gistxlog.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

/*-------------------------------------------------------------------------
 *
 * gistxlog.c
 *	  WAL replay logic for GiST.
 *
 *
 * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 * IDENTIFICATION
 *			 $PostgreSQL: pgsql/src/backend/access/gist/gistxlog.c,v 1.9 2005/09/22 20:44:36 momjian Exp $
 *-------------------------------------------------------------------------
 */
#include "postgres.h"

#include "access/genam.h"
#include "access/gist_private.h"
#include "access/gistscan.h"
#include "access/heapam.h"
#include "catalog/index.h"
#include "commands/vacuum.h"
#include "miscadmin.h"
#include "utils/memutils.h"


typedef struct
{
	gistxlogEntryUpdate *data;
	int			len;
	IndexTuple *itup;
	OffsetNumber *todelete;
} EntryUpdateRecord;

typedef struct
{
	gistxlogPage *header;
	IndexTuple *itup;
} NewPage;

typedef struct
{
	gistxlogPageSplit *data;
	NewPage    *page;
} PageSplitRecord;

/* track for incomplete inserts, idea was taken from nbtxlog.c */

typedef struct gistIncompleteInsert
{
	RelFileNode node;
	BlockNumber origblkno;		/* for splits */
	ItemPointerData key;
	int			lenblk;
	BlockNumber *blkno;
	XLogRecPtr	lsn;
	BlockNumber *path;
	int			pathlen;
} gistIncompleteInsert;


MemoryContext opCtx;
MemoryContext insertCtx;
static List *incomplete_inserts;


#define ItemPointerEQ( a, b )	\
	( \
	ItemPointerGetOffsetNumber(a) == ItemPointerGetOffsetNumber(b) && \
	ItemPointerGetBlockNumber (a) == ItemPointerGetBlockNumber(b) \
		)

static void
pushIncompleteInsert(RelFileNode node, XLogRecPtr lsn, ItemPointerData key,
					 BlockNumber *blkno, int lenblk,
					 PageSplitRecord *xlinfo /* to extract blkno info */ )
{
	MemoryContext oldCxt = MemoryContextSwitchTo(insertCtx);
	gistIncompleteInsert *ninsert = (gistIncompleteInsert *) palloc(sizeof(gistIncompleteInsert));

	ninsert->node = node;
	ninsert->key = key;
	ninsert->lsn = lsn;

	if (lenblk && blkno)
	{
		ninsert->lenblk = lenblk;
		ninsert->blkno = (BlockNumber *) palloc(sizeof(BlockNumber) * ninsert->lenblk);
		memcpy(ninsert->blkno, blkno, sizeof(BlockNumber) * ninsert->lenblk);
		ninsert->origblkno = *blkno;
	}
	else
	{
		int			i;

		Assert(xlinfo);
		ninsert->lenblk = xlinfo->data->npage;
		ninsert->blkno = (BlockNumber *) palloc(sizeof(BlockNumber) * ninsert->lenblk);
		for (i = 0; i < ninsert->lenblk; i++)
			ninsert->blkno[i] = xlinfo->page[i].header->blkno;
		ninsert->origblkno = xlinfo->data->origblkno;
	}
	Assert(ninsert->lenblk > 0);

	incomplete_inserts = lappend(incomplete_inserts, ninsert);
	MemoryContextSwitchTo(oldCxt);
}

static void
forgetIncompleteInsert(RelFileNode node, ItemPointerData key)
{
	ListCell   *l;

	foreach(l, incomplete_inserts)
	{
		gistIncompleteInsert *insert = (gistIncompleteInsert *) lfirst(l);

		if (RelFileNodeEquals(node, insert->node) && ItemPointerEQ(&(insert->key), &(key)))
		{

