * mpool_get 158 * Get a page. 159 */ 160 /*ARGSUSED*/ 161 void * 162 mpool_get(MPOOL *mp, pgno_t pgno, u_int flags) 163 { 164 struct _hqh *head; 165 BKT *bp; 166 off_t off; 167 ssize_t nr; 168 169 /* Check for attempt to retrieve a non-existent page. */ 170 if (pgno >= mp->npages) { 171 errno = EINVAL; 172 return (NULL); 173 } 174 175 #ifdef STATISTICS 176 ++mp->pageget; 177 #endif 178 179 /* Check for a page that is cached. */ 180 if ((bp = mpool_look(mp, pgno)) != NULL) { 181 #ifdef DEBUG 182 if (bp->flags & MPOOL_PINNED) { 183 (void)fprintf(stderr, 184 "mpool_get: page %d already pinned\n", bp->pgno); 185 abort(); 186 } 187 #endif 188 /* 189 * Move the page to the head of the hash chain and the tail 190 * of the lru chain. 191 */ 192 head = &mp->hqh[HASHKEY(bp->pgno)]; 193 CIRCLEQ_REMOVE(head, bp, hq); 194 CIRCLEQ_INSERT_HEAD(head, bp, hq); 195 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 196 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 197 198 /* Return a pinned page. */ 199 bp->flags |= MPOOL_PINNED; 200 return (bp->page); 201 } 202 203 /* Get a page from the cache. */ 204 if ((bp = mpool_bkt(mp)) == NULL) 205 return (NULL); 206 207 /* Read in the contents. */ 208 #ifdef STATISTICS 209 ++mp->pageread; 210 #endif 211 off = mp->pagesize * pgno; 212 if ((nr = pread(mp->fd, bp->page, (size_t)mp->pagesize, off)) != (int)mp->pagesize) { 213 if (nr >= 0) 214 errno = EFTYPE; 215 return (NULL); 216 } 217 218 /* Set the page number, pin the page. */ 219 bp->pgno = pgno; 220 bp->flags = MPOOL_PINNED; 221 222 /* 223 * Add the page to the head of the hash chain and the tail 224 * of the lru chain. 225 */ 226 head = &mp->hqh[HASHKEY(bp->pgno)]; 227 CIRCLEQ_INSERT_HEAD(head, bp, hq); 228 CIRCLEQ_INSERT_TAIL(&mp->lqh, bp, q); 229 230 /* Run through the user's filter. */ 231 if (mp->pgin != NULL) 232 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 233 234 return (bp->page); 235 } 236 237 /* 238 * mpool_put 239 * Return a page. 240 */ 241 /*ARGSUSED*/ 242 int 243 mpool_put(MPOOL *mp, void *page, u_int flags) 244 { 245 BKT *bp; 246 247 #ifdef STATISTICS 248 ++mp->pageput; 249 #endif 250 bp = (BKT *)(void *)((char *)page - sizeof(BKT)); 251 #ifdef DEBUG 252 if (!(bp->flags & MPOOL_PINNED)) { 253 (void)fprintf(stderr, 254 "mpool_put: page %d not pinned\n", bp->pgno); 255 abort(); 256 } 257 #endif 258 bp->flags &= ~MPOOL_PINNED; 259 bp->flags |= flags & MPOOL_DIRTY; 260 return (RET_SUCCESS); 261 } 262 263 /* 264 * mpool_close 265 * Close the buffer pool. 266 */ 267 int 268 mpool_close(MPOOL *mp) 269 { 270 BKT *bp; 271 272 /* Free up any space allocated to the lru pages. */ 273 while ((bp = mp->lqh.cqh_first) != (void *)&mp->lqh) { 274 CIRCLEQ_REMOVE(&mp->lqh, mp->lqh.cqh_first, q); 275 free(bp); 276 } 277 278 /* Free the MPOOL cookie. */ 279 free(mp); 280 return (RET_SUCCESS); 281 } 282 283 /* 284 * mpool_sync 285 * Sync the pool to disk. 286 */ 287 int 288 mpool_sync(MPOOL *mp) 289 { 290 BKT *bp; 291 292 /* Walk the lru chain, flushing any dirty pages to disk. */ 293 for (bp = mp->lqh.cqh_first; 294 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 295 if (bp->flags & MPOOL_DIRTY && 296 mpool_write(mp, bp) == RET_ERROR) 297 return (RET_ERROR); 298 299 /* Sync the file descriptor. */ 300 return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 301 } 302 303 /* 304 * mpool_bkt 305 * Get a page from the cache (or create one). 306 */ 307 static BKT * 308 mpool_bkt(MPOOL *mp) 309 { 310 struct _hqh *head; 311 BKT *bp; 312 313 /* If under the max cached, always create a new page. */ 314 if (mp->curcache < mp->maxcache) 315 goto new; 316 317 /* 318 * If the cache is max'd out, walk the lru list for a buffer we 319 * can flush. If we find one, write it (if necessary) and take it 320 * off any lists. If we don't find anything we grow the cache anyway. 