Azure Security Notes

Transcription

1 Azure Security Notes Lessons Learned from Exploring Microsoft Azure and the Cloud Security Space J.D. Meier, Principal PM, Microsoft Paul Enfield

2 Acknowledgments Thanks to the following external contributors and reviewers: Adam Grocholski; Andy Eunson; Bill Collette; Christopher Seary; Jason Taylor; John Daniels; Juval Lowy; Kevin Lam; Long Le; Michael Smith; Michael Stiefel; Michele Leroux Bustamante; Norman Headlam; Rockford Lhotka; Rudolph Araujo; Sarang Kulkarni; Steven Nagy; Terrance Snyder; Will Clevenger Thanks to the following Microsoft contributors and reviewers: Akshay Aggarwal; Alik Levin; Andreas Fuchsberger; Babur Butter; Bharat Shyam; Dave Brankin; Danny Cohen; Diego Dagum; Don Willits; Eugenio Pace; Gabriel Morgan; Jeff Mueller; John Steer; Julian Gonzalez; Mark Curphey; Mohit Srivastava; Pat Filoteo; Rahul Verma; Raul Rojas; Scott Densmore; Sesha Mani; Serena Yeoh; Sriram Krishnan; Stefan Schackow; Steve Marx; Stuart Kwan; Terri Schmidt; Tobin Titus; Varun Sharma; Vidya Vrat Agarwal; Vikram Bhambri; Yale Li

3 Our Cloud Security Approach Objectives Learn a security frame Learn an approach for applying the security frame to application scenarios Learn key components of the security approach for these notes Overview These notes summarize how we took a look at securing Web applications, web services, and data in the cloud. We basically started by mapping out common application scenarios. We then figured out a way to group threats and attacks into a set of common categories. These categories make it easy for us to analyze and inspect the various parts of an application and engineering decisions. Then we focused on finding common principles, patterns, and practices, while working with customers, field, product teams, and industry experts. Our Cloud Security Approach in a Nutshell We took a pragmatic approach to security, identifying key deltas, building on what we know, and leveraging proven practices. Here are some of the cornerstone concepts we used to focus and drive our security efforts: Visual Application Scenarios and Solutions. We created simple visuals that show end-toend scenarios. You can think of these as how you might put on a whiteboard if you were to draw the end in mind for a particular application scenario. This is a simple way to share authentication and authorization patterns for common application scenarios. Secure the network, host, and application. Ask any security expert, where the problems are, and they ll say it s the stuff that falls between the cracks. Thinking about security holistically across the network, host, and application is the key to both helping you focus your efforts, as well as catching what falls between the cracks. Principles, patterns, and practices. Design principles help you leverage proven practices. Rather than have a rule for everything, you have a smaller set of principles to guide you. Patterns are simply problem and solution pairs in context. They help share strategies and principles by building a shared vocabulary. For example, the Trusted Subsystem pattern encapsulates the idea of trusting the application identity rather than every caller. Practices are simply actions, methods, or techniques. For example, you can use patterns & practices Threat Modeling to shape the security design of your application. Threats and countermeasures. One way to make security actionable and prescriptive is to focus on threats and countermeasures. This means identifying the bad things that can happen, and what your defenses, security controls, or countermeasures are.

4 Security Frame. There are many ways to chunk up and think about security. For example, you might use C.I.A., where C is confidentiality, I is integrity, and A represent availability. Another approach is to focus on the three A s: Authentication, Access control, and Auditing. We ve found it especially useful to use the following categories to organize and share application security principles, patterns, anti-patterns, and practices: Auditing and Logging, Authentication, Authorization, Communication, Configuration Management, Cryptography, Exception Management, Sensitive Data, Session Management, and Validation. People, process, and technology. Another way to look at security is through the people, process, and technology viewpoints. You can also use each of these views when you look at security for your network, host, and application. Application, Infrastructure, and Business. This is another helpful lens for looking at security. You can look at the principles, patterns, and practices for the application, the infrastructure, or the business. This leads to application security patterns, infrastructure security patterns, and business security patterns and practices. At the end of the day, what s important is using the lens that helps you the most, and avoid using the wrong tool for the job, or using a one-size fits all solution. Instead, chunk up the problem, while keeping a holistic view. How We Got Started Tackling cloud security can be overwhelming without an approach. In these notes, we have an approach that simplifies and improves security. It does so by chunking up security in a way that helps us leverage proven practices, while sharing information around emerging practices. One way to simplify thinking about security is to simplify the language. When we first started talking with customers about the cloud and security, it was difficult to have meaningful conversations, simply because we didn t have a shared vocabulary. Using a threats and countermeasures approach to talk about security is a key step to simplifying cloud security. Basically, we can look through our security lens (our security frame), and look for threats, attacks, vulnerabilities, and countermeasures. We can use that same lens and look for principles, patterns, and anti-patterns. This helps us turn security information into insight and action. Another important step is leveraging application types from the patterns & practices Application Architecture Guide, second edition - Rather than focus on queues and worker roles, we can shift our focus to application scenarios, and represent them visually. Our goal is to show how to secure your Web application, Web services, and data in the cloud visually. With these scenarios we can move to the whiteboard and draw end-to-end scenarios.

