Cloud Security Threats and Countermeasures at a Glance
Cloud security has been a hot topic with the introduction of the Microsoft offering of the Windows Azure platform. One of the quickest ways to get your head around security is to cut to the chase and look at the threats, attacks, vulnerabilities and countermeasures. This post is a look at threats and countermeasures from a technical perspective.
The thing to keep in mind with security is that it’s a matter of people, process, and technology. However, focusing on a specific slice, in this case the technical slice, can help you get results. The thing to keep in mind about security from a technical side is that you also need to think holistically in terms of the application, network, and host, as well as how you plug it into your product or development life cycle. For information on plugging it into your life cycle, see the Security Development Lifecycle.
While many of the same security issues that apply to running applications on-premise also apply to the cloud, the context of running in the cloud does change some key things. For example, it might mean taking a deeper look at claims for identity management and access control. It might mean rethinking how you think about your storage. It can mean thinking more about how you access and manage virtualized computing resources. It can mean thinking about how you make calls to services or how you protect calls to your own services.
Here is a fast path through looking at security threats, attacks, vulnerabilities, and countermeasures for the cloud …
- Learn a security frame that applies to the cloud
- Learn top threats/attacks, vulnerabilities and countermeasures for each area within the security frame
- Understand differences between threats, attacks, vulnerabilities and countermeasures
It is important to think like an attacker when designing and implementing an application. Putting yourself in the attacker’s mindset will make you more effective at designing mitigations for vulnerabilities and coding defensively. Below is the cloud security frame. We use the cloud security frame to present threats, attacks, vulnerabilities and countermeasures to make them more actionable and meaningful.
You can also use the cloud security frame to effectively organize principles, practices, patterns, and anti-patterns in a more useful way.
Threats, Attacks, Vulnerabilities, and Countermeasures
These terms are defined as follows:
- Asset. A resource of value such as the data in a database, data on the file system, or a system resource.
- Threat. A potential occurrence – malicious or otherwise – that can harm an asset.
- Vulnerability. A weakness that makes a threat possible.
- Attack. An action taken to exploit vulnerability and realize a threat.
- Countermeasure. A safeguard that addresses a threat and mitigates risk.
Cloud Security Frame
The following key security concepts provide a frame for thinking about security when designing applications to run on the cloud, such as Windows Azure. Understanding these concepts helps you put key security considerations such as authentication, authorization, auditing, confidentiality, integrity, and availability into action.
|Auditing and Logging||Cloud auditing and logging refers to how security-related events are recorded, monitored, audited, exposed, compiled & partitioned across multiple cloud instances. Examples include: Who did what and when and on which VM instance?|
|Authentication||Authentication is the process of proving identity, typically through credentials, such as a user name and password. In the cloud this also encompasses authentication against varying identity stores.|
|Authorization||Authorization is how your application provides access controls for roles, resources and operations. Authorization strategies might involve standard mechanisms, utilize claims and potentially support a federated model.|
|Communication||Communication encompasses how data is transmitted over the wire. Transport security, message encryption, and point to point communication are covered here.|
|Configuration Management||Configuration management refers to how your application handles configuration and administration of your applications from a security perspective. Examples include: Who does your application run as? Which databases does it connect to? How is your application administered? How are these settings secured?|
|Cryptography||Cryptography refers to how your application enforces confidentiality and integrity. Examples include: How are you keeping secrets (confidentiality)? How are you tamper-proofing your data or libraries (integrity)? How are you providing seeds for random values that must be cryptographically strong? Certificates and cert management are in this domain as well.|
|Input and Data Validation||Validation refers to how your application filters, scrubs, or rejects input before additional processing, or how it sanitizes output. It's about constraining input through entry points and encoding output through exit points. Message validation refers to how you verify the message payload against schema, as well as message size, content and character sets. Examples include: How do you know that the input your application receives is valid and safe? Do you trust data from sources such as databases and file shares?|
|Exception Management||Exception management refers to how you handle applications errors and exceptions. Examples include: When your application fails, what does your application do? Does it support graceful failover to other application instances in the cloud? How much information do you reveal? Do you return friendly error information to end users? Do you pass valuable exception information back to the caller?|
|Sensitive Data||Sensitive data refers to how your application handles any data that must be protected either in memory, over the network, or in persistent stores. Examples include: How does your application handle sensitive data? How is sensitive data shared between application instances?|
|Session Management||A session refers to a series of related interactions between a user and your application. Examples include: How does your application handle and protect user sessions?|
Threats and Attacks
|Auditing and Logging||
|Input and Data Validation||
|Auditing and Logging||
|Input and Data Validation||
|Auditing and Logging||
Threats and Attacks Explained
- Brute force attacks. Attacks that use the raw computer processing power to try different permutations of any variable that could expose a security hole. For example, if an attacker knew that access required an 8-character username and a 10-character password, the attacker could iterate through every possible (256 multiplied by itself 18 times) combination in order to attempt to gain access to a system. No intelligence is used to filter or shape for likely combinations.
