From ed5b89aa0bee1ad46ec2c914d7d7b4d56fea4462 Mon Sep 17 00:00:00 2001
From: Niraj Yadav <niryadav@redhat.com>
Date: Wed, 27 Nov 2024 15:08:29 +0530
Subject: [PATCH] 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: #4948

Signed-off-by: Niraj Yadav <niryadav@redhat.com>
---
 internal/rbd/nodeserver.go | 101 +++++++++++++++++++++++++++++++++++--
 1 file changed, 96 insertions(+), 5 deletions(-)

diff --git a/internal/rbd/nodeserver.go b/internal/rbd/nodeserver.go
index 905fcf8affe..cbd1764b8d2 100644
--- a/internal/rbd/nodeserver.go
+++ b/internal/rbd/nodeserver.go
@@ -67,6 +67,10 @@ const (
 	xfsReflinkNoSupport
 	xfsReflinkSupport
 
+	ext4PrezeroedUnset
+	ext4PrezeroedNoSupport
+	ext4PrezeroedSupport
+
 	staticVol        = "staticVolume"
 	volHealerCtx     = "volumeHealerContext"
 	tryOtherMounters = "tryOtherMounters"
@@ -100,10 +104,9 @@ var (
 	// checking the support for reflink.
 	xfsHasReflink = xfsReflinkUnset
 
-	mkfsDefaultArgs = map[string][]string{
-		"ext4": {"-m0", "-Enodiscard,lazy_itable_init=1,lazy_journal_init=1"},
-		"xfs":  {"-K"},
-	}
+	// ext4HasPrezeroed is set by ext4SupportsPrezeroed(), use the function when
+	// checking the support for assume_storage_prezeroed.
+	ext4HasPrezeroedSupport = ext4PrezeroedUnset
 
 	mountDefaultOpts = map[string][]string{
 		"xfs": {"nouuid"},
@@ -793,7 +796,7 @@ func (ns *NodeServer) mountVolumeToStagePath(
 	}
 
 	if existingFormat == "" && !staticVol && !readOnly && !isBlock {
-		args := mkfsDefaultArgs[fsType]
+		args := ns.getMkfsArgs(fsType)
 
 		// if the VolumeContext contains "mkfsOptions", use those as args instead
 		volumeCtx := req.GetVolumeContext()
@@ -1314,6 +1317,55 @@ 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
+	}
+
+	ctx := context.TODO()
+	tempImgFile, err := os.CreateTemp(os.TempDir(), "prezeroed.img")
+	if err != nil {
+		log.WarningLog(ctx, "failed to create temporary image file: %v", err)
+
+		return false
+	}
+	defer os.Remove(tempImgFile.Name())
+
+	if err = createSparseFile(tempImgFile, 1); err != nil {
+		log.WarningLog(ctx, "failed to create sparse file: %v", err)
+
+		return false
+	}
+
+	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",
+		tempImgFile.Name(),
+	).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,
@@ -1397,3 +1449,42 @@ func getDeviceSize(ctx context.Context, devicePath string) (uint64, error) {
 
 	return size, nil
 }
+
+// getMkfsArgs returns the appropriate mkfs arguments for the given filesystem type.
+// Supported filesystem types are "ext4" and "xfs". For "ext4", it checks if the
+// ext4 filesystem supports pre-zeroed storage and returns the corresponding arguments.
+// For "xfs", it returns the arguments to disable discard. If an unknown filesystem
+// type is provided, a warning is logged and empty options are returned.
+func (ns *NodeServer) getMkfsArgs(fsType string) []string {
+	switch fsType {
+	case "ext4":
+		if ns.ext4SupportsPrezeroed() {
+			return []string{"-m0", "-Enodiscard,assume_storage_prezeroed=1"}
+		}
+
+		return []string{"-m0", "-Enodiscard,lazy_itable_init=1,lazy_journal_init=1"}
+	case "xfs":
+		return []string{"-K"}
+	default:
+		log.WarningLogMsg(fmt.Sprintf("unknown fsType: %q, using default mkfs options", fsType))
+
+		return []string{}
+	}
+}
+
+// createSpareFile makes `file` a sparse file of size `sizeMB`.
+func createSparseFile(file *os.File, sizeMB int64) error {
+	sizeBytes := sizeMB * 1024 * 1024
+
+	// seek to the end of the file.
+	if _, err := file.Seek(sizeBytes-1, 0); err != nil {
+		return fmt.Errorf("failed to seek to the end of the file: %w", err)
+	}
+
+	// write a single byte, effectively making it a sparse file.
+	if _, err := file.Write([]byte{0}); err != nil {
+		return fmt.Errorf("failed to write to the end of the file: %w", err)
+	}
+
+	return nil
+}