Cloud Computing Security Risk Assessment
Cloud Computing Security Risk Assessment
Document History
Version 1.0
December 2012
2.0
Rev.B
About ENISA
The European Network and Information Security Agency (ENISA) is a centre of network and information security expertise for the EU, its member states, the private sector and Europes citizens. ENISA works with these groups to develop advice and recommendations on good practice in information security. It assists EU member states in implementing relevant EU legislation and works to improve the resilience of Europes critical information infrastructure and networks. ENISA seeks to enhance existing expertise in EU member states by supporting the development of cross-border communities committed to improving network and information security throughout the EU. More information about ENISA and its work can be found at www.enisa.europa.eu.
Contact details
This report has been edited by: Lionel Dupr, Thomas Haeberlen
For contacting ENISA or for general enquiries about this report, please use the following details: Email: [email protected] Internet: http://www.enisa.europa.eu
Legal notice
Notice must be taken that this publication represents the views and interpretations of the authors and editors, unless stated otherwise. This publication should not be construed to be a legal action of ENISA or the ENISA bodies unless adopted pursuant to the ENISA Regulation (EC) No 460/2004 as lastly amended by Regulation (EU) No 580/2011. This publication does not necessarily represent state-of theart and ENISA may update it from time to time. Third-party sources are quoted as appropriate. ENISA is not responsible for the content of the external sources including external websites referenced in this publication. This publication is intended for information purposes only. It must be accessible free of charge. Neither ENISA nor any person acting on its behalf is responsible for the use that might be made of the information contained in this publication. Reproduction is authorised provided the source is acknowledged. European Network and Information Security Agency (ENISA), 2011
Highly abstracted resources Near instant scalability and flexibility Near instantaneous provisioning Shared resources (hardware, database, memory, etc) Service on demand, usually with a pay as you go billing system Programmatic management (e.g., through WS API).
There are three categories of cloud computing: Software as a service (SaaS): is software offered by a third party provider, available on demand, usually via the Internet configurable remotely. Examples include online word processing and spreadsheet tools, CRM services and web content delivery services (Salesforce CRM, Google Docs, etc). Platform as a service (PaaS): allows customers to develop new applications using APIs deployed and configurable remotely. The platforms offered include development tools, configuration management, and deployment platforms. Examples are Microsoft Azure, Force and Google App engine. Infrastructure as service (IaaS): provides virtual machines and other abstracted hardware and operating systems which may be controlled through a service API. Examples include Amazon EC2 and S3, Terremark Enterprise Cloud, Windows Live Skydrive and Rackspace Cloud.
Clouds may also be divided into: Public: available publicly - any organisation may subscribe Private: services built according to cloud computing principles, but accessible only within a private network Partner or Community: cloud services offered by a provider to a limited and well-defined number of parties.
More timely, effective and efficient updates and defaults: default virtual machine images and software modules used by customers can be pre-hardened and updated with the latest patches and security settings according to fine-tuned processes; IaaS cloud service APIs also allow snapshots of virtual infrastructure to be taken regularly and compared with a baseline. Updates can be rolled out many times more rapidly across a homogenous platform than in traditional client-based systems that rely on the patching model. Rapid, smart scaling of resources: the ability of the cloud provider to dynamically reallocate resources for filtering, traffic shaping, authentication, encryption, etc, to defensive measures (e.g., against DDoS attacks) has obvious advantages for resilience. Benefits of resource concentration: Although the concentration of resources undoubtedly has disadvantages for security [see Risks], it has the obvious advantage of cheaper physical perimiterisation and physical access control (per unit resource) and the easier and cheaper application of many security-related processes.
In certain cases, it also means that using a public cloud infrastructure implies that certain kinds of compliance cannot be achieved (e.g., PCI DSS). Lock-in: there still is little on offer in the way of tools, procedures or standard data formats or services interfaces that could guarantee data, application and service portability. This can make it difficult for the customer to migrate from one provider to another or migrate data and services back to an in-
house IT environment. This introduces a dependency on a particular CP for service provision, especially if data portability, as the most fundamental aspect, is not enabled. Isolation failure: multi-tenancy and shared resources are defining characteristics of cloud computing. This risk category covers the failure of mechanisms separating storage, memory, routing and reputation between different tenants (e.g., so-called guest-hopping attacks). However it should be considered that attacks on resource isolation mechanisms (e.g.,. against hypervisors) are still less numerous and much more difficult for an attacker to put in practice compared to attacks on traditional OSs. Management interface compromise: customer management interfaces of a public cloud provider are accessible through the Internet and mediate access to larger sets of resources (than traditional hosting providers) and therefore pose an increased risk, especially when combined with remote access and web browser vulnerabilities. Data protection: cloud computing poses several data protection risks for cloud customers and providers. In some cases, it may be difficult for the cloud customer (in its role as data controller) to effectively check the data handling practices of the cloud provider and thus to be sure that the data is handled in a lawful way. This problem is exacerbated in cases of multiple transfers of data, e.g., between federated clouds. On the other hand, some cloud providers do provide information on their data handling practices. Some also offer certification summaries on their data processing and data security activities and the data controls they have in place, e.g., SAS70 certification. Insecure or incomplete data deletion: when a request to delete a cloud resource is made, as with most operating systems, this may not result in true wiping of the data. Adequate or timely data deletion may also be impossible (or undesirable from a customer perspective), either because extra copies of data are stored but are not available, or because the disk to be destroyed also stores data from other clients. In the case of multiple tenancies and the reuse of hardware resources, this represents a higher risk to the customer than with dedicated hardware. Malicious insider: while usually less likely, the damage which may be caused by malicious insiders is often far greater. Cloud architectures necessitate certain roles which are extremely high-risk. Examples include CP system administrators and managed security service providers. Customers security expectations: the perception of Security levels by Customers might differentiate from the actual security (and availability) offered by the CP, or the actual temptation of the CP to reduce costs further by sacrificing on some security aspects. Availability Chain: Reliance on Internet Connectivity at Customers end creates a Single point of failure in many cases.
