SYNOPSIS

To execute a command:

      super [ -r reqpath] command [ args ]

      super [ -r reqpath] -o path [ args ]

      command [ args ]

To list available commands:

      super [-H|-f] [-S]

For usage and/or version information:

      super [-h] [-V]

For debugging and development:

      super -b

      super -c [ superfile ]

      super [-d|-D|-t] [-S] [MasqOptions] [-H|-f|command...]

      MasqOptions:

-F file
-T hh:mm/dayname
-G gid
-U uid
-M mach

DESCRIPTION

Super allows specified users to execute scripts (or other commands) as if they were root; or it can set the uid, gid, and/or supplementary groups on a per-command basis before executing the command. It is intended to be a secure alternative to making scripts setuid root. Super also allows ordinary users to supply commands for execution by others; these execute with the uid, gid, and groups of the user offering the command.

Super consults a ``super.tab'' file to see if the user is allowed to execute the requested command. If permission is granted, super will exec pgm [ args ], where pgm is the program that is associated with this command. (Root is allowed execution by default, but can still be denied if a rule excludes root. Ordinary users are disallowed execution by default.)

The most common sort of entry in a super.tab file pairs a simple command with a pgm path. But in fact, the command in the super.tab file is actually treated as a pattern, and any user-entered command that matches this pattern causes the associated pgm to be executed. If the listed pgm contains an asterisk, then the asterisk is replaced with the command entered by the user. One use of this is to let any program in a certain directory be executed by a user. For example, if the entry contains the command/pgm pairs
      CommandPatternProgram

      scripts/*  → /usr/local/super/*

         *       → /usr/local/somedir/*
then the translations made are

      User's CommandExecuted Program

      scripts/xyz  →/usr/local/super/scripts/xyz

      xyz        → /usr/local/somedir/xyz

Some commands can only be run after the user enters his or her password.
These commands can then be run multiple times until some expiration time, at which point the password needs to be re-entered. The list of password-requiring commands and the password durations are set in the same file that records the valid users for each command.

If command is a symbolic link (or hard link, too) to the super program, then typing
      % command args
is equivalent to typing
      % super command args
(The command must not be super, or super will not recognize that it's being invoked via a link.)

Super without any arguments will display the list of commands that may be executed by the user.

For security, the following precautions are taken before exec'ing:

(a)

all descriptors save 0,1,2 are closed;

(b)

all of the user's environment variables are discarded, save for TERM, LINES, and COLUMNS. If TERM contains any characters other than {-/:+._a-zA-Z0-9}, it is discarded. If LINES or COLUMNS contains any characters other than [0-9], it is discarded. To these are added reasonable values for:

USER and LOGNAME: both are set to the username associated with the real uid of the program running under super;

HOME: set to the login directory of the user running super;

ORIG_USER, ORIG_LOGNAME, ORIG_HOME: the values of USER, LOGNAME, and HOME that refer to the user who invoked super. (These values are computed by super, and are not the values set by the caller, so they are a reliable description of the caller. These are normally the same values as USER, LOGNAME, and HOME, but they will differ if the super command changes uid or gid before executing the program.);

IFS: set to blank, tab, newline;

PATH: set to /bin:/usr/bin.

SUPERCMD: set to command.

additional environment variables as specified in the super.tab file (see below).

(c)

all signal handling is reset to the default.

If Super is executed without arguments, it will print the commands that the user may execute, one command per line. Super -H prints a long-winded description of each command that the user may execute.
The CmdPat is the command pattern, and FullPath is the full path that will be executed. The super.tab file can specify initial arguments that that precede any user-supplied arguments; these arguments, if any, are printed after the FullPath column.

User-Defined Super.tab Files

Ordinary users can supply their own super files. This lets users give well-controlled setuid/setgid access to their programs: the user who offers the program gets the assurance of safe IFS settings, safe environment variable settings, etc; and the user who executes the program knows that it will execute under the uid and gid of the offering user. If a command is entered in the form

super loginname:cmd

super looks for cmd in the file .supertab, in the home directory of account loginname. The cmd will be executed using the uid, gid, and supplementary groups (if any) of user loginname.

