f2fs: put directory inodes before checkpoint in roll-forward recovery

BugLink: https://bugs.launchpad.net/bugs/1818797

commit 9e1e6df4 upstream.

Before checkpoint, we'd be better drop any inodes.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Juerg Haefliger <juergh@canonical.com>
Signed-off-by: Khalid Elmously <khalid.elmously@canonical.com>

fs/f2fs/recovery.c

@@ -597,6 +597,9 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 		set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 	mutex_unlock(&sbi->cp_mutex);
 
+	/* let's drop all the directory inodes for clean checkpoint */
+	destroy_fsync_dnodes(&dir_list);
+
 	if (!err && need_writecp) {
 		struct cp_control cpc = {
 			.reason = CP_RECOVERY,
@@ -604,7 +607,6 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 		write_checkpoint(sbi, &cpc);
 	}
 
-	destroy_fsync_dnodes(&dir_list);
 	kmem_cache_destroy(fsync_entry_slab);
 	return ret ? ret: err;
 }