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 * 4. 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 30#if defined(LIBC_SCCS) && !defined(lint) 31static char sccsid[] = "@(#)mpool.c 8.7 (Berkeley) 11/2/95"; 32#endif /* LIBC_SCCS and not lint */ 33#include <sys/cdefs.h> 34__FBSDID("$FreeBSD$"); 35 36#include "namespace.h" 37#include <sys/param.h> 38#include <sys/queue.h> 39#include <sys/stat.h> 40 41#include <errno.h> 42#include <stdio.h> 43#include <stdlib.h> 44#include <string.h> 45#include <unistd.h> 46#include "un-namespace.h" 47 48#include <db.h> 49 50#define __MPOOLINTERFACE_PRIVATE 51#include <mpool.h> 52 53static BKT *mpool_bkt(MPOOL *); 54static BKT *mpool_look(MPOOL *, pgno_t); 55static int mpool_write(MPOOL *, BKT *); 56 57/* 58 * mpool_open -- 59 * Initialize a memory pool. 60 */ 61/* ARGSUSED */ 62MPOOL * 63mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache) 64{ 65 struct stat sb; 66 MPOOL *mp; 67 int entry; 68 69 /* 70 * Get information about the file. 71 * 72 * XXX 73 * We don't currently handle pipes, although we should. 74 */ 75 if (_fstat(fd, &sb)) 76 return (NULL); 77 if (!S_ISREG(sb.st_mode)) { 78 errno = ESPIPE; 79 return (NULL); 80 } 81 82 /* Allocate and initialize the MPOOL cookie. */ 83 if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL) 84 return (NULL); 85 TAILQ_INIT(&mp->lqh); 86 for (entry = 0; entry < HASHSIZE; ++entry) 87 TAILQ_INIT(&mp->hqh[entry]); 88 mp->maxcache = maxcache; 89 mp->npages = sb.st_size / pagesize; 90 mp->pagesize = pagesize; 91 mp->fd = fd; 92 return (mp); 93} 94 95/* 96 * mpool_filter -- 97 * Initialize input/output filters. 98 */ 99void 100mpool_filter(MPOOL *mp, void (*pgin) (void *, pgno_t, void *), 101 void (*pgout) (void *, pgno_t, void *), void *pgcookie) 102{ 103 mp->pgin = pgin; 104 mp->pgout = pgout; 105 mp->pgcookie = pgcookie; 106} 107 108/* 109 * mpool_new -- 110 * Get a new page of memory. 111 */ 112void * 113mpool_new(MPOOL *mp, pgno_t *pgnoaddr, u_int flags) 114{ 115 struct _hqh *head; 116 BKT *bp; 117 118 if (mp->npages == MAX_PAGE_NUMBER) { 119 (void)fprintf(stderr, "mpool_new: page allocation overflow.\n"); 120 abort(); 121 } 122#ifdef STATISTICS 123 ++mp->pagenew; 124#endif 125 /* 126 * Get a BKT from the cache. Assign a new page number, attach 127 * it to the head of the hash chain, the tail of the lru chain, 128 * and return. 129 */ 130 if ((bp = mpool_bkt(mp)) == NULL) 131 return (NULL); 132 if (flags == MPOOL_PAGE_REQUEST) { 133 mp->npages++; 134 bp->pgno = *pgnoaddr; 135 } else 136 bp->pgno = *pgnoaddr = mp->npages++; 137 138 bp->flags = MPOOL_PINNED | MPOOL_INUSE; 139 140 head = &mp->hqh[HASHKEY(bp->pgno)]; 141 TAILQ_INSERT_HEAD(head, bp, hq); 142 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 143 return (bp->page); 144} 145 146int 147mpool_delete(MPOOL *mp, void *page) 148{ 149 struct _hqh *head; 150 BKT *bp; 151 152 bp = (BKT *)((char *)page - sizeof(BKT)); 153 154#ifdef DEBUG 155 if (!(bp->flags & MPOOL_PINNED)) { 156 (void)fprintf(stderr, 157 "mpool_delete: page %d not pinned\n", bp->pgno); 158 abort(); 159 } 160#endif 161 162 /* Remove from the hash and lru queues. */ 163 head = &mp->hqh[HASHKEY(bp->pgno)]; 164 TAILQ_REMOVE(head, bp, hq); 165 TAILQ_REMOVE(&mp->lqh, bp, q); 166 167 free(bp); 168 mp->curcache--; 169 return (RET_SUCCESS); 170} 171 172/* 173 * mpool_get 174 * Get a page. 175 */ 176/* ARGSUSED */ 177void * 178mpool_get(MPOOL *mp, pgno_t pgno, 179 u_int flags) /* XXX not used? */ 180{ 181 struct _hqh *head; 182 BKT *bp; 183 off_t off; 184 int nr; 185 186#ifdef STATISTICS 187 ++mp->pageget; 188#endif 189 190 /* Check for a page that is cached. */ 191 if ((bp = mpool_look(mp, pgno)) != NULL) { 192#ifdef DEBUG 193 if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) { 194 (void)fprintf(stderr, 195 "mpool_get: page %d already pinned\n", bp->pgno); 196 abort(); 197 } 198#endif 199 /* 200 * Move the page to the head of the hash chain and the tail 201 * of the lru chain. 