Merge branch 'master' of github.com:DeathWish5/zCore-Tutorial

code/ch03-04/Cargo.toml

@@ -0,0 +1,6 @@
+[workspace]
+members = [
+    "object",
+    "kernel-hal-unix",
+    "kernel-hal",
+]

code/ch03-04/kernel-hal-unix/Cargo.toml

@@ -0,0 +1,18 @@
+[package]
+name = "kernel-hal-unix"
+version = "0.1.0"
+authors = ["Runji Wang <wangrunji0408@163.com>"]
+edition = "2018"
+description = "Kernel HAL implementation on Linux and macOS."
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+log = "0.4"
+libc = "0.2"
+tempfile = "3"
+bitflags = "1.2"
+lazy_static = "1.4"
+kernel-hal = { path = "../kernel-hal" }
+async-std = "1.9"
+trapframe = "0.8.0"

code/ch03-04/kernel-hal-unix/src/lib.rs

@@ -0,0 +1,302 @@
+#![feature(asm)]
+#![feature(linkage)]
+#![deny(warnings)]
+
+extern crate alloc;
+#[macro_use]
+extern crate log;
+
+use {
+    alloc::boxed::Box,
+    alloc::collections::VecDeque,
+    core::time::Duration,
+    core::{future::Future, pin::Pin},
+    lazy_static::*,
+    std::fmt::{Debug, Formatter},
+    std::fs::{File, OpenOptions},
+    std::io::Error,
+    std::os::unix::io::AsRawFd,
+    std::sync::Mutex,
+    std::time::SystemTime,
+    tempfile::tempdir,
+};
+
+pub use kernel_hal::{defs::*, *};
+
+#[repr(C)]
+pub struct Thread {
+    thread: usize,
+}
+
+impl Thread {
+    #[export_name = "hal_thread_spawn"]
+    pub fn spawn(
+        future: Pin<Box<dyn Future<Output = ()> + Send + 'static>>,
+        _vmtoken: usize,
+    ) -> Self {
+        async_std::task::spawn(future);
+        Thread { thread: 0 }
+    }
+}
+
+/// Get current time.
+#[export_name = "hal_timer_now"]
+pub fn timer_now() -> Duration {
+    SystemTime::now()
+        .duration_since(SystemTime::UNIX_EPOCH)
+        .unwrap()
+}
+
+/// Initialize the HAL.
+///
+/// This function must be called at the beginning.
+pub fn init() {
+    #[cfg(target_os = "macos")]
+    unimplemented!()
+}
+
+#[repr(C)]
+pub struct PhysFrame {
+    paddr: PhysAddr,
+}
+
+impl Debug for PhysFrame {
+    fn fmt(&self, f: &mut Formatter<'_>) -> core::result::Result<(), std::fmt::Error> {
+        write!(f, "PhysFrame({:#x})", self.paddr)
+    }
+}
+
+const PMEM_SIZE: usize = 0x4000_0000; // 1GiB
+const PAGE_SIZE: usize = 0x1000;
+fn page_aligned(x: VirtAddr) -> bool {
+    x % PAGE_SIZE == 0
+}
+
+lazy_static! {
+    static ref FRAME_FILE: File = create_pmem_file();
+}
+
+fn create_pmem_file() -> File {
+    let dir = tempdir().expect("failed to create pmem dir");
+    let path = dir.path().join("pmem");
+
+    // workaround on macOS to avoid permission denied.
+    // see https://jiege.ch/software/2020/02/07/macos-mmap-exec/ for analysis on this problem.
+    #[cfg(target_os = "macos")]
+    std::mem::forget(dir);
+
+    let file = OpenOptions::new()
+        .read(true)
+        .write(true)
+        .create(true)
+        .open(&path)
+        .expect("failed to create pmem file");
+    file.set_len(PMEM_SIZE as u64)
+        .expect("failed to resize file");
+    trace!("create pmem file: path={:?}, size={:#x}", path, PMEM_SIZE);
+    let prot = libc::PROT_READ | libc::PROT_WRITE;
+    mmap(file.as_raw_fd(), 0, PMEM_SIZE, phys_to_virt(0), prot);
+    file
+}
+
+/// Mmap frame file `fd` to `vaddr`.
+fn mmap(fd: libc::c_int, offset: usize, len: usize, vaddr: VirtAddr, prot: libc::c_int) {
+    // workaround on macOS to write text section.
+    #[cfg(target_os = "macos")]
+    let prot = if prot & libc::PROT_EXEC != 0 {
+        prot | libc::PROT_WRITE
+    } else {
+        prot
+    };
+
+    let ret = unsafe {
+        let flags = libc::MAP_SHARED | libc::MAP_FIXED;
+        libc::mmap(vaddr as _, len, prot, flags, fd, offset as _)
+    } as usize;
+    println!(
+        "mmap file: fd={}, offset={:#x}, len={:#x}, vaddr={:#x}, prot={:#b}",
+        fd, offset, len, vaddr, prot,
+    );
+    assert_eq!(ret, vaddr, "failed to mmap: {:?}", Error::last_os_error());
+}
+
+lazy_static! {
+    static ref AVAILABLE_FRAMES: Mutex<VecDeque<usize>> =
+        Mutex::new((PAGE_SIZE..PMEM_SIZE).step_by(PAGE_SIZE).collect());
+}
+
+impl PhysFrame {
+    #[export_name = "hal_frame_alloc"]
+    pub fn alloc() -> Option<Self> {
+        let ret = AVAILABLE_FRAMES
+            .lock()
+            .unwrap()
+            .pop_front()
+            .map(|paddr| PhysFrame { paddr });
+        trace!("frame alloc: {:?}", ret);
+        ret
+    }
+    #[export_name = "hal_zero_frame_paddr"]
+    pub fn zero_frame_addr() -> PhysAddr {
+        0
+    }
+}
+
+impl Drop for PhysFrame {
+    #[export_name = "hal_frame_dealloc"]
+    fn drop(&mut self) {
+        trace!("frame dealloc: {:?}", self);
+        AVAILABLE_FRAMES.lock().unwrap().push_back(self.paddr);
+    }
+}
+
+fn phys_to_virt(paddr: PhysAddr) -> VirtAddr {
+    /// Map physical memory from here.
+    const PMEM_BASE: VirtAddr = 0x8_0000_0000;
+
+    PMEM_BASE + paddr
+}
+
+/// Ensure physical memory are mmapped and accessible.
+fn ensure_mmap_pmem() {
+    FRAME_FILE.as_raw_fd();
+}
+
+/// Read physical memory from `paddr` to `buf`.
+#[export_name = "hal_pmem_read"]
+pub fn pmem_read(paddr: PhysAddr, buf: &mut [u8]) {
+    trace!("pmem read: paddr={:#x}, len={:#x}", paddr, buf.len());
+    assert!(paddr + buf.len() <= PMEM_SIZE);
+    ensure_mmap_pmem();
+    unsafe {
+        (phys_to_virt(paddr) as *const u8).copy_to_nonoverlapping(buf.as_mut_ptr(), buf.len());
+    }
+}
+
+/// Write physical memory to `paddr` from `buf`.
+#[export_name = "hal_pmem_write"]
+pub fn pmem_write(paddr: PhysAddr, buf: &[u8]) {
+    trace!("pmem write: paddr={:#x}, len={:#x}", paddr, buf.len());
+    assert!(paddr + buf.len() <= PMEM_SIZE);
+    ensure_mmap_pmem();
+    unsafe {
+        buf.as_ptr()
+            .copy_to_nonoverlapping(phys_to_virt(paddr) as _, buf.len());
+    }
+}
+
+/// Zero physical memory at `[paddr, paddr + len)`
+#[export_name = "hal_pmem_zero"]
+pub fn pmem_zero(paddr: PhysAddr, len: usize) {
+    trace!("pmem_zero: addr={:#x}, len={:#x}", paddr, len);
+    assert!(paddr + len <= PMEM_SIZE);
+    ensure_mmap_pmem();
+    unsafe {
+        core::ptr::write_bytes(phys_to_virt(paddr) as *mut u8, 0, len);
+    }
+}
+
+/// Copy content of `src` frame to `target` frame
+#[export_name = "hal_frame_copy"]
+pub fn frame_copy(src: PhysAddr, target: PhysAddr) {
+    trace!("frame_copy: {:#x} <- {:#x}", target, src);
+    assert!(src + PAGE_SIZE <= PMEM_SIZE && target + PAGE_SIZE <= PMEM_SIZE);
+    ensure_mmap_pmem();
+    unsafe {
+        let buf = phys_to_virt(src) as *const u8;
+        buf.copy_to_nonoverlapping(phys_to_virt(target) as _, PAGE_SIZE);
+    }
+}
+
+/// Flush the physical frame.
+#[export_name = "hal_frame_flush"]
+pub fn frame_flush(_target: PhysAddr) {
+    // do nothing
+}
+
+/// Page Table
+#[repr(C)]
+pub struct PageTable {
+    table_phys: PhysAddr,
+}
+
+impl PageTable {
+    /// Create a new `PageTable`.
+    #[allow(clippy::new_without_default)]
+    #[export_name = "hal_pt_new"]
+    pub fn new() -> Self {
+        PageTable { table_phys: 0 }
+    }
+}
+
+impl PageTableTrait for PageTable {
+    /// Map the page of `vaddr` to the frame of `paddr` with `flags`.
+    #[export_name = "hal_pt_map"]
+    fn map(&mut self, vaddr: VirtAddr, paddr: PhysAddr, flags: MMUFlags) -> Result<()> {
+        debug_assert!(page_aligned(vaddr));
+        debug_assert!(page_aligned(paddr));
+        let prot = flags.to_mmap_prot();
+        mmap(FRAME_FILE.as_raw_fd(), paddr, PAGE_SIZE, vaddr, prot);
+        Ok(())
+    }
+
+    /// Unmap the page of `vaddr`.
+    #[export_name = "hal_pt_unmap"]
+    fn unmap(&mut self, vaddr: VirtAddr) -> Result<()> {
+        self.unmap_cont(vaddr, 1)
+    }
+
+    /// Change the `flags` of the page of `vaddr`.
+    #[export_name = "hal_pt_protect"]
+    fn protect(&mut self, vaddr: VirtAddr, flags: MMUFlags) -> Result<()> {
+        debug_assert!(page_aligned(vaddr));
+        let prot = flags.to_mmap_prot();
+        let ret = unsafe { libc::mprotect(vaddr as _, PAGE_SIZE, prot) };
+        assert_eq!(ret, 0, "failed to mprotect: {:?}", Error::last_os_error());
+        Ok(())
+    }
+
+    /// Query the physical address which the page of `vaddr` maps to.
+    #[export_name = "hal_pt_query"]
+    fn query(&mut self, vaddr: VirtAddr) -> Result<PhysAddr> {
+        debug_assert!(page_aligned(vaddr));
+        unimplemented!()
+    }
+
+    /// Get the physical address of root page table.
+    #[export_name = "hal_pt_table_phys"]
+    fn table_phys(&self) -> PhysAddr {
+        self.table_phys
+    }
+
+    #[export_name = "hal_pt_unmap_cont"]
+    fn unmap_cont(&mut self, vaddr: VirtAddr, pages: usize) -> Result<()> {
+        if pages == 0 {
+            return Ok(());
+        }
+        debug_assert!(page_aligned(vaddr));
+        let ret = unsafe { libc::munmap(vaddr as _, PAGE_SIZE * pages) };
+        assert_eq!(ret, 0, "failed to munmap: {:?}", Error::last_os_error());
+        Ok(())
+    }
+}
+
+trait FlagsExt {
+    fn to_mmap_prot(&self) -> libc::c_int;
+}
+
+impl FlagsExt for MMUFlags {
+    fn to_mmap_prot(&self) -> libc::c_int {
+        let mut flags = 0;
+        if self.contains(MMUFlags::READ) {
+            flags |= libc::PROT_READ;
+        }
+        if self.contains(MMUFlags::WRITE) {
+            flags |= libc::PROT_WRITE;
+        }
+        if self.contains(MMUFlags::EXECUTE) {
+            flags |= libc::PROT_EXEC;
+        }
+        flags
+    }
+}

code/ch03-04/kernel-hal/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "kernel-hal"
+version = "0.1.0"
+authors = ["Runji Wang <wangrunji0408@163.com>"]
+edition = "2018"
+description = "Kernel HAL interface definations."
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+bitflags = "1.2"
+trapframe = "0.8.0"
+numeric-enum-macro = "0.2"

code/ch03-04/kernel-hal/src/lib.rs

@@ -0,0 +1,311 @@
+#![no_std]
+#![feature(linkage)]
+#![deny(warnings)]
+
+extern crate alloc;
+
+pub use trapframe::{GeneralRegs, UserContext};
+
+use {
+    alloc::{boxed::Box, vec::Vec},
+    core::{future::Future, pin::Pin, time::Duration},
+};
+
+#[derive(Debug)]
+pub struct HalError;
+/// The result type returned by HAL functions.
+pub type Result<T> = core::result::Result<T, HalError>;
+
+pub mod defs {
+    use bitflags::bitflags;
+    use numeric_enum_macro::numeric_enum;
+
+    bitflags! {
+        pub struct MMUFlags: usize {
+            #[allow(clippy::identity_op)]
+            const CACHE_1   = 1 << 0;
+            const CACHE_2   = 1 << 1;
+            const READ      = 1 << 2;
+            const WRITE     = 1 << 3;
+            const EXECUTE   = 1 << 4;
+            const USER      = 1 << 5;
+            const RXW = Self::READ.bits | Self::WRITE.bits | Self::EXECUTE.bits;
+        }
+    }
+    numeric_enum! {
+        #[repr(u32)]
+        #[derive(Debug, PartialEq, Clone, Copy)]
+        pub enum CachePolicy {
+            Cached = 0,
+            Uncached = 1,
+            UncachedDevice = 2,
+            WriteCombining = 3,
+        }
+    }
+    pub const CACHE_POLICY_MASK: u32 = 3;
+
+    impl Default for CachePolicy {
+        fn default() -> Self {
+            Self::Cached
+        }
+    }
+
+    pub type PhysAddr = usize;
+    pub type VirtAddr = usize;
+    pub type DevVAddr = usize;
+    pub const PAGE_SIZE: usize = 0x1000;
+}
+
+pub use self::defs::*;
+
+#[repr(C)]
+pub struct Thread {
+    id: usize,
+}
+
+impl Thread {
+    /// Spawn a new thread.
+    #[linkage = "weak"]
+    #[export_name = "hal_thread_spawn"]
+    pub fn spawn(
+        _future: Pin<Box<dyn Future<Output = ()> + Send + 'static>>,
+        _vmtoken: usize,
+    ) -> Self {
+        unimplemented!()
+    }
+}
+
+#[linkage = "weak"]
+#[export_name = "hal_timer_now"]
+pub fn timer_now() -> Duration {
+    unimplemented!()
+}
+
+#[repr(C)]
+pub struct PhysFrame {
+    paddr: PhysAddr,
+}
+
+impl PhysFrame {
+    #[linkage = "weak"]
+    #[export_name = "hal_frame_alloc"]
+    pub fn alloc() -> Option<Self> {
+        unimplemented!()
+    }
+
+    #[linkage = "weak"]
+    #[export_name = "hal_frame_alloc_contiguous"]
+    pub fn alloc_contiguous_base(_size: usize, _align_log2: usize) -> Option<PhysAddr> {
+        unimplemented!()
+    }
+
+    pub fn alloc_contiguous(size: usize, align_log2: usize) -> Vec<Self> {
+        PhysFrame::alloc_contiguous_base(size, align_log2).map_or(Vec::new(), |base| {
+            (0..size)
+                .map(|i| PhysFrame {
+                    paddr: base + i * PAGE_SIZE,
+                })
+                .collect()
+        })
+    }
+
+    pub fn alloc_zeroed() -> Option<Self> {
+        Self::alloc().map(|f| {
+            pmem_zero(f.addr(), PAGE_SIZE);
+            f
+        })
+    }
+
+    pub fn alloc_contiguous_zeroed(size: usize, align_log2: usize) -> Vec<Self> {
+        PhysFrame::alloc_contiguous_base(size, align_log2).map_or(Vec::new(), |base| {
+            pmem_zero(base, size * PAGE_SIZE);
+            (0..size)
+                .map(|i| PhysFrame {
+                    paddr: base + i * PAGE_SIZE,
+                })
+                .collect()
+        })
+    }
+
+    pub fn addr(&self) -> PhysAddr {
+        self.paddr
+    }
+
+    #[linkage = "weak"]
+    #[export_name = "hal_zero_frame_paddr"]
+    pub fn zero_frame_addr() -> PhysAddr {
+        unimplemented!()
+    }
+}
+
+impl Drop for PhysFrame {
+    #[linkage = "weak"]
+    #[export_name = "hal_frame_dealloc"]
+    fn drop(&mut self) {
+        unimplemented!()
+    }
+}
+
+/// Read physical memory from `paddr` to `buf`.
+#[linkage = "weak"]
+#[export_name = "hal_pmem_read"]
+pub fn pmem_read(_paddr: PhysAddr, _buf: &mut [u8]) {
+    unimplemented!()
+}
+
+/// Write physical memory to `paddr` from `buf`.
+#[linkage = "weak"]
+#[export_name = "hal_pmem_write"]
+pub fn pmem_write(_paddr: PhysAddr, _buf: &[u8]) {
+    unimplemented!()
+}
+
+/// Zero physical memory at `[paddr, paddr + len)`
+#[linkage = "weak"]
+#[export_name = "hal_pmem_zero"]
+pub fn pmem_zero(_paddr: PhysAddr, _len: usize) {
+    unimplemented!()
+}
+
+/// Copy content of `src` frame to `target` frame.
+#[linkage = "weak"]
+#[export_name = "hal_frame_copy"]
+pub fn frame_copy(_src: PhysAddr, _target: PhysAddr) {
+    unimplemented!()
+}
+
+/// Flush the physical frame.
+#[linkage = "weak"]
+#[export_name = "hal_frame_flush"]
+pub fn frame_flush(_target: PhysAddr) {
+    unimplemented!()
+}
+
+pub trait PageTableTrait: Sync + Send {
+    /// Map the page of `vaddr` to the frame of `paddr` with `flags`.
+    fn map(&mut self, _vaddr: VirtAddr, _paddr: PhysAddr, _flags: MMUFlags) -> Result<()>;
+
+    /// Unmap the page of `vaddr`.
+    fn unmap(&mut self, _vaddr: VirtAddr) -> Result<()>;
+
+    /// Change the `flags` of the page of `vaddr`.
+    fn protect(&mut self, _vaddr: VirtAddr, _flags: MMUFlags) -> Result<()>;
+
+    /// Query the physical address which the page of `vaddr` maps to.
+    fn query(&mut self, _vaddr: VirtAddr) -> Result<PhysAddr>;
+
+    /// Get the physical address of root page table.
+    fn table_phys(&self) -> PhysAddr;
+
+    #[cfg(target_arch = "riscv64")]
+    /// Activate this page table
+    fn activate(&self);
+
+    fn map_many(
+        &mut self,
+        mut vaddr: VirtAddr,
+        paddrs: &[PhysAddr],
+        flags: MMUFlags,
+    ) -> Result<()> {
+        for &paddr in paddrs {
+            self.map(vaddr, paddr, flags)?;
+            vaddr += PAGE_SIZE;
+        }
+        Ok(())
+    }
+
+    fn map_cont(
+        &mut self,
+        mut vaddr: VirtAddr,
+        paddr: PhysAddr,
+        pages: usize,
+        flags: MMUFlags,
+    ) -> Result<()> {
+        for i in 0..pages {
+            let paddr = paddr + i * PAGE_SIZE;
+            self.map(vaddr, paddr, flags)?;
+            vaddr += PAGE_SIZE;
+        }
+        Ok(())
+    }
+
+    fn unmap_cont(&mut self, vaddr: VirtAddr, pages: usize) -> Result<()> {
+        for i in 0..pages {
+            self.unmap(vaddr + i * PAGE_SIZE)?;
+        }
+        Ok(())
+    }
+}
+
+/// Page Table
+#[repr(C)]
+pub struct PageTable {
+    table_phys: PhysAddr,
+}
+
+impl PageTable {
+    /// Get current page table
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_current"]
+    pub fn current() -> Self {
+        unimplemented!()
+    }
+
+    /// Create a new `PageTable`.
+    #[allow(clippy::new_without_default)]
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_new"]
+    pub fn new() -> Self {
+        unimplemented!()
+    }
+}
+
+impl PageTableTrait for PageTable {
+    /// Map the page of `vaddr` to the frame of `paddr` with `flags`.
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_map"]
+    fn map(&mut self, _vaddr: VirtAddr, _paddr: PhysAddr, _flags: MMUFlags) -> Result<()> {
+        unimplemented!()
+    }
+    /// Unmap the page of `vaddr`.
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_unmap"]
+    fn unmap(&mut self, _vaddr: VirtAddr) -> Result<()> {
+        unimplemented!()
+    }
+    /// Change the `flags` of the page of `vaddr`.
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_protect"]
+    fn protect(&mut self, _vaddr: VirtAddr, _flags: MMUFlags) -> Result<()> {
+        unimplemented!()
+    }
+    /// Query the physical address which the page of `vaddr` maps to.
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_query"]
+    fn query(&mut self, _vaddr: VirtAddr) -> Result<PhysAddr> {
+        unimplemented!()
+    }
+    /// Get the physical address of root page table.
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_table_phys"]
+    fn table_phys(&self) -> PhysAddr {
+        self.table_phys
+    }
+
+    /// Activate this page table
+    #[cfg(target_arch = "riscv64")]
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_activate"]
+    fn activate(&self) {
+        unimplemented!()
+    }
+
+    #[linkage = "weak"]
+    #[export_name = "hal_pt_unmap_cont"]
+    fn unmap_cont(&mut self, vaddr: VirtAddr, pages: usize) -> Result<()> {
+        for i in 0..pages {
+            self.unmap(vaddr + i * PAGE_SIZE)?;
+        }
+        Ok(())
+    }
+}