5 By using a security frame, focusing on threats and countermeasures, and visual scenarios, we have an effective backdrop for analyzing, evaluating, and talking about cloud security challenges. Threats and Countermeasures Focus One way to make security actionable and prescriptive is to focus on threats, attacks, vulnerabilities and countermeasures: Threat - A potential occurrence malicious or otherwise that may harm an asset. Attack - An action taken to harm an asset. Vulnerability - A weakness that makes a threat possible. Countermeasure - A safeguard that addresses a threat and mitigates risk. This means identifying the bad things that can happen, and what your defenses, security controls, or countermeasures are. By building a knowledgebase of threats, attacks, vulnerabilities, and countermeasures, we can dramatically improve our ability to share security know-how. Example Application Scenario and Solution Each Visual Scenario and Solution consists of a Scenario, Solution, and Solution Summary Table. The Scenario is a simple visual of the application from a deployment standpoint. The Solution is a visual of how you might address authentication, authorization, and communication (the security runtime patterns.) The Solution Summary Table is a quick description of how you would address the authentication, authorization, and communication from a security standpoint. Scenario

6 Solution Solution Summary Table Area Notes Authentication Authenticate users with forms authentication Store users in Azure Tables Use TableStorageMembershipProvider for membership API s Authenticate application against Azure Storage with access key Map ASP.NET users to single Azure Storage connection All access to the database occurs as the application identity Authorization Use TableStorageRoleProvider for roles API s Authorize users in application logic Store roles in Azure Tables Communication Use HTTP port 80 for non-secure connections to ASP.NET application Perform forms authentication over port 443 (SSL) Security Frame A Security Frame acts as a simple lens. It helps organize and share security knowledge in a set of actionable buckets. Here is an explanation of the key buckets used in these notes: Category Auditing and Logging Authentication Authorization Communication Configuration Management Cryptography Exception Management Sensitive Data Notes How security-related events are recorded, monitored, and audited. The process of proving identity, typically through credentials, such as a user name and password How your application provides access controls for roles, resources and operations. How data is transmitted over the wire. Transport security versus message encryption is covered here. How your application handles configuration and administration of your applications from a security perspective. How your application enforces confidentiality and integrity. How you handle applications errors and exceptions. How your application handles any data that must be protected either in

7 Session Management Validation memory, over the network, or in persistent stores A session refers to a series of related interactions between a user and your application. How your application filters, scrubs, or rejects input before additional processing, or how it sanitizes output. The key to the buckets is that they are actionable. Another key is that developers can relate to them. This makes it easier to share security knowledge between security experts and developers in a pragmatic way. Carrying the Story Forward Improving Web Application Security: Threats and Countermeasures ( illustrates the approach of securing a solution by addressing key security hotspots defined by the security frame in Figure 1 below. This visual model allows you to conceptualize specific areas and items in an architecture that need to be addressed during the process of securing an application. The visual model represents a fairly typical on-premise application architecture, and then pins hotspots against it. Securing the Application Auditing and Logging Authentication Authorization Configuration Mgt. Cryptography Exception Management Parameter Manipulation Sensitive Data Session Management Validation Web Server Application Server Database Server Firewall Apps Host Firewall Apps Host Database Host Securing the Network Router Firewall Switch Patches and Updates Services Protocols Securing the Host Accounts Files and Directories Shares Ports Registry Auditing and Logging Threats and Countermeasures Figure 1. Securing the Network, Host and Application On-premise Using this visual model concept and the revised security frame for cloud applications we can use the same approach to evaluate cloud applications. In Figure 2 below, we modify the visual

