This chapter discusses many of the issues to consider when configuring and administering a cell, and directs you to detailed related information available elsewhere in this guide. It is assumed you are already familiar with the material in An Overview of OpenAFS Administration. It is best to read this chapter before installing your cell's first file server machine or performing any other administrative task. AFS differences from UNIX summarized UNIX differences from AFS summarized differences between AFS and UNIX, summarized AFS behaves like a standard UNIX file system in most respects, while also making file sharing easy within and between cells. This section describes some differences between AFS and the UNIX file system, referring you to more detailed information as appropriate. protection AFS compared to UNIX AFS augments the standard UNIX file protection mechanism in two ways: it associates an access control list (ACL) with each directory, and it enables users to define a large number of their own groups, which can be placed on ACLs. AFS uses ACLs to protect files and directories, rather than relying exclusively on the mode bits. This has several implications, which are discussed further in the indicated sections: AFS ACLs use seven access permissions rather than the three UNIX mode bits. See The AFS ACL Permissions. For directories, AFS ignores the UNIX mode bits. For files, AFS uses only the first set of mode bits (the owner bits), and their meaning interacts with permissions on the directory's ACL. See How AFS Interprets the UNIX Mode Bits. A directory's ACL protects all of the files in a directory in the same manner. To apply a more restrictive set of AFS permissions to certain file, place it in directory with a different ACL. If a directory must contain files with different permissions, use symbolic links to point to files stored in directories with different ACLs. Moving a file to a different directory changes its protection. See Differences Between UFS and AFS Data Protection. An ACL can include about 20 entries granting different combinations of permissions to different users or groups, rather than only the three UNIX entities represented by the three sets of mode bits. See Differences Between UFS and AFS Data Protection. You can designate an AFS file as write-only as in the UNIX file system, by setting only the w (write) mode bit. You cannot designate an AFS directory as write-only, because AFS ignores the mode bits on a directory. See How AFS Interprets the UNIX Mode Bits. AFS enables users to create groups and add other users to those groups. Placing these groups on ACLs extends the same permissions to a number of exactly specified users at the same time, which is much more convenient than placing the individuals on the ACLs directly. See Administering the Protection Database. There are also system-defined groups, system:anyuser and system:authuser, whose presence on an ACL extends access to a wide range of users at once. See The System Groups and Using Groups on ACLs. authentication AFS compared to UNIX password AFS compared to UNIX Just as the AFS filespace is distinct from each machine's local file system, AFS authentication is separate from local login. This has two practical implications, which will already be familiar to users and system administrators who use Kerberos for authentication. To access AFS files, users must log into the local machine as normal, obtain Kerberos tickets, and then obtain AFS tokens. This process can often be automated through the system authentication configuration so that the user logs into the system as normal and obtains Kerberos tickets and AFS tokens transparently. If you cannot or chose not to configure the system this way, your users must login and authenticate in separate steps, as detailed in the OpenAFS User Guide. Passwords may be stored in two separate places: the Kerberos KDC and, optionally, each machine's local user database (/etc/passwd or equivalent) for the local system. A user's passwords in the two places can differ if desired. This section summarizes how AFS modifies the functionality of some UNIX commands. The chmod command chmod command AFS compared to UNIX commands chmod (AFS compared to UNIX) setuid programs setting mode bits Only members of the system:administrators group can use this command to turn on the setuid, setgid or sticky mode bits on AFS files. For more information, see Determining if a Client Can Run Setuid Programs. The chown command chown command AFS compared to UNIX commands chown (AFS compared to UNIX) Only members of the system:administrators group can issue this command on AFS files. The chgrp command chgrp command AFS compared to UNIX commands chgrp (AFS compared to UNIX) Only members of the system:administrators can issue this command on AFS files and directories. The groups and id commands groups command AFS compared to UNIX id command AFS compared to UNIX commands groups (AFS compared to UNIX) commands id (AFS compared to UNIX) If the user's AFS tokens are associated with a process authentication group (PAG), the output of these commands may include one or two large numbers. These are artificial groups used by the OpenAFS Cache Manager to track the PAG on some platforms. Other platforms may use other methods, such as native kernel support for a PAG or a similar concept, in which case the large GIDs may not appear. To learn about PAGs, see Identifying AFS Tokens by PAG. The ln command ln command AFS compared to UNIX commands ln (AFS compared to UNIX) This command cannot create hard links between files in different AFS directories. See Creating Hard Links. The sshd daemon and ssh command sshd command AFS compared to UNIX commands sshd (AFS compared to UNIX) ssh command AFS compared to UNIX commands ssh (AFS compared to UNIX) In order for a user to have access to files stored in AFS, that user needs to have Kerberos tickets and an AFS token on the system from which they're accessing AFS. This has an implication for users who log in remotely via protocols such as Secure Shell (SSH): that log-in process must create local Kerberos tickets and an AFS token on the system, or the user will have to separately authenticate to Kerberos and AFS after logging in. The OpenSSH project provides an SSH client and server that uses the GSS-API protocol to pass Kerberos tickets between machines. With a suitable SSH client, this allows users to delegate their Kerberos tickets to the remote machine, and that machine to store those tickets and obtain AFS tokens as part of the log-in process. fsck command AFS compared to UNIX file server machine inode-based file server machine namei-based namei definition commands fsck (AFS compared to UNIX) fsck command AFS version commands fsck (AFS version) directories lost+found lost+found directory The fileserver uses either of two formats for storing data on disk. The inode-based format uses a combination of regular files and extra fields stored in the inode data structures that are normally reserved for use by the operating system. The namei format uses normal file storage and does not use special structures. The choice of storage formats is chosen at compile time and the two formats are incompatible. The inode format is only available on certain platforms. The storage format must be consistent for the fileserver binaries and all vice partitions on a given file server machine. This section on fsck advice only applies to the inode-based fileserver binaries. On servers using namei-based binaries, the vendor-supplied fsck can be used as normal. If you are using AFS fileserver binaries compiled with the inode-based format, never run the standard UNIX fsck command on an AFS file server machine. It does not understand how the File Server organizes volume data on disk, and so moves all AFS data into the lost+found directory on the partition. Instead, use the version of the fsck program that is included in the AFS distribution. The OpenAFS Quick Start Guide explains how to replace the vendor-supplied fsck program with the AFS version as you install each server machine. The AFS version functions like the standard fsck program on data stored on both UFS and AFS partitions. The appearance of a banner like the following as the fsck program initializes confirms that you are running the correct one: --- AFS (R) version fsck--- where version is the AFS version. For correct results, it must match the AFS version of the server binaries in use on the machine. If you ever accidentally run the standard version of the program, contact your AFS support provider, contact the OpenAFS mailing lists, or refer to the OpenAFS support web page for support options. It is sometimes possible to recover volume data from the lost+found directory. If the data is not recoverabled, then restoring from backup is recommended. Running the fsck binary supplied by the operating system vendor on an fileserver using inode-based file storage will result in data corruption! hard link AFS restrictions on restrictions on hard links in AFS AFS does not allow hard links (created with the UNIX ln command) between files that reside in different directories, because in that case it is unclear which of the directory's ACLs to associate with the link. AFS also does not allow hard links to directories, in order to keep the file system organized as a tree. It is possible to create symbolic links (with the UNIX ln -s command) between elements in two different AFS directories, or even between an element in AFS and one in a machine's local UNIX file system. Do not create a symbolic link in AFS to a file whose name begins with either a number sign (#) or a percent sign (%), however. The Cache Manager interprets such links as a mount point to a regular or read/write volume, respectively. fsync system call for files saved on AFS client close system call for files saved on AFS client write system call for files saved on AFS client When an application issues the UNIX close system call on a file, the Cache Manager performs a synchronous write of the data to the File Server that maintains the central copy of the file. It does not return control to the application until the File Server has acknowledged receipt of the data. For the fsync system call, control does not return to the application until the File Server indicates that it has written the data to non-volatile storage on the file server machine. When an application issues the UNIX write system call, the Cache Manager writes modifications to the local AFS client cache only. If the local machine crashes or an application program exits without issuing the close system call, it is possible that the modifications are not recorded in the central copy of the file maintained by the File Server. The Cache Manager does sometimes write this type of modified data from the cache to the File Server without receiving the close or fsync system call, such as when it needs to free cache chunks for new data. However, it is not generally possible to predict when the Cache Manager transfers modified data to the File Server in this way. The implication is that if an application's Save option invokes the write system call rather than close or fsync, the changes are not necessarily stored permanently on the File Server machine. Most application programs issue the close system call for save operations, as well as when they finish handling a file and when they exit. setuid programs restrictions on The UNIX setuid bit is ignored by default for programs run from AFS, but can be enabled by the system administrator on a client machine. The fs setcell command determines whether setuid programs that originate in a particular cell can run on a given client machine. Running setuid binaries from AFS poses a security risk due to weaknesses in the integrity checks of the AFS protocol and should normally not be permitted. See Determining if a Client Can Run Setuid Programs. Set the UNIX setuid bit only for files whose owner is UID 0 (the local superuser root). This does not present an automatic security risk: the local superuser has no special privilege in AFS, but only in the local machine's UNIX file system and kernel. Setting the UNIX setuid bit for files owned with a different UID will have unpredictable resuilts, since that UID will be interpreted as possibly different users on each AFS client machine. Any file can be marked with the setuid bit, but only members of the system:administrators group can issue the chown system call or the chown command, or issue the chmod system call or the chmod command to set the setuid bit. cell name choosing choosing name cell conventions cell name Internet conventions for cell name This section explains how to choose a cell name and explains why choosing an appropriate cell name is important. Your cell name must distinguish your cell from all others in the AFS global namespace. By convention, the cell name is the second element in any AFS pathname; therefore, a unique cell name guarantees that every AFS pathname uniquely identifies a file, even if cells use the same directory names at lower levels in their local AFS filespace. For example, both the Example Corporation cell and the Example Organization cell can have a home directory for the user pat, because the pathnames are distinct: /afs/example.com/usr/pat and /afs/example.org/usr/pat. By convention, cell names follow the Domain Name System (DNS) conventions for domain names. If you are already an Internet site, then it is simplest and strongly recommended to choose your Internet domain name as the cell name. If you are not an Internet site, it is best to choose a unique DNS-style name, particularly if you plan to connect to the Internet in the future. There are a few constraints on AFS cell names: It can contain as many as 64 characters, but shorter names are better because the cell name frequently is part of machine and file names. If your cell name is long, you can reduce pathname length either by creating a symbolic link to the complete cell name, at the second level in your file tree or by using the CellAlias configuration file on a client machine. See The Second (Cellname) Level. To guarantee it is suitable for different operating system types, the cell name can contain only lowercase characters, numbers, underscores, dashes, and periods. Do not include command shell metacharacters. It can include any number of fields, which are conventionally separated by periods (see the examples below). setting cell name cell name setting server machine setting home cell client machine setting home cell The cell name is recorded in two files on the local disk of each file server and client machine. Among other functions, these files define the machine's cell