summaryrefslogtreecommitdiff
path: root/db/mp/mp_fopen.c
blob: 3611ded18f42f7d89056761a404d1c4c3bc0f29a (plain)
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
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 1996, 1997, 1998, 1999, 2000
 *	Sleepycat Software.  All rights reserved.
 */
#include "db_config.h"

#ifndef lint
static const char revid[] = "$Id: mp_fopen.c,v 11.41 2001/01/10 04:50:53 ubell Exp $";
#endif /* not lint */

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

#include <string.h>
#endif

#ifdef  HAVE_RPC
#include "db_server.h"
#endif

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

#ifdef HAVE_RPC
#include "gen_client_ext.h"
#include "rpc_client_ext.h"
#endif

static int __memp_mf_open __P((DB_MPOOL *, const char *,
    size_t, db_pgno_t, DB_MPOOL_FINFO *, u_int32_t, MPOOLFILE **));

/*
 * MEMP_FREMOVE --
 *	Discard an MPOOLFILE and any buffers it references: update the flags
 *	so we never try to write buffers associated with the file, nor can we
 *	find it when looking for files to join.  In addition, clear the ftype
 *	field, there's no reason to post-process pages, they can be discarded
 *	by any thread.
 */
#define	MEMP_FREMOVE(mfp) {						\
	mfp->ftype = 0;							\
	F_SET(mfp, MP_DEADFILE);					\
}

/*
 * memp_fopen --
 *	Open a backing file for the memory pool.
 */
int
memp_fopen(dbenv, path, flags, mode, pagesize, finfop, retp)
	DB_ENV *dbenv;
	const char *path;
	u_int32_t flags;
	int mode;
	size_t pagesize;
	DB_MPOOL_FINFO *finfop;
	DB_MPOOLFILE **retp;
{
	DB_MPOOL *dbmp;
	int ret;

#ifdef HAVE_RPC
	if (F_ISSET(dbenv, DB_ENV_RPCCLIENT))
		return (__dbcl_memp_fopen(dbenv, path, flags,
		    mode, pagesize, finfop, retp));
#endif

	PANIC_CHECK(dbenv);
	ENV_REQUIRES_CONFIG(dbenv, dbenv->mp_handle, DB_INIT_MPOOL);

	dbmp = dbenv->mp_handle;

	/* Validate arguments. */
	if ((ret = __db_fchk(dbenv, "memp_fopen", flags,
	    DB_CREATE |
	    DB_NOMMAP | DB_ODDFILESIZE | DB_RDONLY | DB_TRUNCATE)) != 0)
		return (ret);

	/* Require a non-zero pagesize. */
	if (pagesize == 0 ||
	    (finfop != NULL && finfop->clear_len > pagesize)) {
		__db_err(dbenv, "memp_fopen: illegal page size.");
		return (EINVAL);
	}

	return (__memp_fopen(dbmp,
	    NULL, path, flags, mode, pagesize, 1, finfop, retp));
}

/*
 * __memp_set_unlink -- set unlink on last close flag.
 *
 * PUBLIC: void __memp_set_unlink __P((DB_MPOOLFILE *));
 */
void
__memp_set_unlink(dbmpf)
	DB_MPOOLFILE *dbmpf;
{
	DB_MPOOL *dbmp;
	dbmp = dbmpf->dbmp;

	R_LOCK(dbmp->dbenv, dbmp->reginfo);
	F_SET(dbmpf->mfp, MP_UNLINK);
	R_UNLOCK(dbmp->dbenv, dbmp->reginfo);
}

/*
 * __memp_clear_unlink -- clear unlink on last close flag.
 *
 * PUBLIC: void __memp_clear_unlink __P((DB_MPOOLFILE *));
 */
void
__memp_clear_unlink(dbmpf)
	DB_MPOOLFILE *dbmpf;
{
	DB_MPOOL *dbmp;
	dbmp = dbmpf->dbmp;

