Tom Rhodes Written by MAC Mandatory Access Control MAC &os; 5.X introduced new security extensions from the TrustedBSD project based on the &posix;.1e draft. Two of the most significant new security mechanisms are file system Access Control Lists (ACLs) and Mandatory Access Control (MAC) facilities. Mandatory Access Control allows new access control modules to be loaded, implementing new security policies. Some provide protections of a narrow subset of the system, hardening a particular service. Others provide comprehensive labeled security across all subjects and objects. The mandatory part of the definition comes from the fact that the enforcement of the controls is done by administrators and the system, and is not left up to the discretion of users as is done with discretionary access control (DAC, the standard file and System V IPC permissions on &os;). This chapter will focus on the Mandatory Access Control Framework (MAC Framework), and a set of pluggable security policy modules enabling various security mechanisms. After reading this chapter, you will know: What MAC security policy modules are currently included in &os; and their associated mechanisms. What MAC security policy modules implement as well as the difference between a labeled and non-labeled policy. How to efficiently configure a system to use the MAC framework. How to configure the different security policy modules included with the MAC framework. How to implement a more secure environment using the MAC framework and the examples shown. How to test the MAC configuration to ensure the framework has been properly implemented. Before reading this chapter, you should: Understand &unix; and &os; basics (). Be familiar with the basics of kernel configuration/compilation (). Have some familiarity with security and how it pertains to &os; (). The improper use of the information contained herein may cause loss of system access, aggravation of users, or inability to access the features provided by X11. More importantly, MAC should not be relied upon to completely secure a system. The MAC framework only augments existing security policy; without sound security practices and regular security checks, the system will never be completely secure. It should also be noted that the examples contained within this chapter are just that, examples. It is not recommended that these particular settings be rolled out on a production system. Implementing the various security policy modules takes a good deal of thought and testing. One who does not fully understand exactly how everything works may find him or herself going back through the entire system and reconfiguring many files or directories. This chapter covers a broad range of security issues relating to the MAC framework. The development of new MAC security policy modules will not be covered. A number of security policy modules included with the MAC framework have specific characteristics which are provided for both testing and new module development. These include the &man.mac.test.4;, &man.mac.stub.4; and &man.mac.none.4;. For more information on these security policy modules and the various mechanisms they provide, please review the manual pages. Before reading this chapter, a few key terms must be explained. This will hopefully clear up any confusion that may occur and avoid the abrupt introduction of new terms and information. compartment: A compartment is a set of programs and data to be partitioned or separated, where users are given explicit access to specific components of a system. Also, a compartment represents a grouping, such as a work group, department, project, or topic. Using compartments, it is possible to implement a need-to-know security policy. high water mark: A high water mark policy is one which permits the raising of security levels for the purpose of accessing higher level information. In most cases, the original level is restored after the process is complete. Currently, the &os; MAC framework does not have a policy for this, but the definition is included for completeness. integrity: Integrity, as a key concept, is the level of trust which can be placed on data. As the integrity of the data is elevated, so does the ability to trust that data. label: A label is a security attribute which can be applied to files, directories, or other items in the system. It could be considered a confidentiality stamp; when a label is placed on a file it describes the security properties for that specific file and will only permit access by files, users, resources, etc. with a similar security setting. The meaning and interpretation of label values depends on the policy configuration: while some policies might treat a label as representing the integrity or secrecy of an object, other policies might use labels to hold rules for access. level: The increased or decreased setting of a security attribute. As the level increases, its security is considered to elevate as well. low water mark: A low water mark policy is one which permits lowering of the security levels for the purpose of accessing information which is less secure. In most cases, the original security level of the user is restored after the process is complete. The only security policy module in &os; to use this is &man.mac.lomac.4;. multilabel: The multilabel property is a file system option which can be set in single user mode using the &man.tunefs.8; utility, during the boot operation using the &man.fstab.5; file, or during the creation of a new file system. This option will permit an administrator to apply different MAC labels on different objects. This option only applies to security policy modules which support labeling. object: An object or system object is an entity through which information flows under the direction of a subject. This includes directories, files, fields, screens, keyboards, memory, magnetic storage, printers or any other data storage/moving device. Basically, an object is a data container or a system resource; access to an object effectively means access to the data. policy: A collection of rules which defines how objectives are to be achieved. A policy usually documents how certain items are to be handled. This chapter will consider the term policy in this context as a security policy; i.e. a collection of rules which will control the flow of data and information and define whom will have access to that data and information. sensitivity: Usually used when discussing MLS. A sensitivity level is a term used to describe how important or secret the data should be. As the sensitivity level increases, so does the importance of the secrecy, or confidentiality of the data. single label: A single label is when the entire file system uses one label to enforce access control over the flow of data. When a file system has this set, which is any time when the multilabel option is not set, all files will conform to the same label setting. subject: a subject is any active entity that causes information to flow between objects; e.g., a user, user process, system process, etc. On &os;, this is almost always a thread acting in a process on behalf of a user. With all of these new terms in