- Buffer overflows. The maximum size of a given variable (string or otherwise) is exceeded, forcing unintended program processing. In this case, the attacker uses this behavior to cause insertion and execution of code in such a way that the attacker gains control of the program in which the buffer overflow occurs. Depending on the program’s privileges, the seriousness of the security breach will vary.
- Canonicalization attacks. There are multiple ways to access the same object and an attacker uses a method to bypass any security measures instituted on the primary intended methods of access. Often, the unintended methods of access can be less secure deprecated methods kept for backward compatibility.
- Cookie manipulation. Through various methods, an attacker will alter the cookies stored in the browser. Attackers will then use the cookie to fraudulently authenticate themselves to a service or Web site.
- Cookie replay attacks. Reusing a previously valid cookie to deceive the server into believing that a previously authenticated session is still in progress and valid.
- Credential theft. Stealing the verification part of an authentication pair (identity + credentials = authentication). Passwords are a common credential.
- Cross- Site Request Forgery (CSRF). Interacting with a web site on behalf of another user to perform malicious actions. A site that assumes all requests it receives are intentional is vulnerable to a forged request.
- Cross-site scripting (XSS) . An attacker is able to inject executable code (script) into a stream of data that will be rendered in a browser. The code will be executed in the context of the user’s current session and will gain privileges to the site and information that it would not otherwise have.
- Connection pooling. The practice of creating and then reusing a connection resource as a performance optimization. In a security context, this can result in either the client or server using a connection previously used by a highly privileged user being used for a lower-privileged user or purpose. This can potentially expose vulnerability if the connection is not reauthorized when used by a new identity.
- Data tampering. An attacker violates the integrity of data by modifying it in local memory, in a data-store, or on the network. Modification of this data could provide the attacker with access to a service through a number of the different methods listed in this document.
- Denial of service. Denial of service (DoS) is the process of making a system or application unavailable. For example, a DoS attack might be accomplished by bombarding a server with requests to consume all available system resources, or by passing the server malformed input data that can crash an application process.
- Dictionary attack. Use of a list of likely access methods (usernames, passwords, coding methods) to try and gain access to a system. This approach is more focused and intelligent than the “brute force” attack method, so as to increase the likelihood of success in a shorter amount of time.
- Disclosure of sensitive/confidential data. Sensitive data is exposed in some unintended way to users who do not have the proper privileges to see it. This can often be done through parameterized error messages, where an attacker will force an error and the program will pass sensitive information up through the layers of the program without filtering it. This can be personally identifiable information (i.e., personal data) or system data.
