Merge tag 'realtime-reflink_2024-12-23' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into for-next

xfs: reflink on the realtime device [v6.2 05/14]

This patchset enables use of the file data block sharing feature (i.e.
reflink) on the realtime device.  It follows the same basic sequence as
the realtime rmap series -- first a few cleanups; then  introduction of
the new btree format and inode fork format.  Next comes enabling CoW and
remapping for the rt device; new scrub, repair, and health reporting
code; and at the end we implement some code to lengthen write requests
so that rt extents are always CoWed fully.

This has been running on the djcloud for months with no problems.  Enjoy!

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

1 files changed, 26 insertions, 2 deletions

diff --git a/fs/xfs/libxfs/xfs_rtrmap_btree.c b/fs/xfs/libxfs/xfs_rtrmap_btree.c
index b90901e39e92..e4ec36943cb7 100644
--- a/fs/xfs/libxfs/xfs_rtrmap_btree.c
+++ b/fs/xfs/libxfs/xfs_rtrmap_btree.c
@@ -718,6 +718,7 @@ xfs_rtrmapbt_maxrecs(
 unsigned int
 xfs_rtrmapbt_maxlevels_ondisk(void)
 {
+	unsigned long long	max_dblocks;
 	unsigned int		minrecs[2];
 	unsigned int		blocklen;
 
@@ -726,8 +727,20 @@ xfs_rtrmapbt_maxlevels_ondisk(void)
 	minrecs[0] = xfs_rtrmapbt_block_maxrecs(blocklen, true) / 2;
 	minrecs[1] = xfs_rtrmapbt_block_maxrecs(blocklen, false) / 2;
 
-	/* We need at most one record for every block in an rt group. */
-	return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_RGBLOCKS);
+	/*
+	 * Compute the asymptotic maxlevels for an rtrmapbt on any rtreflink fs.
+	 *
+	 * On a reflink filesystem, each block in an rtgroup can have up to
+	 * 2^32 (per the refcount record format) owners, which means that
+	 * theoretically we could face up to 2^64 rmap records.  However, we're
+	 * likely to run out of blocks in the data device long before that
+	 * happens, which means that we must compute the max height based on
+	 * what the btree will look like if it consumes almost all the blocks
+	 * in the data device due to maximal sharing factor.
+	 */
+	max_dblocks = -1U; /* max ag count */
+	max_dblocks *= XFS_MAX_CRC_AG_BLOCKS;
+	return xfs_btree_space_to_height(minrecs, max_dblocks);
 }
 
 int __init
@@ -766,9 +779,20 @@ xfs_rtrmapbt_compute_maxlevels(
 	 * maximum height is constrained by the size of the data device and
 	 * the height required to store one rmap record for each block in an
 	 * rt group.
+	 *
+	 * On a reflink filesystem, each rt block can have up to 2^32 (per the
+	 * refcount record format) owners, which means that theoretically we
+	 * could face up to 2^64 rmap records.  This makes the computation of
+	 * maxlevels based on record count meaningless, so we only consider the
+	 * size of the data device.
 	 */
 	d_maxlevels = xfs_btree_space_to_height(mp->m_rtrmap_mnr,
 				mp->m_sb.sb_dblocks);
+	if (xfs_has_rtreflink(mp)) {
+		mp->m_rtrmap_maxlevels = d_maxlevels + 1;
+		return;
+	}
+
 	r_maxlevels = xfs_btree_compute_maxlevels(mp->m_rtrmap_mnr,
 				mp->m_groups[XG_TYPE_RTG].blocks);