The risks listed above do not follow a specific order of criticality; they are just ten of the most important cloud computing specific risks identified during the assessment. In terms of criticality, loss of governance is still considered the top risk associated with moving to the Cloud. The risks of using Cloud computing should be compared to the risks of staying with traditional solutions, such as desktop-based models. To facilitate this, the 2009 Cloud Risk Assessment contains estimates of relative risks as compared with a typical traditional environment. These were also reconsidered during the first review round, and in many cases explanations were added. It is often possible, and in some cases advisable, for the cloud customer to transfer risk to the cloud provider.However not all risks can be transferred: If a risk leads to the failure of a business, serious damage to reputation or legal implications, it is hard or impossible for any other party to compensate for this damage. Ultimately, you can outsource responsibility but you can't outsource accountability.
It is hardly necessary to repeat one again about the Put simply, all kinds of security economic, technical, architectural and ecological benefits of measures are cheaper when cloud computing. However, in the direct experience of the implemented on a larger scale. members of our expert group, as well as according to recent Therefore the same amount of news from the real world, an examination of the security investment in security buys better risks of cloud computing must be balanced by a review of its protection. specific security benefits. Cloud computing has significant potential to improve security and resilience. What follows is a description of the key ways in which it can contribute.
10
2.3 More timely and effective and efficient updates and defaults
Virtual machine images and software modules used by customers can be pre-hardened and updated with the latest patches and security settings according to fine-tuned processes; moreover, IaaS cloud service APIs also allow snapshots of virtual infrastructure to be taken regularly and compared with a baseline (e.g., to ensure software firewall rules have not changed). Updates can be rolled out many times more rapidly across a homogenous platform than in traditional client-based systems that rely on the patching model. Finally in PaaS and SaaS models the applications are more likely to have been hardened to run outside the enterprise environment, which makes them likely to be more portable and robust than the equivalent enterprise software (where it exists). They are also more likely to be regularly updated and patched in a centralized fashion minimizing the window of vulnerability.
12
Risk assessment
Likelihood of incident scenario Very Low Low Medium Business Impact High Very High
Low (Unlikely)
Medium (Possible)
High (Likely)
1 2 3 4 5
2 3 4 5 6
3 4 5 6 7
4 5 6 7 8
14
Risks
The following points should be noted in relation to the descriptions of risk below:
-
Risk should always be understood in relation to overall business opportunity and appetite for risk sometimes risk is compensated by opportunity. Cloud services are not only about convenient storage, accessible by multiple devices, but include important benefits such as more convenient communication and instant multi-point collaboration. Therefore, a comparative analysis needs to compare not only the risks of storing data in different places (on premises vs. the cloud) but also the risks when on premises-data stored on premises e.g. a spreadsheet - is emailed to other persons for their contributions, against the security issues of a spreadsheet stored in the cloud and open to collaboration between those persons. Therefore, the risks of using cloud computing should be compared to the risks of staying with traditional solutions, such as desktop-based models. The level of risk will in many cases vary significantly with the type of cloud architecture being considered. It is possible for the cloud customer to transfer risk to the cloud provider and the risks should be considered against the cost benefit received from the services. However not all risks can be transferred: if a risk leads to the failure of a business, serious damage to reputation or legal implications, it is hard or impossible for any other party to compensate for this damage. The risk analysis in this paper applies to cloud technology. It does not apply to any specific cloud computing offering or company. This paper is not meant to replace a project-specific organisational risk assessment. The level of risks is expressed from the perspective of the cloud customer.
The following table shows the distribution of the risk probabilities and impacts:
PROBABILITY 4 5 6 7 8
R.2
R.19 R.1
5 R.10
3 R.3 R.23
R.14
The risks are classified into three categories: Policy and Organizational Technical
16
Legal.
Each risk is presented in tables which include: Probability estimate Impact estimate Level of risk Comparison (where applicable) between the Classic IT and Cloud settings. The arrows show the way the risk changes when moving from the Classic IT to the Cloud setting
As in the original Risk Assessment document, we have not included an overall comparative risk since it is assumed that all the risks selected are higher in the Cloud setting.
Relying strongly on the services of one provider can lead to severe difficulties in changing the provider. Migrating to another provider may even become virtually impossible. Probability: High Impact: Medium Risk: High
Cloud services are often based on proprietary nonstandard data formats and application logic. This can make migrating data and services to another CP difficult or even impossible, A missing exit strategy exacerbates this risk. Lock-in also occurs in the classic IT setting, but in this case the customer usually has more control over the data and services.
