LibVMI is a virtual machine introspection library. This means that it helps you access the memory of a running virtual machine. LibVMI provides primitives for accessing this memory using physical or virtual addresses and kernel symbols. LibVMI also supports accessing memory from a physical memory snapshot, which is helpful for debugging or forensic analysis.
In addition to memory access, LibVMI supports memory events. Events provide notifications when registered regions of memory are executed, written to, or read. Memory events require hypervisor support and are currently only available with Xen.
LibVMI is designed to run on Linux (file, Xen, or KVM access) or Mac OS X (file access only). The most used platform is Linux + Xen, but the others are well tested and worth exploring as well. LibVMI can provide access to physical memory for any operating system, and access to virtual memory and kernel symbols from Windows and Linux.
If you would like higher level semantic information, then we suggest using LibVMI with PyVMI (python wrapper, included with LibVMI) and Volatility. Volatility (http://code.google.com/p/volatility/) is a forensic memory analysis framework supporting both Linux and Windows systems that can aid significantly in performing useful memory analysis tasks. PyVMI includes a Volatility address space plugin that enables you to use Volatility on a live virtual machine.
This file contains very basic instructions to get you up and running. If you want more details about installation, or programming with LibVMI, then see the documentation included in the doc/ subdirectory of LibVMI, or view the documentation online at http://www.libvmi.com.
The following libraries are used in building this code:
- libxc (from Xen, the Xen Control library, required for Xen support)
- libxenstore (from Xen, access to the xenstore, required for Xen support)
- libvirt (from Red Hat, access to KVM guests, 0.8.7 or newer required for KVM support, MUST BE BUILT WITH QMP SUPPORT -- THIS REQUIRES yajl)
- qemu-kvm patch (option 1 for KVM memory access, optional for KVM support, still buggy but faster than alternative option, see Note 2)
- gdb enabled kvm VM (option 2 for KVM memory access, optional for KVM support, more stable than option 1 but slower, see Note 2)
- yacc OR bison (for reading the configuration file)
- lex OR flex (for reading the configuration file)
- glib (version 2.22 or better is required)
Note 1: If you are installing a packaged version of Xen, you will likely need to install something like 'xen-devel' to obtain the files needed from libxc and libxenstore in the dependencies listed above.
Note 2: If you want KVM support then you will need to build your own version of QEMU-KVM or enable GDB support for your VM. See the section on KVM support (below) for additional information.
For complete details on installation and configuration, please see the related online documentation:
http://code.google.com/p/vmitools/wiki/LibVMIInstallation
LibVMI is written in C. If you would rather work with Python, then look at the tools/pyvmi/ directory after installing LibVMI. PyVMI provides a feature complete python interface to LibVMI with a relatively small performance overhead.
If you would like LibVMI to work on Xen domains, you must simply ensure that you have Xen installed along with any Xen development packages. LibVMI should effectively just work on any recent version of Xen.
Compiling LibVMI on a XenServer dom0 can be challenging as there are no development headers and tools present. The recommended way to compile is in a separate CentOS installation using the Xen development packages matching what XenServer runs on. Make sure to remove the libvirt packages from CentOS (yum remove libvirt*) as to avoid the KVM driver getting activated in LibVMI. The compiled LibVMI library and tools can then be transferred to the XenServer dom0 and run natively.
If you would like LibVMI to work on KVM VM's, you must do some additional setup. This is because KVM doesn't have much built-in capability for introspection. For KVM support you need to do the following:
Ensure that you have libvirt version 0.8.7 or newer
Ensure that your libvirt installation supports QMP commands, most prepackaged versions do not support this by default so you may need to install libvirt from source yourself. To enable QMP support when installing from source, ensure that you have libyajl-dev (or the equivalent from your linux distro) installed, then run the configure script from libvirt. Ensure that the configure script reports that it found yajl. Then run make && make install.
Choose a memory access technique:
Patch QEMU-KVM with the provided patch. This technique will provide the fastest memory access, but is buggy and may cause your VM to crash / lose data / etc. To use this method, follow the instructions in the libvmi/tools/qemu-kvm-patch directory.
Enable GDB access to your KVM VM. This is done by adding '-s' to the VM creation line or, by modifying the VM XML definition used by libvirt as follows:
Change:
<domain type='kvm'>
to:
<domain type='kvm' xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'>
Add:
<qemu:commandline> <qemu:arg value='-s'/> </qemu:commandline>
under the <domain> level of the XML.
You only need one memory access technique. LibVMI will first look for the QEMU-KVM patch and use that if it is installed. Otherwise it will fall back to using GDB. So if you want to use GDB, you should both enable GDB and ensure that QEMU-KVM does not have the LibVMI patch.