membership and so affect how programs and processes run on the machine; see Why Choosing the Appropriate Cell Name is Important. The procedure for setting the cell name is different for the two types of machines. For file server machines, the two files that record the cell name are the /usr/afs/etc/ThisCell and /usr/afs/etc/CellServDB files. As described more explicitly in the OpenAFS Quick Start Guide, you set the cell name in both by issuing the bos setcellname command on the first file server machine you install in your cell. It is not usually necessary to issue the command again. If you use the Update Server, it distributes its copy of the ThisCell and CellServDB files to additional server machines that you install. If you do not use the Update Server, the OpenAFS Quick Start Guide explains how to copy the files manually. For client machines, the two files that record the cell name are the /usr/vice/etc/ThisCell and /usr/vice/etc/CellServDB files. You create these files on a per-client basis, either with a text editor or by copying them onto the machine from a central source in AFS. See Maintaining Knowledge of Database Server Machines for details. Change the cell name in these files only when you want to transfer the machine to a different cell (client machines can only have one default cell at a time and server machines can only belong to one cell at a time). If the machine is a file server, follow the complete set of instructions in the OpenAFS Quick Start Guide for configuring a new cell. If the machine is a client, all you need to do is change the files appropriately and reboot the machine. The next section explains further the negative consequences of changing the name of an existing cell. To set the default cell name used by most AFS commands without changing the local /usr/vice/etc/ThisCell file, set the AFSCELL environment variable in the command shell. It is worth setting this variable if you need to complete significant administrative work in a foreign cell. The fs checkservers and fs mkmount commands do not use the AFSCELL variable. The fs checkservers command always defaults to the cell named in the ThisCell file, unless the -cell argument is used. The fs mkmount command defaults to the cell in which the parent directory of the new mount point resides. ThisCell file (client) how used by programs Take care to select a cell name that is suitable for long-term use. Changing a cell name later is complicated. An appropriate cell name is important because it is the second element in the pathname of all files in a cell's file tree. Because each cell name is unique, its presence in an AFS pathname makes the pathname unique in the AFS global namespace, even if multiple cells use similar filespace organization at lower levels. For instance, it means that every cell can have a home directory called /afs/cellname/usr/pat without causing a conflict. The presence of the cell name in pathnames also means that users in every cell use the same pathname to access a file, whether the file resides in their local cell or in a foreign cell. Another reason to choose the correct cell name early in the process of installing your cell is that the cell membership defined in each machine's ThisCell file affects the performance of many programs and processes running on the machine. For instance, AFS commands (fs, pts, and vos commands, for example) by default execute in the cell of the machine on which they are issued. The command interpreters check the ThisCell file on the local disk and then contact the database server machines listed in the CellServDB file or configured in DNS for the indicated cell. (The bos commands work differently because the issuer always has to name of the machine on which to run the command.) The ThisCell file also normally determines the cell for which a user receives an AFS token when he or she logs in to a machine. If you change the cell name, you must change the ThisCell and CellServDB files on every server and client machine. Failure to change them all will cause many commands from the AFS suites to not work as expected. participation in AFS global namespace AFS global namespace global namespace Participating in the AFS global namespace makes your cell's local file tree visible to AFS users in foreign cells and makes other cells' file trees visible to your local users. It makes file sharing across cells just as easy as sharing within a cell. This section outlines the procedures necessary for participating in the global namespace. Participation in the global namespace is not mandatory. Some cells use AFS primarily to facilitate file sharing within the cell, and are not interested in providing their users with access to foreign cells. Making your file tree visible does not mean making it vulnerable. You control how foreign users access your cell using the same protection mechanisms that control local users' access. See Granting and Denying Foreign Users Access to Your Cell. The two aspects of participation are independent. A cell can make its file tree visible without allowing its users to see foreign cells' file trees, or can enable its users to see other file trees without advertising its own. You make your cell visible to others by advertising your database server machines and allowing users at other sites to access your database server and file server machines. See Making Your Cell Visible to Others. You control access to foreign cells on a per-client machine basis. In other words, it is possible to make a foreign cell accessible from one client machine in your cell but not another. See Making Other Cells Visible in Your Cell. conventions AFS pathnames AFS root directory (/afs) on client machine directories /afs directories /afs/cellname cell name at second level in file tree The AFS global namespace appears the same to all AFS cells that participate in it, because they all agree to follow a small set of conventions in constructing pathnames. The first convention is that all AFS pathnames begin with the string /afs to indicate that they belong to the AFS global namespace. The second convention is that the cell name is the second element in an AFS pathname; it indicates where the file resides (that is, the cell in which a file server machine houses the file). As noted, the presence of a cell name in pathnames makes the global namespace possible, because it guarantees that all AFS pathnames are unique even if cells use the same directory names at lower levels in their AFS filespace. What appears at the third and lower levels in an AFS pathname depends on how a cell has chosen to arrange its filespace. There are some suggested conventional directories at the third level; see The Third Level. cell making local visible to foreign local cell making visible to foreign cells foreign cell making local cell visible You make your cell visible to others by advertising your cell name and database server machines. Just like client machines in the local cell, the Cache Manager on machines in foreign cells use the information to reach your cell's Volume Location (VL) Servers when they need volume and file location information. For authenticated access, foreign clients must be configured with the necessary Kerberos version 5 domain-to-realm mappings and Key Distribution Center (KDC) location information for both the local and remote Kerberos version 5 realms. There are two places you can make this information available: files global CellServDB CellServDB file maintained by the AFS Registrar as global update source In the global CellServDB file maintained by the AFS Registrar. This file lists the name and database server machines of every cell that has agreed to make this information available to other cells. This file is available at http://grand.central.org/csdb.html To add or change your cell's listing in this file, follow the instructions at http://grand.central.org/csdb.html. It is a good policy to check the file for changes on a regular schedule. An updated copy of this file is included with new releases of OpenAFS. files CellServDB.local CellServDB.local file A file called CellServDB.local in the /afs/cellname/service/etc directory of your cell's filespace. List only your cell's database server machines. Update the files whenever you change the identity of your cell's database server machines. Also update the copies of the CellServDB files on all of your server machines (in the /usr/afs/etc directory) and client machines (in the /usr/vice/etc directory). For instructions, see Maintaining the Server CellServDB File and Maintaining Knowledge of Database Server Machines. Once you have advertised your database server machines, it can be difficult to make your cell invisible again. You can remove the CellServDB.local file and ask the AFS Registrar to remove your entry from the global CellServDB file, but other cells probably have an entry for your cell in their local CellServDB files already. To make those entries invalid, you must change the names or IP addresses of your database server machines. Your cell does not have to be invisible to be inaccessible, however. To make your cell completely inaccessible to foreign users, remove the system:anyuser group from all ACLs at the top three levels of your filespace; see Granting and Denying Foreign Users Access to Your Cell. cell making foreign visible to local local cell making foreign cells visible in foreign cell making visible in local cell client machine making foreign cell visible To make a foreign cell's filespace visible on a client machine in your cell that is not configured for Freelance Mode or Dynamic Root mode, perform the following three steps: Mount the cell's root.cell volume at the second level in your cell's filespace just below the /afs directory. Use the fs mkmount command with the -cell argument as instructed in To create a cellular mount point. Mount AFS at the /afs directory on the client machine. The afsd program, which initializes the Cache Manager, performs the mount automatically at the directory named in the first field of the local /usr/vice/etc/cacheinfo file or by the command's -mountdir argument. Mounting AFS at an alternate location makes it impossible to reach the filespace of any cell that mounts its root.afs and root.cell volumes at the conventional locations. See Displaying and Setting the Cache Size and Location. Create an entry for the cell in the list of database server machines which the Cache Manager maintains in kernel memory. The /usr/vice/etc/CellServDB file on every client machine's local disk lists the database server machines for the local and foreign cells. The afsd program reads the contents of the CellServDB file into kernel memory as it initializes the Cache Manager. You can also use the fs newcell command to add or alter entries in kernel memory directly between reboots of the machine. See Maintaining Knowledge of Database Server Machines. Non-windows client machines may enable Dynamic Root Mode by using the -dynroot option to afsd. When this option is enabled, all cells listed in the CellServDB file will appear in the /afs directory. The contents of the root.afs volume will be ignored. Windows client machines may enable Freelance Mode during client installation or by setting the FreelanceClient setting under Service Parameters in the Windows Registry as mentioned in the Release Notes. When this option is enabled, the root.afs volume is ignored and a mounpoint for each cell is automatically created in the the \\AFS directory when the folder \\AFS\cellname is accessed and the foreign Volume Location servers can be reached. Note that making a foreign cell visible to client machines does not guarantee that your users can access its filespace. The ACLs in the foreign cell must also grant them the necessary permissions. cell granting local access to foreign users local cell granting foreign users access to Making your cell visible in the AFS global namespace does not take away your control over the way in which users from foreign cells access your file tree. By default, foreign users access your cell as the user anonymous, which means they have only the permissions granted to the system:anyuser group on each directory's ACL. Normally these permissions are limited to the l (lookup) and r (read) permissions. There are three ways to grant wider access to foreign users: Grant additional permissions to the system:anyuser group on certain ACLs. Keep in mind, however, that all users can then access that directory in the indicated way (not just specific foreign users you have in mind). Enable automatic registration for users in the foreign cell. This may be done by creating a cross-realm trust in the Kerberos Database. Then add a PTS group named system:authuser@FOREIGN.REALM and give it a group quota greater than the number of foreign users expected to be registered. After the cross-realm trust and the PTS group are created, the aklog command will automatically register foreign users as needed. Consult the documentation for your Kerberos Server for instructions on how to establish a cross-realm trust. Create a local authentication account for specific foreign users, by creating entries in the Protection Database, the Kerberos Database, and the local password file. cell filespace configuration issues configuring filespace, issues file tree conventions for configuring This section summarizes the issues to consider when configuring your AFS filespace. For a discussion of creating volumes that correspond most efficiently to the filespace's directory structure, see Creating Volumes to Simplify Administration. For Windows users: Windows uses a backslash (\) rather than a forward slash (/) to separate the elements in a pathname. The hierarchical organization of the filespace is however the same as on a UNIX machine. AFS pathnames must follow a few conventions so the AFS global namespace looks the same from any AFS client machine. There are corresponding conventions to follow in building your file tree, not just because pathnames reflect the structure of a file tree, but also because the AFS Cache Manager expects a certain configuration. AFS root directory (/afs) in cell filespace directories /afs The first convention is that the top level in your file tree be called the /afs directory. If you name it something else, then you must use the -mountdir argument with the afsd program to get Cache Managers to mount AFS properly. You cannot participate in the AFS global namespace in that case. cell name at second level in file tree directories /afs/cellname symbolic link at second level of AFS pathname The second convention is that just below the /afs directory you place directories corresponding to each cell whose file tree is