about summary refs log tree commit diff
path: root/db2/mp/mp_sync.c
blob: 6d16cf3cd4f41c0598504d7919bae5562c6f9fd7 (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
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
/*-
 * 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_sync.c	10.19 (Sleepycat) 12/3/97";
#endif /* not lint */

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

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

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

static int __bhcmp __P((const void *, const void *));

/*
 * memp_sync --
 *	Mpool sync function.
 */
int
memp_sync(dbmp, lsnp)
	DB_MPOOL *dbmp;
	DB_LSN *lsnp;
{
	BH *bhp, **bharray;
	DB_ENV *dbenv;
	MPOOL *mp;
	MPOOLFILE *mfp;
	int ar_cnt, cnt, nalloc, next, ret, wrote;

	dbenv = dbmp->dbenv;

	if (dbenv->lg_info == NULL) {
		__db_err(dbenv, "memp_sync: requires logging");
		return (EINVAL);
	}

	/*
	 * We try and write the buffers in page order so that the underlying
	 * filesystem doesn't have to seek and can write contiguous blocks,
	 * plus, we don't want to hold the region lock while we write the
	 * buffers.  Get memory to hold the buffer pointers.  Get a good-size
	 * block, too, because we realloc while holding the region lock if we
	 * run out.
	 */
	if ((bharray =
	    (BH **)__db_malloc((nalloc = 1024) * sizeof(BH *))) == NULL)
		return (ENOMEM);

	LOCKREGION(dbmp);

	/*
	 * If the application is asking about a previous call to memp_sync(),
	 * and we haven't found any buffers that the application holding the
	 * pin couldn't write, return yes or no based on the current count.
	 * Note, if the application is asking about a LSN *smaller* than one
	 * we've already handled or are currently handling, then we return a
	 * result based on the count for the larger LSN.
	 */
	mp = dbmp->mp;
	if (!F_ISSET(mp, MP_LSN_RETRY) && log_compare(lsnp, &mp->lsn) <= 0) {
		if (mp->lsn_cnt == 0) {
			*lsnp = mp->lsn;
			ret = 0;
		} else
			ret = DB_INCOMPLETE;
		goto done;
	}

	/* Else, it's a new checkpoint. */
	F_CLR(mp, MP_LSN_RETRY);

	/*
	 * Save the LSN.  We know that it's a new LSN or larger than the one
	 * for which we were already doing a checkpoint.  (BTW, I don't expect
	 * to see multiple LSN's from the same or multiple processes, but You
	 * Just Never Know.  Responding as if they all called with the largest
	 * of the LSNs specified makes everything work.)
	 *
	 * We don't currently use the LSN we save.  We could potentially save
	 * the last-written LSN in each buffer header and use it to determine
	 * what buffers need to be written.  The problem with this is that it's
	 * sizeof(LSN) more bytes of buffer header.  We currently write all the
	 * dirty buffers instead.
	 *
	 * Walk the list of shared memory segments clearing the count of
	 * buffers waiting to be written.
	 */
	mp->lsn = *lsnp;
	mp->lsn_cnt = 0;
	for (mfp = SH_TAILQ_FIRST(&dbmp->mp->mpfq, __mpoolfile);
	    mfp != NULL; mfp = SH_TAILQ_NEXT(mfp, q, __mpoolfile))
		mfp->lsn_cnt = 0;

	/*
	 * Walk the list of buffers and mark all dirty buffers to be written
	 * and all pinned buffers to be potentially written (we can't know if
	 * we'll need to write them until the holding process returns them to
	 * the cache).  We do this in one pass while holding the region locked
	 * so that processes can't make new buffers dirty, causing us to never
	 * finish.  Since the application may have restarted the sync, clear
	 * any BH_WRITE flags that appear to be left over from previous calls.
	 *
	 * Keep a count of the total number of buffers we need to write in
	 * MPOOL->lsn_cnt, and for each file, in MPOOLFILE->lsn_count.
	 */
	ar_cnt = 0;
	for (bhp = SH_TAILQ_FIRST(&mp->bhq, __bh);
	    bhp != NULL; bhp = SH_TAILQ_NEXT(bhp, q, __bh))
		if (F_ISSET(bhp, BH_DIRTY) || bhp->ref != 0) {
			F_SET(bhp, BH_WRITE);

			++mp->lsn_cnt;

			mfp = R_ADDR(dbmp, bhp->mf_offset);
			++mfp->lsn_cnt;

			/*
			 * If the buffer isn't in use, we should be able to
			 * write it immediately, so save a reference to it.
			 */
			if (bhp->ref == 0) {
				if (ar_cnt == nalloc) {
					nalloc *= 2;
					if ((bharray =
					    (BH **)__db_realloc(bharray,
					    nalloc * sizeof(BH *))) == NULL) {
						ret = ENOMEM;
						goto err;
					}
				}
				bharray[ar_cnt++] = bhp;
			}
		} else
			F_CLR(bhp, BH_WRITE);

	/* If there no buffers we can write immediately, we're done. */
	if (ar_cnt == 0) {
		ret = mp->lsn_cnt ? DB_INCOMPLETE : 0;
		goto done;
	}

