1 /*- 2 * Copyright (c) 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)mpool.c 8.7 (Berkeley) 11/2/95 30 * $FreeBSD: head/lib/libc/db/mpool/mpool.c 194804 2009-06-24 01:15:10Z delphij $ 31 */ 32 33 #include "namespace.h" 34 #include <sys/param.h> 35 #include <sys/queue.h> 36 #include <sys/stat.h> 37 38 #include <errno.h> 39 #include <stdio.h> 40 #include <stdlib.h> 41 #include <string.h> 42 #include <unistd.h> 43 #include "un-namespace.h" 44 45 #include <db.h> 46 47 #define __MPOOLINTERFACE_PRIVATE 48 #include <mpool.h> 49 50 static BKT *mpool_bkt(MPOOL *); 51 static BKT *mpool_look(MPOOL *, pgno_t); 52 static int mpool_write(MPOOL *, BKT *); 53 54 /* 55 * mpool_open -- 56 * Initialize a memory pool. 57 */ 58 /* ARGSUSED */ 59 MPOOL * 60 mpool_open(__unused void *key, int fd, pgno_t pagesize, pgno_t maxcache) 61 { 62 struct stat sb; 63 MPOOL *mp; 64 int entry; 65 66 /* 67 * Get information about the file. 68 * 69 * XXX 70 * We don't currently handle pipes, although we should. 71 */ 72 if (_fstat(fd, &sb)) 73 return (NULL); 74 if (!S_ISREG(sb.st_mode)) { 75 errno = ESPIPE; 76 return (NULL); 77 } 78 79 /* Allocate and initialize the MPOOL cookie. */ 80 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 81 return (NULL); 82 TAILQ_INIT(&mp->lqh); 83 for (entry = 0; entry < HASHSIZE; ++entry) 84 TAILQ_INIT(&mp->hqh[entry]); 85 mp->maxcache = maxcache; 86 mp->npages = sb.st_size / pagesize; 87 mp->pagesize = pagesize; 88 mp->fd = fd; 89 return (mp); 90 } 91 92 /* 93 * mpool_filter -- 94 * Initialize input/output filters. 95 */ 96 void 97 mpool_filter(MPOOL *mp, void (*pgin)(void *, pgno_t, void *), 98 void (*pgout) (void *, pgno_t, void *), void *pgcookie) 99 { 100 mp->pgin = pgin; 101 mp->pgout = pgout; 102 mp->pgcookie = pgcookie; 103 } 104 105 /* 106 * mpool_new -- 107 * Get a new page of memory. 108 */ 109 void * 110 mpool_new(MPOOL *mp, pgno_t *pgnoaddr, unsigned int flags) 111 { 112 struct _hqh *head; 113 BKT *bp; 114 115 if (mp->npages == MAX_PAGE_NUMBER) { 116 fprintf(stderr, "mpool_new: page allocation overflow.\n"); 117 abort(); 118 } 119 #ifdef STATISTICS 120 ++mp->pagenew; 121 #endif 122 /* 123 * Get a BKT from the cache. Assign a new page number, attach 124 * it to the head of the hash chain, the tail of the lru chain, 125 * and return. 126 */ 127 if ((bp = mpool_bkt(mp)) == NULL) 128 return (NULL); 129 if (flags == MPOOL_PAGE_REQUEST) { 130 mp->npages++; 131 bp->pgno = *pgnoaddr; 132 } else 133 bp->pgno = *pgnoaddr = mp->npages++; 134 135 bp->flags = MPOOL_PINNED | MPOOL_INUSE; 136 137 head = &mp->hqh[HASHKEY(bp->pgno)]; 138 TAILQ_INSERT_HEAD(head, bp, hq); 139 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 140 return (bp->page); 141 } 142 143 int 144 mpool_delete(MPOOL *mp, void *page) 145 { 146 struct _hqh *head; 147 BKT *bp; 148 149 bp = (BKT *)((char *)page - sizeof(BKT)); 150 151 #ifdef DEBUG 152 if (!(bp->flags & MPOOL_PINNED)) { 153 fprintf(stderr, 154 "mpool_delete: page %d not pinned\n", bp->pgno); 155 abort(); 156 } 157 #endif 158 159 /* Remove from the hash and lru queues. */ 160 head = &mp->hqh[HASHKEY(bp->pgno)]; 161 TAILQ_REMOVE(head, bp, hq); 162 TAILQ_REMOVE(&mp->lqh, bp, q); 163 164 free(bp); 165 mp->curcache--; 166 return (RET_SUCCESS); 167 } 168 169 /* 170 * mpool_get 171 * Get a page. 172 */ 173 /* ARGSUSED */ 174 void * 175 mpool_get(MPOOL *mp, pgno_t pgno, 176 unsigned int flags) /* XXX not used? */ 177 { 178 struct _hqh *head; 179 BKT *bp; 180 off_t off; 181 int nr; 182 183 #ifdef STATISTICS 184 ++mp->pageget; 185 #endif 186 187 /* Check for a page that is cached. */ 188 if ((bp = mpool_look(mp, pgno)) != NULL) { 189 #ifdef DEBUG 190 if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) { 191 fprintf(stderr, 192 "mpool_get: page %d already pinned\n", bp->pgno); 193 abort(); 194 } 195 #endif 196 /* 197 * Move the page to the head of the hash chain and the tail 198 * of the lru chain. 