visible and accessible from the local cell. Minimally, there must be a directory for the local cell. Each such directory is a mount point to the indicated cell's root.cell volume. For example, in the Example Corporation cell, /afs/example.com is a mount point for the cell's own root.cell volume and example.org is a mount point for the Example Organization cell's root.cell volume. The fs lsmount command displays the mount points. % fs lsmount /afs/example.com '/afs/example.com' is a mount point for volume '#root.cell' % fs lsmount /afs/example.org '/afs/example.org' is a mount point for volume '#example.org:root.cell' To reduce the amount of typing necessary in pathnames, you can create a symbolic link with an abbreviated name to the mount point of each cell your users frequently access (particularly the home cell). In the Example Corporation cell, for instance, /afs/example is a symbolic link to the /afs/example.com mount point, as the fs lsmount command reveals. % fs lsmount /afs/example '/afs/example' is a symbolic link, leading to a mount point for volume '#root.cell' file tree conventions third level directories conventional under /afs/cellname You can organize the third level of your cell's file tree any way you wish. The following list describes directories that appear at this level in the conventional configuration: common This directory contains programs and files needed by users working on machines of all system types, such as text editors, online documentation files, and so on. Its /etc subdirectory is a logical place to keep the central update sources for files used on all of your cell's client machines, such as the ThisCell and CellServDB files. public A directory accessible to anyone who can access your filespace, because its ACL grants the l (lookup) and r (read) permissions to the system:anyuser group. It is useful if you want to enable your users to make selected information available to everyone, but do not want to grant foreign users access to the contents of the usr directory which houses user home directories (and is also at this level). It is conventional to create a subdirectory for each of your cell's users. service This directory contains files and subdirectories that help cells coordinate resource sharing. For a list of the proposed standard files and subdirectories to create, call or write to AFS Product Support. As an example, files that other cells expect to find in this directory's etc subdirectory can include the following: CellServDB.export, a list of database server machines for many cells CellServDB.local, a list of the cell's own database server machines passwd, a copy of the local password file (/etc/passwd or equivalent) kept on the local disk of the cell's client machines group, a copy of the local groups file (/etc/group or equivalent) kept on the local disk of the cell's client machines sys_type A separate directory for storing the server and client binaries for each system type you use in the cell. Configuration is simplest if you use the system type names assigned in the AFS distribution, particularly if you wish to use the @sys variable in pathnames (see Using the @sys Variable in Pathnames). The OpenAFS Release Notes lists the conventional name for each supported system type. Within each such directory, create directories named bin, etc, usr, and so on, to store the programs normally kept in the /bin, /etc and /usr directories on a local disk. Then create symbolic links from the local directories on client machines into AFS; see Configuring the Local Disk. Even if you do not choose to use symbolic links in this way, it can be convenient to have central copies of system binaries in AFS. If binaries are accidentally removed from a machine, you can recopy them onto the local disk from AFS rather than having to recover them from tape usr This directory contains home directories for your local users. As discussed in the previous entry for the public directory, it is often practical to protect this directory so that only locally authenticated users can access it. This keeps the contents of your user's home directories as secure as possible. If your cell is quite large, directory lookup can be slowed if you put all home directories in a single usr directory. For suggestions on distributing user home directories among multiple grouping directories, see Grouping Home Directories. volume name conventions for conventions volume names volume separate for each top level directory file tree creating volumes to match top level directories This section discusses how to create volumes in ways that make administering your system easier. At the top levels of your file tree (at least through the third level), each directory generally corresponds to a separate volume. Some cells also configure the subdirectories of some third level directories as separate volumes. Common examples are the /afs/cellname/common and /afs/cellname/usr directories. You do not have to create a separate volume for every directory level in a tree, but the advantage is that each volume tends to be smaller and easier to move for load balancing. The overhead for a mount point is no greater than for a standard directory, nor does the volume structure itself require much disk space. Most cells find that below the fourth level in the tree, using a separate volume for each directory is no longer efficient. For instance, while each user's home directory (at the fourth level in the tree) corresponds to a separate volume, all of the subdirectories in the home directory normally reside in the same volume. Keep in mind that only one volume can be mounted at a given directory location in the tree. In contrast, a volume can be mounted at several locations, though this is not recommended because it distorts the hierarchical nature of the file tree, potentially causing confusion. volume name restrictions restrictions on volume names volume name two required volume root (root.afs and root.cell) root volumes (root.afs and root.cell) You can name your volumes anything you choose, subject to a few restrictions: Read/write volume names can be up to 22 characters in length. The maximum length for volume names is 31 characters, and there must be room to add the .readonly extension on read-only volumes. Do not add the .readonly and .backup extensions to volume names yourself, even if they are appropriate. The Volume Server adds them automatically as it creates a read-only or backup version of a volume. There must be volumes named root.afs and root.cell, mounted respectively at the top (/afs) level in the filespace and just below that level, at the cell's name (for example, at /afs/example.com in the Example Corporation cell). Deviating from these names only creates confusion and extra work. Changing the name of the root.afs volume, for instance, means that you must use the -rootvol argument to the afsd program on every client machine, to name the alternate volume. Similarly, changing the root.cell volume name prevents users in foreign cells from accessing your filespace, if the mount point for your cell in their filespace refers to the conventional root.cell name. Of course, this is one way to make your cell invisible to other cells. It is best to assign volume names that indicate the type of data they contain, and to use similar names for volumes with similar contents. It is also helpful if the volume name is similar to (or at least has elements in common with) the name of the directory at which it is mounted. Understanding the pattern then enables you accurately to guess what a volume contains and where it is mounted. Many cells find that the most effective volume naming scheme puts a common prefix on the names of all related volumes. Table 1 describes the recommended prefixing scheme. Prefix Contents Example Name Example Mount Point common. popular programs and files common.etc /afs/cellname/common/etc src. source code src.afs /afs/cellname/src/afs proj. project data proj.portafs /afs/cellname/proj/portafs test. testing or other temporary data test.smith /afs/cellname/usr/smith/test user. user home directory data user.terry /afs/cellname/usr/terry sys_type. programs compiled for an operating system type rs_aix42.bin /afs/cellname/rs_aix42/bin Table 2 is a more specific example for a cell's rs_aix42 system volumes and directories: Example Name Example Mount Point rs_aix42.bin /afs/cellname/rs_aix42/bin, /afs/cellname/rs_aix42/bin rs_aix42.etc /afs/cellname/rs_aix42/etc rs_aix42.usr /afs/cellname/rs_aix42/usr rs_aix42.usr.afsws /afs/cellname/rs_aix42/usr/afsws rs_aix42.usr.lib /afs/cellname/rs_aix42/usr/lib rs_aix42.usr.bin /afs/cellname/rs_aix42/usr/bin rs_aix42.usr.etc /afs/cellname/rs_aix42/usr/etc rs_aix42.usr.inc /afs/cellname/rs_aix42/usr/inc rs_aix42.usr.man /afs/cellname/rs_aix42/usr/man rs_aix42.usr.sys /afs/cellname/rs_aix42/usr/sys rs_aix42.usr.local /afs/cellname/rs_aix42/usr/local There are several advantages to this scheme: The volume name is similar to the mount point name in the filespace. In all of the entries in Table 2, for example, the only difference between the volume and mount point name is that the former uses periods as separators and the latter uses slashes. Another advantage is that the volume name indicates the contents, or at least suggests the directory on which to issue the ls command to learn the contents. It makes it easy to manipulate groups of related volumes at one time. In particular, the vos backupsys command's -prefix argument enables you to create a backup version of every volume whose name starts with the same string of characters. Making a backup version of each volume is one of the first steps in backing up a volume with the AFS Backup System, and doing it for many volumes with one command saves you a good deal of typing. For instructions for creating backup volumes, see Creating Backup Volumes, For information on the AFS Backup System, see Configuring the AFS Backup System and Backing Up and Restoring AFS Data. It makes it easy to group related volumes together on a partition. Grouping related volumes together has several advantages of its own, discussed in Grouping Related Volumes on a Partition. volume grouping related on same partition disk partition grouping related volumes on If your cell is large enough to make it practical, consider grouping related volumes together on a partition. In general, you need at least three file server machines for volume grouping to be effective. Grouping has several advantages, which are most obvious when the file server machine becomes inaccessible: If you keep a hardcopy record of the volumes on a partition, you know which volumes are unavailable. You can keep such a record without grouping related volumes, but a list composed of unrelated volumes is much harder to maintain. Note that the record must be on paper, because the outage can prevent you from accessing an online copy or from issuing the vos listvol command, which gives you the same information. The effect of an outage is more localized. For example, if all of the binaries for a given system type are on one partition, then only users of that system type are affected. If a partition houses binary volumes from several system types, then an outage can affect more people, particularly if the binaries that remain available are interdependent with those that are not available. The advantages of grouping related volumes on a partition do not necessarily extend to the grouping of all related volumes on one file server machine. For instance, it is probably unwise in a cell with two file server machines to put all system volumes on one machine and all user volumes on the other. An outage of either machine probably affects everyone. Admittedly, the need to move volumes for load balancing purposes can limit the practicality of grouping related volumes. You need to weigh the complementary advantages case by case. replication appropriate volumes volume type to replicate volume where to place replicated read-only volume selecting site As discussed in Replication, replication refers to making a copy, or clone, of a read/write source volume and then placing the copy on one or more additional file server machines. Replicating a volume can increase the availability of the contents. If one file server machine housing the volume becomes inaccessible, users can still access the copy of the volume stored on a different machine. No one machine is likely to become overburdened with requests for a popular file, either, because the file is available from several machines. However, replication is not appropriate for all cells. If a cell does not have much disk space, replication can be unduly expensive, because each clone not on the same partition as the read/write source takes up as much disk space as its source volume did at the time the clone was made. Also, if you have only one file server machine, replication uses up disk space without increasing availability. Replication is also not appropriate for volumes that change frequently. You must issue the vos release command every time you need to update a read-only volume to reflect changes in its read/write source. For both of these reasons, replication is appropriate only for popular volumes whose contents do not change very often, such as system binaries and other volumes mounted at the upper levels of your filespace. User volumes usually exist only in a read/write version since they change so often. If you are replicating any volumes, you must replicate the root.afs and root.cell volumes, preferably at two or three sites each (even if your cell only has two or three file server machines). The Cache Manager needs to pass through the directories corresponding to the root.afs and root.cell volumes as it interprets any pathname. The unavailability of these volumes makes all other volumes unavailable too, even if the file server machines storing the other volumes are still functioning. Another reason to replicate the root.afs volume is that it can lessen the load on the File Server machine. The Cache Manager has a bias to access a read-only version of the root.afs volume if it is replicate, which puts the Cache Manager onto the read-only path through the AFS filespace. While on the read-only path, the Cache Manager attempts to access a read-only copy of replicated volumes. The File Server needs to track only one callback per Cache Manager for all of the data in a read-only volume, rather than the one callback per file it must track for read/write volumes. Fewer callbacks translate into a smaller load on the File Server. If the root.afs volume is not replicated, the Cache Manager follows a read/write path through the filespace, accessing the read/write version of each