Langue: en

Version: 301707 (debian - 07/07/09)

Section: 5 (Format de fichier)

NAME - database of restricted commands for super(1)


The file contains the restrictions on who can execute commands with super(1). It may also contain options that modify the uid and/or gid under which a command is run; the list of environment variables that are discarded before executing a command, and so on. The reader is expected to be familiar with the super(1) man page.

If you are trying to avoid reading this lengthy man page, you need to know a few simple rules. First, for a user to successfully execute a command by typing super commandName, the minimum entry in the is something like
      commandName   FullPathToCommand  userName
For example, the entry
      cdmount   /usr/local/bin/cdmount  wally,dolly
says that users wally and dolly may execute the /usr/local/bin/cdmount program by typing super cdmount.

Second, when you define options in the file, remember that all options are orthogonal to each other, and it doesn't matter in what order they appear on a control line. Third, global options (defined on a :global or :global_options line) take effect immediately after the defining line, are valid until the end of the input or until there is a countermanding global option or pattern, and are overridden by local (per-command) options, which are defined on a regular control line.

Although super has a great many options that you can set inside the file, none of them are required for security, so you don't have to be intimately familiar with this entire document in order to use super securely. The most important options that you will probably want to use are (a) a logging option like logfile=xxx or syslog=y; and (b) patterns=shell, so that the default regular-expression interpretation of wildcards is changed to the more convenient shell-style glob patterns.

Unless modified with options in the file, super executes commands using effective uid root; unchanged real uid and gid; no supplementary groups; no open file descriptors save 0, 1, and 2; no environment variables except a few with safe values (see super(1)); and signal handling reset to the default.

When super uses a command from a user-supplied super file (.supertab in the user's home directory) these rules are modified: the real and effective uid and gid are set to the owner of the .supertab file, and the supplementary groups are set to the user's login groups.

If the special supplementary file named super.init exists, it is implictly include at the start of and every .supertab file. The file resides in the same directory as and must be owned by root, and should be world-readable. This provides a uniform configuration file applied to every super-executed command. Note that the configuration file should contain entries that are appropriate for use in both the file and any per-user supertab file. It is not a good idea to include commands in the super.init file. The variable IS_USERTAB is defined to be ``yes'' if super is processing a per-user .supertab file, and ``no'' otherwise. The following pair of entries could be used in a super.init file to allow different initialization for regular use and per-user use:

 :if $IS_USERTAB == yes    :include /etc/super.init.per-user
 :if $IS_USERTAB != yes    :include /etc/

(The use of variables and conditional expressions is explained below, in the sections Variables and Conditionally-included Control Lines, respectively.)

The file is formatted as a series of control lines of the form

      CmdPat  FullPath \\
                          Options PermittedUsers PermittedTimes

      CmdPat1::FullPath1 CmdPat2::FullPath2 ... \\
                          Options PermittedUsers PermittedTimes

      :BuiltinCmd arguments

The Options, PermittedUsers, and PermittedTimes may be mixed together in any order. At least one PermittedUsers field is required, but neither Options nor PermittedTimes are required. For each control line, Super matches the following:

• the user-entered command to a CmdPat;
• the {user,group,host} triplet to a PermittedUsers
entry; and
• the current time to a PermittedTimes
entry (the default permits any time).

If these conditions are not satisfied, super ``falls through'' and tries the next control line.

For example, the entry

 renice /etc/renice jack@hill jill@bucket time~8-17

specifies that between hours 8:00 and 17:00, user Jack can renice any process on host hill, and user Jill can do so on host bucket, by simply typing super renice <args> .

Control lines begin in column 1, and the fields are whitespace-separated. Note that you can either use a single CmdPat and FullPath, separated by whitespace, or you can use a series of them in one control line by putting `::' between each CmdPat and FullPath pair.

Whitespace may be included in a field by enclosing text in single- or double-quotes. The quoting is shell-like, in the sense that quotemarks don't have to surround the entire field, and you can switch between quotemark types in a single entry. For instance, X"a b"Y'd e' is equivalent to "Xa bYd e". Comments begin with a `#' and continue to the end of a line.

There can also be blank or comment lines without any control data.

A control line may be continued by preceding the newline with a backslash, and indenting the continuation line with whitespace. When the multiple-line entry is read, the text just before the backslash-newline is not modified (any whitespace before the backslash will be kept), and the backslash-newline-whitespace sequence is either eliminated entirely or treated as whitespace. The rule is simple: following a letter, digit, or underscore, it is treated as whitespace. Otherwise, it is eliminated. This means that

Cmd  File  user1\            user2\            user3


is equivalent to

Cmd  File  user1 user2 user3

On the other hand,

Cmd  File  {user1,\            user2,\            user3}


is equivalent to

Cmd  File  {user1,user2,user3}

That is, the sensible interpretation is done in both cases.

