md.c

关系型数据库 Postgresql 6.5.2

	int			status;
	char	   *path;

#ifndef LET_OS_MANAGE_FILESIZE
	int			nchars;

	/* be sure we have enough space for the '.segno', if any */
	segno = blkno / RELSEG_SIZE;
	if (segno > 0)
		nchars = 10;
	else
		nchars = 0;

	/* construct the path to the file and open it */
	/* system table? then put in system area... */
	if (dbid == (Oid) 0)
	{
		path = (char *) palloc(strlen(DataDir) + sizeof(NameData) + 2 + nchars);
		if (segno == 0)
			sprintf(path, "%s/%s", DataDir, relstr);
		else
			sprintf(path, "%s/%s.%d", DataDir, relstr, segno);
	}
	/* user table? then put in user database area... */
	else if (dbid == MyDatabaseId)
	{
		extern char *DatabasePath;

		path = (char *) palloc(strlen(DatabasePath) + 2 * sizeof(NameData) + 2 + nchars);
		if (segno == 0)
			sprintf(path, "%s%c%s", DatabasePath, SEP_CHAR, relstr);
		else
			sprintf(path, "%s%c%s.%d", DatabasePath, SEP_CHAR, relstr, segno);
	}
	else
/* this is work arround only !!! */
	{
		char		dbpath[MAXPGPATH + 1];
		int4		owner;
		Oid			id;
		char	   *tmpPath;
		int			tmpEncoding;

		GetRawDatabaseInfo(dbstr, &owner, &id, dbpath, &tmpEncoding);

		if (id != dbid)
			elog(FATAL, "mdblindwrt: oid of db %s is not %u", dbstr, dbid);
		tmpPath = ExpandDatabasePath(dbpath);
		if (tmpPath == NULL)
			elog(FATAL, "mdblindwrt: can't expand path for db %s", dbstr);
		path = (char *) palloc(strlen(tmpPath) + 2 * sizeof(NameData) + 2 + nchars);
		if (segno == 0)
			sprintf(path, "%s%c%s", tmpPath, SEP_CHAR, relstr);
		else
			sprintf(path, "%s%c%s.%d", tmpPath, SEP_CHAR, relstr, segno);
		pfree(tmpPath);
	}
#else
	/* construct the path to the file and open it */
	/* system table? then put in system area... */
	if (dbid == (Oid) 0)
	{
		path = (char *) palloc(strlen(DataDir) + sizeof(NameData) + 2);
		sprintf(path, "%s/%s", DataDir, relstr);
	}
	/* user table? then put in user database area... */
	else if (dbid == MyDatabaseId)
	{
		extern char *DatabasePath;

		path = (char *) palloc(strlen(DatabasePath) + 2 * sizeof(NameData) + 2);
		sprintf(path, "%s%c%s", DatabasePath, SEP_CHAR, relstr);
	}
	else
/* this is work arround only !!! */
	{
		char		dbpath[MAXPGPATH + 1];
		int4		owner;
		Oid			id;
		char	   *tmpPath;
		int			tmpEncoding;

		GetRawDatabaseInfo(dbstr, &owner, &id, dbpath, &tmpEncoding);

		if (id != dbid)
			elog(FATAL, "mdblindwrt: oid of db %s is not %u", dbstr, dbid);
		tmpPath = ExpandDatabasePath(dbpath);
		if (tmpPath == NULL)
			elog(FATAL, "mdblindwrt: can't expand path for db %s", dbstr);
		path = (char *) palloc(strlen(tmpPath) + 2 * sizeof(NameData) + 2);
		sprintf(path, "%s%c%s", tmpPath, SEP_CHAR, relstr);
		pfree(tmpPath);
	}
#endif

#ifndef __CYGWIN32__
	if ((fd = open(path, O_RDWR, 0600)) < 0)
#else
	if ((fd = open(path, O_RDWR | O_BINARY, 0600)) < 0)
#endif
		return SM_FAIL;

	/* seek to the right spot */
#ifndef LET_OS_MANAGE_FILESIZE
	seekpos = (long) (BLCKSZ * (blkno % RELSEG_SIZE));
#else
	seekpos = (long) (BLCKSZ * (blkno));
#endif

	if (lseek(fd, seekpos, SEEK_SET) != seekpos)
	{
		close(fd);
		return SM_FAIL;
	}

	status = SM_SUCCESS;

	/* write and sync the block */
	if (write(fd, buffer, BLCKSZ) != BLCKSZ || (pg_fsync(fd) < 0))
		status = SM_FAIL;

	if (close(fd) < 0)
		status = SM_FAIL;

	pfree(path);

	return status;
}

/*
 *	mdnblocks() -- Get the number of blocks stored in a relation.
 *
 *		Important side effect: all segments of the relation are opened
 *		and added to the mdfd_chain list.  If this routine has not been
 *		called, then only segments up to the last one actually touched
 *		are present in the chain...
 *
 *		Returns # of blocks, elog's on error.
 */
int
mdnblocks(Relation reln)
{
	int			fd;
	MdfdVec    *v;
#ifndef LET_OS_MANAGE_FILESIZE
	int			nblocks;
	int			segno;
#endif

	fd = _mdfd_getrelnfd(reln);
	v = &Md_fdvec[fd];

#ifndef LET_OS_MANAGE_FILESIZE
	segno = 0;
	for (;;)
	{
		nblocks = _mdnblocks(v->mdfd_vfd, BLCKSZ);
		if (nblocks > RELSEG_SIZE)
			elog(FATAL, "segment too big in mdnblocks!");
		v->mdfd_lstbcnt = nblocks;
		if (nblocks == RELSEG_SIZE)
		{
			segno++;

			if (v->mdfd_chain == (MdfdVec *) NULL)
			{
				v->mdfd_chain = _mdfd_openseg(reln, segno, O_CREAT);
				if (v->mdfd_chain == (MdfdVec *) NULL)
					elog(ERROR, "cannot count blocks for %s -- open failed",
						 RelationGetRelationName(reln));
			}

			v = v->mdfd_chain;
		}
		else
			return (segno * RELSEG_SIZE) + nblocks;
	}
#else
	return _mdnblocks(v->mdfd_vfd, BLCKSZ);
#endif
}

/*
 *	mdtruncate() -- Truncate relation to specified number of blocks.
 *
 *		Returns # of blocks or -1 on error.
 */
int
mdtruncate(Relation reln, int nblocks)
{
	int			curnblk;
	int			fd;
	MdfdVec    *v;
#ifndef LET_OS_MANAGE_FILESIZE
	MemoryContext oldcxt;
	int			priorblocks;
#endif

	/* NOTE: mdnblocks makes sure we have opened all existing segments,
	 * so that truncate/delete loop will get them all!
	 */
	curnblk = mdnblocks(reln);
	if (nblocks < 0 || nblocks > curnblk)
		return -1;				/* bogus request */
	if (nblocks == curnblk)
		return nblocks;			/* no work */

	fd = _mdfd_getrelnfd(reln);
	v = &Md_fdvec[fd];

#ifndef LET_OS_MANAGE_FILESIZE
	oldcxt = MemoryContextSwitchTo(MdCxt);
	priorblocks = 0;
	while (v != (MdfdVec *) NULL)
	{
		MdfdVec    *ov = v;

		if (priorblocks > nblocks)
		{
			/* This segment is no longer wanted at all (and has already been
			 * unlinked from the mdfd_chain).
			 * We truncate the file before deleting it because if other
			 * backends are holding the file open, the unlink will fail on
			 * some platforms.  Better a zero-size file gets left around than
			 * a big file...
			 */
			FileTruncate(v->mdfd_vfd, 0);
			/* In 6.5, it is not safe to unlink apparently-unused segments,
			 * because another backend could store tuples in one of those
			 * segments before it notices the shared-cache-invalidation
			 * message that would warn it to re-open the file.  So, don't
			 * unlink 'em, just truncate 'em.  This is fixed properly for 6.6
			 * but back-patching the changes was judged too risky.
			 */
#if 0
			FileUnlink(v->mdfd_vfd);
#endif
			v = v->mdfd_chain;
			Assert(ov != &Md_fdvec[fd]); /* we never drop the 1st segment */
			pfree(ov);
		}
		else if (priorblocks + RELSEG_SIZE > nblocks)
		{
			/* This is the last segment we want to keep.
			 * Truncate the file to the right length, and clear chain link
			 * that points to any remaining segments (which we shall zap).
			 * NOTE: if nblocks is exactly a multiple K of RELSEG_SIZE,
			 * we will truncate the K+1st segment to 0 length but keep it.
			 * This is mainly so that the right thing happens if nblocks=0.
			 */
			int lastsegblocks = nblocks - priorblocks;
			if (FileTruncate(v->mdfd_vfd, lastsegblocks * BLCKSZ) < 0)
				return -1;
			v->mdfd_lstbcnt = lastsegblocks;
			v = v->mdfd_chain;
			ov->mdfd_chain = (MdfdVec *) NULL;
		}
		else
		{
			/* We still need this segment and 0 or more blocks beyond it,
			 * so nothing to do here.
			 */
			v = v->mdfd_chain;
		}
		priorblocks += RELSEG_SIZE;
	}
	MemoryContextSwitchTo(oldcxt);
#else
	if (FileTruncate(v->mdfd_vfd, nblocks * BLCKSZ) < 0)
		return -1;
	v->mdfd_lstbcnt = nblocks;
#endif

	return nblocks;

}	/* mdtruncate */

/*
 *	mdcommit() -- Commit a transaction.
 *
 *		All changes to magnetic disk relations must be forced to stable
 *		storage.  This routine makes a pass over the private table of
 *		file descriptors.  Any descriptors to which we have done writes,
 *		but not synced, are synced here.
 *
 *		Returns SM_SUCCESS or SM_FAIL with errno set as appropriate.
 */
int
mdcommit()
{
	int			i;
	MdfdVec    *v;

	for (i = 0; i < CurFd; i++)
	{
#ifndef LET_OS_MANAGE_FILESIZE
		for (v = &Md_fdvec[i]; v != (MdfdVec *) NULL; v = v->mdfd_chain)
#else
		v = &Md_fdvec[i];
		if (v != (MdfdVec *) NULL)
#endif
		{
			if (v->mdfd_flags & MDFD_DIRTY)
			{
				if (FileSync(v->mdfd_vfd) < 0)
					return SM_FAIL;

				v->mdfd_flags &= ~MDFD_DIRTY;
			}
		}
	}

	return SM_SUCCESS;
}

/*
 *	mdabort() -- Abort a transaction.
 *
 *		Changes need not be forced to disk at transaction abort.  We mark
 *		all file descriptors as clean here.  Always returns SM_SUCCESS.
 */
int
mdabort()
{
	int			i;
	MdfdVec    *v;

	for (i = 0; i < CurFd; i++)
	{
#ifndef LET_OS_MANAGE_FILESIZE
		for (v = &Md_fdvec[i]; v != (MdfdVec *) NULL; v = v->mdfd_chain)
			v->mdfd_flags &= ~MDFD_DIRTY;
#else
		v = &Md_fdvec[i];
		v->mdfd_flags &= ~MDFD_DIRTY;
#endif
	}

	return SM_SUCCESS;
}

/*
 *	_fdvec_alloc () -- grab a free (or new) md file descriptor vector.
 *
 */
static
int
_fdvec_alloc()
{
	MdfdVec    *nvec;
	int			fdvec,
				i;
	MemoryContext oldcxt;

	if (Md_Free >= 0)			/* get from free list */
	{
		fdvec = Md_Free;
		Md_Free = Md_fdvec[fdvec].mdfd_nextFree;
		Assert(Md_fdvec[fdvec].mdfd_flags == MDFD_FREE);
		Md_fdvec[fdvec].mdfd_flags = 0;
		if (fdvec >= CurFd)
		{
			Assert(fdvec == CurFd);
			CurFd++;
		}
		return fdvec;
	}

	/* Must allocate more room */

	if (Nfds != CurFd)
		elog(FATAL, "_fdvec_alloc error");

	Nfds *= 2;

	oldcxt = MemoryContextSwitchTo(MdCxt);

	nvec = (MdfdVec *) palloc(Nfds * sizeof(MdfdVec));
	MemSet(nvec, 0, Nfds * sizeof(MdfdVec));
	memmove(nvec, (char *) Md_fdvec, CurFd * sizeof(MdfdVec));
	pfree(Md_fdvec);

	MemoryContextSwitchTo(oldcxt);

	Md_fdvec = nvec;