	/*
	 * This bit is protected in the queue code because the metapage
	 * is locked so we can avoid geting the region lock.
	 * If this gets used from other than the queue code, we cannot.
	 */
	if (!F_ISSET(dbmpf->mfp, MP_UNLINK))
		return;
	R_LOCK(dbmp->dbenv, dbmp->reginfo);
	F_CLR(dbmpf->mfp, MP_UNLINK);
	R_UNLOCK(dbmp->dbenv, dbmp->reginfo);
}

/*
 * __memp_fopen --
 *	Open a backing file for the memory pool; internal version.
 *
 * PUBLIC: int __memp_fopen __P((DB_MPOOL *, MPOOLFILE *, const char *,
 * PUBLIC:    u_int32_t, int, size_t, int, DB_MPOOL_FINFO *, DB_MPOOLFILE **));
 */
int
__memp_fopen(dbmp, mfp, path, flags, mode, pagesize, needlock, finfop, retp)
	DB_MPOOL *dbmp;
	MPOOLFILE *mfp;
	const char *path;
	u_int32_t flags;
	int mode, needlock;
	size_t pagesize;
	DB_MPOOL_FINFO *finfop;
	DB_MPOOLFILE **retp;
{
	DB_ENV *dbenv;
	DB_MPOOLFILE *dbmfp;
	DB_MPOOL_FINFO finfo;
	db_pgno_t last_pgno;
	size_t maxmap;
	u_int32_t mbytes, bytes, oflags;
	int ret;
	u_int8_t idbuf[DB_FILE_ID_LEN];
	char *rpath;

	dbenv = dbmp->dbenv;
	ret = 0;
	rpath = NULL;

	/*
	 * If mfp is provided, we take the DB_MPOOL_FINFO information from
	 * the mfp.  We don't bother initializing everything, because some
	 * of them are expensive to acquire.  If no mfp is provided and the
	 * finfop argument is NULL, we default the values.
	 */
	if (finfop == NULL) {
		memset(&finfo, 0, sizeof(finfo));
		if (mfp != NULL) {
			finfo.ftype = mfp->ftype;
			finfo.pgcookie = NULL;
			finfo.fileid = NULL;
			finfo.lsn_offset = mfp->lsn_off;
			finfo.clear_len = mfp->clear_len;
		} else {
			finfo.ftype = 0;
			finfo.pgcookie = NULL;
			finfo.fileid = NULL;
			finfo.lsn_offset = -1;
			finfo.clear_len = 0;
		}
		finfop = &finfo;
	}

	/* Allocate and initialize the per-process structure. */
	if ((ret = __os_calloc(dbenv, 1, sizeof(DB_MPOOLFILE), &dbmfp)) != 0)
		return (ret);
	dbmfp->dbmp = dbmp;
	dbmfp->ref = 1;
	if (LF_ISSET(DB_RDONLY))
		F_SET(dbmfp, MP_READONLY);

	if (path == NULL) {
		if (LF_ISSET(DB_RDONLY)) {
			__db_err(dbenv,
			    "memp_fopen: temporary files can't be readonly");
			ret = EINVAL;
			goto err;
		}
		last_pgno = 0;
	} else {
		/* Get the real name for this file and open it. */
		if ((ret = __db_appname(dbenv,
		    DB_APP_DATA, NULL, path, 0, NULL, &rpath)) != 0)
			goto err;
		oflags = 0;
		if (LF_ISSET(DB_CREATE))
			oflags |= DB_OSO_CREATE;
		if (LF_ISSET(DB_RDONLY))
			oflags |= DB_OSO_RDONLY;
		if ((ret =
		   __os_open(dbenv, rpath, oflags, mode, &dbmfp->fh)) != 0) {
			if (!LF_ISSET(DB_EXTENT))
				__db_err(dbenv,
				    "%s: %s", rpath, db_strerror(ret));
			goto err;
		}