code/ch03-04/object/Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "ch02-03"
+version = "0.1.0"
+authors = ["Runji Wang <wangrunji0408@163.com>"]
+edition = "2018"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+log = "0.4"
+spin = "0.7"
+downcast-rs = { version = "1.2.0", default-features = false }
+bitflags = "1.2"
+hashbrown = "0.9"
+trapframe = "0.8.0"
+futures = { version = "0.3", default-features = false, features = ["alloc", "async-await"] }
+async-std = { version = "1.9", features = ["attributes", "unstable"] }
+numeric-enum-macro = "0.2"
+kernel-hal = { path = "../kernel-hal" }
+kernel-hal-unix = { path = "../kernel-hal-unix" }

code/ch03-04/object/src/error.rs

@@ -0,0 +1,232 @@
+// ANCHOR: result
+///
+pub type ZxResult<T = ()> = Result<T, ZxError>;
+// ANCHOR_END: result
+
+// ANCHOR: error_begin
+/// Zircon statuses are signed 32 bit integers. The space of values is
+/// divided as follows:
+/// - The zero value is for the OK status.
+/// - Negative values are defined by the system, in this file.
+/// - Positive values are reserved for protocol-specific error values,
+///   and will never be defined by the system.
+#[allow(non_camel_case_types, dead_code)]
+#[repr(i32)]
+#[derive(Debug, Clone, Copy, Eq, PartialEq)]
+pub enum ZxError {
+    OK = 0,
+
+    // ======= Internal failures =======
+    /// The system encountered an otherwise unspecified error
+    /// while performing the operation.
+    INTERNAL = -1,
+
+    /// The operation is not implemented, supported,
+    /// or enabled.
+    NOT_SUPPORTED = -2,
+    // ANCHOR_END: error_begin
+    /// The system was not able to allocate some resource
+    /// needed for the operation.
+    NO_RESOURCES = -3,
+
+    /// The system was not able to allocate memory needed
+    /// for the operation.
+    NO_MEMORY = -4,
+
+    // -5 used to be ZX_ERR_CALL_FAILED.
+    /// The system call was interrupted, but should be
+    /// retried.  This should not be seen outside of the VDSO.
+    INTERNAL_INTR_RETRY = -6,
+
+    // ======= Parameter errors =======
+    /// an argument is invalid, ex. null pointer
+    INVALID_ARGS = -10,
+
+    /// A specified handle value does not refer to a handle.
+    BAD_HANDLE = -11,
+
+    /// The subject of the operation is the wrong type to
+    /// perform the operation.
+    /// Example: Attempting a message_read on a thread handle.
+    WRONG_TYPE = -12,
+
+    /// The specified syscall number is invalid.
+    BAD_SYSCALL = -13,
+
+    /// An argument is outside the valid range for this
+    /// operation.
+    OUT_OF_RANGE = -14,
+
+    /// A caller provided buffer is too small for
+    /// this operation.
+    BUFFER_TOO_SMALL = -15,
+
+    // ======= Precondition or state errors =======
+    /// operation failed because the current state of the
+    /// object does not allow it, or a precondition of the operation is
+    /// not satisfied
+    BAD_STATE = -20,
+
+    /// The time limit for the operation elapsed before
+    /// the operation completed.
+    TIMED_OUT = -21,
+
+    /// The operation cannot be performed currently but
+    /// potentially could succeed if the caller waits for a prerequisite
+    /// to be satisfied, for example waiting for a handle to be readable
+    /// or writable.
+    /// Example: Attempting to read from a channel that has no
+    /// messages waiting but has an open remote will return ZX_ERR_SHOULD_WAIT.
+    /// Attempting to read from a channel that has no messages waiting
+    /// and has a closed remote end will return ZX_ERR_PEER_CLOSED.
+    SHOULD_WAIT = -22,
+
+    /// The in-progress operation (e.g. a wait) has been
+    /// canceled.
+    CANCELED = -23,
+
+    /// The operation failed because the remote end of the
+    /// subject of the operation was closed.
+    PEER_CLOSED = -24,
+
+    /// The requested entity is not found.
+    NOT_FOUND = -25,
+
+    /// An object with the specified identifier
+    /// already exists.
+    /// Example: Attempting to create a file when a file already exists
+    /// with that name.
+    ALREADY_EXISTS = -26,
+
+    /// The operation failed because the named entity
+    /// is already owned or controlled by another entity. The operation
+    /// could succeed later if the current owner releases the entity.
+    ALREADY_BOUND = -27,
+
+    /// The subject of the operation is currently unable
+    /// to perform the operation.
+    /// Note: This is used when there's no direct way for the caller to
+    /// observe when the subject will be able to perform the operation
+    /// and should thus retry.
+    UNAVAILABLE = -28,
+
+    // ======= Permission check errors =======
+    /// The caller did not have permission to perform
+    /// the specified operation.
+    ACCESS_DENIED = -30,
+
+    // ======= Input-output errors =======
+    /// Otherwise unspecified error occurred during I/O.
+    IO = -40,
+
+    /// The entity the I/O operation is being performed on
+    /// rejected the operation.
+    /// Example: an I2C device NAK'ing a transaction or a disk controller
+    /// rejecting an invalid command, or a stalled USB endpoint.
+    IO_REFUSED = -41,
+
+    /// The data in the operation failed an integrity
+    /// check and is possibly corrupted.
+    /// Example: CRC or Parity error.
+    IO_DATA_INTEGRITY = -42,
+
+    /// The data in the operation is currently unavailable
+    /// and may be permanently lost.
+    /// Example: A disk block is irrecoverably damaged.
+    IO_DATA_LOSS = -43,
+
+    /// The device is no longer available (has been
+    /// unplugged from the system, powered down, or the driver has been
+    /// unloaded,
+    IO_NOT_PRESENT = -44,
+
+    /// More data was received from the device than expected.
+    /// Example: a USB "babble" error due to a device sending more data than
+    /// the host queued to receive.
+    IO_OVERRUN = -45,
+
+    /// An operation did not complete within the required timeframe.
+    /// Example: A USB isochronous transfer that failed to complete due to an overrun or underrun.
+    IO_MISSED_DEADLINE = -46,
+
+    /// The data in the operation is invalid parameter or is out of range.
+    /// Example: A USB transfer that failed to complete with TRB Error
+    IO_INVALID = -47,
+
+    // ======== Filesystem Errors ========
+    /// Path name is too long.
+    BAD_PATH = -50,
+
+    /// Object is not a directory or does not support
+    /// directory operations.
+    /// Example: Attempted to open a file as a directory or
+    /// attempted to do directory operations on a file.
+    NOT_DIR = -51,
+
+    /// Object is not a regular file.
+    NOT_FILE = -52,
+
+    /// This operation would cause a file to exceed a
+    /// filesystem-specific size limit
+    FILE_BIG = -53,
+
+    /// Filesystem or device space is exhausted.
+    NO_SPACE = -54,
+
+    /// Directory is not empty.
+    NOT_EMPTY = -55,
+
+    // ======== Flow Control ========
+    // These are not errors, as such, and will never be returned
+    // by a syscall or public API.  They exist to allow callbacks
+    // to request changes in operation.
+    /// Do not call again.
+    /// Example: A notification callback will be called on every
+    /// event until it returns something other than ZX_OK.
+    /// This status allows differentiation between "stop due to
+    /// an error" and "stop because the work is done."
+    STOP = -60,
+
+    /// Advance to the next item.
+    /// Example: A notification callback will use this response
+    /// to indicate it did not "consume" an item passed to it,
+    /// but by choice, not due to an error condition.
+    NEXT = -61,
+
+    /// Ownership of the item has moved to an asynchronous worker.
+    ///
+    /// Unlike ZX_ERR_STOP, which implies that iteration on an object
+    /// should stop, and ZX_ERR_NEXT, which implies that iteration
+    /// should continue to the next item, ZX_ERR_ASYNC implies
+    /// that an asynchronous worker is responsible for continuing iteration.
+    ///
+    /// Example: A notification callback will be called on every
+    /// event, but one event needs to handle some work asynchronously
+    /// before it can continue. ZX_ERR_ASYNC implies the worker is
+    /// responsible for resuming iteration once its work has completed.
+    ASYNC = -62,
+
+    // ======== Network-related errors ========
+    /// Specified protocol is not
+    /// supported.
+    PROTOCOL_NOT_SUPPORTED = -70,
+
+    /// Host is unreachable.
+    ADDRESS_UNREACHABLE = -71,
+
+    /// Address is being used by someone else.
+    ADDRESS_IN_USE = -72,
+
+    /// Socket is not connected.
+    NOT_CONNECTED = -73,
+
+    /// Remote peer rejected the connection.
+    CONNECTION_REFUSED = -74,
+
+    /// Connection was reset.
+    CONNECTION_RESET = -75,
+    // ANCHOR: error_end
+    /// Connection was aborted.
+    CONNECTION_ABORTED = -76,
+}
+// ANCHOR_END: error_end

code/ch03-04/object/src/ipc/channel.rs

@@ -0,0 +1,176 @@
+use {
+    super::*,
+    crate::error::*,
+    crate::object::*,
+    alloc::collections::VecDeque,
+    alloc::sync::{Arc, Weak},
+    alloc::vec::Vec,
+    core::sync::atomic::{AtomicU32, Ordering},
+    spin::Mutex,
+};
+
+pub struct Channel {
+    base: KObjectBase,
+    peer: Weak<Channel>,
+    recv_queue: Mutex<VecDeque<T>>,
+    next_txid: AtomicU32,
+}
+
+type T = MessagePacket;
+type TxID = u32;
+
+impl_kobject!(Channel
+    fn peer(&self) -> ZxResult<Arc<dyn KernelObject>> {
+        let peer = self.peer.upgrade().ok_or(ZxError::PEER_CLOSED)?;
+        Ok(peer)
+    }
+    fn related_koid(&self) -> KoID {
+        self.peer.upgrade().map(|p| p.id()).unwrap_or(0)
+    }
+);
+
+impl Channel {
+    /// Create a channel and return a pair of its endpoints
+    #[allow(unsafe_code)]
+    pub fn create() -> (Arc<Self>, Arc<Self>) {
+        let mut channel0 = Arc::new(Channel {
+            base: KObjectBase::default(),
+            peer: Weak::default(),
+            recv_queue: Default::default(),
+            next_txid: AtomicU32::new(0x8000_0000),
+        });
+        let channel1 = Arc::new(Channel {
+            base: KObjectBase::default(),
+            peer: Arc::downgrade(&channel0),
+            recv_queue: Default::default(),
+            next_txid: AtomicU32::new(0x8000_0000),
+        });
+        // no other reference of `channel0`
+        unsafe {
+            Arc::get_mut_unchecked(&mut channel0).peer = Arc::downgrade(&channel1);
+        }
+        (channel0, channel1)
+    }
+
+    /// Read a packet from the channel if check is ok, otherwise the msg will keep.
+    pub fn read(&self) -> ZxResult<T> {
+        let mut recv_queue = self.recv_queue.lock();
+        if let Some(_) = recv_queue.front() {
+            let msg = recv_queue.pop_front().unwrap();
+            return Ok(msg);
+        }
+        if self.peer_closed() {
+            Err(ZxError::PEER_CLOSED)
+        } else {
+            Err(ZxError::SHOULD_WAIT)
+        }
+    }
+
+    /// Write a packet to the channel
+    pub fn write(&self, msg: T) -> ZxResult {
+        let peer = self.peer.upgrade().ok_or(ZxError::PEER_CLOSED)?;
+        peer.push_general(msg);
+        Ok(())
+    }
+
+    /// Push a message to general queue, called from peer.
+    fn push_general(&self, msg: T) {
+        let mut send_queue = self.recv_queue.lock();
+        send_queue.push_back(msg);
+    }
+
+    /// Generate a new transaction ID for `call`.
+    fn new_txid(&self) -> TxID {
+        self.next_txid.fetch_add(1, Ordering::SeqCst)
+    }
+
+    /// Is peer channel closed?
+    fn peer_closed(&self) -> bool {
+        self.peer.strong_count() == 0
+    }
+}
+
+/// The message transferred in the channel.
+/// See [Channel](struct.Channel.html) for details.
+#[derive(Default)]
+pub struct MessagePacket {
+    /// The transition id of the message packet
+    pub txid: TxID,
+    /// The data carried by the message packet
+    pub data: Vec<u8>,
+    /// See [Channel](struct.Channel.html) for details.
+    pub handles: Vec<Handle>,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_basics() {
+        let (end0, end1) = Channel::create();
+        assert!(Arc::ptr_eq(
+            &end0.peer().unwrap().downcast_arc().unwrap(),
+            &end1
+        ));
+        assert_eq!(end0.related_koid(), end1.id());
+
+        drop(end1);
+        assert_eq!(end0.peer().unwrap_err(), ZxError::PEER_CLOSED);
+        assert_eq!(end0.related_koid(), 0);
+    }
+
+    #[test]
+    fn read_write() {
+        let (channel0, channel1) = Channel::create();
+        // write a message to each other
+        let txid0 = channel0.new_txid();
+        channel0
+            .write(MessagePacket {
+                txid: txid0,
+                data: Vec::from("hello 1"),
+                handles: Vec::new(),
+            })
+            .unwrap();
+        let txid1 = channel1.new_txid();
+        channel1
+            .write(MessagePacket {
+                txid: txid1,
+                data: Vec::from("hello 0"),
+                handles: Vec::new(),
+            })
+            .unwrap();
+
+        // read message should success
+        let recv_msg = channel1.read().unwrap();
+        assert_eq!(recv_msg.txid, txid0);
+        assert_eq!(recv_msg.data.as_slice(), b"hello 1");
+        assert!(recv_msg.handles.is_empty());
+
+        let recv_msg = channel0.read().unwrap();
+        assert_eq!(recv_msg.txid, txid1);
+        assert_eq!(recv_msg.data.as_slice(), b"hello 0");
+        assert!(recv_msg.handles.is_empty());
+
+        // read more message should fail.
+        assert_eq!(channel0.read().err(), Some(ZxError::SHOULD_WAIT));
+        assert_eq!(channel1.read().err(), Some(ZxError::SHOULD_WAIT));
+    }
+
+    #[test]
+    fn peer_closed() {
+        let (channel0, channel1) = Channel::create();
+        // write a message from peer, then drop it
+        channel1.write(MessagePacket::default()).unwrap();
+        drop(channel1);
+        // read the first message should success.
+        channel0.read().unwrap();
+        // read more message should fail.
+        assert_eq!(channel0.read().err(), Some(ZxError::PEER_CLOSED));
+        // write message should fail.
+        assert_eq!(
+            channel0.write(MessagePacket::default()),
+            Err(ZxError::PEER_CLOSED)
+        );
+    }
+}

code/ch03-04/object/src/ipc/mod.rs

@@ -0,0 +1,4 @@
+use super::*;
+
+mod channel;
+pub use self::channel::*;

code/ch03-04/object/src/lib.rs

@@ -0,0 +1,23 @@
+#![no_std]
+#![deny(unused_imports)]
+#![allow(dead_code)]
+#![feature(get_mut_unchecked)]
+#![feature(drain_filter)]
+
+extern crate alloc;
+
+#[cfg(test)]
+#[macro_use]
+extern crate std;
+
+#[macro_use]
+extern crate log;
+
+mod error;
+mod ipc;
+mod object;
+mod task;
+mod util;
+mod vm;
+
+pub use self::error::*;

code/ch03-04/object/src/object/handle.rs

@@ -0,0 +1,21 @@
+// ANCHOR: handle
+use super::{KernelObject, Rights};
+use alloc::sync::Arc;
+
+pub type HandleValue = u32;
+pub const INVALID_HANDLE: HandleValue = 0;
+
+/// 内核对象句柄
+#[derive(Clone)]
+pub struct Handle {
+    pub object: Arc<dyn KernelObject>,
+    pub rights: Rights,
+}
+
+impl Handle {
+    /// 创建一个新句柄
+    pub fn new(object: Arc<dyn KernelObject>, rights: Rights) -> Self {
+        Handle { object, rights }
+    }
+}
+// ANCHOR_END: handle

code/ch03-04/object/src/object/mod.rs

@@ -0,0 +1,185 @@
+use alloc::string::String;
+use alloc::sync::Arc;
+use core::fmt::Debug;
+use core::sync::atomic::*;
+use downcast_rs::{impl_downcast, DowncastSync};
+use spin::Mutex;
+
+mod handle;
+mod rights;
+
+pub use self::handle::*;
+pub use self::rights::*;
+pub use super::*;
+
+/// 内核对象公共接口
+pub trait KernelObject: DowncastSync + Debug {
+    /// 获取对象 ID
+    fn id(&self) -> KoID;
+    /// 获取对象类型名
+    fn type_name(&self) -> &str;
+    /// 获取对象名称
+    fn name(&self) -> String;
+    /// 设置对象名称
+    fn set_name(&self, name: &str);
+    /// 尝试获取对象伙伴
+    ///
+    /// 当前该对象必须是 `Channel`
+    fn peer(&self) -> ZxResult<Arc<dyn KernelObject>> {
+        Err(ZxError::NOT_SUPPORTED)
+    }
+    /// 尝试获取关联对象 id，否则返回 0
+    ///
+    /// 当前该对象必须是 `Channel` 或者 `Task`
+    ///
+    /// 如果该对象是 `Channel`, 将获取伙伴的 id
+    ///
+    /// 如果该对象是 `Task`, 将获取其父 `Task` 的 id
+    fn related_koid(&self) -> KoID {
+        0
+    }
+    /// 尝试获取对应 id 的子对象
+    ///
+    /// 当前该对象必须是 `Job` 或者 `Process`.
+    ///
+    /// 如果该对象是 `Job`，则其直属子 `Job` 以及 `Process` 必须被获取。
+    fn get_child(&self, _id: KoID) -> ZxResult<Arc<dyn KernelObject>> {
+        Err(ZxError::WRONG_TYPE)
+    }
+}
+
+impl_downcast!(sync KernelObject);
+
+/// 对象 ID 类型
+pub type KoID = u64;
+
+/// 内核对象核心结构
+pub struct KObjectBase {
+    /// 对象 ID
+    pub id: KoID,
+    inner: Mutex<KObjectBaseInner>,
+}
+
+/// `KObjectBase` 的内部可变部分
+#[derive(Default)]
+struct KObjectBaseInner {
+    name: String,
+}
+
+impl Default for KObjectBase {
+    /// 创建一个新 `KObjectBase`
+    fn default() -> Self {
+        KObjectBase {
+            id: Self::new_koid(),
+            inner: Default::default(),
+        }
+    }
+}
+
+impl KObjectBase {
+    /// Create a new kernel object base.
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// 生成一个唯一的 ID
+    fn new_koid() -> KoID {
+        static NEXT_KOID: AtomicU64 = AtomicU64::new(1024);
+        NEXT_KOID.fetch_add(1, Ordering::SeqCst)
+    }
+    /// 获取对象名称
+    pub fn name(&self) -> String {
+        self.inner.lock().name.clone()
+    }
+    /// 设置对象名称
+    pub fn set_name(&self, name: &str) {
+        self.inner.lock().name = String::from(name);
+    }
+
+    /// Create a kernel object base with `name`.
+    pub fn with_name(name: &str) -> Self {
+        KObjectBase {
+            id: Self::new_koid(),
+            inner: Mutex::new(KObjectBaseInner {
+                name: String::from(name),
+            }),
+        }
+    }
+}
+
+/// 为内核对象 struct 自动实现 `KernelObject` trait 的宏。
+#[macro_export] // 导出宏，可在 crate 外部使用
+macro_rules! impl_kobject {
+    // 匹配类型名，并可以提供函数覆盖默认实现
+    ($class:ident $( $fn:tt )*) => {
+        // 为对象实现 KernelObject trait，方法直接转发到内部 struct
+        impl KernelObject for $class {
+            fn id(&self) -> KoID {
+                // 直接访问内部的 pub 属性
+                self.base.id
+            }
+            fn type_name(&self) -> &str {
+                // 用 stringify! 宏将输入转成字符串
+                stringify!($class)
+            }
+            // 注意宏里面的类型要写完整路径，例如：alloc::string::String
+            fn name(&self) -> alloc::string::String {
+                self.base.name()
+            }
+            fn set_name(&self, name: &str){
+                // 直接访问内部的 pub 方法
+                self.base.set_name(name)
+            }
+            // 可以传入任意数量的函数，覆盖 trait 的默认实现
+            $( $fn )*
+        }
+        // 为对象实现 Debug trait
+        impl core::fmt::Debug for $class {
+            fn fmt(
+                &self,
+                f: &mut core::fmt::Formatter<'_>,
+            ) -> core::result::Result<(), core::fmt::Error> {
+                // 输出对象类型、ID 和名称
+                f.debug_tuple(&stringify!($class))
+                    .field(&self.id())
+                    .field(&self.name())
+                    .finish()
+            }
+        }
+    };
+}
+
+/// 空对象
+pub struct DummyObject {
+    // 其中必须包含一个名为 `base` 的 `KObjectBase`
+    base: KObjectBase,
+}
+
+// 使用刚才的宏，声明其为内核对象，自动生成必要的代码
+impl_kobject!(DummyObject);
+
+impl DummyObject {
+    /// 创建一个新 `DummyObject`
+    pub fn new() -> Arc<Self> {
+        Arc::new(DummyObject {
+            base: KObjectBase::default(),
+        })
+    }
+}
+
+#[cfg(test)]
+#[test]
+fn impl_kobject() {
+    use alloc::format;
+    let dummy = DummyObject::new();
+    let object: Arc<dyn KernelObject> = dummy;
+    assert_eq!(object.type_name(), "DummyObject");
+    assert_eq!(object.name(), "");
+    object.set_name("dummy");
+    assert_eq!(object.name(), "dummy");
+    assert_eq!(
+        format!("{:?}", object),
+        format!("DummyObject({}, \"dummy\")", object.id())
+    );
+    let _result: Arc<DummyObject> = object.downcast_arc::<DummyObject>().unwrap();
+}