199 */ 200 head = &mp->hqh[HASHKEY(bp->pgno)]; 201 TAILQ_REMOVE(head, bp, hq); 202 TAILQ_INSERT_HEAD(head, bp, hq); 203 TAILQ_REMOVE(&mp->lqh, bp, q); 204 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 205 206 /* Return a pinned page. */ 207 bp->flags |= MPOOL_PINNED; 208 return (bp->page); 209 } 210 211 /* Get a page from the cache. */ 212 if ((bp = mpool_bkt(mp)) == NULL) 213 return (NULL); 214 215 /* Read in the contents. */ 216 off = mp->pagesize * pgno; 217 if ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != (ssize_t)mp->pagesize) { 218 switch (nr) { 219 case -1: 220 /* errno is set for us by pread(). */ 221 free(bp); 222 mp->curcache--; 223 return (NULL); 224 case 0: 225 /* 226 * A zero-length read means you need to create a 227 * new page. 228 */ 229 memset(bp->page, 0, mp->pagesize); 230 break; 231 default: 232 /* A partial read is definitely bad. */ 233 free(bp); 234 mp->curcache--; 235 errno = EINVAL; 236 return (NULL); 237 } 238 } 239 #ifdef STATISTICS 240 ++mp->pageread; 241 #endif 242 243 /* Set the page number, pin the page. */ 244 bp->pgno = pgno; 245 if (!(flags & MPOOL_IGNOREPIN)) 246 bp->flags = MPOOL_PINNED; 247 bp->flags |= MPOOL_INUSE; 248 249 /* 250 * Add the page to the head of the hash chain and the tail 251 * of the lru chain. 252 */ 253 head = &mp->hqh[HASHKEY(bp->pgno)]; 254 TAILQ_INSERT_HEAD(head, bp, hq); 255 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 256 257 /* Run through the user's filter. */ 258 if (mp->pgin != NULL) 259 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 260 261 return (bp->page); 262 } 263 264 /* 265 * mpool_put 266 * Return a page. 267 */ 268 /* ARGSUSED */ 269 int 270 mpool_put(MPOOL *mp, void *page, unsigned int flags) 271 { 272 BKT *bp; 273 274 #ifdef STATISTICS 275 ++mp->pageput; 276 #endif 277 bp = (BKT *)((char *)page - sizeof(BKT)); 278 #ifdef DEBUG 279 if (!(bp->flags & MPOOL_PINNED)) { 280 fprintf(stderr, 281 "mpool_put: page %d not pinned\n", bp->pgno); 282 abort(); 283 } 284 #endif 285 bp->flags &= ~MPOOL_PINNED; 286 if (flags & MPOOL_DIRTY) 287 bp->flags |= flags & MPOOL_DIRTY; 288 return (RET_SUCCESS); 289 } 290 291 /* 292 * mpool_close 293 * Close the buffer pool. 294 */ 295 int 296 mpool_close(MPOOL *mp) 297 { 298 BKT *bp; 299 300 /* Free up any space allocated to the lru pages. */ 301 while (!TAILQ_EMPTY(&mp->lqh)) { 302 bp = TAILQ_FIRST(&mp->lqh); 303 TAILQ_REMOVE(&mp->lqh, bp, q); 304 free(bp); 305 } 306 307 /* Free the MPOOL cookie. */ 308 free(mp); 309 return (RET_SUCCESS); 310 } 311 312 /* 313 * mpool_sync 314 * Sync the pool to disk. 315 */ 316 int 317 mpool_sync(MPOOL *mp) 318 { 319 BKT *bp; 320 321 /* Walk the lru chain, flushing any dirty pages to disk. */ 322 TAILQ_FOREACH(bp, &mp->lqh, q) 323 if (bp->flags & MPOOL_DIRTY && 324 mpool_write(mp, bp) == RET_ERROR) 325 return (RET_ERROR); 326 327 /* Sync the file descriptor. */ 328 return (_fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 329 } 330 331 /* 332 * mpool_bkt 333 * Get a page from the cache (or create one). 334 */ 335 static BKT * 336 mpool_bkt(MPOOL *mp) 337 { 338 struct _hqh *head; 339 BKT *bp; 340 341 /* If under the max cached, always create a new page. */ 342 if (mp->curcache < mp->maxcache) 343 goto new; 344 345 /* 346 * If the cache is max'd out, walk the lru list for a buffer we 347 * can flush. If we find one, write it (if necessary) and take it 348 * off any lists. If we don't find anything we grow the cache anyway. 