		/*
		 * Don't permit files that aren't a multiple of the pagesize,
		 * and find the number of the last page in the file, all the
		 * time being careful not to overflow 32 bits.
		 *
		 * !!!
		 * We can't use off_t's here, or in any code in the mainline
		 * library for that matter.  (We have to use them in the os
		 * stubs, of course, as there are system calls that take them
		 * as arguments.)  The reason is that some customers build in
		 * environments where an off_t is 32-bits, but still run where
		 * offsets are 64-bits, and they pay us a lot of money.
		 */
		if ((ret = __os_ioinfo(dbenv, rpath,
		    &dbmfp->fh, &mbytes, &bytes, NULL)) != 0) {
			__db_err(dbenv, "%s: %s", rpath, db_strerror(ret));
			goto err;
		}

		/*
		 * If we're doing a verify, we might have to cope with
		 * a truncated file;  if the file size is not a multiple
		 * of the page size, round down to a page--we'll
		 * take care of the partial page outside the memp system.
		 */

		/* Page sizes have to be a power-of-two, ignore mbytes. */
		if (bytes % pagesize != 0) {
			if (LF_ISSET(DB_ODDFILESIZE))
				/*
				 * If we're doing a verify, we might
				 * have to cope with a truncated file;
				 * round down, we'll worry about the partial
				 * page outside the memp system.
				 */
				bytes -= (bytes % pagesize);
			else {
				__db_err(dbenv,
		"%s: file size not a multiple of the pagesize",
				    rpath);
				ret = EINVAL;
				goto err;
			}
		}

		last_pgno = mbytes * (MEGABYTE / pagesize);
		last_pgno += bytes / pagesize;

		/* Correction: page numbers are zero-based, not 1-based. */
		if (last_pgno != 0)
			--last_pgno;

		/*
		 * Get the file id if we weren't given one.  Generated file id's
		 * don't use timestamps, otherwise there'd be no chance of any
		 * other process joining the party.
		 */
		if (finfop->fileid == NULL) {
			if ((ret = __os_fileid(dbenv, rpath, 0, idbuf)) != 0)
				goto err;
			finfop->fileid = idbuf;
		}
	}

	/*
	 * If we weren't provided an underlying shared object to join with,
	 * find/allocate the shared file objects.  Also allocate space for
	 * for the per-process thread lock.
	 */
	if (needlock)
		R_LOCK(dbenv, dbmp->reginfo);
	if (mfp == NULL)
		ret = __memp_mf_open(
		    dbmp, path, pagesize, last_pgno, finfop, flags, &mfp);
	else {
		++mfp->mpf_cnt;
		ret = 0;
	}
	if (needlock)
		R_UNLOCK(dbenv, dbmp->reginfo);
	if (ret != 0)
		goto err;

	if (F_ISSET(dbenv, DB_ENV_THREAD)) {
		if ((ret = __db_mutex_alloc(
		    dbenv, dbmp->reginfo, &dbmfp->mutexp)) != 0)
			goto err;
		if ((ret = __db_mutex_init(
		    dbenv, dbmfp->mutexp, 0, MUTEX_THREAD)) != 0)
			goto err;

		/* XXX: KEITH: CLOSE THE FILE ON FAILURE? */
	}

	dbmfp->mfp = mfp;

