XNU内核源码分析（一）：XNU内核初探

What is XNU?
===========
 
XNU kernel is part of the Darwin operating system for use in macOS and iOS operating systems. XNU is an acronym for X is Not Unix.
XNU is a hybrid kernel combining the Mach kernel developed at Carnegie Mellon University with components from FreeBSD and a C++ API for writing drivers called IOKit.
XNU runs on x86_64 for both single processor and multi-processor configurations.
 
XNU Source Tree
===============
 
  * `config` - configurations for exported apis for supported architecture and platform
  * `SETUP` - Basic set of tools used for configuring the kernel, versioning and kextsymbol management.
  * `EXTERNAL_HEADERS` - Headers sourced from other projects to avoid dependency cycles when building. These headers should be regularly synced when source is updated.
  * `libkern` - C++ IOKit library code for handling of drivers and kexts.
  * `libsa` -  kernel bootstrap code for startup
  * `libsyscall` - syscall library interface for userspace programs
  * `libkdd` - source for user library for parsing kernel data like kernel chunked data.
  * `makedefs` - top level rules and defines for kernel build.
  * `osfmk` - Mach kernel based subsystems
  * `pexpert` - Platform specific code like interrupt handling, atomics etc.
  * `security` - Mandatory Access Check policy interfaces and related implementation.
  * `bsd` - BSD subsystems code
  * `tools` - A set of utilities for testing, debugging and profiling kernel.
 
How to build XNU
================
 
Building `DEVELOPMENT` kernel
-----------------------------
 
The xnu make system can build kernel based on `KERNEL_CONFIGS` & `ARCH_CONFIGS` variables as arguments.
Here is the syntax:
 
    make SDKROOT=<sdkroot> ARCH_CONFIGS=<arch> KERNEL_CONFIGS=<variant>
 
Where:
 
  * \<sdkroot>: path to macOS SDK on disk. (defaults to `/`)
  * \<variant>: can be `debug`, `development`, `release`, `profile` and configures compilation flags and asserts throughout kernel code.
  * \<arch>   : can be valid arch to build for. (E.g. `X86_64`)
 
To build a kernel for the same architecture as running OS, just type
 
    $ make
    $ make SDKROOT=macosx.internal
 
Additionally, there is support for configuring architectures through `ARCH_CONFIGS` and kernel configurations with `KERNEL_CONFIGS`.
 
    $ make SDKROOT=macosx.internal ARCH_CONFIGS=X86_64 KERNEL_CONFIGS=DEVELOPMENT
    $ make SDKROOT=macosx.internal ARCH_CONFIGS=X86_64 KERNEL_CONFIGS="RELEASE DEVELOPMENT DEBUG"
 
 
Note:
  * By default, architecture is set to the build machine architecture, and the default kernel
    config is set to build for DEVELOPMENT.
 
 
This will also create a bootable image, kernel.[config],  and a kernel binary
with symbols, kernel.[config].unstripped.
 
To intall the kernel into a DSTROOT, use the `install_kernels` target:
 
    $ make install_kernels DSTROOT=/tmp/xnu-dst
 
Hint:
For a more satisfying kernel debugging experience, with access to all
local variables and arguments, but without all the extra check of the
DEBUG kernel, add something like:
	CFLAGS_DEVELOPMENTARM64="-O0 -g -DKERNEL_STACK_MULTIPLIER=2"
	CXXFLAGS_DEVELOPMENTARM64="-O0 -g -DKERNEL_STACK_MULTIPLIER=2"
to your make command.
Replace DEVELOPMENT and ARM64 with the appropriate build and platform.
 
 
  * To build with RELEASE kernel configuration
 
        make KERNEL_CONFIGS=RELEASE SDKROOT=/path/to/SDK
 
 
Building FAT kernel binary
--------------------------
 
Define architectures in your environment or when running a make command.
 
    $ make ARCH_CONFIGS="X86_64" exporthdrs all
 
Other makefile options
----------------------
 
 * $ make MAKEJOBS=-j8    # this will use 8 processes during the build. The default is 2x the number of active CPUS.
 * $ make -j8             # the standard command-line option is also accepted
 * $ make -w              # trace recursive make invocations. Useful in combination with VERBOSE=YES
 * $ make BUILD_LTO=0     # build without LLVM Link Time Optimization
 * $ make BOUND_CHECKS=1  # enable -fbound-attributes for this build
 * $ make REMOTEBUILD=user@remotehost # perform build on remote host
 * $ make BUILD_JSON_COMPILATION_DATABASE=1 # Build Clang JSON Compilation Database
 
The XNU build system can optionally output color-formatted build output. To enable this, you can either
set the `XNU_LOGCOLORS` environment variable to `y`, or you can pass `LOGCOLORS=y` to the make command.
 
