xfs_aops.c

linux-2.6.15.6

					startio, unmapped, tlast);
	}

	return page_dirty;

error:
	for (i = 0; i < cnt; i++) {
		unlock_buffer(bh_arr[i]);
	}

	/*
	 * If it's delalloc and we have nowhere to put it,
	 * throw it away, unless the lower layers told
	 * us to try again.
	 */
	if (err != -EAGAIN) {
		if (!unmapped) {
			block_invalidatepage(page, 0);
		}
		ClearPageUptodate(page);
	}
	return err;
}

STATIC int
__linvfs_get_block(
	struct inode		*inode,
	sector_t		iblock,
	unsigned long		blocks,
	struct buffer_head	*bh_result,
	int			create,
	int			direct,
	bmapi_flags_t		flags)
{
	vnode_t			*vp = LINVFS_GET_VP(inode);
	xfs_iomap_t		iomap;
	xfs_off_t		offset;
	ssize_t			size;
	int			retpbbm = 1;
	int			error;

	offset = (xfs_off_t)iblock << inode->i_blkbits;
	if (blocks)
		size = (ssize_t) min_t(xfs_off_t, LONG_MAX,
					(xfs_off_t)blocks << inode->i_blkbits);
	else
		size = 1 << inode->i_blkbits;

	VOP_BMAP(vp, offset, size,
		create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
	if (error)
		return -error;

	if (retpbbm == 0)
		return 0;

	if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
		xfs_daddr_t	bn;
		xfs_off_t	delta;

		/* For unwritten extents do not report a disk address on
		 * the read case (treat as if we're reading into a hole).
		 */
		if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
			delta = offset - iomap.iomap_offset;
			delta >>= inode->i_blkbits;

			bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
			bn += delta;
			BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
			bh_result->b_blocknr = bn;
			set_buffer_mapped(bh_result);
		}
		if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
			if (direct)
				bh_result->b_private = inode;
			set_buffer_unwritten(bh_result);
			set_buffer_delay(bh_result);
		}
	}

	/* If this is a realtime file, data might be on a new device */
	bh_result->b_bdev = iomap.iomap_target->pbr_bdev;

	/* If we previously allocated a block out beyond eof and
	 * we are now coming back to use it then we will need to
	 * flag it as new even if it has a disk address.
	 */
	if (create &&
	    ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
	     (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
		set_buffer_new(bh_result);

	if (iomap.iomap_flags & IOMAP_DELAY) {
		BUG_ON(direct);
		if (create) {
			set_buffer_uptodate(bh_result);
			set_buffer_mapped(bh_result);
			set_buffer_delay(bh_result);
		}
	}

	if (blocks) {
		ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
		offset = min_t(xfs_off_t,
				iomap.iomap_bsize - iomap.iomap_delta,
				(xfs_off_t)blocks << inode->i_blkbits);
		bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset);
	}

	return 0;
}

int
linvfs_get_block(
	struct inode		*inode,
	sector_t		iblock,
	struct buffer_head	*bh_result,
	int			create)
{
	return __linvfs_get_block(inode, iblock, 0, bh_result,
					create, 0, BMAPI_WRITE);
}

STATIC int
linvfs_get_blocks_direct(
	struct inode		*inode,
	sector_t		iblock,
	unsigned long		max_blocks,
	struct buffer_head	*bh_result,
	int			create)
{
	return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
					create, 1, BMAPI_WRITE|BMAPI_DIRECT);
}

STATIC void
linvfs_end_io_direct(
	struct kiocb	*iocb,
	loff_t		offset,
	ssize_t		size,
	void		*private)
{
	xfs_ioend_t	*ioend = iocb->private;

	/*
	 * Non-NULL private data means we need to issue a transaction to
	 * convert a range from unwritten to written extents.  This needs
	 * to happen from process contect but aio+dio I/O completion
	 * happens from irq context so we need to defer it to a workqueue.
	 * This is not nessecary for synchronous direct I/O, but we do
	 * it anyway to keep the code uniform and simpler.
	 *
	 * The core direct I/O code might be changed to always call the
	 * completion handler in the future, in which case all this can
	 * go away.
	 */
	if (private && size > 0) {
		ioend->io_offset = offset;
		ioend->io_size = size;
		xfs_finish_ioend(ioend);
	} else {
		ASSERT(size >= 0);
		xfs_destroy_ioend(ioend);
	}