There is currently little on offer in the way of tools, procedures or standard data formats or services interfaces that could guarantee data and service portability. This makes it extremely difficult for a customer to migrate from one provider to another, or to migrate data and services to or from an inhouse IT environment. Furthermore, cloud providers may have an incentive to prevent (directly or indirectly) the portability of their customers services and data. This potential dependency for service provision on a particular CP, depending on the CP's commitments, may lead to a catastrophic business failure should the cloud provider go bankrupt and the content and application migration path to another provider is too costly (financially or time-wise) or insufficient warning is given (no early warning).
18
The acquisition of the cloud provider can also have a similar effect, since it increases the likelihood of sudden changes in provider policy and non-binding agreements such as terms of use (ToU). It is important to understand that the extent and nature of lock-in varies according to the cloud type:
SaaS Lock-in Customer data is typically stored in a custom database schema designed by the SaaS provider. Most SaaS providers offer API calls to read (and thereby export) data records. However, if the provider does not offer a readymade data export routine, the customer will need to develop a program to extract their data and write it to file ready for import to another provider. It should be noted that there are few formal agreements on the structure of business records (e.g., a customer record at one SaaS provider may have different fields than at another provider), although there are common underlying file formats for the export and import of data, e.g., XML. The new provider can normally help with this work at a negotiated cost. However, if the data is to be brought back in-house, the customer will need to write import routines that take care of any required data mapping unless the CP offers such a routine. As customers will evaluate this aspect before making important migration decisions, it is in the long-term business interest of CPs to make data portability as easy, complete and cost-effective as possible. Application lock-in is the most obvious form of lock-in (although it is not specific to cloud services). SaaS providers typically develop a custom application tailored to the needs of their target market. SaaS customers with a large user-base can incur very high switching costs when migrating to another SaaS provider as the end-user experience is impacted (e.g., re-training is necessary). Where the customer has In using cloud infrastructures, the client developed programs to interact with the necessarily cedes control to the CP on a providers API directly (e.g., for integration number of issues which may affect security. with other applications), these will also need For example ToUs may prohibit port scans, to be re-written to take into account the new vulnerability assessment and penetration providers API. testing. Moreover, there may be conflicts between customer hardening procedures and PaaS Lock-in the cloud environment. On the other hand, PaaS lock-in occurs at both the API layer (ie, platform SLAs may not offer a commitment to provide specific API calls) and at the component level. For such services on the part of the cloud provider, example, the PaaS provider may offer a highly thus leaving a gap in security defenses. efficient back-end data store. Not only must the customer develop code using the custom APIs offered Moreover the cloud provider may outsource or by the provider, but they must also code data access sub-contract services to third-parties routines in a way that is compatible with the back-end (unknown providers) which may not offer the same guarantees (such as to provide the service in a lawful way) as issued by the cloud provider. Or the control of the cloud provider changes, so the terms and conditions of their services may also change.
data store. This code will not necessarily be portable across PaaS providers, even if a seemingly compatible API is offered, as the data access model may be different (e.g., relational v hashing). PaaS lock-in at the API layer happens as different providers offer different APIs. PaaS lock-in happens at the runtime layer as standard runtimes are often heavily customised to operate safely in a cloud environment. For example, a Java runtime may have dangerous calls removed or modified for security reasons. The onus is on the customers' developers to understand and take into account these differences. PaaS also suffers from data lock-in, in the same way as in SaaS, but in this case the onus is completely on the customer to create compatible export routines.
IaaS-Lock-in IaaS lock-in varies depending on the specific infrastructure services consumed. For example, a customer using cloud storage will not be impacted by non-compatible virtual machine formats. IaaS computing providers typically offer hypervisor based virtual machines. Software and VM metadata is bundled together for portability typically just within the providers cloud. Migrating between providers is non-trivial until open standards, such as OVF (11), are adopted. IaaS storage provider offerings vary from simplistic key/value based data stores to policy enhanced file based stores. Feature sets can vary significantly, hence so do storage semantics. However application level dependence on specific policy features (e.g., access controls) may limit the customers choice of provider. Data lock-in is the obvious concern with IaaS storage services. As cloud customers push more data to cloud storage, data lock-in increases unless the CP provides for data portability.
Common to all providers is the possibility of a run on the banks scenario for a cloud provider. For this scenario, suppose there is a crisis of confidence in the cloud providers financial position, and therefore a mass exit and withdrawal of content on a first come, first served basis. Then, in a situation where a provider limits the amount of content (data and application code) which can be withdrawn in a given timeframe, some customers will never be able to retrieve their data and applications. As in any new IT market, competitive pressure, an inadequate business strategy, lack of financial support, etc, could lead some providers to go out of business or at least to force them to restructure their service portfolio offering. In other words, it is possible that in the short or medium term some cloud computing services could be terminated. The impact of this threat for the cloud customer is easily understandable, since it could lead to a loss or deterioration of service delivery performance, and quality of service, as well as a loss of investment. Furthermore, failures in the services outsourced to the CP may have a significant impact on the cloud customers ability to meet its duties and obligations to its own customers. The customer of the cloud provider may thus be exposed to contractual and tortuous liability to its customers based on its
20
providers negligence. Failures by the cloud provider may also result in liability by the customer to its employees. Acquisition of the cloud provider could increase the likelihood of a strategic shift and may put nonbinding agreements at risk (e.g., software interfaces, security investments, non-contractual security controls). This could make it impossible to comply with the security requirements. The final impact could be damaging for crucial assets such as: the organizations reputation, customer or patient trust, and employee loyalty and experience.