mind, consider how the MAC framework augments the security of the system as a whole. The various security policy modules provided by the MAC framework could be used to protect the network and file systems, block users from accessing certain ports and sockets, and more. Perhaps the best use of the policy modules is to blend them together, by loading several security policy modules at a time for a multi-layered security environment. In a multi-layered security environment, multiple policy modules are in effect to keep security in check. This is different to a hardening policy, which typically hardens elements of a system that is used only for specific purposes. The only downside is administrative overhead in cases of multiple file system labels, setting network access control user by user, etc. These downsides are minimal when compared to the lasting effect of the framework; for instance, the ability to pick and choose which policies are required for a specific configuration keeps performance overhead down. The reduction of support for unneeded policies can increase the overall performance of the system as well as offer flexibility of choice. A good implementation would consider the overall security requirements and effectively implement the various security policy modules offered by the framework. Thus a system utilizing MAC features should at least guarantee that a user will not be permitted to change security attributes at will; all user utilities, programs and scripts must work within the constraints of the access rules provided by the selected security policy modules; and that total control of the MAC access rules are in the hands of the system administrator. It is the sole duty of the system administrator to carefully select the correct security policy modules. Some environments may need to limit access control over the network; in these cases, the &man.mac.portacl.4;, &man.mac.ifoff.4; and even &man.mac.biba.4; policy modules might make good starting points. In other cases, strict confidentiality of file system objects might be required. Policy modules such as &man.mac.bsdextended.4; and &man.mac.mls.4; exist for this purpose. Policy decisions could be made based on network configuration. Perhaps only certain users should be permitted access to facilities provided by &man.ssh.1; to access the network or the Internet. The &man.mac.portacl.4; would be the policy module of choice for these situations. But what should be done in the case of file systems? Should all access to certain directories be severed from other groups or specific users? Or should we limit user or utility access to specific files by setting certain objects as classified? In the file system case, access to objects might be considered confidential to some users, but not to others. For an example, a large development team might be broken off into smaller groups of individuals. Developers in project A might not be permitted to access objects written by developers in project B. Yet they might need to access objects created by developers in project C; that is quite a situation indeed. Using the different security policy modules provided by the MAC framework; users could be divided into these groups and then given access to the appropriate areas without fear of information leakage. Thus, each security policy module has a unique way of dealing with the overall security of a system. Module selection should be based on a well thought out security policy. In many cases, the overall policy may need to be revised and reimplemented on the system. Understanding the different security policy modules offered by the MAC framework will help administrators choose the best policies for their situations. The default &os; kernel does not include the option for the MAC framework; thus the following kernel option must be added before trying any of the examples or information in this chapter: options MAC And the kernel will require a rebuild and a reinstall. While the various manual pages for MAC policy modules state that they may be built into the kernel, it is possible to lock the system out of the network and more. Implementing MAC is much like implementing a firewall, care must be taken to prevent being completely locked out of the system. The ability to revert back to a previous configuration should be considered while the implementation of MAC remotely should be done with extreme caution. A MAC label is a security attribute which may be applied to subjects and objects throughout the system. When setting a label, the user must be able to comprehend what it is, exactly, that is being done. The attributes available on an object depend on the policy module loaded, and that policy modules interpret their attributes in different ways. If improperly configured due to lack of comprehension, or the inability to understand the implications, the result will be the unexpected and perhaps, undesired, behavior of the system. The security label on an object is used as a part of a security access control decision by a policy. With some policies, the label by itself contains all information necessary to make a decision; in other models, the labels may be processed as part of a larger rule set, etc. For instance, setting the label of biba/low on a file will represent a label maintained by the Biba security policy module, with a value of low. A few policy modules which support the labeling feature in &os; offer three specific predefined labels. These are the low, high, and equal labels. Although they enforce access control in a different manner with each policy module, you can be sure that the low label will be the lowest setting, the equal label will set the subject or object to be disabled or unaffected, and the high label will enforce the highest setting available in the Biba and MLS policy modules. Within single label file system environments, only one label may be used on objects. This will enforce one set of access permissions across the entire system and in many environments may be all that is required. There are a few cases where multiple labels may be set on objects or subjects in the file system. For those cases, the multilabel option may be passed to &man.tunefs.8;. In the case of Biba and MLS, a numeric label may be set to indicate the precise level of hierarchical control. This numeric level is used to partition or sort information into different groups of say, classification only permitting access to that group or a higher group level. In most cases the administrator will only be setting up a single label to use throughout the file system. Hey wait, this is similar to DAC! I thought MAC gave control strictly to the administrator. That statement still holds true, to some extent as root is the one in control and who configures the policies so that users are placed in the appropriate categories/access levels. Alas, many policy modules can restrict the root user as well. Basic control over objects will then be released to the group, but root may revoke or modify the settings at any time. This is the hierarchal/clearance model covered by policies such as Biba and MLS. Virtually all aspects of label policy module configuration will be