openentropy-core 0.12.3

//! Mach IPC timing — entropy from complex Mach messages with OOL descriptors.

#[cfg(target_os = "macos")]
use std::sync::Arc;
#[cfg(target_os = "macos")]
use std::sync::atomic::{AtomicBool, Ordering};
#[cfg(target_os = "macos")]
use std::thread;

use crate::source::{EntropySource, Platform, SourceCategory, SourceInfo};
#[cfg(target_os = "macos")]
use crate::sources::helpers::{extract_timing_entropy, mach_time};

/// Configuration for Mach IPC entropy collection.
///
/// # Example
/// ```
/// # use openentropy_core::sources::ipc::MachIPCConfig;
/// let config = MachIPCConfig {
///     num_ports: 16,               // more ports = more contention
///     ool_size: 8192,              // larger OOL = more VM work
///     use_complex_messages: true,  // OOL messages (recommended)
/// };
/// ```
#[derive(Debug, Clone)]
pub struct MachIPCConfig {
    /// Number of Mach ports to round-robin across.
    ///
    /// More ports create more namespace contention (splay tree operations)
    /// and varied queue depths. Each port is allocated with both receive
    /// and send rights.
    ///
    /// **Range:** 1-64 (clamped to ≥1). **Default:** `8`
    pub num_ports: usize,

    /// Size of out-of-line (OOL) memory descriptors in bytes.
    ///
    /// OOL descriptors force the kernel to perform `vm_map_copyin`/`copyout`
    /// operations, exercising page table updates and physical page allocation.
    /// Larger sizes mean more VM work per message.
    ///
    /// **Range:** 1-65536. **Default:** `4096` (one page)
    pub ool_size: usize,

    /// Use complex messages with OOL descriptors (`true`) or simple port
    /// allocate/deallocate (`false`, legacy behavior).
    ///
    /// Complex messages traverse deeper kernel paths and produce significantly
    /// higher timing variance than simple port operations.
    ///
    /// **Default:** `true`
    pub use_complex_messages: bool,
}

impl Default for MachIPCConfig {
    fn default() -> Self {
        Self {
            num_ports: 8,
            ool_size: 4096,
            use_complex_messages: true,
        }
    }
}

/// Harvests timing jitter from Mach IPC using complex OOL messages.
///
/// # What it measures
/// Nanosecond timing of `mach_msg()` sends with out-of-line memory descriptors,
/// round-robined across a pool of Mach ports.
///
/// # Why it's entropic
/// Complex Mach messages with OOL descriptors traverse deep kernel paths:
/// - **OOL VM remapping** — `vm_map_copyin`/`vm_map_copyout` exercises page
///   tables, physical page allocation, and TLB updates
/// - **Port namespace contention** — round-robin across ports exercises the
///   splay tree with timing dependent on tree depth and rebalancing
/// - **Per-port lock contention** — `ipc_mqueue_send` acquires per-port locks
/// - **Receiver thread wakeup** — cross-thread scheduling decisions affected
///   by ALL runnable threads
///
/// # What makes it unique
/// Mach IPC is unique to XNU/macOS. Unlike higher-level IPC (pipes, sockets),
/// Mach messages go through XNU's `ipc_mqueue` subsystem with entirely different
/// locking and scheduling paths. OOL descriptors add VM operations that no
/// other entropy source exercises.
///
/// # Configuration
/// See [`MachIPCConfig`] for tunable parameters. Key options:
/// - `use_complex_messages`: OOL messages vs simple port ops (recommended: `true`)
/// - `num_ports`: controls namespace contention level
/// - `ool_size`: controls VM remapping workload per message
#[derive(Default)]
pub struct MachIPCSource {
    /// Source configuration. Use `Default::default()` for recommended settings.
    pub config: MachIPCConfig,
}

static MACH_IPC_INFO: SourceInfo = SourceInfo {
    name: "mach_ipc",
    description: "Mach port complex OOL message and VM remapping timing jitter",
    physics: "Sends complex Mach messages with out-of-line (OOL) memory descriptors via \
              mach_msg(), round-robining across multiple ports. OOL descriptors force kernel \
              VM remapping (vm_map_copyin/copyout) which exercises page table operations. \
              Round-robin across ports with varied queue depths creates namespace contention. \
              Timing captures: OOL VM remap latency, port namespace splay tree operations, \
              per-port lock contention, and cross-core scheduling nondeterminism.",
    category: SourceCategory::IPC,
    platform: Platform::MacOS,
    requirements: &[],
    entropy_rate_estimate: 2.0,
    composite: false,
    is_fast: true,
};

impl EntropySource for MachIPCSource {
    fn info(&self) -> &SourceInfo {
        &MACH_IPC_INFO
    }

    fn is_available(&self) -> bool {
        cfg!(target_os = "macos")
    }

    fn collect(&self, n_samples: usize) -> Vec<u8> {
        #[cfg(not(target_os = "macos"))]
        {
            let _ = n_samples;
            Vec::new()
        }

        #[cfg(target_os = "macos")]
        {
            let raw_count = n_samples * 4 + 64;
            let mut timings: Vec<u64> = Vec::with_capacity(raw_count);