R.2 Loss of governance Risk number and name Short description R.2 Loss of governance
When using Cloud services, the CC necessarily cedes control to the CP on a number of issues which may affect security. Probablity: Very High Impact: Very High Risk: Very High
Ceding control to a service provider lies in the nature of Cloud services, and of any kind of outsourcing in general, and is therefore almost certain to occur. Loss of governance might occur in classic IT scenarios, as well. If it occurs, the impact will be similar.
In using cloud infrastructures, the client necessarily cedes control to the CP on a number of issues which may affect security. For example ToUs may prohibit port scans, vulnerability assessment and penetration testing. Moreover, there may be conflicts between customer hardening procedures and the cloud environment. On the other hand, SLAs may not offer a commitment to provide such services on the part of the CP, thus leaving a gap in security defenses. Moreover the CP may outsource or sub- Certain organisations migrating to the cloud contract services to third-parties (unknown have made considerable investments in providers) which may not offer the same achieving certification either for competitive guarantees (such as to provide the service in a advantage or to meet industry standards or lawful way) as issued by the CP (see also regulatory requirements (eg, PCI DSS). below). Or the control of the CP changes, so the terms and conditions of their services may also change. The loss of governance and control could have a potentially severe impact on the organizations strategy and therefore on the capacity to meet its mission and goals. The loss of control and governance can also lead to the impossibility of complying with the security requirements, a lack of confidentiality, integrity and availability of data, and a deterioration of performance and quality of service, not to mention the introduction of compliance challenges, as certain organisations migrating to the cloud have made considerable investments in achieving certification either for competitive
22
advantage or to meet industry standards or regulatory requirements (e.g., PCI DSS). This investment may be put at risk by a migration to the cloud: if the CP cannot provide evidence of their own compliance to the relevant requirements; if the CP does not permit audit by the CC. Resource sharing means that malicious activities carried out by one tenant may affect the reputation of another tenant.
In certain cases, it even means that using a public cloud infrastructure implies that certain kinds of compliance cannot be achieved and hence cloud hosted services cannot be used for services that need them. For example, EC2 says customers would be hard-pressed to achieve PCI compliance on their platform. So EC2 hosted services cannot be used to handle credit card transactions. A CP can also outsource certain specialised tasks of its production chain to third parties. In such a situation the level of security of the CP may depend on the level of security of each one of the links and the level of dependency of the CP on the third party. Any interruption or corruption in the chain or a lack of coordination of responsibilities between all the parties involved can lead to: unavailability of services, loss of data confidentiality, integrity and availability, economic and reputational losses due to failure to meet customer demand, violation of SLA, cascading service failure, etc. An important example here is where a critical dependency exists on a third party single-sign-on or identity management service. In this case, an interruption of the third party service or of the CPs connection to the service or a weakness in their security procedures may compromise the availability or confidentiality of a CC or indeed the entire cloud offering. In general, a lack of transparency in the contract can be a problem for the whole system. If a CP does not declare which core IT services are outsourced - it is not realistic that providers should list the contractors since these may change frequently - the CC is not in a position to properly evaluate the risk he is facing. This lack of transparency could decrease the level of trust in the provider.
R.3 Suppy Chain Failure Risk number and name Short description R.3 Suppy Chain Failure
A CP can outsource parts of its production chain to third parties, or even use other CPs as part of its service. This way, a potential for cascading failures is created Probability: Low Impact: Medium Risk: Medium
Supply chain failiures can occur also in classic IT setups, but the probability is considered to be higher in Cloud scenarios. The potential impact of supply chain failures has to be considered higher than in classic IT setups, because the supply chain is likely to be longer and the customers options fewer.
A CP can outsource certain specialised tasks of its production chain to third parties, or even use another cloud service as a backend. In such a situation the level of security of the cloud service depends on the level of security of each one of the links and the level of dependency of the cloud provider on the third party. Any interruption or corruption in the chain or a lack of coordination of responsibilities between all the parties involved can lead to: unavailability of services, loss of data confidentiality, integrity and availability, economic and reputational losses due to failure to meet customer demand, violation of SLA, cascading service failure, etc. An important example here is where a critical dependency exists on a third party single-sign-on or identity management service. In this case, an interruption of the third party service or of the CPs connection to the service or a weakness in their security procedures may compromise the availability or confidentiality of a cloud customer or indeed the entire cloud offering. In general, a lack of transparency in the contract can be a problem for the whole system. If a CP does not declare which core IT services are outsourced - it is not realistic that providers should list the contractors since these may change frequently - the customer is not in a position to properly evaluate the risk he is facing. This lack of transparency could decrease the level of trust in the provider.
24
R.4 Conflicts between customer hardening procedures and cloud environment Risk number and name R.4 Conflicts between customer hardening procedures and cloud environment
Short description
Certain security measures of a CC may conflict with a CPs environment, making their implementation by the CC impossible. Probability: Medium n.a. Impact: Medium Risk: Medium
n.a.