performed using the base system utilities. These commands provide a simple interface for object or subject configuration or the manipulation and verification of the configuration. All configuration may be done by use of the &man.setfmac.8; and &man.setpmac.8; utilities. The setfmac command is used to set MAC labels on system objects while the setpmac command is used to set the labels on system subjects. Observe: &prompt.root; setfmac biba/high test If no errors occurred with the command above, a prompt will be returned. The only time these commands are not quiescent is when an error occurred; similarly to the &man.chmod.1; and &man.chown.8; commands. In some cases this error may be a Permission denied and is usually obtained when the label is being set or modified on an object which is restricted.Other conditions may produce different failures. For instance, the file may not be owned by the user attempting to relabel the object, the object may not exist or may be read only. A mandatory policy will not allow the process to relabel the file, maybe because of a property of the file, a property of the process, or a property of the proposed new label value. For example: a user running at low integrity tries to change the label of a high integrity file. Or perhaps a user running at low integrity tries to change the label of a low integrity file to a high integrity label. The system administrator may use the following commands to overcome this: &prompt.root; setfmac biba/high test Permission denied &prompt.root; setpmac biba/low setfmac biba/high test &prompt.root; getfmac test test: biba/high As we see above, setpmac can be used to override the policy module's settings by assigning a different label to the invoked process. The getpmac utility is usually used with currently running processes, such as sendmail: although it takes a process ID in place of a command the logic is extremely similar. If users attempt to manipulate a file not in their access, subject to the rules of the loaded policy modules, the Operation not permitted error will be displayed by the mac_set_link function. For the &man.mac.biba.4;, &man.mac.mls.4; and &man.mac.lomac.4; policy modules, the ability to assign simple labels is provided. These take the form of high, equal and low, what follows is a brief description of what these labels provide: The low label is considered the lowest label setting an object or subject may have. Setting this on objects or subjects will block their access to objects or subjects marked high. The equal label should only be placed on objects considered to be exempt from the policy. The high label grants an object or subject the highest possible setting. With respect to each policy module, each of those settings will instate a different information flow directive. Reading the proper manual pages will further explain the traits of these generic label configurations. Numeric grade labels are used for comparison:compartment+compartment; thus the following: biba/10:2+3+6(5:2+3-20:2+3+4+5+6) May be interpreted as: Biba Policy Label/Grade 10 :Compartments 2, 3 and 6: (grade 5 ...) In this example, the first grade would be considered the effective grade with effective compartments, the second grade is the low grade and the last one is the high grade. In most configurations these settings will not be used; indeed, they offered for more advanced configurations. When applied to system objects, they will only have a current grade/compartments as opposed to system subjects as they reflect the range of available rights in the system, and network interfaces, where they are used for access control. The grade and compartments in a subject and object pair are used to construct a relationship referred to as dominance, in which a subject dominates an object, the object dominates the subject, neither dominates the other, or both dominate each other. The both dominate case occurs when the two labels are equal. Due to the information flow nature of Biba, you have rights to a set of compartments, need to know, that might correspond to projects, but objects also have a set of compartments. Users may have to subset their rights using su or setpmac in order to access objects in a compartment from which they are not restricted. Users themselves are required to have labels so that their files and processes may properly interact with the security policy defined on the system. This is configured through the login.conf file by use of login classes. Every policy module that uses labels will implement the user class setting. An example entry containing every policy module setting is displayed below: default:\ :copyright=/etc/COPYRIGHT:\ :welcome=/etc/motd:\ :setenv=MAIL=/var/mail/$,BLOCKSIZE=K:\ :path=~/bin:/sbin:/bin:/usr/sbin:/usr/bin:/usr/local/sbin:/usr/local/bin:\ :manpath=/usr/share/man /usr/local/man:\ :nologin=/usr/sbin/nologin:\ :cputime=1h30m:\ :datasize=8M:\ :vmemoryuse=100M:\ :stacksize=2M:\ :memorylocked=4M:\ :memoryuse=8M:\ :filesize=8M:\ :coredumpsize=8M:\ :openfiles=24:\ :maxproc=32:\ :priority=0:\ :requirehome:\ :passwordtime=91d:\ :umask=022:\ :ignoretime@:\ :label=partition/13,mls/5,biba/10(5-15),lomac/10[2]: The label option is used to set the user class default label which will be enforced by MAC. Users will never be permitted to modify this value, thus it can be considered not optional in the user case. In a real configuration, however, the administrator will never wish to enable every policy module. It is recommended that the rest of this chapter be reviewed before any of this configuration is implemented. Users may change their label after the initial login; however, this change is subject constraints of the policy. The example above tells the Biba policy that a process's minimum integrity is 5, its maximum is 15, but the default effective label is 10. The process will run at 10 until it chooses to change label, perhaps due to the user using the setpmac command, which will be constrained by Biba to the range set at login. In all cases, after a change to login.conf, the login class capability database must be rebuilt using cap_mkdb and this will be reflected throughout every forthcoming example or discussion. It is useful to note that many sites may have a particularly large number of users requiring several different user classes. In depth planning is required as this may get extremely difficult to manage. Labels may also be set on network interfaces to help control the flow of data across the network. In all cases they function in the same way the policies function with respect to objects. Users at high settings in biba, for example, will not be permitted to access network interfaces with a label of low. The maclabel may be passed to ifconfig when setting the MAC label on network interfaces. For example: &prompt.root; ifconfig bge0 maclabel biba/equal will set the MAC label of biba/equal on the &man.bge.4; interface. When using a setting similar to biba/high(low-high) the entire label should be quoted; otherwise an error will be returned. Each policy module which supports labeling has a tunable which may be used to disable the MAC label on network interfaces. Setting the label to equal