sandbox-rs

Lightweight process sandboxing for Linux

Features

• Unprivileged mode — works without root via user namespaces, Landlock, and setrlimit
• Privileged mode — full isolation with cgroups v2, chroot, and all namespace types
• Auto-detection — automatically picks the best mode for the current environment
• Seccomp BPF — six built-in syscall filtering profiles
• Landlock — filesystem access control without root (Linux 5.13+)
• Resource limits — memory, CPU, and PID constraints
• Streaming output — real-time stdout/stderr capture

Requirements

• Linux kernel 5.10+ (5.13+ for Landlock support)
• Root is optional — unprivileged mode uses user namespaces + seccomp + Landlock + setrlimit

Quick Start

Library

[dependencies]
sandbox-rs = "0.2"

use sandbox_rs::{SandboxBuilder, SeccompProfile, PrivilegeMode};
use std::time::Duration;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut sandbox = SandboxBuilder::new("my-sandbox")
        .privilege_mode(PrivilegeMode::Unprivileged)
        .memory_limit_str("256M")?
        .cpu_limit_percent(50)
        .timeout(Duration::from_secs(30))
        .seccomp_profile(SeccompProfile::IoHeavy)
        .build()?;

    let result = sandbox.run("/bin/echo", &["hello world"])?;
    println!("exit={} mem={}B cpu={}μs", result.exit_code, result.memory_peak, result.cpu_time_us);
    Ok(())
}

Note: memory_peak and cpu_time_us require privileged mode (cgroups v2). In unprivileged mode these values are 0.

CLI

# Run a program in a sandbox (auto-detects privilege mode)
sandbox-ctl /bin/echo "hello world"

# Use a security profile with resource limits
sandbox-ctl --profile moderate --memory 512M --cpu 50 python script.py

# Check system capabilities
sandbox-ctl check

# List seccomp profiles
sandbox-ctl seccomp

Seccomp Profiles

Each profile includes all syscalls from profiles below it (cumulative).

Profile	Syscalls
Essential	Process bootstrap only (~40): execve, mmap, brk, read, write, exit, ...
Minimal	Essential + signals, pipes, timers, process control (~110 total)
IoHeavy	Minimal + file manipulation: mkdir, chmod, unlink, rename, fsync, ...
Compute	IoHeavy + scheduling/NUMA: sched_setscheduler, mbind, membarrier, ...
Network	Compute + sockets: socket, bind, listen, connect, sendto, ...
Unrestricted	Network + privileged: ptrace, mount, bpf, setuid, ...

Security

• Defense-in-depth: multiple isolation layers (namespaces, seccomp, Landlock, cgroups)
• Combine with AppArmor or SELinux for production use
• Kernel vulnerabilities can bypass sandbox boundaries — keep your kernel updated
• Not a replacement for VM-level isolation for fully untrusted code

License

MIT — see LICENSE for details.