	/*
	 * If a file:
	 *	+ is read-only
	 *	+ isn't temporary
	 *	+ doesn't require any pgin/pgout support
	 *	+ the DB_NOMMAP flag wasn't set (in either the file open or
	 *	  the environment in which it was opened)
	 *	+ and is less than mp_mmapsize bytes in size
	 *
	 * we can mmap it instead of reading/writing buffers.  Don't do error
	 * checking based on the mmap call failure.  We want to do normal I/O
	 * on the file if the reason we failed was because the file was on an
	 * NFS mounted partition, and we can fail in buffer I/O just as easily
	 * as here.
	 *
	 * XXX
	 * We'd like to test to see if the file is too big to mmap.  Since we
	 * don't know what size or type off_t's or size_t's are, or the largest
	 * unsigned integral type is, or what random insanity the local C
	 * compiler will perpetrate, doing the comparison in a portable way is
	 * flatly impossible.  Hope that mmap fails if the file is too large.
	 */
#define	DB_MAXMMAPSIZE	(10 * 1024 * 1024)	/* 10 Mb. */
	if (F_ISSET(mfp, MP_CAN_MMAP)) {
		if (!F_ISSET(dbmfp, MP_READONLY))
			F_CLR(mfp, MP_CAN_MMAP);
		if (path == NULL)
			F_CLR(mfp, MP_CAN_MMAP);
		if (finfop->ftype != 0)
			F_CLR(mfp, MP_CAN_MMAP);
		if (LF_ISSET(DB_NOMMAP) || F_ISSET(dbenv, DB_ENV_NOMMAP))
			F_CLR(mfp, MP_CAN_MMAP);
		maxmap = dbenv->mp_mmapsize == 0 ?
		    DB_MAXMMAPSIZE : dbenv->mp_mmapsize;
		if (mbytes > maxmap / MEGABYTE ||
		    (mbytes == maxmap / MEGABYTE && bytes >= maxmap % MEGABYTE))
			F_CLR(mfp, MP_CAN_MMAP);
	}
	dbmfp->addr = NULL;
	if (F_ISSET(mfp, MP_CAN_MMAP)) {
		dbmfp->len = (size_t)mbytes * MEGABYTE + bytes;
		if (__os_mapfile(dbenv, rpath,
		    &dbmfp->fh, dbmfp->len, 1, &dbmfp->addr) != 0) {
			dbmfp->addr = NULL;
			F_CLR(mfp, MP_CAN_MMAP);
		}
	}
	if (rpath != NULL)
		__os_freestr(rpath);

	MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);
	TAILQ_INSERT_TAIL(&dbmp->dbmfq, dbmfp, q);
	MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);

	*retp = dbmfp;
	return (0);

err:	/*
	 * Note that we do not have to free the thread mutex, because we
	 * never get to here after we have successfully allocated it.
	 */
	if (rpath != NULL)
		__os_freestr(rpath);
	if (F_ISSET(&dbmfp->fh, DB_FH_VALID))
		(void)__os_closehandle(&dbmfp->fh);
	if (dbmfp != NULL) {
		if (dbmfp->mutexp != NULL)
			__db_mutex_free(dbenv, dbmp->reginfo, dbmfp->mutexp);
		__os_free(dbmfp, sizeof(DB_MPOOLFILE));
	}
	return (ret);
}

/*
 * __memp_mf_open --
 *	Open an MPOOLFILE.
 */
static int
__memp_mf_open(dbmp, path, pagesize, last_pgno, finfop, flags, retp)
	DB_MPOOL *dbmp;
	const char *path;
	size_t pagesize;
	db_pgno_t last_pgno;
	DB_MPOOL_FINFO *finfop;
	u_int32_t flags;
	MPOOLFILE **retp;
{
	MPOOL *mp;
	MPOOLFILE *mfp;
	int ret;
	void *p;

#define	ISTEMPORARY	(path == NULL)

	/*
	 * If not creating a temporary file, walk the list of MPOOLFILE's,
	 * looking for a matching file.  Files backed by temporary files
	 * or previously removed files can't match.
	 *
	 * DB_TRUNCATE support.
	 *
	 * The fileID is a filesystem unique number (e.g., a UNIX dev/inode
	 * pair) plus a timestamp.  If files are removed and created in less
	 * than a second, the fileID can be repeated.  The problem with
	 * repetition happens when the file that previously had the fileID
	 * value still has pages in the pool, since we don't want to use them
	 * to satisfy requests for the new file.
	 *
	 * Because the DB_TRUNCATE flag reuses the dev/inode pair, repeated
	 * opens with that flag set guarantees matching fileIDs when the
	 * machine can open a file and then re-open with truncate within a
	 * second.  For this reason, we pass that flag down, and, if we find
	 * a matching entry, we ensure that it's never found again, and we
	 * create a new entry for the current request.
	 */
	if (!ISTEMPORARY) {
		mp = dbmp->reginfo[0].primary;
		for (mfp = SH_TAILQ_FIRST(&mp->mpfq, __mpoolfile);
		    mfp != NULL; mfp = SH_TAILQ_NEXT(mfp, q, __mpoolfile)) {
			if (F_ISSET(mfp, MP_DEADFILE | MP_TEMP))
				continue;
			if (memcmp(finfop->fileid, R_ADDR(dbmp->reginfo,
			    mfp->fileid_off), DB_FILE_ID_LEN) == 0) {
				if (LF_ISSET(DB_TRUNCATE)) {
					MEMP_FREMOVE(mfp);
					continue;
				}
				if (finfop->clear_len != mfp->clear_len ||
				    pagesize != mfp->stat.st_pagesize) {
					__db_err(dbmp->dbenv,
				    "%s: page size or clear length changed",
					    path);
					return (EINVAL);
				}

