🚀 FastDivert

A driver-backed Windows kernel interception SDK for building professional user-space applications without writing your own drivers.

FastDivert is a modern, ultra-high-performance Windows kernel-mode framework for network packet/ file IO capture, interception, and injection.

it is designed to process 10Gbps+ network traffic with minimal CPU overhead and low latency.

• ⚡ Ultra-High Performance: Eliminates syscalls and lock contention via zero-copy shared memory, per-core lock-free ring buffers, and configurable poll modes (BusyPoll/IoctlWait).

• 🌐 L3/L4 Network Interception: Full support for IPv4/IPv6, stream, socket, and flow-level events, featuring high-fidelity traffic capture and dynamic injection capabilities.

• 📂 File I/O Control: Real-time directory monitoring and synchronous file access evaluation (Allow/Deny) with dynamic path redirection (STATUS_REPARSE).

• 🛡️ Rust Memory Safety: Built with standard WDK crates, eradicating buffer overflows and use-after-free vulnerabilities to ensure kernel-level BSOD resistance.

• 💻 Cross-Architecture: Native support for x64 and ARM64 (Windows on ARM) architectures.

• Windows 10 / 11 / Server (x64 or ARM64)
• Rust toolchain (with WDK build bindings)
• Windows Driver Kit (WDK) installed

Compile both the kernel driver (.sys) and user-mode library (.dll) seamlessly:

cargo wdk build

Ensure Windows Test Signing is enabled (bcdedit /set testsigning on and reboot).

• Packet Counter:

  cd demo/packet_count
  cargo run --release

• File Monitor:

  cd demo/file_monitor
  cargo run --release

use fastdivert::{Divert, Flags, Layer};

fn main() -> anyhow::Result<()> {
    let divert = Divert::open_with_driver_path(
        "true", "FastDivert", "fast_divert.sys", Layer::Network as u32, 0, Flags::RecvOnly as u64,
    )?;

    divert.poll_multi_threads(4, |_, _| {
        // Handle packet data in user-space
    }, || {
        std::thread::sleep(std::time::Duration::from_millis(10));
    });
}

use fastdivert::{Divert, FileCallbackDecision, FileModuleConfig, DefaultWait, FILE_OP_WRITE, FILE_MATCH_GLOB, FILE_ACTION_ALLOW};

fn main() -> anyhow::Result<()> {
    let config = FileModuleConfig::new(5000, FILE_ACTION_ALLOW)
        .add_filter(r"*_block.txt", FILE_OP_WRITE, FILE_MATCH_GLOB, 0, false)?;

    let divert = Divert::open_file(&config)?;
    divert.poll_file_events(|ref_event| {
        println!("PID: {} | Path: {}", ref_event.event.process_id, ref_event.path);
        if ref_event.path.ends_with(".txt") {
            FileCallbackDecision::Deny
        } else {
            FileCallbackDecision::Allow
        }
    }, DefaultWait(&divert))?;
}

• CPU: AMD Ryzen 9 9900X3D (8 vCPUs allocated in VMware Workstation)
• OS: Windows Server 2025
• Network Adapter: VMware VMXNET3

Tested using demo/packet_count under high network load.

• What: Real-time synchronous tracking and active blocking of file write/creation activities combined with network socket monitoring.
• Value: Instantly detects and halts unauthorized file encryption (Ransomware behavior) or exfiltration (Data Loss Prevention) with authoritative kernel-level intervention before any data leaves the host.

• What: Transparent redirection of specific local/remote network traffic to user-space security proxies without changing system-wide proxy settings.
• Value: Enables seamless micro-segmentation, identity-based access control, and inline deep-packet inspection (DPI) with zero context-switch bottlenecks.

• What: Sub-microsecond packet capture, modification, and batch injection directly from user-space threads pinned to CPU cores.
• Value: Delivers ultra-low latency game acceleration (VPNs) and high-frequency trading gateway connections with zero packet drop rates.

• What: Real-time filesystem path redirection (STATUS_REPARSE) based on process ID tracking.
• Value: Easily isolates untrusted applications by transparently redirecting their file modifications to a secure, ephemeral directory without virtual machine overhead.

This project is dual-licensed.

• Open Source Option: Licensed under the AGPLv3 License.
• Commercial Option: Please contact the author for commercial licensing.

If you have any questions, feedback, please feel free to reach out:

• GitHub Issues: For bug reports and feature requests.
• Email: hello@one-api.net