			/* found */
			pfree(insert->blkno);
			incomplete_inserts = list_delete_ptr(incomplete_inserts, insert);
			pfree(insert);
			break;
		}
	}
}

static void
decodeEntryUpdateRecord(EntryUpdateRecord *decoded, XLogRecord *record)
{
	char	   *begin = XLogRecGetData(record),
			   *ptr;
	int			i = 0,
				addpath = 0;

	decoded->data = (gistxlogEntryUpdate *) begin;

	if (decoded->data->ntodelete)
	{
		decoded->todelete = (OffsetNumber *) (begin + sizeof(gistxlogEntryUpdate) + addpath);
		addpath = MAXALIGN(sizeof(OffsetNumber) * decoded->data->ntodelete);
	}
	else
		decoded->todelete = NULL;

	decoded->len = 0;
	ptr = begin + sizeof(gistxlogEntryUpdate) + addpath;
	while (ptr - begin < record->xl_len)
	{
		decoded->len++;
		ptr += IndexTupleSize((IndexTuple) ptr);
	}

	decoded->itup = (IndexTuple *) palloc(sizeof(IndexTuple) * decoded->len);

	ptr = begin + sizeof(gistxlogEntryUpdate) + addpath;
	while (ptr - begin < record->xl_len)
	{
		decoded->itup[i] = (IndexTuple) ptr;
		ptr += IndexTupleSize(decoded->itup[i]);
		i++;
	}
}

/*
 * redo any page update (except page split)
 */
static void
gistRedoEntryUpdateRecord(XLogRecPtr lsn, XLogRecord *record, bool isnewroot)
{
	EntryUpdateRecord xlrec;
	Relation	reln;
	Buffer		buffer;
	Page		page;

	decodeEntryUpdateRecord(&xlrec, record);

	reln = XLogOpenRelation(xlrec.data->node);
	if (!RelationIsValid(reln))
		return;
	buffer = XLogReadBuffer(false, reln, xlrec.data->blkno);
	if (!BufferIsValid(buffer))
		elog(PANIC, "block %u unfound", xlrec.data->blkno);
	page = (Page) BufferGetPage(buffer);

	if (isnewroot)
	{
		if (!PageIsNew((PageHeader) page) && XLByteLE(lsn, PageGetLSN(page)))
		{
			LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
			ReleaseBuffer(buffer);
			return;
		}
	}
	else
	{
		if (PageIsNew((PageHeader) page))
			elog(PANIC, "uninitialized page %u", xlrec.data->blkno);
		if (XLByteLE(lsn, PageGetLSN(page)))
		{
			LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
			ReleaseBuffer(buffer);
			return;
		}
	}

	if (xlrec.data->isemptypage)
	{
		while (!PageIsEmpty(page))
			PageIndexTupleDelete(page, FirstOffsetNumber);

		if (xlrec.data->blkno == GIST_ROOT_BLKNO)
			GistPageSetLeaf(page);
		else
			GistPageSetDeleted(page);
	}
	else
	{
		if (isnewroot)
			GISTInitBuffer(buffer, 0);
		else if (xlrec.data->ntodelete)
		{
			int			i;

			for (i = 0; i < xlrec.data->ntodelete; i++)
				PageIndexTupleDelete(page, xlrec.todelete[i]);
			if (GistPageIsLeaf(page))
				GistMarkTuplesDeleted(page);
		}

		/* add tuples */
		if (xlrec.len > 0)
			gistfillbuffer(reln, page, xlrec.itup, xlrec.len, InvalidOffsetNumber);

		/*
		 * special case: leafpage, nothing to insert, nothing to delete, then
		 * vacuum marks page
		 */
		if (GistPageIsLeaf(page) && xlrec.len == 0 && xlrec.data->ntodelete == 0)
			GistClearTuplesDeleted(page);
	}