				/*
				 * It's possible that our needs for pre- and
				 * post-processing are changing.  For example,
				 * an application created a hash subdatabase
				 * in a database that was previously all btree.
				 */
				if (finfop->ftype != 0)
					mfp->ftype = finfop->ftype;

				++mfp->mpf_cnt;

				*retp = mfp;
				return (0);
			}
		}
	}

	/* Allocate a new MPOOLFILE. */
	if ((ret = __memp_alloc(
	    dbmp, dbmp->reginfo, NULL, sizeof(MPOOLFILE), NULL, &mfp)) != 0)
		goto mem_err;
	*retp = mfp;

	/* Initialize the structure. */
	memset(mfp, 0, sizeof(MPOOLFILE));
	mfp->mpf_cnt = 1;
	mfp->ftype = finfop->ftype;
	mfp->lsn_off = finfop->lsn_offset;
	mfp->clear_len = finfop->clear_len;

	/*
	 * If the user specifies DB_MPOOL_LAST or DB_MPOOL_NEW on a memp_fget,
	 * we have to know the last page in the file.  Figure it out and save
	 * it away.
	 */
	mfp->stat.st_pagesize = pagesize;
	mfp->orig_last_pgno = mfp->last_pgno = last_pgno;

	if (ISTEMPORARY)
		F_SET(mfp, MP_TEMP);
	else {
		/* Copy the file path into shared memory. */
		if ((ret = __memp_alloc(dbmp, dbmp->reginfo,
		    NULL, strlen(path) + 1, &mfp->path_off, &p)) != 0)
			goto err;
		memcpy(p, path, strlen(path) + 1);

		/* Copy the file identification string into shared memory. */
		if ((ret = __memp_alloc(dbmp, dbmp->reginfo,
		    NULL, DB_FILE_ID_LEN, &mfp->fileid_off, &p)) != 0)
			goto err;
		memcpy(p, finfop->fileid, DB_FILE_ID_LEN);

		F_SET(mfp, MP_CAN_MMAP);
	}

	/* Copy the page cookie into shared memory. */
	if (finfop->pgcookie == NULL || finfop->pgcookie->size == 0) {
		mfp->pgcookie_len = 0;
		mfp->pgcookie_off = 0;
	} else {
		if ((ret = __memp_alloc(dbmp, dbmp->reginfo,
		    NULL, finfop->pgcookie->size, &mfp->pgcookie_off, &p)) != 0)
			goto err;
		memcpy(p, finfop->pgcookie->data, finfop->pgcookie->size);
		mfp->pgcookie_len = finfop->pgcookie->size;
	}

	/* Prepend the MPOOLFILE to the list of MPOOLFILE's. */
	mp = dbmp->reginfo[0].primary;
	SH_TAILQ_INSERT_HEAD(&mp->mpfq, mfp, q, __mpoolfile);

	if (0) {
err:		if (mfp->path_off != 0)
			__db_shalloc_free(dbmp->reginfo[0].addr,
			    R_ADDR(dbmp->reginfo, mfp->path_off));
		if (mfp->fileid_off != 0)
			__db_shalloc_free(dbmp->reginfo[0].addr,
			    R_ADDR(dbmp->reginfo, mfp->fileid_off));
		if (mfp != NULL)
			__db_shalloc_free(dbmp->reginfo[0].addr, mfp);
mem_err:	__db_err(dbmp->dbenv,
		    "Unable to allocate memory for mpool file");
	}
	return (ret);
}