- Elevation of privilege. A user with limited privileges assumes the identity of a privileged user to gain privileged access to an application. For example, an attacker with limited privileges might elevate his or her privilege level to compromise and take control of a highly privileged and trusted process or account. More information about this attack in the context of Windows Azure can be found in the Security Best Practices for Developing Windows Azure Applications at http://www.microsoft.com/downloads/details.aspx?displaylang=en&FamilyID=0ff0c25f-dc54-4f56-aae7-481e67504df6
- Encryption. The process of taking sensitive data and changing it in such a way that it is unrecognizable to anyone but those who know how to decode it. Different encryption methods have different strengths based on how easy it is for an attacker to obtain the original information through whatever methods are available.
- Information disclosure. Unwanted exposure of private data. For example, a user views the contents of a table or file that he or she is not authorized to open, or monitors data passed in plaintext over a network. Some examples of information disclosure vulnerabilities include the use of hidden form fields, comments embedded in Web pages that contain database connection strings and connection details, and weak exception handling that can lead to internal system-level details being revealed to the client. Any of this information can be very useful to the attacker.
- Luring attacks. An attacker lures a higher-privileged user into taking an action on his or her behalf. This is not an authorization failure but rather a failure of the system to properly inform the user.
- Man-in-the-middle attacks. A person intercepts both the client and server communications and then acts as an intermediary between the two without each ever knowing. This gives the “middle man” the ability to read and potentially modify messages from either party in order to implement another type of attack listed here.
- Network eavesdropping. Listening to network packets and reassembling the messages being sent back and forth between one or more parties on the network. While not an attack itself, network eavesdropping can easily intercept information for use in specific attacks listed in this document.
- Open Redirects. Attacker provides a URL to a malicious site when allowed to input a URL used in a redirect. This allows the attacker to direct users to sites that perform phishing attacks or other malicious actions.
- Password cracking. If the attacker cannot establish an anonymous connection with the server, he or she will try to establish an authenticated connection. For this, the attacker must know a valid username and password combination. If you use default account names, you are giving the attacker a head start. Then the attacker only has to crack the account’s password. The use of blank or weak passwords makes the attacker’s job even easier.
- Repudiation. The ability of users (legitimate or otherwise) to deny that they performed specific actions or transactions. Without adequate auditing, repudiation attacks are difficult to prove.
- Session hijacking. Also known as man-in-the-middle attacks, session hijacking deceives a server or a client into accepting the upstream host as the actual legitimate host. Instead, the upstream host is an attacker’s host that is manipulating the network so the attacker’s host appears to be the desired destination.
- Session replay. An attacker steals messages off of the network and replays them in order to steal a user’s session.
- Session fixation. An attacker sets (fixates) another person’s session identifier artificially. The attacker must know that a particular Web service accepts any session ID that is set externally; for example, the attacker sets up a URL such as http://unsecurewebservice.com/?sessionID=1234567. The attacker then sends this URL to a valid user, who clicks on it. At this point, a valid session with the ID 1234567 is created on the server. Because the attacker determines this ID, he or she can now hijack the session, which has been authenticated using the valid user’s credentials.
- Spoofing. An attempt to gain access to a system by using a false identity. This can be accomplished by using stolen user credentials or a false IP address. After the attacker successfully gains access as a legitimate user or host, elevation of privileges or abuse using authorization can begin.
- SQL injection. Failure to validate input in cases where the input is used to construct a SQL statement or will modify the construction of a SQL statement in some way. If the attacker can influence the creation of a SQL statement, he or she can gain access to the database with privileges otherwise unavailable and use this in order to steal or modify information or destroy data.
- Throttling. The process of limiting resource usage to keep a particular process from bogging down and/or crashing a system. Relevant as a countermeasure in DoS attacks, where an attacker attempts to crash the system by overloading it with input.
- Assume all input is malicious. Assuming all input is malicious means designing your application to validate all input. User input should never be accepted without being filtered and/or sanitized.
- Audit and log activity through all of the application tiers. Log business critical and security sensitive events. This will help you track security issues down and make sense of security problems. Skilled attackers attempt to cover their tracks, so you’ll want to protect your logs.