will have a similar effect. Review the output from sysctl, the policy manual pages, or even the information found later in this chapter for those tunables. By default the system will use the singlelabel option. But what does this mean to the administrator? There are several differences which, in their own right, offer pros and cons to the flexibility in the systems security model. The singlelabel only permits for one label, for instance biba/high to be used for each subject or object. It provides for lower administration overhead but decreases the flexibility of policies which support labeling. Many administrators may want to use the multilabel option in their security policy. The multilabel option will permit each subject or object to have its own independent MAC label in place of the standard singlelabel option which will allow only one label throughout the partition. The multilabel and single label options are only required for the policies which implement the labeling feature, including the Biba, Lomac, MLS and SEBSD policies. In many cases, the multilabel may not need to be set at all. Consider the following situation and security model: &os; web-server using the MAC framework and a mix of the various policies. This machine only requires one label, biba/high, for everything in the system. Here the file system would not require the multilabel option as a single label will always be in effect. But, this machine will be a web server and should have the web server run at biba/low to prevent write up capabilities. The Biba policy and how it works will be discussed later, so if the previous comment was difficult to interpret just continue reading and return. The server could use a separate partition set at biba/low for most if not all of its runtime state. Much is lacking from this example, for instance the restrictions on data, configuration and user settings; however, this is just a quick example to prove the aforementioned point. If any of the non-labeling policies are to be used, then the multilabel option would never be required. These include the seeotheruids, portacl and partition policies. It should also be noted that using multilabel with a partition and establishing a security model based on multilabel functionality could open the doors for higher administrative overhead as everything in the file system would have a label. This includes directories, files, and even device nodes. The following command will set multilabel on the file systems to have multiple labels. This may only be done in single user mode: &prompt.root; tunefs -l enable / This is not a requirement for the swap file system. Some users have experienced problems with setting the multilabel flag on the root partition. If this is the case, please review the of this chapter. Whenever a new technology is implemented, a planning phase is always a good idea. During the planning stages, an administrator should in general look at the big picture, trying to keep in view at least the following: The implementation requirements; The implementation goals; For MAC installations, these include: How to classify information and resources available on the target systems. What sorts of information or resources to restrict access to along with the type of restrictions that should be applied. Which MAC module or modules will be required to achieve this goal. It is always possible to reconfigure and change the system resources and security settings, it is quite often very inconvenient to search through the system and fix existing files and user accounts. Planning helps to ensure a trouble-free and efficient trusted system implementation. A trial run of the trusted system, including the configuration, is often vital and definitely beneficial before a MAC implementation is used on production systems. The idea of just letting loose on a system with MAC is like setting up for failure. Different environments may have explicit needs and requirements. Establishing an in depth and complete security profile will decrease the need of changes once the system goes live. As such, the future sections will cover the different modules available to administrators; describe their use and configuration; and in some cases provide insight on what situations they would be most suitable for. For instance, a web server might roll out the &man.mac.biba.4; and &man.mac.bsdextended.4; policies. In other cases, a machine with very few local users, the &man.mac.partition.4; might be a good choice. Every module included with the MAC framework may be either compiled into the kernel as noted above or loaded as a run-time kernel module. The recommended method is to add the module name to the /boot/loader.conf file so that it will load during the initial boot operation. The following sections will discuss the various MAC modules and cover their features. Implementing them into a specific environment will also be a consideration of this chapter. Some modules support the use of labeling, which is controlling access by enforcing a label such as this is allowed and this is not. A label configuration file may control how files may be accessed, network communication can be exchanged, and more. The previous section showed how the multilabel flag could be set on file systems to enable per-file or per-partition access control. A single label configuration would enforce only one label across the system, that is why the tunefs option is called multilabel. MAC See Other UIDs Policy Module name: mac_seeotheruids.ko Kernel configuration line: options MAC_SEEOTHERUIDS Boot option: mac_seeotheruids_load="YES" The &man.mac.seeotheruids.4; module mimics and extends the security.bsd.see_other_uids and security.bsd.see_other_gids sysctl tunables. This option does not require any labels to be set before configuration and can operate transparently with the other modules. After loading the module, the following sysctl tunables may be used to control the features: security.mac.seeotheruids.enabled will enable the module's features and use the default settings. These default settings will deny users the ability to view processes and sockets owned by other users. security.mac.seeotheruids.specificgid_enabled will allow a certain group to be exempt from this policy. To exempt specific groups from this policy, use the security.mac.seeotheruids.specificgid=XXX sysctl tunable. In the above example, the XXX should be replaced with the numeric group ID to be exempted. security.mac.seeotheruids.primarygroup_enabled is used to exempt specific primary groups from this policy. When using this tunable, the security.mac.seeotheruids.specificgid_enabled may not be set. MAC File System Firewall Policy Module name: mac_bsdextended.ko Kernel configuration line: options MAC_BSDEXTENDED Boot option: mac_bsdextended_load="YES" The &man.mac.bsdextended.4; module enforces the file system firewall. This module's policy provides an extension to the standard file system permissions model, permitting an administrator to create a firewall-like ruleset to protect files, utilities, and directories in the file system hierarchy. When access to a file system object is attempted, the list of rules is iterated