by Nicolas Delhaye, Vulnerability Researcher
The Bromium vSentry solution is a product deployed on end-user workstations which takes advantage of hardware virtualisation features in order to isolate untrusted or exposed software.
As part of this analysis, it is possible for a non-privileged user to cause a Denial of Service (DoS) in the client side application of Bromium vSentry 18.104.22.1680. A race condition permits an Out of Bound (OoB) read, resulting in kernel memory leak or Denial of Service depending on whether a read access violation occurs.
The vulnerability is located in « uxenv4vlib.sys » version 2017.1128.23039.33944. This driver is in charge of managing the uXen V4V between the host and the guest. More detail on this will be discussed later in this article.
It starts with the possibility to establish a communication channel with one of the Bromium drivers through the symbolic device name « \\.\v4vdev » from standard user rights. It is then possible to create specific IOCTLs related to the Bromium data ring management. Before going into more technical details, we need a little more explanation about the V4V technology.
Here is an extract of the v4v technology documentation
“The V4V technology is a new approach to inter-domain communications on a Xen virtualisation platform. Most existing inter-domain communications frameworks use shared memory between domains and some form of event channel interrupt mechanism. V4V is a departure from this approach, instead relying on the hypervisor to broker all communications. Domains manage their own data rings and there is no sharing of actual memory. Domains communicate with one or more other domains (or themselves in the trivial case) using source and destination addresses and ports.
The basic construct in V4V is the v4v_ring. A domain that wants to communicate with other domains must register a v4v_ring with the V4V management code in the hypervisor.”*
Source: OpenXT github
- Field “len” is the length of the v4v ring
- Field “rx_ptr” is the receive pointer into the ring where the next message to be read by the domain is located
- Field “tx_ptr” is the transmit pointer into the ring indicating where the next received message can be written into the ring
- Field “ring” is the data buffer and starts at ring
As you might guess, the vulnerability is related to the Bromium data ring management.
We will see later that it is possible to corrupt all the v4v_ring structure. Especially the “len” field, the “rx_ptr” and “tx_ptr” fields. By controlling these values, we could write or read more data than the real size of the ring buffer. Moreover, the initial size of “v4v_ring” structure can also be controlled through an IOCTL allowing the V4V initialisation for a domain. Hence, due to some checks already present in the « uxenv4vlib.sys » driver, the vulnerability cannot be triggered during a write operation but only for a read operation. Bad luck! Actually we could have expected a buffer overflow leading to a privilege escalation.
We also analysed where the vulnerability is located and how to trigger it. The latter is in « gh_v4v_process_datagram_reads » function which checks if the ring is not mapped by a user process. In this case the driver deals with the operation, otherwise a user process will be notified.
Please see below screenshot of the disassembled code highlighting the check (in green) and the ring information (in red) on access.
We can see that there is no synchronisation mechanism when the data is accessed. In fact, a race condition could occur when a thread attempts to read and another thread attempts to modify the ring data.
This scenario can be achieved because a user process can map the ring and so, can directly manipulate headers of the ring data from the user buffer. All changes are implicitly replicated to the kernel buffer.
Please see below steps for the scenario leading to the vulnerability:
- A reading thread attempts to read data. No user process has mapped the v4v ring. The driver should deal with the operation. Hence, before accessing the ring data, the execution thread is interrupted and a second thread is scheduled.
- The second thread maps the ring and then attempts to corrupt data, especially fields such as “len”, “rx_ptr” and “tx_ptr” of the “v4v_ring” structure.
- Finally, the second thread is scheduled in favour of the reading thread. This one is now using corrupted data.
As a result, we see an Out of Bound read, leading in kernel memory leak or Denial of Service depending on whether a read access violation occurs.
An important point should be taken into account to successfully exploit this vulnerability. We must use asynchronous operations. If threads open the device in synchronous mode, they stay blocked until data are available. Therefore the race condition cannot be triggered. So, the attribute FILE_FLAG_OVERLAPPED must be set to open the device for asynchronous IO.