	PageSetLSN(page, lsn);
	PageSetTLI(page, ThisTimeLineID);
	GistPageGetOpaque(page)->rightlink = InvalidBlockNumber;
	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
	WriteBuffer(buffer);

	if (ItemPointerIsValid(&(xlrec.data->key)))
	{
		if (incomplete_inserts != NIL)
			forgetIncompleteInsert(xlrec.data->node, xlrec.data->key);

		if (!isnewroot && xlrec.data->blkno != GIST_ROOT_BLKNO)
			pushIncompleteInsert(xlrec.data->node, lsn, xlrec.data->key,
								 &(xlrec.data->blkno), 1,
								 NULL);
	}
}

static void
decodePageSplitRecord(PageSplitRecord *decoded, XLogRecord *record)
{
	char	   *begin = XLogRecGetData(record),
			   *ptr;
	int			j,
				i = 0;

	decoded->data = (gistxlogPageSplit *) begin;
	decoded->page = (NewPage *) palloc(sizeof(NewPage) * decoded->data->npage);

	ptr = begin + sizeof(gistxlogPageSplit);
	for (i = 0; i < decoded->data->npage; i++)
	{
		Assert(ptr - begin < record->xl_len);
		decoded->page[i].header = (gistxlogPage *) ptr;
		ptr += sizeof(gistxlogPage);

		decoded->page[i].itup = (IndexTuple *)
			palloc(sizeof(IndexTuple) * decoded->page[i].header->num);
		j = 0;
		while (j < decoded->page[i].header->num)
		{
			Assert(ptr - begin < record->xl_len);
			decoded->page[i].itup[j] = (IndexTuple) ptr;
			ptr += IndexTupleSize((IndexTuple) ptr);
			j++;
		}
	}
}

static void
gistRedoPageSplitRecord(XLogRecPtr lsn, XLogRecord *record)
{
	PageSplitRecord xlrec;
	Relation	reln;
	Buffer		buffer;
	Page		page;
	int			i;
	int			flags = 0;

	decodePageSplitRecord(&xlrec, record);
	reln = XLogOpenRelation(xlrec.data->node);
	if (!RelationIsValid(reln))
		return;

	/* first of all wee need get F_LEAF flag from original page */
	buffer = XLogReadBuffer(false, reln, xlrec.data->origblkno);
	if (!BufferIsValid(buffer))
		elog(PANIC, "block %u unfound", xlrec.data->origblkno);
	page = (Page) BufferGetPage(buffer);
	if (PageIsNew((PageHeader) page))
		elog(PANIC, "uninitialized page %u", xlrec.data->origblkno);

	flags = (GistPageIsLeaf(page)) ? F_LEAF : 0;
	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
	ReleaseBuffer(buffer);

	/* loop around all pages */
	for (i = 0; i < xlrec.data->npage; i++)
	{
		NewPage    *newpage = xlrec.page + i;
		bool		isorigpage = (xlrec.data->origblkno == newpage->header->blkno) ? true : false;

		buffer = XLogReadBuffer(!isorigpage, reln, newpage->header->blkno);
		if (!BufferIsValid(buffer))
			elog(PANIC, "block %u unfound", newpage->header->blkno);
		page = (Page) BufferGetPage(buffer);

		if (XLByteLE(lsn, PageGetLSN(page)))
		{
			LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
			ReleaseBuffer(buffer);
			continue;
		}

		/* ok, clear buffer */
		GISTInitBuffer(buffer, flags);

		/* and fill it */
		gistfillbuffer(reln, page, newpage->itup, newpage->header->num, FirstOffsetNumber);

