Filesystem Journaling: Write-Ahead Logging

The Problem: Crashes During Writes

Imagine you're updating a file when suddenly—power failure! Without protection, your filesystem could be left in an inconsistent state:

• Half-written metadata: Directory entries point to freed blocks
• Orphaned data: Allocated blocks with no file reference
• Corrupted structures: Inconsistent inode tables, bitmaps

Traditional filesystems required full disk scans (fsck) after crashes—potentially hours on large drives. Journaling solves this with write-ahead logging.

The Journaling Solution

Core Idea: Before making any changes, write your intentions to a journal (transaction log). If a crash occurs, replay the journal to complete or undo partial operations.

Think of it like a chef's prep notes: write down what you're about to cook before you start. If interrupted, check your notes to know what state you're in.

How Journaling Works: Interactive Exploration

See the journaling mechanism in action—from transaction start to commit, and crash recovery:

Journal Modes: Safety vs Performance

Different journaling modes offer varying guarantees:

1. Journal Mode (Full Journaling)

• What's journaled: Both metadata AND data
• Process: Write data to journal → Write metadata to journal → Commit → Write to final location
• Safety: Highest - complete consistency
• Performance: Slowest - everything written twice
• Use case: Critical data (financial systems)

2. Ordered Mode (Default)

• What's journaled: Only metadata
• Process: Write data to disk → Write metadata to journal → Commit → Write metadata to final location
• Safety: High - metadata consistent, data may be old
• Performance: Balanced - data written once
• Use case: Most systems (ext4 default, XFS)

3. Writeback Mode

• What's journaled: Only metadata
• Process: Write metadata to journal → Commit → Write data and metadata to disk (any order)
• Safety: Lower - metadata consistent, data may be garbage
• Performance: Fastest - no ordering constraints
• Use case: Non-critical data, scratch disks

The Transaction Lifecycle

1. Transaction Start
   ↓
2. Write Intent to Journal (WAL)
   ↓
3. Wait for Journal Flush
   ↓
4. Commit Transaction (Commit Record)
   ↓
5. Apply Changes to Filesystem
   ↓
6. Mark Journal Entries as Completed (Checkpoint)
   ↓
7. Reuse Journal Space (Circular Buffer)

Recovery After Crash

When a filesystem mounts after a crash:

1. Scan Journal: Read from last checkpoint
2. Check Commit Records: Find complete vs incomplete transactions
3. Replay Complete: Apply committed but not-yet-applied changes
4. Rollback Incomplete: Ignore uncommitted transactions
5. Mount Filesystem: Now in consistent state

Recovery Time: Seconds to minutes (scanning journal only), not hours (scanning entire disk).

Journaling in Different Filesystems

ext4

• Journal: Dedicated journal inode or external device
• Modes: journal, ordered (default), writeback
• Journal size: Configurable (default ~128MB)
• Command: tune2fs -o journal_data /dev/sda1

XFS

• Journal: Metadata-only (always ordered mode)
• Log size: Configurable with -l size=128m
• Real-time log: Optional separate device for sync writes
• Efficient: Only logs metadata changes

NTFS

• $LogFile: Transaction log for metadata
• Mode: Metadata journaling only
• Recovery: Automatic on mount (chkdsk if needed)
• USN Journal: Separate change journal for applications

Btrfs / ZFS

• No traditional journal: Use Copy-on-Write instead
• Atomic operations: CoW provides transaction semantics
• See: Copy-on-Write mechanism

Performance Impact

Journal Placement

# Internal journal (default)
mkfs.ext4 /dev/sda1

# External journal (faster, separate device)
mkfs.ext4 -J device=/dev/sdb1 /dev/sda1

External journal benefits:

• Reduced seek time (journal on SSD, data on HDD)
• Parallel I/O
• Better for write-heavy workloads

Journal Size Tuning

# Larger journal = more buffering, less frequent commits
tune2fs -J size=400 /dev/sda1  # 400MB journal

# For databases: larger journal reduces checkpoint frequency

Best Practices

1. Use ordered mode for most workloads (default)
2. Enable full journaling only for critical data
3. External journal on separate SSD for performance
4. Monitor journal wraps: dumpe2fs /dev/sda1 | grep -i journal
5. Disable journaling only for scratch/tmp filesystems

Journaling vs Copy-on-Write

When to choose:

• Journaling (ext4, XFS): Maximum maturity, proven reliability
• CoW (Btrfs, ZFS): Want snapshots, checksums, modern features

Related concepts