NFV network function virtualization security considerations

I have been asked to write down a few things related to NFV and security. NFV is relatively a new thing in the IT world. It has been on the news in 2012 and since then it has followed the developing path common to the virtualization technologies.

Virtualization has made dramatic improvement in the last years. It all started at first with simple virtualization platforms, of course VMware on top of our mind, but not only. The idea was to abstract HW platforms from software ones.

Developing the idea, the abstraction growth covering multiple hw platforms moving also to the multisite in WAN and geographical development. We call this sort of implementation nowadays cloud, but all the cloud story started from the old virtualization idea.

While this platform change was taking place, the world of services was experimenting different clientserver options (web services and so on).

With the new platforms taking place it was clear the network part would have followed this trend, moving to software and virtual shores.

Form the network point of view the first step has been the SDN (Software Defined Network).

Software defined networks (SDN) allow dynamic changes of network configuration that can alter network function characteristics and behaviors. For example, SDN can render real-time topological changes of a network path. A SDN-enabled network provides a platform on which to implement a dynamic chain of virtualized network services that make up an end-to-end network service.

SDN basically allow to centrally administer, manage, configure network services creating policies that can be related to different needs and able to adapt to a changing environment.

But this level of abstraction was not enough to cover the needed flexibility of the new implementation of modern datacenter, cloud and virtualized environment.

In a SDN environment the network gears remain mainly real solid box in an environment that is way more virtualized.

The first attempt to hybridize the physical network with the virtual one was the introduction of the first virtual network element as switches and firewalls. Those components were sometimes part of the hypervisor of the virtualizing platform, sometimes virtual appliances able to run inside a virtual environment as virtual appliances.

Those solutions were (are, since actually exist) good t target specific needs but were not covering the needed flexibility, resilience and scalability required to modern virtualization systems. Products like VMware’s vShield, Cisco’s ASA 1000v and F5 Networks‘ vCMP brought improvements in management and licensing more suited to service provider needs. Each used different architectures to accomplish those goals, making a blending of approaches difficult. But the lack of a comprehensive approach was making difficult to expand those services extensively.

The natural step of the process of virtualization would have be to define something to address in a more comprehensive way the need to transfer part of the network function inside the virtual environment.

Communications service providers and network operators came together through ETSI to try to address the management issues around virtual appliances that handle network functions.

NFV represents a decoupling of the software implementation of network functions from the underlying hardware by leveraging virtualization techniques. NFV offers a variety of network functions and elements, including routing, content delivery networks, network address translation, virtual private networks (VPNs), load balancing, intrusion detection and prevention systems (IDPS), and firewalls. Multiple network functions can be consolidated into the same hardware or server. NFV allows network operators and users to provision and execute on-demand network functions on commodity hardware or CSP platforms.

NFV does not depend on SDN (and vice-versa) and can be implemented without it. However, SDN can improve performance and enable a rich feature set known as Dynamic Virtual Network Function Service Chaining (or VNF Service Chaining). This capability simplifies and accelerates deployment of NFV-based network functions.

Based on the framework introduced by the European Telecommunications Standards Institute (ETSI), NFV is built on three main domains:

VNF,
NFV infrastructure, and
NFV management and orchestration (MANO).

VNF can be considered as a container of network services provisioned by software, very similar to a VM operational model. The infrastructure part of NFV includes all physical resources (e.g., CPU, memory, and I/O) required for storage, computing and networking to prepare the execution of VNFs. The management of all virtualization-specific tasks in NFV framework is performed by NFV management and orchestration domain. For instance, this domain orchestrates and manages the lifecycle of resources and VNFs, and also controls the automatic remote installation of VNFs.

The resulting environment now is a little bit more complicated than a few years before.

Where in the past we used to have

physical servers running Operative Systems as Linux, Unix or Windows bound to the specific hardware platform, and almost monolithic services running on those solutions,
physical storage unit running on different technologies and network (Ethernet, iscasi, fiber optic and so on),
network connected through physical devices, with some specific unit providing external access (VPN servers)
and protected by some sort of security unit providing some sort of control (firewall, IPSIDS, 802.1x, AAA and so on)
managed quite independently trough different interfaces or programs

now we moved to a world where we have

a virtualized environment where services (think as an example at Docker implementations) or entire operating systems run on a virtual machines (VMs) that manage the abstraction with the hardware

and is able to allocate resources dynamically in terms of performance and even geographic locations,

a network environment which services are partially virtualized (as in VNF implementation) and partially physical and interact with the virtual environment dynamically

a network configured dynamically through control software (SDN) which can dynamically and easily modify the network topology itself in order to respond to the changing request coming from the environment (users, services, processes).

Nowadays, the impressive effects of network functions virtualization (NFV) are evident in the wide range of applications from IP node implementations (e.g., future Internet architecture) to mobile core networks. NFV allows network functions (e.g., packet forwarding and dropping) to be performed in virtual machines (VMs) in a cloud infrastructure rather than in dedicated devices. NFV as an agile and automated network is desirable for network operators due to the ability of easily developing new services and the capabilities of self-management and network programmability via software-defined networking (SDN). Furthermore, co-existence with current networks and services leads to improve customer experience, and reduces the complexity, capital expenditure (CAPEX), and operational expenditure (OPEX).

In theory, virtualization broadly describes the separation of resources or requests for a service from the underlying physical delivery of that service. In this view, NFV involves the implementation of network functions in software that can run on a range of hardware, which can be moved without the need for installation of new equipment. Therefore, all low-level physical network details are hidden and the users are provided with the dynamic configuration of network tasks.

Everything seems so better and easy, but all those transformation does not come out without a price in terms of security.