If you would like LibVMI to work on physical memory snapshots saved to a file, then you don't need any special setup.
(Don't mix up with VM snapshot file) This technique will provide a very fast and coherent memory access, except the creation of shm-snapshot can take 0.2 ~ 1.4 seconds (KVM) when the memory size of guest VM expands from 512MB to 3GB. Shm-snapshot supports both KVM and Xen. However,shm-snapshot for Xen is currently created by LibVMI, hence unreal. Moreover,it takes more time (about 3 seconds in 1GB guest memory settings) to create Xen "shm-snapshot" because we have to probe unmmapable memory page holes one by one. Shm-snapshot is shiped with direct guest memory access, a non-copy access technique that can drastically reduce the latency of guest memory access. For KVM, we support both vmi_get_dgpma() and vmi_get_dgvma(); for Xen, however, due to the unreal shm-snapshot, we only support vmi_get_dgpma() now.
If you would like LibVMI to work on a shm-snapshot, then you need to do the following:
(P.S: If you use Xen, just to start on step 3)
- ensure that your libvirt installation supports QMP commands.
- patch QEMU-KVM with the provided shm-snapshot patch.
- cd qemu-1.6 patch -p1 < [libvmi_dir]/tools/qemu-kvm-patch/kvm-physmem-access-physmem-snapshot_1.6.0.patch make make install
- ./configure --enable-shm-snapshot
- Choose a setup method :
- Add VMI_INIT_SHM_SNAPSHOT flag to vmi_int(), then vmi_init() will create a shm-snapshot and enter shm-snapshot mode automatically. Once LibVMI enters the shm-snapshot mode, memory access will be redirect to the shared memory shm-snapshot, rather than your live guest VM.
- After the vmi_init() has been called, invoke vmi_snapshot_create(vmi) to snaphsot your guest VM and enter shm-snapshot mode.
No matter which method you choose, you can turn LibVMI back to live mode by calling vmi_shm_snapshot_destroy(vmi).
Even if you didn't call vmi_shm_snapshot_destroy(vmi), vmi_destroy(vmi) will teardown the shm-snapshot if existed.
- (optional but valuable) replace your guest memory access function.
- If you ever used vmi_read_pa() like the following:
- void* buf = malloc(100); int size = vmi_read_pa(vmi, 0x1000, buf, 100); process_anything(buf, size); free(buf);
- Just feel free to change your code like the following:
- void* buf = NULL; int size = vmi_get_dgpma(vmi, 0x1000, &buf, 100); process_anything(buf, size);
For vmi_read_va(), the replacement is very similar, but only capable for KVM at present.
LibVMI also supports the use of Rekall profiles for introspecting Windows. By using Rekall profiles, LibVMI is able to bypass the use if the in-memory KdDebuggerData (KDBG) structure normally used by memory forensics tools and thus allows introspecting domains where this structure is either corrupted, or encoded (like in the case of Windows 8 x64). However, Rekall profiles have to be created for each kernel version, and therefore if an update is made to the kernel, the profile has to be re-generated, thus it's a bit less stable as the standard LibVMI configuration entries which are generally stable for specific releases of Windows.
Rekall is available at https://github.com/google/rekall. You will also need to install libjson-c-dev from your distribution's repository or compile it from source that can be found at https://github.com/json-c/json-c.
To create a Rekall profile for Windows you need to determine the PDB filename and GUID of the kernel. This can be done either by running the win-guid example shipped with LibVMI, or by accessing the kernel executable on disk (normally found in Windows' System32 folder as ntoskrnl.exe).
If you need to examine an on-disk version of the kernel (or any other PE executable), you can run the following the Rekall command:
rekall peinfo <path/to/file>
Once the PDB filename and GUID is known, creating the Rekall profile is done in two steps:
rekall fetch_pdb --pdb_filename <PDB filename> --guid <GUID> -D . rekall parse_pdb <PDB filename> > rekall-profile.json
The Rekall profile can be used directly in the LibVMI config via an additional rekall_profile entry pointing to this file with an absolute path. There is no need to specify any of the offsets normally required as those offsets will be available via the profile itself.
LibVMI uses the standard GNU build system. To compile this library, simply follow the steps below:
./autogen.sh ./configure make
The example code will work without installing LibVMI. However, you may choose to install the library into the prefix specified to 'configure' by:
make install
The default installation prefix is /usr/local. You may need to run 'ldconfig' after performing a 'make install'.
To enable LibVMI debug output, modify the libvmi/debug.h header file and recompile libvmi.
The LibVMI forums are available at https://groups.google.com/forum/#!forum/vmitools
If you are just making the transition form XenAccess, please see the transition documentation online:
http://code.google.com/p/vmitools/wiki/TransitionFromXenAccess