The usual super options (such as -H) can be applied to a user's .supertab file. For example, help information about one command can be had by using:
      super -H loginname:cmd
Likewise, help information about all of loginname's commands can be obtained with:
      super -H loginname:

Links to per-user commands can be created and used in a manner similar to making symlinks to super itself. If command is a symbolic link to a user's .supertab file, and that .supertab file is

(a) executable, and

(b) begins with

#! /path/to/super -o

then the following pair are completely equivalent:

% super loginname:command
% command

If the #!-line would be longer than the typical Unix limit of 32 characters, you can instead start the .supertab file with:

#! /bin/sh
# Keep this backslash -> \
     exec /long/path/to/the/super/executable -o $0 ${1+"$@"}

(The above takes advantage of the fact that super allows comments to be backslash-continued, but the shell doesn't.)

Per-user .supertab linking works as follows: if /path/to/xyz is a symlink to some user's .supertab file, and the .supertab file begins with #! /path/to/super -o, then the shell will invoke super with arguments something like

super -o /path/to/xyz [args]

Super checks that /path/to/xyz is a link to a real .supertab file, and then always turns the last part of the path (here xyz) into the command to execute.

** Security Warning **

Note that if you use symlinks to a per-user .supertab file, then you must trust that the .supertab file will actually execute a super command, instead of doing something nasty. That is because super itself isn't invoked until the shell has opened the .supertab file and done whatever the .supertab file tells it to do. By contrast, the direct command super loginname:cmd doesn't involve any shell processing of the .supertab file.

REGULAR OPTIONS

-V

Print the super version number.

-S

When super prompts for a password, this forces it to prompt on stdin, even if the default (/dev/tty) is readable and writable. Note: This only applies to password-type authentication --- that is, the older type of authentication wherein super itself prompts for the password; PAM authentication is handled by your system's PAM modules.

-f

This requests a list of available commands in a terse format useful for processing by scripts. (-f stands for facts, as in ``just the facts, m'am'').

-rreqpath

Tells super to generate an error if the program associated with this command is not reqpath. This helps you write scripts that ensure that super only executes what they expect it to execute. See step 4 of the section, ``Creating Super Scripts'', for an example of its use.

-H

Causes super to print a verbose listing of the commands available to the user. It prints both the command and its translation to a program pgm. If the displayed pgm contains an asterisk, then the actual program executed is formed by replacing the asterisk with the command entered by the user. The following examples show the kinds of lines that may be displayed with the -H option:

Example 1.
      super skill → /usr/local/bin/skill

Typing super skill will execute /usr/local/bin/skill.

Example 2.
      super {lp*} → /usr/bin/*

This example contains asterisks on both the left and right sides. The left side shows the valid pattern you must match to execute the command shown on the right-hand side. Usually, the right-hand side has no asterisk, just a full path to a command to execute. If there is an asterisk present, it is replaced by the command you entered, thereby forming the actual executed command. Thus, if you type super lpxxx (where xxx is any string), super will execute /usr/bin/lpxxx.

Example 3.
      super {co*} → /usr/bin/compress

The asterisk on the left-hand side means you can enter super coxxx (where xxx is any string), but since the right-hand side doesn't contain an asterisk, coxxx will always execute /usr/bin/compress.

-t

This enables ``test'' mode. It does all normal checks except for those requiring user input (passwords and variables that the user must enter), but doesn't execute any command. Instead, it exits with status code 0 if the command is ok to execute, else 1. All normal error message output is generated in the usual way, but no special debug messages are generated. Thus, it is a useful means for a script to check if a command is likely to work, and hence reasonable to exec super. Let's say that a script /usr/local/bin/foo wants to invoke itself using super foo (See the section ``Creating Super Scripts'' for how to avoid infinite loops when doing this!) the script can use the -r option to ensure that super foo refers to the correct file, and it can use test mode to ensure that super foo is a valid command:

prog=`basename $0`
/usr/local/bin/super -t -r $0 $prog
case $? in
0 ) exec /usr/local/bin/super -t -r $0 $prog ;;
* ) echo "Super $prog doesn't work!"
    ... So take appropriate action ...
    ;;
esac

DEBUG AND DEVELOPMENT OPTIONS

These options are useful when creating and debugging super.tab files. They have little or no value to the everyday user. With the exception of the -b option, they can be combined with the regular options, above.

-b

Print the names and values of built-in variables, then exit. Useful for administrators to learn the values against which builtin variables can be tested.

-c[superfile]

Tells super to check the syntax of the entries in the superfile, but not to execute any command. If no superfile is given, the regular super.tab is checked. The exit code is 0 if the file's syntax is ok; otherwise the exit code is 1 (and an error message is printed). After modifying a super file, you should use this option to check its integrity.