CPs must set out a clear segregation of responsibilities that articulates the minimum actions customers must undertake. The failure of customers to properly secure their environments may pose a vulnerability to the cloud platform if the CP has not taken the necessary steps to provide isolation. CPs Customers must realize and assume their should further articulate their isolation mechanisms responsibility as failure to do so would place and provide best practice guidelines to assist their data and resources at further risk. customers to secure their resources. Customers must realize and assume their responsibility as failure to do so would place their data and resources at further risk. In some cases CC have inappropriately assumed that the CP was responsible for, and was conducting, all activities required to ensure security of their data. This assumption by the customer, and/or a lack of clear articulation by the CP, placed unnecessary risk on the customers data. It is imperative that CCs identify their responsibilities and comply with them. CPs, by their very nature, are tasked with providing a multi-tenant environment, whether this is via virtualization on a server or the common network shared by the customers. The co-location of many customers inevitably causes conflict for the CP as customers communication security requirements are likely to be divergent from each other.
Take, for example, the case of two customers on a shared traditional network infrastructure. If one customer wishes the network firewall to block all traffic except for SSH, but another customer is running a web server farm and requires passage of HTTP and HTTPS, who wins? This same type of issue is raised by customers who have competing and conflicting compliance requirements. This type of challenge only worsens as the number of tenants and the disparity of their requirements increase. Therefore, CPs must be in a position to deal with these challenges by way of technology, policy and transparency (where appropriate).
26
R.5 Social engineering attacks Risk number and name Short description R.5 Social engineering attacks
Social engineering is understood to mean the art of manipulating people into performing actions or divulging confidential information. While it is similar to a confidence trick or simple fraud, it is typically trickery or deception for the purpose of information gathering, fraud, or computer system access; in most cases the attacker never comes face-to-face with the victims. Probability: Medium Impact: High Risk: Medium
Due to the involvement of different organisations, the probablility for social engineering attacks is considered higher. This is mostly due to the greater attack surface created by the interaction between two different entities. If a social engineering attack occurs, the impact will be the same in both classic IT and Cloud settings.
As Cloud services are on-demand services, there is the possibility that the CP wont be able to meet an increased demand in a certain shared resource, or to maintain a given service level. Impact: Medium Risk: Medium
Ris rating for failing to meet Probability: Medium an increase in demand Risk rating for failing to Probability: Low maintain the current service level Probability in Comparison to classic IT
Impact: High
Risk: Medium
Resource exhaustion can also occur in classic IT settings. As the CP is supposed to have an effective capacity management in place that will make it unlikely that a resource exhaustion event occurs, the probability is considered lower in a Cloud setting. If a resource exhaustion event occurs, the impact will be the same in both classic IT and Cloud settings.
Cloud services are on-demand services. Therefore there is a level of calculated risk in allocating all the resources of a cloud service, because resources are allocated according to statistical projections. Inaccurate modelling of resources usage - common resources allocation algorithms are vulnerable to distortions of fairness - or inadequate resource provisioning and inadequate investments in infrastructure can lead, from the CP perspective, to:
Service unavailability: failure in certain highly specific application scenarios which use a particular resource very intensively (ie, CPU/Memory intensive number crunching or simulation (eg. forecasting stock prices;
28
Access control compromised: in some cases it may be possible to force a system to fail open in the event of resource exhaustion. Economic and reputational losses: due to failure to meet customer demand. The opposite consequences of inaccurate estimation of resource needs could lead to: Infrastructure oversize: excessive provisioning leading to economic losses and loss of profitability.
From the CC perspective, a poor provider selection and lack of supplier redundancy could lead to:
Service unavailability: failure in the delivery (or degrading performance) of services both in real time and not in real time; Access control system compromised: put the confidentiality and Integrity of data at risk; Economic and reputational losses: due to failure to meet customer demand, violation of SLA, cascading service failure, etc. There is a level of calculated risk in allocating all the resources of a cloud service, because resources are allocated according to statistical projections.
Note: this risk could be also a consequence of a DDoS attack and of misbehaving applications due to poor application compartmentalization in some CPs systems.
R.7 Isolation failure Risk number and name Short description R.7 Isolation failure
In shared environments, errors or attacks can lead to situations where one tenant has access to another tenants resources or data. In the case of attacks, an attacker gets access to the resources or data of a specific customer, or even of all customers of the Cloud service. Probability: High Impact: High Risk: High
Isolation failures can also occur in classic IT settings, but due to the shared nature of Cloud services and the exposed interfaces, the probability is considered higher for Cloud services. The impact of an isolation failure is considered the same for both classic IT and Cloud settings.
Multi-tenancy and shared resources are two of the defining characteristics of cloud computing environments. Computing capacity, storage, and network are shared between multiple users. This class of risks includes the failure of mechanisms separating storage, memory, routing, and even reputation between different tenants of the shared infrastructure (e.g., so-called guest-hopping attacks, SQL injection attacks exposing multiple customers data stored in the same table, and side channel attacks). Note that the likelihood (probability) of this incident scenario depends on the cloud model considered; it is likely to be low for private clouds and higher (medium) in the case of public clouds. The impact can be a loss of valuable or sensitive data, reputation damage and service interruption for cloud providers and their clients. Resource sharing also means that malicious activities carried out by one tenant may affect the reputation of another tenant. For example, spamming, port scanning or the serving of malicious content from cloud infrastructure can lead to:
30
a range of IP addresses being blocked, including the attacker and other innocent tenants of an infrastructure; confiscation of resources due to neighbour activities (neighbour subpoenaed).
