[08/12] xfs_fsr: skip the xattr/forkoff levering with the newer swapext implementations

Commit Message

Darrick J. Wong July 2, 2024, 12:55 a.m. UTC

From: Darrick J. Wong <[e-mail]>

The newer swapext implementations in the kernel run at a high enough
level (above the bmap layer) that it's no longer required to manipulate
bs_forkoff by creating garbage xattrs to get the extent tree that we
want.  If we detect the newer algorithms, skip this error prone step.

Signed-off-by: Darrick J. Wong <[e-mail]>
---
 fsr/xfs_fsr.c |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

Comments

Christoph Hellwig July 2, 2024, 5:14 a.m. UTC | #1

Looks good:

Reviewed-by: Christoph Hellwig <[e-mail]>

Patch

diff --git a/fsr/xfs_fsr.c b/fsr/xfs_fsr.c
index 2989e4c989bf..7d0639868258 100644
--- a/fsr/xfs_fsr.c
+++ b/fsr/xfs_fsr.c
@@ -999,6 +999,20 @@  fsr_setup_attr_fork(
 	if (!(bstatp->bs_xflags & FS_XFLAG_HASATTR))
 		return 0;
 
+	/*
+	 * If the filesystem has the ability to perform atomic file mapping
+	 * exchanges, the file extent swap implementation uses a higher level
+	 * algorithm that calls into the bmap code instead of playing games
+	 * with swapping the extent forks.
+	 *
+	 * This new functionality does not require specific values of
+	 * bs_forkoff, unlike the old fork swap code.  Leave the extended
+	 * attributes alone if we know we're not using the old fork swap
+	 * strategy.  This eliminates a major source of runtime errors in fsr.
+	 */
+	if (fsgeom.flags & XFS_FSOP_GEOM_FLAGS_EXCHANGE_RANGE)
+		return 0;
+
 	/*
 	 * use the old method if we have attr1 or the kernel does not yet
 	 * support passing the fork offset in the bulkstat data.