code/ch03-04/object/src/object/rights.rs

@@ -0,0 +1,58 @@
+// ANCHOR: rights
+use bitflags::bitflags;
+
+bitflags! {
+    /// 句柄权限
+    #[derive(Default)]
+    pub struct Rights: u32 {
+        const DUPLICATE = 1 << 0;
+        const TRANSFER = 1 << 1;
+        const READ = 1 << 2;
+        const WRITE = 1 << 3;
+        const EXECUTE = 1 << 4;
+        const MAP = 1 << 5;
+        const GET_PROPERTY = 1 << 6;
+        const SET_PROPERTY = 1 << 7;
+        const ENUMERATE = 1 << 8;
+        const DESTROY = 1 << 9;
+        const SET_POLICY = 1 << 10;
+        const GET_POLICY = 1 << 11;
+        const SIGNAL = 1 << 12;
+        const SIGNAL_PEER = 1 << 13;
+        const WAIT = 1 << 14;
+        const INSPECT = 1 << 15;
+        const MANAGE_JOB = 1 << 16;
+        const MANAGE_PROCESS = 1 << 17;
+        const MANAGE_THREAD = 1 << 18;
+        const APPLY_PROFILE = 1 << 19;
+        const SAME_RIGHTS = 1 << 31;
+
+        const BASIC = Self::TRANSFER.bits | Self::DUPLICATE.bits | Self::WAIT.bits | Self::INSPECT.bits;
+        const IO = Self::READ.bits | Self::WRITE.bits;
+
+        /// GET_PROPERTY ｜ SET_PROPERTY
+        const PROPERTY = Self::GET_PROPERTY.bits | Self::SET_PROPERTY.bits;
+
+        /// GET_POLICY ｜ SET_POLICY
+        const POLICY = Self::GET_POLICY.bits | Self::SET_POLICY.bits;
+
+        /// BASIC & !Self::DUPLICATE | IO | SIGNAL | SIGNAL_PEER
+        const DEFAULT_CHANNEL = Self::BASIC.bits & !Self::DUPLICATE.bits | Self::IO.bits | Self::SIGNAL.bits | Self::SIGNAL_PEER.bits;
+
+        /// BASIC | IO | PROPERTY | ENUMERATE | DESTROY | SIGNAL | MANAGE_PROCESS | MANAGE_THREAD
+        const DEFAULT_PROCESS = Self::BASIC.bits | Self::IO.bits | Self::PROPERTY.bits | Self::ENUMERATE.bits | Self::DESTROY.bits
+            | Self::SIGNAL.bits | Self::MANAGE_PROCESS.bits | Self::MANAGE_THREAD.bits;
+
+        /// BASIC | IO | PROPERTY | DESTROY | SIGNAL | MANAGE_THREAD
+        const DEFAULT_THREAD = Self::BASIC.bits | Self::IO.bits | Self::PROPERTY.bits | Self::DESTROY.bits | Self::SIGNAL.bits | Self::MANAGE_THREAD.bits;
+
+        /// BASIC | WAIT
+        const DEFAULT_VMAR = Self::BASIC.bits & !Self::WAIT.bits;
+
+        /// BASIC | IO | PROPERTY | POLICY | ENUMERATE | DESTROY | SIGNAL | MANAGE_JOB | MANAGE_PROCESS | MANAGE_THREAD
+        const DEFAULT_JOB = Self::BASIC.bits | Self::IO.bits | Self::PROPERTY.bits | Self::POLICY.bits | Self::ENUMERATE.bits
+            | Self::DESTROY.bits | Self::SIGNAL.bits | Self::MANAGE_JOB.bits | Self::MANAGE_PROCESS.bits | Self::MANAGE_THREAD.bits;
+
+    }
+}
+// ANCHOR_END: rights

code/ch03-04/object/src/task/job.rs

@@ -0,0 +1,409 @@
+use {
+    super::job_policy::*,
+    super::process::Process,
+    super::*,
+    crate::error::*,
+    crate::object::*,
+    crate::task::Task,
+    alloc::sync::{Arc, Weak},
+    alloc::vec::Vec,
+    spin::Mutex,
+};
+
+/// Job 对象
+#[allow(dead_code)]
+pub struct Job {
+    base: KObjectBase,
+    parent: Option<Arc<Job>>,
+    parent_policy: JobPolicy,
+    inner: Mutex<JobInner>,
+}
+
+impl_kobject!(Job
+    fn get_child(&self, id: KoID) -> ZxResult<Arc<dyn KernelObject>> {
+        let inner = self.inner.lock();
+        if let Some(job) = inner.children.iter().filter_map(|o|o.upgrade()).find(|o| o.id() == id) {
+            return Ok(job);
+        }
+        if let Some(proc) = inner.processes.iter().find(|o| o.id() == id) {
+            return Ok(proc.clone());
+        }
+        Err(ZxError::NOT_FOUND)
+    }
+    fn related_koid(&self) -> KoID {
+        self.parent.as_ref().map(|p| p.id()).unwrap_or(0)
+    }
+);
+
+#[derive(Default)]
+struct JobInner {
+    policy: JobPolicy,
+    children: Vec<Weak<Job>>,
+    processes: Vec<Arc<Process>>,
+    // if the job is killed, no more child creation should works
+    killed: bool,
+    self_ref: Weak<Job>,
+}
+
+impl Job {
+    /// Create the root job.
+    pub fn root() -> Arc<Self> {
+        let job = Arc::new(Job {
+            base: KObjectBase::new(),
+            parent: None,
+            parent_policy: JobPolicy::default(),
+            inner: Mutex::new(JobInner::default()),
+        });
+        job.inner.lock().self_ref = Arc::downgrade(&job);
+        job
+    }
+
+    /// Create a new child job object.
+    pub fn create_child(self: &Arc<Self>) -> ZxResult<Arc<Self>> {
+        let mut inner = self.inner.lock();
+        if inner.killed {
+            return Err(ZxError::BAD_STATE);
+        }
+        let child = Arc::new(Job {
+            base: KObjectBase::new(),
+            parent: Some(self.clone()),
+            parent_policy: inner.policy.merge(&self.parent_policy),
+            inner: Mutex::new(JobInner::default()),
+        });
+        let child_weak = Arc::downgrade(&child);
+        child.inner.lock().self_ref = child_weak.clone();
+        inner.children.push(child_weak);
+        Ok(child)
+    }
+
+    fn remove_child(&self, to_remove: &Weak<Job>) {
+        let mut inner = self.inner.lock();
+        inner.children.retain(|child| !to_remove.ptr_eq(child));
+        if inner.killed && inner.processes.is_empty() && inner.children.is_empty() {
+            drop(inner);
+            self.terminate()
+        }
+    }
+
+    /// Get the policy of the job.
+    pub fn policy(&self) -> JobPolicy {
+        self.inner.lock().policy.merge(&self.parent_policy)
+    }
+
+    /// Get the parent job.
+    pub fn parent(&self) -> Option<Arc<Self>> {
+        self.parent.clone()
+    }
+
+    /// Sets one or more security and/or resource policies to an empty job.
+    ///
+    /// The job's effective policies is the combination of the parent's
+    /// effective policies and the policies specified in policy.
+    ///
+    /// After this call succeeds any new child process or child job will have
+    /// the new effective policy applied to it.
+    pub fn set_policy_basic(
+        &self,
+        options: SetPolicyOptions,
+        policies: &[BasicPolicy],
+    ) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if !inner.is_empty() {
+            return Err(ZxError::BAD_STATE);
+        }
+        for policy in policies {
+            if self.parent_policy.get_action(policy.condition).is_some() {
+                match options {
+                    SetPolicyOptions::Absolute => return Err(ZxError::ALREADY_EXISTS),
+                    SetPolicyOptions::Relative => {}
+                }
+            } else {
+                inner.policy.apply(*policy);
+            }
+        }
+        Ok(())
+    }
+
+    /// Add a process to the job.
+    pub(super) fn add_process(&self, process: Arc<Process>) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if inner.killed {
+            return Err(ZxError::BAD_STATE);
+        }
+        inner.processes.push(process);
+        Ok(())
+    }
+
+    /// Remove a process from the job.
+    pub(super) fn remove_process(&self, id: KoID) {
+        let mut inner = self.inner.lock();
+        inner.processes.retain(|proc| proc.id() != id);
+        if inner.killed && inner.processes.is_empty() && inner.children.is_empty() {
+            drop(inner);
+            self.terminate()
+        }
+    }
+
+    /// Check whether this job is root job.
+    pub fn check_root_job(&self) -> ZxResult {
+        if self.parent.is_some() {
+            Err(ZxError::ACCESS_DENIED)
+        } else {
+            Ok(())
+        }
+    }
+
+    /// Get KoIDs of Processes.
+    pub fn process_ids(&self) -> Vec<KoID> {
+        self.inner.lock().processes.iter().map(|p| p.id()).collect()
+    }
+
+    /// Get KoIDs of children Jobs.
+    pub fn children_ids(&self) -> Vec<KoID> {
+        self.inner
+            .lock()
+            .children
+            .iter()
+            .filter_map(|j| j.upgrade())
+            .map(|j| j.id())
+            .collect()
+    }
+
+    /// Return true if this job has no processes and no child jobs.
+    pub fn is_empty(&self) -> bool {
+        self.inner.lock().is_empty()
+    }
+
+    /// The job finally terminates.
+    fn terminate(&self) {
+        if let Some(parent) = self.parent.as_ref() {
+            parent.remove_child(&self.inner.lock().self_ref)
+        }
+    }
+}
+
+impl Task for Job {
+    /// Kill the job. The job do not terminate immediately when killed.
+    /// It will terminate after all its children and processes are terminated.
+    fn kill(&self) {
+        let (children, processes) = {
+            let mut inner = self.inner.lock();
+            if inner.killed {
+                return;
+            }
+            inner.killed = true;
+            (inner.children.clone(), inner.processes.clone())
+        };
+        if children.is_empty() && processes.is_empty() {
+            self.terminate();
+            return;
+        }
+        for child in children {
+            if let Some(child) = child.upgrade() {
+                child.kill();
+            }
+        }
+        for proc in processes {
+            proc.kill();
+        }
+    }
+
+    fn suspend(&self) {
+        panic!("job do not support suspend");
+    }
+
+    fn resume(&self) {
+        panic!("job do not support resume");
+    }
+}
+
+impl JobInner {
+    fn is_empty(&self) -> bool {
+        self.processes.is_empty() && self.children.is_empty()
+    }
+}
+
+impl Drop for Job {
+    fn drop(&mut self) {
+        self.terminate();
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::task::TASK_RETCODE_SYSCALL_KILL;
+
+    #[test]
+    fn create() {
+        let root_job = Job::root();
+        let job = Job::create_child(&root_job).expect("failed to create job");
+
+        let child = root_job
+            .get_child(job.id())
+            .unwrap()
+            .downcast_arc()
+            .unwrap();
+        assert!(Arc::ptr_eq(&child, &job));
+        assert_eq!(job.related_koid(), root_job.id());
+        assert_eq!(root_job.related_koid(), 0);
+
+        root_job.kill();
+        assert_eq!(root_job.create_child().err(), Some(ZxError::BAD_STATE));
+    }
+
+    #[test]
+    fn set_policy() {
+        let root_job = Job::root();
+
+        // default policy
+        assert_eq!(
+            root_job.policy().get_action(PolicyCondition::BadHandle),
+            None
+        );
+
+        // set policy for root job
+        let policy = &[BasicPolicy {
+            condition: PolicyCondition::BadHandle,
+            action: PolicyAction::Deny,
+        }];
+        root_job
+            .set_policy_basic(SetPolicyOptions::Relative, policy)
+            .expect("failed to set policy");
+        assert_eq!(
+            root_job.policy().get_action(PolicyCondition::BadHandle),
+            Some(PolicyAction::Deny)
+        );
+
+        // override policy should success
+        let policy = &[BasicPolicy {
+            condition: PolicyCondition::BadHandle,
+            action: PolicyAction::Allow,
+        }];
+        root_job
+            .set_policy_basic(SetPolicyOptions::Relative, policy)
+            .expect("failed to set policy");
+        assert_eq!(
+            root_job.policy().get_action(PolicyCondition::BadHandle),
+            Some(PolicyAction::Allow)
+        );
+
+        // create a child job
+        let job = Job::create_child(&root_job).expect("failed to create job");
+
+        // should inherit parent's policy.
+        assert_eq!(
+            job.policy().get_action(PolicyCondition::BadHandle),
+            Some(PolicyAction::Allow)
+        );
+
+        // setting policy for a non-empty job should fail.
+        assert_eq!(
+            root_job.set_policy_basic(SetPolicyOptions::Relative, &[]),
+            Err(ZxError::BAD_STATE)
+        );
+
+        // set new policy should success.
+        let policy = &[BasicPolicy {
+            condition: PolicyCondition::WrongObject,
+            action: PolicyAction::Allow,
+        }];
+        job.set_policy_basic(SetPolicyOptions::Relative, policy)
+            .expect("failed to set policy");
+        assert_eq!(
+            job.policy().get_action(PolicyCondition::WrongObject),
+            Some(PolicyAction::Allow)
+        );
+
+        // relatively setting existing policy should be ignored.
+        let policy = &[BasicPolicy {
+            condition: PolicyCondition::BadHandle,
+            action: PolicyAction::Deny,
+        }];
+        job.set_policy_basic(SetPolicyOptions::Relative, policy)
+            .expect("failed to set policy");
+        assert_eq!(
+            job.policy().get_action(PolicyCondition::BadHandle),
+            Some(PolicyAction::Allow)
+        );
+
+        // absolutely setting existing policy should fail.
+        assert_eq!(
+            job.set_policy_basic(SetPolicyOptions::Absolute, policy),
+            Err(ZxError::ALREADY_EXISTS)
+        );
+    }
+
+    #[test]
+    fn parent_child() {
+        let root_job = Job::root();
+        let job = Job::create_child(&root_job).expect("failed to create job");
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+
+        assert_eq!(root_job.get_child(job.id()).unwrap().id(), job.id());
+        assert_eq!(root_job.get_child(proc.id()).unwrap().id(), proc.id());
+        assert_eq!(
+            root_job.get_child(root_job.id()).err(),
+            Some(ZxError::NOT_FOUND)
+        );
+        assert!(Arc::ptr_eq(&job.parent().unwrap(), &root_job));
+
+        let job1 = root_job.create_child().expect("failed to create job");
+        let proc1 = Process::create(&root_job, "proc1").expect("failed to create process");
+        assert_eq!(root_job.children_ids(), vec![job.id(), job1.id()]);
+        assert_eq!(root_job.process_ids(), vec![proc.id(), proc1.id()]);
+
+        root_job.kill();
+        assert_eq!(root_job.create_child().err(), Some(ZxError::BAD_STATE));
+    }
+
+    #[test]
+    fn check() {
+        let root_job = Job::root();
+        assert!(root_job.is_empty());
+        let job = root_job.create_child().expect("failed to create job");
+        assert_eq!(root_job.check_root_job(), Ok(()));
+        assert_eq!(job.check_root_job(), Err(ZxError::ACCESS_DENIED));
+
+        assert!(!root_job.is_empty());
+        assert!(job.is_empty());
+
+        let _proc = Process::create(&job, "proc").expect("failed to create process");
+        assert!(!job.is_empty());
+    }
+
+    #[test]
+    fn kill() {
+        let root_job = Job::root();
+        let job = Job::create_child(&root_job).expect("failed to create job");
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+        let current_thread = CurrentThread(thread.clone());
+
+        root_job.kill();
+        assert!(root_job.inner.lock().killed);
+        assert!(job.inner.lock().killed);
+        assert_eq!(proc.status(), Status::Exited(TASK_RETCODE_SYSCALL_KILL));
+        assert_eq!(thread.state(), ThreadState::Dying);
+
+        std::mem::drop(current_thread);
+        assert!(root_job.inner.lock().killed);
+        assert!(job.inner.lock().killed);
+        assert_eq!(proc.status(), Status::Exited(TASK_RETCODE_SYSCALL_KILL));
+        assert_eq!(thread.state(), ThreadState::Dead);
+
+        // The job has no children.
+        let root_job = Job::root();
+        root_job.kill();
+        assert!(root_job.inner.lock().killed);
+
+        // The job's process have no threads.
+        let root_job = Job::root();
+        let job = Job::create_child(&root_job).expect("failed to create job");
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        root_job.kill();
+        assert!(root_job.inner.lock().killed);
+        assert!(job.inner.lock().killed);
+        assert_eq!(proc.status(), Status::Exited(TASK_RETCODE_SYSCALL_KILL));
+    }
+}

code/ch03-04/object/src/task/job_policy.rs

@@ -0,0 +1,108 @@
+/// Security and resource policies of a job.
+#[derive(Default, Copy, Clone)]
+pub struct JobPolicy {
+    // TODO: use bitset
+    action: [Option<PolicyAction>; 15],
+}
+
+impl JobPolicy {
+    /// Get the action of a policy `condition`.
+    pub fn get_action(&self, condition: PolicyCondition) -> Option<PolicyAction> {
+        self.action[condition as usize]
+    }
+
+    /// Apply a basic policy.
+    pub fn apply(&mut self, policy: BasicPolicy) {
+        self.action[policy.condition as usize] = Some(policy.action);
+    }
+
+    /// Merge the policy with `parent`'s.
+    pub fn merge(&self, parent: &Self) -> Self {
+        let mut new = *self;
+        for i in 0..15 {
+            if parent.action[i].is_some() {
+                new.action[i] = parent.action[i];
+            }
+        }
+        new
+    }
+}
+
+/// Control the effect in the case of conflict between
+/// the existing policies and the new policies when setting new policies.
+#[derive(Debug, Copy, Clone)]
+pub enum SetPolicyOptions {
+    /// Policy is applied for all conditions in policy or the call fails.
+    Absolute,
+    /// Policy is applied for the conditions not specifically overridden by the parent policy.
+    Relative,
+}
+
+/// The policy type.
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct BasicPolicy {
+    /// Condition when the policy is applied.
+    pub condition: PolicyCondition,
+    ///
+    pub action: PolicyAction,
+}
+
+/// The condition when a policy is applied.
+#[repr(u32)]
+#[derive(Debug, Copy, Clone)]
+pub enum PolicyCondition {
+    /// A process under this job is attempting to issue a syscall with an invalid handle.
+    /// In this case, `PolicyAction::Allow` and `PolicyAction::Deny` are equivalent:
+    /// if the syscall returns, it will always return the error ZX_ERR_BAD_HANDLE.
+    BadHandle = 0,
+    /// A process under this job is attempting to issue a syscall with a handle that does not support such operation.
+    WrongObject = 1,
+    /// A process under this job is attempting to map an address region with write-execute access.
+    VmarWx = 2,
+    /// A special condition that stands for all of the above ZX_NEW conditions
+    /// such as NEW_VMO, NEW_CHANNEL, NEW_EVENT, NEW_EVENTPAIR, NEW_PORT, NEW_SOCKET, NEW_FIFO,
+    /// And any future ZX_NEW policy.
+    /// This will include any new kernel objects which do not require a parent object for creation.
+    NewAny = 3,
+    /// A process under this job is attempting to create a new vm object.
+    NewVMO = 4,
+    /// A process under this job is attempting to create a new channel.
+    NewChannel = 5,
+    /// A process under this job is attempting to create a new event.
+    NewEvent = 6,
+    /// A process under this job is attempting to create a new event pair.
+    NewEventPair = 7,
+    /// A process under this job is attempting to create a new port.
+    NewPort = 8,
+    /// A process under this job is attempting to create a new socket.
+    NewSocket = 9,
+    /// A process under this job is attempting to create a new fifo.
+    NewFIFO = 10,
+    /// A process under this job is attempting to create a new timer.
+    NewTimer = 11,
+    /// A process under this job is attempting to create a new process.
+    NewProcess = 12,
+    /// A process under this job is attempting to create a new profile.
+    NewProfile = 13,
+    /// A process under this job is attempting to use zx_vmo_replace_as_executable()
+    /// with a ZX_HANDLE_INVALID as the second argument rather than a valid ZX_RSRC_KIND_VMEX.
+    AmbientMarkVMOExec = 14,
+}
+
+/// The action taken when the condition happens specified by a policy.
+#[repr(u32)]
+#[derive(Debug, Copy, Clone, Eq, PartialEq)]
+pub enum PolicyAction {
+    /// Allow condition.
+    Allow = 0,
+    /// Prevent condition.
+    Deny = 1,
+    /// Generate an exception via the debug port. An exception generated this
+    /// way acts as a breakpoint. The thread may be resumed after the exception.
+    AllowException = 2,
+    /// Just like `AllowException`, but after resuming condition is denied.
+    DenyException = 3,
+    /// Terminate the process.
+    Kill = 4,
+}