volume. The File Server distributes and tracks a separate callback for each file in a read/write volume, imposing a greater load on it. For more on read/write and read-only paths, see The Rules of Mount Point Traversal. It also makes sense to replicate system binary volumes in many cases, as well as the volume corresponding to the /afs/cellname/usr directory and the volumes corresponding to the /afs/cellname/common directory and its subdirectories. It is a good idea to place a replica on the same partition as the read/write source. In this case, the read-only volume is a clone (like a backup volume): it is a copy of the source volume's vnode index, rather than a full copy of the volume contents. Only if the read/write volume moves to another partition or changes substantially does the read-only volume consume significant disk space. Read-only volumes kept on other servers' partitions always consume the full amount of disk space that the read/write source consumed when the read-only volume was created. You cannot have a replica volume on a different partition of the same server hosting the read/write volume. "Cheap" read-only volumes must be on the same partition as the read/write; all other read-only volumes must be on different servers. Every AFS volume has associated with it a quota that limits the amount of disk space the volume is allowed to use. To set and change quota, use the commands described in Setting and Displaying Volume Quota and Current Size. By default, every new volume is assigned a space quota of 5000 KB blocks unless you include the -maxquota argument to the vos create command. Also by default, the ACL on the root directory of every new volume grants all permissions to the members of the system:administrators group. To learn how to change these values when creating an account with individual commands, see To create one user account with individual commands. When using uss commands to create accounts, you can specify alternate ACL and quota values in the template file's V instruction; see Creating a Volume with the V Instruction. server machine configuration issues configuring file server machine, issues roles for server machine summary server machine roles for summary server machine first installed This section discusses some issues to consider when configuring server machines, which store AFS data, transfer it to client machines on request, and house the AFS administrative databases. To learn about client machines, see Configuring Client Machines. If your cell has more than one AFS server machine, you can configure them to perform specialized functions. A machine can assume one or more of the roles described in the following list. For more details, see The Four Roles for File Server Machines. A simple file server machine runs only the processes that store and deliver AFS files to client machines. You can run as many simple file server machines as you need to satisfy your cell's performance and disk space requirements. A database server machine runs the four database server processes that maintain AFS's replicated administrative databases: the Authentication, Backup, Protection, and Volume Location (VL) Server processes. A binary distribution machine distributes the AFS server binaries for its system type to all other server machines of that system type. The single system control machine distributes common server configuration files to all other server machines in the cell. The OpenAFS Quick Beginnings explains how to configure your cell's first file server machine to assume all four roles. The OpenAFS Quick Beginnings chapter on installing additional server machines also explains how to configure them to perform one or more roles. database server machine reason to run three distribution of databases databases, distributed distributed databases The AFS administrative databases are housed on database server machines and store information that is crucial for correct cell functioning. Both server processes and Cache Managers access the information frequently: Every time a Cache Manager fetches a file from a directory that it has not previously accessed, it must look up the file's location in the Volume Location Database (VLDB). Every time a user obtains an AFS token from the Authentication Server, the server looks up the user's password in the Authentication Database. The first time that a user accesses a volume housed on a specific file server machine, the File Server contacts the Protection Server for a list of the user's group memberships as recorded in the Protection Database. Every time you back up a volume using the AFS Backup System, the Backup Server creates records for it in the Backup Database. Maintaining your cell is simplest if the first machine has the lowest IP address of any machine you plan to use as a database server machine. If you later decide to use a machine with a lower IP address as a database server machine, you must update the CellServDB file on all clients before introducing the new machine. If your cell has more than one server machine, it is best to run more than one as a database server machine (but more than three are rarely necessary). Replicating the administrative databases in this way yields the same benefits as replicating volumes: increased availability and reliability. If one database server machine or process stops functioning, the information in the database is still available from others. The load of requests for database information is spread across multiple machines, preventing any one from becoming overloaded. Unlike replicated volumes, however, replicated databases do change frequently. Consistent system performance demands that all copies of the database always be identical, so it is not acceptable to record changes in only some of them. To synchronize the copies of a database, the database server processes use AFS's distributed database technology, Ubik. See Replicating the OpenAFS Administrative Databases. If your cell has only one file server machine, it must also serve as a database server machine. If you cell has two file server machines, it is not always advantageous to run both as database server machines. If a server, process, or network failure interrupts communications between the database server processes on the two machines, it can become impossible to update the information in the database because neither of them can alone elect itself as the synchronization site. server machine protecting directories on local disk local disk protecting on file server machine It is generally simplest to store the binaries for all AFS server processes in the /usr/afs/bin directory on every file server machine, even if some processes do not actively run on the machine. This makes it easier to reconfigure a machine to fill a new role. For security reasons, the /usr/afs directory on a file server machine and all of its subdirectories and files must be owned by the local superuser root and have only the first w (write) mode bit turned on. Some files even have only the first r (read) mode bit turned on (for example, the /usr/afs/etc/KeyFile file, which lists the AFS server encryption keys). Each time the BOS Server starts, it checks that the mode bits on certain files and directories match the expected values. For a list, see the OpenAFS Quick Beginnings section about protecting sensitive AFS directories, or the discussion of the output from the bos status command in To display the status of server processes and their BosConfig entries. For a description of the contents of all AFS directories on a file server machine's local disk, see Administering Server Machines. The partitions that house AFS volumes on a file server machine must be mounted at directories named /vicepindex where index is one or two lowercase letters. By convention, the first AFS partition created is mounted at the /vicepa directory, the second at the /vicepb directory, and so on through the /vicepz directory. The names then continue with /vicepaa through /vicepaz, /vicepba through /vicepbz, and so on, up to the maximum supported number of server partitions, which is specified in the OpenAFS Release Notes. Each /vicepx directory must correspond to an entire partition or logical volume, and must be a subdirectory of the root directory (/). It is not acceptable to configure part of (for example) the /usr partition as an AFS server partition and mount it on a directory called /usr/vicepa. Also, do not store non-AFS files on AFS server partitions. The File Server and Volume Server expect to have available all of the space on the partition. Sharing space also creates competition between AFS and the local UNIX file system for access to the partition, particularly if the UNIX files are frequently used. server machine monitoring file server machine rebooting, about rebooting file server machine, limiting weekly restart of BOS Server (automatic) about restart times for BOS Server about AFS provides several tools for monitoring the File Server, including the scout and afsmonitor programs. You can configure them to alert you when certain threshold values are exceeded, for example when a server partition is more than 95% full. See Monitoring and Auditing AFS Performance. Rebooting a file server machine requires shutting down the AFS processes and so inevitably causes a service outage. Reboot file server machines as infrequently as possible. For instructions, see Rebooting a Server Machine. The BOS Server checks each morning at 5:00 a.m. for any newly installed binary files in the /usr/afs/bin directory. It compares the timestamp on each binary file to the time at which the corresponding process last restarted. If the timestamp on the binary is later, the BOS Server restarts the corresponding process to start using it. The BOS server also supports performing a weekly restart of all AFS server processes, including itself. This functionality is disabled on new installs, but historically it was set to 4:00am on Sunday. Administrators may find that installations predating OpenAFS 1.6.0 have weekly restarts enabled. The default times are in the early morning hours when the outage that results from restarting a process is likely to disturb the fewest number of people. You can display the restart times for each machine with the bos getrestart command, and set them with the bos setrestart command. The latter command enables you to disable automatic restarts entirely, by setting the time to never. See Setting the BOS Server's Restart Times. client machine configuration issues configuring client machine, issues This section summarizes issues to consider as you install and configure client machines in your cell. client machine files required on local disk local disk files required on client machine file required on client machine local disk You can often free up significant amounts of local disk space on AFS client machines by storing standard UNIX files in AFS and creating symbolic links to them from the local disk. The @sys pathname variable can be useful in links to system-specific files; see Using the @sys Variable in Pathnames. There are two types of files that must actually reside on the local disk: boot sequence files needed before the afsd program is invoked, and files that can be helpful during file server machine outages. During a reboot, AFS is inaccessible until the afsd program executes and initializes the Cache Manager. (In the conventional configuration, the AFS initialization file is included in the machine's initialization sequence and invokes the afsd program.) Files needed during reboot prior to that point must reside on the local disk. They include the following, but this list is not necessarily exhaustive. Standard UNIX utilities including the following or their equivalents: Machine initialization files (stored in the /etc or /sbin directory on many system types) The fstab file The mount command binary The umount command binary All subdirectories and files in the /usr/vice directory, including the following: The /usr/vice/cache directory The /usr/vice/etc/afsd command binary The /usr/vice/etc/cacheinfo file The /usr/vice/etc/CellServDB file The /usr/vice/etc/ThisCell file For more information on these files, see Configuration and Cache-Related Files on the Local Disk. The other type of files and programs to retain on the local disk are those you need when diagnosing and fixing problems caused by a file server outage, because the outage can make inaccessible the copies stored in AFS. Examples include the binaries for a text editor (such as ed or vi) and for the fs and bos commands. Store copies of AFS command binaries in the /usr/vice/etc directory as well as including them in the /usr/afsws directory, which is normally a link into AFS. Then place the /usr/afsws directory before the /usr/vice/etc directory in users' PATH environment variable definition. When AFS is functioning normally, users access the copy in the /usr/afsws directory, which is more likely to be current than a local copy. client machine enabling access to foreign cell As detailed in Making Other Cells Visible in Your Cell, you enable the Cache Manager to access a cell's AFS filespace by storing a list of the cell's database server machines in the local /usr/vice/etc/CellServDB file. The Cache Manager reads the list into kernel memory at reboot for faster retrieval. You can change the list in kernel memory between reboots by using the fs newcell command. Because each client machine maintains its own copy of the CellServDB file, you can in theory enable access to different foreign cells on different client machines. This is not usually practical, however, especially if users do not always work on the same machine. at-sys (@sys) variable in pathnames sys (@sys) variable in pathnames variables @sys in pathnames When creating symbolic links into AFS on the local disk, it is often practical to use the @sys variable in pathnames. The Cache Manager automatically substitutes the local machine's AFS system name (CPU/operating system type) for the @sys variable. This means you can place the same links on machines of various system types and still have each machine access the binaries for its system type. For example, the Cache Manager on a machine running AIX 4.2 converts /afs/example.com/@sys to /afs/example.com/rs_aix42, whereas a machine running Solaris 10 converts it to /afs/example.com/sun4x_510. If you want to use the @sys variable, it is simplest to use the conventional AFS system type names as specified in the OpenAFS Release Notes. The Cache Manager records the local machine's system type name in kernel memory during initialization. If you do not use the conventional names, you must use the fs sysname command to change the value