sequence.c 24.4 KB
Newer Older
1 2
/*-------------------------------------------------------------------------
 *
3
 * sequence.c
4
 *	  PostgreSQL sequences support code.
5
 *
B
Bruce Momjian 已提交
6
 * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
7 8 9 10
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
11
 *	  $Header: /cvsroot/pgsql/src/backend/commands/sequence.c,v 1.86 2002/09/03 18:50:54 petere Exp $
12
 *
13 14
 *-------------------------------------------------------------------------
 */
15
#include "postgres.h"
16

17
#include "access/heapam.h"
18
#include "catalog/namespace.h"
19
#include "catalog/pg_type.h"
20
#include "commands/defrem.h"
21
#include "commands/tablecmds.h"
22
#include "commands/sequence.h"
B
Bruce Momjian 已提交
23
#include "miscadmin.h"
24
#include "utils/acl.h"
B
Bruce Momjian 已提交
25
#include "utils/builtins.h"
26

27

28 29 30 31 32 33
#ifndef INT64_IS_BUSTED
#ifdef HAVE_LL_CONSTANTS
#define SEQ_MAXVALUE	((int64) 0x7FFFFFFFFFFFFFFFLL)
#else
#define SEQ_MAXVALUE	((int64) 0x7FFFFFFFFFFFFFFF)
#endif
34
#else							/* INT64_IS_BUSTED */
35
#define SEQ_MAXVALUE	((int64) 0x7FFFFFFF)
36
#endif   /* INT64_IS_BUSTED */
37 38

#define SEQ_MINVALUE	(-SEQ_MAXVALUE)
39

V
Vadim B. Mikheev 已提交
40
/*
41
 * We don't want to log each fetching of a value from a sequence,
V
Vadim B. Mikheev 已提交
42 43 44
 * so we pre-log a few fetches in advance. In the event of
 * crash we can lose as much as we pre-logged.
 */
B
Bruce Momjian 已提交
45
#define SEQ_LOG_VALS	32
46

47 48 49 50 51
/*
 * The "special area" of a sequence's buffer page looks like this.
 */
#define SEQ_MAGIC	  0x1717

52 53
typedef struct sequence_magic
{
54
	uint32		magic;
55
} sequence_magic;
56

57 58 59 60 61 62 63 64 65 66
/*
 * We store a SeqTable item for every sequence we have touched in the current
 * session.  This is needed to hold onto nextval/currval state.  (We can't
 * rely on the relcache, since it's only, well, a cache, and may decide to
 * discard entries.)
 *
 * XXX We use linear search to find pre-existing SeqTable entries.  This is
 * good when only a small number of sequences are touched in a session, but
 * would suck with many different sequences.  Perhaps use a hashtable someday.
 */
67 68
typedef struct SeqTableData
{
69 70 71 72 73 74 75
	struct SeqTableData *next;	/* link to next SeqTable object */
	Oid			relid;			/* pg_class OID of this sequence */
	TransactionId xid;			/* xact in which we last did a seq op */
	int64		last;			/* value last returned by nextval */
	int64		cached;			/* last value already cached for nextval */
	/* if last != cached, we have not used up all the cached values */
	int64		increment;		/* copy of sequence's increment field */
76
} SeqTableData;
77 78 79

typedef SeqTableData *SeqTable;

80
static SeqTable seqtab = NULL;	/* Head of list of SeqTable items */
81

82 83 84 85 86

static void init_sequence(const char *caller, RangeVar *relation,
						  SeqTable *p_elm, Relation *p_rel);
static Form_pg_sequence read_info(const char *caller, SeqTable elm,
								  Relation rel, Buffer *buf);
87
static void init_params(CreateSeqStmt *seq, Form_pg_sequence new);
88
static void do_setval(RangeVar *sequence, int64 next, bool iscalled);
89 90

/*
B
Bruce Momjian 已提交
91
 * DefineSequence
92
 *				Creates a new sequence relation
93 94
 */
void
95
DefineSequence(CreateSeqStmt *seq)
96
{
97
	FormData_pg_sequence new;
98
	CreateStmt *stmt = makeNode(CreateStmt);
99
	Oid			seqoid;
100 101 102
	Relation	rel;
	Buffer		buf;
	PageHeader	page;
103
	sequence_magic *sm;
104 105 106 107 108
	HeapTuple	tuple;
	TupleDesc	tupDesc;
	Datum		value[SEQ_COL_LASTCOL];
	char		null[SEQ_COL_LASTCOL];
	int			i;
109
	NameData	name;
110 111 112 113 114

	/* Check and set values */
	init_params(seq, &new);

	/*
115
	 * Create relation (and fill *null & *value)
116 117 118
	 */
	stmt->tableElts = NIL;
	for (i = SEQ_COL_FIRSTCOL; i <= SEQ_COL_LASTCOL; i++)
119
	{
120 121 122
		ColumnDef  *coldef;
		TypeName   *typnam;