The indentation requirement for continuation lines helps super catch typos. Comments may be placed before each backslash-newline pair, not just at the end of a continued control line.

The CmdPat Field

In response to the user typing

super cmd [ args ],

the cmd is searched for in the file. The cmd is matched against each pattern CmdPat.

Basic Use: Simple Command Patterns without Wildcards.

Typically, a CmdPat just uses plain characters without any special pattern-matching characters, so a cmd must be the same as the CmdPat string. For example:

skill /usr/local/bin/skill user1 user2 user3

Here, any user in the list {user1, user2, user3} may type super skill to execute /usr/local/bin/skill.

Advanced Use: Command Patterns with Wildcards.

More generally, a CmdPat can be either an ed-style pattern (``regex'' --- the BSD 4.x syntax used in the re_comp()/re_exec() routines), POSIX regular expressions, or a Bourne-shell-style pattern. You can set the pattern style using the global option patterns (see below). The default is ``regex'' for historical reasons; however, shell-style patterns are easier to use without errors, and most sites should use shell-style patterns. In all cases the patterns are extended to support csh-style brace expansion. For instance, a{x,y,z}b stands for the set of patterns axb ayb azb. Don't put any whitespace inside the braces!

For convenience, there is always an implied set of braces around an entire pattern. This means that any comma-separated list a,b,c is interpreted as {a,b,c}, and is very helpful so that you don't have to worry about getting braces just right when you build complex lists of out of (say) variables containing other lists of users.

All pattern matching is ``anchored'': patterns are forced to match the entire cmd string.

If the CmdPat is matched and the other conditions are met (such as the user being in the PermittedUser list to execute this command), FullPath gives the name of the actual command that will be executed with execve(). If FullPath contains an asterisk, the asterisk is first replaced by the user's cmd.

If you put special pattern-matching characters into the CmdPat, but don't put an asterisk into FullPath, you have simply given more ways a user can execute the same FullPath. This isn't useful, and in fact isn't a good idea at all. The power of using patterns in the CmdPat comes when FullPath includes an asterisk. For instance, a SysV-based host might have an entry in the file that looks like:

/usr/bin/{lp,lpstat,disable,enable,cancel} * \
               :operators uid=0

This would allow anybody in the "operators" group to have root access to the line printer commands (the uid=0 tells super to set the real uid to 0, not just the effective uid). For instance, if the user typed:

super /usr/bin/disable some_printer

then the FullPath (``*'') would be replaced by /usr/bin/disable, which would be the command to exec.

More conveniently, the file could have a line like:

{lp,lpstat,disable,enable,cancel} /usr/bin/* \
               :operators uid=0

In this case, the user can type

super disable some_printer

The asterisk-replacement ability also lets a user execute any of an entire directory of commands, using a entry like the following:

op/* /usr/local/super/scripts/* :operators uid=0

In this case, the user can type

super op/xyz

and super will execute /usr/local/super/scripts/op/xyz.

The asterisk-replacement capability is useful but potentially dangerous, because it may unintentionally open the door to programs you hadn't intended to give away. It is a sensible precaution to restrict asterisk-replacement to entries where the valid-user list includes trusted users only.

If you completely trust some users, but want logging of all actions executed as root, you could use:

/*   *   ReallyReallyTrustedUsers
/.*   *   ReallyReallyTrustedUsers

(depending on whether patterns=shell, patterns=regex, or patterns=posix).

The trusted users can now execute any command. Note that the pattern begins with a slash, to ensure that the cmd must be an absolute path --- this helps avoid accidental execs of the wrong program.

If you were really going to give everything away as shown above, you'd probably want to exclude any public-area workstations, require the trusted users to periodically give their passwords, and set the real uid=root (instead of just the effective uid), so the entry might be modified to read something like:

 /*   *   TrustedUsers \
          !{PatternsOfPublicWorkstations} \
          password=y timeout=5 uid=0

The FullPath Field

The FullPath field gives the name of the actual command that will be executed, and if it contains an asterisk, the asterisk is first replaced by the user's cmd string. The FullPath can optionally contain a list of initial arguments that precede any arguments passed by the user. For example,

xyz    "/usr/local/bin/blah -o1 -o2 -xrm 'a b c'" ...

specifies that when a valid user types super xyz, the command to execute is /usr/local/bin/blah and its arguments will be

argv[1]: -o1
argv[2]: -o2
argv[3]: -xrm
argv[4]: a b c
followed by any arguments that the user put on the super command line. Note: asterisk replacement is only done on the filename part of the FullPath, not on the arguments. You can safely include asterisks in the argument list. For security, the user's cmd may not contain any whitespace or backslashes.