8 model by hosting both the application and data layers in the managed infrastructure, which in this case is the cloud. With a managed infrastructure we can remove checks from our list because they are items handled by the managed infrastructure. For example, a Windows Azure application will not have permissions to create user accounts, or elevate privileges. This alleviates the need to manage accounts at the host level. Securing the Application Auditing and Logging Authentication Authorization Communication Configuration Management Cryptography Exception Management Sensitive Data Session Management Validation Application Layer Database Layer Firewall Apps Cloud Host Data Managed Infrastructure Securing the Network Router Firewall Switch Securing the Host Patches and Updates Accounts Services Files and Directories Protocols Shares Ports Registry Auditing and Logging Threats and Countermeasures Figure 2. Securing the Network, Host and Application in the Cloud The above diagram represents the canonical case architecture. To evaluate other architectures such as ones involving federated identity or hybrid on-premise and cloud implementations, you can use the same approach. Represent your architecture with a base diagram, and overlay the frame on it. Once the frame is overlaid, you can evaluate each item for applicability and quickly scope out categories not needing attention.

9 Cloud Security Threats and Countermeasures Objectives 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. Overview 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. This section is a reference of the security frame, and its pieces pinned against the cloud as a platform. We also present threats, vulnerabilities and countermeasures against the cloud organized by hot spot such that you can see how they relate to each other in each category. Threats, Attacks, Vulnerabilities, and Countermeasures This chapter analyzes security from the perspectives of threats, vulnerabilities, attacks, and countermeasures against the cloud. 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 and architecting Windows Azure applications. Understanding these concepts helps you put core security features such as authentication, authorization, auditing, confidentiality, integrity, and availability in action. Hot Spot Auditing and Logging Authentication Description Auditing and logging in Windows Azure 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 is the process of proving identity, typically through credentials, such as a user name and password. In the cloud this also encompasses

10 Authorization Communication Configuration Management Cryptography Input and Data Validation Exception Management Sensitive Data authentication against varying identity stores. Authorization in Windows Azure 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 encompasses how data is transmitted over the wire. Transport security, message encryption, and point to point communication are covered here. 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 in Windows Azure 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. 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 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 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? Threats and Attacks Category Auditing and Logging Items Disclosure of confidential information. An attacker gathers sensitive information from log files. Denial of service (DoS). An attacker overwhelms logs with excessive entries or very large log entries.

11 Repudiation. An attacker denies performing an operation, exploits an application without trace, or covers his or her tracks. Authentication Network eavesdropping. An attacker steals identity and/or credentials off the network by reading network traffic not intended for them. Brute force attacks. An attacker guesses identity and/or credentials through the use of brute force. Dictionary attacks. An attacker guesses identity and/or credentials through the use of common terms in a dictionary designed for that purpose. Cookie replay attacks. An attacker gains access to an authenticated session through the reuse of a stolen cookie containing session information. Credential theft. An attacker gains access to credentials through data theft; for instance, phishing or social engineering. Authorization Elevation of privilege. An attacker enters a system as a lower-level user, but is able to obtain higher-level access. Disclosure of confidential data. An attacker accesses confidential information because of authorization failure on a resource or operation. Data tampering. An attacker modifies sensitive data because of authorization failure on a resource or operation. Luring attacks. An attacker lures a higher-privileged user into taking an action on their behalf. This is not an authorization failure but rather a failure of the system to properly inform the user. Token stealing. An attacker steals the credentials or token of another user in order to gain authorization to resources or operations they would not otherwise be able to access. Communication Failure to encrypt messages. An attacker is able to read message content off the network because it is not encrypted. Theft of encryption keys. An attacker is able to decrypt sensitive data because he or she has the keys. Man-in-the-middle attack. An attacker can read and then modify messages between the client and the service. Session replay. An attacker steals messages off the network and replays them in order to steal a user's session. Data tampering. An attacker modifies the data in a message in order to attack the client or the service Configuration Management Unauthorized access to configuration stores. An attacker gains access to configuration files and is able to modify binding settings, etc. Retrieval of clear text configuration secrets. An attacker gains access to configuration files and is able to retrieve sensitive information such as database connection strings. Cryptography Encryption cracking. Breaking an encryption algorithm and gaining access to the encrypted data. Loss of decryption keys. Obtaining decryption keys and using them to access protected data. Exception Management Information disclosure. Sensitive system or application details are revealed through exception information. Denial of service. An attacker uses error conditions to stop your service or place it in an unrecoverable error state.