/*
 * memp_fclose --
 *	Close a backing file for the memory pool.
 */
int
memp_fclose(dbmfp)
	DB_MPOOLFILE *dbmfp;
{
	DB_ENV *dbenv;
	DB_MPOOL *dbmp;
	MPOOLFILE *mfp;
	char *rpath;
	int ret, t_ret;

	dbmp = dbmfp->dbmp;
	dbenv = dbmp->dbenv;
	ret = 0;

	PANIC_CHECK(dbenv);

#ifdef HAVE_RPC
	if (F_ISSET(dbenv, DB_ENV_RPCCLIENT))
		return (__dbcl_memp_fclose(dbmfp));
#endif

	/*
	 * Remove the DB_MPOOLFILE from the queue.  This has to happen before
	 * we perform any action that can fail, otherwise __memp_close may
	 * loop infinitely when calling us to discard all of the DB_MPOOLFILEs.
	 */
	for (;;) {
		MUTEX_THREAD_LOCK(dbenv, dbmp->mutexp);

		/*
		 * We have to reference count DB_MPOOLFILE structures as other
		 * threads may be using them.  The problem only happens if the
		 * application makes a bad design choice.  Here's the path:
		 *
		 * Thread A opens a database.
		 * Thread B uses thread A's DB_MPOOLFILE to write a buffer
		 *    in order to free up memory in the mpool cache.
		 * Thread A closes the database while thread B is using the
		 *    DB_MPOOLFILE structure.
		 *
		 * By opening all databases before creating the threads, and
		 * closing them after the threads have exited, applications
		 * get better performance and avoid the problem path entirely.
		 *
		 * Regardless, holding the DB_MPOOLFILE to flush a dirty buffer
		 * is a short-term lock, even in worst case, since we better be
		 * the only thread of control using the DB_MPOOLFILE structure
		 * to read pages *into* the cache.  Wait until we're the only
		 * reference holder and remove the DB_MPOOLFILE structure from
		 * the list, so nobody else can even find it.
		 */
		if (dbmfp->ref == 1) {
			TAILQ_REMOVE(&dbmp->dbmfq, dbmfp, q);
			break;
		}
		MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);

		(void)__os_sleep(dbenv, 1, 0);
	}
	MUTEX_THREAD_UNLOCK(dbenv, dbmp->mutexp);

	/* Complain if pinned blocks never returned. */
	if (dbmfp->pinref != 0)
		__db_err(dbenv, "%s: close: %lu blocks left pinned",
		    __memp_fn(dbmfp), (u_long)dbmfp->pinref);

	/* Discard any mmap information. */
	if (dbmfp->addr != NULL &&
	    (ret = __os_unmapfile(dbenv, dbmfp->addr, dbmfp->len)) != 0)
		__db_err(dbenv, "%s: %s", __memp_fn(dbmfp), db_strerror(ret));

	/* Close the file; temporary files may not yet have been created. */
	if (F_ISSET(&dbmfp->fh, DB_FH_VALID) &&
	    (t_ret = __os_closehandle(&dbmfp->fh)) != 0) {
		__db_err(dbenv, "%s: %s", __memp_fn(dbmfp), db_strerror(t_ret));
		if (ret != 0)
			t_ret = ret;
	}

	/* Discard the thread mutex. */
	if (dbmfp->mutexp != NULL)
		__db_mutex_free(dbenv, dbmp->reginfo, dbmfp->mutexp);