 
Debug information formats
=========================
 
By default, a DWARF debug information repository is created during the install phase; this is a "bundle" named kernel.development.\<variant>.dSYM
To select the older STABS debug information format (where debug information is embedded in the kernel.development.unstripped image), set the BUILD_STABS environment variable.
 
    $ export BUILD_STABS=1
    $ make
 
 
Building KernelCaches
=====================
 
To test the xnu kernel, you need to build a kernelcache that links the kexts and
kernel together into a single bootable image.
To build a kernelcache you can use the following mechanisms:
 
  * Using automatic kernelcache generation with `kextd`.
    The kextd daemon keeps watching for changing in `/System/Library/Extensions` directory.
    So you can setup new kernel as
 
        $ cp BUILD/obj/DEVELOPMENT/X86_64/kernel.development /System/Library/Kernels/
        $ touch /System/Library/Extensions
        $ ps -e | grep kextd
 
  * Manually invoking `kextcache` to build new kernelcache.
 
        $ kextcache -q -z -a x86_64 -l -n -c /var/tmp/kernelcache.test -K /var/tmp/kernel.test /System/Library/Extensions
 
 
 
Running KernelCache on Target machine
=====================================
 
The development kernel and iBoot supports configuring boot arguments so that we can safely boot into test kernel and, if things go wrong, safely fall back to previously used kernelcache.
Following are the steps to get such a setup:
 
  1. Create kernel cache using the kextcache command as `/kernelcache.test`
  2. Copy exiting boot configurations to alternate file
 
         $ cp /Library/Preferences/SystemConfiguration/com.apple.Boot.plist /next_boot.plist
 
  3. Update the kernelcache and boot-args for your setup
 
         $ plutil -insert "Kernel Cache" -string "kernelcache.test" /next_boot.plist
         $ plutil -replace "Kernel Flags" -string "debug=0x144 -v kernelsuffix=test " /next_boot.plist
 
  4. Copy the new config to `/Library/Preferences/SystemConfiguration/`
 
         $ cp /next_boot.plist /Library/Preferences/SystemConfiguration/boot.plist
 
  5. Bless the volume with new configs.
 
         $ sudo -n bless  --mount / --setBoot --nextonly --options "config=boot"
 
     The `--nextonly` flag specifies that use the `boot.plist` configs only for one boot.
     So if the kernel panic's you can easily power reboot and recover back to original kernel.
Creating tags and cscope
========================
Set up your build environment and from the top directory, run:
    $ make tags     # this will build ctags and etags on a case-sensitive volume, only ctags on case-insensitive
    $ make TAGS     # this will build etags
    $ make cscope   # this will build cscope database
How to install a new header file from XNU
=========================================
To install IOKit headers, see additional comments in [iokit/IOKit/Makefile]().
XNU installs header files at the following locations -
    a. $(DSTROOT)/System/Library/Frameworks/Kernel.framework/Headers
    b. $(DSTROOT)/System/Library/Frameworks/Kernel.framework/PrivateHeaders
    c. $(DSTROOT)/usr/include/
    d. $(DSTROOT)/usr/local/include/
    e. $(DSTROOT)/System/DriverKit/usr/include/
    f. $(DSTROOT)/System/Library/Frameworks/System.framework/PrivateHeaders
`Kernel.framework` is used by kernel extensions.\
The `System.framework`, `/usr/include` and `/usr/local/include` are used by user level applications. \
`/System/DriverKit/usr/include` is used by userspace drivers. \
The header files in framework's `PrivateHeaders` are only available for ** Apple Internal Development **.
 
The directory containing the header file should have a Makefile that
creates the list of files that should be installed at different locations.
If you are adding the first header file in a directory, you will need to
create Makefile similar to `xnu/bsd/sys/Makefile`.
 
Add your header file to the correct file list depending on where you want
to install it. The default locations where the header files are installed
from each file list are -
 
    a. `DATAFILES` : To make header file available in user level -
       `$(DSTROOT)/usr/include`
       `$(DSTROOT)/System/Library/Frameworks/System.framework/PrivateHeaders`
 
    b. `DRIVERKIT_DATAFILES` : To make header file available to DriverKit userspace drivers -
       `$(DSTROOT)/System/DriverKit/usr/include`
 
    c. `PRIVATE_DATAFILES` : To make header file available to Apple internal in
       user level -
       `$(DSTROOT)/System/Library/Frameworks/System.framework/PrivateHeaders`
 
    d. `EMBEDDED_PRIVATE_DATAFILES` : To make header file available in user
       level for macOS as `EXTRA_DATAFILES`, but Apple internal in user level
       for embedded OSes as `EXTRA_PRIVATE_DATAFILES` -
       `$(DSTROOT)/usr/include` (`EXTRA_DATAFILES`)
       `$(DSTROOT)/usr/local/include` (`EXTRA_PRIVATE_DATAFILES`)
 
