xref: /freebsd/lib/libc/db/mpool/mpool.c (revision 6137b5f7b8c183ee8806d79b3f1d8e5e3ddb3df3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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 "namespace.h"
33 #include <sys/param.h>
34 #include <sys/queue.h>
35 #include <sys/stat.h>
36 
37 #include <errno.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <unistd.h>
42 #include "un-namespace.h"
43 
44 #include <db.h>
45 
46 #define	__MPOOLINTERFACE_PRIVATE
47 #include <mpool.h>
48 
49 static BKT *mpool_bkt(MPOOL *);
50 static BKT *mpool_look(MPOOL *, pgno_t);
51 static int  mpool_write(MPOOL *, BKT *);
52 
53 /*
54  * mpool_open --
55  *	Initialize a memory pool.
56  */
57 /* ARGSUSED */
58 MPOOL *
59 mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache)
60 {
61 	struct stat sb;
62 	MPOOL *mp;
63 	int entry;
64 
65 	/*
66 	 * Get information about the file.
67 	 *
68 	 * XXX
69 	 * We don't currently handle pipes, although we should.
70 	 */
71 	if (_fstat(fd, &sb))
72 		return (NULL);
73 	if (!S_ISREG(sb.st_mode)) {
74 		errno = ESPIPE;
75 		return (NULL);
76 	}
77 
78 	/* Allocate and initialize the MPOOL cookie. */
79 	if ((mp = (MPOOL *)calloc(1, sizeof(MPOOL))) == NULL)
80 		return (NULL);
81 	TAILQ_INIT(&mp->lqh);
82 	for (entry = 0; entry < HASHSIZE; ++entry)
83 		TAILQ_INIT(&mp->hqh[entry]);
84 	mp->maxcache = maxcache;
85 	mp->npages = sb.st_size / pagesize;
86 	mp->pagesize = pagesize;
87 	mp->fd = fd;
88 	return (mp);
89 }
90 
91 /*
92  * mpool_filter --
93  *	Initialize input/output filters.
94  */
95 void
96 mpool_filter(MPOOL *mp, void (*pgin) (void *, pgno_t, void *),
97     void (*pgout) (void *, pgno_t, void *), void *pgcookie)
98 {
99 	mp->pgin = pgin;
100 	mp->pgout = pgout;
101 	mp->pgcookie = pgcookie;
102 }
103 
104 /*
105  * mpool_new --
106  *	Get a new page of memory.
107  */
108 void *
109 mpool_new(MPOOL *mp, pgno_t *pgnoaddr, u_int flags)
110 {
111 	struct _hqh *head;
112 	BKT *bp;
113 
114 	if (mp->npages == MAX_PAGE_NUMBER) {
115 		(void)fprintf(stderr, "mpool_new: page allocation overflow.\n");
116 		abort();
117 	}
118 #ifdef STATISTICS
119 	++mp->pagenew;
120 #endif
121 	/*
122 	 * Get a BKT from the cache.  Assign a new page number, attach
123 	 * it to the head of the hash chain, the tail of the lru chain,
124 	 * and return.
125 	 */
126 	if ((bp = mpool_bkt(mp)) == NULL)
127 		return (NULL);
128 	if (flags == MPOOL_PAGE_REQUEST) {
129 		mp->npages++;
130 		bp->pgno = *pgnoaddr;
131 	} else
132 		bp->pgno = *pgnoaddr = mp->npages++;
133 
134 	bp->flags = MPOOL_PINNED | MPOOL_INUSE;
135 
136 	head = &mp->hqh[HASHKEY(bp->pgno)];
137 	TAILQ_INSERT_HEAD(head, bp, hq);
138 	TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
139 	return (bp->page);
140 }
141 
142 int
143 mpool_delete(MPOOL *mp, void *page)
144 {
145 	struct _hqh *head;
146 	BKT *bp;
147 
148 	bp = (BKT *)((char *)page - sizeof(BKT));
149 
150 #ifdef DEBUG
151 	if (!(bp->flags & MPOOL_PINNED)) {
152 		(void)fprintf(stderr,
153 		    "mpool_delete: page %d not pinned\n", bp->pgno);
154 		abort();
155 	}
156 #endif
157 
158 	/* Remove from the hash and lru queues. */
159 	head = &mp->hqh[HASHKEY(bp->pgno)];
160 	TAILQ_REMOVE(head, bp, hq);
161 	TAILQ_REMOVE(&mp->lqh, bp, q);
162 
163 	free(bp);
164 	mp->curcache--;
165 	return (RET_SUCCESS);
166 }
167 
168 /*
169  * mpool_get
170  *	Get a page.