- Avoid storing secrets. Design around storing secrets. If necessary, sometimes they can be avoided by storing them after using a one-way hash algorithm.
- Avoid storing sensitive data in the Web space. Anything exposed to the public Internet is considered “web space.” Sensitive data stored in a location that might be compromised by any member of the public places it at much higher risk.
- Back up and regularly analyze log files. Some attacks can occur over time. Regular analysis of logs will allow you to recognize with sufficient time to address them. Performing regular backups lowers the risk of an attacker covering his tracks by deleting logging of his activities.
- Be able to disable accounts. The ability to reactively defend an attack by shutting out a user should be supported through the ability to disable an account.
- Be careful with canonicalization issues. Predictable naming of file resources is convenient for programming, but is also very convenient for malicious parties to attack. Application logic should not be exposed to users in this manner. Instead, you use file names derived from the original names or fed through a one-way hashing algorithm.
- Catch exceptions. Unhandled exceptions are at risk of passing too much information to the client. Handle exceptions when possible.
- Centralize your input and data validation. Input and data validation should be performed using a common set of code such as a validation library.
- Consider a centralized exception management framework. Exception handling frameworks are available publically and provide an established and tested means for handling exceptions.
- Consider authorization granularity. Every object needs to have an authorization control that authorizes access based on the identity of the authenticated party requesting access. Fine grained authorization will control access to each resource, while coarse grained authorization will control access to groups of resources or functional areas of the application.
- Consider identity flow. Auditing should be traceable back to the authenticated party. Take note of identity transitions imposed by design decisions like impersonation.
- Constrain input. Limit user input to expected ranges and formats.
- Constrain, reject, and sanitize your input. Constrain, reject and sanitize should be primary techniques in handling input data.
- Cycle your keys periodically. Expiring encryption keys lowers the risk of stolen keys.
- Disable anonymous access and authenticate every principle. When possible, require all interactions to occur as an authenticated party as opposed to an anonymous one. This will help facilitate more effective auditing.
- Do not develop your own cryptography. Custom cryptography is not difficult for experts to crack. Established cryptography is preferred because it is known to be safe.
- Do not leak information to the client. Exception data can potentially contain sensitive data or information exposing program logic. Provide clients only with the error data they need for the UI.
- Do not log private data such as passwords. Log files are an attack vector for malicious parties. Limit the risk of their being compromised by not logging sensitive data in the log.
- Do not pass sensitive data using the HTTP-GET protocol. Data passed using HTTP GET is appended to the querystring. When users share links by copying and pasting them from the browser address bar, sensitive data may also be inadvertently passed. Pass sensitive data in the body of a POST to avoid this.
- Do not rely on client-side validation. Any code delivered to a client is at risk of being compromised. Because of this, it should always be assumed that input validation on the client might have been bypassed.
- Do not send passwords over the wire in plaintext. Authentication information communicated over the wire should always be encrypted. This may mean encrypting the values, or encrypting the entire channel with SSL.
- Do not store credentials in plaintext. Credentials are sometimes stored in application configuration files, repositories, or sent over email. Always encrypt credentials before storing them.
- Do not store database connections, passwords, or keys in plaintext. Configuration secrets should always be stored in encrypted form, external to the code.
- Do not store passwords in user stores. In the event that the user store is compromised, an attack should never be able to access passwords. A derivative of a password should be stored instead. A common approach to this is to encrypt a version of the password using a one-way hash with a SALT. Upon authentication, the encrypted password can be re-generated with the SALT and the result can be compared to the original encrypted password.
- Do not store secrets in code. Secrets such as configuration settings are convenient to store in code, but are more likely to be stolen. Instead, store them in a secure location such as a secret store.
- Do not store sensitive data in persistent cookies. Persistent cookies are stored client-side and provide attackers with ample opportunity to steal sensitive data, be it through encryption cracking or any other means.
