Go to file
Josh Poimboeuf ff28767295 kmod: error handling cleanup
Cleanup the error handling a little bit and make the flow a little
clearer.
2014-04-23 14:36:15 -05:00
contrib remove systemd service unit file 2014-04-15 13:50:18 -05:00
kmod kmod: error handling cleanup 2014-04-23 14:36:15 -05:00
kpatch kpatch: look for core module in git dir 2014-04-21 21:38:00 -05:00
kpatch-build fixup review comments 2014-04-22 11:01:01 -05:00
man kpatch: allow kpatch load/unload/info for files 2014-04-15 14:59:47 -05:00
test add automated testing system 2014-03-10 14:36:11 -05:00
.gitignore add man .gz files to gitignore 2014-03-27 16:29:14 -05:00
COPYING Add a COPYING file to make it trivial to find the license in a standard place 2014-03-04 10:51:58 +01:00
Makefile kpatch: install to initrd and load patch modules on boot 2014-04-01 15:47:22 -05:00
Makefile.inc fix DRACUTDIR to be in /usr/lib/dracut 2014-04-14 12:39:19 -05:00
README.md readme: recommend use of the kpatch utility 2014-04-21 21:44:40 -05:00

README.md

kpatch: dynamic kernel patching

kpatch is a Linux dynamic kernel patching tool which allows you to patch a running kernel without rebooting or restarting any processes. It enables sysadmins to apply critical security patches to the kernel immediately, without having to wait for long-running tasks to complete, users to log off, or for scheduled reboot windows. It gives more control over uptime without sacrificing security or stability.

kpatch is currently in active development. For now, it should not be used in production environments.

WARNING: Use with caution! Kernel crashes, spontaneous reboots, and data loss may occur!

Installation

NOTE: These installation instructions are currently Fedora-specific. Support for other distributions is planned soon.

Install the dependencies for compiling kpatch:

sudo yum install gcc kernel-devel elfutils elfutils-devel

NOTE: Ensure you have elfutils-0.158 or newer.

Install the dependencies for the "kpatch-build" command:

sudo yum install rpmdevtools pesign
sudo yum-builddep kernel

# optional, but highly recommended
sudo yum install ccache

Compile kpatch:

make

OPTIONAL: Install kpatch to /usr/local:

sudo make install

Alternatively, the kpatch and kpatch-build scripts can be run directly from the git tree.

Quick start

NOTE: While kpatch is designed to work with any recent Linux kernel on any distribution, the "kpatch-build" command currently only works on Fedora.

Make a source patch against the kernel tree:

# from a kernel git tree:
git diff > /path/to/foo.patch

Build the patch module:

kpatch-build /path/to/foo.patch

This outputs a patch module named kpatch-foo.ko in the current directory. Now apply it to the running kernel:

sudo kpatch load kpatch-foo.ko

Done! The kernel is now patched.

How it works

kpatch works at a function granularity: old functions are replaced with new ones. It has four main components:

  • kpatch-build: a collection of tools which convert a source diff patch to a patch module. They work by compiling the kernel both with and without the source patch, comparing the binaries, and generating a patch module which includes new binary versions of the functions to be replaced.

  • patch module: a kernel module (.ko file) which includes the replacement functions and metadata about the original functions.

  • kpatch core module: a kernel module (.ko file) which provides an interface for the patch modules to register new functions for replacement. It uses the kernel ftrace subsystem to hook into the original function's mcount call instruction, so that a call to the original function is redirected to the replacement function.

  • kpatch utility: a command-line tool which allows a user to manage a collection of patch modules. One or more patch modules may be configured to load at boot time, so that a system can remain patched even after a reboot into the same version of the kernel.

kpatch-build

The "kpatch-build" command converts a source-level diff patch file to a kernel patch module. Most of its work is performed by the kpatch-build script which uses a collection of utilities: create-diff-object, add-patch-section, and link-vmlinux-syms.

The primary steps in kpatch-build are:

  • Build the unstripped vmlinux for the kernel
  • Patch the source tree
  • Rebuild vmlinux and monitor which objects are being rebuilt. These are the "changed objects".
  • Recompile each changed object with -ffunction-sections -fdata-sections, resulting in the changed patched objects
  • Unpatch the source tree
  • Recompile each changed object with -ffunction-sections -fdata-sections, resulting in the changed original objects
  • Use create-diff-object to analyze each original/patched object pair for patchability and generate an output object containing modified sections
  • Link all the output objects into a cumulative object
  • Use add-patches-section to add the .patches section that the kpatch core module uses to determine the list of functions that need to be redirected using ftrace
  • Generate the patch module
  • Use link-vmlinux-syms to hardcode non-exported kernel symbols into the symbol table of the patch module

Patching

The patch modules register with the core module (kpatch.ko). They provide information about original functions that need to be replaced, and corresponding function pointers to the replacement functions.

