Tech Study Guide

Linux Filesystems and IO

VFS, inodes, dentries, page cache, block devices, mounts, filesystem checks, IO schedulers, and storage troubleshooting.

linux filesystems storage io

Linux Filesystems and IO

Filesystems are where application abstractions meet storage reality. The full stack includes VFS objects, inodes, dentries, page cache, block devices, device mapper, filesystem journals, IO schedulers, and hardware queues.

Command Examples

findmnt
df -h
df -i
lsblk -f
iostat -xz 1
journalctl -k -g 'I/O error|EXT4|XFS|blk|nvme|scsi'

Example output and meaning:

Command Example output What it does
findmnt Device names, filesystems, mountpoints, latency, errors, or health fields. Connects storage symptoms to device and filesystem evidence.
df -h Device names, filesystems, mountpoints, latency, errors, or health fields. Connects storage symptoms to device and filesystem evidence.
df -i Device names, filesystems, mountpoints, latency, errors, or health fields. Connects storage symptoms to device and filesystem evidence.

VFS Model

The Virtual Filesystem layer gives Linux one interface over many filesystems. Applications call open, read, write, fsync, rename, and stat; VFS routes those operations to the mounted filesystem and underlying block or network storage.

Important objects:

  • inode: metadata for a file object,
  • dentry: directory-entry name lookup cache,
  • superblock: mounted filesystem metadata,
  • page cache: cached file data,
  • file descriptor: process handle to an opened file.

Page Cache and Durability

A successful write() often means data reached kernel memory, not durable storage. fsync() or equivalent database durability paths matter when power loss or kernel crash safety is required.

Dirty pages are eventually written back. Dirty writeback tuning, slow storage, and congested queues can create latency spikes far from the code that caused the writes.

flowchart LR
  App[read/write syscall] --> VFS[VFS and filesystem]
  VFS --> Cache[Page cache]
  Cache -->|cache hit| App
  Cache -->|dirty pages| Writeback[writeback threads]
  Writeback --> Block[block layer scheduler]
  Block --> Device[SSD / HDD / network volume]
  Device --> Complete[IO completion]

fio examples:

fio --name=randread --filename=/mnt/testfile --size=2G --rw=randread --bs=4k --iodepth=32 --direct=1
fio --name=seqwrite --filename=/mnt/testfile --size=2G --rw=write --bs=1M --iodepth=8 --direct=1
fio --name=fsync --filename=/mnt/testfile --size=512M --rw=write --bs=4k --fsync=1

Use --direct=1 when you want device behavior more than cache behavior. Use fsync-heavy tests for databases and metadata-sensitive workloads.

Mounts and Namespaces

Mounts define where filesystems appear. Containers may have different mount namespaces from the host, so a path can exist in one namespace but not another. Bind mounts and overlay filesystems are common in container runtimes.

Failure Modes

  • inode exhaustion with free bytes still available,
  • file descriptor leak,
  • deleted-but-open log files consuming disk,
  • read-only remount after filesystem error,
  • saturated block device queue,
  • slow fsync,
  • page cache hiding disk read cost during tests,
  • mount namespace mismatch between host and container.

Study Cards

Question

What does VFS provide?

Answer

A common kernel interface for file operations across many filesystem implementations.

Question

Why can df show free space while writes fail?

Answer

The filesystem may be out of inodes, quotas, or writable mount state.

Question

Why does fsync matter?

Answer

It asks the system to make file data or metadata durable beyond page cache.

References