until either a matching rule is located or the end is reached. This behavior may be changed by the use of a &man.sysctl.8; parameter, security.mac.bsdextended.firstmatch_enabled. Similar to other firewall modules in &os;, a file containing access control rules can be created and read by the system at boot time using an &man.rc.conf.5; variable. The rule list may be entered using a utility, &man.ugidfw.8;, that has a syntax similar to that of &man.ipfw.8;. More tools can be written by using the functions in the &man.libugidfw.3; library. Extreme caution should be taken when working with this module; incorrect use could block access to certain parts of the file system. After the &man.mac.bsdextended.4; module has been loaded, the following command may be used to list the current rule configuration: &prompt.root; ugidfw list 0 slots, 0 rules As expected, there are no rules defined. This means that everything is still completely accessible. To create a rule which will block all access by users but leave root unaffected, simply run the following command: &prompt.root; ugidfw add subject not uid root new object not uid root mode n This is a very bad idea as it will block all users from issuing even the most simple commands, such as ls. A more patriotic list of rules might be: &prompt.root; ugidfw set 2 subject uid user1 object uid user2 mode n &prompt.root; ugidfw set 3 subject uid user1 object gid user2 mode n This will block any and all access, including directory listings, to user2's home directory from the username user1. In place of user1, the not uid user2 could be passed. This will enforce the same access restrictions above for all users in place of just one user. The root user will be unaffected by these changes. This should provide a general idea of how the &man.mac.bsdextended.4; module may be used to help fortify a file system. For more information, see the &man.mac.bsdextended.4; and the &man.ugidfw.8; manual pages. MAC Interface Silencing Policy Module name: mac_ifoff.ko Kernel configuration line: options MAC_IFOFF Boot option: mac_ifoff_load="YES" The &man.mac.ifoff.4; module exists solely to disable network interfaces on the fly and keep network interfaces from being brought up during the initial system boot. It does not require any labels to be set up on the system, nor does it have a dependency on other MAC modules. Most of the control is done through the sysctl tunables listed below. security.mac.ifoff.lo_enabled will enable/disable all traffic on the loopback (&man.lo.4;) interface. security.mac.ifoff.bpfrecv_enabled will enable/disable all traffic on the Berkeley Packet Filter interface (&man.bpf.4;) security.mac.ifoff.other_enabled will enable/disable traffic on all other interfaces. One of the most common uses of &man.mac.ifoff.4; is network monitoring in an environment where network traffic should not be permitted during the boot sequence. Another suggested use would be to write a script which uses security/aide to automatically block network traffic if it finds new or altered files in protected directories. MAC Port Access Control List Policy Module name: mac_portacl.ko Kernel configuration line: MAC_PORTACL Boot option: mac_portacl_load="YES" The &man.mac.portacl.4; module is used to limit binding to local TCP and UDP ports using a variety of sysctl variables. In essence &man.mac.portacl.4; makes it possible to allow non-root users to bind to specified privileged ports, i.e., ports below 1024. Once loaded, this module will enable the MAC policy on all sockets. The following tunables are available: security.mac.portacl.enabled will enable/disable the policy completely. security.mac.portacl.port_high will set the highest port number that &man.mac.portacl.4; will enable protection for. security.mac.portacl.suser_exempt will, when set to a non-zero value, exempt the root user from this policy. security.mac.portacl.rules will specify the actual mac_portacl policy; see below. The actual mac_portacl policy, as specified in the security.mac.portacl.rules sysctl, is a text string of the form: rule[,rule,...] with as many rules as needed. Each rule is of the form: idtype:id:protocol:port. The idtype parameter can be uid or gid and used to interpret the id parameter as either a user id or group id, respectively. The protocol parameter is used to determine if the rule should apply to TCP or UDP by setting the parameter to tcp or udp. The final port parameter is the port number to allow the specified user or group to bind to. Since the ruleset is interpreted directly by the kernel only numeric values can be used for the user ID, group ID, and port parameters. Names cannot be used for users, groups, or services. By default, on &unix;-like systems, ports below 1024 can only be used by/bound to privileged processes, i.e., those run as root. For &man.mac.portacl.4; to allow non-privileged processes to bind to ports below 1024 this standard &unix; restriction has to be disabled. This can be accomplished by setting the &man.sysctl.8; variables net.inet.ip.portrange.reservedlow and net.inet.ip.portrange.reservedhigh to zero. See the examples below or review the &man.mac.portacl.4; manual page for further information. The following examples should illuminate the above discussion a little better: &prompt.root; sysctl security.mac.portacl.port_high=1023 &prompt.root; sysctl net.inet.ip.portrange.reservedlow=0 net.inet.ip.portrange.reservedhigh=0 First we set &man.mac.portacl.4; to cover the standard privileged ports and disable the normal &unix; bind restrictions. &prompt.root; sysctl security.mac.portacl.suser_exempt=1 The root user should not be crippled by this policy, thus set the security.mac.portacl.suser_exempt to a non-zero value. The &man.mac.portacl.4; module has now been set up to behave the same way &unix;-like systems behave by default. &prompt.root; sysctl security.mac.portacl.rules=uid:80:tcp:80 Allow the user with UID 80 (normally the www user) to bind to port 80. This can be used to allow the www user to run a web server without ever having root privilege. &prompt.root; sysctl security.mac.portacl.rules=uid:1001:tcp:110,uid:1001:tcp:995 Permit the user with the UID of 1001 to bind to the TCP ports 110 (pop3) and 995 (pop3s). This will permit this user to start a server that accepts connections on ports 110 and 995. MAC Process Partition Policy Module name: mac_partition.ko Kernel configuration line: options MAC_PARTITION Boot option: mac_partition_load="YES" The &man.mac.partition.4; policy will drop processes into specific partitions based on their MAC label. Think of it as a special type of &man.jail.8;, though that is hardly a worthy comparison. This is one module that should be added to the &man.loader.conf.5; file so that it loads and enables the policy during the boot process. Most configuration for this policy is done using the &man.setpmac.8; utility which will be explained below. The following sysctl tunable is available for this policy: security.mac.partition.enabled will enable the enforcement of MAC process partitions. When this policy is enabled, users will only be permitted to see their processes, and any others within their partition, but