202 */ 203 head = &mp->hqh[HASHKEY(bp->pgno)]; 204 TAILQ_REMOVE(head, bp, hq); 205 TAILQ_INSERT_HEAD(head, bp, hq); 206 TAILQ_REMOVE(&mp->lqh, bp, q); 207 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 208 209 /* Return a pinned page. */ 210 bp->flags |= MPOOL_PINNED; 211 return (bp->page); 212 } 213 214 /* Get a page from the cache. */ 215 if ((bp = mpool_bkt(mp)) == NULL) 216 return (NULL); 217 218 /* Read in the contents. */ 219 off = mp->pagesize * pgno; 220 if ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != (ssize_t)mp->pagesize) { 221 switch (nr) { 222 case -1: 223 /* errno is set for us by pread(). */ 224 free(bp); 225 mp->curcache--; 226 return (NULL); 227 case 0: 228 /* 229 * A zero-length read means you need to create a 230 * new page. 231 */ 232 memset(bp->page, 0, mp->pagesize); 233 break; 234 default: 235 /* A partial read is definitely bad. */ 236 free(bp); 237 mp->curcache--; 238 errno = EINVAL; 239 return (NULL); 240 } 241 } 242#ifdef STATISTICS 243 ++mp->pageread; 244#endif 245 246 /* Set the page number, pin the page. */ 247 bp->pgno = pgno; 248 if (!(flags & MPOOL_IGNOREPIN)) 249 bp->flags = MPOOL_PINNED; 250 bp->flags |= MPOOL_INUSE; 251 252 /* 253 * Add the page to the head of the hash chain and the tail 254 * of the lru chain. 255 */ 256 head = &mp->hqh[HASHKEY(bp->pgno)]; 257 TAILQ_INSERT_HEAD(head, bp, hq); 258 TAILQ_INSERT_TAIL(&mp->lqh, bp, q); 259 260 /* Run through the user's filter. */ 261 if (mp->pgin != NULL) 262 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 263 264 return (bp->page); 265} 266 267/* 268 * mpool_put 269 * Return a page. 270 */ 271/* ARGSUSED */ 272int 273mpool_put(MPOOL *mp, void *page, u_int flags) 274{ 275 BKT *bp; 276 277#ifdef STATISTICS 278 ++mp->pageput; 279#endif 280 bp = (BKT *)((char *)page - sizeof(BKT)); 281#ifdef DEBUG 282 if (!(bp->flags & MPOOL_PINNED)) { 283 (void)fprintf(stderr, 284 "mpool_put: page %d not pinned\n", bp->pgno); 285 abort(); 286 } 287#endif 288 bp->flags &= ~MPOOL_PINNED; 289 if (flags & MPOOL_DIRTY) 290 bp->flags |= flags & MPOOL_DIRTY; 291 return (RET_SUCCESS); 292} 293 294/* 295 * mpool_close 296 * Close the buffer pool. 297 */ 298int 299mpool_close(MPOOL *mp) 300{ 301 BKT *bp; 302 303 /* Free up any space allocated to the lru pages. */ 304 while (!TAILQ_EMPTY(&mp->lqh)) { 305 bp = TAILQ_FIRST(&mp->lqh); 306 TAILQ_REMOVE(&mp->lqh, bp, q); 307 free(bp); 308 } 309 310 /* Free the MPOOL cookie. */ 311 free(mp); 312 return (RET_SUCCESS); 313} 314 315/* 316 * mpool_sync 317 * Sync the pool to disk. 318 */ 319int 320mpool_sync(MPOOL *mp) 321{ 322 BKT *bp; 323 324 /* Walk the lru chain, flushing any dirty pages to disk. */ 325 TAILQ_FOREACH(bp, &mp->lqh, q) 326 if (bp->flags & MPOOL_DIRTY && 327 mpool_write(mp, bp) == RET_ERROR) 328 return (RET_ERROR); 329 330 /* Sync the file descriptor. */ 331 return (_fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); 332} 333 334/* 335 * mpool_bkt 336 * Get a page from the cache (or create one). 337 */ 338static BKT * 339mpool_bkt(MPOOL *mp) 340{ 341 struct _hqh *head; 342 BKT *bp; 343 344 /* If under the max cached, always create a new page. */ 345 if (mp->curcache < mp->maxcache) 346 goto new; 347 348 /* 349 * If the cache is max'd out, walk the lru list for a buffer we 350 * can flush. If we find one, write it (if necessary) and take it 351 * off any lists. If we don't find anything we grow the cache anyway. 