In a world without Windows, PAM guards the doors.
If you are into British detective fiction and names like Sherlock Holmes, Sexton Blake, Mr. J. G. Reeder, Miss Marple, Hercule Poirot, Father Brown, Dr. John Evelyn Thorndyke and Lord Peter Wimsey mean anything to you, you also probably will recognize E. W. Hornung's (brother-in-law to Sir Arthur Conan Doyle, the creator of Sherlock Holmes) character: the white-glove thief, Raffles. In the “A Jubilee Present” short story, the thief is fascinated with an antique gold cup, displayed at the British Museum. Upon finding only one guard, Raffles questions him on the perceived lack of security and gets the confident answer, “You see, sir, it's early as yet; in a few minutes these here rooms will fill up; and there's safety in numbers, as they say.” With Linux, rather than security by numbers (which eventually is no good for the poor guard; see Resources for a link to the complete story), security is managed by Pluggable Authentication Modules (PAM). In this article, we study PAM's features, configuration and usage.
Let's start at the beginning and consider how an application authenticates a user. Without a common, basic mechanism, each application would need to be programmed with particular authentication logic, such as checking the /etc/passwd for a valid user and password. But, what if you have several different applications that need authentication? Do you include the same specific logic in all of them? And, what if your security requirements vary? Would you then have to modify and recompile all those applications? This wouldn't be a practical method and surely would become a vulnerability. How would you be sure that all applications were duly updated and correctly implemented your new specifications?
The PAM Project provides a solution by adding an extra layer. Programs that need authentication use a standard library or API (Application Programming Interface), and system administrators can configure what checks will be done by that library separately. (Checks are implemented via independent modules; you even can program your own modules.) This way, you can change your security checks dynamically, and all utilities will follow your new rules automatically. In other words, you can modify the authentication mechanism used by any PAM-aware application, without ever touching the application itself. For programmers, this also is a good thing, because they need not be concerned with the mechanisms that will be used. Simply by using the PAM libraries, whenever the application is run, the appropriate checks will be made (Figure 1).
The PAM library breaks down authentication in four areas or groups (Table 1). Note that all applications won't always require the four previous actions. For example, the passwd command will require only the last group. (Quick tip: how can you learn whether an application uses PAM? Use ldd to print the shared libraries required by the program, and check for libpam.so; see Listing 1 for an example.)
Figure 1. Whenever an application does an authentication request, the PAM library executes whatever modules are specified in the configuration file and decides whether to approve (success) or reject (failure) the request.
Table 1. PAM has four groups of checks, organized as stacks. The groups that will be used depend on what the user requires.
| auth | Related to user identification, such as when a user needs to enter a password. This is usually the first set of checks. |
| account | Has to do with user account management, including checking whether a password has expired or whether there are time-access restrictions. Once users have been identified by the authentication modules, the account modules will determine whether they can be granted access. |
| session | Deals with connection management, with actions such as logging entries or activities, or doing some cleanup actions after the session ends. |
| password | Includes functions such as updating users' passwords. |
Table 2. For each stack, modules are executed in sequence, depending on their control flags. You must specify whether the corresponding check is mandatory, optional and so on.
| required | This module must end successfully. If it doesn't, the overall result will be failure. If all modules are labeled as required, any single failure will deny authentication, although the other modules in the stack will be tried anyway. |
| requisite | Works like required, but in case of failure, returns immediately, without going through the rest of the stack. |
| sufficient | If this module ends successfully, other modules will be skipped, and the overall result will be successful. |
| optional | If this module fails, the overall result will depend upon the other modules. If there are no required or sufficient modules, at least one optional module should end successfully to allow authentication. |
For each service (such as login or SSH), you must define which checks will be done for each group. That list of actions is called a stack. Depending on the results of the actions in each stack, users will succeed or fail, and whatever they attempted to do will be allowed or rejected. You can specify each action in the stack for each service using a specific file at /etc/pam.d (the more current method) or by editing the single, catchall file /etc/pam.conf (the older method); in this article, we use the former method.
Each stack is built out of modules, executed sequentially in the given order. For each module, you can specify whether it's necessary (failure automatically denies access), sufficient (success automatically grants access) or optative (allows for alternative checks). Table 2 shows the actual control flags. The file for each service consists of a list of rules, each on its own line. (Longer lines can be split by ending with a \, but this is seldom required.) Lines that start with a hash character (#) are considered to be comments and, thus, are ignored. Each rule contains three fields: the context area (Table 1), the control flag (Table 2) and the module that will be run, along with possible (optional) extra parameters. Thus, the specification for the PAM checks for login would be found in the /etc/pam.d/login file.
