FUSE Subsystem Tutorial

中文

[English]

Main contents

Source baseline: Linux 7.2-rc6.

Chapter 1: FUSE subsystem design

Chapter 1 explains why FUSE needs its objects, protocol identities, cache leases, and queue rules without depending on individual source functions.

1. Boundaries, layers, and end-to-end architecture
2. Core objects, protocol identity, and capability negotiation
3. Caching, I/O modes, and coherency
4. Concurrency, backpressure, security, and failure

Chapter 2: Linux implementation

Chapter 2 follows Linux source code through object creation, request submission, daemon replies, and resource release.

1. Source map, module initialization, mount, and FUSE_INIT
2. Path walking, directory operations, attributes, and references
3. OPEN, cached I/O, direct I/O, writeback, and RELEASE
4. Request allocation, queue state machine, and /dev/fuse transport
5. Ordinary mmap and virtio-fs DAX
6. Teardown, abort, debugging, and performance analysis

Learning objectives

After completing the tutorial, you should be able to:

• distinguish a VFS inode, a FUSE nodeid, its generation, and an open file handle;
• explain request movement through pending, io, processing, and finished states;
• analyze dentry, attribute, and data-cache validity and invalidation;
• trace a VFS operation to fuse_simple_request() and /dev/fuse;
• determine when cached I/O, direct I/O, passthrough, or DAX is selected;
• use fusectl and FUSE tracepoints to diagnose daemon stalls, queue buildup, and reply errors.