	/* Lock down the buffers and their contents. */
	for (cnt = 0; cnt < ar_cnt; ++cnt)
		++bharray[cnt]->ref;

	UNLOCKREGION(dbmp);

	/* Sort the buffers we're going to write. */
	qsort(bharray, ar_cnt, sizeof(BH *), __bhcmp);

	LOCKREGION(dbmp);

	/* Walk the array, writing buffers. */
	for (next = 0; next < ar_cnt; ++next) {
		/*
		 * It's possible for a thread to have gotten the buffer since
		 * we listed it for writing.  If the reference count is still
		 * 1, we're the only ones using the buffer, go ahead and write.
		 * If it's >1, then skip the buffer and assume that it will be
		 * written when it's returned to the cache.
		 */
		if (bharray[next]->ref > 1) {
			--bharray[next]->ref;
			continue;
		}

		/* Write the buffer. */
		mfp = R_ADDR(dbmp, bharray[next]->mf_offset);
		ret = __memp_bhwrite(dbmp, mfp, bharray[next], NULL, &wrote);

		/* Release the buffer. */
		--bharray[next]->ref;

		/* If there's an error, release the rest of the buffers. */
		if (ret != 0 || !wrote) {
			/*
			 * Any process syncing the shared memory buffer pool
			 * had better be able to write to any underlying file.
			 * Be understanding, but firm, on this point.
			 */
			if (ret == 0) {
				__db_err(dbenv, "%s: unable to flush page: %lu",
				    __memp_fns(dbmp, mfp),
				    (u_long)bharray[next]->pgno);
				ret = EPERM;
			}

			while (++next < ar_cnt)
				--bharray[next]->ref;
			goto err;
		}
	}
	ret = mp->lsn_cnt ? DB_INCOMPLETE : 0;

done:
	if (0) {
err:		/*
		 * On error, clear:
		 *	MPOOL->lsn_cnt (the total sync count)
		 *	MPOOLFILE->lsn_cnt (the per-file sync count)
		 *	BH_WRITE flag (the scheduled for writing flag)
		 */
		mp->lsn_cnt = 0;
		for (mfp = SH_TAILQ_FIRST(&dbmp->mp->mpfq, __mpoolfile);
		    mfp != NULL; mfp = SH_TAILQ_NEXT(mfp, q, __mpoolfile))
			mfp->lsn_cnt = 0;
		for (bhp = SH_TAILQ_FIRST(&mp->bhq, __bh);
		    bhp != NULL; bhp = SH_TAILQ_NEXT(bhp, q, __bh))
			F_CLR(bhp, BH_WRITE);
	}
	UNLOCKREGION(dbmp);
	__db_free(bharray);
	return (ret);
}

/*
 * memp_fsync --
 *	Mpool file sync function.
 */
int
memp_fsync(dbmfp)
	DB_MPOOLFILE *dbmfp;
{
	BH *bhp, **bharray;
	DB_MPOOL *dbmp;
	size_t mf_offset;
	int ar_cnt, cnt, nalloc, next, pincnt, ret, wrote;

	dbmp = dbmfp->dbmp;

	/*
	 * If this handle doesn't have a file descriptor that's open for
	 * writing, or if the file is a temporary, there's no reason to
	 * proceed further.
	 */
	if (F_ISSET(dbmfp, MP_READONLY))
		return (0);

	ret = 0;
	LOCKREGION(dbmp);
	if (F_ISSET(dbmfp->mfp, MP_TEMP))
		ret = 1;
	UNLOCKREGION(dbmp);
	if (ret)
		return (0);

	mf_offset = R_OFFSET(dbmp, dbmfp->mfp);

	/*
	 * We try and write the buffers in page order so that the underlying
	 * filesystem doesn't have to seek and can write contiguous blocks,
	 * plus, we don't want to hold the region lock while we write the
	 * buffers.  Get memory to hold the buffer pointers.  Get a good-size
	 * block, too, because we realloc while holding the region lock if we
	 * run out.
	 */
	nalloc = 1024;
	if ((bharray =
	    (BH **)__db_malloc((size_t)nalloc * sizeof(BH *))) == NULL)
		return (ENOMEM);

	LOCKREGION(dbmp);

	/*
	 * Walk the LRU list of buffer headers, and get a list of buffers to
	 * write for this MPOOLFILE.
	 */
	ar_cnt = pincnt = 0;
	for (bhp = SH_TAILQ_FIRST(&dbmp->mp->bhq, __bh);
	    bhp != NULL; bhp = SH_TAILQ_NEXT(bhp, q, __bh)) {
		if (!F_ISSET(bhp, BH_DIRTY) || bhp->mf_offset != mf_offset)
			continue;
		if (bhp->ref != 0 || F_ISSET(bhp, BH_LOCKED)) {
			++pincnt;
			continue;
		}

		if (ar_cnt == nalloc) {
			nalloc *= 2;
			if ((bharray = (BH **)__db_realloc(bharray,
			    nalloc * sizeof(BH *))) == NULL) {
				ret = ENOMEM;
				goto err;
			}
		}

		bharray[ar_cnt++] = bhp;
	}

	/* Lock down the buffers and their contents. */
	for (cnt = 0; cnt < ar_cnt; ++cnt)
		++bharray[cnt]->ref;