The control flag field actually can be more complicated, but I won't cover all the details here. See Resources if you are interested. Also, you can use include, as in auth include common-account, which means to include rules from other files.
There is a special, catchall service called other, that is used for services without specific rules. A good start from a security point of view would be creating /etc/pam.d/other, as shown in Listing 2. All attempts are denied, and a warning is sent to the administrator. If you want to be more forgiving, substitute pam_unix2.so for pam_deny.so, and then the standard Linux authentication method will be used, although a warning will still be sent (Listing 3). If you don't care about security, substitute pam_permit.so instead, which allows entry to everybody, but don't say I didn't warn you.
Finally, give the files in /etc/pam.d a quick once-over. If you find configuration files for applications you don't use, simply rename the files, so PAM will fall back to your “other” configuration. Should you discover later that you really needed the application, change the configuration file back to its original name, and everything will be okay again.
To get a handle on all this, let's consider an actual application. I wanted to be able to access my machine remotely with SSH, but I didn't want to allow any other users (Listing 4). So, I configured my /etc/pam.d/sshd file. See the Modules, Modules Everywhere sidebar for more details on these and other modules. Here are some of the modules I used:
pam_unix2.so: provides traditional password, rights, session and password-changing methods, in the classic UNIX way.
pam_nologin.so: disallows login if the file /etc/nologin exists.
pam_access.so: implements extra rules for access control (more later in this article on how I used this).
pam_limits.so: enforces limits for users or groups according to the file /etc/security/limits.conf.
pam_umask.so: sets the file mode creation mask for the current environment (do info umask for more information).
pam_pwcheck: enforces password-strength checks (more details on further uses of this module later in this article).
If you check your own /etc/pam.d/sshd file, it probably will look like this, except for the pam_access module, which is the interesting part. This module implements added security controls based on the /etc/security/access.conf file. I edited it in order to specify who could access my machine (Listing 5). The first line means that anybody (ALL) can log in to my machine from within the internal network at home. The second line allows the remoteKereki user to access my machine from anywhere in the world, and the final line is a catchall that disables access to anybody not included specifically in these lines. I created the remoteKereki user with minimum rights to allow myself entry to the machine, and then I execute su and work as myself or even as root, if needed. If people guess the correct password for remoteKereki, it won't help them much, because attackers still will have to guess the password for the other, more useful, users. As it is, it provides an extra barrier before intruders can do serious damage.
I had to modify /etc/ssh/sshd_config by adding a line UsePAM yes, so sshd would use the PAM configuration. I had to restart SSH with /etc/init.d/sshd restart so the configuration would be used. For even more secure connections, you also could change the SSH standard port (22) to a different value, forbid root remote logins and limit retries to hinder brute-force attacks, but those topics are beyond the scope of this article. Do man ssh_config for more details.
Left on their own, most users will (trustingly and unknowingly) use easily guessable and never-changed passwords, simplifying the job for intruders. With PAM, you can enforce several good practices for password management by using the password stack and the pam_pwcheck.so module. This module does several checks on the strength of your password:
Is the new password too short?
Is the new password too similar to the old one?
Is the new password merely the old password, reversed or rotated (for example, safe123 and 123safe)?
Is the new password the same as the old one, with only case changes (such as sEcReT and SEcrET)?
Was the new password already used before? (Old passwords are stored in the /etc/security/opasswd file.)
You can add several parameters to the module (do man pam_pwcheck for complete documentation) for extra rules, such as:
minlen=aNumber: specifies the minimum length (by default, five characters) for the new password. If you set it to zero, all password lengths are accepted.
cracklib=pathToDictionaries: allows use of the cracklib library for password checks. If the new password is in a dictionary, a simple brute-force attack quickly will guess it.
tries=aNumber: sets how many attempts to allow, if previous attempts were rejected because they were too easy.
remember=aNumber: defines how many previous passwords will be remembered.
Another module provides similar functionality, pam_cracklib.so, but it has some different parameters. For example, you might specify how many characters must differ between your old and new password and whether you want to include digits, uppercase, lowercase and nonalphabetic characters. Do man pam_cracklib for more information.
There might be security in numbers (as the poor British Museum guard thought when he tried to deter Raffles from stealing the cup), but for Linux, PAM is the way to go. Without even resorting to rolling out your own modules, you can add plenty of flexibility to your security by setting up a few configuration files and rest assured that those rules will be obeyed globally.