Apache Server 2.4 Windows Site Security Technical Implementation Guide

Web server management includes the ability to control the number of users and user sessions that utilize a web server. Limiting the number of allowed users and sessions per user is helpful in limiting risks related to several types of denial-of-service (DoS) attacks. Although there is some latitude concerning the settings, they should follow DoD-recommended values, but the settings should be configurable to allow for future DoD direction. While the DoD will specify recommended values, the values can be adjusted to accommodate the operational requirement of a given system.

Session management is the practice of protecting the bulk of the user authorization and identity information. Storing of this data can occur on the client system or on the server. When the session information is stored on the client, the session ID, along with the user authorization and identity information, is sent along with each client request and is stored in a cookie, embedded in the uniform resource locator (URL), or placed in a hidden field on the displayed form. Each of these offers advantages and disadvantages. The biggest disadvantage to all three is the hijacking of a session along with all of the user's credentials. When the user authorization and identity information is stored on the server in a protected and encrypted database, the communication between the client and web server will only send the session identifier, and the server can then retrieve user credentials for the session when needed. If, during transmission, the session were to be hijacked, the user's credentials would not be compromised.

Web server logging capability is critical for accurate forensic analysis. Without sufficient and accurate information, a correct replay of the events cannot be determined. Ascertaining the correct type of event that occurred is important during forensic analysis. The correct determination of the event and when it occurred is important in relation to other events that happened at that same time. Without sufficient information establishing what type of log event occurred, investigation into the cause of event is severely hindered. Log record content that may be necessary to satisfy the requirement of this control includes but is not limited to time stamps, source and destination IP addresses, user/process identifiers, event descriptions, application-specific events, success/fail indications, file names involved, access control, and flow control rules invoked. Satisfies: SRG-APP-000095-WSR-000056, SRG-APP-000098-WSR-000060, SRG-APP-000099-WSR-000061, SRG-APP-000100-WSR-000064

User management and authentication can be an essential part of any application hosted by the web server. Along with authenticating users, the user management function must perform several other tasks such as password complexity, locking users after a configurable number of failed logons, and management of temporary and emergency accounts. All of this must be done enterprise-wide. The web server contains a minimal user management function, but the web server user management function does not offer enterprise-wide user management, and user management is not the primary function of the web server. User management for the hosted applications should be done through a facility that is built for enterprise-wide user management, such as LDAP and Active Directory.

Resource mapping is the process of tying a particular file type to a process in the web server that can serve that type of file to a requesting client and to identify which file types are not to be delivered to a client. By not specifying which files can and cannot be served to a user, the web server could deliver to a user web server configuration files, log files, password files, etc. The web server must only allow hosted application file types to be served to a user, and all other types must be disabled. Satisfies: SRG-APP-000141-WSR-000081, SRG-APP-000141-WSR-000083

Scripts allow server-side processing on behalf of the hosted application user or as processes needed in the implementation of hosted applications. Removing scripts not needed for application operation or deemed vulnerable helps to secure the web server. To ensure scripts are not added to the web server and run maliciously, script mappings that are not needed or used by the web server for hosted application operation must be removed.

A web server is designed to deliver content and execute scripts or applications on the request of a client or user. Containing user requests to files in the directory tree of the hosted web application and limiting the execution of scripts and applications guarantees that the user is not accessing information protected outside the application's realm. The web server must also prohibit users from jumping outside the hosted application directory tree through access to the user's home directory, symbolic links or shortcuts, or through search paths for missing files.

The web server must be configured to listen on a specified IP address and port. Without specifying an IP address and port for the web server to use, the web server will listen on all IP addresses available to the hosting server. If the web server has multiple IP addresses, i.e., a management IP address, the web server will also accept connections on the management IP address. Accessing the hosted application through an IP address normally used for non-application functions opens the possibility of user access to resources, utilities, files, ports, and protocols that are protected on the desired application IP address.

A certificate's certification path is the path from the end entity certificate to a trusted root certification authority (CA). Certification path validation is necessary for a relying party to make an informed decision regarding acceptance of an end entity certificate. Certification path validation includes checks such as certificate issuer trust, time validity, and revocation status for each certificate in the certification path. Revocation status information for CA and subject certificates in a certification path is commonly provided via certificate revocation lists (CRLs) or online certificate status protocol (OCSP) responses.

The web server's private key is used to prove the identity of the server to clients and securely exchange the shared secret key used to encrypt communications between the web server and clients. By gaining access to the private key, an attacker can pretend to be an authorized server and decrypt the SSL traffic between a client and the web server.

As a rule, accounts on a web server are to be kept to a minimum. Only administrators, web managers, developers, auditors, and web authors require accounts on the machine hosting the web server. The resources to which these accounts have access must also be closely monitored and controlled. Only the system administrator needs access to all the system's capabilities, while the web administrator and associated staff require access and control of the web content and web server configuration files.