code/ch03-04/object/src/task/mod.rs

@@ -0,0 +1,24 @@
+use super::*;
+
+mod job;
+mod job_policy;
+mod process;
+mod thread;
+
+pub use {self::job::*, self::job_policy::*, self::process::*, self::thread::*};
+
+/// Task (Thread, Process, or Job)
+pub trait Task: Sync + Send {
+    /// Kill the task. The task do not terminate immediately when killed.
+    /// It will terminate after all its children are terminated or some cleanups are finished.
+    fn kill(&self);
+
+    /// Suspend the task. Currently only thread or process handles may be suspended.
+    fn suspend(&self);
+
+    /// Resume the task
+    fn resume(&self);
+}
+
+/// The return code set when a task is killed via zx_task_kill().
+pub const TASK_RETCODE_SYSCALL_KILL: i64 = -1028;

code/ch03-04/object/src/task/process.rs

@@ -0,0 +1,505 @@
+use {
+    super::{job::Job, job_policy::*, thread::*, *},
+    crate::{error::*, object::*, vm::*},
+    alloc::{sync::Arc, vec::Vec},
+    hashbrown::HashMap,
+    spin::Mutex,
+};
+
+pub struct Process {
+    base: KObjectBase,
+    job: Arc<Job>,
+    policy: JobPolicy,
+    vmar: Arc<VmAddressRegion>,
+    inner: Mutex<ProcessInner>,
+}
+
+impl_kobject!(Process
+    fn get_child(&self, id: KoID) -> ZxResult<Arc<dyn KernelObject>> {
+        let inner = self.inner.lock();
+        let thread = inner.threads.iter().find(|o| o.id() == id).ok_or(ZxError::NOT_FOUND)?;
+        Ok(thread.clone())
+    }
+    fn related_koid(&self) -> KoID {
+        self.job.id()
+    }
+);
+
+#[derive(Default)]
+struct ProcessInner {
+    max_handle_id: u32,
+    status: Status,
+    handles: HashMap<HandleValue, Handle>,
+    threads: Vec<Arc<Thread>>,
+}
+
+/// Status of a process.
+#[derive(Debug, Copy, Clone, Eq, PartialEq)]
+pub enum Status {
+    /// Initial state, no thread present in process.
+    Init,
+    /// First thread has started and is running.
+    Running,
+    /// Process has exited with the code.
+    Exited(i64),
+}
+
+impl Default for Status {
+    fn default() -> Self {
+        Status::Init
+    }
+}
+
+impl Process {
+    /// Create a new process in the `job`.
+    pub fn create(job: &Arc<Job>, name: &str) -> ZxResult<Arc<Self>> {
+        let proc = Arc::new(Process {
+            base: KObjectBase::with_name(name),
+            job: job.clone(),
+            policy: job.policy(),
+            vmar: VmAddressRegion::new_root(),
+            inner: Mutex::new(ProcessInner::default()),
+        });
+        job.add_process(proc.clone())?;
+        Ok(proc)
+    }
+
+    /// Get a handle from the process
+    fn get_handle(&self, handle_value: HandleValue) -> ZxResult<Handle> {
+        self.inner.lock().get_handle(handle_value)
+    }
+
+    /// 添加一个新的对象句柄
+    pub fn add_handle(&self, handle: Handle) -> HandleValue {
+        self.inner.lock().add_handle(handle)
+    }
+
+    /// 删除一个对象句柄
+    pub fn remove_handle(&self, handle_value: HandleValue) -> ZxResult<Handle> {
+        self.inner.lock().remove_handle(handle_value)
+    }
+
+    /// Add all handles to the process
+    pub fn add_handles(&self, handles: Vec<Handle>) -> Vec<HandleValue> {
+        let mut inner = self.inner.lock();
+        handles.into_iter().map(|h| inner.add_handle(h)).collect()
+    }
+
+    /// Remove all handles from the process.
+    pub fn remove_handles(&self, handle_values: &[HandleValue]) -> ZxResult<Vec<Handle>> {
+        let mut inner = self.inner.lock();
+        handle_values
+            .iter()
+            .map(|h| inner.remove_handle(*h))
+            .collect()
+    }
+
+    /// Get the kernel object corresponding to this `handle_value`
+    pub fn get_object<T: KernelObject>(&self, handle_value: HandleValue) -> ZxResult<Arc<T>> {
+        let handle = self.get_handle(handle_value)?;
+        let object = handle
+            .object
+            .downcast_arc::<T>()
+            .map_err(|_| ZxError::WRONG_TYPE)?;
+        Ok(object)
+    }
+
+    /// 根据句柄值查找内核对象，并检查权限
+    pub fn get_object_with_rights<T: KernelObject>(
+        &self,
+        handle_value: HandleValue,
+        desired_rights: Rights,
+    ) -> ZxResult<Arc<T>> {
+        let handle = self.get_handle(handle_value)?;
+        // check type before rights
+        let object = handle
+            .object
+            .downcast_arc::<T>()
+            .map_err(|_| ZxError::WRONG_TYPE)?;
+        if !handle.rights.contains(desired_rights) {
+            return Err(ZxError::ACCESS_DENIED);
+        }
+        Ok(object)
+    }
+
+    /// Get the kernel object corresponding to this `handle_value` and this handle's rights.
+    pub fn get_object_and_rights<T: KernelObject>(
+        &self,
+        handle_value: HandleValue,
+    ) -> ZxResult<(Arc<T>, Rights)> {
+        let handle = self.get_handle(handle_value)?;
+        let object = handle
+            .object
+            .downcast_arc::<T>()
+            .map_err(|_| ZxError::WRONG_TYPE)?;
+        Ok((object, handle.rights))
+    }
+
+    /// Remove a handle referring to a kernel object of the given type from the process.
+    pub fn remove_object<T: KernelObject>(&self, handle_value: HandleValue) -> ZxResult<Arc<T>> {
+        let handle = self.remove_handle(handle_value)?;
+        let object = handle
+            .object
+            .downcast_arc::<T>()
+            .map_err(|_| ZxError::WRONG_TYPE)?;
+        Ok(object)
+    }
+
+    pub fn start(
+        &self,
+        thread: &Arc<Thread>,
+        entry: usize,
+        stack: usize,
+        arg1: Option<Handle>,
+        arg2: usize,
+        thread_fn: ThreadFn,
+    ) -> ZxResult {
+        let handle_value;
+        {
+            let mut inner = self.inner.lock();
+            if !inner.contains_thread(thread) {
+                return Err(ZxError::ACCESS_DENIED);
+            }
+            if inner.status != Status::Init {
+                return Err(ZxError::BAD_STATE);
+            }
+            inner.status = Status::Running;
+            handle_value = arg1.map_or(INVALID_HANDLE, |handle| inner.add_handle(handle));
+        }
+        thread.set_first_thread();
+        match thread.start(entry, stack, handle_value as usize, arg2, thread_fn) {
+            Ok(_) => Ok(()),
+            Err(err) => {
+                let mut inner = self.inner.lock();
+                if handle_value != INVALID_HANDLE {
+                    inner.remove_handle(handle_value).ok();
+                }
+                Err(err)
+            }
+        }
+    }
+
+    /// Exit current process with `retcode`.
+    /// The process do not terminate immediately when exited.
+    /// It will terminate after all its child threads are terminated.
+    pub fn exit(&self, retcode: i64) {
+        let mut inner = self.inner.lock();
+        if let Status::Exited(_) = inner.status {
+            return;
+        }
+        inner.status = Status::Exited(retcode);
+        if inner.threads.is_empty() {
+            inner.handles.clear();
+            drop(inner);
+            self.terminate();
+            return;
+        }
+        for thread in inner.threads.iter() {
+            thread.kill();
+        }
+        inner.handles.clear();
+    }
+
+    /// The process finally terminates.
+    fn terminate(&self) {
+        let mut inner = self.inner.lock();
+        let _retcode = match inner.status {
+            Status::Exited(retcode) => retcode,
+            _ => {
+                inner.status = Status::Exited(0);
+                0
+            }
+        };
+        self.job.remove_process(self.base.id);
+    }
+
+    /// Check whether `condition` is allowed in the parent job's policy.
+    pub fn check_policy(&self, condition: PolicyCondition) -> ZxResult {
+        match self
+            .policy
+            .get_action(condition)
+            .unwrap_or(PolicyAction::Allow)
+        {
+            PolicyAction::Allow => Ok(()),
+            PolicyAction::Deny => Err(ZxError::ACCESS_DENIED),
+            _ => unimplemented!(),
+        }
+    }
+
+    /// Get process status.
+    pub fn status(&self) -> Status {
+        self.inner.lock().status
+    }
+
+    /// Get the `VmAddressRegion` of the process.
+    pub fn vmar(&self) -> Arc<VmAddressRegion> {
+        self.vmar.clone()
+    }
+
+    /// Get the job of the process.
+    pub fn job(&self) -> Arc<Job> {
+        self.job.clone()
+    }
+
+    /// Add a thread to the process.
+    pub(super) fn add_thread(&self, thread: Arc<Thread>) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if let Status::Exited(_) = inner.status {
+            return Err(ZxError::BAD_STATE);
+        }
+        inner.threads.push(thread);
+        Ok(())
+    }
+
+    /// Remove a thread from the process.
+    ///
+    /// If no more threads left, exit the process.
+    pub(super) fn remove_thread(&self, tid: KoID) {
+        let mut inner = self.inner.lock();
+        inner.threads.retain(|t| t.id() != tid);
+        if inner.threads.is_empty() {
+            drop(inner);
+            self.terminate();
+        }
+    }
+
+    /// Get KoIDs of Threads.
+    pub fn thread_ids(&self) -> Vec<KoID> {
+        self.inner.lock().threads.iter().map(|t| t.id()).collect()
+    }
+
+    /// Get information of this process.
+    pub fn get_info(&self) -> ProcessInfo {
+        let mut info = ProcessInfo {
+            ..Default::default()
+        };
+        match self.inner.lock().status {
+            Status::Init => {
+                info.started = false;
+                info.has_exited = false;
+            }
+            Status::Running => {
+                info.started = true;
+                info.has_exited = false;
+            }
+            Status::Exited(ret) => {
+                info.return_code = ret;
+                info.has_exited = true;
+                info.started = true;
+            }
+        }
+        info
+    }
+}
+
+/// Information of a process.
+#[allow(missing_docs)]
+#[repr(C)]
+#[derive(Default)]
+pub struct ProcessInfo {
+    pub return_code: i64,
+    pub started: bool,
+    pub has_exited: bool,
+}
+
+impl Task for Process {
+    fn kill(&self) {
+        self.exit(TASK_RETCODE_SYSCALL_KILL);
+    }
+
+    fn suspend(&self) {
+        let inner = self.inner.lock();
+        for thread in inner.threads.iter() {
+            thread.suspend();
+        }
+    }
+
+    fn resume(&self) {
+        let inner = self.inner.lock();
+        for thread in inner.threads.iter() {
+            thread.resume();
+        }
+    }
+}
+
+impl ProcessInner {
+    /// Add a handle to the process
+    fn add_handle(&mut self, handle: Handle) -> HandleValue {
+        let key = (self.max_handle_id << 2) | 0x3u32;
+        self.max_handle_id += 1;
+        self.handles.insert(key, handle);
+        key
+    }
+
+    fn remove_handle(&mut self, handle_value: HandleValue) -> ZxResult<Handle> {
+        let handle = self
+            .handles
+            .remove(&handle_value)
+            .ok_or(ZxError::BAD_HANDLE)?;
+        Ok(handle)
+    }
+
+    fn get_handle(&mut self, handle_value: HandleValue) -> ZxResult<Handle> {
+        let handle = self.handles.get(&handle_value).ok_or(ZxError::BAD_HANDLE)?;
+        Ok(handle.clone())
+    }
+
+    /// Whether `thread` is in this process.
+    fn contains_thread(&self, thread: &Arc<Thread>) -> bool {
+        self.threads.iter().any(|t| Arc::ptr_eq(t, thread))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn create() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+
+        assert_eq!(proc.related_koid(), root_job.id());
+        assert!(Arc::ptr_eq(&root_job, &proc.job()));
+    }
+
+    #[test]
+    fn handle() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let handle = Handle::new(proc.clone(), Rights::DEFAULT_PROCESS);
+
+        let handle_value = proc.add_handle(handle);
+
+        // getting object should success
+        let object: Arc<Process> = proc
+            .get_object_with_rights(handle_value, Rights::DEFAULT_PROCESS)
+            .expect("failed to get object");
+        assert!(Arc::ptr_eq(&object, &proc));
+
+        let (object, rights) = proc
+            .get_object_and_rights::<Process>(handle_value)
+            .expect("failed to get object");
+        assert!(Arc::ptr_eq(&object, &proc));
+        assert_eq!(rights, Rights::DEFAULT_PROCESS);
+
+        // getting object with an extra rights should fail.
+        assert_eq!(
+            proc.get_object_with_rights::<Process>(handle_value, Rights::MANAGE_JOB)
+                .err(),
+            Some(ZxError::ACCESS_DENIED)
+        );
+
+        // getting object with invalid type should fail.
+        assert_eq!(
+            proc.get_object_with_rights::<Job>(handle_value, Rights::DEFAULT_PROCESS)
+                .err(),
+            Some(ZxError::WRONG_TYPE)
+        );
+
+        proc.remove_handle(handle_value).unwrap();
+
+        // getting object with invalid handle should fail.
+        assert_eq!(
+            proc.get_object_with_rights::<Process>(handle_value, Rights::DEFAULT_PROCESS)
+                .err(),
+            Some(ZxError::BAD_HANDLE)
+        );
+
+        let handle1 = Handle::new(proc.clone(), Rights::DEFAULT_PROCESS);
+        let handle2 = Handle::new(proc.clone(), Rights::DEFAULT_PROCESS);
+
+        let handle_values = proc.add_handles(vec![handle1, handle2]);
+        let object1: Arc<Process> = proc
+            .get_object_with_rights(handle_values[0], Rights::DEFAULT_PROCESS)
+            .expect("failed to get object");
+        assert!(Arc::ptr_eq(&object1, &proc));
+
+        proc.remove_handles(&handle_values).unwrap();
+        assert_eq!(
+            proc.get_object_with_rights::<Process>(handle_values[0], Rights::DEFAULT_PROCESS)
+                .err(),
+            Some(ZxError::BAD_HANDLE)
+        );
+    }
+
+    #[test]
+    fn get_child() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        assert_eq!(proc.get_child(thread.id()).unwrap().id(), thread.id());
+        assert_eq!(proc.get_child(proc.id()).err(), Some(ZxError::NOT_FOUND));
+
+        let thread1 = Thread::create(&proc, "thread1").expect("failed to create thread");
+        assert_eq!(proc.thread_ids(), vec![thread.id(), thread1.id()]);
+    }
+
+    #[test]
+    fn contains_thread() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        let proc1 = Process::create(&root_job, "proc1").expect("failed to create process");
+        let thread1 = Thread::create(&proc1, "thread1").expect("failed to create thread");
+
+        let inner = proc.inner.lock();
+        assert!(inner.contains_thread(&thread) && !inner.contains_thread(&thread1));
+    }
+
+    #[test]
+    fn check_policy() {
+        let root_job = Job::root();
+        let policy1 = BasicPolicy {
+            condition: PolicyCondition::BadHandle,
+            action: PolicyAction::Allow,
+        };
+        let policy2 = BasicPolicy {
+            condition: PolicyCondition::NewChannel,
+            action: PolicyAction::Deny,
+        };
+
+        assert!(root_job
+            .set_policy_basic(SetPolicyOptions::Absolute, &[policy1, policy2])
+            .is_ok());
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+
+        assert!(proc.check_policy(PolicyCondition::BadHandle).is_ok());
+        assert!(proc.check_policy(PolicyCondition::NewProcess).is_ok());
+        assert_eq!(
+            proc.check_policy(PolicyCondition::NewChannel).err(),
+            Some(ZxError::ACCESS_DENIED)
+        );
+
+        let _job = root_job.create_child().unwrap();
+        assert_eq!(
+            root_job
+                .set_policy_basic(SetPolicyOptions::Absolute, &[policy1, policy2])
+                .err(),
+            Some(ZxError::BAD_STATE)
+        );
+    }
+
+    #[test]
+    fn exit() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        let info = proc.get_info();
+        assert!(!info.has_exited && !info.started && info.return_code == 0);
+
+        proc.exit(666);
+        let info = proc.get_info();
+        assert!(info.has_exited && info.started && info.return_code == 666);
+        assert_eq!(thread.state(), ThreadState::Dying);
+        // TODO: when is the thread dead?
+
+        assert_eq!(
+            Thread::create(&proc, "thread1").err(),
+            Some(ZxError::BAD_STATE)
+        );
+    }
+}