	/*
	 * Discard our reference on the the underlying MPOOLFILE, and close
	 * it if it's no longer useful to anyone.
	 *
	 * If we're not discarding it, and it's a temp file, this means
	 * all the outstanding references belong to unflushed buffers.
	 * (A temp file can only be referenced by one DB_MPOOLFILE).
	 * We don't care about preserving any of those buffers, so mark
	 * the MPOOLFILE as dead so that when we try to flush them,
	 * even the dirty ones just get discarded.
	 */
	R_LOCK(dbenv, dbmp->reginfo);
	mfp = dbmfp->mfp;
	if (--mfp->mpf_cnt == 0) {
		if (F_ISSET(mfp, MP_UNLINK)) {
			MEMP_FREMOVE(mfp);
			if ((t_ret = __db_appname(dbmp->dbenv,
			    DB_APP_DATA, NULL, R_ADDR(dbmp->reginfo,
			    mfp->path_off), 0, NULL, &rpath)) != 0 && ret == 0)
				ret = t_ret;
			if (t_ret == 0 && (t_ret =
			    __os_unlink(dbmp->dbenv, rpath) != 0 && ret == 0))
				ret = t_ret;
			__os_free(rpath, 0);
		}
		if (mfp->block_cnt == 0)
			__memp_mf_discard(dbmp, mfp);
	}
	else if (F_ISSET(mfp, MP_TEMP))
		MEMP_FREMOVE(mfp);
	R_UNLOCK(dbenv, dbmp->reginfo);

	/* Discard the DB_MPOOLFILE structure. */
	__os_free(dbmfp, sizeof(DB_MPOOLFILE));

	return (ret);
}

/*
 * __memp_mf_discard --
 *	Discard an MPOOLFILE.
 *
 * PUBLIC: void __memp_mf_discard __P((DB_MPOOL *, MPOOLFILE *));
 */
void
__memp_mf_discard(dbmp, mfp)
	DB_MPOOL *dbmp;
	MPOOLFILE *mfp;
{
	MPOOL *mp;

	mp = dbmp->reginfo[0].primary;

	/* Delete from the list of MPOOLFILEs. */
	SH_TAILQ_REMOVE(&mp->mpfq, mfp, q, __mpoolfile);

	/* Free the space. */
	if (mfp->path_off != 0)
		__db_shalloc_free(dbmp->reginfo[0].addr,
		    R_ADDR(dbmp->reginfo, mfp->path_off));
	if (mfp->fileid_off != 0)
		__db_shalloc_free(dbmp->reginfo[0].addr,
		    R_ADDR(dbmp->reginfo, mfp->fileid_off));
	if (mfp->pgcookie_off != 0)
		__db_shalloc_free(dbmp->reginfo[0].addr,
		    R_ADDR(dbmp->reginfo, mfp->pgcookie_off));
	__db_shalloc_free(dbmp->reginfo[0].addr, mfp);
}

/*
 * __memp_fremove --
 *	Remove an underlying file from the system.
 *
 * PUBLIC: int __memp_fremove __P((DB_MPOOLFILE *));
 */
int
__memp_fremove(dbmfp)
	DB_MPOOLFILE *dbmfp;
{
	DB_ENV *dbenv;
	DB_MPOOL *dbmp;
	MPOOLFILE *mfp;

	dbmp = dbmfp->dbmp;
	dbenv = dbmp->dbenv;
	mfp = dbmfp->mfp;

	PANIC_CHECK(dbenv);

	R_LOCK(dbenv, dbmp->reginfo);

	MEMP_FREMOVE(mfp);

	R_UNLOCK(dbenv, dbmp->reginfo);

	return (0);
}

/*
 * __memp_fn --
 *	On errors we print whatever is available as the file name.
 *
 * PUBLIC: char * __memp_fn __P((DB_MPOOLFILE *));
 */
char *
__memp_fn(dbmfp)
	DB_MPOOLFILE *dbmfp;
{
	return (__memp_fns(dbmfp->dbmp, dbmfp->mfp));
}

/*
 * __memp_fns --
 *	On errors we print whatever is available as the file name.
 *
 * PUBLIC: char * __memp_fns __P((DB_MPOOL *, MPOOLFILE *));
 *
 */
char *
__memp_fns(dbmp, mfp)
	DB_MPOOL *dbmp;
	MPOOLFILE *mfp;
{
	if (mfp->path_off == 0)
		return ((char *)"temporary");

	return ((char *)R_ADDR(dbmp->reginfo, mfp->path_off));
}