The separation of user functionality from web server management can be accomplished by moving management functions to a separate IP address or port. To further separate the management functions, separate authentication methods and certificates should be used. By moving the management functionality, the possibility of accidental discovery of the management functions by non-privileged users during hosted application use is minimized.

Captured sessions can be reused in "replay" attacks. This requirement limits the ability of adversaries from capturing and continuing to employ previously valid session IDs. Session IDs are tokens generated by web applications to uniquely identify an application user's session. Unique session IDs help to reduce predictability of said identifiers. When a user logs out, or when any other session termination event occurs, the web server must terminate the user session to minimize the potential for an attacker to hijack that particular user session.

Cookies are used to exchange data between the web server and the client. Cookies, such as a session cookie, may contain session information and user credentials used to maintain a persistent connection between the user and the hosted application since HTTP/HTTPS is a stateless protocol. When the cookie parameters are not set properly (i.e., domain and path parameters), cookies can be shared within hosted applications residing on the same web server or to applications hosted on different web servers residing on the same domain.

Communication between a client and the web server is done using the HTTP protocol, but HTTP is a stateless protocol. To maintain a connection or session, a web server will generate a session identifier (ID) for each client session when the session is initiated. The session ID allows the web server to track a user session and, in many cases, the user, if the user previously logged on to a hosted application. When a web server accepts session identifiers that are not generated by the web server, the web server creates an environment where session hijacking, such as session fixation, could be used to access hosted applications through session IDs that have already been authenticated. Forcing the web server to only accept web server-generated session IDs and to create new session IDs once a user is authenticated will limit session hijacking.

Communication between a client and the web server is done using the HTTP protocol, but HTTP is a stateless protocol. To maintain a connection or session, a web server will generate a session identifier (ID) for each client session when the session is initiated. The session ID allows the web server to track a user session and, in many cases, the user, if the user previously logged on to a hosted application. By being able to guess session IDs, an attacker can easily perform a man-in-the-middle attack. To truly generate random session identifiers that cannot be reproduced, the web server session ID generator, when used twice with the same input criteria, must generate an unrelated random ID. The session ID generator also needs to be a FIPS 140-2 approved generator. Satisfies: SRG-APP-000224-WSR-000136, SRG-APP-000224-WSR-000137

Generating a session identifier (ID) that is not easily guessed through brute force is essential to deter several types of session attacks. By knowing the session ID, an attacker can hijack a user session that has already been user authenticated by the hosted application. The attacker does not need to guess user identifiers and passwords or have a secure token since the user session has already been authenticated. By generating session IDs that contain as much of the character set as possible, i.e., A-Z, a-z, and 0-9, the session ID becomes exponentially harder to guess.

Making certain that the web server has not been updated by an unauthorized user is always a concern. Adding patches, functions, and modules that are untested and not part of the baseline opens the possibility for security risks. The web server must offer, and not hinder, a method that allows for the quick and easy reinstallation of a verified and patched baseline to guarantee the production web server is up-to-date and has not been modified to add functionality or expose security risks. When the web server does not offer a method to roll back to a clean baseline, external methods, such as a baseline snapshot or virtualizing the web server, can be used.

The web server may host applications that display information that cannot be disrupted, such as information that is time critical or life threatening. In these cases, a web server that shuts down or ceases to be accessible when there is a failure is not acceptable. In these types of cases, clustering of web servers is used. Clustering of multiple web servers is a common approach to providing fail-safe application availability. To ensure application availability, the web server must provide clustering or some form of failover functionality. Satisfies: SRG-APP-000225-WSR-000141, SRG-APP-000356-WSR-000007

A web server is used to deliver content on the request of a client. The content delivered to a client must be controlled, allowing only hosted application files to be accessed and delivered. To allow a client access to system files of any type is a major security risk that is entirely avoidable. Obtaining such access is the goal of directory traversal and URL manipulation vulnerabilities. To facilitate such access by misconfiguring the web document (home) directory is a serious error. In addition, having the path on the same drive as the system folder compounds potential attacks, such as drive space exhaustion.

The goal is to completely control the web user's experience in navigating any portion of the web document root directories. Ensuring all web content directories have at least the equivalent of an "index.html" file is a significant factor to accomplish this end. Enumeration techniques, such as URL parameter manipulation, rely upon being able to obtain information about the web server's directory structure by locating directories without default pages. In the scenario, the web server will display to the user a listing of the files in the directory being accessed. By having a default hosted application web page, the anonymous web user will not obtain directory browsing information or an error message that reveals the server type and version.

Information needed by an attacker to begin looking for possible vulnerabilities in a web server includes any information about the web server, backend systems being accessed, and plug-ins or modules being used. Web servers will often display error messages to client users displaying enough information to aid in the debugging of the error. The information given back in error messages may display the web server type, version, patches installed, plug-ins and modules installed, type of code being used by the hosted application, and any backends being used for data storage. This information could be used by an attacker to blueprint what type of attacks might be successful. The information given to users must be minimized to not aid in the blueprinting of the web server.