By now, we will see the most important code portions which are used in this exploitation.
1. Opening the device
First, we have to open the device on the vulnerable driver.
2. Initialise V4V (V4V_IOCTL_INITIALIZE)
Now, we will craft the V4V_IOCTL_INITIALIZE IOCTL to initialise a V4V ring.
From the above code, the value 0x01FC0 refers to the ring buffer size in bytes. In reality, as the v4v_ring structure has a size of 0x40 bytes, the kernel will allocate 0x2000 bytes. This value was intentionally chosen in order to be aligned on the memory page size.
3. Bind the V4V ring with a VM (V4V_IOCTL_BIND)
The next step is to bind our V4V ring with a micro-VM ID. The following IOCTL V4V_IOCTL_BIND allows us to do this task.
We focus in particular on the first three fields of the “InputBuffer”. The first one represents the VM ID, the two following have been retrieved by reverse engineering. In fact, this represents a “v4v_addr” structure as described below:
4. Writing data into the ring buffer
With correct configuration, we could write in the device thanks to this piece of code.
As you might understand, a simple write in the device fills the ring buffer. However, a short header must be used to set the VM ID in order to target a specific micro-VM.
5. Creating the “MapRing” and the “Read” threads
We are ready to start two concurrency threads in charge of respectively mapping the ring and reading the data previously written.
You will notice that the two threads are created in suspended mode and resume at the “same” time. Thus, we assure a better chance of causing the race condition. Implementation of these threads are described below.
6. MapRing thread (V4V_IOCTL_MAPRING)
The V4V ring could me mapped in user land through the V4V_IOCTL_MAPRING IOCTL.
Once we have retrieved a user address on the ring, we will overwrite data before the read operation occurs. In this case, the values of “len”, “rx_ptr” and “tx_ptr” fields are set to cause a nice BSOD.
7. Read thread
The race condition could be triggered thanks to a read operation.
8. Repeat, repeat and so on
Finally, loop infinitely with this scenario by incrementing the VM ID. After a few moments, we should get a BSOD or an Out of Bound memory read depending on the values crafted in the mapped ring.
This demonstration has been tested on Windows 7 x64 Home Basic and Enterprise.
I have noticed that the hardware configuration, in particular the amount of CPUs, can affect the success of this exploitation. On large hardware configuration, we have to launch several instances of the POC to trig the vulnerability.
As soon as the vulnerability is triggered, we could get the following BSOD:
For in-depth analysis, more accurate information can be obtained through a Windows debugger.
In conclusion, as this bug is a race condition, it has been quite difficult to find it. But it is always fun to discover this type of vulnerability. Finally, we are dealing with a design mistake because the developer omits to protect data from concurrency access. Another remediation could be to hide access on the ring header from the user and so, prevent the alteration of critical fields. Nevertheless, this vulnerability is not exploitable for a privilege escalation but could help to reach this effect thanks to the memory leak.
Airbus CyberSecurity follows the widely accepted 90-day vulnerability disclosure policy; meaning Airbus CyberSecurity won’t engage any public communication about the reported vulnerability during that time frame without any prior public communication or fix. Please note that Airbus CyberSecurity’s general position is that as soon a working and active relationship is established, there is no need to blindly push for the 90-day vulnerability disclosure if it’s not necessary.
Bromium reacted quickly to this issue both to confirm the vulnerability and fix the issue.
July 22th, 2019 Vulnerability Reported
July 23th, 2019 Initial response from Bromium:
– Issue is confirmed and root cause is identified
– Patch is scheduled for end of August
– Airbus will be invited to test the forthcoming release
July 30th, 2019 Bromium committed a patch on “https://github.com/uxen-virt/uxen”
Sept 10th, 2019 Bromium made the fix available to all customers in 4.1.7 update 1
Oct 9th, 2019 Bromium provided us more details on the fix
Nov 4th, 2019 CVE process is initiated
Feb 7th, 2020 Airbus publishes the report associated with the CVE
Feb 10th, 2020 Update the Disclosure Timeline with more accurate information