will not be permitted to work with utilities outside the scope of this partition. For instance, a user in the insecure class above will not be permitted to access the top command as well as many other commands that must spawn a process. To set or drop utilities into a partition label, use the setpmac utility: &prompt.root; setpmac partition/13 top This will add the top command to the label set on users in the insecure class. Note that all processes spawned by users in the insecure class will stay in the partition/13 label. The following command will show you the partition label and the process list: &prompt.root; ps Zax This next command will allow the viewing of another user's process partition label and that user's currently running processes: &prompt.root; ps -ZU trhodes Users can see processes in root's label unless the &man.mac.seeotheruids.4; policy is loaded. A really crafty implementation could have all of the services disabled in /etc/rc.conf and started by a script that starts them with the proper labeling set. The following policies support integer settings in place of the three default labels offered. These options, including their limitations, are further explained in the module manual pages. MAC Multi-Level Security Policy Module name: mac_mls.ko Kernel configuration line: options MAC_MLS Boot option: mac_mls_load="YES" The &man.mac.mls.4; policy controls access between subjects and objects in the system by enforcing a strict information flow policy. In MLS environments, a clearance level is set in each subject or objects label, along with compartments. Since these clearance or sensibility levels can reach numbers greater than six thousand; it would be a daunting task for any system administrator to thoroughly configure each subject or object. Thankfully, three instant labels are already included in this policy. These labels are mls/low, mls/equal and mls/high. Since these labels are described in depth in the manual page, they will only get a brief description here: The mls/low label contains a low configuration which permits it to be dominated by all other objects. Anything labeled with mls/low will have a low clearance level and not be permitted to access information of a higher level. In addition, this label will prevent objects of a higher clearance level from writing or passing information on to them. The mls/equal label should be placed on objects considered to be exempt from the policy. The mls/high label is the highest level of clearance possible. Objects assigned this label will hold dominance over all other objects in the system; however, they will not permit the leaking of information to objects of a lower class. MLS provides for: A hierarchical security level with a set of non hierarchical categories; Fixed rules: no read up, no write down (a subject can have read access to objects on its own level or below, but not above. Similarly, a subject can have write access to objects on its own level or above but not beneath.); Secrecy (preventing inappropriate disclosure of data); Basis for the design of systems that concurrently handle data at multiple sensitivity levels (without leaking information between secret and confidential). The following sysctl tunables are available for the configuration of special services and interfaces: security.mac.mls.enabled is used to enable/disable the MLS policy. security.mac.mls.ptys_equal will label all &man.pty.4; devices as mls/equal during creation. security.mac.mls.revocation_enabled is used to revoke access to objects after their label changes to a label of a lower grade. security.mac.mls.max_compartments is used to set the maximum number of compartment levels with objects; basically the maximum compartment number allowed on a system. To manipulate the MLS labels, the &man.setfmac.8; command has been provided. To assign a label to an object, issue the following command: &prompt.root; setfmac mls/5 test To get the MLS label for the file test issue the following command: &prompt.root; getfmac test This is a summary of the MLS policy's features. Another approach is to create a master policy file in /etc which specifies the MLS policy information and to feed that file into the setfmac command. This method will be explained after all policies are covered. With the Multi-Level Security Policy Module, an administrator plans for controlling the flow of sensitive information. By default, with its block read up block write down nature, the system defaults everything to a low state. Everything is accessible and an administrator slowly changes this during the configuration stage; augmenting the confidentiality of the information. Beyond the three basic label options above, an administrator may group users and groups as required to block the information flow between them. It might be easier to look at the information in clearance levels familiarized with words, for instance classifications such as Confidential, Secret, and Top Secret. Some administrators might just create different groups based on project levels. Regardless of classification method, a well thought out plan must exist before implementing such a restrictive policy. Some example situations for this security policy module could be an e-commerce web server, a file server holding critical company information, and financial institution environments. The most unlikely place would be a personal workstation with only two or three users. MAC Biba Integrity Policy Module name: mac_biba.ko Kernel configuration line: options MAC_BIBA Boot option: mac_biba_load="YES" The &man.mac.biba.4; module loads the MAC Biba policy. This policy works much like that of the MLS policy with the exception that the rules for information flow are slightly reversed. This is said to prevent the downward flow of sensitive information whereas the MLS policy prevents the upward flow of sensitive information; thus, much of this section can apply to both policies. In Biba environments, an integrity label is set on each subject or object. These labels are made up of hierarchal grades, and non-hierarchal components. As an object's or subject's grade ascends, so does its integrity. Supported labels are biba/low, biba/equal, and biba/high; as explained below: The biba/low label is considered the lowest integrity an object or subject may have. Setting this on objects or subjects will block their write access to objects or subjects marked high. They still have read access though. The biba/equal label should only be placed on objects considered to be exempt from the policy. The biba/high label will permit writing to objects set at a lower label, but not permit reading that object. It is recommended that this label be placed on objects that affect the integrity of the entire system. Biba provides for: Hierarchical integrity level with a set of non hierarchical integrity categories; Fixed rules: no write up, no read down (opposite of MLS). A subject can have write access to objects on its own level or below, but not above. Similarly, a subject can have read access to objects on its own level or above, but not below; Integrity (preventing inappropriate modification of data); Integrity levels (instead