xref: /openbsd-src/lib/libc/db/mpool/mpool.c (revision 5ad04d351680822078003e2b066cfc9680d6157d)
1 /*	$OpenBSD: mpool.c,v 1.19 2013/12/02 02:28:21 krw 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/param.h>
33 #include <sys/queue.h>
34 #include <sys/stat.h>
35 
36 #include <errno.h>
37 #include <stdio.h>
38 #include <stdlib.h>
39 #include <string.h>
40 #include <unistd.h>
41 
42 #include <db.h>
43 
44 #define	__MPOOLINTERFACE_PRIVATE
45 #include <mpool.h>
46 
47 static BKT *mpool_bkt(MPOOL *);
48 static BKT *mpool_look(MPOOL *, pgno_t);
49 static int  mpool_write(MPOOL *, BKT *);
50 
51 /*
52  * mpool_open --
53  *	Initialize a memory pool.
54  */
55 /* ARGSUSED */
56 MPOOL *
57 mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache)
58 {
59 	struct stat sb;
60 	MPOOL *mp;
61 	int entry;
62 
63 	/*
64 	 * Get information about the file.
65 	 *
66 	 * XXX
67 	 * We don't currently handle pipes, although we should.
68 	 */
69 	if (fstat(fd, &sb))
70 		return (NULL);
71 	if (!S_ISREG(sb.st_mode)) {
72 		errno = ESPIPE;
73 		return (NULL);
74 	}
75 
76 	/* Allocate and initialize the MPOOL cookie. */
77 	if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL)
78 		return (NULL);
79 	TAILQ_INIT(&mp->lqh);
80 	for (entry = 0; entry < HASHSIZE; ++entry)
81 		TAILQ_INIT(&mp->hqh[entry]);
82 	mp->maxcache = maxcache;
83 	mp->npages = sb.st_size / pagesize;
84 	mp->pagesize = pagesize;
85 	mp->fd = fd;
86 	return (mp);
87 }
88 
89 /*
90  * mpool_filter --
91  *	Initialize input/output filters.
92  */
93 void
94 mpool_filter(MPOOL *mp, void (*pgin) (void *, pgno_t, void *),
95     void (*pgout) (void *, pgno_t, void *), void *pgcookie)
96 {
97 	mp->pgin = pgin;
98 	mp->pgout = pgout;
99 	mp->pgcookie = pgcookie;
100 }
101 
102 /*
103  * mpool_new --
104  *	Get a new page of memory.
105  */
106 void *
107 mpool_new(MPOOL *mp, pgno_t *pgnoaddr, u_int flags)
108 {
109 	struct _hqh *head;
110 	BKT *bp;
111 
112 	if (mp->npages == MAX_PAGE_NUMBER) {
113 		(void)fprintf(stderr, "mpool_new: page allocation overflow.\n");
114 		abort();
115 	}
116 #ifdef STATISTICS
117 	++mp->pagenew;
118 #endif
119 	/*
120 	 * Get a BKT from the cache.  Assign a new page number, attach
121 	 * it to the head of the hash chain, the tail of the lru chain,
122 	 * and return.
123 	 */
124 	if ((bp = mpool_bkt(mp)) == NULL)
125 		return (NULL);
126 	if (flags == MPOOL_PAGE_REQUEST) {
127 		mp->npages++;
128 		bp->pgno = *pgnoaddr;
129 	} else
130 		bp->pgno = *pgnoaddr = mp->npages++;
131 
132 	bp->flags = MPOOL_PINNED | MPOOL_INUSE;
133 
134 	head = &mp->hqh[HASHKEY(bp->pgno)];
135 	TAILQ_INSERT_HEAD(head, bp, hq);
136 	TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
137 	return (bp->page);
138 }
139 
140 int
141 mpool_delete(MPOOL *mp, void *page)
142 {
143 	struct _hqh *head;
144 	BKT *bp;
145 
146 	bp = (BKT *)((char *)page - sizeof(BKT));
147 
148 #ifdef DEBUG
149 	if (!(bp->flags & MPOOL_PINNED)) {
150 		(void)fprintf(stderr,
151 		    "mpool_delete: page %d not pinned\n", bp->pgno);
152 		abort();
153 	}
154 #endif
155 
156 	/* Remove from the hash and lru queues. */
157 	head = &mp->hqh[HASHKEY(bp->pgno)];
158 	TAILQ_REMOVE(head, bp, hq);
159 	TAILQ_REMOVE(&mp->lqh, bp, q);
160 
161 	free(bp);
162 	mp->curcache--;
163 	return (RET_SUCCESS);
164 }
165 
166 /*
167  * mpool_get
168  *	Get a page.
169  */
170 /* ARGSUSED */
171 void *
172 mpool_get(MPOOL *mp, pgno_t pgno,
173     u_int flags)		/* XXX not used? */
174 {
175 	struct _hqh *head;
176 	BKT *bp;
177 	off_t off;
178 	int nr;
179 
180 #ifdef STATISTICS
181 	++mp->pageget;
182 #endif
183 
184 	/* Check for a page that is cached. */
185 	if ((bp = mpool_look(mp, pgno)) != NULL) {
186 #ifdef DEBUG
187 		if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) {
188 			(void)fprintf(stderr,
189 			    "mpool_get: page %d already pinned\n", bp->pgno);
190 			abort();
191 		}
192 #endif
193 		/*
194 		 * Move the page to the head of the hash chain and the tail
195 		 * of the lru chain.