in kernel memory from its default just after Cache Manager initialization, on every client machine of the relevant system type. The fs sysname command also displays the current value; see Displaying and Setting the System Type Name. In pathnames in the AFS filespace itself, use the @sys variable carefully and sparingly, because it can lead to unexpected results. It is generally best to restrict its use to only one level in the filespace. The third level is a common choice, because that is where many cells store the binaries for different machine types. Multiple instances of the @sys variable in a pathname are especially dangerous to people who must explicitly change directories (with the cd command, for example) into directories that store binaries for system types other than the machine on which they are working, such as administrators or developers who maintain those directories. After changing directories, it is recommended that such people verify they are in the desired directory. The Cache Manager stores a table of preferences for file server machines in kernel memory. A preference rank pairs a file server machine interface's IP address with an integer in the range from 1 to 65,534. When it needs to access a file, the Cache Manager compares the ranks for the interfaces of all machines that house the file, and first attempts to access the file via the interface with the best rank. As it initializes, the Cache Manager sets default ranks that bias it to access files via interfaces that are close to it in terms of network topology. You can adjust the preference ranks to improve performance if you wish. The Cache Manager also uses similar preferences for Volume Location (VL) Server machines. Use the fs getserverprefs command to display preference ranks and the fs setserverprefs command to set them. See Maintaining Server Preference Ranks. user account configuration issues This section discusses some of the issues to consider when configuring AFS user accounts. Because AFS is separate from the UNIX file system, a user's AFS account is separate from her UNIX account. The preferred method for creating a user account is with the uss suite of commands. With a single command, you can create all the components of one or many accounts, after you have prepared a template file that guides the account creation. See Creating and Deleting User Accounts with the uss Command Suite. Alternatively, you can issue the individual commands that create each component of an account. For instructions, along with instructions for removing user accounts and changing user passwords, user volume quotas and usernames, see Administering User Accounts. When users leave your system, it is often good policy to remove their accounts. Instructions appear in Deleting Individual Accounts with the uss delete Command and Removing a User Account. An AFS user account consists of the following components, which are described in greater detail in The Components of an AFS User Account. A Protection Database entry An Authentication Database entry A volume A home directory at which the volume is mounted Ownership of the home directory and full permissions on its ACL An entry in the local password file (/etc/passwd or equivalent) of each machine the user needs to log into Optionally, standard files and subdirectories that make the account more useful By creating some components but not others, you can create accounts at different levels of functionality, using either uss commands as described in Creating and Deleting User Accounts with the uss Command Suite or individual commands as described in Administering User Accounts. The levels of functionality include the following An authentication-only account enables the user to obtain AFS tokens and so to access protected AFS data and to issue privileged commands. It consists only of entries in the Authentication and Protection Database. This type of account is suitable for administrative accounts and for users from foreign cells who need to access protected data. Local users generally also need a volume and home directory. A basic user account includes a volume for the user, in addition to Authentication and Protection Database entries. The volume is mounted in the AFS filespace as the user's home directory, and provides a repository for the user's personal files. A full account adds configuration files for basic functions such as logging in, printing, and mail delivery to a basic account, making it more convenient and useful. For a discussion of some useful types of configuration files, see Creating Standard Files in New AFS Accounts. If your users have UNIX user accounts that predate the introduction of AFS in the cell, you possibly want to convert them into AFS accounts. There are three main issues to consider: Making UNIX and AFS UIDs match Setting the password field in the local password file appropriately Moving files from the UNIX file system into AFS For further discussion, see Converting Existing UNIX Accounts with uss or Converting Existing UNIX Accounts. username choosing user name username choosing name user anonymous user AFS UID reserved AFS UID reserved anonymous user This section suggests schemes for choosing usernames, AFS UIDs, user volume names and mount point names, and also outlines some restrictions on your choices. AFS imposes very few restrictions on the form of usernames. It is best to keep usernames short, both because many utilities and applications can handle usernames of no more than eight characters and because by convention many components of and AFS account incorporate the name. These include the entries in the Protection and Authentication Databases, the volume, and the mount point. Depending on your electronic mail delivery system, the username can become part of the user's mailing address. The username is also the string that the user types when logging in to a client machine. Some common choices for usernames are last names, first names, initials, or a combination, with numbers sometimes added. It is also best to avoid using the following characters, many of which have special meanings to the command shell. The comma (,) The colon (:), because AFS reserves it as a field separator in protection group names; see The Two Types of User-Defined Groups The semicolon (;) The "at-sign" (@); this character is reserved for Internet mailing addresses Spaces The newline character The period (.); it is conventional to use this character only in the special username that an administrator adopts while performing privileged tasks, such as pat.admin AFS associates a unique identification number, the AFS UID, with every username, recording the mapping in the user's Protection Database entry. The AFS UID functions within AFS much as the UNIX UID does in the local file system: the AFS server processes and the Cache Manager use it internally to identify a user, rather than the username. Every AFS user also must have a UNIX UID recorded in the local password file (/etc/passwd or equivalent) of each client machine they log onto. Both administration and a user's AFS access are simplest if the AFS UID and UNIX UID match. One important consequence of matching UIDs is that the owner reported by the ls -l command matches the AFS username. It is usually best to allow the Protection Server to allocate the AFS UID as it creates the Protection Database entry. However, both the pts createuser command and the uss commands that create user accounts enable you to assign AFS UIDs explicitly. This is appropriate in two cases: You wish to group together the AFS UIDs of related users You are converting an existing UNIX account into an AFS account and want to make the AFS UID match the existing UNIX UID After the Protection Server initializes for the first time on a cell's first file server machine, it starts assigning AFS UIDs at a default value. To change the default before creating any user accounts, or at any time, use the pts setmax command to reset the max user id counter. To display the counter, use the pts listmax command. See Displaying and Setting the AFS UID and GID Counters. AFS reserves one AFS UID, 32766, for the user anonymous. The AFS server processes assign this identity and AFS UID to any user who does not possess a token for the local cell. Do not assign this AFS UID to any other user or hardcode its current value into any programs or a file's owner field, because it is subject to change in future releases. username part of volume name choosing name user volume Like any volume name, a user volume's base (read/write) name cannot exceed 22 characters in length or include the .readonly or .backup extension. See Creating Volumes to Simplify Administration. By convention, user volume names have the format user.username. Using the user. prefix not only makes it easy to identify the volume's contents, but also to create a backup version of all user volumes by issuing a single vos backupsys command. mount point choosing name for user volume choosing name user volume mount point By convention, the mount point for a user's volume is named after the username. Many cells follow the convention of mounting user volumes in the /afs/cellname/usr directory, as discussed in The Third Level. Very large cells sometimes find that mounting all user volumes in the same directory slows directory lookup, however; for suggested alternatives, see the following section. directories for grouping user home directories user account suggestions for grouping home directories Mounting user volumes in the /afs/cellname/usr directory is an AFS-appropriate variation on the standard UNIX practice of putting user home directories under the /usr subdirectory. However, cells with more than a few hundred users sometimes find that mounting all user volumes in a single directory results in slow directory lookup. The solution is to distribute user volume mount points into several directories; there are a number of alternative methods to accomplish this. Distribute user home directories into multiple directories that reflect organizational divisions, such as academic or corporate departments. For example, a company can create group directories called usr/marketing, usr/research, usr/finance. A good feature of this scheme is that knowing a user's department is enough to find the user's home directory. Also, it makes it easy to set the ACL to limit access to members of the department only. A potential drawback arises if departments are of sufficiently unequal size that users in large departments experience slower lookup than users in small departments. This scheme is also not appropriate in cells where users frequently change between divisions. Distribute home directories into alphabetic subdirectories of the usr directory (the usr/a subdirectory, the usr/b subdirectory, and so on), based on the first letter of the username. If the cell is very large, create subdirectories under each letter that correspond to the second letter in the user name. This scheme has the same advantages and disadvantages of a department-based scheme. Anyone who knows the user's username can find the user's home directory, but users with names that begin with popular letters sometimes experience slower lookup. Distribute home directories randomly but evenly into more than one grouping directory. One cell that uses this scheme has over twenty such directories called the usr1 directory, the usr2 directory, and so on. This scheme is especially appropriate in cells where the other two schemes do not seem feasible. It eliminates the potential problem of differences in lookup speed, because all directories are about the same size. Its disadvantage is that there is no way to guess which directory a given user's volume is mounted in, but a solution is to create a symbolic link in the regular usr directory that references the actual mount point. For example, if user smith's volume is mounted at the /afs/bigcell.example.com/usr17/smith directory, then the /afs/bigcell.example.com/usr/smith directory is a symbolic link to the ../usr17/smith directory. This way, if someone does not know which directory the user smith is in, he or she can access it through the link called usr/smith; people who do know the appropriate directory save lookup time by specifying it. For instructions on how to implement the various schemes when using the uss program to create user accounts, see Evenly Distributing User Home Directories with the G Instruction and Creating a Volume with the V Instruction. Mounting the backup version of a user's volume is a simple way to enable users themselves to restore data they have accidentally removed or deleted. It is conventional to mount the backup version at a subdirectory of the user's home directory (called perhaps the OldFiles subdirectory), but other schemes are possible. Once per day you create a new backup version to capture the changes made that day, overwriting the previous day's backup version with the new one. Users can always retrieve the previous day's copy of a file without your assistance, freeing you to deal with more pressing tasks. Users sometimes want to delete the mount point to their backup volume, because they erroneously believe that the backup volume's contents count against their quota. Remind them that the backup volume is separate, so the only space it uses in the user volume is the amount needed for the mount point. For further discussion of backup volumes, see Backing Up AFS Data and Creating Backup Volumes. file creating standard ones in new user account user account creating standard files in creating standard files in new user account From your experience as a UNIX administrator, you are probably familiar with the use of login and shell initialization files (such as the .login and .cshrc files) to make an account easier to use. It is often practical to add some AFS-specific directories to the definition of the user's PATH environment variable, including the following: The path to a bin subdirectory in the user's home directory for binaries the user has created (that is, /afs/cellname/usr/username/bin) The /usr/afsws/bin path, which conventionally includes programs like fs, klog, kpasswd, pts, tokens, and unlog The /usr/afsws/etc path, if the user is an administrator; it usually houses the AFS command suites that require privilege (the backup, butc, kas, uss, vos commands), and others. If you are not using an AFS-modified login utility, it can be helpful to users to invoke the klog command in their .login file so that they obtain AFS tokens as part of logging in. In the following example command sequence, the first line echoes the string klog to the standard output stream, so that the user understands the purpose of the Password: