xfs: allow reflink on the rt volume when extent size is larger than 1 rt blockrealtime-reflink-extsize_2021-09-17

Make the necessary tweaks to the reflink remapping code to support
remapping on the realtime volume when the rt extent size is larger than
a single rt block.  We need to check that the remap arguments from
userspace are aligned to a rt extent boundary, and that the length
is always aligned, even if the kernel tried to round it up to EOF for
us.  XFS can only map and remap full rt extents, so we have to be a
little more strict about the alignment there.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>

4 files changed, 81 insertions, 9 deletions

diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
index b93faa819894..0c73aa441c47 100644
--- a/fs/xfs/xfs_reflink.c
+++ b/fs/xfs/xfs_reflink.c
@@ -1348,6 +1348,13 @@ xfs_reflink_remap_blocks(
 	len = min_t(xfs_filblks_t, XFS_B_TO_FSB(mp, remap_len),
 			XFS_MAX_FILEOFF);
 
+	/*
+	 * Make sure the end is aligned with a rt extent (if desired), since
+	 * the end of the range could be EOF.
+	 */
+	if (xfs_inode_has_bigrtextents(dest))
+		len = roundup_64(len, mp->m_sb.sb_rextsize);
+
 	trace_xfs_reflink_remap_blocks(src, srcoff, len, dest, destoff);
 
 	while (len > 0) {
@@ -1421,6 +1428,50 @@ xfs_reflink_zero_posteof(
 			&xfs_buffered_write_iomap_ops);
 }
 
+/* Adjust the length of the remap operation to end on a rt extent boundary. */
+STATIC int
+xfs_reflink_remap_adjust_rtlen(
+	struct xfs_inode	*src,
+	loff_t			pos_in,
+	struct xfs_inode	*dest,
+	loff_t			pos_out,
+	loff_t			*len,
+	unsigned int		remap_flags)
+{
+	struct xfs_mount	*mp = src->i_mount;
+	uint32_t		mod;
+
+	div_u64_rem(*len, XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize), &mod);
+
+	/*
+	 * We previously checked the rtextent alignment of both offsets, so we
+	 * now have to check the alignment of the length.  The VFS remap prep
+	 * function can change the length on us, so we can only make length
+	 * adjustments after that.  If the length is aligned to an rtextent,
+	 * we're trivially good to go.
+	 *
+	 * Otherwise, the length is not aligned to an rt extent.  If the source
+	 * file's range ends at EOF, the VFS ensured that the dest file's range
+	 * also ends at EOF.  The actual remap function will round the (byte)
+	 * length up to the nearest rtextent unit, so we're ok here too.
+	 */
+	if (mod == 0 || pos_in + *len == i_size_read(VFS_I(src)))
+		return 0;
+
+	/*
+	 * Otherwise, the only thing we can do is round the request length down
+	 * to an rt extent boundary.  If the caller doesn't allow that, we are
+	 * finished.
+	 */
+	if (!(remap_flags & REMAP_FILE_CAN_SHORTEN))
+		return -EINVAL;
+
+	/* Back off by a single extent. */
+	(*len) -= mod;
+	trace_xfs_reflink_remap_adjust_rtlen(src, pos_in, *len, dest, pos_out);
+	return 0;
+}
+
 /*
  * Prepare two files for range cloning.  Upon a successful return both inodes
  * will have the iolock and mmaplock held, the page cache of the out file will
@@ -1480,11 +1531,22 @@ xfs_reflink_remap_prep(
 	if (IS_DAX(inode_in) || IS_DAX(inode_out))
 		goto out_unlock;
 
-	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
-			len, remap_flags);
+	ASSERT(is_power_of_2(xfs_inode_alloc_unitsize(dest)));
+
+	ret = __generic_remap_file_range_prep(file_in, pos_in, file_out,
+			pos_out, len, remap_flags,
+			xfs_inode_alloc_unitsize(dest));
 	if (ret || *len == 0)
 		goto out_unlock;
 
+	/* Make sure the end is aligned with a rt extent. */
+	if (xfs_inode_has_bigrtextents(src)) {
+		ret = xfs_reflink_remap_adjust_rtlen(src, pos_in, dest,
+				pos_out, len, remap_flags);
+		if (ret || *len == 0)
+			goto out_unlock;
+	}
+
 	/* Attach dquots to dest inode before changing block map */
 	ret = xfs_qm_dqattach(dest);
 	if (ret)
diff --git a/fs/xfs/xfs_rtalloc.c b/fs/xfs/xfs_rtalloc.c
index 15f5a405d7d4..0a1114df236f 100644
--- a/fs/xfs/xfs_rtalloc.c
+++ b/fs/xfs/xfs_rtalloc.c
@@ -1039,7 +1039,7 @@ xfs_growfs_rt(
 	if (!xfs_has_metadir(mp) && (xfs_has_rmapbt(mp) || xfs_has_reflink(mp)))
 		return -EOPNOTSUPP;
 
-	if (xfs_has_reflink(mp) && in->extsize != 1)
+	if (xfs_has_reflink(mp) && !is_power_of_2(mp->m_sb.sb_rextsize))
 		return -EOPNOTSUPP;
 
 	nrblocks = in->newblocks;
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index a70862935988..1c924fe18f3e 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -1682,13 +1682,23 @@ xfs_fs_fill_super(
 
 	if (xfs_has_reflink(mp)) {
 		/*
-		 * Reflink doesn't support rt extent sizes larger than a single
-		 * block because we would have to perform unshare-around for
-		 * rtext-unaligned write requests.
+		 * Reflink doesn't support pagecache pages that span multiple
+		 * realtime extents because iomap doesn't track subpage dirty
+		 * state.  This means that we cannot dirty all the pages
+		 * backing an rt extent without dirtying the adjoining rt
+		 * extents.  If those rt extents are shared and extend into
+		 * other pages, this leads to crazy write amplification.  The
+		 * VFS remap_range checks assume power-of-two block sizes, so
+		 * we don't support that either.
+		 *
+		 * Hence we only support rt extent sizes that are an integer
+		 * power of two because we know those will align with the page
+		 * size.
 		 */
-		if (xfs_has_realtime(mp) && mp->m_sb.sb_rextsize != 1) {
+		if (xfs_has_realtime(mp) &&
+		    !is_power_of_2(mp->m_sb.sb_rextsize)) {
 			xfs_alert(mp,
-	"reflink not compatible with realtime extent size %u!",
+	"reflink not compatible with non-power-of-2 realtime extent size %u!",
 					mp->m_sb.sb_rextsize);
 			error = -EINVAL;
 			goto out_filestream_unmount;
@@ -1707,7 +1717,6 @@ xfs_fs_fill_super(
 		}
 	}
 
-
 	error = xfs_mountfs(mp);
 	if (error)
 		goto out_filestream_unmount;
diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
index e3d4111ac387..de536012a04d 100644
--- a/fs/xfs/xfs_trace.h
+++ b/fs/xfs/xfs_trace.h
@@ -3616,6 +3616,7 @@ TRACE_EVENT(xfs_reflink_remap_blocks,
 		  __entry->dest_lblk)
 );
 DEFINE_DOUBLE_IO_EVENT(xfs_reflink_remap_range);
+DEFINE_DOUBLE_IO_EVENT(xfs_reflink_remap_adjust_rtlen);
 DEFINE_INODE_ERROR_EVENT(xfs_reflink_remap_range_error);
 DEFINE_INODE_ERROR_EVENT(xfs_reflink_set_inode_flag_error);
 DEFINE_INODE_ERROR_EVENT(xfs_reflink_update_inode_size_error);