Information needed by an attacker to begin looking for possible vulnerabilities in a web server includes any information about the web server and plug-ins or modules being used. When debugging or trace information is enabled in a production web server, information about the web server, such as web server type, version, patches installed, plug-ins and modules installed, type of code being used by the hosted application, and any backends being used for data storage may be displayed. Since this information may be placed in logs and general messages during normal operation of the Apache web server, an attacker does not need to cause an error condition to gain this information.

Leaving sessions open indefinitely is a major security risk. An attacker can easily use an already authenticated session to access the hosted application as the previously authenticated user. By closing sessions after an absolute period of time, the user is forced to reauthenticate, guaranteeing the session is still in use. Enabling an absolute timeout for sessions closes sessions that are still active. Examples would be a runaway process accessing the web server or an attacker using a hijacked session to slowly probe the web server.

Leaving sessions open indefinitely is a major security risk. An attacker can easily use an already authenticated session to access the hosted application as the previously authenticated user. Timeouts for completing the TLS handshake, receiving the request headers and/or the request body from the client. If the client fails to complete each of these stages within the configured time, a 408 REQUEST TIME OUT error is sent. For SSL virtual hosts, the handshake timeout values is the time needed to do the initial SSL handshake. If the user's browser is configured to query certificate revocation lists and the CRL server is not reachable, the initial SSL handshake may take a significant time until the browser gives up waiting for the CRL. Therefore the handshake timeout should take this possible overhead into consideration for SSL virtual hosts (if necessary). The body timeout values include the time needed for SSL renegotiation (if necessary).

Remote access to the web server is any access that communicates through an external, non-organization-controlled network. Remote access can be used to access hosted applications or to perform management functions. A web server can be accessed remotely and must be capable of restricting access from what the DoD defines as nonsecure zones. Nonsecure zones are defined as any IP, subnet, or region that is defined as a threat to the organization. The nonsecure zones must be defined for public web servers logically located in a DMZ, as well as private web servers with perimeter protection devices. By restricting access from nonsecure zones, through the internal web server access list, the web server can stop or slow denial-of-service (DoS) attacks on the web server.

By separating web server security functions from non-privileged users, roles can be developed that can then be used to administer the web server. Forcing users to change from a non-privileged account to a privileged account when operating on the web server or on security-relevant information forces users to only operate as a web server administrator when necessary. Operating in this manner allows for better logging of changes and better forensic information and limits accidental changes to the web server.

Web servers provide numerous processes, features, and functionalities that use TCP/IP ports. Some of these processes may be deemed unnecessary or too unsecure to run on a production system. The web server must provide the capability to disable or deactivate network-related services that are deemed to be non-essential to the server mission, are too unsecure, or are prohibited by the Ports, Protocols, and Services Management (PPSM) Category Assurance List (CAL) and vulnerability assessments.

Non-DoD-approved PKIs have not been evaluated to ensure that they have security controls and identity vetting procedures in place that are sufficient for DoD systems to rely on the identity asserted in the certificate. PKIs lacking sufficient security controls and identity vetting procedures risk being compromised and issuing certificates that enable adversaries to impersonate legitimate users.

A denial of service (DoS) can occur when the web server is so overwhelmed that it can no longer respond to additional requests. A web server not properly tuned may become overwhelmed and cause a DoS condition even with expected traffic from users. To avoid a DoS, the web server must be tuned to handle the expected traffic for the hosted applications. Satisfies: SRG-APP-000435-WSR-000148, SRG-APP-000246-WSR-000149

A cookie is used when a web server needs to share data with the client's browser. The data is often used to remember the client when the client returns to the hosted application at a later date. A session cookie is a special type of cookie used to remember the client during the session. The cookie will contain the session identifier (ID) and may contain authentication data to the hosted application. To protect this data from easily being compromised, the cookie can be encrypted. When a cookie is sent encrypted via SSL/TLS, an attacker must spend a great deal of time and resources to decrypt the cookie. If, along with encryption, the cookie is compressed, the attacker can now use a combination of plaintext injection and inadvertent information leakage through data compression to reduce the time needed to decrypt the cookie. This attack is called Compression Ratio Info-leak Made Easy (CRIME). Cookies shared between the web server and the client when encrypted should not also be compressed.

Cookies can be sent to a client using TLS/SSL to encrypt the cookies, but TLS/SSL is not used by every hosted application since the data being displayed does not require the encryption of the transmission. To safeguard against cookies, especially session cookies, being sent in plaintext, a cookie can be encrypted before transmission. To force a cookie to be encrypted before transmission, the cookie "Secure" property can be set.

A cookie can be read by client-side scripts easily if cookie properties are not set properly. By allowing cookies to be read by the client-side scripts, information such as session identifiers could be compromised and used by an attacker who intercepts the cookie. Setting cookie properties (i.e., HttpOnly property) to disallow client-side scripts from reading cookies better protects the information inside the cookie.