12 Elevation of privilege. Your service encounters an error and fails to an insecure state; for instance, failing to revert impersonation. Input and Data Validation Canonicalization attacks. Canonicalization attacks can occur anytime validation is performed on a different form of the input than that which is used for later processing. For instance, a validation check may be performed on an encoded string, which is later decoded and used as a file path or URL. Cross-site scripting. Cross-site scripting can occur if you fail to encode user input before echoing back to a client that will render it as HTML. SQL injection. Failure to validate input can result in SQL injection if the input is used to construct a SQL statement, or if it will modify the construction of a SQL statement in some way. Cross-Site Request Forgery: CSRF attacks involve forged transactions submitted to a site on behalf of another party. XPath injection. XPath injection can result if the input sent to the Web service is used to influence or construct an XPath statement. The input can also introduce unintended results if the XPath statement is used by the Web service as part of some larger operation, such as applying an XQuery or an XSLT transformation to an XML document. XML bomb. XML bomb attacks occur when specific, small XML messages are parsed by a service resulting in data that feeds on itself and grows exponentially. An attacker sends an XML bomb with the intent of overwhelming a Web service s XML parser and resulting in a denial of service attack. Sensitive Data Memory dumping. An attacker is able to read sensitive data out of memory or from local files. Network eavesdropping. An attacker sniffs unencrypted sensitive data off the network. Configuration file sniffing. An attacker steals sensitive information, such as connection strings, out of configuration files. Vulnerabilities Category Items Auditing and Failing to audit failed logons. Logging Failing to secure log files. Storing sensitive information in log files. Failing to audit across application tiers. Failure to throttle log files. Authentication Using weak passwords. Storing clear text credentials in configuration files. Passing clear text credentials over the network. Permitting prolonged session lifetime. Mixing personalization with authentication. Using weak authentication mechanisms (e.g., using basic authentication over

13 an untrusted network). Authorization Relying on a single gatekeeper (e.g., relying on client-side validation only). Failing to lock down system resources against application identities. Failing to limit database access to specified stored procedures. Using inadequate separation of privileges. Connection pooling. Permitting overprivileged accounts. Configuration Management Using insecure custom administration interfaces. Failing to secure configuration files on the server. Storing sensitive information in the clear text. Having too many administrators. Using overprivileged process accounts and service accounts. Communication Not encrypting messages. Using custom cryptography. Distributing keys insecurely. Managing or storing keys insecurely. Failure to use a mechanism to detect message replays. Not using either message or transport security. Cryptography Using custom cryptography Failing to secure encryption keys Using the wrong algorithm or a key size that is too small Using the same key for a prolonged period of time Distributing keys in an insecure manner Exception Management Input and Data Validation Failure to use structured exception handling (try/catch). Revealing too much information to the client. Failure to specify fault contracts with the client. Failure to use a global exception handler. Using non-validated input used to generate SQL queries. Relying only on client-side validation. Using input file names, URLs, or usernames for security decisions. Using application-only filters for malicious input. Looking for known bad patterns of input. Trusting data read from databases, file shares, and other network resources. Failing to validate input from all sources including cookies, headers, parameters, databases, and network resources. Sensitive Data Storing secrets when you do not need to. Storing secrets in code. Storing secrets in clear text in files, registry, or configuration. Passing sensitive data in clear text over networks. Countermeasures Category Items