The FullPath string is parsed using rules similar to the Bourne shell rules for backslashes in quoted strings, namely:

(a) backslash-newline is discarded;
(b) Otherwise, if outside a quoted substring, \x → plain x, which will not be treated as a delimiter, quotemark, or comment character;
(c) Otherwise, inside a quoted substring of FullPath:
\qq, where q is the quote character that encloses the FullPath string;
other backslashes are preserved: \x\x.

After writing a command with such backslash escapes, you should certainly use ``super -d cmd'' to check that your args are being parsed as expected before you offer this command to your users.

The same cmd can be listed several times in the file, in which case the first entry that allows the user to execute the command is chosen. The EXAMPLES section shows how this can be useful.

Permitted Users

Permitted users are those who may execute the specified Cmd's. Entries for any number of permitted users are given after the CmdPat and FullPath fields. Local options --- options that apply to this command only --- may also appear anywhere after the FullPath. Options are distinguished from permitted users because options all have the form key=value, whereas permitted-user entries may not contain unescaped equal signs. Each whitespace-separated word is a pattern in one of the formats




Note that the user~ part is optional. The user's login name must match the user pattern; the user must belong to a group whose name matches the group pattern; and the hostname must match the host pattern. If the user, group, or host part is not given, there are no corresponding restrictions.

If the user is root, super acts as if all permitted-user patterns are preceded by the pattern user~root --- that is, root's rule is default-allow, instead of the default-deny rule that applies to all ordinary users.

By default, the ``group'' field is first matched against named groups to which the user belongs, and then against the user's decimal gid --- this allows the user to be put in a group in the /etc/passwd file that isn't given a name in the /etc/group file. (If you want to change the rules for using decimal gid, see the use of MATCH_DECIMAL_UID and MATCH_DECIMAL_GID in super.c for details.)

Since you can restrict users to particular hosts, a single file can be shared among many different machines. If the host part is of the form +xyz, then xyz is interpreted as a netgroup name and any host in netgroup xyz is matched. In that case, xyz is taken literally, and not interpreted as a pattern to be matched. Note: netgroup lookup is only implemented if the function innetgr() is available.

If the host part doesn't match the hostname, it might be because the pattern and actual hostnames contain two different (but both valid) incomplete versions of the fully-qualified domain name (FQDN). By default, if the host part fails to match the hostname, the FQDN is looked up and all of the ways of naming the host are matched against the pattern. For example, if the FQDN is, super will first try, then spacely.sprockets, and finally spacely. This can be turned off; see option gethostbyname. (You might want to turn it off because using nameserver lookup can reduce security a bit --- your host may query a nameserver on another host to obtain the FQDN, and (a) that nameserver or an intermediate host along the way could have been subverted, or (b) another host could impersonate the nameserver. In either case your computer could receive incorrect hostnames.)

The patterns for valid users, groups, and hosts follow the same rules for the CmdPats, described above: they can be entered with either ed-style or Bourne-shell-style patterns (depending on the setting of the patterns global option); csh-style brace expansion is allowed; and all pattern matching is ``anchored'': patterns are forced to match the entire username, groupname, or hostname.

To make it easy to exclude some users/groups/hosts, any of these patterns can be negated by prefixing the pattern with `!'. If a negated pattern is matched, the user may not execute the command, even if there was a previous non-negated pattern that the user matched. All patterns are read left-to-right, and the last matched pattern ``wins''. Thus if the user/group/host list is

j.* !jo
user~j.* !user~jo

then the first entry allows any username beginning with `j' to execute the command, but the second entry disallows user `jo'. If entered in the reverse order,

!jo j.*

then the second entry, `j.*', allows all users beginning `j', and therefore the first entry has no effect.

Permitted Times

The time condition restricts the days and times during which this command may be matched. If the execution time isn't acceptable, then super ignores the control line, and ``falls through'' to inspect the next entry, just as it does if the user/group/host aren't acceptable. A time condition looks like:

with patterns that look like:

Pattern                       Example

hh[:mm]-hh[:mm][/dayname]    13:30-17/monday

{<,>,<=,>=}hh[:mm][/dayname] <17/tues

dayname                      Friday

The first form explicitly specifies an interval during which the command may be used. Times may not go past midnight; to specify the night between Monday and Tuesday, you must do something like:


The second form for time patterns allows you to use logical operators. The Monday-night example could equally have been rendered as:

(There is a tiny difference in the two examples above: in the first example, the time range includes 17:30 and 8:00; in the second example, the time range is 17:31-07:59. Use time~{>=17:30/mon,<=8/tues} to make the interpretation identical to the first example.)

