summary refs log tree commit diff
path: root/db2/mp/mp_region.c
blob: c20e66974965f291c27cc66371694e83bf748fd2 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 1996, 1997
 *	Sleepycat Software.  All rights reserved.
 */
#include "config.h"

#ifndef lint
static const char sccsid[] = "@(#)mp_region.c	10.18 (Sleepycat) 11/29/97";
#endif /* not lint */

#ifndef NO_SYSTEM_INCLUDES
#include <sys/types.h>
#include <sys/stat.h>

#include <errno.h>
#include <fcntl.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#endif

#include "db_int.h"
#include "shqueue.h"
#include "db_shash.h"
#include "mp.h"
#include "common_ext.h"

/*
 * __memp_ralloc --
 *	Allocate some space in the mpool region.
 *
 * PUBLIC: int __memp_ralloc __P((DB_MPOOL *, size_t, size_t *, void *));
 */
int
__memp_ralloc(dbmp, len, offsetp, retp)
	DB_MPOOL *dbmp;
	size_t len, *offsetp;
	void *retp;
{
	BH *bhp, *nbhp;
	MPOOL *mp;
	MPOOLFILE *mfp;
	size_t fsize, total;
	int nomore, restart, ret, wrote;
	void *p;

	mp = dbmp->mp;

	nomore = 0;
alloc:	if ((ret = __db_shalloc(dbmp->addr, len, MUTEX_ALIGNMENT, &p)) == 0) {
		if (offsetp != NULL)
			*offsetp = R_OFFSET(dbmp, p);
		*(void **)retp = p;
		return (0);
	}
	if (nomore) {
		__db_err(dbmp->dbenv, "%s", strerror(ret));
		return (ret);
	}

	/* Look for a buffer on the free list that's the right size. */
	for (bhp =
	    SH_TAILQ_FIRST(&mp->bhfq, __bh); bhp != NULL; bhp = nbhp) {
		nbhp = SH_TAILQ_NEXT(bhp, q, __bh);

		if (__db_shsizeof(bhp) == len) {
			SH_TAILQ_REMOVE(&mp->bhfq, bhp, q, __bh);
			if (offsetp != NULL)
				*offsetp = R_OFFSET(dbmp, bhp);
			*(void **)retp = bhp;
			return (0);
		}
	}

	/* Discard from the free list until we've freed enough memory. */
	total = 0;
	for (bhp =
	    SH_TAILQ_FIRST(&mp->bhfq, __bh); bhp != NULL; bhp = nbhp) {
		nbhp = SH_TAILQ_NEXT(bhp, q, __bh);

		SH_TAILQ_REMOVE(&mp->bhfq, bhp, q, __bh);
		__db_shalloc_free(dbmp->addr, bhp);
		--mp->stat.st_page_clean;

		/*
		 * Retry as soon as we've freed up sufficient space.  If we
		 * have to coalesce of memory to satisfy the request, don't
		 * try until it's likely (possible?) that we'll succeed.
		 */
		total += fsize = __db_shsizeof(bhp);
		if (fsize >= len || total >= 3 * len)
			goto alloc;
	}

retry:	/* Find a buffer we can flush; pure LRU. */
	total = 0;
	for (bhp =
	    SH_TAILQ_FIRST(&mp->bhq, __bh); bhp != NULL; bhp = nbhp) {
		nbhp = SH_TAILQ_NEXT(bhp, q, __bh);

		/* Ignore pinned or locked (I/O in progress) buffers. */
		if (bhp->ref != 0 || F_ISSET(bhp, BH_LOCKED))
			continue;

		/* Find the associated MPOOLFILE. */
		mfp = R_ADDR(dbmp, bhp->mf_offset);

		/*
		 * Write the page if it's dirty.
		 *
		 * If we wrote the page, fall through and free the buffer.  We
		 * don't have to rewalk the list to acquire the buffer because
		 * it was never available for any other process to modify it.
		 * If we didn't write the page, but we discarded and reacquired
		 * the region lock, restart the buffer list walk.  If we neither
		 * wrote the buffer nor discarded the region lock, continue down
		 * the buffer list.
		 */
		if (F_ISSET(bhp, BH_DIRTY)) {
			if ((ret = __memp_bhwrite(dbmp,
			    mfp, bhp, &restart, &wrote)) != 0)
				return (ret);

