xfs: implement CoW for directio writes

For O_DIRECT writes to shared blocks, we have to CoW them just like
we would with buffered writes.  For writes that are not block-aligned,
just bounce them to the page cache.

For block-aligned writes, however, we can do better than that.  Use
the same mechanisms that we employ for buffered CoW to set up a
delalloc reservation, allocate all the blocks at once, issue the
writes against the new blocks and use the same ioend functions to
remap the blocks after the write.  This should be fairly performant.

Christoph discovered that xfs_reflink_allocate_cow_range may stumble
over invalid entries in the extent array given that it drops the ilock
but still expects the index to be stable.  Simple fixing it to a new
lookup for every iteration still isn't correct given that
xfs_bmapi_allocate will trigger a BUG_ON() if hitting a hole, and
there is nothing preventing a xfs_bunmapi_cow call removing extents
once we dropped the ilock either.

This patch duplicates the inner loop of xfs_bmapi_allocate into a
helper for xfs_reflink_allocate_cow_range so that it can be done under
the same ilock critical section as our CoW fork delayed allocation.
The directio CoW warts will be revisited in a later patch.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

1 files changed, 83 insertions, 7 deletions

diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 2f4a15ec9ee9..0e2a931fa725 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -40,6 +40,7 @@
 /* flags for direct write completions */
 #define XFS_DIO_FLAG_UNWRITTEN	(1 << 0)
 #define XFS_DIO_FLAG_APPEND	(1 << 1)
+#define XFS_DIO_FLAG_COW	(1 << 2)
 
 /*
  * structure owned by writepages passed to individual writepage calls
@@ -1190,18 +1191,24 @@ xfs_map_direct(
 	struct inode		*inode,
 	struct buffer_head	*bh_result,
 	struct xfs_bmbt_irec	*imap,
-	xfs_off_t		offset)
+	xfs_off_t		offset,
+	bool			is_cow)
 {
 	uintptr_t		*flags = (uintptr_t *)&bh_result->b_private;
 	xfs_off_t		size = bh_result->b_size;
 
 	trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
-		ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap);
+		ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : is_cow ? XFS_IO_COW :
+		XFS_IO_OVERWRITE, imap);
 
 	if (ISUNWRITTEN(imap)) {
 		*flags |= XFS_DIO_FLAG_UNWRITTEN;
 		set_buffer_defer_completion(bh_result);
-	} else if (offset + size > i_size_read(inode) || offset + size < 0) {
+	} else if (is_cow) {
+		*flags |= XFS_DIO_FLAG_COW;
+		set_buffer_defer_completion(bh_result);
+	}
+	if (offset + size > i_size_read(inode) || offset + size < 0) {
 		*flags |= XFS_DIO_FLAG_APPEND;
 		set_buffer_defer_completion(bh_result);
 	}
@@ -1247,6 +1254,44 @@ xfs_map_trim_size(
 	bh_result->b_size = mapping_size;
 }
 
+/* Bounce unaligned directio writes to the page cache. */
+static int
+xfs_bounce_unaligned_dio_write(
+	struct xfs_inode	*ip,
+	xfs_fileoff_t		offset_fsb,
+	struct xfs_bmbt_irec	*imap)
+{
+	struct xfs_bmbt_irec	irec;
+	xfs_fileoff_t		delta;
+	bool			shared;
+	bool			x;
+	int			error;
+
+	irec = *imap;
+	if (offset_fsb > irec.br_startoff) {
+		delta = offset_fsb - irec.br_startoff;
+		irec.br_blockcount -= delta;
+		irec.br_startblock += delta;
+		irec.br_startoff = offset_fsb;
+	}
+	error = xfs_reflink_trim_around_shared(ip, &irec, &shared, &x);
+	if (error)
+		return error;
+
+	/*
+	 * We're here because we're trying to do a directio write to a
+	 * region that isn't aligned to a filesystem block.  If any part
+	 * of the extent is shared, fall back to buffered mode to handle
+	 * the RMW.  This is done by returning -EREMCHG ("remote addr
+	 * changed"), which is caught further up the call stack.
+	 */
+	if (shared) {
+		trace_xfs_reflink_bounce_dio_write(ip, imap);
+		return -EREMCHG;
+	}
+	return 0;
+}
+
 STATIC int
 __xfs_get_blocks(
 	struct inode		*inode,
@@ -1266,6 +1311,8 @@ __xfs_get_blocks(
 	xfs_off_t		offset;
 	ssize_t			size;
 	int			new = 0;
+	bool			is_cow = false;
+	bool			need_alloc = false;
 
 	BUG_ON(create && !direct);
 
@@ -1291,8 +1338,26 @@ __xfs_get_blocks(
 	end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
 	offset_fsb = XFS_B_TO_FSBT(mp, offset);
 
-	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
-				&imap, &nimaps, XFS_BMAPI_ENTIRE);
+	if (create && direct && xfs_is_reflink_inode(ip))
+		is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap,
+					&need_alloc);
+	if (!is_cow) {
+		error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
+					&imap, &nimaps, XFS_BMAPI_ENTIRE);
+		/*
+		 * Truncate an overwrite extent if there's a pending CoW
+		 * reservation before the end of this extent.  This
+		 * forces us to come back to get_blocks to take care of
+		 * the CoW.
+		 */
+		if (create && direct && nimaps &&
+		    imap.br_startblock != HOLESTARTBLOCK &&
+		    imap.br_startblock != DELAYSTARTBLOCK &&
+		    !ISUNWRITTEN(&imap))
+			xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb,
+					&imap);
+	}
+	ASSERT(!need_alloc);
 	if (error)
 		goto out_unlock;
 
@@ -1344,6 +1409,13 @@ __xfs_get_blocks(
 	if (imap.br_startblock != HOLESTARTBLOCK &&
 	    imap.br_startblock != DELAYSTARTBLOCK &&
 	    (create || !ISUNWRITTEN(&imap))) {
+		if (create && direct && !is_cow) {
+			error = xfs_bounce_unaligned_dio_write(ip, offset_fsb,
+					&imap);
+			if (error)
+				return error;
+		}
+
 		xfs_map_buffer(inode, bh_result, &imap, offset);
 		if (ISUNWRITTEN(&imap))
 			set_buffer_unwritten(bh_result);
@@ -1352,7 +1424,8 @@ __xfs_get_blocks(
 			if (dax_fault)
 				ASSERT(!ISUNWRITTEN(&imap));
 			else
-				xfs_map_direct(inode, bh_result, &imap, offset);
+				xfs_map_direct(inode, bh_result, &imap, offset,
+						is_cow);
 		}
 	}
 
@@ -1478,7 +1551,10 @@ xfs_end_io_direct_write(
 		trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
 
 		error = xfs_iomap_write_unwritten(ip, offset, size);
-	} else if (flags & XFS_DIO_FLAG_APPEND) {
+	}
+	if (flags & XFS_DIO_FLAG_COW)
+		error = xfs_reflink_end_cow(ip, offset, size);
+	if (flags & XFS_DIO_FLAG_APPEND) {
 		trace_xfs_end_io_direct_write_append(ip, offset, size);
 
 		error = xfs_setfilesize(ip, offset, size);