The impact can be deterioration in service delivery and data loss, as well as problems for the organizations reputation.
R.8 Cloud provider malicious insider - abuse of high privilege roles Risk number and name R.8 Cloud provider malicious insider - abuse of high privilege roles
Short description
Malicious insiders at the CP can cause various kinds of damage to a CCs assets. Probability: Medium Impact: Very high Risk: High
The malicious insider threat also exists in classic IT settings, but the likelihood of an incident has to be considered higher because admin roles at CPs provide more temptation and more opportunities. The impact caused by mailicious insider is considered the same for both classic IT and Cloud settings.
The malicious activities of an insider could potentially have an impact on: the confidentiality, integrity and availability of all kind of data, IP, all kind of services and therefore indirectly on the organizations reputation, customer trust and the experiences of employees. This can be considered especially important in the case of cloud computing due to the fact that cloud architectures necessitate certain roles which are extremely high-risk. Examples of such roles include CP system administrators and auditors and managed security service providers dealing with intrusion detection reports and incident response. As cloud use increases, employees of cloud providers increasingly become targets for criminal gangs (as has been witnessed in the financial services industry with call centre workers).
32
R.9 Management interface compromise (manipulation, availability of infrastructure) Risk number and name R.9 Management interface compromise availability of infrastructure) (manipulation,
Short description
The customer management interfaces of public cloud providers are Internet accessible and mediate access to larger sets of resources (than traditional hosting providers) and therefore pose an increased risk especially when combined with remote access and web browser vulnerabilities. Probability: Medium Impact: Very high Risk: High
If management interfaces are compromised in classic IT settings, the impact is the same as for Cloud settings.
The customer management interfaces of public cloud providers are Internet accessible and mediate access to larger sets of resources (than traditional hosting providers) and therefore pose an increased The customer management interfaces of public risk especially when combined with remote access cloud providers are Internet accessible and and web browser vulnerabilities. This includes mediate access to larger sets of resources customer interfaces controlling a number of virtual (than traditional hosting providers) and machines and, most importantly, CP interfaces therefore pose an increased risk especially controlling the operation of the overall cloud system. when combined with remote access and web Of course, this risk may be mitigated by more browser vulnerabilities. investment in security by providers.
R.10
Whenever data is transferred between different computers or sites, there is the possibility that the transfer can be intercepted, This is especially relevant in shared environments and when data is transferred between sites (e.g. between CC and CP). Probability: Medium Impact: High Risk: Medium
The risk of data being intercepted is considerably higher in Cloud environments, because these always involve both a shared environment and transfers between sites. If data is intercepted, the impact is the same in both Cloud and classic IT settings.
Cloud computing, being a distributed architecture, implies more data in transit than traditional infrastructures. For example, data must be transferred in order to synchronise multiple distributed machine images, images distributed across multiple physical machines, between cloud infrastructure and remote web clients, etc. Furthermore, most use of data-centre hosted computing is implemented using a secure VPN-like connection environment, a practice not always followed in the cloud context. Sniffing, spoofing, man-inthe-middle attacks, side channel and replay attacks have to be considered as possible threat sources. Moreover, in some cases the CP does not offer a confidentiality or non-disclosure clause or these clauses are not sufficient to guarantee respect for the protection of the customers secret information and know-how that will circulate in the cloud.
34
R.11
Deleting data from Cloud storage does not in fact mean that the data is removed from the storage or eventual backup media. If disk storage is not encrypted, the data could be accessed at at later time by another customer of a Cloud provider Probability: Medium Impact: Very high Risk: High
Scaling up or down resources and moving data around is a characteristic of the Cloud deployment model. In this context, the probability of a data exposure due to ineffective deletion is considered higher than in a classic IT setting. The impact of a data exposure is considered the same for both Cloud and classic IT settings.
Whenever a provider is changed, resources are scaled down, physical hardware is reallocated, etc, data may be available beyond the lifetime specified in the security policy. It may be impossible to carry out the procedures specified by the security policy, since full data deletion is only possible by destroying a disk which also stores data from other clients. When a request to delete a cloud resource is made, this may not result in true wiping of the data There are several different scenarios in which a (as with most operating systems). Where true data cloud customer's resources may be used by wiping is required, special procedures must be other parties in a malicious way that has an followed and this may not be supported by the economic impact. standard API (or at all). If effective encryption is used then the level of risk may be considered to be lower.
R.12
Distributed Denial of Service attacks aim at overloading a resource (network or service interface) by flooding it with requests from many sources distributed across a wide geographical or topological area, so that the legitimate users are unable to use the resource as intended. Probability: Medium Impact: High Risk: Medium
The probability of a DDoS attack affecting a particular customer is considered lower than the probability of a standalone service being affected by a targeted DDoS attack. The reason is that a Cloud provider is expected to have better resources and mitigation techniques available than an individual company could have. The impact of a DDoS attack is considered to be the same in both classic IT and Cloud settings.