			/*
			 * It's possible that another process wants this buffer
			 * and incremented the ref count while we were writing
			 * it.
			 */
			if (bhp->ref != 0)
				goto retry;

			if (wrote)
				++mp->stat.st_rw_evict;
			else {
				if (restart)
					goto retry;
				continue;
			}
		} else
			++mp->stat.st_ro_evict;

		/*
		 * Check to see if the buffer is the size we're looking for.
		 * If it is, simply reuse it.
		 */
		total += fsize = __db_shsizeof(bhp);
		if (fsize == len) {
			__memp_bhfree(dbmp, mfp, bhp, 0);

			if (offsetp != NULL)
				*offsetp = R_OFFSET(dbmp, bhp);
			*(void **)retp = bhp;
			return (0);
		}

		/* Free the buffer. */
		__memp_bhfree(dbmp, mfp, bhp, 1);

		/*
		 * Retry as soon as we've freed up sufficient space.  If we
		 * have to coalesce of memory to satisfy the request, don't
		 * try until it's likely (possible?) that we'll succeed.
		 */
		if (fsize >= len || total >= 3 * len)
			goto alloc;

		/* Restart the walk if we discarded the region lock. */
		if (restart)
			goto retry;
	}
	nomore = 1;
	goto alloc;
}

/*
 * __memp_ropen --
 *	Attach to, and optionally create, the mpool region.
 *
 * PUBLIC: int __memp_ropen
 * PUBLIC:    __P((DB_MPOOL *, const char *, size_t, int, int));
 */
int
__memp_ropen(dbmp, path, cachesize, mode, flags)
	DB_MPOOL *dbmp;
	const char *path;
	size_t cachesize;
	int mode, flags;
{
	MPOOL *mp;
	size_t rlen;
	int fd, newregion, ret, retry_cnt;

	/*
	 * Unlike other DB subsystems, mpool can't simply grow the region
	 * because it returns pointers into the region to its clients.  To
	 * "grow" the region, we'd have to allocate a new region and then
	 * store a region number in the structures that reference regional
	 * objects.  It's reasonable that we fail regardless, as clients
	 * shouldn't have every page in the region pinned, so the only
	 * "failure" mode should be a performance penalty because we don't
	 * find a page in the cache that we'd like to have found.
	 *
	 * Up the user's cachesize by 25% to account for our overhead.
	 */
	if (cachesize < DB_CACHESIZE_MIN)
		if (cachesize == 0)
			cachesize = DB_CACHESIZE_DEF;
		else
			cachesize = DB_CACHESIZE_MIN;
	rlen = cachesize + cachesize / 4;

	/* Map in the region. */
	retry_cnt = newregion = 0;
retry:	if (LF_ISSET(DB_CREATE)) {
		/*
		 * If it's a private mpool, use malloc, it's a lot faster than
		 * instantiating a region.
		 *
		 * XXX
		 * If we're doing locking and don't have spinlocks for this
		 * architecture, we'd have to instantiate the file, we need
		 * the file descriptor for locking.  However, it should not
		 * be possible for DB_THREAD to be set if HAVE_SPINLOCKS aren't
		 * defined.
		 *
		 * XXX
		 * HP-UX won't permit mutexes to live in anything but shared
		 * memory.  So, instantiate the shared mpool region file on
		 * that architecture, regardless.  If this turns out to be a
		 * performance problem, we could use anonymous memory instead.
		 */
#if !defined(__hppa)
		if (F_ISSET(dbmp, MP_ISPRIVATE))
			if ((dbmp->maddr = __db_malloc(rlen)) == NULL)
				ret = ENOMEM;
			else {
				F_SET(dbmp, MP_MALLOC);
				ret = __db_rinit(dbmp->dbenv,
				    dbmp->maddr, 0, rlen, 0);
			}
		else
#endif
			ret = __db_rcreate(dbmp->dbenv, DB_APP_NONE, path,
			    DB_DEFAULT_MPOOL_FILE, mode, rlen,
			    F_ISSET(dbmp, MP_ISPRIVATE) ? DB_TEMPORARY : 0,
			    &fd, &dbmp->maddr);
		if (ret == 0) {
			/* Put the MPOOL structure first in the region. */
			mp = dbmp->maddr;

