Persistence

Key concepts

• › Files are sequences of bytes; the FS imposes structure on top
• › Inodes hold metadata + pointers to data blocks; directories are just files with name→inode mappings
• › Crash consistency requires either fsck (slow) or journaling (the modern default)
• › SSDs require erase-before-write; the FTL hides that behind a block-device interface
• › RAID trades drive capacity for redundancy; parity-based schemes (RAID 5/6) are common in storage arrays

Reference template

// Read path for one byte
1. read(fd, buf, 1)        → kernel
2. Lookup VFS / file entry → resolves to inode
3. Compute block address    → inode walk
4. Block in page cache?    → return
5. Block on disk?          → schedule I/O, sleep
6. Disk interrupt          → wake the process, finish copy
7. Update access time      → journaled if metadata journaling on

Adapt to your problem; the structure is the load-bearing part.

Common pitfalls

• ⚠ Treating write() as durable — it's not until fsync() returns
• ⚠ Forgetting that the page cache is between you and the disk on every read and write
• ⚠ Trusting RAID as backup — RAID handles drive failure, not user error or corruption
• ⚠ Ignoring SSD wear leveling — write-heavy patterns degrade endurance

Related topics

Items (11)

• I/O Devices and Drivers

  The canonical device protocol, interrupts vs polling, DMA, memory-mapped I/O, and the driver as an OS abstraction.

  Building Block Foundational
• Hard Disk Drives

  Geometry, seek + rotational latency + transfer time, disk scheduling (SSTF, SCAN, C-SCAN), and the I/O-cost math.

  Building Block Foundational
• RAID — Striping, Mirroring, Parity

  RAID 0/1/4/5/6 — capacity, performance, reliability trade-offs and where parity-based schemes break down.

  Building Block Intermediate
• Files and Directories

  The UNIX file abstraction, descriptors, inodes, hard vs symbolic links, fsync, mount, and the permission-bit model.

  Building Block Foundational
• File System Implementation

  Inode + data-block layout, free-space tracking, directory structures, access paths, caching, and the I/O cost per system call.

  Building Block Intermediate
• The Fast File System (FFS)

  Cylinder groups, locality policy, large-file exception — the design that made UNIX file systems orders of magnitude faster.

  System Intermediate
• Crash Consistency — fsck and Journaling

  Write-ahead logging, metadata-only vs full journaling, ordered mode, soft updates, and why fsck stopped scaling.

  Building Block Intermediate
• Log-Structured File System (LFS)

  Sequential writes, segments, inode maps, garbage collection — the design that influenced flash file systems and modern databases.

  System Advanced
• Flash SSDs and the Flash Translation Layer

  Cells, banks, planes; erase-before-write; the FTL log-structured mapping; garbage collection; wear leveling; trim.

  Building Block Advanced
• Data Integrity — Checksums and Scrubbing

  Latent sector errors, silent corruption, checksums, mismatched-write protection, periodic scrubbing — bit-rot defense.

  Building Block Intermediate
• GitLab 2017 — The Database Outage

  How a 'wrong terminal' rm on a primary led to ~6 hours of data loss; backups that didn't work; the public postmortem.

  Postmortem Foundational