123 124
		typnam = makeNode(TypeName);
		typnam->setof = FALSE;
125
		typnam->arrayBounds = NIL;
B
Bruce Momjian 已提交
126
		typnam->typmod = -1;
127

128 129
		coldef = makeNode(ColumnDef);
		coldef->typename = typnam;
130
		coldef->is_inherited = false;
131
		coldef->is_not_null = true;
132 133
		coldef->raw_default = NULL;
		coldef->cooked_default = NULL;
134 135 136
		coldef->constraints = NIL;
		coldef->support = NULL;

137 138 139 140
		null[i - 1] = ' ';

		switch (i)
		{
141
			case SEQ_COL_NAME:
142
				typnam->typeid = NAMEOID;
143
				coldef->colname = "sequence_name";
144
				namestrcpy(&name, seq->sequence->relname);
145
				value[i - 1] = NameGetDatum(&name);
146 147
				break;
			case SEQ_COL_LASTVAL:
148
				typnam->typeid = INT8OID;
149
				coldef->colname = "last_value";
150
				value[i - 1] = Int64GetDatumFast(new.last_value);
151 152
				break;
			case SEQ_COL_INCBY:
153
				typnam->typeid = INT8OID;
154
				coldef->colname = "increment_by";
155
				value[i - 1] = Int64GetDatumFast(new.increment_by);
156 157
				break;
			case SEQ_COL_MAXVALUE:
158
				typnam->typeid = INT8OID;
159
				coldef->colname = "max_value";
160
				value[i - 1] = Int64GetDatumFast(new.max_value);
161 162
				break;
			case SEQ_COL_MINVALUE:
163
				typnam->typeid = INT8OID;
164
				coldef->colname = "min_value";
165
				value[i - 1] = Int64GetDatumFast(new.min_value);
166 167
				break;
			case SEQ_COL_CACHE:
168
				typnam->typeid = INT8OID;
169
				coldef->colname = "cache_value";
170
				value[i - 1] = Int64GetDatumFast(new.cache_value);
171
				break;
V
Vadim B. Mikheev 已提交
172
			case SEQ_COL_LOG:
173
				typnam->typeid = INT8OID;
V
Vadim B. Mikheev 已提交
174
				coldef->colname = "log_cnt";
175
				value[i - 1] = Int64GetDatum((int64) 1);
V
Vadim B. Mikheev 已提交
176
				break;
177
			case SEQ_COL_CYCLE:
178
				typnam->typeid = BOOLOID;
179
				coldef->colname = "is_cycled";
180
				value[i - 1] = BoolGetDatum(new.is_cycled);
181 182
				break;
			case SEQ_COL_CALLED:
183
				typnam->typeid = BOOLOID;
184
				coldef->colname = "is_called";
185
				value[i - 1] = BoolGetDatum(false);
186
				break;
187 188 189 190
		}
		stmt->tableElts = lappend(stmt->tableElts, coldef);
	}

191 192
	stmt->relation = seq->sequence;
	stmt->inhRelations = NIL;
193
	stmt->constraints = NIL;
194
	stmt->hasoids = false;
195

196
	seqoid = DefineRelation(stmt, RELKIND_SEQUENCE);
197

198
	rel = heap_open(seqoid, AccessExclusiveLock);
199
	tupDesc = RelationGetDescr(rel);
200

201 202
	/* Initialize first page of relation with special magic number */

203 204 205
	buf = ReadBuffer(rel, P_NEW);

	if (!BufferIsValid(buf))
206
		elog(ERROR, "DefineSequence: ReadBuffer failed");
207

208 209
	Assert(BufferGetBlockNumber(buf) == 0);

210 211 212 213 214 215
	page = (PageHeader) BufferGetPage(buf);

	PageInit((Page) page, BufferGetPageSize(buf), sizeof(sequence_magic));
	sm = (sequence_magic *) PageGetSpecialPointer(page);
	sm->magic = SEQ_MAGIC;

216 217 218
	/* hack: ensure heap_insert will insert on the just-created page */
	rel->rd_targblock = 0;

219
	/* Now form & insert sequence tuple */
220
	tuple = heap_formtuple(tupDesc, value, null);
221
	simple_heap_insert(rel, tuple);
222

223 224
	Assert(ItemPointerGetOffsetNumber(&(tuple->t_self)) == FirstOffsetNumber);

225
	/*
226 227 228 229 230 231 232 233 234 235 236 237 238
	 * Two special hacks here:
	 *
	 * 1. Since VACUUM does not process sequences, we have to force the tuple
	 * to have xmin = FrozenTransactionId now.  Otherwise it would become
	 * invisible to SELECTs after 2G transactions.  It is okay to do this
	 * because if the current transaction aborts, no other xact will ever
	 * examine the sequence tuple anyway.
	 *
	 * 2. Even though heap_insert emitted a WAL log record, we have to emit
	 * an XLOG_SEQ_LOG record too, since (a) the heap_insert record will
	 * not have the right xmin, and (b) REDO of the heap_insert record
	 * would re-init page and sequence magic number would be lost.  This
	 * means two log records instead of one :-(
239
	 */
240
	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
241