	/*
	 * blockdev_direct_IO can return an error even afer the I/O
	 * completion handler was called.  Thus we need to protect
	 * against double-freeing.
	 */
	iocb->private = NULL;
}

STATIC ssize_t
linvfs_direct_IO(
	int			rw,
	struct kiocb		*iocb,
	const struct iovec	*iov,
	loff_t			offset,
	unsigned long		nr_segs)
{
	struct file	*file = iocb->ki_filp;
	struct inode	*inode = file->f_mapping->host;
	vnode_t		*vp = LINVFS_GET_VP(inode);
	xfs_iomap_t	iomap;
	int		maps = 1;
	int		error;
	ssize_t		ret;

	VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
	if (error)
		return -error;

	iocb->private = xfs_alloc_ioend(inode);

	ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
		iomap.iomap_target->pbr_bdev,
		iov, offset, nr_segs,
		linvfs_get_blocks_direct,
		linvfs_end_io_direct);

	if (unlikely(ret <= 0 && iocb->private))
		xfs_destroy_ioend(iocb->private);
	return ret;
}


STATIC sector_t
linvfs_bmap(
	struct address_space	*mapping,
	sector_t		block)
{
	struct inode		*inode = (struct inode *)mapping->host;
	vnode_t			*vp = LINVFS_GET_VP(inode);
	int			error;

	vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);

	VOP_RWLOCK(vp, VRWLOCK_READ);
	VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
	VOP_RWUNLOCK(vp, VRWLOCK_READ);
	return generic_block_bmap(mapping, block, linvfs_get_block);
}

STATIC int
linvfs_readpage(
	struct file		*unused,
	struct page		*page)
{
	return mpage_readpage(page, linvfs_get_block);
}

STATIC int
linvfs_readpages(
	struct file		*unused,
	struct address_space	*mapping,
	struct list_head	*pages,
	unsigned		nr_pages)
{
	return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
}

STATIC void
xfs_count_page_state(
	struct page		*page,
	int			*delalloc,
	int			*unmapped,
	int			*unwritten)
{
	struct buffer_head	*bh, *head;

	*delalloc = *unmapped = *unwritten = 0;

	bh = head = page_buffers(page);
	do {
		if (buffer_uptodate(bh) && !buffer_mapped(bh))
			(*unmapped) = 1;
		else if (buffer_unwritten(bh) && !buffer_delay(bh))
			clear_buffer_unwritten(bh);
		else if (buffer_unwritten(bh))
			(*unwritten) = 1;
		else if (buffer_delay(bh))
			(*delalloc) = 1;
	} while ((bh = bh->b_this_page) != head);
}


/*
 * writepage: Called from one of two places:
 *
 * 1. we are flushing a delalloc buffer head.
 *
 * 2. we are writing out a dirty page. Typically the page dirty
 *    state is cleared before we get here. In this case is it
 *    conceivable we have no buffer heads.
 *
 * For delalloc space on the page we need to allocate space and
 * flush it. For unmapped buffer heads on the page we should
 * allocate space if the page is uptodate. For any other dirty
 * buffer heads on the page we should flush them.
 *
 * If we detect that a transaction would be required to flush
 * the page, we have to check the process flags first, if we
 * are already in a transaction or disk I/O during allocations
 * is off, we need to fail the writepage and redirty the page.
 */

STATIC int
linvfs_writepage(
	struct page		*page,
	struct writeback_control *wbc)
{
	int			error;
	int			need_trans;
	int			delalloc, unmapped, unwritten;
	struct inode		*inode = page->mapping->host;

	xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);