352 * The cache never shrinks. 353 */ 354 TAILQ_FOREACH(bp, &mp->lqh, q) 355 if (!(bp->flags & MPOOL_PINNED)) { 356 /* Flush if dirty. */ 357 if (bp->flags & MPOOL_DIRTY && 358 mpool_write(mp, bp) == RET_ERROR) 359 return (NULL); 360#ifdef STATISTICS 361 ++mp->pageflush; 362#endif 363 /* Remove from the hash and lru queues. */ 364 head = &mp->hqh[HASHKEY(bp->pgno)]; 365 TAILQ_REMOVE(head, bp, hq); 366 TAILQ_REMOVE(&mp->lqh, bp, q); 367#ifdef DEBUG 368 { void *spage; 369 spage = bp->page; 370 memset(bp, 0xff, sizeof(BKT) + mp->pagesize); 371 bp->page = spage; 372 } 373#endif 374 bp->flags = 0; 375 return (bp); 376 } 377 378new: if ((bp = (BKT *)calloc(1, sizeof(BKT) + mp->pagesize)) == NULL) 379 return (NULL); 380#ifdef STATISTICS 381 ++mp->pagealloc; 382#endif 383 bp->page = (char *)bp + sizeof(BKT); 384 bp->flags = 0; 385 ++mp->curcache; 386 return (bp); 387} 388 389/* 390 * mpool_write 391 * Write a page to disk. 392 */ 393static int 394mpool_write(MPOOL *mp, BKT *bp) 395{ 396 off_t off; 397 398#ifdef STATISTICS 399 ++mp->pagewrite; 400#endif 401 402 /* Run through the user's filter. */ 403 if (mp->pgout) 404 (mp->pgout)(mp->pgcookie, bp->pgno, bp->page); 405 406 off = mp->pagesize * bp->pgno; 407 if (pwrite(mp->fd, bp->page, mp->pagesize, off) != (ssize_t)mp->pagesize) 408 return (RET_ERROR); 409 410 /* 411 * Re-run through the input filter since this page may soon be 412 * accessed via the cache, and whatever the user's output filter 413 * did may screw things up if we don't let the input filter 414 * restore the in-core copy. 415 */ 416 if (mp->pgin) 417 (mp->pgin)(mp->pgcookie, bp->pgno, bp->page); 418 419 bp->flags &= ~MPOOL_DIRTY; 420 return (RET_SUCCESS); 421} 422 423/* 424 * mpool_look 425 * Lookup a page in the cache. 426 */ 427static BKT * 428mpool_look(MPOOL *mp, pgno_t pgno) 429{ 430 struct _hqh *head; 431 BKT *bp; 432 433 head = &mp->hqh[HASHKEY(pgno)]; 434 TAILQ_FOREACH(bp, head, hq) 435 if ((bp->pgno == pgno) && 436 ((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) { 437#ifdef STATISTICS 438 ++mp->cachehit; 439#endif 440 return (bp); 441 } 442#ifdef STATISTICS 443 ++mp->cachemiss; 444#endif 445 return (NULL); 446} 447 448#ifdef STATISTICS 449/* 450 * mpool_stat 451 * Print out cache statistics. 452 */ 453void 454mpool_stat(MPOOL *mp) 455{ 456 BKT *bp; 457 int cnt; 458 char *sep; 459 460 (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); 461 (void)fprintf(stderr, 462 "page size %lu, cacheing %lu pages of %lu page max cache\n", 463 mp->pagesize, mp->curcache, mp->maxcache); 464 (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", 465 mp->pageput, mp->pageget, mp->pagenew); 466 (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", 467 mp->pagealloc, mp->pageflush); 468 if (mp->cachehit + mp->cachemiss) 469 (void)fprintf(stderr, 470 "%.0f%% cache hit rate (%lu hits, %lu misses)\n", 471 ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) 472 * 100, mp->cachehit, mp->cachemiss); 473 (void)fprintf(stderr, "%lu page reads, %lu page writes\n", 474 mp->pageread, mp->pagewrite); 475 476 sep = ""; 477 cnt = 0; 478 TAILQ_FOREACH(bp, &mp->lqh, q) { 479 (void)fprintf(stderr, "%s%d", sep, bp->pgno); 480 if (bp->flags & MPOOL_DIRTY) 481 (void)fprintf(stderr, "d"); 482 if (bp->flags & MPOOL_PINNED) 483 (void)fprintf(stderr, "P"); 484 if (++cnt == 10) { 485 sep = "\n"; 486 cnt = 0; 487 } else 488 sep = ", "; 489 490 } 491 (void)fprintf(stderr, "\n"); 492} 493#endif 494