xfs_aops.c

来自「linux-2.4.29操作系统的源码」· C语言 代码 · 共 1,300 行 · 第 1/3 页

	ssize_t			size;
	loff_t			offset = (loff_t)iblock << inode->i_blkbits;

	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;
		loff_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)) {
			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_dev = iomap.iomap_target->pbr_kdev;

	/* 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)))) {
		set_buffer_new(bh_result);
	}

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

	return 0;
}

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

STATIC int
linvfs_get_block_direct(
	struct inode		*inode,
	long			iblock,
	struct buffer_head	*bh_result,
	int			create)
{
	return linvfs_get_block_core(inode, iblock, bh_result,
					create, 1, BMAPI_WRITE|BMAPI_DIRECT);
}

STATIC int
linvfs_bmap(
	struct address_space	*mapping,
	long			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_direct);
}

STATIC int
linvfs_readpage(
	struct file		*unused,
	struct page		*page)
{
	return block_read_full_page(page, 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)
{
	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() || PFLAGS_TEST_NOIO()) && 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, inode->i_dev, 1 << inode->i_blkbits);

	/*
	 * Convert delayed allocate, unwritten or unmapped space
	 * to real space and flush out to disk.
	 */
	if (need_trans)
		PFLAGS_SET_NOIO();
	error = xfs_page_state_convert(inode, page, 1, unmapped);
	if (need_trans)
		PFLAGS_CLEAR_NOIO();
	if (error == -EAGAIN)
		goto out_fail;

	if (unlikely(error < 0)) {
		unlock_page(page);
		return error;
	}

	return 0;

out_fail:
	SetPageDirty(page);
	unlock_page(page);
	return 0;
}

/*
 * 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,
	int			gfp_mask)
{
	struct inode		*inode = page->mapping->host;
	int			dirty, delalloc, unmapped, unwritten;

	xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);

	xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
	if (!delalloc && !unwritten)
		return 1;

	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() || PFLAGS_TEST_NOIO())
		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, 0, 0);
	return (dirty == 0 && !unwritten) ? 1 : 0;
}

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);
}

/*
 * Initiate I/O on a kiobuf of user memory
 */
STATIC int
linvfs_direct_IO(
	int			rw,
	struct inode		*inode,
	struct kiobuf		*iobuf,
	unsigned long		blocknr,
	int			blocksize)
{
	struct page		**maplist;
	size_t			page_offset;
	xfs_buf_t		*pb;
	xfs_iomap_t		iomap;
	int			niomap, error = 0;
	int			pb_flags, map_flags, pg_index = 0;
	size_t			length, total;
	loff_t			offset, map_size;
	size_t			size;
	vnode_t			*vp = LINVFS_GET_VP(inode);

	/* Note - although the iomap could have a 64-bit size,
	 * kiobuf->length is only an int, so the min(map_size, length)
	 * test will keep us from overflowing the pagebuf size_t size.
	 */
	total = length = iobuf->length;
	offset = blocknr;
	offset <<= inode->i_blkbits;

	maplist = iobuf->maplist;
	page_offset = iobuf->offset;

	map_flags = (rw ? BMAPI_WRITE : BMAPI_READ) | BMAPI_DIRECT;
	pb_flags = (rw ? PBF_WRITE : PBF_READ) | PBF_FORCEIO | PBF_DIRECTIO;
	while (length) {
		niomap = 1;
		VOP_BMAP(vp, offset, length, map_flags, &iomap, &niomap, error);
		if (unlikely(!error && niomap &&
				(iomap.iomap_flags & IOMAP_DELAY))) {
#ifdef DEBUG
			printk(
	"XFS: %s - direct IO (%lld:%ld) into a delayed allocate extent?\n",
				__FUNCTION__, (long long)offset, (long)length);
			xfs_stack_trace();
#endif
			VOP_BMAP(vp, offset, length, BMAPI_ALLOCATE,
					&iomap, &niomap, error);
		}
		if (error)
			break;

		if (rw == WRITE)
			VMODIFY(vp);

		BUG_ON(niomap != 1);
		BUG_ON(iomap.iomap_flags & IOMAP_DELAY);

		map_size = iomap.iomap_bsize - iomap.iomap_delta;
		size = (size_t)min(map_size, (loff_t)length);

		if ((iomap.iomap_flags & IOMAP_HOLE) ||
		    ((iomap.iomap_flags & IOMAP_UNWRITTEN) && rw == READ)) {
			size_t	zero_len = size;

			if (rw == WRITE)
				break;

			/* Need to zero it all */
			while (zero_len) {
				struct page	*page;
				size_t		pg_len;

				pg_len = min((size_t)
						(PAGE_CACHE_SIZE - page_offset),
						zero_len);

				page = maplist[pg_index];

				memset(kmap(page) + page_offset, 0, pg_len);
				flush_dcache_page(page);
				kunmap(page);

				zero_len -= pg_len;
				if ((pg_len + page_offset) == PAGE_CACHE_SIZE) {
					pg_index++;
					page_offset = 0;
				} else {
					page_offset = (page_offset + pg_len) &
							~PAGE_CACHE_MASK;
				}
			}
		} else {
			int	pg_count;

			pg_count = (size + page_offset + PAGE_CACHE_SIZE - 1)
					>> PAGE_CACHE_SHIFT;
			if ((pb = pagebuf_lookup(iomap.iomap_target, offset,
						size, pb_flags)) == NULL) {
				error = ENOMEM;
				break;
			}
			/* Need to hook up pagebuf to kiobuf pages */
			pb->pb_pages = &maplist[pg_index];
			pb->pb_offset = page_offset;
			pb->pb_page_count = pg_count;
			pb->pb_bn = iomap.iomap_bn + (iomap.iomap_delta >> BBSHIFT);

			error = pagebuf_iostart(pb, pb_flags);
			if (!error && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
				VOP_BMAP(vp, XFS_BUF_OFFSET(pb),
					XFS_BUF_SIZE(pb),
					BMAPI_UNWRITTEN, NULL, NULL, error);
			}

			pagebuf_rele(pb);

			if (error)
				break;

			page_offset = (page_offset + size) & ~PAGE_CACHE_MASK;
			if (page_offset)
				pg_count--;
			pg_index += pg_count;
		}

		offset += size;
		length -= size;
	}

	if (error)
		return -error;
	return (int)(total - length);
}


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