	/*
	 * We need a transaction if:
	 *  1. There are delalloc buffers on the page
	 *  2. The page is uptodate and we have unmapped buffers
	 *  3. The page is uptodate and we have no buffers
	 *  4. There are unwritten buffers on the page
	 */

	if (!page_has_buffers(page)) {
		unmapped = 1;
		need_trans = 1;
	} else {
		xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
		if (!PageUptodate(page))
			unmapped = 0;
		need_trans = delalloc + unmapped + unwritten;
	}

	/*
	 * If we need a transaction and the process flags say
	 * we are already in a transaction, or no IO is allowed
	 * then mark the page dirty again and leave the page
	 * as is.
	 */
	if (PFLAGS_TEST_FSTRANS() && need_trans)
		goto out_fail;

	/*
	 * Delay hooking up buffer heads until we have
	 * made our go/no-go decision.
	 */
	if (!page_has_buffers(page))
		create_empty_buffers(page, 1 << inode->i_blkbits, 0);

	/*
	 * Convert delayed allocate, unwritten or unmapped space
	 * to real space and flush out to disk.
	 */
	error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
	if (error == -EAGAIN)
		goto out_fail;
	if (unlikely(error < 0))
		goto out_unlock;

	return 0;

out_fail:
	redirty_page_for_writepage(wbc, page);
	unlock_page(page);
	return 0;
out_unlock:
	unlock_page(page);
	return error;
}

STATIC int
linvfs_invalidate_page(
	struct page		*page,
	unsigned long		offset)
{
	xfs_page_trace(XFS_INVALIDPAGE_ENTER,
			page->mapping->host, page, offset);
	return block_invalidatepage(page, offset);
}

/*
 * Called to move a page into cleanable state - and from there
 * to be released. Possibly the page is already clean. We always
 * have buffer heads in this call.
 *
 * Returns 0 if the page is ok to release, 1 otherwise.
 *
 * Possible scenarios are:
 *
 * 1. We are being called to release a page which has been written
 *    to via regular I/O. buffer heads will be dirty and possibly
 *    delalloc. If no delalloc buffer heads in this case then we
 *    can just return zero.
 *
 * 2. We are called to release a page which has been written via
 *    mmap, all we need to do is ensure there is no delalloc
 *    state in the buffer heads, if not we can let the caller
 *    free them and we should come back later via writepage.
 */
STATIC int
linvfs_release_page(
	struct page		*page,
	gfp_t			gfp_mask)
{
	struct inode		*inode = page->mapping->host;
	int			dirty, delalloc, unmapped, unwritten;
	struct writeback_control wbc = {
		.sync_mode = WB_SYNC_ALL,
		.nr_to_write = 1,
	};

	xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);

	xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
	if (!delalloc && !unwritten)
		goto free_buffers;

	if (!(gfp_mask & __GFP_FS))
		return 0;

	/* If we are already inside a transaction or the thread cannot
	 * do I/O, we cannot release this page.
	 */
	if (PFLAGS_TEST_FSTRANS())
		return 0;

	/*
	 * Convert delalloc space to real space, do not flush the
	 * data out to disk, that will be done by the caller.
	 * Never need to allocate space here - we will always
	 * come back to writepage in that case.
	 */
	dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
	if (dirty == 0 && !unwritten)
		goto free_buffers;
	return 0;

free_buffers:
	return try_to_free_buffers(page);
}

STATIC int
linvfs_prepare_write(
	struct file		*file,
	struct page		*page,
	unsigned int		from,
	unsigned int		to)
{
	return block_prepare_write(page, from, to, linvfs_get_block);
}

struct address_space_operations linvfs_aops = {
	.readpage		= linvfs_readpage,
	.readpages		= linvfs_readpages,
	.writepage		= linvfs_writepage,
	.sync_page		= block_sync_page,
	.releasepage		= linvfs_release_page,
	.invalidatepage		= linvfs_invalidate_page,
	.prepare_write		= linvfs_prepare_write,
	.commit_write		= generic_commit_write,
	.bmap			= linvfs_bmap,
	.direct_IO		= linvfs_direct_IO,
};