            // SAFETY: mach_task_self() returns the current task port (always valid).
            let task = unsafe { mach_task_self() };
            let num_ports = self.config.num_ports.max(1);

            let mut ports: Vec<u32> = Vec::with_capacity(num_ports);
            for _ in 0..num_ports {
                let mut port: u32 = 0;
                // SAFETY: mach_port_allocate allocates a receive right.
                let kr = unsafe {
                    mach_port_allocate(task, 1 /* MACH_PORT_RIGHT_RECEIVE */, &mut port)
                };
                if kr == 0 {
                    // SAFETY: port is a valid receive right we just allocated.
                    let kr2 = unsafe {
                        mach_port_insert_right(
                            task, port, port, 20, /* MACH_MSG_TYPE_MAKE_SEND */
                        )
                    };
                    if kr2 == 0 {
                        ports.push(port);
                    } else {
                        unsafe {
                            mach_port_mod_refs(task, port, 1, -1);
                        }
                    }
                }
            }

            if ports.is_empty() {
                return self.collect_simple(n_samples);
            }

            if self.config.use_complex_messages {
                let ool_size = self.config.ool_size.max(1);
                let ool_buf = vec![0xBEu8; ool_size];

                let stop = Arc::new(AtomicBool::new(false));
                let stop2 = stop.clone();
                let recv_ports = ports.clone();
                let receiver = thread::spawn(move || {
                    let mut recv_buf = vec![0u8; 1024 + ool_size * 2];
                    // Safety net: receiver exits after 30s even if stop is never set
                    // (e.g. if the main thread panics between spawn and stop.store).
                    let deadline = std::time::Instant::now() + std::time::Duration::from_secs(30);
                    while !stop2.load(Ordering::Relaxed) && std::time::Instant::now() < deadline {
                        for &port in &recv_ports {
                            // SAFETY: recv_buf is large enough. Non-blocking receive.
                            unsafe {
                                let hdr = recv_buf.as_mut_ptr() as *mut MachMsgHeader;
                                (*hdr).msgh_local_port = port;
                                (*hdr).msgh_size = recv_buf.len() as u32;
                                let kr =
                                    mach_msg(hdr, 2 | 0x100, 0, recv_buf.len() as u32, port, 0, 0);
                                // If we received a complex message, the kernel mapped OOL
                                // data into our address space. Deallocate it to prevent
                                // VM memory leaks proportional to message count.
                                if kr == 0 && ((*hdr).msgh_bits & 0x80000000) != 0 {
                                    let recv_ool_ptr = recv_buf.as_ptr().add(
                                        std::mem::size_of::<MachMsgHeader>()
                                            + std::mem::size_of::<MachMsgBody>(),
                                    )
                                        as *const MachMsgOOLDescriptor;
                                    let recv_ool = std::ptr::read_unaligned(recv_ool_ptr);
                                    let addr = recv_ool.address as usize;
                                    let size = recv_ool.size as usize;
                                    if addr != 0 && size > 0 {
                                        vm_deallocate(mach_task_self(), addr, size);
                                    }
                                }
                            }
                        }
                        std::thread::yield_now();
                    }
                });

                for i in 0..raw_count {
                    let port = ports[i % ports.len()];

                    let mut msg = MachMsgOOL::zeroed();
                    msg.header.msgh_bits = 0x80000000 | 17; // COMPLEX | COPY_SEND
                    msg.header.msgh_size = std::mem::size_of::<MachMsgOOL>() as u32;
                    msg.header.msgh_remote_port = port;
                    msg.header.msgh_local_port = 0;
                    msg.header.msgh_id = i as i32;
                    msg.body.msgh_descriptor_count = 1;
                    msg.ool.address = ool_buf.as_ptr() as *mut _;
                    msg.ool.size = ool_size as u32;
                    msg.ool.deallocate = 0;
                    msg.ool.copy = 1; // MACH_MSG_VIRTUAL_COPY
                    msg.ool.ool_type = 1; // MACH_MSG_OOL_DESCRIPTOR

                    let t0 = mach_time();
                    // SAFETY: msg is properly initialized. MACH_SEND_TIMEOUT prevents blocking.
                    unsafe {
                        mach_msg(&mut msg.header, 1 | 0x80, msg.header.msgh_size, 0, 0, 10, 0);
                    }
                    let t1 = mach_time();
                    timings.push(t1.wrapping_sub(t0));
                }

                stop.store(true, Ordering::Relaxed);
                let _ = receiver.join();
            } else {
                for i in 0..raw_count {
                    let t0 = mach_time();
                    let base_port = ports[i % ports.len()];

                    let mut new_port: u32 = 0;
                    // SAFETY: standard Mach port operations.
                    let kr = unsafe { mach_port_allocate(task, 1, &mut new_port) };
                    if kr == 0 {
                        // Drop the receive right. mach_port_allocate with
                        // MACH_PORT_RIGHT_RECEIVE creates a receive right only;
                        // use mod_refs to release it (not mach_port_deallocate,
                        // which is for send/send-once rights).
                        unsafe {
                            mach_port_mod_refs(task, new_port, 1, -1);
                        }
                    }
                    unsafe {
                        let mut ptype: u32 = 0;
                        mach_port_type(task, base_port, &mut ptype);
                    }
                    let t1 = mach_time();
                    timings.push(t1.wrapping_sub(t0));
                }
            }