Note that super -c isn't a complete check that you've correctly set up an entry, because you can create syntactically valid entries that don't do exactly what you want. Therefore, you should also use super -d cmd to make sure that the command you've entered will be executed with the correct arguments, uid, gid, umask, and so on.

-d

This enables debug mode, in which case (a) debugging information is printed while checking a user for validity, and (b) the command isn't actually executed. Useful to check if a new entry in the super.tab file (see below) has been handled properly.

-D

Same as -d, plus prints more information about variables defined in the super.tab file.

-Fsuperfile

This option is only used for debugging, and lets you test a superfile before installing it. No command will actually be executed. It also turns on a non-verbose debugging, showing the matched command names and reasons for accepting or rejecting the command.

-Ggid

This option is also used for debugging, and tells super to act as if the caller's groupid or groupname was gid. It carries the same restrictions and debug info as the -F option.

-Uuid

This option is also used for debugging, and tells super to act as if the caller's uid or username was uid. It carries the same restrictions and debug info as the -F option.

-Mmach

This option is also used for debugging, and tells super to act as if the caller's host (machine) was mach. It carries the same restrictions and debug info as the -F option.

-Thh:mm/dayname

This option is also used for debugging, and tells super to act as if the execution time is hh:mm/dayname. This lets you check if a time specification in the super.tab file is properly restricting execution. It carries the same restrictions and debug info as the -F option.

FILES

/etc/super.tab

contains the list of commands that super may execute, along with the names of the user/group combinations who may execute each command. The valid-user line can restrict use to particular users or groups on different hosts, so a single super.tab file can be used across a network.

/run/superstamps/username

is used as a timestamp for the last time that the user entered his or her password.

CREATING SUPER SCRIPTS

You must be exceedingly careful when writing scripts for super. A surprising variety of ordinary commands can, when run setuid-root, be exploited for nasty purposes. Always make your scripts do as little as possible, and give the user as few options as possible.

Think twice about side-effects and alternative uses of these scripts. For instance, make sure your script doesn't quietly invoke the user's .cshrc or similar file. Or, you might write a script to allow users to mount cd-rom's by executing mount(8). But if you don't write it carefully, a user could mount a floppy disk containing, say, a setuid-root shell.

Security issues aside, here are some hints on creating super scripts:

1.

Scripts must begin with #! interpreter-path.

2.

Some variants of csh will not run setuid scripts unless the -b flag (force a "break" from option processing) is set:
      #!/bin/csh -fb
Similarly, if your super.tab file starts a shell such as csh or tcsh, you may want to include the -b option in the super.tab file, so that you don't have to remember to type it on the command line every time; use a line like the following in the super.tab file:
      SHELL "/usr/bin/csh -fb" some_priv_user
N.B. This is by way of example only; it's not a very good idea to really let somebody become root without any password check.

3.

Better still, avoid csh scripts entirely -- they are harder to write safely than Bourne-shell scripts.

4.

It's nice to make the super call transparent to users, so that they can type
      % cdmount args
instead of
      % super cdmount args
You can make a script super itself by beginning the script in the following way:

#!/bin/sh
prog=`basename $0`
test "X$SUPERCMD" = "X$prog" ||
                    exec /usr/local/bin/super -r $0 $prog ${1+"$@"}

Here, the path that is exec'd should be replaced with the path at your site that leads to super. The option -r$0 is a sanity-check option: it tells super that it's an error if ``super $prog'' doesn't execute ``$0'', ie this self-same program. (Also, see the -t option for how a script can check that super $prog will work before doing an exec super.)

5.

Some programs need certain directories in the path. Your super scripts may have to add directories like /etc or /usr/etc to make commands work. For instance, SunOS 4.1 needs /usr/etc in the path before it can mount filesystems of type ``hsfs''.

6.

By default, super only changes the effective uid. Some programs (e.g. exportfs under SunOS 4.1.x) require the real uid to be root. In that case, you should put an option like ``uid=root'' or ``u+g=root'' into the super.tab file.

AUTHOR

Will Deich
[email protected]

NOTES

If the super.tab file isn't owned by root, or if it is group- or world-writable, super won't run setuid-root. (If the user's real uid is root, super won't run at all; otherwise, the effective uid reverts to real uid.)

There is a race condition when using password-requiring commands, but it doesn't affect security: if a user is running two copies of super simultaneously, and both processes try to update the user's password timestamp file at the same time, then it is possible for one of the super commands to fail. Workaround: a single user shouldn't execute two password-requiring super programs simultaneously.