of MLS sensitivity levels). The following sysctl tunables can be used to manipulate the Biba policy. security.mac.biba.enabled may be used to enable/disable enforcement of the Biba policy on the target machine. security.mac.biba.ptys_equal may be used to disable the Biba policy on &man.pty.4; devices. security.mac.biba.revocation_enabled will force the revocation of access to objects if the label is changed to dominate the subject. To access the Biba policy setting on system objects, use the setfmac and getfmac commands: &prompt.root; setfmac biba/low test &prompt.root; getfmac test test: biba/low Integrity, different from sensitivity, guarantees that the information will never be manipulated by untrusted parties. This includes information passed between subjects, objects, and both. It ensures that users will only be able to modify and in some cases even access information they explicitly need to. The &man.mac.biba.4; security policy module permits an administrator to address which files and programs a user or users may see and invoke while assuring that the programs and files are free from threats and trusted by the system for that user, or group of users. During the initial planning phase, an administrator must be prepared to partition users into grades, levels, and areas. Users will be blocked access not only to data but programs and utilities both before and after they start. The system will default to a high label once this policy module is enabled, and it is up to the administrator to configure the different grades and levels for users. Instead of using clearance levels as described above, a good planning method could include topics. For instance, only allow developers modification access to the source code repository, source code compiler, and other development utilities. While other users would be grouped into other categories such as testers, designers, or just ordinary users and would only be permitted read access. With its natural security control, a lower integrity subject is unable to write to a higher integrity subject; a higher integrity subject cannot observe or read a lower integrity object. Setting a label at the lowest possible grade could make it inaccessible to subjects. Some prospective environments for this security policy module would include a constrained web server, development and test machine, and source code repository. A less useful implementation would be a personal workstation, a machine used as a router, or a network firewall. MAC LOMAC Module name: mac_lomac.ko Kernel configuration line: options MAC_LOMAC Boot option: mac_lomac_load="YES" Unlike the MAC Biba policy, the &man.mac.lomac.4; policy permits access to lower integrity objects only after decreasing the integrity level to not disrupt any integrity rules. The MAC version of the Low-watermark integrity policy, not to be confused with the older &man.lomac.4; implementation, works almost identically to Biba, but with the exception of using floating labels to support subject demotion via an auxiliary grade compartment. This secondary compartment takes the form of [auxgrade]. When assigning a lomac policy with an auxiliary grade, it should look a little bit like: lomac/10[2] where the number two (2) is the auxiliary grade. The MAC LOMAC policy relies on the ubiquitous labeling of all system objects with integrity labels, permitting subjects to read from low integrity objects and then downgrading the label on the subject to prevent future writes to high integrity objects. This is the [auxgrade] option discussed above, thus the policy may provide for greater compatibility and require less initial configuration than Biba. Like the Biba and MLS policies; the setfmac and setpmac utilities may be used to place labels on system objects: &prompt.root; setfmac /usr/home/trhodes lomac/high[low] &prompt.root; getfmac /usr/home/trhodes lomac/high[low] Notice the auxiliary grade here is low, this is a feature provided only by the MAC LOMAC policy. Nagios in a MAC Jail The following demonstration will implement a secure environment using various MAC modules with properly configured policies. This is only a test and should not be considered the complete answer to everyone's security woes. Just implementing a policy and ignoring it never works and could be disastrous in a production environment. Before beginning this process, the multilabel option must be set on each file system as stated at the beginning of this chapter. Not doing so will result in errors. While at it, ensure that the net-mngt/nagios-plugins, net-mngt/nagios, and www/apache22 ports are all installed, configured, and working correctly. Begin the procedure by adding the following user class to the /etc/login.conf file: insecure:\ :copyright=/etc/COPYRIGHT:\ :welcome=/etc/motd:\ :setenv=MAIL=/var/mail/$,BLOCKSIZE=K:\ :path=~/bin:/sbin:/bin:/usr/sbin:/usr/bin:/usr/local/sbin:/usr/local/bin :manpath=/usr/share/man /usr/local/man:\ :nologin=/usr/sbin/nologin:\ :cputime=1h30m:\ :datasize=8M:\ :vmemoryuse=100M:\ :stacksize=2M:\ :memorylocked=4M:\ :memoryuse=8M:\ :filesize=8M:\ :coredumpsize=8M:\ :openfiles=24:\ :maxproc=32:\ :priority=0:\ :requirehome:\ :passwordtime=91d:\ :umask=022:\ :ignoretime@:\ :label=biba/10(10-10): And adding the following line to the default user class: :label=biba/high: Once this is completed, the following command must be issued to rebuild the database: &prompt.root; cap_mkdb /etc/login.conf Do not reboot yet, just add the following lines to /boot/loader.conf so the required modules will load during system initialization: mac_biba_load="YES" mac_seeotheruids_load="YES" Set the root user to the default class using: &prompt.root; pw usermod root -L default All user accounts that are not root or system users will now require a login class. The login class is required otherwise users will be refused access to common commands such as &man.vi.1;. The following sh script should do the trick: &prompt.root; for x in `awk -F: '($3 >= 1001) && ($3 != 65534) { print $1 }' \ /etc/passwd`; do pw usermod $x -L default; done; Drop the nagios and www users into the insecure class: &prompt.root; pw usermod nagios -L insecure &prompt.root; pw usermod www -L insecure A contexts file should now be created; the following example file should be placed in /etc/policy.contexts. # This is the default BIBA policy for this system. # System: /var/run biba/equal /var/run/* biba/equal /dev biba/equal /dev/* biba/equal /var biba/equal /var/spool biba/equal /var/spool/* biba/equal /var/log biba/equal /var/log/* biba/equal /tmp biba/equal /tmp/* biba/equal /var/tmp biba/equal /var/tmp/* biba/equal /var/spool/mqueue biba/equal /var/spool/clientmqueue biba/equal # For Nagios: /usr/local/etc/nagios /usr/local/etc/nagios/* biba/10 /var/spool/nagios biba/10 /var/spool/nagios/* biba/10 # For apache /usr/local/etc/apache biba/10 /usr/local/etc/apache/* biba/10 This policy will enforce security by setting restrictions on the flow of information. In this specific configuration, users, root and others, should never be allowed to access Nagios. Configuration files and processes that are a part of Nagios will be completely self contained or jailed. This file