242
	START_CRIT_SECTION();
243 244 245 246 247 248 249 250 251 252 253 254 255 256 257

	{
		/*
		 * Note that the "tuple" structure is still just a local tuple record
		 * created by heap_formtuple; its t_data pointer doesn't point at the
		 * disk buffer.  To scribble on the disk buffer we need to fetch the
		 * item pointer.  But do the same to the local tuple, since that will
		 * be the source for the WAL log record, below.
		 */
		ItemId		itemId;
		Item		item;

		itemId = PageGetItemId((Page) page, FirstOffsetNumber);
		item = PageGetItem((Page) page, itemId);

258
		HeapTupleHeaderSetXmin((HeapTupleHeader) item, FrozenTransactionId);
259 260
		((HeapTupleHeader) item)->t_infomask |= HEAP_XMIN_COMMITTED;

261
		HeapTupleHeaderSetXmin(tuple->t_data, FrozenTransactionId);
262 263 264
		tuple->t_data->t_infomask |= HEAP_XMIN_COMMITTED;
	}

265 266
	/* XLOG stuff */
	if (!rel->rd_istemp)
267
	{
268 269 270 271
		xl_seq_rec	xlrec;
		XLogRecPtr	recptr;
		XLogRecData rdata[2];
		Form_pg_sequence newseq = (Form_pg_sequence) GETSTRUCT(tuple);
272 273

		/* We do not log first nextval call, so "advance" sequence here */
274
		/* Note we are scribbling on local tuple, not the disk buffer */
275
		newseq->is_called = true;
276 277 278 279 280 281 282 283 284
		newseq->log_cnt = 0;

		xlrec.node = rel->rd_node;
		rdata[0].buffer = InvalidBuffer;
		rdata[0].data = (char *) &xlrec;
		rdata[0].len = sizeof(xl_seq_rec);
		rdata[0].next = &(rdata[1]);

		rdata[1].buffer = InvalidBuffer;
285
		rdata[1].data = (char *) tuple->t_data;
286 287 288 289 290 291 292 293
		rdata[1].len = tuple->t_len;
		rdata[1].next = NULL;

		recptr = XLogInsert(RM_SEQ_ID, XLOG_SEQ_LOG | XLOG_NO_TRAN, rdata);

		PageSetLSN(page, recptr);
		PageSetSUI(page, ThisStartUpID);
	}
294

295
	END_CRIT_SECTION();
296

297
	LockBuffer(buf, BUFFER_LOCK_UNLOCK);
298 299
	WriteBuffer(buf);
	heap_close(rel, NoLock);
300 301 302
}


303 304
Datum
nextval(PG_FUNCTION_ARGS)
305
{
306
	text	   *seqin = PG_GETARG_TEXT_P(0);
307
	RangeVar   *sequence;
308
	SeqTable	elm;
309
	Relation	seqrel;
310
	Buffer		buf;
311
	Page		page;
312
	Form_pg_sequence seq;
313
	int64		incby,
314 315
				maxv,
				minv,
V
Vadim B. Mikheev 已提交
316 317 318 319
				cache,
				log,
				fetch,
				last;
320
	int64		result,
321 322
				next,
				rescnt = 0;
V
Vadim B. Mikheev 已提交
323
	bool		logit = false;
324

325 326 327
	sequence = makeRangeVarFromNameList(textToQualifiedNameList(seqin,
																"nextval"));

V
Vadim B. Mikheev 已提交
328
	/* open and AccessShareLock sequence */
329
	init_sequence("nextval", sequence, &elm, &seqrel);
330

331 332
	if (pg_class_aclcheck(elm->relid, GetUserId(), ACL_UPDATE) != ACLCHECK_OK)
		elog(ERROR, "%s.nextval: you don't have permissions to set sequence %s",
333
			 sequence->relname, sequence->relname);
334 335 336 337

	if (elm->last != elm->cached)		/* some numbers were cached */
	{
		elm->last += elm->increment;
338
		relation_close(seqrel, NoLock);
339
		PG_RETURN_INT64(elm->last);
340
	}
341

342 343
	/* lock page' buffer and read tuple */
	seq = read_info("nextval", elm, seqrel, &buf);
344
	page = BufferGetPage(buf);
345

V
Vadim B. Mikheev 已提交
346
	last = next = result = seq->last_value;
347 348 349
	incby = seq->increment_by;
	maxv = seq->max_value;
	minv = seq->min_value;
V
Vadim B. Mikheev 已提交
350 351
	fetch = cache = seq->cache_value;
	log = seq->log_cnt;
352

353
	if (!seq->is_called)
V
Vadim B. Mikheev 已提交
354
	{
355
		rescnt++;				/* last_value if not called */
V
Vadim B. Mikheev 已提交
356 357 358
		fetch--;
		log--;
	}
359

