rbd: Use assume_storage_prezeroed when formatting

Instead of passing lazy_itable_init=1 and lazy_journal_init=1 to
mkfs.ext4, pass assume_storage_prezeroed=1 which is
stronger and allows the filesystem to skip inode table zeroing
completely instead of simply doing it lazily.

The support for this flag is checked by trying to format a fake
temporary image with mkfs.ext4 and checking its STDERR.

Closes: ceph#4948

Signed-off-by: Niraj Yadav <[email protected]>

internal/rbd/nodeserver.go

@@ -67,6 +67,10 @@ const (
 	xfsReflinkNoSupport
 	xfsReflinkSupport
 
+	ext4PrezeroedUnset
+	ext4PrezeroedNoSupport
+	ext4PrezeroedSupport
+
 	staticVol        = "staticVolume"
 	volHealerCtx     = "volumeHealerContext"
 	tryOtherMounters = "tryOtherMounters"
@@ -100,6 +104,9 @@ var (
 	// checking the support for reflink.
 	xfsHasReflink = xfsReflinkUnset
 
+	// ext4HasPrezeroed is set by ext4SupportsPrezeroed()
+	ext4HasPrezeroedSupport = ext4PrezeroedUnset
+
 	mkfsDefaultArgs = map[string][]string{
 		"ext4": {"-m0", "-Enodiscard,lazy_itable_init=1,lazy_journal_init=1"},
 		"xfs":  {"-K"},
@@ -794,6 +801,9 @@ func (ns *NodeServer) mountVolumeToStagePath(
 
 	if existingFormat == "" && !staticVol && !readOnly && !isBlock {
 		args := mkfsDefaultArgs[fsType]
+		if fsType == "ext4" && ns.ext4SupportsPrezeroed() {
+			args = []string{"-m0", "-Enodiscard,assume_storage_prezeroed=1"}
+		}
 
 		// if the VolumeContext contains "mkfsOptions", use those as args instead
 		volumeCtx := req.GetVolumeContext()
@@ -1314,6 +1324,44 @@ func (ns *NodeServer) xfsSupportsReflink() bool {
 	return false
 }
 
+// ext4SupportsPrezeroed checks if the ext4 filesystem supports the
+// "assume_storage_prezeroed" option. It does this by creating a temporary
+// image file, attempting to format it with the ext4 filesystem using the
+// "assume_storage_prezeroed" option, and checking the output for errors.
+func (ns *NodeServer) ext4SupportsPrezeroed() bool {
+	if ext4HasPrezeroedSupport != ext4PrezeroedUnset {
+		return ext4HasPrezeroedSupport == ext4PrezeroedSupport
+	}
+
+	tempImage := "/tmp/prezeroed.img"
+	ctx := context.TODO()
+	_, _, err := util.ExecCommand(ctx, "dd", "if=/dev/zero", "of="+tempImage, "bs=1M", "count=1")
+	if err != nil {
+		log.ErrorLog(ctx, "failed to create temporary image %s: %v", tempImage, err)
+
+		return false
+	}
+	defer os.Remove(tempImage)
+
+	diskMounter := &mount.SafeFormatAndMount{Interface: ns.Mounter, Exec: utilexec.New()}
+
+	// `-n` Causes mke2fs to not actually create a file system, but display what it would do if it were to create a file system.
+	out, err := diskMounter.Exec.Command("mkfs.ext4", "-n", "-Enodiscard,assume_storage_prezeroed=1", tempImage).CombinedOutput()
+	if err != nil {
+		if strings.Contains(string(out), "Bad option(s) specified") {
+			ext4HasPrezeroedSupport = ext4PrezeroedNoSupport
+		}
+
+		// if the error is not due to the missing option, we can't be sure
+		// if the option is supported or not, return false optimistically
+		return false
+	}
+
+	ext4HasPrezeroedSupport = ext4PrezeroedSupport
+
+	return true
+}
+
 // NodeGetVolumeStats returns volume stats.
 func (ns *NodeServer) NodeGetVolumeStats(
 	ctx context.Context,