	UNLOCKREGION(dbmp);

	/* Sort the buffers we're going to write. */
	qsort(bharray, ar_cnt, sizeof(BH *), __bhcmp);

	LOCKREGION(dbmp);

	/* Walk the array, writing buffers. */
	for (next = 0; next < ar_cnt; ++next) {
		/*
		 * It's possible for a thread to have gotten the buffer since
		 * we listed it for writing.  If the reference count is still
		 * 1, we're the only ones using the buffer, go ahead and write.
		 * If it's >1, then skip the buffer and assume that it will be
		 * written when it's returned to the cache.
		 */
		if (bharray[next]->ref > 1) {
			++pincnt;

			--bharray[next]->ref;
			continue;
		}

		/* Write the buffer. */
		ret = __memp_pgwrite(dbmfp, bharray[next], NULL, &wrote);

		/* Release the buffer. */
		--bharray[next]->ref;

		/* If there's an error, release the rest of the buffers. */
		if (ret != 0) {
			while (++next < ar_cnt)
				--bharray[next]->ref;
			goto err;
		}
		if (!wrote)
			++pincnt;
	}

err:	UNLOCKREGION(dbmp);

	__db_free(bharray);

	/*
	 * Sync the underlying file as the last thing we do, so that the OS
	 * has maximal opportunity to flush buffers before we request it.
	 *
	 * XXX:
	 * Don't lock the region around the sync, fsync(2) has no atomicity
	 * issues.
	 */
	if (ret == 0)
		return (pincnt == 0 ? __db_fsync(dbmfp->fd) : DB_INCOMPLETE);
	return (ret);

}

/*
 * memp_trickle --
 *	Keep a specified percentage of the buffers clean.
 */
int
memp_trickle(dbmp, pct, nwrotep)
	DB_MPOOL *dbmp;
	int pct, *nwrotep;
{
	BH *bhp;
	MPOOL *mp;
	MPOOLFILE *mfp;
	u_long total;
	int ret, wrote;

	mp = dbmp->mp;
	if (nwrotep != NULL)
		*nwrotep = 0;

	if (pct < 1 || pct > 100)
		return (EINVAL);

	LOCKREGION(dbmp);

	/*
	 * If there are sufficient clean buffers, or no buffers or no dirty
	 * buffers, we're done.
	 *
	 * XXX
	 * Using st_page_clean and st_page_dirty is our only choice at the
	 * moment, but it's not as correct as we might like in the presence
	 * of pools with more than one buffer size, as a free 512-byte buffer
	 * isn't the same as a free 8K buffer.
	 */
loop:	total = mp->stat.st_page_clean + mp->stat.st_page_dirty;
	if (total == 0 || mp->stat.st_page_dirty == 0 ||
	    (mp->stat.st_page_clean * 100) / total >= (u_long)pct) {
		UNLOCKREGION(dbmp);
		return (0);
	}

	/* Loop until we write a buffer. */
	for (bhp = SH_TAILQ_FIRST(&mp->bhq, __bh);
	    bhp != NULL; bhp = SH_TAILQ_NEXT(bhp, q, __bh)) {
		if (bhp->ref != 0 ||
		    !F_ISSET(bhp, BH_DIRTY) || F_ISSET(bhp, BH_LOCKED))
			continue;

		mfp = R_ADDR(dbmp, bhp->mf_offset);

		/*
		 * We can't write to temporary files -- see the comment in
		 * mp_bh.c:__memp_bhwrite().
		 */
		if (F_ISSET(mfp, MP_TEMP))
			continue;

		if ((ret = __memp_bhwrite(dbmp, mfp, bhp, NULL, &wrote)) != 0)
			goto err;

		/*
		 * Any process syncing the shared memory buffer pool had better
		 * be able to write to any underlying file.  Be understanding,
		 * but firm, on this point.
		 */
		if (!wrote) {
			__db_err(dbmp->dbenv, "%s: unable to flush page: %lu",
			    __memp_fns(dbmp, mfp), (u_long)bhp->pgno);
			ret = EPERM;
			goto err;
		}

		++mp->stat.st_page_trickle;
		if (nwrotep != NULL)
			++*nwrotep;
		goto loop;
	}

	/* No more buffers to write. */
	return (0);

err:	UNLOCKREGION(dbmp);
	return (ret);
}

static int
__bhcmp(p1, p2)
	const void *p1, *p2;
{
	BH *bhp1, *bhp2;

	bhp1 = *(BH **)p1;
	bhp2 = *(BH **)p2;

	/* Sort by file (shared memory pool offset). */
	if (bhp1->mf_offset < bhp2->mf_offset)
		return (-1);
	if (bhp1->mf_offset > bhp2->mf_offset)
		return (1);

	/* Sort by page in file. */
	return (bhp1->pgno < bhp2->pgno ? -1 : 1);
}