360 361 362 363 364 365 366 367 368 369
	/*
	 * Decide whether we should emit a WAL log record.  If so, force up
	 * the fetch count to grab SEQ_LOG_VALS more values than we actually
	 * need to cache.  (These will then be usable without logging.)
	 *
	 * If this is the first nextval after a checkpoint, we must force
	 * a new WAL record to be written anyway, else replay starting from the
	 * checkpoint would fail to advance the sequence past the logged
	 * values.  In this case we may as well fetch extra values.
	 */
V
Vadim B. Mikheev 已提交
370 371
	if (log < fetch)
	{
372 373
		/* forced log to satisfy local demand for values */
		fetch = log = fetch + SEQ_LOG_VALS;
V
Vadim B. Mikheev 已提交
374 375
		logit = true;
	}
376 377 378 379 380 381 382 383 384 385 386
	else
	{
		XLogRecPtr	redoptr = GetRedoRecPtr();

		if (XLByteLE(PageGetLSN(page), redoptr))
		{
			/* last update of seq was before checkpoint */
			fetch = log = fetch + SEQ_LOG_VALS;
			logit = true;
		}
	}
V
Vadim B. Mikheev 已提交
387

B
Bruce Momjian 已提交
388
	while (fetch)				/* try to fetch cache [+ log ] numbers */
389
	{
390 391 392 393
		/*
		 * Check MAXVALUE for ascending sequences and MINVALUE for
		 * descending sequences
		 */
394
		if (incby > 0)
395
		{
396
			/* ascending sequence */
397 398 399 400
			if ((maxv >= 0 && next > maxv - incby) ||
				(maxv < 0 && next + incby > maxv))
			{
				if (rescnt > 0)
V
Vadim B. Mikheev 已提交
401
					break;		/* stop fetching */
402
				if (!seq->is_cycled)
403 404 405 406 407 408
				{
					char buf[100];
					snprintf(buf, 100, INT64_FORMAT, maxv);
					elog(ERROR, "%s.nextval: reached MAXVALUE (%s)",
						 sequence->relname, buf);
				}
409 410 411 412 413 414 415
				next = minv;
			}
			else
				next += incby;
		}
		else
		{
416
			/* descending sequence */
417 418 419 420
			if ((minv < 0 && next < minv - incby) ||
				(minv >= 0 && next + incby < minv))
			{
				if (rescnt > 0)
V
Vadim B. Mikheev 已提交
421
					break;		/* stop fetching */
422
				if (!seq->is_cycled)
423 424 425 426 427 428
				{
					char buf[100];
					snprintf(buf, 100, INT64_FORMAT, minv);
					elog(ERROR, "%s.nextval: reached MINVALUE (%s)",
						 sequence->relname, buf);
				}
429 430 431 432 433
				next = maxv;
			}
			else
				next += incby;
		}
V
Vadim B. Mikheev 已提交
434 435 436 437 438 439
		fetch--;
		if (rescnt < cache)
		{
			log--;
			rescnt++;
			last = next;
B
Bruce Momjian 已提交
440 441
			if (rescnt == 1)	/* if it's first result - */
				result = next;	/* it's what to return */
V
Vadim B. Mikheev 已提交
442
		}
443 444
	}

445 446 447
	log -= fetch;				/* adjust for any unfetched numbers */
	Assert(log >= 0);

448 449
	/* save info in local cache */
	elm->last = result;			/* last returned number */
V
Vadim B. Mikheev 已提交
450 451
	elm->cached = last;			/* last fetched number */

452
	START_CRIT_SECTION();
453 454 455

	/* XLOG stuff */
	if (logit && !seqrel->rd_istemp)
V
Vadim B. Mikheev 已提交
456 457 458
	{
		xl_seq_rec	xlrec;
		XLogRecPtr	recptr;
B
Bruce Momjian 已提交
459
		XLogRecData rdata[2];
V
Vadim B. Mikheev 已提交
460

461
		xlrec.node = seqrel->rd_node;
462
		rdata[0].buffer = InvalidBuffer;
B
Bruce Momjian 已提交
463
		rdata[0].data = (char *) &xlrec;
464 465 466
		rdata[0].len = sizeof(xl_seq_rec);
		rdata[0].next = &(rdata[1]);

467
		/* set values that will be saved in xlog */
468
		seq->last_value = next;
469
		seq->is_called = true;
470
		seq->log_cnt = 0;
471

472
		rdata[1].buffer = InvalidBuffer;
B
Bruce Momjian 已提交
473 474 475
		rdata[1].data = (char *) page + ((PageHeader) page)->pd_upper;
		rdata[1].len = ((PageHeader) page)->pd_special -
			((PageHeader) page)->pd_upper;
476 477
		rdata[1].next = NULL;

B
Bruce Momjian 已提交
478
		recptr = XLogInsert(RM_SEQ_ID, XLOG_SEQ_LOG | XLOG_NO_TRAN, rdata);
V
Vadim B. Mikheev 已提交
479

480 481
		PageSetLSN(page, recptr);
		PageSetSUI(page, ThisStartUpID);
V
Vadim B. Mikheev 已提交
482
	}
483