    d. `KERNELFILES` : To make header file available in kernel level -
       `$(DSTROOT)/System/Library/Frameworks/Kernel.framework/Headers`
       `$(DSTROOT)/System/Library/Frameworks/Kernel.framework/PrivateHeaders`
 
    e. `PRIVATE_KERNELFILES` : To make header file available to Apple internal
       for kernel extensions -
       `$(DSTROOT)/System/Library/Frameworks/Kernel.framework/PrivateHeaders`
 
    f. `MODULEMAPFILES` : To make module map file available in user level -
       `$(DSTROOT)/usr/include`
 
    g. `PRIVATE_MODULEMAPFILES` : To make module map file available to Apple
       internal in user level -
       `$(DSTROOT)/usr/local/include`
 
The Makefile combines the file lists mentioned above into different
install lists which are used by build system to install the header files. There
are two types of install lists: machine-dependent and machine-independent.
These lists are indicated by the presence of `MD` and `MI` in the build
setting, respectively. If your header is architecture-specific, then you should
use a machine-dependent install list (e.g. `INSTALL_MD_LIST`). If your header
should be installed for all architectures, then you should use a
machine-independent install list (e.g. `INSTALL_MI_LIST`).
 
If the install list that you are interested does not exist, create it
by adding the appropriate file lists.  The default install lists, its
member file lists and their default location are described below -
 
    a. `INSTALL_MI_LIST`, `INSTALL_MODULEMAP_MI_LIST` : Installs header and module map
       files to a location that is available to everyone in user level.
       Locations -
           $(DSTROOT)/usr/include
       Definition -
           INSTALL_MI_LIST = ${DATAFILES}
           INSTALL_MODULEMAP_MI_LIST = ${MODULEMAPFILES}
 
    b. `INSTALL_DRIVERKIT_MI_LIST` : Installs header file to a location that is
        available to DriverKit userspace drivers.
        Locations -
           $(DSTROOT)/System/DriverKit/usr/include
       Definition -
           INSTALL_DRIVERKIT_MI_LIST = ${DRIVERKIT_DATAFILES}
 
    c.  `INSTALL_MI_LCL_LIST`, `INSTALL_MODULEMAP_MI_LCL_LIST` : Installs header and
       module map files to a location that is available for Apple internal in user level.
       Locations -
           $(DSTROOT)/usr/local/include
       Definition -
           INSTALL_MI_LCL_LIST =
           INSTALL_MODULEMAP_MI_LCL_LIST = ${PRIVATE_MODULEMAPFILES}
 
    d.  `INSTALL_SF_MI_LCL_LIST` : Installs header file to a location that is available
       for Apple internal in user level.
       Locations -
           $(DSTROOT)/System/Library/Frameworks/System.framework/PrivateHeaders
       Definition -
           INSTALL_SF_MI_LCL_LIST = ${DATAFILES} ${PRIVATE_DATAFILES}
 
    e. `INSTALL_KF_MI_LIST` : Installs header file to location that is available
       to everyone for kernel extensions.
       Locations -
            $(DSTROOT)/System/Library/Frameworks/Kernel.framework/Headers
       Definition -
            INSTALL_KF_MI_LIST = ${KERNELFILES}
 
    f. `INSTALL_KF_MI_LCL_LIST` : Installs header file to location that is
       available for Apple internal for kernel extensions.
       Locations -
            $(DSTROOT)/System/Library/Frameworks/Kernel.framework/PrivateHeaders
       Definition -
            INSTALL_KF_MI_LCL_LIST = ${KERNELFILES} ${PRIVATE_KERNELFILES}
 
    g. `EXPORT_MI_LIST` : Exports header file to all of xnu (bsd/, osfmk/, etc.)
       for compilation only. Does not install anything into the SDK.
       Definition -
            EXPORT_MI_LIST = ${KERNELFILES} ${PRIVATE_KERNELFILES}
 
If you want to install the header file in a sub-directory of the paths
described in (1), specify the directory name using two variables
`INSTALL_MI_DIR` and `EXPORT_MI_DIR` as follows -
 
    INSTALL_MI_DIR = dirname
    EXPORT_MI_DIR = dirname
 
If you want to install the module map file in a sub-directory, specify the
directory name using the variable `INSTALL_MODULEMAP_MI_DIR` as follows -
 
    INSTALL_MODULEMAP_MI_DIR = dirname
 
A single header file can exist at different locations using the steps
mentioned above.  However it might not be desirable to make all the code
in the header file available at all the locations.  For example, you
want to export a function only to kernel level but not user level.
 