171  */
172 /* ARGSUSED */
173 void *
174 mpool_get(MPOOL *mp, pgno_t pgno,
175     u_int flags)		/* XXX not used? */
176 {
177 	struct _hqh *head;
178 	BKT *bp;
179 	off_t off;
180 	int nr;
181 
182 #ifdef STATISTICS
183 	++mp->pageget;
184 #endif
185 
186 	/* Check for a page that is cached. */
187 	if ((bp = mpool_look(mp, pgno)) != NULL) {
188 #ifdef DEBUG
189 		if (!(flags & MPOOL_IGNOREPIN) && bp->flags & MPOOL_PINNED) {
190 			(void)fprintf(stderr,
191 			    "mpool_get: page %d already pinned\n", bp->pgno);
192 			abort();
193 		}
194 #endif
195 		/*
196 		 * Move the page to the head of the hash chain and the tail
197 		 * of the lru chain.
198 		 */
199 		head = &mp->hqh[HASHKEY(bp->pgno)];
200 		TAILQ_REMOVE(head, bp, hq);
201 		TAILQ_INSERT_HEAD(head, bp, hq);
202 		TAILQ_REMOVE(&mp->lqh, bp, q);
203 		TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
204 
205 		/* Return a pinned page. */
206 		bp->flags |= MPOOL_PINNED;
207 		return (bp->page);
208 	}
209 
210 	/* Get a page from the cache. */
211 	if ((bp = mpool_bkt(mp)) == NULL)
212 		return (NULL);
213 
214 	/* Read in the contents. */
215 	off = mp->pagesize * pgno;
216 	if ((nr = pread(mp->fd, bp->page, mp->pagesize, off)) != (ssize_t)mp->pagesize) {
217 		switch (nr) {
218 		case -1:
219 			/* errno is set for us by pread(). */
220 			free(bp);
221 			mp->curcache--;
222 			return (NULL);
223 		case 0:
224 			/*
225 			 * A zero-length read means you need to create a
226 			 * new page.
227 			 */
228 			memset(bp->page, 0, mp->pagesize);
229 			break;
230 		default:
231 			/* A partial read is definitely bad. */
232 			free(bp);
233 			mp->curcache--;
234 			errno = EINVAL;
235 			return (NULL);
236 		}
237 	}
238 #ifdef STATISTICS
239 	++mp->pageread;
240 #endif
241 
242 	/* Set the page number, pin the page. */
243 	bp->pgno = pgno;
244 	if (!(flags & MPOOL_IGNOREPIN))
245 		bp->flags = MPOOL_PINNED;
246 	bp->flags |= MPOOL_INUSE;
247 
248 	/*
249 	 * Add the page to the head of the hash chain and the tail
250 	 * of the lru chain.
251 	 */
252 	head = &mp->hqh[HASHKEY(bp->pgno)];
253 	TAILQ_INSERT_HEAD(head, bp, hq);
254 	TAILQ_INSERT_TAIL(&mp->lqh, bp, q);
255 
256 	/* Run through the user's filter. */
257 	if (mp->pgin != NULL)
258 		(mp->pgin)(mp->pgcookie, bp->pgno, bp->page);
259 
260 	return (bp->page);
261 }
262 
263 /*
264  * mpool_put
265  *	Return a page.