code/ch03-04/object/src/task/thread.rs

@@ -0,0 +1,567 @@
+use {
+    super::process::Process,
+    super::*,
+    crate::object::*,
+    alloc::{boxed::Box, sync::Arc},
+    bitflags::bitflags,
+    core::{
+        future::Future,
+        ops::Deref,
+        pin::Pin,
+        task::{Context, Poll, Waker},
+    },
+    spin::Mutex,
+    trapframe::UserContext,
+};
+
+pub use self::thread_state::*;
+
+mod thread_state;
+
+pub struct Thread {
+    base: KObjectBase,
+    proc: Arc<Process>,
+    inner: Mutex<ThreadInner>,
+}
+
+impl_kobject!(Thread
+    fn related_koid(&self) -> KoID {
+        self.proc.id()
+    }
+);
+
+#[derive(Default)]
+struct ThreadInner {
+    /// Thread context
+    ///
+    /// It will be taken away when running this thread.
+    context: Option<Box<UserContext>>,
+    /// The number of existing `SuspendToken`.
+    suspend_count: usize,
+    /// The waker of task when suspending.
+    waker: Option<Waker>,
+    /// Thread state
+    ///
+    /// NOTE: This variable will never be `Suspended`. On suspended, the
+    /// `suspend_count` is non-zero, and this represents the state before suspended.
+    state: ThreadState,
+    /// Should The ProcessStarting exception generated at start of this thread
+    first_thread: bool,
+    /// Should The ThreadExiting exception do not block this thread
+    killed: bool,
+    /// The time this thread has run on cpu
+    time: u128,
+    flags: ThreadFlag,
+}
+
+impl ThreadInner {
+    fn state(&self) -> ThreadState {
+        // Dying > Exception > Suspend > Blocked
+        if self.suspend_count == 0
+            || self.context.is_none()
+            || self.state == ThreadState::BlockedException
+            || self.state == ThreadState::Dying
+            || self.state == ThreadState::Dead
+        {
+            self.state
+        } else {
+            ThreadState::Suspended
+        }
+    }
+
+    /// Change state and update signal.
+    fn change_state(&mut self, state: ThreadState) {
+        self.state = state;
+    }
+}
+
+bitflags! {
+    /// Thread flags.
+    #[derive(Default)]
+    pub struct ThreadFlag: usize {
+        /// The thread currently has a VCPU.
+        const VCPU = 1 << 3;
+    }
+}
+
+/// The type of a new thread function.
+pub type ThreadFn = fn(thread: CurrentThread) -> Pin<Box<dyn Future<Output = ()> + Send + 'static>>;
+
+impl Thread {
+    /// Create a new thread.
+    pub fn create(proc: &Arc<Process>, name: &str) -> ZxResult<Arc<Self>> {
+        let thread = Arc::new(Thread {
+            base: KObjectBase::with_name(name),
+            proc: proc.clone(),
+            inner: Mutex::new(ThreadInner {
+                context: Some(Box::new(UserContext::default())),
+                ..Default::default()
+            }),
+        });
+        proc.add_thread(thread.clone())?;
+        Ok(thread)
+    }
+
+    /// Get the process.
+    pub fn proc(&self) -> &Arc<Process> {
+        &self.proc
+    }
+
+    /// Start execution on the thread.
+    pub fn start(
+        self: &Arc<Self>,
+        entry: usize,
+        stack: usize,
+        arg1: usize,
+        arg2: usize,
+        thread_fn: ThreadFn,
+    ) -> ZxResult {
+        {
+            let mut inner = self.inner.lock();
+            let context = inner.context.as_mut().ok_or(ZxError::BAD_STATE)?;
+            context.general.rip = entry;
+            context.general.rsp = stack;
+            context.general.rdi = arg1;
+            context.general.rsi = arg2;
+            context.general.rflags |= 0x3202;
+            inner.change_state(ThreadState::Running);
+        }
+        kernel_hal::Thread::spawn(thread_fn(CurrentThread(self.clone())), 0);
+        Ok(())
+    }
+
+    /// Stop the thread. Internal implementation of `exit` and `kill`.
+    ///
+    /// The thread do not terminate immediately when stopped. It is just made dying.
+    /// It will terminate after some cleanups (when `terminate` are called **explicitly** by upper layer).
+    fn stop(&self, killed: bool) {
+        let mut inner = self.inner.lock();
+        if inner.state == ThreadState::Dead {
+            return;
+        }
+        if killed {
+            inner.killed = true;
+        }
+        if inner.state == ThreadState::Dying {
+            return;
+        }
+        inner.change_state(ThreadState::Dying);
+        if let Some(waker) = inner.waker.take() {
+            waker.wake();
+        }
+    }
+
+    /// Read one aspect of thread state.
+    pub fn read_state(&self, kind: ThreadStateKind, buf: &mut [u8]) -> ZxResult<usize> {
+        let inner = self.inner.lock();
+        let state = inner.state();
+        if state != ThreadState::BlockedException && state != ThreadState::Suspended {
+            return Err(ZxError::BAD_STATE);
+        }
+        let context = inner.context.as_ref().ok_or(ZxError::BAD_STATE)?;
+        context.read_state(kind, buf)
+    }
+
+    /// Write one aspect of thread state.
+    pub fn write_state(&self, kind: ThreadStateKind, buf: &[u8]) -> ZxResult {
+        let mut inner = self.inner.lock();
+        let state = inner.state();
+        if state != ThreadState::BlockedException && state != ThreadState::Suspended {
+            return Err(ZxError::BAD_STATE);
+        }
+        let context = inner.context.as_mut().ok_or(ZxError::BAD_STATE)?;
+        context.write_state(kind, buf)
+    }
+
+    /// Get the thread's information.
+    pub fn get_thread_info(&self) -> ThreadInfo {
+        let inner = self.inner.lock();
+        ThreadInfo {
+            state: inner.state() as u32,
+        }
+    }
+    /// Get the thread state.
+    pub fn state(&self) -> ThreadState {
+        self.inner.lock().state()
+    }
+
+    /// Add the parameter to the time this thread has run on cpu.
+    pub fn time_add(&self, time: u128) {
+        self.inner.lock().time += time;
+    }
+
+    /// Get the time this thread has run on cpu.
+    pub fn get_time(&self) -> u64 {
+        self.inner.lock().time as u64
+    }
+
+    /// Set this thread as the first thread of a process.
+    pub(super) fn set_first_thread(&self) {
+        self.inner.lock().first_thread = true;
+    }
+
+    /// Whether this thread is the first thread of a process.
+    pub fn is_first_thread(&self) -> bool {
+        self.inner.lock().first_thread
+    }
+
+    /// Get the thread's flags.
+    pub fn flags(&self) -> ThreadFlag {
+        self.inner.lock().flags
+    }
+
+    /// Apply `f` to the thread's flags.
+    pub fn update_flags(&self, f: impl FnOnce(&mut ThreadFlag)) {
+        f(&mut self.inner.lock().flags)
+    }
+
+    /// Set the thread local fsbase register on x86_64.
+    pub fn set_fsbase(&self, fsbase: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        let context = inner.context.as_mut().ok_or(ZxError::BAD_STATE)?;
+        context.general.fsbase = fsbase;
+        Ok(())
+    }
+
+    /// Set the thread local gsbase register on x86_64.
+    pub fn set_gsbase(&self, gsbase: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        let context = inner.context.as_mut().ok_or(ZxError::BAD_STATE)?;
+        context.general.gsbase = gsbase;
+        Ok(())
+    }
+}
+
+impl Task for Thread {
+    fn kill(&self) {
+        self.stop(true)
+    }
+
+    fn suspend(&self) {
+        let mut inner = self.inner.lock();
+        inner.suspend_count += 1;
+        // let state = inner.state;
+        // inner.change_state(state);
+    }
+
+    fn resume(&self) {
+        let mut inner = self.inner.lock();
+        assert_ne!(inner.suspend_count, 0);
+        inner.suspend_count -= 1;
+        if inner.suspend_count == 0 {
+            // let state = inner.state;
+            // inner.change_state(state);
+            if let Some(waker) = inner.waker.take() {
+                waker.wake();
+            }
+        }
+    }
+}
+
+/// A handle to current thread.
+///
+/// This is a wrapper of [`Thread`] that provides additional methods for the thread runner.
+/// It can only be obtained from the argument of `thread_fn` in a new thread started by [`Thread::start`].
+///
+/// It will terminate current thread on drop.
+///
+/// [`Thread`]: crate::task::Thread
+/// [`Thread::start`]: crate::task::Thread::start
+pub struct CurrentThread(pub(super) Arc<Thread>);
+
+impl Deref for CurrentThread {
+    type Target = Arc<Thread>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl Drop for CurrentThread {
+    /// Terminate the current running thread.
+    fn drop(&mut self) {
+        let mut inner = self.inner.lock();
+        inner.change_state(ThreadState::Dead);
+        self.proc().remove_thread(self.base.id);
+    }
+}
+
+impl CurrentThread {
+    /// Exit the current thread.
+    ///
+    /// The thread do not terminate immediately when exited. It is just made dying.
+    /// It will terminate after some cleanups on this struct drop.
+    pub fn exit(&self) {
+        self.stop(false);
+    }
+
+    /// Wait until the thread is ready to run (not suspended),
+    /// and then take away its context to run the thread.
+    pub fn wait_for_run(&self) -> impl Future<Output = Box<UserContext>> {
+        #[must_use = "wait_for_run does nothing unless polled/`await`-ed"]
+        struct RunnableChecker {
+            thread: Arc<Thread>,
+        }
+        impl Future for RunnableChecker {
+            type Output = Box<UserContext>;
+
+            fn poll(self: Pin<&mut Self>, cx: &mut Context) -> Poll<Self::Output> {
+                let mut inner = self.thread.inner.lock();
+                if inner.state() != ThreadState::Suspended {
+                    // resume:  return the context token from thread object
+                    // There is no need to call change_state here
+                    // since take away the context of a non-suspended thread won't change it's state
+                    Poll::Ready(inner.context.take().unwrap())
+                } else {
+                    // suspend: put waker into the thread object
+                    inner.waker = Some(cx.waker().clone());
+                    Poll::Pending
+                }
+            }
+        }
+        RunnableChecker {
+            thread: self.0.clone(),
+        }
+    }
+
+    /// The thread ends running and takes back the context.
+    pub fn end_running(&self, context: Box<UserContext>) {
+        let mut inner = self.inner.lock();
+        inner.context = Some(context);
+        // let state = inner.state;
+        // inner.change_state(state);
+    }
+
+    /// Access saved context of current thread.
+    ///
+    /// Will panic if the context is not availiable.
+    pub fn with_context<T, F>(&self, f: F) -> T
+    where
+        F: FnOnce(&mut UserContext) -> T,
+    {
+        let mut inner = self.inner.lock();
+        let mut cx = inner.context.as_mut().unwrap();
+        f(&mut cx)
+    }
+}
+
+/// The thread state.
+#[derive(Debug, Clone, Copy, Eq, PartialEq)]
+pub enum ThreadState {
+    /// The thread has been created but it has not started running yet.
+    New = 0,
+    /// The thread is running user code normally.
+    Running = 1,
+    /// Stopped due to `zx_task_suspend()`.
+    Suspended = 2,
+    /// In a syscall or handling an exception.
+    Blocked = 3,
+    /// The thread is in the process of being terminated, but it has not been stopped yet.
+    Dying = 4,
+    /// The thread has stopped running.
+    Dead = 5,
+    /// The thread is stopped in an exception.
+    BlockedException = 0x103,
+    /// The thread is stopped in `zx_nanosleep()`.
+    BlockedSleeping = 0x203,
+    /// The thread is stopped in `zx_futex_wait()`.
+    BlockedFutex = 0x303,
+    /// The thread is stopped in `zx_port_wait()`.
+    BlockedPort = 0x403,
+    /// The thread is stopped in `zx_channel_call()`.
+    BlockedChannel = 0x503,
+    /// The thread is stopped in `zx_object_wait_one()`.
+    BlockedWaitOne = 0x603,
+    /// The thread is stopped in `zx_object_wait_many()`.
+    BlockedWaitMany = 0x703,
+    /// The thread is stopped in `zx_interrupt_wait()`.
+    BlockedInterrupt = 0x803,
+    /// Pager.
+    BlockedPager = 0x903,
+}
+
+impl Default for ThreadState {
+    fn default() -> Self {
+        ThreadState::New
+    }
+}
+
+/// The thread information.
+#[repr(C)]
+pub struct ThreadInfo {
+    state: u32,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::job::Job;
+    use super::*;
+    use core::time::Duration;
+    use kernel_hal::timer_now;
+    use kernel_hal::GeneralRegs;
+
+    #[test]
+    fn create() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+        assert_eq!(thread.flags(), ThreadFlag::empty());
+
+        assert_eq!(thread.related_koid(), proc.id());
+        let child = proc.get_child(thread.id()).unwrap().downcast_arc().unwrap();
+        assert!(Arc::ptr_eq(&child, &thread));
+    }
+
+    #[async_std::test]
+    async fn start() {
+        kernel_hal_unix::init();
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+        let thread1 = Thread::create(&proc, "thread1").expect("failed to create thread");
+
+        // function for new thread
+        async fn new_thread(thread: CurrentThread) {
+            let cx = thread.wait_for_run().await;
+            assert_eq!(cx.general.rip, 1);
+            assert_eq!(cx.general.rsp, 4);
+            assert_eq!(cx.general.rdi, 3);
+            assert_eq!(cx.general.rsi, 2);
+            async_std::task::sleep(Duration::from_millis(10)).await;
+            thread.end_running(cx);
+        }
+
+        // start a new thread
+        let handle = Handle::new(proc.clone(), Rights::DEFAULT_PROCESS);
+        proc.start(&thread, 1, 4, Some(handle.clone()), 2, |thread| {
+            Box::pin(new_thread(thread))
+        })
+        .expect("failed to start thread");
+
+        // check info and state
+        let info = proc.get_info();
+        assert!(info.started && !info.has_exited && info.return_code == 0);
+        assert_eq!(proc.status(), Status::Running);
+        assert_eq!(thread.state(), ThreadState::Running);
+
+        // start again should fail
+        assert_eq!(
+            proc.start(&thread, 1, 4, Some(handle.clone()), 2, |thread| Box::pin(
+                new_thread(thread)
+            )),
+            Err(ZxError::BAD_STATE)
+        );
+
+        // start another thread should fail
+        assert_eq!(
+            proc.start(&thread1, 1, 4, Some(handle.clone()), 2, |thread| Box::pin(
+                new_thread(thread)
+            )),
+            Err(ZxError::BAD_STATE)
+        );
+
+        // wait 100ms for the new thread to exit
+        async_std::task::sleep(core::time::Duration::from_millis(100)).await;
+
+        // no other references to `Thread`
+        assert_eq!(Arc::strong_count(&thread), 1);
+        assert_eq!(thread.state(), ThreadState::Dead);
+    }
+
+    #[test]
+    fn info() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        let info = thread.get_thread_info();
+        assert!(info.state == thread.state() as u32);
+    }
+
+    #[test]
+    fn read_write_state() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        const SIZE: usize = core::mem::size_of::<GeneralRegs>();
+        let mut buf = [0; 10];
+        assert_eq!(
+            thread.read_state(ThreadStateKind::General, &mut buf).err(),
+            Some(ZxError::BAD_STATE)
+        );
+        assert_eq!(
+            thread.write_state(ThreadStateKind::General, &buf).err(),
+            Some(ZxError::BAD_STATE)
+        );
+
+        thread.suspend();
+
+        assert_eq!(
+            thread.read_state(ThreadStateKind::General, &mut buf).err(),
+            Some(ZxError::BUFFER_TOO_SMALL)
+        );
+        assert_eq!(
+            thread.write_state(ThreadStateKind::General, &buf).err(),
+            Some(ZxError::BUFFER_TOO_SMALL)
+        );
+
+        let mut buf = [0; SIZE];
+        assert!(thread
+            .read_state(ThreadStateKind::General, &mut buf)
+            .is_ok());
+        assert!(thread.write_state(ThreadStateKind::General, &buf).is_ok());
+        // TODO
+    }
+
+    #[async_std::test]
+    async fn wait_for_run() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        assert_eq!(thread.state(), ThreadState::New);
+
+        thread
+            .start(0, 0, 0, 0, |thread| Box::pin(new_thread(thread)))
+            .unwrap();
+        async fn new_thread(thread: CurrentThread) {
+            assert_eq!(thread.state(), ThreadState::Running);
+
+            // without suspend
+            let context = thread.wait_for_run().await;
+            thread.end_running(context);
+
+            // with suspend
+            thread.suspend();
+            thread.suspend();
+            assert_eq!(thread.state(), ThreadState::Suspended);
+            async_std::task::spawn({
+                let thread = (*thread).clone();
+                async move {
+                    async_std::task::sleep(Duration::from_millis(10)).await;
+                    thread.resume();
+                    async_std::task::sleep(Duration::from_millis(10)).await;
+                    thread.resume();
+                }
+            });
+            let time = timer_now();
+            let _context = thread.wait_for_run().await;
+            assert!(timer_now() - time >= Duration::from_millis(20));
+        }
+        // FIX ME
+        // let thread: Arc<dyn KernelObject> = thread;
+        // thread.wait_signal(Signal::THREAD_TERMINATED).await;
+    }
+
+    #[test]
+    fn time() {
+        let root_job = Job::root();
+        let proc = Process::create(&root_job, "proc").expect("failed to create process");
+        let thread = Thread::create(&proc, "thread").expect("failed to create thread");
+
+        assert_eq!(thread.get_time(), 0);
+        thread.time_add(10);
+        assert_eq!(thread.get_time(), 10);
+    }
+}

code/ch03-04/object/src/task/thread/thread_state.rs

@@ -0,0 +1,64 @@
+use crate::{ZxError, ZxResult};
+use kernel_hal::UserContext;
+use numeric_enum_macro::numeric_enum;
+
+numeric_enum! {
+    #[repr(u32)]
+    /// Possible values for "kind" in zx_thread_read_state and zx_thread_write_state.
+    #[allow(missing_docs)]
+    #[derive(Debug, Copy, Clone)]
+    pub enum ThreadStateKind {
+        General = 0,
+        FS = 6,
+        GS = 7,
+    }
+}
+
+pub(super) trait ContextExt {
+    fn read_state(&self, kind: ThreadStateKind, buf: &mut [u8]) -> ZxResult<usize>;
+    fn write_state(&mut self, kind: ThreadStateKind, buf: &[u8]) -> ZxResult;
+}
+
+impl ContextExt for UserContext {
+    fn read_state(&self, kind: ThreadStateKind, buf: &mut [u8]) -> ZxResult<usize> {
+        match kind {
+            ThreadStateKind::General => buf.write_struct(&self.general),
+            ThreadStateKind::FS => buf.write_struct(&self.general.fsbase),
+            ThreadStateKind::GS => buf.write_struct(&self.general.gsbase),
+        }
+    }
+
+    fn write_state(&mut self, kind: ThreadStateKind, buf: &[u8]) -> ZxResult {
+        match kind {
+            ThreadStateKind::General => self.general = buf.read_struct()?,
+            ThreadStateKind::FS => self.general.fsbase = buf.read_struct()?,
+            ThreadStateKind::GS => self.general.gsbase = buf.read_struct()?,
+        }
+        Ok(())
+    }
+}
+
+trait BufExt {
+    fn read_struct<T>(&self) -> ZxResult<T>;
+    fn write_struct<T: Copy>(&mut self, value: &T) -> ZxResult<usize>;
+}
+
+#[allow(unsafe_code)]
+impl BufExt for [u8] {
+    fn read_struct<T>(&self) -> ZxResult<T> {
+        if self.len() < core::mem::size_of::<T>() {
+            return Err(ZxError::BUFFER_TOO_SMALL);
+        }
+        Ok(unsafe { (self.as_ptr() as *const T).read() })
+    }
+
+    fn write_struct<T: Copy>(&mut self, value: &T) -> ZxResult<usize> {
+        if self.len() < core::mem::size_of::<T>() {
+            return Err(ZxError::BUFFER_TOO_SMALL);
+        }
+        unsafe {
+            *(self.as_mut_ptr() as *mut T) = *value;
+        }
+        Ok(core::mem::size_of::<T>())
+    }
+}

code/ch03-04/object/src/util.rs

@@ -0,0 +1,102 @@
+#![allow(dead_code)]
+
+/// Given a range and iterate sub-range for each block
+pub struct BlockIter {
+    pub begin: usize,
+    pub end: usize,
+    pub block_size_log2: u8,
+}
+
+#[derive(Debug, Eq, PartialEq)]
+pub struct BlockRange {
+    pub block: usize,
+    pub begin: usize,
+    pub end: usize,
+    pub block_size_log2: u8,
+}
+
+impl BlockRange {
+    pub fn len(&self) -> usize {
+        self.end - self.begin
+    }
+    pub fn is_full(&self) -> bool {
+        self.len() == (1usize << self.block_size_log2)
+    }
+    pub fn is_empty(&self) -> bool {
+        self.len() == 0
+    }
+    pub fn origin_begin(&self) -> usize {
+        (self.block << self.block_size_log2) + self.begin
+    }
+    pub fn origin_end(&self) -> usize {
+        (self.block << self.block_size_log2) + self.end
+    }
+}
+
+impl Iterator for BlockIter {
+    type Item = BlockRange;
+
+    fn next(&mut self) -> Option<<Self as Iterator>::Item> {
+        if self.begin >= self.end {
+            return None;
+        }
+        let block_size_log2 = self.block_size_log2;
+        let block_size = 1usize << self.block_size_log2;
+        let block = self.begin / block_size;
+        let begin = self.begin % block_size;
+        let end = if block == self.end / block_size {
+            self.end % block_size
+        } else {
+            block_size
+        };
+        self.begin += end - begin;
+        Some(BlockRange {
+            block,
+            begin,
+            end,
+            block_size_log2,
+        })
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn block_iter() {
+        let mut iter = BlockIter {
+            begin: 0x123,
+            end: 0x2018,
+            block_size_log2: 12,
+        };
+        assert_eq!(
+            iter.next(),
+            Some(BlockRange {
+                block: 0,
+                begin: 0x123,
+                end: 0x1000,
+                block_size_log2: 12
+            })
+        );
+        assert_eq!(
+            iter.next(),
+            Some(BlockRange {
+                block: 1,
+                begin: 0,
+                end: 0x1000,
+                block_size_log2: 12
+            })
+        );
+        assert_eq!(
+            iter.next(),
+            Some(BlockRange {
+                block: 2,
+                begin: 0,
+                end: 0x18,
+                block_size_log2: 12
+            })
+        );
+        assert_eq!(iter.next(), None);
+    }
+}

code/ch03-04/object/src/vm/mod.rs

@@ -0,0 +1,52 @@
+//! Objects for Virtual Memory Management.
+
+mod vmar;
+mod vmo;
+
+pub use self::{vmar::*, vmo::*};
+
+/// Physical Address
+pub type PhysAddr = usize;
+
+/// Virtual Address
+pub type VirtAddr = usize;
+
+/// Device Address
+pub type DevVAddr = usize;
+
+/// Size of a page
+pub const PAGE_SIZE: usize = 0x1000;
+
+/// log2(PAGE_SIZE)
+pub const PAGE_SIZE_LOG2: usize = 12;
+
+/// Check whether `x` is a multiple of `PAGE_SIZE`.
+pub fn page_aligned(x: usize) -> bool {
+    check_aligned(x, PAGE_SIZE)
+}
+
+/// Check whether `x` is a multiple of `align`.
+pub fn check_aligned(x: usize, align: usize) -> bool {
+    x % align == 0
+}
+
+/// How many pages the `size` needs.
+/// To avoid overflow and pass more unit tests, use wrapping add
+pub fn pages(size: usize) -> usize {
+    ceil(size, PAGE_SIZE)
+}
+
+/// How many `align` the `x` needs.
+pub fn ceil(x: usize, align: usize) -> usize {
+    x.wrapping_add(align - 1) / align
+}
+
+/// Round up `size` to a multiple of `PAGE_SIZE`.
+pub fn roundup_pages(size: usize) -> usize {
+    pages(size) * PAGE_SIZE
+}
+
+/// Round down `size` to a multiple of `PAGE_SIZE`.
+pub fn round_down_pages(size: usize) -> usize {
+    size / PAGE_SIZE * PAGE_SIZE
+}