- Do not trust fields that the client can manipulate (query strings, form fields, cookies, or HTTP headers) . All information sent from a client should always be assumed to be malicious. All information from a client should always be validated and sanitized before it is used.
- Do not trust HTTP header information. Http header manipulation is a threat that can be mitigated by building application logic that assumes HTTP headers are compromised and validates the HTTP headers before using them.
- Encrypt communication channels to protect authentication tokens. Authentication tokens are often the target of eavesdropping, theft or replay type attacks. To reduce the risk in these types of attacks, it is useful to encrypt the channel the tokens are communicated over. Typically this means protecting a login page with SSL encryption.
- Encrypt sensitive cookie state. Sensitive data contained within cookies should always be encrypted.
- Encrypt the contents of the authentication cookies. In the case where cookies are compromised, they should not contain clear-text session data. Encrypt sensitive data within the session cookie.
- Encrypt the data or secure the communication channel. Sensitive data should only be passed in encrypted form. This can be accomplished by encrypting the individual items that are sent over the wire, or encrypting the entire channel as with SSL.
- Enforce separation of privileges. Avoid building generic roles with privileges to perform a wide range of actions. Roles should be designed for specific tasks and provided the minimum privileges required for those tasks.
- Enforce unique transactions. Identify each transaction from a client uniquely to help prevent replay and forgery attacks.
- Identify malicious behavior. Monitoring site interactions that fall outside of normal usage patterns, you can quickly identify malicious behavior. This is closely related to “Know what good traffic looks like.
- Keep unencrypted data close to the algorithm. Use decrypted data as soon as it is decrypted, and then dispose of it promptly. Unencrypted data should not be held in memory in code.
- Know what good traffic looks like. Active auditing and logging of a site will allow you know recognize what regular traffic and usage patterns are. This is a required step in order to be able to identify malicious behavior.
- Limit session lifetime. Longer session lifetimes provide greater opportunity for Cross-Site Scripting or Cross-Site Request Forgery attacks to add activity onto an old session.
- Log detailed error messages. Highly detailed error message logging can provide clues to attempted attacks.
- Log key events. Profile your application and note key or sensitive operations and/or events, and log these events during application operation.
- Maintain separate administration privileges. Consider granularity of authorization in the administrative interfaces as well. Avoid combining administrator roles with distinctly different roles such as development, test or deployment.
- Make sure that users do not bypass your checks. Bypassing checks can be accomplished by canonicalization attacks, or bypassing client-side validation. Application design should avoid exposing application logic, and segregating application logic into flow that can be interrupted. For example, an ASPX page that performs only validations and then redirects. Instead, validation routines should be tightly bound to the data they are validating.
- Pass Forms authentication cookies only over HTTPS connections. Cookies are at risk of theft and replay type attacks. Encrypting them with SSL helps reduce the risk of these types of attacks.
- Protect authentication cookies. Cookies can be manipulated with Cross-Site Scripting attacks, encrypt sensitive data in cookies, and use browser features such as the HttpOnly cookie attribute.
- Provide strong access controls on sensitive data stores. Access to secret stores should but authorized. Protect the secret store as you would other secure resources by requiring authentication and authorization as appropriate.
- Reject known bad input. Rejecting known bad input involves screening input for values that are known to be problematic or malicious. NOTE: Rejecting should never be the primary means of screening bad input, it should always be used in conjunction with input sanitization.
- Require strong passwords. Enforce password complexity requirement by requiring long passwords with a combination of upper case, lower case, numeric and special (for example punctuation) characters. This helps mitigate the threat posed by dictionary attacks. If possible, also enforce automatic password expiry.
- Restrict user access to system-level resources. Users should not be touching system resources directly. This should be accomplished through an intermediary such as the application. System resources should be restricted to application access.
- Retrieve sensitive data on demand. Sensitive data stored in application memory provides attackers another location they can attempt to access the data. Often this data is used in unencrypted form also. To minimize risk of sensitive data theft, sensitive data should be used immediately and then cleared from memory.