266  */
267 /* ARGSUSED */
268 int
269 mpool_put(MPOOL *mp, void *page, u_int flags)
270 {
271 	BKT *bp;
272 
273 #ifdef STATISTICS
274 	++mp->pageput;
275 #endif
276 	bp = (BKT *)((char *)page - sizeof(BKT));
277 #ifdef DEBUG
278 	if (!(bp->flags & MPOOL_PINNED)) {
279 		(void)fprintf(stderr,
280 		    "mpool_put: page %d not pinned\n", bp->pgno);
281 		abort();
282 	}
283 #endif
284 	bp->flags &= ~MPOOL_PINNED;
285 	if (flags & MPOOL_DIRTY)
286 		bp->flags |= flags & MPOOL_DIRTY;
287 	return (RET_SUCCESS);
288 }
289 
290 /*
291  * mpool_close
292  *	Close the buffer pool.
293  */
294 int
295 mpool_close(MPOOL *mp)
296 {
297 	BKT *bp;
298 
299 	/* Free up any space allocated to the lru pages. */
300 	while (!TAILQ_EMPTY(&mp->lqh)) {
301 		bp = TAILQ_FIRST(&mp->lqh);
302 		TAILQ_REMOVE(&mp->lqh, bp, q);
303 		free(bp);
304 	}
305 
306 	/* Free the MPOOL cookie. */
307 	free(mp);
308 	return (RET_SUCCESS);
309 }
310 
311 /*
312  * mpool_sync
313  *	Sync the pool to disk.
314  */
315 int
316 mpool_sync(MPOOL *mp)
317 {
318 	BKT *bp;
319 
320 	/* Walk the lru chain, flushing any dirty pages to disk. */
321 	TAILQ_FOREACH(bp, &mp->lqh, q)
322 		if (bp->flags & MPOOL_DIRTY &&
323 		    mpool_write(mp, bp) == RET_ERROR)
324 			return (RET_ERROR);
325 
326 	/* Sync the file descriptor. */
327 	return (_fsync(mp->fd) ? RET_ERROR : RET_SUCCESS);
328 }
329 
330 /*
331  * mpool_bkt
332  *	Get a page from the cache (or create one).
333  */
334 static BKT *
335 mpool_bkt(MPOOL *mp)
336 {
337 	struct _hqh *head;
338 	BKT *bp;
339 
340 	/* If under the max cached, always create a new page. */
341 	if (mp->curcache < mp->maxcache)
342 		goto new;
343 
344 	/*
345 	 * If the cache is max'd out, walk the lru list for a buffer we
346 	 * can flush.  If we find one, write it (if necessary) and take it
347 	 * off any lists.  If we don't find anything we grow the cache anyway.
348 	 * The cache never shrinks.
349 	 */
350 	TAILQ_FOREACH(bp, &mp->lqh, q)
351 		if (!(bp->flags & MPOOL_PINNED)) {
352 			/* Flush if dirty. */
353 			if (bp->flags & MPOOL_DIRTY &&
354 			    mpool_write(mp, bp) == RET_ERROR)
355 				return (NULL);
356 #ifdef STATISTICS
357 			++mp->pageflush;
358 #endif
359 			/* Remove from the hash and lru queues. */
360 			head = &mp->hqh[HASHKEY(bp->pgno)];
361 			TAILQ_REMOVE(head, bp, hq);
362 			TAILQ_REMOVE(&mp->lqh, bp, q);
363 #ifdef DEBUG
364 			{ void *spage;
365 				spage = bp->page;
366 				memset(bp, 0xff, sizeof(BKT) + mp->pagesize);
367 				bp->page = spage;
368 			}
369 #endif
370 			bp->flags = 0;
371 			return (bp);
372 		}
373 
374 new:	if ((bp = (BKT *)calloc(1, sizeof(BKT) + mp->pagesize)) == NULL)
375 		return (NULL);
376 #ifdef STATISTICS
377 	++mp->pagealloc;
378 #endif
379 	bp->page = (char *)bp + sizeof(BKT);
380 	bp->flags = 0;
381 	++mp->curcache;
382 	return (bp);
383 }
384 
385 /*
386  * mpool_write
387  *	Write a page to disk.