		PageSetLSN(page, lsn);
		PageSetTLI(page, ThisTimeLineID);
		LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
		WriteBuffer(buffer);
	}

	if (ItemPointerIsValid(&(xlrec.data->key)))
	{
		if (incomplete_inserts != NIL)
			forgetIncompleteInsert(xlrec.data->node, xlrec.data->key);

		pushIncompleteInsert(xlrec.data->node, lsn, xlrec.data->key,
							 NULL, 0,
							 &xlrec);
	}
}

static void
gistRedoCreateIndex(XLogRecPtr lsn, XLogRecord *record)
{
	RelFileNode *node = (RelFileNode *) XLogRecGetData(record);
	Relation	reln;
	Buffer		buffer;
	Page		page;

	reln = XLogOpenRelation(*node);
	if (!RelationIsValid(reln))
		return;
	buffer = XLogReadBuffer(true, reln, GIST_ROOT_BLKNO);
	if (!BufferIsValid(buffer))
		elog(PANIC, "root block unfound");
	page = (Page) BufferGetPage(buffer);

	if (!PageIsNew((PageHeader) page) && XLByteLE(lsn, PageGetLSN(page)))
	{
		LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
		ReleaseBuffer(buffer);
		return;
	}

	GISTInitBuffer(buffer, F_LEAF);

	PageSetLSN(page, lsn);
	PageSetTLI(page, ThisTimeLineID);
	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
	WriteBuffer(buffer);
}

static void
gistRedoCompleteInsert(XLogRecPtr lsn, XLogRecord *record)
{
	char	   *begin = XLogRecGetData(record),
			   *ptr;
	gistxlogInsertComplete *xlrec;

	xlrec = (gistxlogInsertComplete *) begin;

	ptr = begin + sizeof(gistxlogInsertComplete);
	while (ptr - begin < record->xl_len)
	{
		Assert(record->xl_len - (ptr - begin) >= sizeof(ItemPointerData));
		forgetIncompleteInsert(xlrec->node, *((ItemPointerData *) ptr));
		ptr += sizeof(ItemPointerData);
	}
}

void
gist_redo(XLogRecPtr lsn, XLogRecord *record)
{
	uint8		info = record->xl_info & ~XLR_INFO_MASK;

	MemoryContext oldCxt;

	oldCxt = MemoryContextSwitchTo(opCtx);
	switch (info)
	{
		case XLOG_GIST_ENTRY_UPDATE:
		case XLOG_GIST_ENTRY_DELETE:
			gistRedoEntryUpdateRecord(lsn, record, false);
			break;
		case XLOG_GIST_NEW_ROOT:
			gistRedoEntryUpdateRecord(lsn, record, true);
			break;
		case XLOG_GIST_PAGE_SPLIT:
			gistRedoPageSplitRecord(lsn, record);
			break;
		case XLOG_GIST_CREATE_INDEX:
			gistRedoCreateIndex(lsn, record);
			break;
		case XLOG_GIST_INSERT_COMPLETE:
			gistRedoCompleteInsert(lsn, record);
			break;
		default:
			elog(PANIC, "gist_redo: unknown op code %u", info);
	}

	MemoryContextSwitchTo(oldCxt);
	MemoryContextReset(opCtx);
}

static void
out_target(char *buf, RelFileNode node, ItemPointerData key)
{
	sprintf(buf + strlen(buf), "rel %u/%u/%u; tid %u/%u",
			node.spcNode, node.dbNode, node.relNode,
			ItemPointerGetBlockNumber(&key),
			ItemPointerGetOffsetNumber(&key));
}

static void
out_gistxlogEntryUpdate(char *buf, gistxlogEntryUpdate *xlrec)
{
	out_target(buf, xlrec->node, xlrec->key);
	sprintf(buf + strlen(buf), "; block number %u",
			xlrec->blkno);
}

static void
out_gistxlogPageSplit(char *buf, gistxlogPageSplit *xlrec)
{
	strcat(buf, "page_split: ");
	out_target(buf, xlrec->node, xlrec->key);
	sprintf(buf + strlen(buf), "; block number %u splits to %d pages",
			xlrec->origblkno, xlrec->npage);
}

void
gist_desc(char *buf, uint8 xl_info, char *rec)
{
	uint8		info = xl_info & ~XLR_INFO_MASK;

	switch (info)
	{
		case XLOG_GIST_ENTRY_UPDATE: