md.c

PostgreSQL7.4.6 for Linux

	int			status;
	long		seekpos;
	int			fd;

	fd = _mdfd_blind_getseg(rnode, blkno);

	if (fd < 0)
		return SM_FAIL;

#ifndef LET_OS_MANAGE_FILESIZE
	seekpos = (long) (BLCKSZ * (blkno % ((BlockNumber) RELSEG_SIZE)));
#ifdef DIAGNOSTIC
	if (seekpos >= BLCKSZ * RELSEG_SIZE)
		elog(FATAL, "seekpos too big");
#endif
#else
	seekpos = (long) (BLCKSZ * (blkno));
#endif

	errno = 0;
	if (lseek(fd, seekpos, SEEK_SET) != seekpos)
	{
		elog(LOG, "lseek(%ld) failed: %m", seekpos);
		close(fd);
		return SM_FAIL;
	}

	status = SM_SUCCESS;

	/* write the block */
	errno = 0;
	if (write(fd, buffer, BLCKSZ) != BLCKSZ)
	{
		/* if write didn't set errno, assume problem is no disk space */
		if (errno == 0)
			errno = ENOSPC;
		elog(LOG, "write() failed: %m");
		status = SM_FAIL;
	}

	if (close(fd) < 0)
	{
		elog(LOG, "close() failed: %m");
		status = SM_FAIL;
	}

	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, ereport's on error.
 */
BlockNumber
mdnblocks(Relation reln)
{
	int			fd;
	MdfdVec    *v;

#ifndef LET_OS_MANAGE_FILESIZE
	BlockNumber nblocks;
	BlockNumber segno;
#endif

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

#ifndef LET_OS_MANAGE_FILESIZE
	segno = 0;

	/*
	 * Skip through any segments that aren't the last one, to avoid
	 * redundant seeks on them.  We have previously verified that these
	 * segments are exactly RELSEG_SIZE long, and it's useless to recheck
	 * that each time. (NOTE: this assumption could only be wrong if
	 * another backend has truncated the relation.	We rely on higher code
	 * levels to handle that scenario by closing and re-opening the md
	 * fd.)
	 */
	while (v->mdfd_chain != (MdfdVec *) NULL)
	{
		segno++;
		v = v->mdfd_chain;
	}

	for (;;)
	{
		nblocks = _mdnblocks(v->mdfd_vfd, BLCKSZ);
		if (nblocks > ((BlockNumber) RELSEG_SIZE))
			elog(FATAL, "segment too big");
		if (nblocks < ((BlockNumber) RELSEG_SIZE))
			return (segno * ((BlockNumber) RELSEG_SIZE)) + nblocks;

		/*
		 * If segment is exactly RELSEG_SIZE, advance to next one.
		 */
		segno++;

		if (v->mdfd_chain == (MdfdVec *) NULL)
		{
			/*
			 * Because we pass O_CREAT, we will create the next segment
			 * (with zero length) immediately, if the last segment is of
			 * length REL_SEGSIZE.	This is unnecessary but harmless, and
			 * testing for the case would take more cycles than it seems
			 * worth.
			 */
			v->mdfd_chain = _mdfd_openseg(reln, segno, O_CREAT);
			if (v->mdfd_chain == (MdfdVec *) NULL)
				elog(ERROR, "could not count blocks of \"%s\": %m",
					 RelationGetRelationName(reln));
		}

		v = v->mdfd_chain;
	}
#else
	return _mdnblocks(v->mdfd_vfd, BLCKSZ);
#endif
}

/*
 *	mdtruncate() -- Truncate relation to specified number of blocks.
 *
 *		Returns # of blocks or InvalidBlockNumber on error.
 */
BlockNumber
mdtruncate(Relation reln, BlockNumber nblocks)
{
	int			fd;
	MdfdVec    *v;
	BlockNumber curnblk;

#ifndef LET_OS_MANAGE_FILESIZE
	BlockNumber 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 > curnblk)
		return InvalidBlockNumber;		/* bogus request */
	if (nblocks == curnblk)
		return nblocks;			/* no work */

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

#ifndef LET_OS_MANAGE_FILESIZE
	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);
			FileUnlink(v->mdfd_vfd);
			v = v->mdfd_chain;
			Assert(ov != &Md_fdvec[fd]);		/* we never drop the 1st
												 * segment */
			pfree(ov);
		}
		else if (priorblocks + ((BlockNumber) 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.
			 */
			BlockNumber lastsegblocks = nblocks - priorblocks;

			if (FileTruncate(v->mdfd_vfd, lastsegblocks * BLCKSZ) < 0)
				return InvalidBlockNumber;
			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;
	}
#else
	if (FileTruncate(v->mdfd_vfd, nblocks * BLCKSZ) < 0)
		return InvalidBlockNumber;
#endif

	return nblocks;
}

/*
 *	mdcommit() -- Commit a transaction.
 *
 *		Returns SM_SUCCESS or SM_FAIL with errno set as appropriate.
 */
int
mdcommit(void)
{
	/*
	 * We don't actually have to do anything here...
	 */
	return SM_SUCCESS;
}