code/ch03-04/object/src/vm/vmar.rs

@@ -0,0 +1,849 @@
+use {
+    super::*,
+    crate::object::*,
+    alloc::sync::Arc,
+    alloc::vec,
+    alloc::vec::Vec,
+    bitflags::bitflags,
+    kernel_hal::{MMUFlags, PageTableTrait},
+    spin::Mutex,
+};
+
+bitflags! {
+    /// Creation flags for VmAddressRegion.
+    pub struct VmarFlags: u32 {
+        #[allow(clippy::identity_op)]
+        /// When randomly allocating subregions, reduce sprawl by placing allocations
+        /// near each other.
+        const COMPACT               = 1 << 0;
+        /// Request that the new region be at the specified offset in its parent region.
+        const SPECIFIC              = 1 << 1;
+        /// Like SPECIFIC, but permits overwriting existing mappings.  This
+        /// flag will not overwrite through a subregion.
+        const SPECIFIC_OVERWRITE    = 1 << 2;
+        /// Allow VmMappings to be created inside the new region with the SPECIFIC or
+        /// OFFSET_IS_UPPER_LIMIT flag.
+        const CAN_MAP_SPECIFIC      = 1 << 3;
+        /// Allow VmMappings to be created inside the region with read permissions.
+        const CAN_MAP_READ          = 1 << 4;
+        /// Allow VmMappings to be created inside the region with write permissions.
+        const CAN_MAP_WRITE         = 1 << 5;
+        /// Allow VmMappings to be created inside the region with execute permissions.
+        const CAN_MAP_EXECUTE       = 1 << 6;
+        /// Require that VMO backing the mapping is non-resizable.
+        const REQUIRE_NON_RESIZABLE = 1 << 7;
+        /// Treat the offset as an upper limit when allocating a VMO or child VMAR.
+        const ALLOW_FAULTS          = 1 << 8;
+
+        /// Allow VmMappings to be created inside the region with read, write and execute permissions.
+        const CAN_MAP_RXW           = Self::CAN_MAP_READ.bits | Self::CAN_MAP_EXECUTE.bits | Self::CAN_MAP_WRITE.bits;
+        /// Creation flags for root VmAddressRegion
+        const ROOT_FLAGS            = Self::CAN_MAP_RXW.bits | Self::CAN_MAP_SPECIFIC.bits;
+    }
+}
+
+/// Virtual Memory Address Regions
+pub struct VmAddressRegion {
+    flags: VmarFlags,
+    base: KObjectBase,
+    addr: VirtAddr,
+    size: usize,
+    parent: Option<Arc<VmAddressRegion>>,
+    page_table: Arc<Mutex<dyn PageTableTrait>>,
+    /// If inner is None, this region is destroyed, all operations are invalid.
+    inner: Mutex<Option<VmarInner>>,
+}
+
+impl_kobject!(VmAddressRegion);
+
+/// The mutable part of `VmAddressRegion`.
+#[derive(Default)]
+struct VmarInner {
+    children: Vec<Arc<VmAddressRegion>>,
+    mappings: Vec<Arc<VmMapping>>,
+}
+
+impl VmAddressRegion {
+    /// Create a new root VMAR.
+    pub fn new_root() -> Arc<Self> {
+        let (addr, size) = {
+            use core::sync::atomic::*;
+            static VMAR_ID: AtomicUsize = AtomicUsize::new(0);
+            let i = VMAR_ID.fetch_add(1, Ordering::SeqCst);
+            (0x2_0000_0000 + 0x100_0000_0000 * i, 0x100_0000_0000)
+        };
+        Arc::new(VmAddressRegion {
+            flags: VmarFlags::ROOT_FLAGS,
+            base: KObjectBase::new(),
+            addr,
+            size,
+            parent: None,
+            page_table: Arc::new(Mutex::new(kernel_hal::PageTable::new())), //hal PageTable
+            inner: Mutex::new(Some(VmarInner::default())),
+        })
+    }
+
+    /// Create a kernel root VMAR.
+    pub fn new_kernel() -> Arc<Self> {
+        let kernel_vmar_base = KERNEL_ASPACE_BASE as usize;
+        let kernel_vmar_size = KERNEL_ASPACE_SIZE as usize;
+        Arc::new(VmAddressRegion {
+            flags: VmarFlags::ROOT_FLAGS,
+            base: KObjectBase::new(),
+            addr: kernel_vmar_base,
+            size: kernel_vmar_size,
+            parent: None,
+            page_table: Arc::new(Mutex::new(kernel_hal::PageTable::new())),
+            inner: Mutex::new(Some(VmarInner::default())),
+        })
+    }
+
+    /// Create a child VMAR at the `offset`.
+    pub fn allocate_at(
+        self: &Arc<Self>,
+        offset: usize,
+        len: usize,
+        flags: VmarFlags,
+        align: usize,
+    ) -> ZxResult<Arc<Self>> {
+        self.allocate(Some(offset), len, flags, align)
+    }
+
+    /// Create a child VMAR with optional `offset`.
+    pub fn allocate(
+        self: &Arc<Self>,
+        offset: Option<usize>,
+        len: usize,
+        flags: VmarFlags,
+        align: usize,
+    ) -> ZxResult<Arc<Self>> {
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        let offset = self.determine_offset(inner, offset, len, align)?;
+        let child = Arc::new(VmAddressRegion {
+            flags,
+            base: KObjectBase::new(),
+            addr: self.addr + offset,
+            size: len,
+            parent: Some(self.clone()),
+            page_table: self.page_table.clone(),
+            inner: Mutex::new(Some(VmarInner::default())),
+        });
+        inner.children.push(child.clone());
+        Ok(child)
+    }
+
+    /// Map the `vmo` into this VMAR at given `offset`.
+    pub fn map_at(
+        &self,
+        vmar_offset: usize,
+        vmo: Arc<VmObject>,
+        vmo_offset: usize,
+        len: usize,
+        flags: MMUFlags,
+    ) -> ZxResult<VirtAddr> {
+        self.map(Some(vmar_offset), vmo, vmo_offset, len, flags)
+    }
+
+    /// Map the `vmo` into this VMAR.
+    pub fn map(
+        &self,
+        vmar_offset: Option<usize>,
+        vmo: Arc<VmObject>,
+        vmo_offset: usize,
+        len: usize,
+        flags: MMUFlags,
+    ) -> ZxResult<VirtAddr> {
+        self.map_ext(
+            vmar_offset,
+            vmo,
+            vmo_offset,
+            len,
+            MMUFlags::RXW,
+            flags,
+            false,
+            true,
+        )
+    }
+
+    /// Map the `vmo` into this VMAR.
+    #[allow(clippy::too_many_arguments)]
+    pub fn map_ext(
+        &self,
+        vmar_offset: Option<usize>,
+        vmo: Arc<VmObject>,
+        vmo_offset: usize,
+        len: usize,
+        permissions: MMUFlags,
+        flags: MMUFlags,
+        overwrite: bool,
+        map_range: bool,
+    ) -> ZxResult<VirtAddr> {
+        if !page_aligned(vmo_offset) || !page_aligned(len) || vmo_offset.overflowing_add(len).1 {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        if !permissions.contains(flags & MMUFlags::RXW) {
+            return Err(ZxError::ACCESS_DENIED);
+        }
+        if vmo_offset > vmo.len() || len > vmo.len() - vmo_offset {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        // Simplify: overwrite == false && map_range == true
+        if overwrite || !map_range {
+            warn!("Simplify: overwrite == false && map_range == true");
+            return Err(ZxError::INVALID_ARGS);
+        }
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        let offset = self.determine_offset(inner, vmar_offset, len, PAGE_SIZE)?;
+        let addr = self.addr + offset;
+        let flags = flags | MMUFlags::from_bits_truncate(vmo.cache_policy() as u32 as usize);
+        // align = 1K? 2K? 4K? 8K? ...
+        if !self.test_map(inner, offset, len, PAGE_SIZE) {
+            return Err(ZxError::NO_MEMORY);
+        }
+        let mapping = VmMapping::new(
+            addr,
+            len,
+            vmo,
+            vmo_offset,
+            permissions,
+            flags,
+            self.page_table.clone(),
+        );
+        mapping.map()?;
+        inner.mappings.push(mapping);
+        Ok(addr)
+    }
+
+    /// Unmaps all VMO mappings and destroys all sub-regions within the absolute range
+    /// including `addr` and ending before exclusively at `addr + len`.
+    /// Any sub-region that is in the range must be fully in the range
+    /// (i.e. partial overlaps are an error).
+    /// NOT SUPPORT:
+    /// If a mapping is only partially in the range, the mapping is split and the requested
+    /// portion is unmapped.
+    pub fn unmap(&self, addr: VirtAddr, len: usize) -> ZxResult {
+        if !page_aligned(addr) || !page_aligned(len) || len == 0 {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        let begin = addr;
+        let end = addr + len;
+        // check partial overlapped sub-regions
+        if inner
+            .children
+            .iter()
+            .any(|vmar| vmar.partial_overlap(begin, end))
+        {
+            return Err(ZxError::INVALID_ARGS);
+        }
+
+        if inner
+            .mappings
+            .iter()
+            .any(|map| map.partial_overlap(begin, end))
+        {
+            warn!("Simplify: Not support partial unmap.");
+            return Err(ZxError::INVALID_ARGS);
+        }
+        inner.mappings.drain_filter(|map| map.within(begin, end));
+
+        for vmar in inner.children.drain_filter(|vmar| vmar.within(begin, end)) {
+            vmar.destroy_internal()?;
+        }
+        Ok(())
+    }
+
+    /// Change protections on a subset of the region of memory in the containing
+    /// address space.  If the requested range overlaps with a subregion,
+    /// protect() will fail.
+    pub fn protect(&self, addr: usize, len: usize, flags: MMUFlags) -> ZxResult {
+        if !page_aligned(addr) || !page_aligned(len) {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        let end_addr = addr + len;
+        // check if there are overlapping subregion
+        if inner
+            .children
+            .iter()
+            .any(|child| child.overlap(addr, end_addr))
+        {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        let length = inner.mappings.iter().fold(0, |acc, map| {
+            acc + end_addr
+                .min(map.end_addr())
+                .checked_sub(addr.max(map.addr()))
+                .unwrap_or(0)
+        });
+        if length != len {
+            return Err(ZxError::NOT_FOUND);
+        }
+        // check if protect flags is valid
+        if inner
+            .mappings
+            .iter()
+            .filter(|map| map.overlap(addr, end_addr)) // get mappings in range: [addr, end_addr]
+            .any(|map| !map.is_valid_mapping_flags(flags))
+        {
+            return Err(ZxError::ACCESS_DENIED);
+        }
+        inner
+            .mappings
+            .iter()
+            .filter(|map| map.overlap(addr, end_addr))
+            .for_each(|map| {
+                let start_index = pages(addr.max(map.addr()) - map.addr());
+                let end_index = pages(end_addr.min(map.end_addr()) - map.addr());
+                map.protect(flags, start_index, end_index);
+            });
+        Ok(())
+    }
+
+    /// Unmap all mappings and destroy all sub-regions of VMAR.
+    pub fn clear(&self) -> ZxResult {
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        for vmar in inner.children.drain(..) {
+            vmar.destroy_internal()?;
+        }
+        inner.mappings.clear();
+        Ok(())
+    }
+
+    /// Destroy but do not remove self from parent.
+    fn destroy_internal(&self) -> ZxResult {
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+        for vmar in inner.children.drain(..) {
+            vmar.destroy_internal()?;
+        }
+        inner.mappings.clear();
+        *guard = None;
+        Ok(())
+    }
+
+    /// Unmap all mappings within the VMAR, and destroy all sub-regions of the region.
+    pub fn destroy(self: &Arc<Self>) -> ZxResult {
+        self.destroy_internal()?;
+        // remove from parent
+        if let Some(parent) = &self.parent {
+            let mut guard = parent.inner.lock();
+            let inner = guard.as_mut().ok_or(ZxError::BAD_STATE)?;
+            inner.children.retain(|vmar| !Arc::ptr_eq(self, vmar));
+        }
+        Ok(())
+    }
+
+    /// Get physical address of the underlying page table.
+    pub fn table_phys(&self) -> PhysAddr {
+        self.page_table.lock().table_phys()
+    }
+
+    /// Get start address of this VMAR.
+    pub fn addr(&self) -> usize {
+        self.addr
+    }
+
+    /// Whether this VMAR is dead.
+    pub fn is_dead(&self) -> bool {
+        self.inner.lock().is_none()
+    }
+
+    /// Whether this VMAR is alive.
+    pub fn is_alive(&self) -> bool {
+        !self.is_dead()
+    }
+
+    /// Determine final address with given input `offset` and `len`.
+    fn determine_offset(
+        &self,
+        inner: &VmarInner,
+        offset: Option<usize>,
+        len: usize,
+        align: usize,
+    ) -> ZxResult<VirtAddr> {
+        if !check_aligned(len, align) {
+            Err(ZxError::INVALID_ARGS)
+        } else if let Some(offset) = offset {
+            if check_aligned(offset, align) && self.test_map(inner, offset, len, align) {
+                Ok(offset)
+            } else {
+                Err(ZxError::INVALID_ARGS)
+            }
+        } else if len > self.size {
+            Err(ZxError::INVALID_ARGS)
+        } else {
+            match self.find_free_area(inner, 0, len, align) {
+                Some(offset) => Ok(offset),
+                None => Err(ZxError::NO_MEMORY),
+            }
+        }
+    }
+
+    /// Test if can create a new mapping at `offset` with `len`.
+    fn test_map(&self, inner: &VmarInner, offset: usize, len: usize, align: usize) -> bool {
+        debug_assert!(check_aligned(offset, align));
+        debug_assert!(check_aligned(len, align));
+        let begin = self.addr + offset;
+        let end = begin + len;
+        if end > self.addr + self.size {
+            return false;
+        }
+        // brute force
+        if inner.children.iter().any(|vmar| vmar.overlap(begin, end)) {
+            return false;
+        }
+        if inner.mappings.iter().any(|map| map.overlap(begin, end)) {
+            return false;
+        }
+        true
+    }
+
+    /// Find a free area with `len`.
+    fn find_free_area(
+        &self,
+        inner: &VmarInner,
+        offset_hint: usize,
+        len: usize,
+        align: usize,
+    ) -> Option<usize> {
+        // TODO: randomize
+        debug_assert!(check_aligned(offset_hint, align));
+        debug_assert!(check_aligned(len, align));
+        // brute force:
+        // try each area's end address as the start
+        core::iter::once(offset_hint)
+            .chain(inner.children.iter().map(|map| map.end_addr() - self.addr))
+            .chain(inner.mappings.iter().map(|map| map.end_addr() - self.addr))
+            .find(|&offset| self.test_map(inner, offset, len, align))
+    }
+
+    fn end_addr(&self) -> VirtAddr {
+        self.addr + self.size
+    }
+
+    fn overlap(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        !(self.addr >= end || self.end_addr() <= begin)
+    }
+
+    fn within(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        begin <= self.addr && self.end_addr() <= end
+    }
+
+    fn partial_overlap(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        self.overlap(begin, end) && !self.within(begin, end)
+    }
+
+    fn contains(&self, vaddr: VirtAddr) -> bool {
+        self.addr <= vaddr && vaddr < self.end_addr()
+    }
+
+    /// Get information of this VmAddressRegion
+    pub fn get_info(&self) -> VmarInfo {
+        // pub fn get_info(&self, va: usize) -> VmarInfo {
+        // let _r = self.page_table.lock().query(va);
+        VmarInfo {
+            base: self.addr(),
+            len: self.size,
+        }
+    }
+
+    /// Get VmarFlags of this VMAR.
+    pub fn get_flags(&self) -> VmarFlags {
+        self.flags
+    }
+
+    #[cfg(test)]
+    fn count(&self) -> usize {
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().unwrap();
+        println!("m = {}, c = {}", inner.mappings.len(), inner.children.len());
+        inner.mappings.len() + inner.children.len()
+    }
+
+    #[cfg(test)]
+    fn used_size(&self) -> usize {
+        let mut guard = self.inner.lock();
+        let inner = guard.as_mut().unwrap();
+        let map_size: usize = inner.mappings.iter().map(|map| map.size()).sum();
+        let vmar_size: usize = inner.children.iter().map(|vmar| vmar.size).sum();
+        println!("size = {:#x?}", map_size + vmar_size);
+        map_size + vmar_size
+    }
+}
+
+/// Information of a VmAddressRegion.
+#[repr(C)]
+#[derive(Debug)]
+pub struct VmarInfo {
+    base: usize,
+    len: usize,
+}
+
+/// Virtual Memory Mapping
+pub struct VmMapping {
+    /// The permission limitation of the vmar
+    permissions: MMUFlags,
+    vmo: Arc<VmObject>,
+    page_table: Arc<Mutex<dyn PageTableTrait>>,
+    inner: Mutex<VmMappingInner>,
+}
+
+#[derive(Debug, Clone)]
+struct VmMappingInner {
+    /// The actual flags used in the mapping of each page
+    flags: Vec<MMUFlags>,
+    addr: VirtAddr,
+    size: usize,
+    vmo_offset: usize,
+}
+
+impl core::fmt::Debug for VmMapping {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        let inner = self.inner.lock();
+        f.debug_struct("VmMapping")
+            .field("addr", &inner.addr)
+            .field("size", &inner.size)
+            .field("permissions", &self.permissions)
+            .field("flags", &inner.flags)
+            .field("vmo_id", &self.vmo.id())
+            .field("vmo_offset", &inner.vmo_offset)
+            .finish()
+    }
+}
+
+impl VmMapping {
+    fn new(
+        addr: VirtAddr,
+        size: usize,
+        vmo: Arc<VmObject>,
+        vmo_offset: usize,
+        permissions: MMUFlags,
+        flags: MMUFlags,
+        page_table: Arc<Mutex<dyn PageTableTrait>>,
+    ) -> Arc<Self> {
+        let mapping = Arc::new(VmMapping {
+            inner: Mutex::new(VmMappingInner {
+                flags: vec![flags; pages(size)],
+                addr,
+                size,
+                vmo_offset,
+            }),
+            permissions,
+            page_table,
+            vmo: vmo.clone(),
+        });
+        vmo.append_mapping(Arc::downgrade(&mapping));
+        mapping
+    }
+
+    /// Map range and commit.
+    /// Commit pages to vmo, and map those to frames in page_table.
+    /// Temporarily used for development. A standard procedure for
+    /// vmo is: create_vmo, op_range(commit), map
+    fn map(self: &Arc<Self>) -> ZxResult {
+        self.vmo.commit_pages_with(&mut |commit| {
+            let inner = self.inner.lock();
+            let mut page_table = self.page_table.lock();
+            let page_num = inner.size / PAGE_SIZE;
+            let vmo_offset = inner.vmo_offset / PAGE_SIZE;
+            for i in 0..page_num {
+                let paddr = commit(vmo_offset + i, inner.flags[i])?;
+                //通过 PageTableTrait 的 hal_pt_map 进行页表映射
+                page_table
+                    .map(inner.addr + i * PAGE_SIZE, paddr, inner.flags[i])
+                    .expect("failed to map");
+            }
+            Ok(())
+        })
+    }
+
+    fn unmap(&self) {
+        let inner = self.inner.lock();
+        let pages = inner.size / PAGE_SIZE;
+        // TODO inner.vmo_offset unused?
+        self.page_table
+            .lock()
+            .unmap_cont(inner.addr, pages)
+            .expect("failed to unmap")
+    }
+
+    fn overlap(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        let inner = self.inner.lock();
+        !(inner.addr >= end || inner.end_addr() <= begin)
+    }
+
+    fn within(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        let inner = self.inner.lock();
+        begin <= inner.addr && inner.end_addr() <= end
+    }
+
+    fn partial_overlap(&self, begin: VirtAddr, end: VirtAddr) -> bool {
+        self.overlap(begin, end) && !self.within(begin, end)
+    }
+
+    fn contains(&self, vaddr: VirtAddr) -> bool {
+        let inner = self.inner.lock();
+        inner.addr <= vaddr && vaddr < inner.end_addr()
+    }
+
+    fn is_valid_mapping_flags(&self, flags: MMUFlags) -> bool {
+        self.permissions.contains(flags & MMUFlags::RXW)
+    }
+
+    fn protect(&self, flags: MMUFlags, start_index: usize, end_index: usize) {
+        let mut inner = self.inner.lock();
+        let mut pg_table = self.page_table.lock();
+        for i in start_index..end_index {
+            inner.flags[i] = (inner.flags[i] & !MMUFlags::RXW) | (flags & MMUFlags::RXW);
+            pg_table
+                .protect(inner.addr + i * PAGE_SIZE, inner.flags[i])
+                .unwrap();
+        }
+    }
+
+    fn size(&self) -> usize {
+        self.inner.lock().size
+    }
+
+    fn addr(&self) -> VirtAddr {
+        self.inner.lock().addr
+    }
+
+    fn end_addr(&self) -> VirtAddr {
+        self.inner.lock().end_addr()
+    }
+
+    /// Get MMUFlags of this VmMapping.
+    pub fn get_flags(&self, vaddr: usize) -> ZxResult<MMUFlags> {
+        if self.contains(vaddr) {
+            let page_id = (vaddr - self.addr()) / PAGE_SIZE;
+            Ok(self.inner.lock().flags[page_id])
+        } else {
+            Err(ZxError::NO_MEMORY)
+        }
+    }
+}
+
+impl VmMappingInner {
+    fn end_addr(&self) -> VirtAddr {
+        self.addr + self.size
+    }
+}
+
+impl Drop for VmMapping {
+    fn drop(&mut self) {
+        self.unmap();
+    }
+}
+
+/// The base of kernel address space
+/// In x86 fuchsia this is 0xffff_ff80_0000_0000 instead
+pub const KERNEL_ASPACE_BASE: u64 = 0xffff_ff02_0000_0000;
+/// The size of kernel address space
+pub const KERNEL_ASPACE_SIZE: u64 = 0x0000_0080_0000_0000;
+/// The base of user address space
+pub const USER_ASPACE_BASE: u64 = 0;
+// pub const USER_ASPACE_BASE: u64 = 0x0000_0000_0100_0000;
+/// The size of user address space
+pub const USER_ASPACE_SIZE: u64 = (1u64 << 47) - 4096 - USER_ASPACE_BASE;
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn create_child() {
+        let root_vmar = VmAddressRegion::new_root();
+        let child = root_vmar
+            .allocate_at(0, 0x2000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+            .expect("failed to create child VMAR");
+
+        // test invalid argument
+        assert_eq!(
+            root_vmar
+                .allocate_at(0x2001, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .err(),
+            Some(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            root_vmar
+                .allocate_at(0x2000, 1, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .err(),
+            Some(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            root_vmar
+                .allocate_at(0, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .err(),
+            Some(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            child
+                .allocate_at(0x1000, 0x2000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .err(),
+            Some(ZxError::INVALID_ARGS)
+        );
+    }
+
+    /// A valid virtual address base to mmap.
+    const MAGIC: usize = 0xdead_beaf;
+
+    #[test]
+    #[allow(unsafe_code)]
+    fn map() {
+        let vmar = VmAddressRegion::new_root();
+        let vmo = VmObject::new_paged(4);
+        let flags = MMUFlags::READ | MMUFlags::WRITE;
+
+        // invalid argument
+        assert_eq!(
+            vmar.map_at(0, vmo.clone(), 0x4000, 0x1000, flags),
+            Err(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            vmar.map_at(0, vmo.clone(), 0, 0x5000, flags),
+            Err(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            vmar.map_at(0, vmo.clone(), 0x1000, 1, flags),
+            Err(ZxError::INVALID_ARGS)
+        );
+        assert_eq!(
+            vmar.map_at(0, vmo.clone(), 1, 0x1000, flags),
+            Err(ZxError::INVALID_ARGS)
+        );
+        vmar.map_at(0, vmo.clone(), 0, 0x4000, flags).unwrap();
+        vmar.map_at(0x12000, vmo.clone(), 0x2000, 0x1000, flags)
+            .unwrap();
+        unsafe {
+            ((vmar.addr() + 0x2000) as *mut usize).write(MAGIC);
+            assert_eq!(((vmar.addr() + 0x12000) as *const usize).read(), MAGIC);
+        }
+    }
+
+    /// ```text
+    /// +--------+--------+--------+--------+
+    /// |           root              ....  |
+    /// +--------+--------+--------+--------+
+    /// |      child1     | child2 |
+    /// +--------+--------+--------+
+    /// | g-son1 | g-son2 |
+    /// +--------+--------+
+    /// ```
+    struct Sample {
+        root: Arc<VmAddressRegion>,
+        child1: Arc<VmAddressRegion>,
+        child2: Arc<VmAddressRegion>,
+        grandson1: Arc<VmAddressRegion>,
+        grandson2: Arc<VmAddressRegion>,
+    }
+
+    impl Sample {
+        fn new() -> Self {
+            let root = VmAddressRegion::new_root();
+            let child1 = root
+                .allocate_at(0, 0x2000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .unwrap();
+            let child2 = root
+                .allocate_at(0x2000, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .unwrap();
+            let grandson1 = child1
+                .allocate_at(0, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .unwrap();
+            let grandson2 = child1
+                .allocate_at(0x1000, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+                .unwrap();
+            Sample {
+                root,
+                child1,
+                child2,
+                grandson1,
+                grandson2,
+            }
+        }
+    }
+
+    #[test]
+    fn unmap_vmar() {
+        let s = Sample::new();
+        let base = s.root.addr();
+        s.child1.unmap(base, 0x1000).unwrap();
+        assert!(s.grandson1.is_dead());
+        assert!(s.grandson2.is_alive());
+
+        // partial overlap sub-region should fail.
+        let s = Sample::new();
+        let base = s.root.addr();
+        assert_eq!(
+            s.root.unmap(base + 0x1000, 0x2000),
+            Err(ZxError::INVALID_ARGS)
+        );
+
+        // unmap nothing should success.
+        let s = Sample::new();
+        let base = s.root.addr();
+        s.child1.unmap(base + 0x8000, 0x1000).unwrap();
+    }
+
+    #[test]
+    fn destroy() {
+        let s = Sample::new();
+        s.child1.destroy().unwrap();
+        assert!(s.child1.is_dead());
+        assert!(s.grandson1.is_dead());
+        assert!(s.grandson2.is_dead());
+        assert!(s.child2.is_alive());
+        // address space should be released
+        assert!(s
+            .root
+            .allocate_at(0, 0x1000, VmarFlags::CAN_MAP_RXW, PAGE_SIZE)
+            .is_ok());
+    }
+
+    #[test]
+    fn unmap_mapping() {
+        //   +--------+--------+--------+--------+--------+
+        // 1 [--------------------------|xxxxxxxx|--------]
+        // 2 [xxxxxxxx|-----------------]
+        // 3          [--------|xxxxxxxx]
+        // 4          [xxxxxxxx]
+        let vmar = VmAddressRegion::new_root();
+        let base = vmar.addr();
+        let vmo = VmObject::new_paged(5);
+        let flags = MMUFlags::READ | MMUFlags::WRITE;
+        vmar.map_at(0, vmo, 0, 0x5000, flags).unwrap();
+        assert_eq!(vmar.count(), 1);
+        assert_eq!(vmar.used_size(), 0x5000);
+
+        // 0. unmap none.
+        vmar.unmap(base + 0x5000, 0x1000).unwrap();
+        assert_eq!(vmar.count(), 1);
+        assert_eq!(vmar.used_size(), 0x5000);
+
+        // // 1. unmap middle.
+        // vmar.unmap(base + 0x3000, 0x1000).unwrap();
+        // assert_eq!(vmar.count(), 2);
+        // assert_eq!(vmar.used_size(), 0x4000);
+
+        // // 2. unmap prefix.
+        // vmar.unmap(base, 0x1000).unwrap();
+        // assert_eq!(vmar.count(), 2);
+        // assert_eq!(vmar.used_size(), 0x3000);
+
+        // // 3. unmap postfix.
+        // vmar.unmap(base + 0x2000, 0x1000).unwrap();
+        // assert_eq!(vmar.count(), 2);
+        // assert_eq!(vmar.used_size(), 0x2000);
+
+        // 4. unmap all.
+        vmar.unmap(base, 0x5000).unwrap();
+        assert_eq!(vmar.count(), 0);
+        assert_eq!(vmar.used_size(), 0x0);
+    }
+}