196 		 */
197 		head = &mp->hqh[HASHKEY(bp->pgno)];
198 		TAILQ_REMOVE(head, bp, hq);
199 		TAILQ_INSERT_HEAD(head, bp, hq);
200 		TAILQ_REMOVE(&mp->lqh, bp, q);
201 		TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
202 
203 		/* Return a pinned page. */
204 		bp->flags |= MPOOL_PINNED;
205 		return (bp->page);
206 	}
207 
208 	/* Get a page from the cache. */
209 	if ((bp = mpool_bkt(mp)) == NULL)
210 		return (NULL);
211 
212 	/* Read in the contents. */
213 	off = mp->pagesize * pgno;
214 	if ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != mp->pagesize) {
215 		switch (nr) {
216 		case -1:
217 			/* errno is set for us by pread(). */
218 			free(bp);
219 			mp->curcache--;
220 			return (NULL);
221 		case 0:
222 			/*
223 			 * A zero-length read means you need to create a
224 			 * new page.
225 			 */
226 			memset(bp->page, 0, mp->pagesize);
227 			break;
228 		default:
229 			/* A partial read is definitely bad. */
230 			free(bp);
231 			mp->curcache--;
232 			errno = EINVAL;
233 			return (NULL);
234 		}
235 	}
236 #ifdef STATISTICS
237 	++mp->pageread;
238 #endif
239 
240 	/* Set the page number, pin the page. */
241 	bp->pgno = pgno;
242 	if (!(flags & MPOOL_IGNOREPIN))
243 		bp->flags = MPOOL_PINNED;
244 	bp->flags |= MPOOL_INUSE;
245 
246 	/*
247 	 * Add the page to the head of the hash chain and the tail
248 	 * of the lru chain.
249 	 */
250 	head = &mp->hqh[HASHKEY(bp->pgno)];
251 	TAILQ_INSERT_HEAD(head, bp, hq);
252 	TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
253 
254 	/* Run through the user's filter. */
255 	if (mp->pgin != NULL)
256 		(mp->pgin)(mp->pgcookie, bp->pgno, bp->page);
257 
258 	return (bp->page);
259 }
260 
261 /*
262  * mpool_put
263  *	Return a page.
264  */
265 /* ARGSUSED */
266 int
267 mpool_put(MPOOL *mp, void *page, u_int flags)
268 {
269 	BKT *bp;
270 
271 #ifdef STATISTICS
272 	++mp->pageput;
273 #endif
274 	bp = (BKT *)((char *)page - sizeof(BKT));
275 #ifdef DEBUG
276 	if (!(bp->flags & MPOOL_PINNED)) {
277 		(void)fprintf(stderr,
278 		    "mpool_put: page %d not pinned\n", bp->pgno);
279 		abort();
280 	}
281 #endif
282 	bp->flags &= ~MPOOL_PINNED;
283 	if (flags & MPOOL_DIRTY)
284 		bp->flags |= flags & MPOOL_DIRTY;
285 	return (RET_SUCCESS);
286 }
287 
288 /*
289  * mpool_close
290  *	Close the buffer pool.
291  */
292 int
293 mpool_close(MPOOL *mp)
294 {
295 	BKT *bp;
296 
297 	/* Free up any space allocated to the lru pages. */
298 	while ((bp = TAILQ_FIRST(&mp->lqh))) {
299 		TAILQ_REMOVE(&mp->lqh, bp, q);
300 		free(bp);
301 	}
302 
303 	/* Free the MPOOL cookie. */
304 	free(mp);
305 	return (RET_SUCCESS);
306 }
307 
308 /*
309  * mpool_sync
310  *	Sync the pool to disk.
311  */
312 int
313 mpool_sync(MPOOL *mp)
314 {
315 	BKT *bp;
316 
317 	/* Walk the lru chain, flushing any dirty pages to disk. */
318 	TAILQ_FOREACH(bp, &mp->lqh, q)
319 		if (bp->flags & MPOOL_DIRTY &&
320 		    mpool_write(mp, bp) == RET_ERROR)
321 			return (RET_ERROR);
322 
323 	/* Sync the file descriptor. */
324 	return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS);
325 }
326 
327 /*
328  * mpool_bkt
329  *	Get a page from the cache (or create one).
330  */
331 static BKT *
332 mpool_bkt(MPOOL *mp)
333 {
334 	struct _hqh *head;
335 	BKT *bp;
336 
337 	/* If under the max cached, always create a new page. */
338 	if (mp->curcache < mp->maxcache)
339 		goto new;
340 
341 	/*
342 	 * If the cache is max'd out, walk the lru list for a buffer we
343 	 * can flush.  If we find one, write it (if necessary) and take it
344 	 * off any lists.  If we don't find anything we grow the cache anyway.
345 	 * The cache never shrinks.
