Cover V01, I01
Listing 1
Listing 2
Listing 3
Sidebar 1
Sidebar 2


How UNIX Password Controls Work

Chris Hare

One of the problems with writing articles about security-related issues is the fear that someone will use what you have written to attack a system. The good side is that the more system administrators know about how the tools work, the better able they will be to protect their systems from attack.

This article focuses on the UNIX password encryption system. I describe how the system works and discuss some of the methods the bad guys use to circumvent it.

Ensuring that users select good passwords is the administrator's primary mechanism for controlling access to the machine. This is the most common breakpoint in any computing environment where the password is set by the user. The opposite scenario, in which the user has no control over password assignment, often results in the user writing the password down somewhere, immediately snapping the security chain.

Good passwords are at least six characters long and contain a mix of upper- and lower-case characters, a number, and, potentially, a control character. Many versions on UNIX also accept a space as a valid character in a password. Bad passwords violate one or more of these guidelines. In addition, a password consisting of any piece of information that can be found out about a user, or of the name of a visible object on the user's desk, forms a gaping hole just waiting to be exploited.

Communicating the guidelines for creating good passwords to users is one side of this issue; getting the operating system to enforce them is the other.

Password Aging Controls

Many versions of UNIX provide for password aging. This mechanism controls when users can change their passwords by means of a value inserted into the password file after the encrypted password. This value defines the minimum period of time that must pass before a user can change a password and the maximum period of time that can elapse before the password has expired. A timeline provides a good representation of the concept.

0          CANNOT       min       MAY        max      MUST

Password aging control information is stored with the encrypted password as a series of printable characters. The controls are included after the password, preceded by a comma. The characters represent:

  • The maximum number of weeks the password is valid
  • The minimum number of weeks that must elapse before the user can change the password again
  • When the password was most recently changed.

    See the sidebar "Using Password-ese" for examples of this mechanism.

    The aging control mechanisms recognize two special conditions: one which forces the user to change the password on next login, and one which prevents the user from being able to change it.

    To force a user to change a password, as in the case of a new user, the user's password field would be modified to include a comma followed by two periods representing the maximum and minimum time frames. These values force the user to change the password at the next login. When the password has been changed, the "force" control information is removed from the password entry.

    The second special case prohibits a user from being able to change a password. This condition is established by setting the maximum value to be less than the minimum value (i.e., first < second). In this case, the user is informed that the password cannot be changed.

    Some of the newer, more secure versions of UNIX now on the market use the term "password lifetime." This denotes a grace period after the maximum time period has elapsed during which the user may still login using the expired password. Once the lifetime has been reached, the account will be disabled.

    The password lifetime mechanism doesn't prevent a user from changing a password and then changing it back to the old one later. Only a few UNIX system versions keep track of the passwords a user has used. The actual process of implementing password aging is version dependent. To implement it on your system, consult your system documentation.

    The program in Listing 1,, provides advance notice of password expiration dates, so that users can be prepared for the day when the system demands a new password. Note, however, that this version of the program is for standard System V UNIX, where a shadow password file is not used. addresses the aging mechanism that was implemented in versions of UNIX up to System V Release 3.2. At this level, some variations took place in the interests of increasing system security. The AT&T camp moved to using the /etc/shadow file for storing the password information, while the SCO camp moved to using the Trusted Computing Base facilities form SecureWare. In SCO's UNIX 3.2v4 product, the aging control information may be in /etc/passwd, /etc/shadow, or the Trusted Computing Base files, depending upon which level of security you configure.

    How Passwords Are Encrypted

    The real issue at hand is how the UNIX password mechanism is implemented. At one time, the passwords were stored in an unecrypted format, and only the administrator and system software had access to the file. The problem came when the password file (/etc/passwd) was being edited. Since most editors create a temporary file for editing purposes, during this time the file would be world-readable, giving the passwords for all of the accounts away.

    As a result, a method of encrypting the passwords using a one-way encryption algorithm and storing the encrypted values instead of the clear text was provided. However, the security of the system is only as good as the encryption method chosen.

    When a user logs into a UNIX system, the getty program prompts for username and then executes the login program. The login program prompts for the password, but does not decrypt it. In fact, the login program encrypts the password, and then compares the newly encrypted value to the one stored in /etc/passwd. If they match, then the user supplied the correct one.

    So how does it work? The UNIX encryption method for password encryption is accessed via the system call crypt(3) (because of U.S. licensing issues, the crypt routines may not be available on your machine).

    The actual value stored in /etc/passwd results from using the user's password to encrypt a 64-bit block of zeroes via the crypt(3) call. The "clear text" is the user's password, which is the key to the operation. The text being encrypted is 64 bits of zeroes, and the resulting "cipher text" is the encrypted password.

    The crypt(3) algorithm is based upon the Data Encryption Standard (DES) developed by the National Institute of Standards and Technology (NIST). In normal operation, according to the DES, a 56-bit key, such as eight 7-bit characters, is used to encrypt the original text, commonly called "clear text." This clear text is typically 64 bits in length. The resulting cipher text cannot be decrypted easily without knowledge of the original key.