36
R.13
As a consequence of attacks, poor budget planning, or misconfigurations, the cost of a Cloud service can strain the financial resources of a CC to an extent that the service is no longer affordable. Probability: Low Impact: High Risk: Medium
Although this risk is to some extent also present in classic IT settings, it becomes much more relevant in Cloud settings, because resources are allocated and billed dynamically. If an EDoS event occurs, the impact will be the same both in a classic IT and in Cloud setting.
There are several different scenarios in which a CC's resources may be used by other parties in a malicious way that has an economic impact: Identity theft: an attacker uses an account and uses the customer's resources for his own gain or in order to damage the customer economically. The CC has not set effective limits on the use of paid resources and experiences unexpected loads on these resources through no malicious actions. An attacker uses a public channel to use up the CC's metered resources - for example, where the customer pays per HTTP request; a DDoS attack can have this effect.
EDoS destroys economic resources; the worst case scenario would be the bankruptcy of the customer or a serious economic impact.
R.14
The service engine is a fundamental part of a Cloud service. A compromise of the service engine will give an attacker access to the data of all customers, resulting in a potential complete loss of data or denial of service. Probability: Low Impact: Very high Risk: High
n.a.
Each cloud architecture relies on a highly specialized platform, the service engine that sits above the physical hardware resources and manages customer resources at different levels of abstraction. For example, in IaaS clouds this software component can be the hypervisor. The service engine is developed and supported by cloud platform vendors and the open source community in some cases. It can be further customized by the CPs. Like any other software layer, the service engine code can have vulnerabilities and is prone to attacks or unexpected failure. An attacker can compromise Cloud providers must set out a clear the service engine by hacking it from inside a virtual segregation of responsibilities that articulates machine (IaaS clouds), the runtime environment the minimum actions customers must (PaaS clouds), the application pool (SaaS clouds), or undertake. through its APIs. Hacking the service engine may be useful to escape the isolation between different customer environments (jailbreak) and gain access to the data contained inside them, to monitor and modify the information inside them in a transparent way (without direct interaction with the application inside the customer environment), or to reduce the resources assigned to them, causing a denial of service.
38
R.15
The loss or compromise of cryptographic keys used for encryption, authentication or digital signatures can lead to data loss, denial of services, or financial damages Probability: Low Impact: High Risk: Medium
This risk is also present in classic IT settings, but cryptographic techniques are more likely to be used in Cloud settings If a loss of cryptographic keys occurs, the impact is considered the same both in classic IT and in Cloud settings.
This includes disclosure of secret keys (SSL, file encryption, customer private keys, etc) or passwords to malicious parties, the loss or corruption of those keys, or their unauthorised use for authentication and non-repudiation (digital signature).
R.16
Non Cloud-Specific Network-Related Technical Failures or Attacks R.16 Non Cloud-Specific Network-Related Technical Failures or Attacks
Short description
Cloud services can be affected by a number of network-related technical failures that can also occur on classic IT settings. Examples include the loss of Internet connectivity due to failures at the CCs site or the CCs Internet service provider, temporarily reduced network bandwidth on the path between CC and CP, disruptions in the global Internet routing infrastructure leading to the loss of the network path between CC and CP, and failures of the CPs Internet connectivity. Probability: Medium Impact: Medium Risk: Medium
Network-related technical failures or attacks can occur both in Cloud and in classic IT settings.
The impact of network-related failures is considered higher in a Cloud setting because of the rather immediate effect on the performance or even the availability of the Cloud service.
40
R.17
Short description
The backups a CP makes of its customers data can get lost, damaged, or the physical media on which the backup is stored can get stolen. Probability: Low Impact: High Risk: Medium
The probability of a loss of backups is considered lower in a Cloud setting because it is assumed that a CP will have better protection mechanisms in place than an average CC. If backups are lost or stolen, the impact is the same for both Cloud and classic IT settings.
R.18
Natural disasters like flooding, earthquakes, tsunamis can affect the infrastructure of a CP. This way, a CC might be affected by natural disasters occurring far away from its own location. Probability: Very low Impact: High Risk: Medium
The probability of a natural disaster affecting a Cloud service is considered lower for a Cloud service than for classic IT because it is assumed that a CP has more redundancy built into its infrastructure than an average CC. If a natural disaster does affect a Cloud service, the impact is considered higher than in a classic IT setting, because the recovery efforts of a CP will not be focussed on restoring service to a particular customer, but rather to restore the service in general. Therefore, the time to recovery may be higher than in a classic IT setting.
Generally speaking, the risk from natural disasters is lower compared to traditional infrastructures because CPs often offer multiple redundant sites and network paths by default. This is illustrated e.g. by the 2011 Japan earthquake where classic IT often failed, but no large datacentre went down.
42
Law enforcement authorities may ask operators of IT infrastructures to provide information pertaining to criminal cases, or information may have to be provided during civil lawsuits. In some cases, storage media or other hardware might be seized as evidence. Probability: High Impact: Medium Risk: High
Due to the shared usage of hardware resources in a Cloud setting, a CC can also be affected by subpoenas or civil law suits directed at the CP or third parties, not just those directed at the CC himself. The probability is therefore higher in a Cloud setting than in a classic IT setting. The impact of subpoena or e-discovery is considered to be the same in both classic IT and Cloud settings.