code/ch03-04/object/src/vm/vmo/mod.rs

@@ -0,0 +1,469 @@
+use {
+    self::{paged::*, physical::*, slice::*},
+    super::*,
+    crate::object::*,
+    alloc::{
+        sync::{Arc, Weak},
+        vec::Vec,
+    },
+    bitflags::bitflags,
+    core::ops::Deref,
+    kernel_hal::{CachePolicy, MMUFlags},
+    spin::Mutex,
+};
+
+mod paged;
+mod physical;
+mod slice;
+
+/// Virtual Memory Object Trait
+#[allow(clippy::len_without_is_empty)]
+pub trait VMObjectTrait: Sync + Send {
+    /// Read memory to `buf` from VMO at `offset`.
+    fn read(&self, offset: usize, buf: &mut [u8]) -> ZxResult;
+
+    /// Write memory from `buf` to VMO at `offset`.
+    fn write(&self, offset: usize, buf: &[u8]) -> ZxResult;
+
+    /// Resets the range of bytes in the VMO from `offset` to `offset+len` to 0.
+    fn zero(&self, offset: usize, len: usize) -> ZxResult;
+
+    /// Get the length of VMO.
+    fn len(&self) -> usize;
+
+    /// Set the length of VMO.
+    fn set_len(&self, len: usize) -> ZxResult;
+
+    /// Commit a page.
+    fn commit_page(&self, page_idx: usize, flags: MMUFlags) -> ZxResult<PhysAddr>;
+
+    /// Commit pages with an external function f.
+    /// the vmo is internally locked before it calls f,
+    /// allowing `VmMapping` to avoid deadlock
+    fn commit_pages_with(
+        &self,
+        f: &mut dyn FnMut(&mut dyn FnMut(usize, MMUFlags) -> ZxResult<PhysAddr>) -> ZxResult,
+    ) -> ZxResult;
+
+    /// Commit allocating physical memory.
+    fn commit(&self, offset: usize, len: usize) -> ZxResult;
+
+    /// Decommit allocated physical memory.
+    fn decommit(&self, offset: usize, len: usize) -> ZxResult;
+
+    /// Create a child VMO.
+    fn create_child(&self, offset: usize, len: usize) -> ZxResult<Arc<dyn VMObjectTrait>>;
+
+    /// Append a mapping to the VMO's mapping list.
+    fn append_mapping(&self, _mapping: Weak<VmMapping>) {}
+
+    /// Remove a mapping from the VMO's mapping list.
+    fn remove_mapping(&self, _mapping: Weak<VmMapping>) {}
+
+    /// Complete the VmoInfo.
+    fn complete_info(&self, info: &mut VmoInfo);
+
+    /// Get the cache policy.
+    fn cache_policy(&self) -> CachePolicy;
+
+    /// Set the cache policy.
+    fn set_cache_policy(&self, policy: CachePolicy) -> ZxResult;
+
+    /// Count committed pages of the VMO.
+    fn committed_pages_in_range(&self, start_idx: usize, end_idx: usize) -> usize;
+
+    /// Pin the given range of the VMO.
+    fn pin(&self, _offset: usize, _len: usize) -> ZxResult {
+        Err(ZxError::NOT_SUPPORTED)
+    }
+
+    /// Unpin the given range of the VMO.
+    fn unpin(&self, _offset: usize, _len: usize) -> ZxResult {
+        Err(ZxError::NOT_SUPPORTED)
+    }
+
+    /// Returns true if the object is backed by a contiguous range of physical memory.
+    fn is_contiguous(&self) -> bool {
+        false
+    }
+
+    /// Returns true if the object is backed by RAM.
+    fn is_paged(&self) -> bool {
+        false
+    }
+}
+
+/// Virtual memory containers
+///
+/// ## SYNOPSIS
+///
+/// A Virtual Memory Object (VMO) represents a contiguous region of virtual memory
+/// that may be mapped into multiple address spaces.
+pub struct VmObject {
+    base: KObjectBase,
+    resizable: bool,
+    trait_: Arc<dyn VMObjectTrait>,
+    inner: Mutex<VmObjectInner>,
+}
+
+impl_kobject!(VmObject);
+
+#[derive(Default)]
+struct VmObjectInner {
+    parent: Weak<VmObject>,
+    children: Vec<Weak<VmObject>>,
+    mapping_count: usize,
+    content_size: usize,
+}
+
+impl VmObject {
+    /// Create a new VMO backing on physical memory allocated in pages.
+    pub fn new_paged(pages: usize) -> Arc<Self> {
+        Self::new_paged_with_resizable(false, pages)
+    }
+
+    /// Create a new VMO, which can be resizable, backing on physical memory allocated in pages.
+    pub fn new_paged_with_resizable(resizable: bool, pages: usize) -> Arc<Self> {
+        let base = KObjectBase::new();
+        Arc::new(VmObject {
+            resizable,
+            trait_: VMObjectPaged::new(pages),
+            inner: Mutex::new(VmObjectInner::default()),
+            base,
+        })
+    }
+
+    /// Create a new VMO representing a piece of contiguous physical memory.
+    pub fn new_physical(paddr: PhysAddr, pages: usize) -> Arc<Self> {
+        Arc::new(VmObject {
+            base: KObjectBase::new(),
+            resizable: false,
+            trait_: VMObjectPhysical::new(paddr, pages),
+            inner: Mutex::new(VmObjectInner::default()),
+        })
+    }
+
+    /// Create a VM object referring to a specific contiguous range of physical frame.  
+    pub fn new_contiguous(pages: usize, align_log2: usize) -> ZxResult<Arc<Self>> {
+        let vmo = Arc::new(VmObject {
+            base: KObjectBase::new(),
+            resizable: false,
+            trait_: VMObjectPaged::new_contiguous(pages, align_log2)?,
+            inner: Mutex::new(VmObjectInner::default()),
+        });
+        Ok(vmo)
+    }
+
+    /// Create a child VMO.
+    pub fn create_child(
+        self: &Arc<Self>,
+        resizable: bool,
+        offset: usize,
+        len: usize,
+    ) -> ZxResult<Arc<Self>> {
+        let base = KObjectBase::with_name(&self.base.name());
+        let trait_ = self.trait_.create_child(offset, len)?;
+        let child = Arc::new(VmObject {
+            base,
+            resizable,
+            trait_,
+            inner: Mutex::new(VmObjectInner {
+                parent: Arc::downgrade(self),
+                ..VmObjectInner::default()
+            }),
+        });
+        self.add_child(&child);
+        Ok(child)
+    }
+
+    /// Create a child slice as an VMO
+    pub fn create_slice(self: &Arc<Self>, offset: usize, p_size: usize) -> ZxResult<Arc<Self>> {
+        let size = roundup_pages(p_size);
+        // why 32 * PAGE_SIZE? Refered to zircon source codes
+        if size < p_size || size > usize::MAX & !(32 * PAGE_SIZE) {
+            return Err(ZxError::OUT_OF_RANGE);
+        }
+        // child slice must be wholly contained
+        let parent_size = self.trait_.len();
+        if !page_aligned(offset) {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        if offset > parent_size || size > parent_size - offset {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        if self.resizable {
+            return Err(ZxError::NOT_SUPPORTED);
+        }
+        if self.trait_.cache_policy() != CachePolicy::Cached && !self.trait_.is_contiguous() {
+            return Err(ZxError::BAD_STATE);
+        }
+        let child = Arc::new(VmObject {
+            base: KObjectBase::with_name(&self.base.name()),
+            resizable: false,
+            trait_: VMObjectSlice::new(self.trait_.clone(), offset, size),
+            inner: Mutex::new(VmObjectInner {
+                parent: Arc::downgrade(self),
+                ..VmObjectInner::default()
+            }),
+        });
+        self.add_child(&child);
+        Ok(child)
+    }
+
+    /// Add child to the list and signal if ZeroChildren signal is active.
+    /// If the number of children turns 0 to 1, signal it
+    fn add_child(&self, child: &Arc<VmObject>) {
+        let mut inner = self.inner.lock();
+        inner.children.retain(|x| x.strong_count() != 0);
+        inner.children.push(Arc::downgrade(child));
+        // if inner.children.len() == 1 {
+        //     self.base.signal_clear(Signal::VMO_ZERO_CHILDREN);
+        // }
+    }
+
+    /// Set the length of this VMO if resizable.
+    pub fn set_len(&self, len: usize) -> ZxResult {
+        let size = roundup_pages(len);
+        if size < len {
+            return Err(ZxError::OUT_OF_RANGE);
+        }
+        if !self.resizable {
+            return Err(ZxError::UNAVAILABLE);
+        }
+        self.trait_.set_len(size)
+    }
+
+    /// Set the size of the content stored in the VMO in bytes, resize vmo if needed
+    pub fn set_content_size_and_resize(
+        &self,
+        size: usize,
+        zero_until_offset: usize,
+    ) -> ZxResult<usize> {
+        let mut inner = self.inner.lock();
+        let content_size = inner.content_size;
+        let len = self.trait_.len();
+        if size < content_size {
+            return Ok(content_size);
+        }
+        let required_len = roundup_pages(size);
+        let new_content_size = if required_len > len && self.set_len(required_len).is_err() {
+            len
+        } else {
+            size
+        };
+        let zero_until_offset = zero_until_offset.min(new_content_size);
+        if zero_until_offset > content_size {
+            self.trait_
+                .zero(content_size, zero_until_offset - content_size)?;
+        }
+        inner.content_size = new_content_size;
+        Ok(new_content_size)
+    }
+
+    /// Get the size of the content stored in the VMO in bytes.
+    pub fn content_size(&self) -> usize {
+        let inner = self.inner.lock();
+        inner.content_size
+    }
+
+    /// Get the size of the content stored in the VMO in bytes.
+    pub fn set_content_size(&self, size: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        inner.content_size = size;
+        Ok(())
+    }
+
+    /// Get information of this VMO.
+    pub fn get_info(&self) -> VmoInfo {
+        let inner = self.inner.lock();
+        let mut ret = VmoInfo {
+            koid: self.base.id,
+            name: {
+                let mut arr = [0u8; 32];
+                let name = self.base.name();
+                let length = name.len().min(32);
+                arr[..length].copy_from_slice(&name.as_bytes()[..length]);
+                arr
+            },
+            size: self.trait_.len() as u64,
+            parent_koid: inner.parent.upgrade().map(|p| p.id()).unwrap_or(0),
+            num_children: inner.children.len() as u64,
+            flags: if self.resizable {
+                VmoInfoFlags::RESIZABLE
+            } else {
+                VmoInfoFlags::empty()
+            },
+            cache_policy: self.trait_.cache_policy() as u32,
+            share_count: inner.mapping_count as u64,
+            ..Default::default()
+        };
+        self.trait_.complete_info(&mut ret);
+        ret
+    }
+
+    /// Set the cache policy.
+    pub fn set_cache_policy(&self, policy: CachePolicy) -> ZxResult {
+        let inner = self.inner.lock();
+        if !inner.children.is_empty() {
+            return Err(ZxError::BAD_STATE);
+        }
+        if inner.mapping_count != 0 {
+            return Err(ZxError::BAD_STATE);
+        }
+        self.trait_.set_cache_policy(policy)
+    }
+
+    /// Append a mapping to the VMO's mapping list.
+    pub fn append_mapping(&self, mapping: Weak<VmMapping>) {
+        self.inner.lock().mapping_count += 1;
+        self.trait_.append_mapping(mapping);
+    }
+
+    /// Remove a mapping from the VMO's mapping list.
+    pub fn remove_mapping(&self, mapping: Weak<VmMapping>) {
+        self.inner.lock().mapping_count -= 1;
+        self.trait_.remove_mapping(mapping);
+    }
+
+    /// Returns an estimate of the number of unique VmAspaces that this object
+    /// is mapped into.
+    pub fn share_count(&self) -> usize {
+        let inner = self.inner.lock();
+        inner.mapping_count
+    }
+
+    /// Returns true if the object size can be changed.
+    pub fn is_resizable(&self) -> bool {
+        self.resizable
+    }
+
+    /// Returns true if the object is backed by a contiguous range of physical memory.
+    pub fn is_contiguous(&self) -> bool {
+        self.trait_.is_contiguous()
+    }
+}
+
+impl Deref for VmObject {
+    type Target = Arc<dyn VMObjectTrait>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.trait_
+    }
+}
+
+impl Drop for VmObject {
+    fn drop(&mut self) {
+        let mut inner = self.inner.lock();
+        let parent = match inner.parent.upgrade() {
+            Some(parent) => parent,
+            None => return,
+        };
+        for child in inner.children.iter() {
+            if let Some(child) = child.upgrade() {
+                child.inner.lock().parent = Arc::downgrade(&parent);
+            }
+        }
+        let mut parent_inner = parent.inner.lock();
+        let children = &mut parent_inner.children;
+        children.append(&mut inner.children);
+        children.retain(|c| c.strong_count() != 0);
+        for child in children.iter() {
+            let child = child.upgrade().unwrap();
+            let mut inner = child.inner.lock();
+            inner.children.retain(|c| c.strong_count() != 0);
+        }
+    }
+}
+
+/// Describes a VMO.
+#[repr(C)]
+#[derive(Default)]
+pub struct VmoInfo {
+    /// The koid of this VMO.
+    koid: KoID,
+    /// The name of this VMO.
+    name: [u8; 32],
+    /// The size of this VMO; i.e., the amount of virtual address space it
+    /// would consume if mapped.
+    size: u64,
+    /// If this VMO is a clone, the koid of its parent. Otherwise, zero.
+    parent_koid: KoID,
+    /// The number of clones of this VMO, if any.
+    num_children: u64,
+    /// The number of times this VMO is currently mapped into VMARs.
+    num_mappings: u64,
+    /// The number of unique address space we're mapped into.
+    share_count: u64,
+    /// Flags.
+    pub flags: VmoInfoFlags,
+    /// Padding.
+    padding1: [u8; 4],
+    /// If the type is `PAGED`, the amount of
+    /// memory currently allocated to this VMO; i.e., the amount of physical
+    /// memory it consumes. Undefined otherwise.
+    committed_bytes: u64,
+    /// If `flags & ZX_INFO_VMO_VIA_HANDLE`, the handle rights.
+    /// Undefined otherwise.
+    pub rights: Rights,
+    /// VMO mapping cache policy.
+    cache_policy: u32,
+}
+
+bitflags! {
+    #[derive(Default)]
+    /// Values used by ZX_INFO_PROCESS_VMOS.
+    pub struct VmoInfoFlags: u32 {
+        /// The VMO points to a physical address range, and does not consume memory.
+        /// Typically used to access memory-mapped hardware.
+        /// Mutually exclusive with TYPE_PAGED.
+        const TYPE_PHYSICAL = 0;
+
+        #[allow(clippy::identity_op)]
+        /// The VMO is backed by RAM, consuming memory.
+        /// Mutually exclusive with TYPE_PHYSICAL.
+        const TYPE_PAGED    = 1 << 0;
+
+        /// The VMO is resizable.
+        const RESIZABLE     = 1 << 1;
+
+        /// The VMO is a child, and is a copy-on-write clone.
+        const IS_COW_CLONE  = 1 << 2;
+
+        /// When reading a list of VMOs pointed to by a process, indicates that the
+        /// process has a handle to the VMO, which isn't necessarily mapped.
+        const VIA_HANDLE    = 1 << 3;
+
+        /// When reading a list of VMOs pointed to by a process, indicates that the
+        /// process maps the VMO into a VMAR, but doesn't necessarily have a handle to
+        /// the VMO.
+        const VIA_MAPPING   = 1 << 4;
+
+        /// The VMO is a pager owned VMO created by zx_pager_create_vmo or is
+        /// a clone of a VMO with this flag set. Will only be set on VMOs with
+        /// the ZX_INFO_VMO_TYPE_PAGED flag set.
+        const PAGER_BACKED  = 1 << 5;
+
+        /// The VMO is contiguous.
+        const CONTIGUOUS    = 1 << 6;
+    }
+}
+
+/// Different operations that `range_change` can perform against any VmMappings that are found.
+#[allow(dead_code)]
+#[derive(PartialEq, Eq, Clone, Copy)]
+pub(super) enum RangeChangeOp {
+    Unmap,
+    RemoveWrite,
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    pub fn read_write(vmo: &VmObject) {
+        let mut buf = [0u8; 4];
+        vmo.write(0, &[0, 1, 2, 3]).unwrap();
+        vmo.read(0, &mut buf).unwrap();
+        assert_eq!(&buf, &[0, 1, 2, 3]);
+    }
+}