349 * The cache never shrinks. 350 */ 351 TAILQ_FOREACH(bp, &mp->lqh, q) 352 if (!(bp->flags & MPOOL_PINNED)) { 353 /* Flush if dirty. */ 354 if (bp->flags & MPOOL_DIRTY && 355 mpool_write(mp, bp) == RET_ERROR) 356 return (NULL); 357 #ifdef STATISTICS 358 ++mp->pageflush; 359 #endif 360 /* Remove from the hash and lru queues. */ 361 head = &mp->hqh[HASHKEY(bp->pgno)]; 362 TAILQ_REMOVE(head, bp, hq); 363 TAILQ_REMOVE(&mp->lqh, bp, q); 364 #ifdef DEBUG 365 { void *spage; 366 spage = bp->page; 367 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 368 bp->page = spage; 369 } 370 #endif 371 bp->flags = 0; 372 return (bp); 373 } 374 375 new: if ((bp = (BKT *)calloc(1, sizeof(BKT) + mp->pagesize)) == NULL) 376 return (NULL); 377 #ifdef STATISTICS 378 ++mp->pagealloc; 379 #endif 380 bp->page = (char *)bp + sizeof(BKT); 381 bp->flags = 0; 382 ++mp->curcache; 383 return (bp); 384 } 385 386 /* 387 * mpool_write 388 * Write a page to disk. 389 */ 390 static int 391 mpool_write(MPOOL *mp, BKT *bp) 392 { 393 off_t off; 394 395 #ifdef STATISTICS 396 ++mp->pagewrite; 397 #endif 398 399 /* Run through the user's filter. */ 400 if (mp->pgout) 401 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 402 403 off = mp->pagesize * bp->pgno; 404 if (pwrite(mp->fd, bp->page, mp->pagesize, off) != (ssize_t)mp->pagesize) 405 return (RET_ERROR); 406 407 /* 408 * Re-run through the input filter since this page may soon be 409 * accessed via the cache, and whatever the user's output filter 410 * did may screw things up if we don't let the input filter 411 * restore the in-core copy. 412 */ 413 if (mp->pgin) 414 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 415 416 bp->flags &= ~MPOOL_DIRTY; 417 return (RET_SUCCESS); 418 } 419 420 /* 421 * mpool_look 422 * Lookup a page in the cache. 423 */ 424 static BKT * 425 mpool_look(MPOOL *mp, pgno_t pgno) 426 { 427 struct _hqh *head; 428 BKT *bp; 429 430 head = &mp->hqh[HASHKEY(pgno)]; 431 TAILQ_FOREACH(bp, head, hq) 432 if ((bp->pgno == pgno) && 433 ((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) { 434 #ifdef STATISTICS 435 ++mp->cachehit; 436 #endif 437 return (bp); 438 } 439 #ifdef STATISTICS 440 ++mp->cachemiss; 441 #endif 442 return (NULL); 443 } 444 445 #ifdef STATISTICS 446 /* 447 * mpool_stat 448 * Print out cache statistics. 449 */ 450 void 451 mpool_stat(MPOOL *mp) 452 { 453 BKT *bp; 454 int cnt; 455 char *sep; 456 457 fprintf(stderr, "%lu pages in the file\n", mp->npages); 458 fprintf(stderr, 459 "page size %lu, caching %lu pages of %lu page max cache\n", 460 mp->pagesize, mp->curcache, mp->maxcache); 461 fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 462 mp->pageput, mp->pageget, mp->pagenew); 463 fprintf(stderr, "%lu page allocs, %lu page flushes\n", 464 mp->pagealloc, mp->pageflush); 465 if (mp->cachehit + mp->cachemiss) 466 fprintf(stderr, 467 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 468 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 469 * 100, mp->cachehit, mp->cachemiss); 470 fprintf(stderr, "%lu page reads, %lu page writes\n", 471 mp->pageread, mp->pagewrite); 472 473 sep = ""; 474 cnt = 0; 475 TAILQ_FOREACH(bp, &mp->lqh, q) { 476 fprintf(stderr, "%s%d", sep, bp->pgno); 477 if (bp->flags & MPOOL_DIRTY) 478 fprintf(stderr, "d"); 479 if (bp->flags & MPOOL_PINNED) 480 fprintf(stderr, "P"); 481 if (++cnt == 10) { 482 sep = "\n"; 483 cnt = 0; 484 } else 485 sep = ", "; 486 } 487 fprintf(stderr, "\n"); 488 } 489 #endif 490