In the event of the confiscation of physical hardware as a result of subpoena by law-enforcement agencies or civil suits, the centralisation of storage as well as shared tenancy of physical hardware means many more clients are at risk of the disclosure of their data to unwanted parties. At the same time, it may become impossible for the agency of a single nation to confiscate a cloud given pending advances around long distance hypervisor migration.
R.20
When data is stored or processed in a data centre located in a country other than the CCs, there are numerous ways in which the change in jurisdiction could affect the security of the information. Examples include: Data might be seized or the operations of a service disrupted due to reasons that dont exist in the CCs country. In some cases, national security interests of the hosting country might be cited as a reason for seizing data. Additionally, a CP might be subject to law enforcement or national security actions from the country its business headquarters is based in, not just those from the countries where its data centres are located. Impact: High Risk: High
Probability: High
This risk is considered to be considerably higher in a Cloud setting, because a potentially large number of different countries jurisdictions might be involved. The impact of an event related to this risk is considered the same in both classic IT and Cloud settings.
Customer data may be held in multiple jurisdictions, some of which may be high risk. If data centres are located in high-risk countries, e.g., those. lacking the rule of law and having an unpredictable legal framework and enforcement, autocratic police states, states that do not respect international agreements, etc, sites could be raided by local authorities and data or systems subject to enforced disclosure or seizure. Note that we are not implying here that all subpoena law-enforcement measures are unacceptable, merely that some may be so and that some legitimate seizures of hardware (which appear to be rare) may affect more customers than the targets of a law-enforcement action depending on how the data is stored.
44
R.21
Processing data in another country may incur difficulties regarding data protection legislation, or might even be considered unlawful by the responsible Data Protection authority. Probability: High Impact: High Risk: High
As a CP might move data and processing applications between data centres located in different countries even without notifying the CC, the probability has to be considered higher in a Cloud setting than in a classic IT setting, where the data processor has full control over the location. If difficulties regarding data protection regulations are encountered, the impact will be the same in both classic IT and Cloud setting.
Cloud computing poses several data protection risks for CCs and CPs. It can be difficult for the CC (in its role of data controller) to effectively check the data processing that the CP carries out, and thus be sure that the data is handled in a lawful way. It has to be clear that the CC will be the main person responsible for the processing of personal data, even when such processing is carried out by the CP in its role of external processor. Failure to comply with data protection law may lead to administrative, civil and also criminal sanctions, which vary from country to country, for the data controller. This problem is exacerbated in the case of multiple transfers of data e.g., between federated clouds. On the other hand, some CPs do provide information on the data processing that they carry out. Some also offer certification summaries of their data processing and data security activities and the data controls they have in place, e.g., SAS70 certification providers. There may be data security breaches which are not notified to the controller by the CP. The CC may lose control of the data processed by the CP. This issue is increased in the case of multiple transfers of data (e.g., between federated CPs). The CP may receive data that have not been lawfully collected by its customer (the controller).
46
R.22
Violating a software suppliers licensing agreements can result in significant financial penalties or disruptions of service. Probability: Medium Impact: Medium Risk: Medium
As many software licensing agreements are not yet cloud aware, the probability of incidents related to licensing has to be considered higher in a Cloud setting. The impact of a licensing issue is considered the same in both settings.
Licensing conditions, such as per-seat agreements, and online licensing checks may become unworkable in a cloud environment. For example, if software is charged on a per instance basis every time a new machine is instantiated then the cloud customers licensing costs may increase exponentially even though they are using the same number of machine instances for the same duration.
48
R.23
Both in the Cloud and when using certain software and service environments within the own infrastructure, there is the possibility for creating original work (new applications, software etc.) tied to this specific environment. As with all intellectual property, if not protected by the appropriate contractual clauses, this original work may be at risk. Probability: Low Impact: Medium Risk: Medium
The impact of intellectual property issues is considered the same in both classic IT and Cloud settings.
Cloud computing is a new way of delivering computing resources, not a new technology. Computing services ranging from data storage and processing to software, such as email handling, are now available instantly, commitment-free and on-demand. This new economic model for computing has found fertile ground and is seeing massive global investment. According to IDCs analysis, the worldwide forecast for cloud services in 2009 will be in the order of $17.4bn. The estimation for 2013 amounts to $44.2bn, with the European market ranging from 971m in 2008 to 6,005m in 2013. For cloud computing to reach the full potential promised by the technology, it must offer solid information security, and therefore, proper consideration and management of risks. ENISA has played an important role in giving stakeholders an overview of the information security risks when going cloud . Our 2009 cloud security risk assessment is widely referred to, across EU member states, and outside the EU. This year we decided to update the assessment of risk and benefits, to better reflect the current situation. After the first round of reviews, we conclude that the main risks, when adopting cloud computing, have not changed. We have restructured the risks with the goal of making the descriptions more uniform. The updated risk assessment will now be subjected to a second review round by a group of experts set up by ENISA. At the same time, there will be a more in-depth review of the legal and data protection aspects of Cloud computing, which were excluded from the first round of review. ENISA will continue to monitor the developments related to risks of Cloud computing and update the Risk Assessment as necessary.
50
P.O.
Heraklion,