- Sanitize input. Sanitizing input is the opposite of rejecting bad input. Sanitizing input is the process of filtering input data to only accept values that are known to be safe. Alternatively, input can be rendered innocuous by converting it to safe output through output encoding methods.
- Secure access to log files. Log files should only be accessible to administrators, auditors, or administrative interfaces. An attacker with access to the logs might be able to glean sensitive data or program logic from logs.
- Secure the communication channel for remote administration. Eavesdropping and replay attacks can target administration interfaces as well. If using a web based administration interface, use SSL.
- Secure your configuration store. The configuration store should require authenticated access and should store sensitive settings or information in an encrypted format.
- Secure your encryption keys. Encryption keys should be treated as secrets or sensitive data. They should be secured in a secret store or key repository.
- Separate public and restricted areas. Applications that contain public front-ends as well as content that requires authentication to access should be partitioned in the same manner. Public facing pages should be hosted in a separate file structure, directory or domain from private content.
- Store keys in a restricted location. Protect keys with authorization policies.
- Support password expiration periods. User passwords and account credentials are commonly compromised. Expiration policies help mitigate attacks from stolen accounts, or disgruntled employees who have been terminated.
- Use account lockout policies for end-user accounts. Account login attempts should have a cap on failed attempts. After the cap is exceeded the account should prevent further login attempts. Lockout helps prevent dictionary and brute force attacks.
- Use application instrumentation to expose behavior that can be monitored: Application transactions that are more likely to be targeted by malicious interactions should be logged or monitored. Examples of this might be adding logging code to an exception handler, or logging individual API calls. By providing a means to watch these transactions you have a higher likelihood of being able to identify malicious behavior quickly.
- Use authentication mechanisms that do not require clear text credentials to be passed over the network: A variety of authentication approaches exist for use with web based applications some involve the use of tokens while others will pass user credentials (user name/id and password) over the wire. When possible, it is safer to use an authentication mechanism that does not pass the credentials. If credentials must be passed, it is preferable to encrypt them, and/or send them over an encrypted channel such as SSL.
- Use least privileged accounts. The privileges granted to the authenticated party should be the minimum required to perform all required tasks. Be careful of using existing roles that have permissions beyond what is required.
- Use least privileged process and service accounts. Allocate accounts specifically for process and service accounts. Lock down the privileges of these accounts separately from other accounts.
- Use multiple gatekeepers. Passing the authentication system should not provide a golden ticket to any/all functionality. System and/or application resources should have restricted levels of access depending on the authenticated party. Some design patterns might also enforce multiple authentications, sometimes distributed through application tiers.
- Use SSL to protect session authentication cookies. Session authentication cookies contain data that can be used in a number of different attacks such as replay, Cross-Site Scripting or Cross-Site Request Forgery. Protecting these cookies helps mitigate these risks.
- Use strong authentication and authorization on administration interfaces. Always require authenticated access to administrative interfaces. When applicable, also enforce separation of privileges within the administrative interfaces.
- Use structured exception handling. A structured approach to exception handling lowers the risk of unexpected exceptions from going unhandled.
- Use the correct algorithm and correct key length. Different encryption algorithms are preferred for varying data types and scenarios.
- Use tried and tested platform features. Many cryptographic features are available through the .NET Framework. These are proven features and should be used in favor of custom methods.
- Validate all values sent from the client. Similar to not relying on client-side validation, any input from a client should always be assumed to have been tampered with. This input should always be validated before it is used. This encompasses user input, cookie values, HTTP headers, and anything else that is sent over the wires from the client.
- Validate data for type, length, format, and range. Data validation should encompass these primary tenets. Validate for data type, string lengths, string or numeric formats, and numeric ranges.
For more information on preferred encryption algorithms and key lengths, see the Security Development Lifecycle at http://www.microsoft.com/security/sdl/ .