346 	 */
347 	TAILQ_FOREACH(bp, &mp->lqh, q)
348 		if (!(bp->flags & MPOOL_PINNED)) {
349 			/* Flush if dirty. */
350 			if (bp->flags & MPOOL_DIRTY &&
351 			    mpool_write(mp, bp) == RET_ERROR)
352 				return (NULL);
353 #ifdef STATISTICS
354 			++mp->pageflush;
355 #endif
356 			/* Remove from the hash and lru queues. */
357 			head = &mp->hqh[HASHKEY(bp->pgno)];
358 			TAILQ_REMOVE(head, bp, hq);
359 			TAILQ_REMOVE(&mp->lqh, bp, q);
360 #ifdef DEBUG
361 			{ void *spage;
362 				spage = bp->page;
363 				memset(bp, 0xff, sizeof(BKT) + mp->pagesize);
364 				bp->page = spage;
365 			}
366 #endif
367 			bp->flags = 0;
368 			return (bp);
369 		}
370 
371 new:	if ((bp = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL)
372 		return (NULL);
373 #ifdef STATISTICS
374 	++mp->pagealloc;
375 #endif
376 	memset(bp, 0xff, sizeof(BKT) + mp->pagesize);
377 	bp->page = (char *)bp + sizeof(BKT);
378 	bp->flags = 0;
379 	++mp->curcache;
380 	return (bp);
381 }
382 
383 /*
384  * mpool_write
385  *	Write a page to disk.
386  */
387 static int
388 mpool_write(MPOOL *mp, BKT *bp)
389 {
390 	off_t off;
391 
392 #ifdef STATISTICS
393 	++mp->pagewrite;
394 #endif
395 
396 	/* Run through the user's filter. */
397 	if (mp->pgout)
398 		(mp->pgout)(mp->pgcookie, bp->pgno, bp->page);
399 
400 	off = mp->pagesize * bp->pgno;
401 	if (pwrite(mp->fd, bp->page, mp->pagesize, off) != mp->pagesize)
402 		return (RET_ERROR);
403 
404 	/*
405 	 * Re-run through the input filter since this page may soon be
406 	 * accessed via the cache, and whatever the user's output filter
407 	 * did may screw things up if we don't let the input filter
408 	 * restore the in-core copy.
409 	 */
410 	if (mp->pgin)
411 		(mp->pgin)(mp->pgcookie, bp->pgno, bp->page);
412 
413 	bp->flags &= ~MPOOL_DIRTY;
414 	return (RET_SUCCESS);
415 }
416 
417 /*
418  * mpool_look
419  *	Lookup a page in the cache.
420  */
421 static BKT *
422 mpool_look(MPOOL *mp, pgno_t pgno)
423 {
424 	struct _hqh *head;
425 	BKT *bp;
426 
427 	head = &mp->hqh[HASHKEY(pgno)];
428 	TAILQ_FOREACH(bp, head, hq)
429 		if ((bp->pgno == pgno) &&
430 			((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) {
431 #ifdef STATISTICS
432 			++mp->cachehit;
433 #endif
434 			return (bp);
435 		}
436 #ifdef STATISTICS
437 	++mp->cachemiss;
438 #endif
439 	return (NULL);
440 }
441 
442 #ifdef STATISTICS
443 /*
444  * mpool_stat
445  *	Print out cache statistics.
446  */
447 void
448 mpool_stat(MPOOL *mp)
449 {
450 	BKT *bp;
451 	int cnt;
452 	char *sep;
453 
454 	(void)fprintf(stderr, "%lu pages in the file\n", mp->npages);
455 	(void)fprintf(stderr,
456 	    "page size %lu, cacheing %lu pages of %lu page max cache\n",
457 	    mp->pagesize, mp->curcache, mp->maxcache);
458 	(void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n",
459 	    mp->pageput, mp->pageget, mp->pagenew);
460 	(void)fprintf(stderr, "%lu page allocs, %lu page flushes\n",
461 	    mp->pagealloc, mp->pageflush);
462 	if (mp->cachehit + mp->cachemiss)
463 		(void)fprintf(stderr,
464 		    "%.0f%% cache hit rate (%lu hits, %lu misses)\n",
465 		    ((double)mp->cachehit / (mp->cachehit + mp->cachemiss))
466 		    * 100, mp->cachehit, mp->cachemiss);
467 	(void)fprintf(stderr, "%lu page reads, %lu page writes\n",
468 	    mp->pageread, mp->pagewrite);
469 
470 	sep = "";
471 	cnt = 0;
472 	TAILQ_FOREACH(bp, &mp->lqh, q) {
473 		(void)fprintf(stderr, "%s%d", sep, bp->pgno);
474 		if (bp->flags & MPOOL_DIRTY)
475 			(void)fprintf(stderr, "d");
476 		if (bp->flags & MPOOL_PINNED)
477 			(void)fprintf(stderr, "P");
478 		if (++cnt == 10) {
479 			sep = "\n";
480 			cnt = 0;
481 		} else
482 			sep = ", ";
483 
484 	}
485 	(void)fprintf(stderr, "\n");
486 }
487 #endif
488