 You can use C language's pre-processor directive (#ifdef, #endif, #ifndef)
 to control the text generated before a header file is installed.  The kernel
 only includes the code if the conditional macro is TRUE and strips out
 code for FALSE conditions from the header file.
 Some pre-defined macros and their descriptions are -
    a. `PRIVATE` : If defined, enclosed definitions are considered System
	Private Interfaces. These are visible within xnu and
	exposed in user/kernel headers installed within the AppleInternal
	"PrivateHeaders" sections of the System and Kernel frameworks.
    b. `KERNEL_PRIVATE` : If defined, enclosed code is available to all of xnu
	kernel and Apple internal kernel extensions and omitted from user
	headers.
    c. `BSD_KERNEL_PRIVATE` : If defined, enclosed code is visible exclusively
	within the xnu/bsd module.
    d. `MACH_KERNEL_PRIVATE`: If defined, enclosed code is visible exclusively
	within the xnu/osfmk module.
    e. `XNU_KERNEL_PRIVATE`: If defined, enclosed code is visible exclusively
	within xnu.
    f. `KERNEL` :  If defined, enclosed code is available within xnu and kernel
       extensions and is not visible in user level header files.  Only the
       header files installed in following paths will have the code -
            $(DSTROOT)/System/Library/Frameworks/Kernel.framework/Headers
            $(DSTROOT)/System/Library/Frameworks/Kernel.framework/PrivateHeaders
    g. `DRIVERKIT`: If defined, enclosed code is visible exclusively in the
    DriverKit SDK headers used by userspace drivers.
Module map file name convention
===============================
In the simple case, a subdirectory of `usr/include` or `usr/local/include`
can be represented by a standalone module. Where this is the case, set
`INSTALL_MODULEMAP_MI_DIR` to `INSTALL_MI_DIR` and install a `module.modulemap`
file there. `module.modulemap` is used even for private modules in
`usr/local/include`; `module.private.modulemap` is not used. Caveat: in order
to stay in the simple case, the module name needs to be exactly the same as
the directory name. If that's not possible, then the following method will
need to be applied.
 
`xnu` contributes to the modules defined in CoreOSModuleMaps by installing
module map files that are sourced from `usr/include/module.modulemap` and
`usr/local/include/module.modulemap`. The naming convention for the `xnu`
module map files are as follows.
 
    a. Ideally the module map file covers an entire directory. A module map
       file covering `usr/include/a/b/c` would be named `a_b_c.modulemap`.
       `usr/local/include/a/b/c` would be `a_b_c_private.modulemap`.
    b. Some headers are special and require their own module. In that case,
       the module map file would be named after the module it defines.
       A module map file defining the module `One.Two.Three` would be named
       `one_two_three.modulemap`.
 
Conditional compilation
=======================
 
`xnu` offers the following mechanisms for conditionally compiling code:
 
    a. *CPU Characteristics* If the code you are guarding has specific
    characterstics that will vary only based on the CPU architecture being
    targeted, use this option. Prefer checking for features of the
    architecture (e.g. `__LP64__`, `__LITTLE_ENDIAN__`, etc.).
    b. *New Features* If the code you are guarding, when taken together,
    implements a feature, you should define a new feature in `config/MASTER`
    and use the resulting `CONFIG` preprocessor token (e.g. for a feature
    named `config_virtual_memory`, check for `#if CONFIG_VIRTUAL_MEMORY`).
    This practice ensures that existing features may be brought to other
    platforms by simply changing a feature switch.
    c. *Existing Features* You can use existing features if your code is
    strongly tied to them (e.g. use `SECURE_KERNEL` if your code implements
    new functionality that is exclusively relevant to the trusted kernel and
    updates the definition/understanding of what being a trusted kernel means).
 
It is recommended that you avoid compiling based on the target platform. `xnu`
does not define the platform macros from `TargetConditionals.h`
(`TARGET_OS_OSX`, `TARGET_OS_IOS`, etc.).
 
 
Debugging xnu
=============
 
By default, the kernel reboots in the event of a panic.
This behavior can be overriden by the `debug` boot-arg -- `debug=0x14e` will cause a panic to wait for a debugger to attach.
To boot a kernel so it can be debugged by an attached machine, override the `kdp_match_name` boot-arg with the appropriate `ifconfig` interface.
Ethernet, Thunderbolt, and serial debugging are supported, depending on the hardware.
 
Use LLDB to debug the kernel:
 
    ; xcrun -sdk macosx lldb <path-to-unstripped-kernel>
    (lldb) gdb-remote [<host-ip>:]<port>
 
The debug info for the kernel (dSYM) comes with a set of macros to support kernel debugging.
To load these macros automatically when attaching to the kernel, add the following to `~/.lldbinit`:
 
    settings set target.load-script-from-symbol-file true
 
`tools/lldbmacros` contains the source for these commands.
See the README in that directory for their usage, or use the built-in LLDB help with:
 
    (lldb) help showcurrentstacks