prompt that appears when the second line is executed. The -setpag flag associates the new tokens with a process authentication group (PAG), which is discussed further in Identifying AFS Tokens by PAG. echo -n "klog " klog -setpag The following sequence of commands has a similar effect, except that the pagsh command forks a new shell with which the PAG and tokens are associated. pagsh echo -n "klog " klog If you use an AFS-modified login utility, this sequence is not necessary, because such utilities both log a user in locally and obtain AFS tokens. group AFS GID group restrictions group privacy flags privacy flags on Protection Database entry AFS enables users to define their own groups of other users or machines. The groups are placed on ACLs to grant the same permissions to many users without listing each user individually. For group creation instructions, see Administering the Protection Database. Groups have AFS ID numbers, just as users do, but an AFS group ID (GID) is a negative integer whereas a user's AFS UID is a positive integer. By default, the Protection Server allocates a new group's AFS GID automatically, but members of the system:administrators group can assign a GID when issuing the pts creategroup command. Before explicitly assigning a GID, it is best to verify that it is not already in use. A group cannot belong to another group, but it can own another group or even itself as long as it (the owning group) has at least one member. The current owner of a group can transfer ownership of the group to another user or group, even without the new owner's permission. At that point the former owner loses administrative control over the group. By default, each user can create 20 groups. A system administrator can increase or decrease this group creation quota with the pts setfields command. Each Protection Database entry (group or user) is protected by a set of five privacy flagswhich limit who can administer the entry and what they can do. The default privacy flags are fairly restrictive, especially for user entries. See Setting the Privacy Flags on Database Entries. system:administrators group about system:anyuser group about system:authuser group about group system-defined As the Protection Server initializes for the first time on a cell's first database server machine, it automatically creates three group entries: the system:anyuser, system:authuser, and system:administrators groups. AFS UID reserved system-defined groups The first two system groups are unlike any other groups in the Protection Database in that they do not have a stable membership: The system:anyuser group includes everyone who can access a cell's AFS filespace: users who have tokens for the local cell, users who have logged in on a local AFS client machine but not obtained tokens (such as the local superuser root), and users who have connected to a local machine from outside the cell. Placing the system:anyuser group on an ACL grants access to the widest possible range of users. It is the only way to extend access to users from foreign AFS cells that do not have local accounts. The system:authuser group includes everyone who has a valid token obtained from the cell's AFS authentication service. Because the groups do not have a stable membership, the pts membership command produces no output for them. Similarly, they do not appear in the list of groups to which a user belongs. The system:administrators group does have a stable membership, consisting of the cell's privileged administrators. Members of this group can issue any pts command, and are the only ones who can issue several other restricted commands (such as the chown command on AFS files). By default, they also implicitly have the a (administer) and l (lookup) permissions on every ACL in the filespace. For information about changing this default, see Administering the system:administrators Group. For a discussion of how to use system groups effectively on ACLs, see Using Groups on ACLs. All users can create regular groups. A regular group name has two fields separated by a colon, the first of which must indicate the group's ownership. The Protection Server refuses to create or change the name of a group if the result does not accurately indicate the ownership. Members of the system:administrators group can create prefix-less groups whose names do not have the first field that indicates ownership. For suggestions on using the two types of groups effectively, see Using Groups Effectively. authentication AFS separate from UNIX AFS authentication separate from UNIX As explained in Differences in Authentication, AFS authentication is separate from UNIX authentication because the two file systems are separate. The separation has two practical implications: To access AFS files, users must both log into the local file system and authenticate with the AFS authentication service. (Logging into the local file system is necessary because the only way to access the AFS filespace is through a Cache Manager, which resides in the local machine's kernel.) Passwords are stored in two separate places: in the Kerberos Database for AFS and in the each machine's local password file (the /etc/passwd file or equivalent) for the local file system. When a user successfully authenticates, the AFS authentication service passes a token to the user's Cache Manager. The token is a small collection of data that certifies that the user has correctly provided the password associated with a particular AFS identity. The Cache Manager presents the token to AFS server processes along with service requests, as proof that the user is genuine. To learn about the mutual authentication procedure they use to establish identity, see A More Detailed Look at Mutual Authentication. The Cache Manager stores tokens in the user's credential structure in kernel memory. To distinguish one user's credential structure from another's, the Cache Manager identifies each one either by the user's UNIX UID or by a process authentication group (PAG), which is an identification number guaranteed to be unique in the cell. For further discussion, see Identifying AFS Tokens by PAG. tokens one-per-cell rule A user can have only one token per cell in each separately identified credential structure. To obtain a second token for the same cell, the user must either log into a different machine or obtain another credential structure with a different identifier than any existing credential structure, which is most easily accomplished by issuing the pagsh command (see Identifying AFS Tokens by PAG). In a single credential structure, a user can have one token for each of many cells at the same time. As this implies, authentication status on one machine or PAG is independent of authentication status on another machine or PAG, which can be very useful to a user or system administrator. The AFS distribution includes library files that enable each system type's login utility to authenticate users with AFS and log them into the local file system in one step. If you do not configure an AFS-modified login utility on a client machine, its users must issue the klog command to authenticate with AFS after logging in. The AFS-modified libraries do not necessarily support all features available in an operating system's proprietary login utility. In some cases, it is not possible to support a utility at all. For more information about the supported utilities in each AFS version, see the OpenAFS Release Notes. commands pagsh pagsh command commands klog with -setpag flag klog command with -setpag flag PAG creating with klog or pagsh command creating PAG with klog or pagsh command process authentication group PAG As noted, the Cache Manager identifies user credential structures either by UNIX UID or by PAG. Using a PAG is preferable because it guaranteed to be unique: the Cache Manager allocates it based on a counter that increments with each use. In contrast, multiple users on a machine can share or assume the same UNIX UID, which creates potential security problems. The following are two common such situations: The local superuser root can always assume any other user's UNIX UID simply by issuing the su command, without providing the user's password. If the credential structure is associated with the user's UNIX UID, then assuming the UID means inheriting the AFS tokens. Two users working on different NFS client machines can have the same UNIX UID in their respective local file systems. If they both access the same NFS/AFS Translator machine, and the Cache Manager there identifies them by their UNIX UID, they become indistinguishable. To eliminate this problem, the Cache Manager on a translator machine automatically generates a PAG for each user and uses it, rather than the UNIX UID, to tell users apart. Yet another advantage of PAGs over UIDs is that processes spawned by the user inherit the PAG and so share the token; thus they gain access to AFS as the authenticated user. In many environments, for example, printer and other daemons run under identities (such as the local superuser root) that the AFS server processes recognize only as the anonymous user. Unless PAGs are used, such daemons cannot access files for which the system:anyuser group does not have the necessary ACL permissions. Once a user has a PAG, any new tokens the user obtains are associated with the PAG. The PAG expires two hours after any associated tokens expire or are discarded. If the user issues the klog command before the PAG expires, the new token is associated with the existing PAG (the PAG is said to be recycled in this case). AFS-modified login utilities automatically generate a PAG, as described in the following section. If you use a standard login utility, your users must issue the pagsh command before the klog command, or include the latter command's -setpag flag. For instructions, see Using Two-Step Login and Authentication. Users can also use either command at any time to create a new PAG. The difference between the two commands is that the klog command replaces the PAG associated with the current command shell and tokens. The pagsh command initializes a new command shell before creating a new PAG. If the user already had a PAG, any running processes or jobs continue to use the tokens associated with the old PAG whereas any new jobs or processes use the new PAG and its associated tokens. When you exit the new shell (by pressing <Ctrl-d>, for example), you return to the original PAG and shell. By default, the pagsh command initializes a Bourne shell, but you can include the -c argument to initialize a C shell (the /bin/csh program on many system types) or Korn shell (the /bin/ksh program) instead. login utility AFS version As previously mentioned, an AFS-modified login utility simultaneously obtains an AFS token and logs the user into the local file system. This section outlines the login and authentication process and its interaction with the value in the password field of the local password file. An AFS-modified login utility performs a sequence of steps similar to the following; details can vary for different operating systems: It checks the user's entry in the local password file (the /etc/passwd file or equivalent). If no entry exists, or if an asterisk (*) appears in the entry's password field, the login attempt fails. If the entry exists, the attempt proceeds to the next step. The utility obtains a PAG. The utility converts the password provided by the user into an encryption key and encrypts a packet of data with the key. It sends the packet to the AFS authentication service (the AFS Authentication Server in the conventional configuration). The authentication service decrypts the packet and, depending on the success of the decryption, judges the password to be correct or incorrect. (For more details, see A More Detailed Look at Mutual Authentication.) If the authentication service judges the password incorrect, the user does not receive an AFS token. The PAG is retained, ready to be associated with any tokens obtained later. The attempt proceeds to Step 6. If the authentication service judges the password correct, it issues a token to the user as proof of AFS authentication. The login utility logs the user into the local UNIX file system. Some login utilities echo the following banner to the screen to alert the user to authentication with AFS. Step 6 is skipped. AFS(R) version Login If no AFS token was granted in Step 4, the login utility attempts to log the user into the local file system, by comparing the password provided to the local password file. If the password is incorrect or any value other than an encrypted 13-character string appears in the password field, the login attempt fails. If the password is correct, the user is logged into the local file system only. local password file when using AFS--modified login utility login utility AFS version's interaction with local password file password local password file As indicated, when you use an AFS-modified login utility, the password field in the local password file is no longer the primary gate for access to your system. If the user provides the correct AFS password, then the program never consults the local password file. However, you can still use the password field to control access, in the following way: To prevent both local login and AFS authentication, place an asterisk (*) in the field. This is useful mainly in emergencies, when you want to prevent a certain user from logging into the machine. To prevent login to the local file system if the user does not provide the correct AFS password, place a character string of any length other than the standard thirteen characters in the field. This is appropriate if you want to permit only people with local AFS accounts to login on your machines. A single X or other character is the most easily recognizable way to do this. To enable a user to log into the local file system even after providing an incorrect AFS password, record a standard UNIX encrypted password in the field by issuing the standard UNIX password-setting command (passwd or equivalent). Systems that use a Pluggable Authentication Module (PAM) for login and AFS authentication do not necessarily consult the local password file at all, in which case they do not use the password field to control authentication and login attempts. Instead, instructions in the PAM configuration file (on many system types, /etc/pam.conf) fill the same function. See the instructions in the OpenAFS Quick Beginnings for installing AFS-modified login utilities. local password file when not using AFS-modified login utility In cells that do not use an AFS-modified login utility, users must issue separate commands to login and