Temporary and Pseudo Filesystems

Coverage: procfs/sysfs/debugfs/tracefs → tmpfs (shmem) → devtmpfs → ramfs → kernfs framework → filesystem implementation approaches (pseudo vs. real) Kernel versions: 2.6 ~ 6.x

Overview

"Pseudo-filesystems" do not store data on disk—their data is dynamically generated by the kernel upon access. The most common examples are /proc (process information), /sys (device model), and /sys/kernel/debug (debug interfaces). They appear as regular files, but a read() call triggers callback functions in the kernel rather than disk I/O.

The core value of pseudo-filesystems is to provide a unified, tool-friendly interface. cat /proc/meminfo is more intuitive than using system calls, ioctl, or special device files.

procfs: Process Information

// fs/proc/
// /proc/<pid>/ directory for each process:
//   task_struct → proc_dir_entry → inode → dentry
//   Accessing a file → triggers proc-specific read handler

// Example: /proc/<pid>/status
static const struct file_operations proc_status_operations = {
    .open       = proc_single_open,
    .read       = seq_read,       // Uses seq_file interface
    .release    = proc_single_release,
};

// seq_file interface (fs/seq_file.c):
//   A generic framework simplifying kernel text output generation
//   Automatically handles buffering, paging, and seeking
//   .start() / .next() / .stop() / .show()

// Implementation: task_state() → seq_printf() outputs each field
//   The process sees a plain text file, but the content is generated in real-time

Key proc Files

// /proc/<pid>/maps: VMA list (iterates mm->mmap)
// /proc/<pid>/smaps: Detailed memory statistics per VMA
// /proc/<pid>/fd/: Symbolic links for each fd → file path
// /proc/<pid>/attr/: Security attributes (SELinux/AppArmor)
// /proc/<pid>/mem: Process memory read/write (alternative to ptrace)

sysfs: Device Model

// fs/sysfs/
// sysfs is based on the kernfs framework (refactored in 6.x+)
// Each kobject → one sysfs directory/file

// kernfs_node is the core of sysfs:
struct kernfs_node {
    const char          *name;
    struct kernfs_node  *parent;
    const struct kernfs_ops *ops;       // File operations
    struct kernfs_elem_dir dir;         // Child nodes (if directory)
    struct kernfs_elem_symlink symlink; // Symbolic link
    struct kernfs_elem_attr attr;       // Attribute file
};

// sysfs attribute files:
//   .show(): cat /sys/... → generates text
//   .store(): echo xxx > /sys/... → parses and applies

// Tree structure of the device model:
//   /sys/devices/          ← Real device tree
//   /sys/bus/              ← Bus types → links to devices/
//   /sys/class/            ← Categorized by function → links to devices/
//   /sys/block/            ← Block devices

debugfs / tracefs

// fs/debugfs/
// Unstable API: Debug interface for developers
// Not relied upon in production systems (content/paths may change across versions)

struct dentry *debugfs_create_file(const char *name, umode_t mode,
                                     struct dentry *parent, void *data,
                                     const struct file_operations *fops);

// tracefs (fs/tracefs/):
//   Mount point for ftrace / tracepoints
//   /sys/kernel/tracing/ ← tracefs is mounted here
//   Replaces the old /sys/kernel/debug/tracing (debugfs)

tmpfs: Memory Filesystem

// mm/shmem.c
// tmpfs = shmem filesystem
// Mount: mount -t tmpfs tmpfs /tmp

// Data is stored in page cache (anonymous pages + swap-backed)
//   No disk backing → data lost on power failure
//   Supports swap: writes to swap when memory is tight

// Use cases:
//   /tmp:       Temporary files
//   /dev/shm:   POSIX shared memory
//   /run:       PID files and sockets for daemons
//   /var/run:   Backward-compatible symbolic link → /run

// Features:
//   Auto-growing (upper limit: size mount option, or 50% of RAM)
//   Supports POSIX ACL, xattr, O_TMPFILE

devtmpfs

// drivers/base/devtmpfs.c
// Dynamic device node management for /dev
// At kernel boot: devtmpfs is mounted → populated with known devices (e.g., /dev/sda, /dev/tty)
// Device hotplug: udev receives uevent notifications → creates/removes device nodes

// devtmpfs ensures that even during early boot (before udev is running)
//   basic device nodes are available
//   Handled automatically by kernels 2.6.32+

ramfs

// fs/ramfs/
// Extremely simple memory filesystem
//   No swap backing → data always stays in memory
//   No size limit → can exhaust memory (dangerous!)
//   No complexity of inode operations → shortest code

// Source code is only ~200 lines (inode.c + file-mmu.c)
// Mainly used in specific scenarios (inside initramfs, rootfs)

kernfs Framework (6.x+)

// fs/kernfs/
// Shared lower-level layer for sysfs and cgroupfs
// Provides:
//   1. Hierarchical namespace (directory tree)
//   2. File operations (read/write/mmap/poll)
//   3. Symbolic and hard links
//   4. RCU-safe traversal

// Why is kernfs needed?
//   The old sysfs and cgroupfs independently implemented their own VFS glue layers
//   → Duplicated code, each with its own bugs
//   kernfs unifies VFS interaction → reduces duplication, improves reliability

Approaches to Implementing Pseudo-Filesystems

Approach 1: seq_file (for text output in proc)
  → Suitable for: /proc/<pid>/status and other text generated line-by-line

Approach 2: kernfs (for sysfs/cgroupfs)
  → Suitable for: Attribute files (.show/.store), tree structures

Approach 3: debugfs (for debugging)
  → Suitable for: Simple file_operations registration

Approach 4: libfs (simple filesystem)
  → Suitable for: Pseudo-filesystems with only a few files
  → fs/libfs.c: simple_fill_super(), simple_lookup_dir(), ...
  → Minimal glue code

Debugging

# proc filesystem statistics
cat /proc/filesystems | grep nodev  # All pseudo-filesystems

# sysfs device tree
tree /sys/devices/

# tmpfs usage
df -t tmpfs

# Usage of debugfs in kernel (requires root)
mount -t debugfs none /sys/kernel/debug

References and Further Reading

Kernel Documentation: Documentation/filesystems/proc.rst, Documentation/filesystems/sysfs.rst, Documentation/filesystems/tmpfs.rst, Documentation/filesystems/debugfs.rst
Source Code:
- fs/proc/ — procfs
- fs/sysfs/ — sysfs
- fs/kernfs/ — kernfs framework
- fs/debugfs/ — debugfs
- mm/shmem.c — tmpfs
- drivers/base/devtmpfs.c — devtmpfs
- fs/libfs.c — Simple filesystem utilities

Keywords: procfs, sysfs, debugfs, tracefs, tmpfs, devtmpfs, kernfs, seq_file, pseudo-filesystem