code/ch03-04/object/src/vm/vmo/paged.rs

@@ -0,0 +1,337 @@
+use {
+    super::*,
+    crate::util::BlockIter,
+    alloc::sync::Arc,
+    alloc::vec::Vec,
+    core::ops::Range,
+    kernel_hal::{MMUFlags, PhysFrame, PAGE_SIZE},
+    spin::Mutex,
+};
+
+/// The main VM object type, holding a list of pages.
+pub struct VMObjectPaged {
+    inner: Mutex<VMObjectPagedInner>,
+}
+
+/// The mutable part of `VMObjectPaged`.
+#[derive(Default)]
+struct VMObjectPagedInner {
+    /// Physical frames of this VMO.
+    frames: Vec<PhysFrame>,
+    /// Cache Policy
+    cache_policy: CachePolicy,
+    /// Is contiguous
+    contiguous: bool,
+    /// Sum of pin_count
+    pin_count: usize,
+    /// All mappings to this VMO.
+    mappings: Vec<Weak<VmMapping>>,
+}
+
+impl VMObjectPaged {
+    /// Create a new VMO backing on physical memory allocated in pages.
+    pub fn new(pages: usize) -> Arc<Self> {
+        let mut frames = Vec::new();
+        frames.resize_with(pages, || PhysFrame::alloc_zeroed().unwrap());
+        Arc::new(VMObjectPaged {
+            inner: Mutex::new(VMObjectPagedInner {
+                frames,
+                ..Default::default()
+            }),
+        })
+    }
+
+    /// Create a list of contiguous pages
+    pub fn new_contiguous(pages: usize, align_log2: usize) -> ZxResult<Arc<Self>> {
+        let frames = PhysFrame::alloc_contiguous_zeroed(pages, align_log2 - PAGE_SIZE_LOG2);
+        if frames.is_empty() {
+            return Err(ZxError::NO_MEMORY);
+        }
+        Ok(Arc::new(VMObjectPaged {
+            inner: Mutex::new(VMObjectPagedInner {
+                frames,
+                contiguous: true,
+                ..Default::default()
+            }),
+        }))
+    }
+}
+
+impl VMObjectTrait for VMObjectPaged {
+    fn read(&self, offset: usize, buf: &mut [u8]) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if inner.cache_policy != CachePolicy::Cached {
+            return Err(ZxError::BAD_STATE);
+        }
+        inner.for_each_page(offset, buf.len(), |paddr, buf_range| {
+            kernel_hal::pmem_read(paddr, &mut buf[buf_range]);
+        });
+        Ok(())
+    }
+
+    fn write(&self, offset: usize, buf: &[u8]) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if inner.cache_policy != CachePolicy::Cached {
+            return Err(ZxError::BAD_STATE);
+        }
+        inner.for_each_page(offset, buf.len(), |paddr, buf_range| {
+            kernel_hal::pmem_write(paddr, &buf[buf_range]);
+        });
+        Ok(())
+    }
+
+    fn zero(&self, offset: usize, len: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if inner.cache_policy != CachePolicy::Cached {
+            return Err(ZxError::BAD_STATE);
+        }
+        inner.for_each_page(offset, len, |paddr, buf_range| {
+            kernel_hal::pmem_zero(paddr, buf_range.len());
+        });
+        Ok(())
+    }
+
+    fn len(&self) -> usize {
+        let inner = self.inner.lock();
+        inner.frames.len() * PAGE_SIZE
+    }
+
+    fn set_len(&self, len: usize) -> ZxResult {
+        assert!(page_aligned(len));
+        let mut inner = self.inner.lock();
+        inner.frames.resize_with(len / PAGE_SIZE, || {
+            PhysFrame::alloc().ok_or(ZxError::NO_MEMORY).unwrap()
+        });
+        Ok(())
+    }
+
+    fn commit_page(&self, page_idx: usize, _flags: MMUFlags) -> ZxResult<PhysAddr> {
+        let inner = self.inner.lock();
+        Ok(inner.frames[page_idx].addr())
+    }
+
+    fn commit_pages_with(
+        &self,
+        f: &mut dyn FnMut(&mut dyn FnMut(usize, MMUFlags) -> ZxResult<PhysAddr>) -> ZxResult,
+    ) -> ZxResult {
+        let inner = self.inner.lock();
+        f(&mut |page_idx, _| Ok(inner.frames[page_idx].addr()))
+    }
+
+    fn commit(&self, _offset: usize, _len: usize) -> ZxResult {
+        Ok(())
+    }
+
+    fn decommit(&self, _offset: usize, _len: usize) -> ZxResult {
+        Ok(())
+    }
+
+    fn create_child(&self, offset: usize, len: usize) -> ZxResult<Arc<dyn VMObjectTrait>> {
+        assert!(page_aligned(offset));
+        assert!(page_aligned(len));
+        let mut inner = self.inner.lock();
+        let child = inner.create_child(offset, len)?;
+        Ok(child)
+    }
+
+    fn append_mapping(&self, mapping: Weak<VmMapping>) {
+        let mut inner = self.inner.lock();
+        inner.mappings.push(mapping);
+    }
+
+    fn remove_mapping(&self, mapping: Weak<VmMapping>) {
+        let mut inner = self.inner.lock();
+        inner
+            .mappings
+            .drain_filter(|x| x.strong_count() == 0 || Weak::ptr_eq(x, &mapping));
+    }
+
+    fn complete_info(&self, info: &mut VmoInfo) {
+        let inner = self.inner.lock();
+        info.flags |= VmoInfoFlags::TYPE_PAGED;
+        inner.complete_info(info);
+    }
+
+    fn cache_policy(&self) -> CachePolicy {
+        let inner = self.inner.lock();
+        inner.cache_policy
+    }
+
+    fn set_cache_policy(&self, policy: CachePolicy) -> ZxResult {
+        // conditions for allowing the cache policy to be set:
+        // 1) vmo either has no pages committed currently or is transitioning from being cached
+        // 2) vmo has no pinned pages
+        // 3) vmo has no mappings
+        // 4) vmo has no children (TODO)
+        // 5) vmo is not a child
+        let mut inner = self.inner.lock();
+        if !inner.frames.is_empty() && inner.cache_policy != CachePolicy::Cached {
+            return Err(ZxError::BAD_STATE);
+        }
+        if inner.pin_count != 0 {
+            return Err(ZxError::BAD_STATE);
+        }
+        if inner.cache_policy == CachePolicy::Cached && policy != CachePolicy::Cached {
+            for frame in inner.frames.iter() {
+                kernel_hal::frame_flush(frame.addr());
+            }
+        }
+        inner.cache_policy = policy;
+        Ok(())
+    }
+
+    fn committed_pages_in_range(&self, start_idx: usize, end_idx: usize) -> usize {
+        end_idx - start_idx
+    }
+
+    fn pin(&self, offset: usize, len: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if offset + len > inner.frames.len() * PAGE_SIZE {
+            return Err(ZxError::OUT_OF_RANGE);
+        }
+        if len == 0 {
+            return Ok(());
+        }
+        inner.pin_count += pages(len);
+        Ok(())
+    }
+
+    fn unpin(&self, offset: usize, len: usize) -> ZxResult {
+        let mut inner = self.inner.lock();
+        if offset + len > inner.frames.len() * PAGE_SIZE {
+            return Err(ZxError::OUT_OF_RANGE);
+        }
+        if len == 0 {
+            return Ok(());
+        }
+        inner.pin_count -= pages(len);
+        Ok(())
+    }
+
+    fn is_contiguous(&self) -> bool {
+        let inner = self.inner.lock();
+        inner.contiguous
+    }
+
+    fn is_paged(&self) -> bool {
+        true
+    }
+}
+
+impl VMObjectPagedInner {
+    /// Helper function to split range into sub-ranges within pages.
+    ///
+    /// ```text
+    /// VMO range:
+    /// |----|----|----|----|----|
+    ///
+    /// buf:
+    ///            [====len====]
+    /// |--offset--|
+    ///
+    /// sub-ranges:
+    ///            [===]
+    ///                [====]
+    ///                     [==]
+    /// ```
+    ///
+    /// `f` is a function to process in-page ranges.
+    /// It takes 2 arguments:
+    /// * `paddr`: the start physical address of the in-page range.
+    /// * `buf_range`: the range in view of the input buffer.
+    fn for_each_page(
+        &mut self,
+        offset: usize,
+        buf_len: usize,
+        mut f: impl FnMut(PhysAddr, Range<usize>),
+    ) {
+        let iter = BlockIter {
+            begin: offset,
+            end: offset + buf_len,
+            block_size_log2: 12,
+        };
+        for block in iter {
+            let paddr = self.frames[block.block].addr();
+            let buf_range = block.origin_begin() - offset..block.origin_end() - offset;
+            f(paddr + block.begin, buf_range);
+        }
+    }
+
+    /// Create a snapshot child VMO.
+    fn create_child(&mut self, offset: usize, len: usize) -> ZxResult<Arc<VMObjectPaged>> {
+        // clone contiguous vmo is no longer permitted
+        // https://fuchsia.googlesource.com/fuchsia/+/e6b4c6751bbdc9ed2795e81b8211ea294f139a45
+        if self.contiguous {
+            return Err(ZxError::INVALID_ARGS);
+        }
+        if self.cache_policy != CachePolicy::Cached || self.pin_count != 0 {
+            return Err(ZxError::BAD_STATE);
+        }
+        let mut frames = Vec::with_capacity(pages(len));
+        for _ in 0..pages(len) {
+            frames.push(PhysFrame::alloc().ok_or(ZxError::NO_MEMORY)?);
+        }
+        for (i, frame) in frames.iter().enumerate() {
+            if let Some(src_frame) = self.frames.get(pages(offset) + i) {
+                kernel_hal::frame_copy(src_frame.addr(), frame.addr())
+            } else {
+                kernel_hal::pmem_zero(frame.addr(), PAGE_SIZE);
+            }
+        }
+        // create child VMO
+        let child = Arc::new(VMObjectPaged {
+            inner: Mutex::new(VMObjectPagedInner {
+                frames,
+                ..Default::default()
+            }),
+        });
+        Ok(child)
+    }
+
+    fn complete_info(&self, info: &mut VmoInfo) {
+        if self.contiguous {
+            info.flags |= VmoInfoFlags::CONTIGUOUS;
+        }
+        // info.num_children = if self.type_.is_hidden() { 2 } else { 0 };
+        info.committed_bytes = (self.frames.len() * PAGE_SIZE) as u64;
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn read_write() {
+        let vmo = VmObject::new_paged(2);
+        super::super::tests::read_write(&*vmo);
+    }
+
+    #[test]
+    fn create_child() {
+        let vmo = VmObject::new_paged(1);
+        let child_vmo = vmo.create_child(false, 0, PAGE_SIZE).unwrap();
+
+        // write to parent and make sure clone doesn't see it
+        vmo.test_write(0, 1);
+        assert_eq!(vmo.test_read(0), 1);
+        assert_eq!(child_vmo.test_read(0), 0);
+
+        // write to clone and make sure parent doesn't see it
+        child_vmo.test_write(0, 2);
+        assert_eq!(vmo.test_read(0), 1);
+        assert_eq!(child_vmo.test_read(0), 2);
+    }
+
+    impl VmObject {
+        pub fn test_write(&self, page: usize, value: u8) {
+            self.write(page * PAGE_SIZE, &[value]).unwrap();
+        }
+
+        pub fn test_read(&self, page: usize) -> u8 {
+            let mut buf = [0; 1];
+            self.read(page * PAGE_SIZE, &mut buf).unwrap();
+            buf[0]
+        }
+    }
+}

code/ch03-04/object/src/vm/vmo/physical.rs

@@ -0,0 +1,129 @@
+use {super::*, alloc::sync::Arc, kernel_hal::MMUFlags, spin::Mutex};
+
+/// VMO representing a physical range of memory.
+pub struct VMObjectPhysical {
+    paddr: PhysAddr,
+    pages: usize,
+    /// Lock this when access physical memory.
+    data_lock: Mutex<()>,
+    inner: Mutex<VMObjectPhysicalInner>,
+}
+
+struct VMObjectPhysicalInner {
+    cache_policy: CachePolicy,
+}
+
+impl VMObjectPhysicalInner {
+    pub fn new() -> VMObjectPhysicalInner {
+        VMObjectPhysicalInner {
+            cache_policy: CachePolicy::Uncached,
+        }
+    }
+}
+
+impl VMObjectPhysical {
+    /// Create a new VMO representing a piece of contiguous physical memory.
+    /// You must ensure nobody has the ownership of this piece of memory yet.
+    pub fn new(paddr: PhysAddr, pages: usize) -> Arc<Self> {
+        assert!(page_aligned(paddr));
+        Arc::new(VMObjectPhysical {
+            paddr,
+            pages,
+            data_lock: Mutex::default(),
+            inner: Mutex::new(VMObjectPhysicalInner::new()),
+        })
+    }
+}
+
+impl VMObjectTrait for VMObjectPhysical {
+    fn read(&self, offset: usize, buf: &mut [u8]) -> ZxResult {
+        let _ = self.data_lock.lock();
+        assert!(offset + buf.len() <= self.len());
+        kernel_hal::pmem_read(self.paddr + offset, buf);
+        Ok(())
+    }
+
+    fn write(&self, offset: usize, buf: &[u8]) -> ZxResult {
+        let _ = self.data_lock.lock();
+        assert!(offset + buf.len() <= self.len());
+        kernel_hal::pmem_write(self.paddr + offset, buf);
+        Ok(())
+    }
+
+    fn zero(&self, offset: usize, len: usize) -> ZxResult {
+        let _ = self.data_lock.lock();
+        assert!(offset + len <= self.len());
+        kernel_hal::pmem_zero(self.paddr + offset, len);
+        Ok(())
+    }
+
+    fn len(&self) -> usize {
+        self.pages * PAGE_SIZE
+    }
+
+    fn set_len(&self, _len: usize) -> ZxResult {
+        unimplemented!()
+    }
+
+    fn commit_page(&self, page_idx: usize, _flags: MMUFlags) -> ZxResult<PhysAddr> {
+        Ok(self.paddr + page_idx * PAGE_SIZE)
+    }
+
+    fn commit_pages_with(
+        &self,
+        f: &mut dyn FnMut(&mut dyn FnMut(usize, MMUFlags) -> ZxResult<PhysAddr>) -> ZxResult,
+    ) -> ZxResult {
+        f(&mut |page_idx, _flags| Ok(self.paddr + page_idx * PAGE_SIZE))
+    }
+
+    fn commit(&self, _offset: usize, _len: usize) -> ZxResult {
+        // do nothing
+        Ok(())
+    }
+
+    fn decommit(&self, _offset: usize, _len: usize) -> ZxResult {
+        // do nothing
+        Ok(())
+    }
+
+    fn create_child(&self, _offset: usize, _len: usize) -> ZxResult<Arc<dyn VMObjectTrait>> {
+        Err(ZxError::NOT_SUPPORTED)
+    }
+
+    fn complete_info(&self, _info: &mut VmoInfo) {
+        warn!("VmoInfo for physical is unimplemented");
+    }
+
+    fn cache_policy(&self) -> CachePolicy {
+        let inner = self.inner.lock();
+        inner.cache_policy
+    }
+
+    fn set_cache_policy(&self, policy: CachePolicy) -> ZxResult {
+        let mut inner = self.inner.lock();
+        inner.cache_policy = policy;
+        Ok(())
+    }
+
+    fn committed_pages_in_range(&self, _start_idx: usize, _end_idx: usize) -> usize {
+        0
+    }
+
+    fn is_contiguous(&self) -> bool {
+        true
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    #![allow(unsafe_code)]
+    use super::*;
+    use kernel_hal::CachePolicy;
+
+    #[test]
+    fn read_write() {
+        let vmo = VmObject::new_physical(0x1000, 2);
+        assert_eq!(vmo.cache_policy(), CachePolicy::Uncached);
+        super::super::tests::read_write(&vmo);
+    }
+}

code/ch03-04/object/src/vm/vmo/slice.rs

@@ -0,0 +1,112 @@
+use {super::*, kernel_hal::MMUFlags};
+
+pub struct VMObjectSlice {
+    /// Parent node.
+    parent: Arc<dyn VMObjectTrait>,
+    /// The offset from parent.
+    offset: usize,
+    /// The size in bytes.
+    size: usize,
+}
+
+impl VMObjectSlice {
+    pub fn new(parent: Arc<dyn VMObjectTrait>, offset: usize, size: usize) -> Arc<Self> {
+        Arc::new(VMObjectSlice {
+            parent,
+            offset,
+            size,
+        })
+    }
+
+    fn check_range(&self, offset: usize, len: usize) -> ZxResult {
+        if offset + len >= self.size {
+            return Err(ZxError::OUT_OF_RANGE);
+        }
+        Ok(())
+    }
+}
+
+impl VMObjectTrait for VMObjectSlice {
+    fn read(&self, offset: usize, buf: &mut [u8]) -> ZxResult {
+        self.check_range(offset, buf.len())?;
+        self.parent.read(offset + self.offset, buf)
+    }
+
+    fn write(&self, offset: usize, buf: &[u8]) -> ZxResult {
+        self.check_range(offset, buf.len())?;
+        self.parent.write(offset + self.offset, buf)
+    }
+
+    fn zero(&self, offset: usize, len: usize) -> ZxResult {
+        self.check_range(offset, len)?;
+        self.parent.zero(offset + self.offset, len)
+    }
+
+    fn len(&self) -> usize {
+        self.size
+    }
+
+    fn set_len(&self, _len: usize) -> ZxResult {
+        unimplemented!()
+    }
+
+    fn commit_page(&self, page_idx: usize, flags: MMUFlags) -> ZxResult<usize> {
+        self.parent
+            .commit_page(page_idx + self.offset / PAGE_SIZE, flags)
+    }
+
+    fn commit_pages_with(
+        &self,
+        f: &mut dyn FnMut(&mut dyn FnMut(usize, MMUFlags) -> ZxResult<PhysAddr>) -> ZxResult,
+    ) -> ZxResult {
+        self.parent.commit_pages_with(f)
+    }
+
+    fn commit(&self, offset: usize, len: usize) -> ZxResult {
+        self.parent.commit(offset + self.offset, len)
+    }
+
+    fn decommit(&self, offset: usize, len: usize) -> ZxResult {
+        self.parent.decommit(offset + self.offset, len)
+    }
+
+    fn create_child(&self, _offset: usize, _len: usize) -> ZxResult<Arc<dyn VMObjectTrait>> {
+        Err(ZxError::NOT_SUPPORTED)
+    }
+
+    fn complete_info(&self, info: &mut VmoInfo) {
+        self.parent.complete_info(info);
+    }
+
+    fn cache_policy(&self) -> CachePolicy {
+        self.parent.cache_policy()
+    }
+
+    fn set_cache_policy(&self, _policy: CachePolicy) -> ZxResult {
+        Ok(())
+    }
+
+    fn committed_pages_in_range(&self, start_idx: usize, end_idx: usize) -> usize {
+        let po = pages(self.offset);
+        self.parent
+            .committed_pages_in_range(start_idx + po, end_idx + po)
+    }
+
+    fn pin(&self, offset: usize, len: usize) -> ZxResult {
+        self.check_range(offset, len)?;
+        self.parent.pin(offset + self.offset, len)
+    }
+
+    fn unpin(&self, offset: usize, len: usize) -> ZxResult {
+        self.check_range(offset, len)?;
+        self.parent.unpin(offset + self.offset, len)
+    }
+
+    fn is_contiguous(&self) -> bool {
+        self.parent.is_contiguous()
+    }
+
+    fn is_paged(&self) -> bool {
+        self.parent.is_paged()
+    }
+}