    The UNIX crypt(3) call uses a modified version of this method by stating that the clear text be encrypted in a block of zeroes. The process is complicated by taking the resulting cipher text and encrypting it again with user's password as the key. This process is performed 25 times, and, when it is complete, the resulting 64 bits are split into 11 printable characters and then saved in the password file.

    Despite the fact that source for crypt can be obtained from many vendors (even though the distribution of this is limited outside the United States), there is no known method for translating the cipher text or encrypted value back to its original clear text.

    Robert Morris, Sr., and Ken Thompson, who originally implemented the UNIX crypt(3) technology, were afraid that with the advent of hardware DES chips, the security of the UNIX system could easily be bypassed. By using a "grain of salt," they managed to disarm this threat.

    The "grain of salt" is a 12-bit number which is used to modify the result of the DES function. The value of this number ranges from 0 to 4095. The result is that each possible password could be represented in any one of 4096 ways in the password file. Multiple users on the same machine could be using the same password, and no one, including the system manager, would be the wiser.

    When a user runs the /bin/passwd program to establish a new password, the /bin/passwd program picks a salt value, based upon the time of day, which is then used to modify the user's password.

    To prevent problems in the unlikely case that the user's next login generates the same salt value, UNIX stores the salt in /etc/passwd as well. In fact, this value makes up the first two characters of the encrypted password.

    The Validity of the Password Cracker

    A password cracker is a program that attempts to "guess" the passwords in the /etc/passwd file by comparing them to words in a dictionary. The success of the cracker program is dependent upon the CPU resources, the quality of the dictionary, and the fact that the user has a copy of /etc/passwd.

    It is fairly easy to write a password cracker: approximately 60 lines of C or 40 lines of Perl code will do it. If, as part of your attempt to provide for better passwords on your system, you decide to write such a program, be warned that you may be inviting disaster. Since an efficient cracker program could be stolen and subsequently used to gain access to other machines, if your program works, you may have further compromised the security of your machine. (Even as I was writing this article, one such program was posted to comp.sources.misc. The author clearly states in his documentation that he assumes no responsibility for the use of his program, but yet claims that it is highly efficient.)

    A much better approach is to protect the system using the logical security tools.

    Making /bin/passwd More Intelligent

    Over the last couple of years, as secure versions of UNIX have evolved, /bin/passwd has undergone changes. Indeed, some of the older versions were rather intelligent, and could pick up some of the common problems, such as passwords that were too short, didn't include numbers, etc.

    A number of UNIX manufacturers now provide passwd programs that will validate the usability of the password. These programs check for length, username, circular shifts, and common words (without implementing a dictionary examination).

    In addition, /bin/passwd can be configured to generate passwords and to check for obviousness in those provided by the user.

    Unfortunately, many UNIX systems still in use today do not offer these advantages, and therefore must be coaxed to provide them, so we enter the realm of system security for the programmer.

    Programmer Security

    Rewriting /bin/passwd is not a simple job: password aging, salts, and file access are just three of the issues involved. Neither is it really feasible to build a wrapper around the /bin/passwd program.

    Unlike many programs, which simply read from standard input to get data from the user and the keyboard, /bin/passwd reads from the generic device /dev/tty. This strategy protects the system from intruder programs that attempt to make /bin/passwd "believe" that input is coming from the keyboard.

    Furthermore, the wrapper could not be written as a shell script, but would require either the use of C (or another language that allows for the use of setuid system calls) or the SUID bit on the executable's permissions. When the program is not a binary executable, SUID bits are ignored because there is no way to ensure that the code being executed is what the programmer wrote, or that /bin/sh is not itself a SUID program running the commands with the authority of the user running the the shell.

    The book Programming PERL, by Larry Wall and Randal Schwartz, has such a program. It is 15 pages of PERL source code, and makes a valiant effort to provide for all of the possibilities that should be evaluated. The program in Listing 2 (with its configuration file, pwd.cfg, in Listing 3), named, is an example of what needs to be done to make /bin/passwd more intelligent. If you are seriously considering installing a "smarter" version of /bin/passwd, I suggest using the implementation from Programming PERL. The program I present in Listing 2 is an example only, and while it works, I wouldn't bet my life on it.

    The examples from Programming PERL are available from a number of places, including the host, in /published/nutshell/perl/perl.tar.Z.

    I would prefer to see UNIX vendors increase the level of thoroughness in /bin/passwd rather than leave it to users to do. Too many things can go wrong, and quite a lot about the /bin/passwd program isn't documented. The Santa Cruz Operation, for one, and AT&T with the Multi-Level Security (MLS) features, all provide smarter versions of /bin/passwd and some form of additional password security.

    Above all, be wise about the passwords you choose -- they are the most vital link to system access.

    About the Author

    Chris Hare is Ottawa Technical Services Manager for Choreo Systems, Inc. He has worked in the UNIX environment since 1986 and in 1988 became one the first SCO authorized instructors in Canada. He teaches UNIX introductory, system administration, and programming classes. His current focus is on networking, Perl, and X.