IO

Linux file operations written in pure x86_64 assembly with a C interface as a Shared-Library.

IO — Minimal, High-Performance I/O Library in x86_64 Assembly

Designed for performance-critical applications that need direct I/O (O_DIRECT), synchronized writes (O_DSYNC).

Features

• Pure x86_64 assembly implementation — minimal overhead, full syscall ABI control
• EINTR-safe — all operations correctly handle interrupted syscalls (except close, which follows modern best practices)
• Direct I/O ready — proper alignment handling via fpathconf(_PC_REC_XFER_ALIGN)
• Zero-copy / low-level focus — io_pread / io_pwrite with partial-transfer retry loops

Quick Start

1. Build the library and demo

# Use the convenience script in the root folder:
sh ./io_make.sh
# Or use make in the individual folders
cd ./io/id
make clean; make    # builds libio.so
cd ../demo
make clean; make    # builds the demo binary

2. Run the demo

cd ./demo
./go_demo.sh

The demo creates ./out.txt (containing 10,000 integers (0–9999) stored in fixed 16-byte records) using O_CREAT | O_DIRECT | O_RDWR | O_DSYNC | O_TRUNC, then re-opens the file and reads the data back.

API

All functions are declared in io/io.h:

int   io_close(int fd);
int   io_creat(const char *path, int flags, mode_t mode);
int   io_data_sync(int fd);
off_t io_lseek (int fd, off_t offset, int whence);
int   io_open(const char *path, int flags);
ssize_t io_pread(int fd, void *buf, size_t nbyte, off_t offset);
ssize_t io_pwrite(int fd, const void *buf, size_t nbyte, off_t offset);
ssize_t io_read(int fd, void *buf, size_t nbyte);
int   io_sync(int fd);
ssize_t io_write(int fd, const void *buf, size_t nbyte);

Key Design Notes

• io_creat uses the open syscall (more flexible than the legacy creat syscall)
• io_pread / io_pwrite / io_read / io_write automatically retry on partial transfers and handle EINTR
• io_pwrite and io_write treat a zero-byte return from pwrite64 or write as an error (EIO)
• io_close does not retry on EINTR (modern Linux best practice to avoid fd-reuse races)