authenticate, as detailed in the OpenAFS User Guide: They use the standard login program to login to the local file system, providing the password listed in the local password file (the /etc/passwd file or equivalent). They must issue the klog command to authenticate with the AFS authentication service, including its -setpag flag to associate the new tokens with a process authentication group (PAG). As mentioned in Creating Standard Files in New AFS Accounts, you can invoke the klog -setpag command in a user's .login file (or equivalent) so that the user does not have to remember to issue the command after logging in. The user still must type a password twice, once at the prompt generated by the login utility and once at the klog command's prompt. This implies that the two passwords can differ, but it is less confusing if they do not. Another effect of not using an AFS-modified login utility is that the AFS servers recognize the standard login program as the anonymous user. If the login program needs to access any AFS files (such as the .login file in a user's home directory), then the ACL that protects the file must include an entry granting the l (lookup) and r (read) permissions to the system:anyuser group. When you do not use an AFS-modified login utility, an actual (scrambled) password must appear in the local password file for each user. Use the /bin/passwd file to insert or change these passwords. It is simpler if the password in the local password file matches the AFS password, but it is not required. tokens displaying for user tokens command commands tokens listing tokens held by issuer commands klog klog command server process creating ticket (tokens) for tickets tokens tokens creating for server process authenticated identity acquiring with klog command unlog command commands unlog discarding tokens tokens discarding with unlog command Once logged in, a user can obtain a token at any time with the klog command. If a valid token already exists, the new one overwrites it. If a PAG already exists, the new token is associated with it. By default, the klog command authenticates the issuer using the identity currently logged in to the local file system. To authenticate as a different identity, use the -principal argument. To obtain a token for a foreign cell, use the -cell argument (it can be combined with the -principal argument). See the OpenAFS User Guide and the entry for the klog command in the OpenAFS Administration Reference. To discard either all tokens or the token for a particular cell, issue the unlog command. The command affects only the tokens associated with the current command shell. See the OpenAFS User Guideand the entry for the unlog command in the OpenAFS Administration Reference. To display the tokens associated with the current command shell, issue the tokens command. The following examples illustrate its output in various situations. If the issuer is not authenticated in any cell: % tokens Tokens held by the Cache Manager: --End of list-- The following shows the output for a user with AFS UID 1000 in the Example Corporation cell: % tokens Tokens held by the Cache Manager: User's (AFS ID 1000) tokens for afs@example.com [Expires Jun 2 10:00] --End of list-- The following shows the output for a user who is authenticated in Example Corporation cell, the Example Organization cell and the Example Network cell. The user has different AFS UIDs in the three cells. Tokens for the last cell are expired: % tokens Tokens held by the Cache Manager: User's (AFS ID 1000) tokens for afs@example.com [Expires Jun 2 10:00] User's (AFS ID 4286) tokens for afs@example.org [Expires Jun 3 1:34] User's (AFS ID 22) tokens for afs@example.net [>>Expired<<] --End of list-- The Kerberos version of the tokens command (the tokens.krb command), also reports information on the ticket-granting ticket, including the ticket's owner, the ticket-granting service, and the expiration date, as in the following example. Also see Support for Kerberos Authentication. % tokens.krb Tokens held by the Cache Manager: User's (AFS ID 1000) tokens for afs@example.com [Expires Jun 2 10:00] User smith's tokens for krbtgt.EXAMPLE.COM@example.com [Expires Jun 2 10:00] --End of list-- tokens setting default lifetimes for users The maximum lifetime of a user token is the smallest of the ticket lifetimes recorded in the following three Authentication Database entries. The kas examine command reports the lifetime as Max ticket lifetime. Administrators who have the ADMIN flag on their Authentication Database entry can use the -lifetime argument to the kas setfields command to set an entry's ticket lifetime. The afs entry, which corresponds to the AFS server processes. The default is 100 hours. The krbtgt.cellname entry, which corresponds to the ticket-granting ticket used internally in generating the token. The default is 720 hours (30 days). The entry for the user of the AFS-modified login utility or issuer of the klog command. The default is 25 hours for user entries created using the AFS 3.1 or later version of the Authentication Server, and 100 hours for user entries created using the AFS 3.0 version of the Authentication Server. A user can use the kas examine command to display his or her own Authentication Database entry. An AFS-modified login utility always grants a token with a lifetime calculated from the previously described three values. When issuing the klog command, a user can request a lifetime shorter than the default by using the -lifetime argument. For further information, see the OpenAFS User Guide and the klog reference page in the OpenAFS Administration Reference. password changing in AFS kpasswd command commands kpasswd kas commands setpassword commands kas setpassword Regular AFS users can change their own passwords by using either the kpasswd or kas setpassword command. The commands prompt for the current password and then twice for the new password, to screen out typing errors. Administrators who have the ADMIN flag on their Authentication Database entries can change any user's password, either by using the kpasswd command (which requires knowing the current password) or the kas setpassword command. If your cell does not use an AFS-modified login utility, remember also to change the local password, using the operating system's password-changing command. For more instructions on changing passwords, see Changing AFS Passwords. You can help to make your cell more secure by imposing restrictions on user passwords and authentication attempts. To impose the restrictions as you create an account, use the A instruction in the uss template file as described in Increasing Account Security with the A Instruction. To set or change the values on an existing account, use the kas setfields command as described in Improving Password and Authentication Security. password expiration password lifetime kas commands setfields commands kas setfields Authentication Database password lifetime, setting password restricting reuse By default, AFS passwords never expire. Limiting password lifetime can help improve security by decreasing the time the password is subject to cracking attempts. You can choose an lifetime from 1 to 254 days after the password was last changed. It automatically applies to each new password as it is set. When the user changes passwords, you can also insist that the new password is not similar to any of the 20 passwords previously used. password consequences of multiple failed authentication attempts kas commands setfields commands kas setfields authentication consequences of multiple failures Unscrupulous users can try to gain access to your AFS cell by guessing an authorized user's password. To protect against this type of attack, you can limit the number of times that a user can consecutively fail to provide the correct password. When the limit is exceeded, the authentication service refuses further authentication attempts for a specified period of time (the lockout time). To reenable authentication attempts before the lockout time expires, an administrator must issue the kas unlock command. password checking quality of kpasswd command commands kpasswd kas commands setpassword kpwvalid program In addition to settings on user's authentication accounts, you can improve security by automatically checking the quality of new user passwords. The kpasswd and kas setpassword commands pass the proposed password to a program or script called kpwvalid, if it exists. The kpwvalid performs quality checks and returns a code to indicate whether the password is acceptable. You can create your own program or modified the sample program included in the AFS distribution. See the kpwvalid reference page in the OpenAFS Administration Reference. There are several types of quality checks that can improve password quality. The password is a minimum length The password is not a word The password contains both numbers and letters Kerberos support for in AFS commands klog.krb commands pagsh.krb commands tokens.krb klog.krb command pagsh.krb command tokens.krb command If your site is using standard Kerberos authentication rather than the AFS Authentication Server, use the modified versions of the klog, pagsh, and tokens commands that support Kerberos authentication. The binaries for the modified version of these commands have the same name as the standard binaries with the addition of a .krb extension. Use either the Kerberos version or the standard command throughout the cell; do not mix the two versions. AFS Product Support can provide instructions on installing the Kerberos version of these four commands. For information on the differences between the two versions of these commands, see the OpenAFS Administration Reference. AFS incorporates several features to ensure that only authorized users gain access to data. This section summarizes the most important of them and suggests methods for improving security in your cell. security AFS features AFS security features Files in AFS are protected by the access control list (ACL) associated with their parent directory. The ACL defines which users or groups can access the data in the directory, and in what way. See Managing Access Control Lists. When an AFS client and server process communicate, each requires the other to prove its identity during mutual authentication, which involves the exchange of encrypted information that only valid parties can decrypt and respond to. For a detailed description of the mutual authentication process, see A More Detailed Look at Mutual Authentication. AFS server processes mutually authenticate both with one another and with processes that represent human users. After mutual authentication is complete, the server and client have established an authenticated connection, across which they can communicate repeatedly without having to authenticate again until the connection expires or one of the parties closes it. Authenticated connections have varying lifetimes. In order to access AFS files, users must prove their identities to the AFS authentication service by providing the correct AFS password. If the password is correct, the authentication service sends the user a token as evidence of authenticated status. See Login and Authentication in AFS. Servers assign the user identity anonymous to users and processes that do not have a valid token. The anonymous identity has only the access granted to the system:anyuser group on ACLs. Mutual authentication establishes that two parties communicating with one another are actually who they claim to be. For many functions, AFS server processes also check that the client whose identity they have verified is also authorized to make the request. Different requests require different kinds of privilege. See Three Types of Privilege. network encrypted communication in AFS encrypted network communication security encrypted network communication The AFS server processes encrypt particularly sensitive information before sending it back to clients. Even if an unauthorized party is able to eavesdrop on an authenticated connection, they cannot decipher encrypted data without the proper key. The following AFS commands encrypt data because they involve server encryption keys and passwords: The bos addkey command, which adds a server encryption key to the /usr/afs/etc/KeyFile file The bos listkeys command, which lists the server encryption keys from the /usr/afs/etc/KeyFile file The kpasswd command, which changes a password in the Authentication Database Most commands in the kas command suite In addition, the Update Server encrypts sensitive information (such as the contents of KeyFile) when distributing it. Other commands in the bos suite and the commands in the fs, pts and vos suites do not encrypt data before transmitting it. AFS uses three separate types of privilege for the reasons discussed in The Reason for Separate Privileges. Membership in the system:administrators group. Members are entitled to issue any pts command and those fs commands that set volume quota. By default, they also implicitly have the a (administer) and l (lookup) permissions on every ACL in the file tree even if the ACL does not include an entry for them. The ADMIN flag on the Authentication Database entry. An administrator with this flag can issue any kas command. Inclusion in the /usr/afs/etc/UserList file. An administrator whose username appears in this file can issue any bos, vos, or backup command (although some backup commands require additional privilege as described in Granting Administrative Privilege to Backup Operators). AFS distinguishes between authentication and authorization checking. Authentication refers to the process of proving identity. Authorization checking refers to the process of verifying that an authenticated identity is allowed to perform a certain action. AFS implements authentication at the level of connections. Each time two parties establish a new connection, they mutually authenticate. In general, each issue of an AFS command establishes a new connection between AFS server process and client. AFS implements authorization checking at the level of server machines. If authorization checking is enabled on a server machine, then all of the server processes running on it provide services only to authorized users. If authorization checking is disabled on a server machine, then all of the server processes perform any action for anyone. Obviously, disabling authorization checking is an extreme security exposure. For more information, see Managing Authentication and Authorization Requirements. security suggestions for improving You can improve the level of security in your cell by configuring user accounts, server machines, and system administrator