484
	/* update on-disk data */
V
Vadim B. Mikheev 已提交
485
	seq->last_value = last;		/* last fetched number */
486
	seq->is_called = true;
V
Vadim B. Mikheev 已提交
487
	seq->log_cnt = log;			/* how much is logged */
488

489
	END_CRIT_SECTION();
490

V
Vadim B. Mikheev 已提交
491 492
	LockBuffer(buf, BUFFER_LOCK_UNLOCK);

B
Bruce Momjian 已提交
493
	WriteBuffer(buf);
494

495 496
	relation_close(seqrel, NoLock);

497
	PG_RETURN_INT64(result);
498 499
}

500 501
Datum
currval(PG_FUNCTION_ARGS)
502
{
503
	text	   *seqin = PG_GETARG_TEXT_P(0);
504
	RangeVar   *sequence;
505
	SeqTable	elm;
506
	Relation	seqrel;
507
	int64		result;
508

509 510 511
	sequence = makeRangeVarFromNameList(textToQualifiedNameList(seqin,
																"currval"));

V
Vadim B. Mikheev 已提交
512
	/* open and AccessShareLock sequence */
513
	init_sequence("currval", sequence, &elm, &seqrel);
514

515 516
	if (pg_class_aclcheck(elm->relid, GetUserId(), ACL_SELECT) != ACLCHECK_OK)
		elog(ERROR, "%s.currval: you don't have permissions to read sequence %s",
517
			 sequence->relname, sequence->relname);
518

519
	if (elm->increment == 0)	/* nextval/read_info were not called */
520
		elog(ERROR, "%s.currval is not yet defined in this session",
521
			 sequence->relname);
522 523 524

	result = elm->last;

525 526
	relation_close(seqrel, NoLock);

527
	PG_RETURN_INT64(result);
528 529
}

B
Bruce Momjian 已提交
530
/*
531 532 533 534
 * Main internal procedure that handles 2 & 3 arg forms of SETVAL.
 *
 * Note that the 3 arg version (which sets the is_called flag) is
 * only for use in pg_dump, and setting the is_called flag may not
B
Bruce Momjian 已提交
535
 * work if multiple users are attached to the database and referencing
536 537
 * the sequence (unlikely if pg_dump is restoring it).
 *
B
Bruce Momjian 已提交
538
 * It is necessary to have the 3 arg version so that pg_dump can
539 540 541 542
 * restore the state of a sequence exactly during data-only restores -
 * it is the only way to clear the is_called flag in an existing
 * sequence.
 */
B
Bruce Momjian 已提交
543
static void
544
do_setval(RangeVar *sequence, int64 next, bool iscalled)
M
 
Marc G. Fournier 已提交
545 546
{
	SeqTable	elm;
547
	Relation	seqrel;
548
	Buffer		buf;
549
	Form_pg_sequence seq;
M
 
Marc G. Fournier 已提交
550

551
	/* open and AccessShareLock sequence */
552
	init_sequence("setval", sequence, &elm, &seqrel);
553 554

	if (pg_class_aclcheck(elm->relid, GetUserId(), ACL_UPDATE) != ACLCHECK_OK)
M
 
Marc G. Fournier 已提交
555
		elog(ERROR, "%s.setval: you don't have permissions to set sequence %s",
556
			 sequence->relname, sequence->relname);
M
 
Marc G. Fournier 已提交
557

558
	/* lock page' buffer and read tuple */
559
	seq = read_info("setval", elm, seqrel, &buf);
M
 
Marc G. Fournier 已提交
560

561
	if ((next < seq->min_value) || (next > seq->max_value))
562 563 564 565 566 567 568 569
	{
		char bufv[100], bufm[100], bufx[100];
		snprintf(bufv, 100, INT64_FORMAT, next);
		snprintf(bufm, 100, INT64_FORMAT, seq->min_value);
		snprintf(bufx, 100, INT64_FORMAT, seq->max_value);
		elog(ERROR, "%s.setval: value %s is out of bounds (%s,%s)",
			 sequence->relname, bufv, bufm, bufx);
	}
M
 
Marc G. Fournier 已提交
570 571 572

	/* save info in local cache */
	elm->last = next;			/* last returned number */
B
Bruce Momjian 已提交
573 574
	elm->cached = next;			/* last cached number (forget cached
								 * values) */
M
 
Marc G. Fournier 已提交
575

576
	START_CRIT_SECTION();
577 578 579

	/* XLOG stuff */
	if (!seqrel->rd_istemp)
V
Vadim B. Mikheev 已提交
580 581 582
	{
		xl_seq_rec	xlrec;
		XLogRecPtr	recptr;
B
Bruce Momjian 已提交
583
		XLogRecData rdata[2];
584
		Page		page = BufferGetPage(buf);
V
Vadim B. Mikheev 已提交
585