321 * The cache never shrinks. 322 */ 323 for (bp = mp->lqh.cqh_first; 324 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) 325 if (!(bp->flags & MPOOL_PINNED)) { 326 /* Flush if dirty. */ 327 if (bp->flags & MPOOL_DIRTY && 328 mpool_write(mp, bp) == RET_ERROR) 329 return (NULL); 330 #ifdef STATISTICS 331 ++mp->pageflush; 332 #endif 333 /* Remove from the hash and lru queues. */ 334 head = &mp->hqh[HASHKEY(bp->pgno)]; 335 CIRCLEQ_REMOVE(head, bp, hq); 336 CIRCLEQ_REMOVE(&mp->lqh, bp, q); 337 #ifdef DEBUG 338 { 339 void *spage = bp->page; 340 (void)memset(bp, 0xff, 341 (size_t)(sizeof(BKT) + mp->pagesize)); 342 bp->page = spage; 343 } 344 #endif 345 return (bp); 346 } 347 348 new: if ((bp = calloc(1, (size_t)(sizeof(BKT) + mp->pagesize))) == NULL) 349 return (NULL); 350 #ifdef STATISTICS 351 ++mp->pagealloc; 352 #endif 353 #if defined(DEBUG) || defined(PURIFY) 354 (void)memset(bp, 0xff, (size_t)(sizeof(BKT) + mp->pagesize)); 355 #endif 356 bp->page = (char *)(void *)bp + sizeof(BKT); 357 ++mp->curcache; 358 return (bp); 359 } 360 361 /* 362 * mpool_write 363 * Write a page to disk. 364 */ 365 static int 366 mpool_write(MPOOL *mp, BKT *bp) 367 { 368 off_t off; 369 370 #ifdef STATISTICS 371 ++mp->pagewrite; 372 #endif 373 374 /* Run through the user's filter. */ 375 if (mp->pgout) 376 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 377 378 off = mp->pagesize * bp->pgno; 379 if (pwrite(mp->fd, bp->page, (size_t)mp->pagesize, off) != (int)mp->pagesize) 380 return (RET_ERROR); 381 382 /* 383 * Re-run through the input filter since this page may soon be 384 * accessed via the cache, and whatever the user's output filter 385 * did may screw things up if we don't let the input filter 386 * restore the in-core copy. 387 */ 388 if (mp->pgin) 389 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 390 391 bp->flags &= ~MPOOL_DIRTY; 392 return (RET_SUCCESS); 393 } 394 395 /* 396 * mpool_look 397 * Lookup a page in the cache. 398 */ 399 static BKT * 400 mpool_look(MPOOL *mp, pgno_t pgno) 401 { 402 struct _hqh *head; 403 BKT *bp; 404 405 head = &mp->hqh[HASHKEY(pgno)]; 406 for (bp = head->cqh_first; bp != (void *)head; bp = bp->hq.cqe_next) 407 if (bp->pgno == pgno) { 408 #ifdef STATISTICS 409 ++mp->cachehit; 410 #endif 411 return (bp); 412 } 413 #ifdef STATISTICS 414 ++mp->cachemiss; 415 #endif 416 return (NULL); 417 } 418 419 #ifdef STATISTICS 420 /* 421 * mpool_stat 422 * Print out cache statistics. 423 */ 424 void 425 mpool_stat(mp) 426 MPOOL *mp; 427 { 428 BKT *bp; 429 int cnt; 430 const char *sep; 431 432 (void)fprintf(stderr, "%lu pages in the file\n", (u_long)mp->npages); 433 (void)fprintf(stderr, 434 "page size %lu, cacheing %lu pages of %lu page max cache\n", 435 (u_long)mp->pagesize, (u_long)mp->curcache, (u_long)mp->maxcache); 436 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 437 mp->pageput, mp->pageget, mp->pagenew); 438 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 439 mp->pagealloc, mp->pageflush); 440 if (mp->cachehit + mp->cachemiss) 441 (void)fprintf(stderr, 442 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 443 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 444 * 100, mp->cachehit, mp->cachemiss); 445 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 446 mp->pageread, mp->pagewrite); 447 448 sep = ""; 449 cnt = 0; 450 for (bp = mp->lqh.cqh_first; 451 bp != (void *)&mp->lqh; bp = bp->q.cqe_next) { 452 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 453 if (bp->flags & MPOOL_DIRTY) 454 (void)fprintf(stderr, "d"); 455 if (bp->flags & MPOOL_PINNED) 456 (void)fprintf(stderr, "P"); 457 if (++cnt == 10) { 458 sep = "\n"; 459 cnt = 0; 460 } else 461 sep = ", "; 462 463 } 464 (void)fprintf(stderr, "\n"); 465 } 466 #endif 467