388  */
389 static int
390 mpool_write(MPOOL *mp, BKT *bp)
391 {
392 	off_t off;
393 
394 #ifdef STATISTICS
395 	++mp->pagewrite;
396 #endif
397 
398 	/* Run through the user's filter. */
399 	if (mp->pgout)
400 		(mp->pgout)(mp->pgcookie, bp->pgno, bp->page);
401 
402 	off = mp->pagesize * bp->pgno;
403 	if (pwrite(mp->fd, bp->page, mp->pagesize, off) != (ssize_t)mp->pagesize)
404 		return (RET_ERROR);
405 
406 	/*
407 	 * Re-run through the input filter since this page may soon be
408 	 * accessed via the cache, and whatever the user's output filter
409 	 * did may screw things up if we don't let the input filter
410 	 * restore the in-core copy.
411 	 */
412 	if (mp->pgin)
413 		(mp->pgin)(mp->pgcookie, bp->pgno, bp->page);
414 
415 	bp->flags &= ~MPOOL_DIRTY;
416 	return (RET_SUCCESS);
417 }
418 
419 /*
420  * mpool_look
421  *	Lookup a page in the cache.
422  */
423 static BKT *
424 mpool_look(MPOOL *mp, pgno_t pgno)
425 {
426 	struct _hqh *head;
427 	BKT *bp;
428 
429 	head = &mp->hqh[HASHKEY(pgno)];
430 	TAILQ_FOREACH(bp, head, hq)
431 		if ((bp->pgno == pgno) &&
432 			((bp->flags & MPOOL_INUSE) == MPOOL_INUSE)) {
433 #ifdef STATISTICS
434 			++mp->cachehit;
435 #endif
436 			return (bp);
437 		}
438 #ifdef STATISTICS
439 	++mp->cachemiss;
440 #endif
441 	return (NULL);
442 }
443 
444 #ifdef STATISTICS
445 /*
446  * mpool_stat
447  *	Print out cache statistics.
448  */
449 void
450 mpool_stat(MPOOL *mp)
451 {
452 	BKT *bp;
453 	int cnt;
454 	char *sep;
455 
456 	(void)fprintf(stderr, "%lu pages in the file\n", mp->npages);
457 	(void)fprintf(stderr,
458 	    "page size %lu, cacheing %lu pages of %lu page max cache\n",
459 	    mp->pagesize, mp->curcache, mp->maxcache);
460 	(void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n",
461 	    mp->pageput, mp->pageget, mp->pagenew);
462 	(void)fprintf(stderr, "%lu page allocs, %lu page flushes\n",
463 	    mp->pagealloc, mp->pageflush);
464 	if (mp->cachehit + mp->cachemiss)
465 		(void)fprintf(stderr,
466 		    "%.0f%% cache hit rate (%lu hits, %lu misses)\n",
467 		    ((double)mp->cachehit / (mp->cachehit + mp->cachemiss))
468 		    * 100, mp->cachehit, mp->cachemiss);
469 	(void)fprintf(stderr, "%lu page reads, %lu page writes\n",
470 	    mp->pageread, mp->pagewrite);
471 
472 	sep = "";
473 	cnt = 0;
474 	TAILQ_FOREACH(bp, &mp->lqh, q) {
475 		(void)fprintf(stderr, "%s%d", sep, bp->pgno);
476 		if (bp->flags & MPOOL_DIRTY)
477 			(void)fprintf(stderr, "d");
478 		if (bp->flags & MPOOL_PINNED)
479 			(void)fprintf(stderr, "P");
480 		if (++cnt == 10) {
481 			sep = "\n";
482 			cnt = 0;
483 		} else
484 			sep = ", ";
485 
486 	}
487 	(void)fprintf(stderr, "\n");
488 }
489 #endif
490