586
		xlrec.node = seqrel->rd_node;
587
		rdata[0].buffer = InvalidBuffer;
B
Bruce Momjian 已提交
588
		rdata[0].data = (char *) &xlrec;
589 590 591
		rdata[0].len = sizeof(xl_seq_rec);
		rdata[0].next = &(rdata[1]);

592
		/* set values that will be saved in xlog */
593
		seq->last_value = next;
594
		seq->is_called = true;
595
		seq->log_cnt = 0;
596

597
		rdata[1].buffer = InvalidBuffer;
B
Bruce Momjian 已提交
598 599 600
		rdata[1].data = (char *) page + ((PageHeader) page)->pd_upper;
		rdata[1].len = ((PageHeader) page)->pd_special -
			((PageHeader) page)->pd_upper;
601 602
		rdata[1].next = NULL;

B
Bruce Momjian 已提交
603
		recptr = XLogInsert(RM_SEQ_ID, XLOG_SEQ_LOG | XLOG_NO_TRAN, rdata);
604 605 606

		PageSetLSN(page, recptr);
		PageSetSUI(page, ThisStartUpID);
V
Vadim B. Mikheev 已提交
607
	}
608

609 610
	/* save info in sequence relation */
	seq->last_value = next;		/* last fetched number */
611
	seq->is_called = iscalled;
612
	seq->log_cnt = (iscalled) ? 0 : 1;
613

614
	END_CRIT_SECTION();
M
 
Marc G. Fournier 已提交
615

V
Vadim B. Mikheev 已提交
616 617
	LockBuffer(buf, BUFFER_LOCK_UNLOCK);

B
Bruce Momjian 已提交
618
	WriteBuffer(buf);
619 620

	relation_close(seqrel, NoLock);
621 622
}

623 624 625 626
/*
 * Implement the 2 arg setval procedure.
 * See do_setval for discussion.
 */
627 628 629 630
Datum
setval(PG_FUNCTION_ARGS)
{
	text	   *seqin = PG_GETARG_TEXT_P(0);
631
	int64		next = PG_GETARG_INT64(1);
632 633 634 635
	RangeVar   *sequence;

	sequence = makeRangeVarFromNameList(textToQualifiedNameList(seqin,
																"setval"));
636

637
	do_setval(sequence, next, true);
638

639
	PG_RETURN_INT64(next);
640 641
}

642 643 644 645
/*
 * Implement the 3 arg setval procedure.
 * See do_setval for discussion.
 */
646 647 648 649
Datum
setval_and_iscalled(PG_FUNCTION_ARGS)
{
	text	   *seqin = PG_GETARG_TEXT_P(0);
650
	int64		next = PG_GETARG_INT64(1);
651
	bool		iscalled = PG_GETARG_BOOL(2);
652
	RangeVar   *sequence;
653

654 655
	sequence = makeRangeVarFromNameList(textToQualifiedNameList(seqin,
																"setval"));
656

657
	do_setval(sequence, next, iscalled);
658

659
	PG_RETURN_INT64(next);
M
 
Marc G. Fournier 已提交
660 661
}

662

663 664 665 666 667 668 669
/*
 * Given a relation name, open and lock the sequence.  p_elm and p_rel are
 * output parameters.
 */
static void
init_sequence(const char *caller, RangeVar *relation,
			  SeqTable *p_elm, Relation *p_rel)
670
{
671
	Oid			relid = RangeVarGetRelid(relation, false);
672 673
	TransactionId thisxid = GetCurrentTransactionId();
	SeqTable	elm;
674
	Relation	seqrel;
675 676
	
	/* Look to see if we already have a seqtable entry for relation */
677
	for (elm = seqtab; elm != NULL; elm = elm->next)
678
	{
679
		if (elm->relid == relid)
680 681 682
			break;
	}

683 684 685 686 687 688 689 690
	/*
	 * Open the sequence relation, acquiring AccessShareLock if we don't
	 * already have a lock in the current xact.
	 */
	if (elm == NULL || elm->xid != thisxid)
		seqrel = relation_open(relid, AccessShareLock);
	else
		seqrel = relation_open(relid, NoLock);
691

692
	if (seqrel->rd_rel->relkind != RELKIND_SEQUENCE)
693 694
		elog(ERROR, "%s.%s: %s is not a sequence",
			 relation->relname, caller, relation->relname);
695

696
	/*
697
	 * Allocate new seqtable entry if we didn't find one.
698 699 700 701 702
	 *
	 * NOTE: seqtable entries remain in the list for the life of a backend.
	 * If the sequence itself is deleted then the entry becomes wasted memory,
	 * but it's small enough that this should not matter.
	 */ 
703
	if (elm == NULL)
704
	{
705 706
		/*
		 * Time to make a new seqtable entry.  These entries live as long
707 708 709
		 * as the backend does, so we use plain malloc for them.
		 */
		elm = (SeqTable) malloc(sizeof(SeqTableData));
T
Tom Lane 已提交
710 711
		if (elm == NULL)
			elog(ERROR, "Memory exhausted in init_sequence");
712
		elm->relid = relid;
713 714 715 716
		/* increment is set to 0 until we do read_info (see currval) */
		elm->last = elm->cached = elm->increment = 0;
		elm->next = seqtab;
		seqtab = elm;
717 718
	}