may now be read into our system by issuing the following command: &prompt.root; setfsmac -ef /etc/policy.contexts / &prompt.root; setfsmac -ef /etc/policy.contexts / The above file system layout may be different depending on environment; however, it must be run on every single file system. The /etc/mac.conf file requires the following modifications in the main section: default_labels file ?biba default_labels ifnet ?biba default_labels process ?biba default_labels socket ?biba Add the following line to /boot/loader.conf: security.mac.biba.trust_all_interfaces=1 And the following to the network card configuration stored in rc.conf. If the primary Internet configuration is done via DHCP, this may need to be configured manually after every system boot: maclabel biba/equal MAC Configuration Testing Ensure that the web server and Nagios will not be started on system initialization, and reboot. Ensure the root user cannot access any of the files in the Nagios configuration directory. If root can issue an &man.ls.1; command on /var/spool/nagios, then something is wrong. Otherwise a permission denied error should be returned. If all seems well, Nagios, Apache, and Sendmail can now be started in a way fitting of the security policy. The following commands will make this happen: &prompt.root; cd /etc/mail && make stop && \ setpmac biba/equal make start && setpmac biba/10\(10-10\) apachectl start && \ setpmac biba/10\(10-10\) /usr/local/etc/rc.d/nagios.sh forcestart Double check to ensure that everything is working properly. If not, check the log files or error messages. Use the &man.sysctl.8; utility to disable the &man.mac.biba.4; security policy module enforcement and try starting everything again, like normal. The root user can change the security enforcement and edit the configuration files without fear. The following command will permit the degradation of the security policy to a lower grade for a newly spawned shell: &prompt.root; setpmac biba/10 csh To block this from happening, force the user into a range via &man.login.conf.5;. If &man.setpmac.8; attempts to run a command outside of the compartment's range, an error will be returned and the command will not be executed. In this case, setting root to biba/high(high-high). This example considers a relatively small, fewer than fifty users, storage system. Users would have login capabilities, and be permitted to not only store data but access resources as well. For this scenario, the &man.mac.bsdextended.4; mixed with &man.mac.seeotheruids.4; could co-exist and block access not only to system objects, but to hide user processes as well. Begin by adding the following line to /boot/loader.conf: mac_seeotheruids_load="YES" The &man.mac.bsdextended.4; security policy module may be activated through the use of the following rc.conf variable: ugidfw_enable="YES" Default rules stored in /etc/rc.bsdextended will be loaded at system initialization; however, the default entries may need modification. Since this machine is expected only to service users, everything may be left commented out except the last two. These will force the loading of user owned system objects by default. Add the required users to this machine and reboot. For testing purposes, try logging in as a different user across two consoles. Run the ps aux command to see if processes of other users are visible. Try to run &man.ls.1; on another users home directory, it should fail. Do not try to test with the root user unless the specific sysctls have been modified to block super user access. When a new user is added, their &man.mac.bsdextended.4; rule will not be in the ruleset list. To update the ruleset quickly, simply unload the security policy module and reload it again using the &man.kldunload.8; and &man.kldload.8; utilities. MAC Troubleshooting During the development stage, a few users reported problems with normal configuration. Some of these problems are listed below: The multilabel flag does not stay enabled on my root (/) partition! It seems that one out of every fifty users has this problem, indeed, we had this problem during our initial configuration. Further observation of this so called bug has lead me to believe that it is a result of either incorrect documentation or misinterpretation of the documentation. Regardless of why it happened, the following steps may be taken to resolve it: Edit /etc/fstab and set the root partition at ro for read-only. Reboot into single user mode. Run tunefs -l enable on /. Reboot the system into normal mode. Run mount -urw / and change the ro back to rw in /etc/fstab and reboot the system again. Double-check the output from the mount to ensure that multilabel has been properly set on the root file system. After establishing a secure environment with MAC, I am no longer able to start X! This could be caused by the MAC partition policy or by a mislabeling in one of the MAC labeling policies. To debug, try the following: Check the error message; if the user is in the insecure class, the partition policy may be the culprit. Try setting the user's class back to the default class and rebuild the database with the cap_mkdb command. If this does not alleviate the problem, go to step two. Double-check the label policies. Ensure that the policies are set correctly for the user in question, the X11 application, and the /dev entries. If neither of these resolve the problem, send the error message and a description of your environment to the TrustedBSD discussion lists located at the TrustedBSD website or to the &a.questions; mailing list. When I attempt to switch from the root user to another user in the system, the error message _secure_path: unable to state .login_conf appears. This message is usually shown when the user has a higher label setting then that of the user whom they are attempting to become. For instance a user on the system, joe, has a default label of biba/low. The root user, who has a label of biba/high, cannot view joe's home directory. This will happen regardless if root has used the su command to become joe, or not. In this scenario, the Biba integrity model will not permit root to view objects set at a lower integrity level. In normal or even single user mode, the root is not recognized. The whoami command returns 0 (zero) and su returns who are you?. What could be going on? This can happen if a labeling policy has been disabled, either by a &man.sysctl.8; or the policy module was unloaded. If the policy is being disabled or has been temporarily disabled, then the login capabilities database needs to be reconfigured with the label option being removed. Double check the login.conf file to ensure that all label options have been removed and rebuild the database with the cap_mkdb command. This may also happen if a policy restricts access to the master.passwd file or database. Usually caused by an administrator altering the file under a label which conflicts with the general policy being used by the system. In these cases, the user information would be read by the system and access would be blocked as the file has inherited the new label. Disable the policy via a &man.sysctl.8; and everything should return to normal.