            for &port in &ports {
                unsafe {
                    // Release the send right created by mach_port_insert_right.
                    mach_port_mod_refs(task, port, 0 /* MACH_PORT_RIGHT_SEND */, -1);
                    // Release the receive right created by mach_port_allocate.
                    mach_port_mod_refs(task, port, 1 /* MACH_PORT_RIGHT_RECEIVE */, -1);
                }
            }

            extract_timing_entropy(&timings, n_samples)
        }
    }
}

#[cfg(target_os = "macos")]
impl MachIPCSource {
    fn collect_simple(&self, n_samples: usize) -> Vec<u8> {
        let raw_count = n_samples * 4 + 64;
        let mut timings: Vec<u64> = Vec::with_capacity(raw_count);
        let task = unsafe { mach_task_self() };

        for _ in 0..raw_count {
            let t0 = mach_time();
            let mut port: u32 = 0;
            let kr = unsafe { mach_port_allocate(task, 1, &mut port) };
            if kr == 0 {
                // Drop the receive right directly via mod_refs.
                // mach_port_deallocate is for send rights and would leave
                // the port name invalid before the subsequent mod_refs call.
                unsafe {
                    mach_port_mod_refs(task, port, 1, -1);
                }
            }
            let t1 = mach_time();
            timings.push(t1.wrapping_sub(t0));
        }
        extract_timing_entropy(&timings, n_samples)
    }
}

// Mach message structures for complex OOL messages.
#[cfg(target_os = "macos")]
#[repr(C)]
struct MachMsgHeader {
    msgh_bits: u32,
    msgh_size: u32,
    msgh_remote_port: u32,
    msgh_local_port: u32,
    msgh_voucher_port: u32,
    msgh_id: i32,
}

#[cfg(target_os = "macos")]
#[repr(C)]
struct MachMsgBody {
    msgh_descriptor_count: u32,
}

#[cfg(target_os = "macos")]
#[repr(C)]
struct MachMsgOOLDescriptor {
    address: *mut u8,
    deallocate: u8,
    copy: u8,
    _pad: u8,
    ool_type: u8, // mach_msg_descriptor_type_t — must be last in this group
    size: u32,
}

#[cfg(target_os = "macos")]
#[repr(C)]
struct MachMsgOOL {
    header: MachMsgHeader,
    body: MachMsgBody,
    ool: MachMsgOOLDescriptor,
}

// MachMsgOOL contains a raw pointer (ool.address) and is intentionally !Send.
// It is only used on the same thread that owns the pointed-to buffer.

#[cfg(target_os = "macos")]
impl MachMsgOOL {
    fn zeroed() -> Self {
        // SAFETY: All-zeros is valid for this repr(C) struct.
        unsafe { std::mem::zeroed() }
    }
}

#[cfg(target_os = "macos")]
unsafe extern "C" {
    fn mach_task_self() -> u32;
    fn mach_port_allocate(task: u32, right: i32, name: *mut u32) -> i32;
    fn mach_port_mod_refs(task: u32, name: u32, right: i32, delta: i32) -> i32;
    fn mach_port_insert_right(task: u32, name: u32, poly: u32, poly_poly: u32) -> i32;
    fn mach_port_type(task: u32, name: u32, ptype: *mut u32) -> i32;
    fn mach_msg(
        msg: *mut MachMsgHeader,
        option: i32,
        send_size: u32,
        rcv_size: u32,
        rcv_name: u32,
        timeout: u32,
        notify: u32,
    ) -> i32;
    fn vm_deallocate(target: u32, addr: usize, size: usize) -> i32;
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn info() {
        let src = MachIPCSource::default();
        assert_eq!(src.name(), "mach_ipc");
        assert_eq!(src.info().category, SourceCategory::IPC);
        assert!(!src.info().composite);
    }

    #[test]
    fn default_config() {
        let config = MachIPCConfig::default();
        assert_eq!(config.num_ports, 8);
        assert_eq!(config.ool_size, 4096);
        assert!(config.use_complex_messages);
    }

    #[test]
    fn custom_config() {
        let src = MachIPCSource {
            config: MachIPCConfig {
                num_ports: 4,
                ool_size: 8192,
                use_complex_messages: false,
            },
        };
        assert_eq!(src.config.num_ports, 4);
        assert!(!src.config.use_complex_messages);
    }

    #[test]
    #[ignore] // Uses Mach ports
    fn collects_bytes() {
        let src = MachIPCSource::default();
        assert!(src.is_available());
        let data = src.collect(64);
        assert!(!data.is_empty());
        assert!(data.len() <= 64);
    }

    #[test]
    #[ignore] // Uses Mach ports
    fn simple_mode_collects_bytes() {
        let src = MachIPCSource {
            config: MachIPCConfig {
                use_complex_messages: false,
                ..MachIPCConfig::default()
            },
        };
        assert!(!src.collect(64).is_empty());
    }
}