14 Auditing and Logging Identify malicious behavior. Know your baseline (know what good traffic looks like). Use application instrumentation to expose behavior that can be monitored. Throttle logging. Strip sensitive data before logging. Authentication Use strong password policies. Do not store credentials in an insecure manner. Use authentication mechanisms that do not require clear text credentials to be passed over the network. Encrypt communication channels to secure authentication tokens. Use Secure HTTP (HTTPS) only with forms authentication cookies. Separate anonymous from authenticated pages. Using cryptographic random number generators to generate session IDs. Authorization Use least-privileged accounts. Tie authentication to authorization on the same tier. Consider granularity of access. Enforce separation of privileges. Use multiple gatekeepers. Secure system resources against system identities. Configuration Management Using insecure custom administration interfaces. Failing to secure configuration files on the server. Storing sensitive information in the clear text. Having too many administrators. Using overprivileged process accounts and service accounts. Communication Use message security or transport security to encrypt your messages. Use proven platform-provided cryptography. Periodically change your keys. Use any platform-provided replay detection features. Consider creating custom code if the platform does not provide a detection mechanism. Turn on message or transport security. Cryptography Do not develop and use proprietary algorithms (XOR is not encryption. Use established cryptography such as RSA) Avoid key management. Use the RNGCryptoServiceProvider method to generate random numbers Periodically change your keys Exception Management Use structured exception handling (by using try/catch blocks). Catch and wrap exceptions only if the operation adds value/information. Do not reveal sensitive system or application information. Implement a global exception handler. Do not log private data such as passwords. Sensitive Data Do not store secrets in software. Encrypt sensitive data over the network. Secure the channel. Encrypt sensitive data in configuration files. Validation Do not trust client input.

15 Validate input: length, range, format, and type. Validate XML streams. Constrain, reject, and sanitize input. Encode output. Restrict the size, length, and depth of parsed XML messages. Threats and Attacks Explained 1. 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. 2. 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. 3. 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. 4. 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. 5. 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. 6. Credential theft. Stealing the verification part of an authentication pair (identity + credentials = authentication). Passwords are a common credential. 7. 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. 8. 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. 9. 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 lowerprivileged user or purpose. This can potentially expose vulnerability if the connection is not reauthorized when used by a new identity.

16 10. 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. 11. 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. 12. 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. 13. 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. 14. 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 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. 16. 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. 17. 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. 18. 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.

17 19. 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. 20. 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. 21. 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. 22. 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. 23. 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. 24. Session replay. An attacker steals messages off of the network and replays them in order to steal a user s session. 25. 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 The attacker then sends this URL to a valid user, who clicks on it. At this point, a valid session with the ID 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. 26. 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. 27. 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. 28. 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.

18 Countermeasures Explained 1. 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. 2. 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. 3. Avoid storing secrets. Design around storing secrets. If necessary, sometimes they can be avoided by storing them after using a one-way hash algorithm. 4. 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. 5. 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. 6. 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. 7. 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. 8. Catch exceptions. Unhandled exceptions are at risk of passing too much information to the client. Handle exceptions when possible. 9. Centralize your input and data validation. Input and data validation should be performed using a common set of code such as a validation library. 10. Consider a centralized exception management framework. Exception handling frameworks are available publically and provide an established and tested means for handling exceptions. 11. 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. 12. Consider identity flow. Auditing should be traceable back to the authenticated party. Take note of identity transitions imposed by design decisions like impersonation. 13. Constrain input. Limit user input to expected ranges and formats. 14. Constrain, reject, and sanitize your input. Constrain, reject and sanitize should be primary techniques in handling input data. 15. Cycle your keys periodically. Expiring encryption keys lowers the risk of stolen keys.

19 16. 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. 17. 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. 18. 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. 19. 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. 20. 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. 21. 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. 22. 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. 23. Do not store credentials in plaintext. Credentials are sometimes stored in application configuration files, repositories, or sent over . Always encrypt credentials before storing them. 24. Do not store database connections, passwords, or keys in plaintext. Configuration secrets should always be stored in encrypted form, external to the code. 25. 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 regenerated with the SALT and the result can be compared to the original encrypted password. 26. 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. 27. 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. 28. 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. 29. 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.

20 30. 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. 31. Encrypt sensitive cookie state. Sensitive data contained within cookies should always be encrypted. 32. 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. 33. 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. 34. 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. 35. Enforce unique transactions. Identify each transaction from a client uniquely to help prevent replay and forgery attacks. 36. 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. 37. 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. 38. 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. 39. 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. 40. Log detailed error messages. Highly detailed error message logging can provide clues to attempted attacks. 41. Log key events. Profile your application and note key or sensitive operations and/or events, and log these events during application operation. 42. 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. 43. 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. 44. 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.