The official home of the LibVMI project is at
C Perl C++ Python Yacc M4 Other
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.


LibVMI: Simplified Virtual Machine Introspection

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 the LibVMI Python bindings and Volatility. Volatility ( is a forensic memory analysis framework supporting both Linux and Windows systems that can aid significantly in performing useful memory analysis tasks. The LibVMI Python bindings 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

Join the chat at


The following libraries are used in building this code:

  • autotools
  • libtool Generic library support script
  • yacc OR bison (optional, for reading the configuration file)
  • lex OR flex (optional, for reading the configuration file)
  • glib (>= 2.22)
  • libvirt (>= 0.8.7)
  • libjson-c

Installing the dependencies on Ubuntu:

$ sudo apt-get install autotools-dev automake flex bison libglib2.0-dev libvirt-dev libjson-c-dev libyajl-dev


LibVMI uses the standard GNU build system. To compile this library, simply follow the steps below:


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.

Installation and Configuration

For complete details on installation and configuration, please see the related online documentation:

Xen support

LibVMI provides support for Xen out-of-the-box. If you install Xen from source, make sure the Xen libraries compiled from source are in your LD_LIBRARY_PATH. You don't have to recompile LibVMI if you update Xen as LibVMI is able to detect what version of Xen you have dynamically at runtime.

XenServer Support

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. The compiled LibVMI library and tools can then be transferred to the XenServer dom0 and run natively.

KVM support

LibVMI will have KVM support if libvirt is available during compile time. 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.

Currently there is no native VMI support in KVM, so you have two options for adding it:

  1. 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.

  2. 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'>


      <domain type='kvm' xmlns:qemu=''>
    • Add:

        <qemu:arg value='-s'/>

      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.

Python bindings

LibVMI is written in C. If you would rather work with Python, then look at the libvmi/python` repository. They provide an almost feature complete python interface to LibVMI with a relatively small performance overhead.

File / Snapshot Support

If you would like LibVMI to work on physical memory snapshots saved to a file, then you don't need any special setup.

Shm-snapshot Support

(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 only KVM. 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();

If you would like LibVMI to work on a shm-snapshot, then you need to do the following:

  1. ensure that your libvirt installation supports QMP commands.
  2. 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
  3. ./configure --enable-shm-snapshot
  4. Choose a setup method :
  1. 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.
  2. 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.

  1. (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

Rekall profiles

LibVMI also supports the use of Rekall profiles for introspecting Windows and Linux. 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.

Rekall is available at

To create a Rekall profile for Windows you can use the script that ships with LibVMI. See for more details. 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 -f <path/to/ntoskrnl.exe>

Once the PDB filename and GUID is known, creating the Rekall profile is done in two steps:

rekall fetch_pdb <PDB filename> <GUID>
rekall parse_pdb <PDB filename> > rekall-profile.json

The PDB filename should not have the .pdb extension in the above commands.

To create a Rekall profile for Linux follow the instructions at

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.


To enable LibVMI debug output, modify the libvmi/debug.h header file and recompile libvmi.


The LibVMI forums are available at!forum/vmitools