accounts in the indicated way. Use an AFS-modified login utility, or include the -setpag flag to the klog command, to associate the credential structure that houses tokens with a PAG rather than a UNIX UID. This prevents users from inheriting someone else's tokens by assuming their UNIX identity. For further discussion, see Identifying AFS Tokens by PAG. Encourage users to issue the unlog command to destroy their tokens before logging out. This forestalls attempts to access tokens left behind kernel memory. Consider including the unlog command in every user's .logout file or equivalent. Disable authorization checking only in emergencies or for very brief periods of time. It is best to work at the console of the affected machine during this time, to prevent anyone else from accessing the machine through the keyboard. Change the AFS server encryption key on a frequent and regular schedule. Make it difficult to guess (a long string including nonalphabetic characters, for instance). Unlike user passwords, the password from which the AFS key is derived can be longer than eight characters, because it is never used during login. The kas setpassword command accepts a password hundreds of characters long. For instructions, see Managing Server Encryption Keys. As much as possible, limit the number of people who can login at a server machine's console or remotely. Imposing this limit is an extra security precaution rather than a necessity. The machine cannot serve as an AFS client in this case. Particularly limit access to the local superuser root account on a server machine. The local superuser root has free access to important administrative subdirectories of the /usr/afs directory, as described in AFS Files on the Local Disk. root superuser limiting logins As in any computing environment, server machines must be located in a secured area. Any other security measures are effectively worthless if unauthorized people can access the computer hardware. Limit the number of system administrators in your cell. Limit the use of system administrator accounts on publicly accessible workstations. Such machines are not secure, so unscrupulous users can install programs that try to steal tokens or passwords. If administrators must use publicly accessible workstations at times, they must issue the unlog command before leaving the machine. Create an administrative account for each administrator separate from the personal account, and assign AFS privileges only to the administrative account. The administrators must authenticate to the administrative accounts to perform duties that require privilege, which provides a useful audit trail as well. Administrators must not leave a machine unattended while they have valid tokens. Issue the unlog command before leaving. Use the -lifetime argument to the kas setfields command to set the token lifetime for administrative accounts to a fairly short amount of time. The default lifetime for AFS tokens is 25 hours, but 30 or 60 minutes is possibly a more reasonable lifetime for administrative tokens. The tokens for administrators who initiate AFS Backup System operations must last somewhat longer, because it can take several hours to complete some dump operations, depending on the speed of the tape device and the network connecting it to the file server machines that house the volumes is it accessing. Limit administrators' use of the telnet program. It sends unencrypted passwords across the network. Similarly, limit use of other remote programs such as rsh and rcp, which send unencrypted tokens across the network. mutual authentication distributed file system security issues shared secret server encryption key defined As in any file system, security is a prime concern in AFS. A file system that makes file sharing easy is not useful if it makes file sharing mandatory, so AFS incorporates several features that prevent unauthorized users from accessing data. Security in a networked environment is difficult because almost all procedures require transmission of information across wires that almost anyone can tap into. Also, many machines on networks are powerful enough that unscrupulous users can monitor transactions or even intercept transmissions and fake the identity of one of the participants. The most effective precaution against eavesdropping and information theft or fakery is for servers and clients to accept the claimed identity of the other party only with sufficient proof. In other words, the nature of the network forces all parties on the network to assume that the other party in a transaction is not genuine until proven so. Mutual authentication is the means through which parties prove their genuineness. Because the measures needed to prevent fakery must be quite sophisticated, the implementation of mutual authentication procedures is complex. The underlying concept is simple, however: parties prove their identities by demonstrating knowledge of a shared secret. A shared secret is a piece of information known only to the parties who are mutually authenticating (they can sometimes learn it in the first place from a trusted third party or some other source). The party who originates the transaction presents the shared secret and refuses to accept the other party as valid until it shows that it knows the secret too. The most common form of shared secret in AFS transactions is the encryption key, also referred to simply as a key. The two parties use their shared key to encrypt the packets of information they send and to decrypt the ones they receive. Encryption using keys actually serves two related purposes. First, it protects messages as they cross the network, preventing anyone who does not know the key from eavesdropping. Second, ability to encrypt and decrypt messages successfully indicates that the parties are using the key (it is their shared secret). If they are using different keys, messages remain scrambled and unintelligible after decryption. The following sections describe AFS's mutual authentication procedures in more detail. Feel free to skip these sections if you are not interested in the mutual authentication process. Simple mutual authentication involves only one encryption key and two parties, generally a client and server. The client contacts the server by sending a challenge message encrypted with a key known only to the two of them. The server decrypts the message using its key, which is the same as the client's if they really do share the same secret. The server responds to the challenge and uses its key to encrypt its response. The client uses its key to decrypt the server's response, and if it is correct, then the client can be sure that the server is genuine: only someone who knows the same key as the client can decrypt the challenge and answer it correctly. On its side, the server concludes that the client is genuine because the challenge message made sense when the server decrypted it. AFS uses simple mutual authentication to verify user identities during the first part of the login procedure. In that case, the key is based on the user's password. Complex mutual authentication involves three encryption keys and three parties. All secure AFS transactions (except the first part of the login process) employ complex mutual authentication. ticket-granter server encryption key tokens data in When a client wishes to communicate with a server, it first contacts a third party called a ticket-granter. The ticket-granter and the client mutually authenticate using the simple procedure. When they finish, the ticket-granter gives the client a server ticket (or simply ticket) as proof that it (the ticket-granter) has preverified the identity of the client. The ticket-granter encrypts the ticket with the first of the three keys, called the server encryption key because it is known only to the ticket-granter and the server the client wants to contact. The client does not know this key. The ticket-granter sends several other pieces of information along with the ticket. They enable the client to use the ticket effectively despite being unable to decrypt the ticket itself. Along with the ticket, the items constitute a token: A session key, which is the second encryption key involved in mutual authentication. The ticket-granter invents the session key at random as the shared secret between client and server. For reasons explained further below, the ticket-granter also puts a copy of the session key inside the ticket. The client and server use the session key to encrypt messages they send to one another during their transactions. The ticket-granter invents a different session key for each connection between a client and a server (there can be several transactions during a single connection). session key The name of the server for which the ticket is valid (and so which server encryption key encrypts the ticket itself). A ticket lifetime indicator. The default lifetime of AFS server tickets is 100 hours. If the client wants to contact the server again after the ticket expires, it must contact the ticket-granter to get a new ticket. The ticket-granter seals the entire token with the third key involved in complex mutual authentication--the key known only to it (the ticket-granter) and the client. In some cases, this third key is derived from the password of the human user whom the client represents. Now that the client has a valid server ticket, it is ready to contact the server. It sends the server two things: The server ticket. This is encrypted with the server encryption key. Its request message, encrypted with the session key. Encrypting the message protects it as it crosses the network, since only the server/client pair for whom the ticket-granter invented the session key know it. At this point, the server does not know the session key, because the ticket-granter just created it. However, the ticket-granter put a copy of the session key inside the ticket. The server uses the server encryption key to decrypts the ticket and learns the session key. It then uses the session key to decrypt the client's request message. It generates a response and sends it to the client. It encrypts the response with the session key to protect it as it crosses the network. This step is the heart of mutual authentication between client and server, because it proves to both parties that they know the same secret: The server concludes that the client is authorized to make a request because the request message makes sense when the server decrypts it using the session key. If the client uses a different session key than the one the server finds inside the ticket, then the request message remains unintelligible even after decryption. The two copies of the session key (the one inside the ticket and the one the client used) can only be the same if they both came from the ticket-granter. The client cannot fake knowledge of the session key because it cannot look inside the ticket, sealed as it is with the server encryption key known only to the server and the ticket-granter. The server trusts the ticket-granter to give tokens only to clients with whom it (the ticket-granter) has authenticated, so the server decides the client is legitimate. (Note that there is no direct communication between the ticket-granter and the server, even though their relationship is central to ticket-based mutual authentication. They interact only indirectly, via the client's possession of a ticket sealed with their shared secret.) The client concludes that the server is genuine and trusts the response it gets back from the server, because the response makes sense after the client decrypts it using the session key. This indicates that the server encrypted the response with the same session key as the client knows. The only way for the server to learn that matching session key is to decrypt the ticket first. The server can only decrypt the ticket because it shares the secret of the server encryption key with the ticket-granter. The client trusts the ticket-granter to give out tickets only for legitimate servers, so the client accepts a server that can decrypt the ticket as genuine, and accepts its response. AFS provides two related facilities that help the administrator back up AFS data: backup volumes and the AFS Backup System. The first facility is the backup volume, which you create by cloning a read/write volume. The backup volume is read-only and so preserves the state of the read/write volume at the time the clone is made. Backup volumes can ease administration if you mount them in the file system and make their contents available to users. For example, it often makes sense to mount the backup version of each user volume as a subdirectory of the user's home directory. A conventional name for this mount point is OldFiles. Create a new version of the backup volume (that is, reclone the read/write) once a day to capture any changes that were made since the previous backup. If a user accidentally removes or changes data, the user can restore it from the backup volume, rather than having to ask you to restore it. The OpenAFS User Guide does not mention backup volumes, so regular users do not know about them if you decide not to use them. This implies that if you do make backup versions of user volumes, you need to tell your users about how the backup works and where you have mounted it. Users are often concerned that the data in a backup volume counts against their volume quota and some of them even want to remove the OldFiles mount point. It does not, because the backup volume is a separate volume. The only amount of space it uses in the user's volume is the amount needed for the mount point, which is about the same as the amount needed for a standard directory element. Backup volumes are discussed in detail in Creating Backup Volumes. Backup volumes can reduce restoration requests, but they reside on disk and so do not protect data from loss due to hardware failure. Like any file system, AFS is vulnerable to this sort of data loss. To protect your cell's users from permanent loss of data, you are strongly urged to back up your file system to tape on a regular and frequent schedule. The AFS Backup System is available to ease the administration and performance of backups. For detailed information about the AFS Backup System, see Configuring the AFS Backup System and Backing Up and Restoring AFS Data. Users of NFS client machines can access the AFS filespace by mounting the /afs directory of an AFS client machine that is running the NFS/AFS Translator. This is a particular advantage in cells already running NFS who want to access AFS using client machines for which AFS is not available. See Appendix A, Managing the NFS/AFS Translator.