719 720 721 722 723
	/* Flag that we have a lock in the current xact. */
	elm->xid = thisxid;

	*p_elm = elm;
	*p_rel = seqrel;
724 725 726
}


727 728 729 730
/* Given an opened relation, lock the page buffer and find the tuple */
static Form_pg_sequence
read_info(const char *caller, SeqTable elm,
		  Relation rel, Buffer *buf)
731
{
732 733 734 735 736
	PageHeader	page;
	ItemId		lp;
	HeapTupleData tuple;
	sequence_magic *sm;
	Form_pg_sequence seq;
737

738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764
	if (rel->rd_nblocks > 1)
		elog(ERROR, "%s.%s: invalid number of blocks in sequence",
			 RelationGetRelationName(rel), caller);

	*buf = ReadBuffer(rel, 0);
	if (!BufferIsValid(*buf))
		elog(ERROR, "%s.%s: ReadBuffer failed",
			 RelationGetRelationName(rel), caller);

	LockBuffer(*buf, BUFFER_LOCK_EXCLUSIVE);

	page = (PageHeader) BufferGetPage(*buf);
	sm = (sequence_magic *) PageGetSpecialPointer(page);

	if (sm->magic != SEQ_MAGIC)
		elog(ERROR, "%s.%s: bad magic (%08X)",
			 RelationGetRelationName(rel), caller, sm->magic);

	lp = PageGetItemId(page, FirstOffsetNumber);
	Assert(ItemIdIsUsed(lp));
	tuple.t_data = (HeapTupleHeader) PageGetItem((Page) page, lp);

	seq = (Form_pg_sequence) GETSTRUCT(&tuple);

	elm->increment = seq->increment_by;

	return seq;
765 766 767
}


768
static void
769
init_params(CreateSeqStmt *seq, Form_pg_sequence new)
770
{
771 772 773 774 775 776
	DefElem    *last_value = NULL;
	DefElem    *increment_by = NULL;
	DefElem    *max_value = NULL;
	DefElem    *min_value = NULL;
	DefElem    *cache_value = NULL;
	List	   *option;
777

778
	new->is_cycled = false;
779 780
	foreach(option, seq->options)
	{
781
		DefElem    *defel = (DefElem *) lfirst(option);
782

783
		if (strcmp(defel->defname, "increment") == 0)
784
			increment_by = defel;
785
		else if (strcmp(defel->defname, "start") == 0)
786
			last_value = defel;
787
		else if (strcmp(defel->defname, "maxvalue") == 0)
788
			max_value = defel;
789
		else if (strcmp(defel->defname, "minvalue") == 0)
790
			min_value = defel;
791
		else if (strcmp(defel->defname, "cache") == 0)
792
			cache_value = defel;
793
		else if (strcmp(defel->defname, "cycle") == 0)
794 795
		{
			if (defel->arg != (Node *) NULL)
796
				elog(ERROR, "DefineSequence: CYCLE ??");
797
			new->is_cycled = true;
798 799
		}
		else
800
			elog(ERROR, "DefineSequence: option \"%s\" not recognized",
801 802 803 804 805
				 defel->defname);
	}

	if (increment_by == (DefElem *) NULL)		/* INCREMENT BY */
		new->increment_by = 1;
806
	else if ((new->increment_by = defGetInt64(increment_by)) == 0)
807
		elog(ERROR, "DefineSequence: can't INCREMENT by 0");
808 809

	if (max_value == (DefElem *) NULL)	/* MAXVALUE */
810
	{
811 812 813
		if (new->increment_by > 0)
			new->max_value = SEQ_MAXVALUE;		/* ascending seq */
		else
814
			new->max_value = -1;	/* descending seq */
815
	}
816
	else
817
		new->max_value = defGetInt64(max_value);
818

819
	if (min_value == (DefElem *) NULL)	/* MINVALUE */
820
	{
821 822 823 824
		if (new->increment_by > 0)
			new->min_value = 1; /* ascending seq */
		else
			new->min_value = SEQ_MINVALUE;		/* descending seq */
825
	}
826
	else
827
		new->min_value = defGetInt64(min_value);
828 829

	if (new->min_value >= new->max_value)
830 831 832 833 834 835 836
	{
		char bufm[100], bufx[100];
		snprintf(bufm, 100, INT64_FORMAT, new->min_value);
		snprintf(bufx, 100, INT64_FORMAT, new->max_value);
		elog(ERROR, "DefineSequence: MINVALUE (%s) must be less than MAXVALUE (%s)",
			 bufm, bufx);
	}
837 838

	if (last_value == (DefElem *) NULL) /* START WITH */
839
	{
840 841 842 843
		if (new->increment_by > 0)
			new->last_value = new->min_value;	/* ascending seq */
		else
			new->last_value = new->max_value;	/* descending seq */
844
	}
845
	else
846
		new->last_value = defGetInt64(last_value);
847 848