The core module registers a trampoline function with ftrace. The trampoline function is called by ftrace immediately before the original function begins executing. This occurs with the help of the reserved mcount call at the beginning of every function, created by the gcc -mfentry flag. The trampoline function then modifies the return instruction pointer (IP) address on the stack and returns to ftrace, which then restores the original function's arguments and stack, and "returns" to the new function.

Limitations

  • Patches which modify kernel modules are not supported (yet). Only functions in the vmlinux file (listed in System.map) can be patched.

  • Patches to functions which are always on the stack of at least one process in the system are not supported. Examples: schedule(), sys_poll(), sys_select(), sys_read(), sys_nanosleep(). Attempting to apply such a patch will cause the insmod of the patch module to return an error.

  • Patches which modify init functions (annotated with __init) are not supported. kpatch-build will return an error if the patch attempts to do so.

  • Patches which modify statically allocated data are not supported. kpatch-build will detect that and return an error. (In the future we will add a facility to support it. It will probably require the user to write code which runs at patch module loading time which manually updates the data.)

  • Patches which change the way a function interacts with dynamically allocated data might be safe, or might not. It isn't possible for kpatch-build to verify the safety of this kind of patch. It's up to the user to understand what the patch does, whether the new functions interact with dynamically allocated data in a different way than the old functions did, and whether it would be safe to atomically apply such a patch to a running kernel.

Frequently Asked Questions

Q. Isn't this just a virus/rootkit injection framework?

kpatch uses kernel modules to replace code. It requires the CAP_SYS_MODULE capability. If you already have that capability, then you already have the ability to arbitrarily modify the kernel, with or without kpatch.

Q. How can I detect if somebody has patched the kernel?

We hope to create a new kernel TAINT flag which will get set whenever a patch module is loaded.

Also, many distros ship with cryptographically signed kernel modules, and will taint the kernel anyway if you load an unsigned module.

Q. Will it destabilize my system?

No, as long as the patch is chosen carefully. See the Limitations section above.

Q. Why does kpatch use ftrace to jump to the replacement function instead of adding the jump directly?

ftrace owns the first "call mcount" instruction of every kernel function. In order to keep compatibility with ftrace, we go through ftrace rather than updating the instruction directly.

Q Is kpatch compatible with <insert kernel debugging subsystem here>?

We aim to be good kernel citizens and maintain compatibility. A hot patch replacement function is no different than a function loaded by any other kernel module. Each replacement function has its own symbol name and kallsyms entry, so it looks like a normal function to the kernel.

  • oops stack traces: Yes. If the replacement function is involved in an oops, the stack trace will show the function and kernel module name of the replacement function, just like any other kernel module function. The oops message will also show the taint flag. [TODO: taint flag]
  • kdump/crash: Yes. Replacement functions are normal functions, so crash will have no issues. [TODO: create patch module debuginfo symbols and crash warning message]
  • ftrace: Yes, see previous question.
  • systemtap/kprobes: TODO: try it out
  • perf: TODO: try it out

Q. Why not use something like kexec instead?

If you want to avoid a hardware reboot, but are ok with restarting processes, kexec is a good alternative.

Q. If an application can't handle a reboot, it's designed wrong.

That's a good poi... [system reboots]

Q. What changes are needed in other upstream projects?

We hope to make the following changes to other projects:

  • kernel:

    • ftrace improvements to close any windows that would allow a patch to be inadvertently disabled
    • hot patch taint flag
    • possibly the kpatch core module itself
  • crash:

    • make it glaringly obvious that you're debugging a patched kernel
    • point it to where the patch modules and corresponding debug symbols live on the file system

Q: Is it possible to register a function that gets called atomically with stop_machine when the patch module loads and unloads?

We do have plans to implement something like that.

Q. What kernels are supported?

kpatch needs gcc >= 4.6 and Linux >= 3.7 for use of the -mfentry flag.

Q. Is it possible to remove a patch?

Yes. Just unload the patch module and the original function will be restored.

Q. Can you apply multiple patches?

Yes. Also, a single function can even be patched multiple times if needed.

Demonstration

A low-level demonstration of kpatch is available on Youtube:

http://www.youtube.com/watch?v=WeSmG-XirC4

This demonstration completes each step in the previous section in a manual fashion. However, from a end-user perspective, most of these steps are hidden by the "kpatch-build" command.

Get involved

If you have questions or feedback, join the mailing list and say hi.

Contributions are very welcome. Feel free to open issues or PRs on github. For big PRs, it's a good idea to discuss them first in github issues or on the mailing list before you write a lot of code.

License

kpatch is under the GPLv2 license.

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.