Every step into virtualization bring security concerns, related to the control plane (think of hypervisor security, orchestrator security), the communication plane, the virtual environment itself (that often inherit the same problem of the physical platform), and the transition interface between the physical and virtual world.

Despite many advantages, therefore NFV introduces new security challenges. Since all software-based virtual functions in NFV can be configured or controlled by an external entity (e.g., third-party provider or user), the whole network could be potentially compromised or destroyed. For example, in order to properly reduce hosts’ heavy workloads, a hypervisor in NFV can dynamically try to achieve the load-balance of assigned loads for multiple VMs through a flexible and programmable networking layer which is known as virtual switch; however, if the hypervisor is compromised, all network functions can be disabled completely (a good old Ddos) or priority can be provided to some services instead others.

Also, NFV’s attack surface is considerably increased, compared with traditional network systems. Besides network resources (e.g., routers, switches, etc.) in the traditional networks, virtualization environments, live migration, and multi-tenant common infrastructure could also be attacked in NFV. For example, an at- tacker can snare a dedicated virtualized network function (VNF) and then spread out its bots in a victim’s whole network using the migration and multicast ability of NFV. To make matters worse, the access to a common infrastructure for a multi-tenant network based on NFV inherently allows for other security risks due to the shared resources between VMs. For example, in a data center network (DCN), side-channels (e.g., cache-based side channel) attacks and/or operational interference could be introduced unless the shared resources between VMs is securely controlled with proper security policies. In practice, it is not easy to provide a complete isolation of VNFs in DCNs.

The challenge related to secure a VFN are complex because are related to all the element that compose the environment: physical, virtual and control.

According to CSA Securing this environment is challenging for at least the following reasons:

Hypervisor dependencies: Today, only a few hypervisor vendors dominate the marketplace, with many vendors hoping to become market players. Like their operating system vendor counterparts, these vendors must address security vulnerabilities in their code. Diligent patching is critical. These vendors must also understand the underlying architecture, e.g., how packets flow within the network fabric, various types of encryption and so forth.
Elastic network boundaries: In NFV, the network fabric accommodates multiple functions. Placement of physical controls are limited by location and cable length. These boundaries are blurred or non-existent in NFV architecture, which complicates security matters due to the unclear boundaries. VLANs are not traditionally considered secure, so physical segregation may still be required for some purposes.
Dynamic workloads: NFV’s appeal is in its agility and dynamic capabilities. Traditional security models are static and unable to evolve as network topology changes in response to demand. Inserting security services into NFV often involves relying on an overlay model that does not easily coexist across vendor boundaries.
Service insertion: NFV promises elastic, transparent networks since the fabric intelligently routes packets that meet configurable criteria. Traditional security controls are deployed logically and physically inline. With NFV, there is often no simple insertion point for security services that are not already layered into the hypervisor.
Stateful versus stateless inspection: Today’s networks require redundancy at a system level and along a network path. This path redundancy cause asymmetric flows that pose challenges for stateful devices that need to see every packet in order to provide access controls. Security operations during the last decade have been based on the premise that stateful inspection is more advanced and superior to stateless access controls. NFV may add complexity where security controls cannot deal with the asymmetries created by multiple, redundant network paths and devices.
Scalability of available resources: As earlier noted, NFV’s appeal lies in its ability to do more with less data center rack space, power, and cooling.

Dedicating cores to workloads and network resources enables resource consolidation. Deeper inspection technologies—next-generation firewalls and Transport Layer Security (TLS) decryption, for example—are resource intensive and do not always scale without offload capability. Security controls must be pervasive to be effective, and they often require significant compute resources.

Together, SDN and NFV create additional complexity and challenges for security controls. It is not uncommon to couple an SDN model with some method of centralized control to deploy network services in the virtual layer. This approach leverages both SDN and NFV as part of the current trend toward data center consolidation.

NFV Security Framework try to address those problems.

If we want to dig the security part a little deeper we can analyze

Network function-specific security issues

and

Generic virtualization-related security issues

Network function-specific threats refer to attacks on network functions and/or resources (e.g., spoofing, sniffing and denial of service).

The foundation of NFV is set on network virtualization. In this NFV environment, a single physical infrastructure is logically shared by multiple VNFs. For these VNFs, providing a shared, hosted network infrastructure introduces new security vulnerabilities. The general platform of network virtualization consists of three entities; the providers of the network infrastructure, VNF providers, and users. Since the system consists of different operators, undoubtedly, their cooperation cannot be perfect and each entity may behave in a non-cooperative or greedy way to gain benefits.

The virtualization threats of NFV can be originated by each entity and may target the whole or part of the system.

In this view, we need to consider the threats, such as side-channel or flooding attacks as common attacks, and hypervisor, malware injection or VM migration related attacks as the virtualization and cloud specific attacks.

Basically VNF add a new layer of security concerns to the virtualizedcloud platforms for at least 3 reasons:

It inherits all the classic network security issues and expand them to cloud level

This means once a VNF is compromised there are good chances it can spread the attack or problem to the whole environment affecting not only the resources directly assigned but anything connected to the virtual environment. Think, as an example, the level of damage that can be provided performing a Ddos that deplete rapidly all the cloud network resources modifying, as an example, the Qos parameters and not using the traditional flooding techniques (which are anyway available).

It depends to several layers of abstraction and controls

Orchestrator and hypervisor are, as a matter of fact, great attack point since can

It requires a better planned implementation than the classic physical one,

With a tighter control on who is managing the management interfaces since, in common with SDN, VNF is more exposed to unauthorized access and configuration-related issues.

Still VNF requires studies and analysis from security perspective, the good part is that this is a new technology under development therefore there are big space for improvement.