	if (new->last_value < new->min_value)
849 850 851 852 853 854 855
	{
		char bufs[100], bufm[100];
		snprintf(bufs, 100, INT64_FORMAT, new->last_value);
		snprintf(bufm, 100, INT64_FORMAT, new->min_value);
		elog(ERROR, "DefineSequence: START value (%s) can't be less than MINVALUE (%s)",
			 bufs, bufm);
	}
856
	if (new->last_value > new->max_value)
857 858 859 860 861 862 863
	{
		char bufs[100], bufm[100];
		snprintf(bufs, 100, INT64_FORMAT, new->last_value);
		snprintf(bufm, 100, INT64_FORMAT, new->max_value);
		elog(ERROR, "DefineSequence: START value (%s) can't be greater than MAXVALUE (%s)",
			 bufs, bufm);
	}
864 865 866

	if (cache_value == (DefElem *) NULL)		/* CACHE */
		new->cache_value = 1;
867
	else if ((new->cache_value = defGetInt64(cache_value)) <= 0)
868 869 870 871 872 873
	{
		char buf[100];
		snprintf(buf, 100, INT64_FORMAT, new->cache_value);
		elog(ERROR, "DefineSequence: CACHE (%s) can't be <= 0",
			 buf);
	}
874 875 876

}

V
Vadim B. Mikheev 已提交
877

B
Bruce Momjian 已提交
878 879
void
seq_redo(XLogRecPtr lsn, XLogRecord *record)
V
Vadim B. Mikheev 已提交
880
{
B
Bruce Momjian 已提交
881 882 883 884 885 886 887
	uint8		info = record->xl_info & ~XLR_INFO_MASK;
	Relation	reln;
	Buffer		buffer;
	Page		page;
	char	   *item;
	Size		itemsz;
	xl_seq_rec *xlrec = (xl_seq_rec *) XLogRecGetData(record);
888
	sequence_magic *sm;
V
Vadim B. Mikheev 已提交
889

890
	if (info != XLOG_SEQ_LOG)
891
		elog(PANIC, "seq_redo: unknown op code %u", info);
V
Vadim B. Mikheev 已提交
892 893 894 895 896

	reln = XLogOpenRelation(true, RM_SEQ_ID, xlrec->node);
	if (!RelationIsValid(reln))
		return;

897
	buffer = XLogReadBuffer(true, reln, 0);
V
Vadim B. Mikheev 已提交
898
	if (!BufferIsValid(buffer))
899
		elog(PANIC, "seq_redo: can't read block of %u/%u",
B
Bruce Momjian 已提交
900
			 xlrec->node.tblNode, xlrec->node.relNode);
V
Vadim B. Mikheev 已提交
901 902 903

	page = (Page) BufferGetPage(buffer);

904 905
	/* Always reinit the page and reinstall the magic number */
	/* See comments in DefineSequence */
906 907 908
	PageInit((Page) page, BufferGetPageSize(buffer), sizeof(sequence_magic));
	sm = (sequence_magic *) PageGetSpecialPointer(page);
	sm->magic = SEQ_MAGIC;
V
Vadim B. Mikheev 已提交
909

B
Bruce Momjian 已提交
910
	item = (char *) xlrec + sizeof(xl_seq_rec);
911 912
	itemsz = record->xl_len - sizeof(xl_seq_rec);
	itemsz = MAXALIGN(itemsz);
B
Bruce Momjian 已提交
913
	if (PageAddItem(page, (Item) item, itemsz,
914
					FirstOffsetNumber, LP_USED) == InvalidOffsetNumber)
915
		elog(PANIC, "seq_redo: failed to add item to page");
V
Vadim B. Mikheev 已提交
916 917 918 919 920 921

	PageSetLSN(page, lsn);
	PageSetSUI(page, ThisStartUpID);
	UnlockAndWriteBuffer(buffer);
}

B
Bruce Momjian 已提交
922 923
void
seq_undo(XLogRecPtr lsn, XLogRecord *record)
V
Vadim B. Mikheev 已提交
924 925 926
{
}

B
Bruce Momjian 已提交
927 928
void
seq_desc(char *buf, uint8 xl_info, char *rec)
V
Vadim B. Mikheev 已提交
929
{
B
Bruce Momjian 已提交
930 931
	uint8		info = xl_info & ~XLR_INFO_MASK;
	xl_seq_rec *xlrec = (xl_seq_rec *) rec;
V
Vadim B. Mikheev 已提交
932 933 934 935 936 937 938 939 940

	if (info == XLOG_SEQ_LOG)
		strcat(buf, "log: ");
	else
	{
		strcat(buf, "UNKNOWN");
		return;
	}

941
	sprintf(buf + strlen(buf), "node %u/%u",
B
Bruce Momjian 已提交
942
			xlrec->node.tblNode, xlrec->node.relNode);
V
Vadim B. Mikheev 已提交
943
}