	/* Set new free list */
	for (i = CurFd; i < Nfds; i++)
	{
		Md_fdvec[i].mdfd_nextFree = i + 1;
		Md_fdvec[i].mdfd_flags = MDFD_FREE;
	}
	Md_fdvec[Nfds - 1].mdfd_nextFree = -1;
	Md_Free = CurFd + 1;

	fdvec = CurFd;
	CurFd++;
	Md_fdvec[fdvec].mdfd_flags = 0;

	return fdvec;
}

/*
 *	_fdvec_free () -- free md file descriptor vector.
 *
 */
static
void
_fdvec_free(int fdvec)
{

	Assert(Md_Free < 0 || Md_fdvec[Md_Free].mdfd_flags == MDFD_FREE);
	Assert(Md_fdvec[fdvec].mdfd_flags != MDFD_FREE);
	Md_fdvec[fdvec].mdfd_nextFree = Md_Free;
	Md_fdvec[fdvec].mdfd_flags = MDFD_FREE;
	Md_Free = fdvec;

}

static MdfdVec *
_mdfd_openseg(Relation reln, int segno, int oflags)
{
	MemoryContext oldcxt;
	MdfdVec    *v;
	int			fd;
	bool		dofree;
	char	   *path,
			   *fullpath;

	/* be sure we have enough space for the '.segno', if any */
	path = relpath(RelationGetRelationName(reln)->data);

	dofree = false;
	if (segno > 0)
	{
		dofree = true;
		fullpath = (char *) palloc(strlen(path) + 12);
		sprintf(fullpath, "%s.%d", path, segno);
	}
	else
		fullpath = path;

	/* open the file */
#ifndef __CYGWIN32__
	fd = FileNameOpenFile(fullpath, O_RDWR | oflags, 0600);
#else
	fd = FileNameOpenFile(fullpath, O_RDWR | O_BINARY | oflags, 0600);
#endif

	if (dofree)
		pfree(fullpath);

	if (fd < 0)
		return (MdfdVec *) NULL;

	/* allocate an mdfdvec entry for it */
	oldcxt = MemoryContextSwitchTo(MdCxt);
	v = (MdfdVec *) palloc(sizeof(MdfdVec));
	MemoryContextSwitchTo(oldcxt);

	/* fill the entry */
	v->mdfd_vfd = fd;
	v->mdfd_flags = (uint16) 0;
	v->mdfd_lstbcnt = _mdnblocks(fd, BLCKSZ);
#ifndef LET_OS_MANAGE_FILESIZE
	v->mdfd_chain = (MdfdVec *) NULL;

#ifdef DIAGNOSTIC
	if (v->mdfd_lstbcnt > RELSEG_SIZE)
		elog(FATAL, "segment too big on open!");
#endif
#endif

	/* all done */
	return v;
}

/* Get the fd for the relation, opening it if it's not already open */

static int
_mdfd_getrelnfd(Relation reln)
{
	int			fd;

	fd = RelationGetFile(reln);
	if (fd < 0)
	{
		if ((fd = mdopen(reln)) < 0)
			elog(ERROR, "cannot open relation %s",
				 RelationGetRelationName(reln));
		reln->rd_fd = fd;
	}
	return fd;
}

/* Find the segment of the relation holding the specified block */

static MdfdVec *
_mdfd_getseg(Relation reln, int blkno)
{
	MdfdVec    *v;
	int			segno;
	int			fd;
	int			i;

	fd = _mdfd_getrelnfd(reln);

#ifndef LET_OS_MANAGE_FILESIZE
	for (v = &Md_fdvec[fd], segno = blkno / RELSEG_SIZE, i = 1;
		 segno > 0;
		 i++, segno--)
	{

		if (v->mdfd_chain == (MdfdVec *) NULL)
		{
			v->mdfd_chain = _mdfd_openseg(reln, i, O_CREAT);

			if (v->mdfd_chain == (MdfdVec *) NULL)
				elog(ERROR, "cannot open segment %d of relation %s",
					 i, RelationGetRelationName(reln));
		}
		v = v->mdfd_chain;
	}
#else
	v = &Md_fdvec[fd];
#endif

	return v;
}

static BlockNumber
_mdnblocks(File file, Size blcksz)
{
	long		len;

	len = FileSeek(file, 0L, SEEK_END) - 1;
	return (BlockNumber) ((len < 0) ? 0 : 1 + len / blcksz);
}