			SH_TAILQ_INIT(&mp->bhq);
			SH_TAILQ_INIT(&mp->bhfq);
			SH_TAILQ_INIT(&mp->mpfq);

			/* Initialize the rest of the region as free space. */
			dbmp->addr = (u_int8_t *)dbmp->maddr + sizeof(MPOOL);
			__db_shalloc_init(dbmp->addr, rlen - sizeof(MPOOL));

			/*
			 *
			 * Pretend that the cache will be broken up into 4K
			 * pages, and that we want to keep it under, say, 10
			 * pages on each chain.  This means a 256MB cache will
			 * allocate ~6500 offset pairs.
			 */
			mp->htab_buckets =
			    __db_tablesize((cachesize / (4 * 1024)) / 10);

			/* Allocate hash table space and initialize it. */
			if ((ret = __db_shalloc(dbmp->addr,
			    mp->htab_buckets * sizeof(DB_HASHTAB),
			    0, &dbmp->htab)) != 0)
				goto err;
			__db_hashinit(dbmp->htab, mp->htab_buckets);
			mp->htab = R_OFFSET(dbmp, dbmp->htab);

			ZERO_LSN(mp->lsn);
			mp->lsn_cnt = 0;

			memset(&mp->stat, 0, sizeof(mp->stat));
			mp->stat.st_cachesize = cachesize;

			mp->flags = 0;

			newregion = 1;
		} else if (ret != EEXIST)
			return (ret);
	}

	/* If we didn't or couldn't create the region, try and join it. */
	if (!newregion &&
	    (ret = __db_ropen(dbmp->dbenv, DB_APP_NONE,
	    path, DB_DEFAULT_MPOOL_FILE, 0, &fd, &dbmp->maddr)) != 0) {
		/*
		 * If we failed because the file wasn't available, wait a
		 * second and try again.
		 */
		if (ret == EAGAIN && ++retry_cnt < 3) {
			(void)__db_sleep(1, 0);
			goto retry;
		}
		return (ret);
	}

	/* Set up the common pointers. */
	dbmp->mp = dbmp->maddr;
	dbmp->addr = (u_int8_t *)dbmp->maddr + sizeof(MPOOL);

	/*
	 * If not already locked, lock the region -- if it's a new region,
	 * then either __db_rcreate() locked it for us or we malloc'd it
	 * instead of creating a region, neither of which requires locking
	 * here.
	 */
	if (!newregion)
		LOCKREGION(dbmp);

	/*
	 * Get the hash table address; it's on the shared page, so we have
	 * to lock first.
	 */
	dbmp->htab = R_ADDR(dbmp, dbmp->mp->htab);

	dbmp->fd = fd;

	/* If we locked the region, release it now. */
	if (!F_ISSET(dbmp, MP_MALLOC))
		UNLOCKREGION(dbmp);
	return (0);

err:	if (fd != -1) {
		dbmp->fd = fd;
		(void)__memp_rclose(dbmp);
	}

	if (newregion)
		(void)memp_unlink(path, 1, dbmp->dbenv);
	return (ret);
}

/*
 * __memp_rclose --
 *	Close the mpool region.
 *
 * PUBLIC: int __memp_rclose __P((DB_MPOOL *));
 */
int
__memp_rclose(dbmp)
	DB_MPOOL *dbmp;
{
	if (F_ISSET(dbmp, MP_MALLOC)) {
		__db_free(dbmp->maddr);
		return (0);
	}
	return (__db_rclose(dbmp->dbenv, dbmp->fd, dbmp->maddr));
}