/*
 *	mdabort() -- Abort a transaction.
 *
 *		Changes need not be forced to disk at transaction abort.
 */
int
mdabort(void)
{
	/*
	 * We don't actually have to do anything here...
	 */
	return SM_SUCCESS;
}

/*
 *	mdsync() -- Sync previous writes to stable storage.
 */
int
mdsync(void)
{
	sync();
	if (IsUnderPostmaster)
		sleep(2);
	sync();
	return SM_SUCCESS;
}

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

	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;

	nvec = (MdfdVec *) MemoryContextAlloc(MdCxt, Nfds * sizeof(MdfdVec));
	MemSet(nvec, 0, Nfds * sizeof(MdfdVec));
	memcpy(nvec, (char *) Md_fdvec, CurFd * sizeof(MdfdVec));
	pfree(Md_fdvec);

	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, BlockNumber segno, int oflags)
{
	MdfdVec    *v;
	int			fd;
	char	   *path,
			   *fullpath;

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

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

	/* open the file */
	fd = FileNameOpenFile(fullpath, O_RDWR | PG_BINARY | oflags, 0600);

	pfree(fullpath);

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

	/* allocate an mdfdvec entry for it */
	v = (MdfdVec *) MemoryContextAlloc(MdCxt, sizeof(MdfdVec));

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

#ifdef DIAGNOSTIC
	if (_mdnblocks(fd, BLCKSZ) > ((BlockNumber) RELSEG_SIZE))
		elog(FATAL, "segment too big");
#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, "could not open relation \"%s\": %m",
				 RelationGetRelationName(reln));
		reln->rd_fd = fd;
	}
	return fd;
}

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

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

#ifndef LET_OS_MANAGE_FILESIZE
	BlockNumber segno;
	BlockNumber i;
#endif

	fd = _mdfd_getrelnfd(reln);

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

		if (v->mdfd_chain == (MdfdVec *) NULL)
		{
			/*
			 * We will create the next segment only if the target block is
			 * within it.  This prevents Sorcerer's Apprentice syndrome if
			 * a bug at higher levels causes us to be handed a
			 * ridiculously large blkno --- otherwise we could create many
			 * thousands of empty segment files before reaching the
			 * "target" block.	We should never need to create more than
			 * one new segment per call, so this restriction seems
			 * reasonable.
			 */
			v->mdfd_chain = _mdfd_openseg(reln, i, (segno == 1) ? O_CREAT : 0);

			if (v->mdfd_chain == (MdfdVec *) NULL)
				elog(ERROR, "could not open segment %u of relation \"%s\" (target block %u): %m",
					 i, RelationGetRelationName(reln), blkno);
		}
		v = v->mdfd_chain;
	}
#else
	v = &Md_fdvec[fd];
#endif

	return v;
}

/*
 * Find the segment of the relation holding the specified block.
 *
 * This performs the same work as _mdfd_getseg() except that we must work
 * "blind" with no Relation struct.  We assume that we are not likely to
 * touch the same relation again soon, so we do not create an FD entry for
 * the relation --- we just open a kernel file descriptor which will be
 * used and promptly closed.  We also assume that the target block already
 * exists, ie, we need not extend the relation.
 *
 * The return value is the kernel descriptor, or -1 on failure.
 */

static int
_mdfd_blind_getseg(RelFileNode rnode, BlockNumber blkno)
{
	char	   *path;
	int			fd;

#ifndef LET_OS_MANAGE_FILESIZE
	BlockNumber segno;
#endif

	path = relpath(rnode);

#ifndef LET_OS_MANAGE_FILESIZE
	/* append the '.segno', if needed */
	segno = blkno / ((BlockNumber) RELSEG_SIZE);
	if (segno > 0)
	{
		char	   *segpath = (char *) palloc(strlen(path) + 12);

		sprintf(segpath, "%s.%u", path, segno);
		pfree(path);
		path = segpath;
	}
#endif

	/* call fd.c to allow other FDs to be closed if needed */
	fd = BasicOpenFile(path, O_RDWR | PG_BINARY, 0600);
	if (fd < 0)
		elog(LOG, "could not open \"%s\": %m", path);

	pfree(path);

	return fd;
}

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

	len = FileSeek(file, 0L, SEEK_END);
	if (len < 0)
		return 0;				/* on failure, assume file is empty */
	return (BlockNumber) (len / blcksz);
}