1 files changed, 38 insertions, 24 deletions

diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index 207e0dadffc3..40289fe6f5b2 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -573,17 +573,43 @@ xfs_stat_blksize(
 }
 
 static void
-xfs_get_atomic_write_attr(
+xfs_report_dioalign(
 	struct xfs_inode	*ip,
-	unsigned int		*unit_min,
-	unsigned int		*unit_max)
+	struct kstat		*stat)
 {
-	if (!xfs_inode_can_atomicwrite(ip)) {
-		*unit_min = *unit_max = 0;
-		return;
-	}
+	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
+	struct block_device	*bdev = target->bt_bdev;
+
+	stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN;
+	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
 
-	*unit_min = *unit_max = ip->i_mount->m_sb.sb_blocksize;
+	/*
+	 * For COW inodes, we can only perform out of place writes of entire
+	 * allocation units (blocks or RT extents).
+	 * For writes smaller than the allocation unit, we must fall back to
+	 * buffered I/O to perform read-modify-write cycles.  At best this is
+	 * highly inefficient; at worst it leads to page cache invalidation
+	 * races.  Tell applications to avoid this by reporting the larger write
+	 * alignment in dio_offset_align, and the smaller read alignment in
+	 * dio_read_offset_align.
+	 */
+	stat->dio_read_offset_align = bdev_logical_block_size(bdev);
+	if (xfs_is_cow_inode(ip))
+		stat->dio_offset_align = xfs_inode_alloc_unitsize(ip);
+	else
+		stat->dio_offset_align = stat->dio_read_offset_align;
+}
+
+static void
+xfs_report_atomic_write(
+	struct xfs_inode	*ip,
+	struct kstat		*stat)
+{
+	unsigned int		unit_min = 0, unit_max = 0;
+
+	if (xfs_inode_can_atomicwrite(ip))
+		unit_min = unit_max = ip->i_mount->m_sb.sb_blocksize;
+	generic_fill_statx_atomic_writes(stat, unit_min, unit_max);
 }
 
 STATIC int
@@ -647,22 +673,10 @@ xfs_vn_getattr(
 		stat->rdev = inode->i_rdev;
 		break;
 	case S_IFREG:
-		if (request_mask & STATX_DIOALIGN) {
-			struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
-			struct block_device	*bdev = target->bt_bdev;
-
-			stat->result_mask |= STATX_DIOALIGN;
-			stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
-			stat->dio_offset_align = bdev_logical_block_size(bdev);
-		}
-		if (request_mask & STATX_WRITE_ATOMIC) {
-			unsigned int	unit_min, unit_max;
-
-			xfs_get_atomic_write_attr(ip, &unit_min,
-					&unit_max);
-			generic_fill_statx_atomic_writes(stat,
-					unit_min, unit_max);
-		}
+		if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN))
+			xfs_report_dioalign(ip, stat);
+		if (request_mask & STATX_WRITE_ATOMIC)
+			xfs_report_atomic_write(ip, stat);
 		fallthrough;
 	default:
 		stat->blksize = xfs_stat_blksize(ip);