If there are a series of time patterns, they are evaluated left-to-right, and the rightmost matching pattern is used. To permit execution between 17:30 Monday and 8:00 Tuesday, but exclude 0:00 to 1:00 Tuesday, use:

time~{>=17:30/mon,<=8/tues}  !time~{0-1/tues}

By default, valid daynames are English, but if your system supports the setlocale(3) function, the global option lang=zzz will set the valid names to those of locale zzz. Valid daynames are either (a) the full names of the chosen language; (b) an official abbreviated day name for that language; (c) a 3-or-more character abbreviation of the full weekday; or (d) *, meaning any day. (You can check on super's valid daynames by executing super -d, which will show the default names of the weekdays at or near the top of its debugging output, and show the new weekday names that take effect when the lang=zzz option is encountered.)

Time patterns have a special defaulting rule when the execution time is not in one of the intervals in the time list:

• if all time patterns are negated, super permits execution at any time not in one of the listed intervals;
• otherwise, there is at least one non-negated pattern, and super defaults to deny execution at times outside the specified acceptable intervals.
The reason is that the natural interpretation of a series of negated conditions, such as
!time~{0-8,17-24} !time~{sat,sun}
is to infer that all other times are acceptable for execution. On the other hand, if there are any ordinary, non-negated times, such as
the natural interpretation is that any times not explicitly mentioned are not acceptable.

Note that explicit braces are important in the above list. If they were missing, the implied braces around the entire pattern would render this equivalent to

time~8-17/mon time~tues time~wed time~thu time~fri
That is, the permitted times are 8-17h on Monday, and any time on Tuesday through Friday.

Global Conditions

Global options and conditions affect the overall super processing. To set them, you must use a line like

:global            Global Options And Patterns
:global_options    Global Options And Patterns

For backwards compatibility, you can alternatively use

/        /         GlobalOptionsAndPatterns

but this use is discouraged.

These so-called ``global'' options and conditions actually take effect immediately after the line on which they appear, and are valid until the end of the input or until there is a countermanding global option or pattern. To have a global option or pattern affect the entire file, you must place it as the first non-comment line of the file.

If there are any PermittedUser or PermittedTime conditions on the global settings line, they are applied to each following command in the file. The conditions look like

:global  cond cond ... <> cond cond ...

PermittedUser and PermittedTime conditions to the left of ``<>'' are processed before the local (per-command) conditions; conditions to the right of ``<>'' are processed after the per-command conditions. If ``<>'' is missing, all conditions are processed after the local conditions. Example 1:

:global  jan <> !@+badhosts

says that user
jan can usually execute any command, but under no circumstances will a user on any host in netgroup badhosts be allowed to execute any command. (User jan will not be allowed to execute a command if the per-command conditions disallow it, or if jan is on one of the badhosts computers).

Example 2:

:global  !root <>

root's default setting from default-allow to default-deny, just like ordinary users. Root will only be given execute permission for entries that explicitly allow root on the per-entry line.

Global PermittedUser (PermittedTime) conditions take effect on the line on which they are defined, and are good until another set of global PermittedUser (PermittedTime) conditions is entered on another :global_option line. That is, a new global condition line replaces any previous global conditions.

One sensible approach to using global conditions is to put conditions that allow users to execute commands before the per-command conditions are processed, and to put negating patterns (for users/groups/hosts that are never to be allowed to execute anything) after the per-command conditions.

(Global options are discussed below, together with local options).

Conditionally-included Control Lines

Control lines can be conditionally included by using the :if control line. This can be helpful when using a single file for hosts with different architectures, or hosts in different NIS domains, etc. The syntax is:

:if left   op   right     moreText

The expression left op right is evaluated, and if true, moreText is evaluated as an ordinary control line. The valid comparison operators are:

   ==   string equality

   !=   string inequality

   ~    glob match string
left against pattern right
   !~   negated glob-match.

For example, you could include a file of commands only valid on Sun4-type machines as follows:

 :if $UNAME_MACHINE ~ sun* \
          :include  /Path/To/Sun4/File
(UNAME_MACHINE is a variable automatically defined by super; see the following section on variables.)

If you wanted to exclude Sun4c-type machines from using the file, you could modify this to be:

 :if $UNAME_MACHINE ~ sun* \
     :if $UNAME_MACHINE != sun4c \
          :include  /Path/To/Sun4/File

There are no boolean operators provided, but note that there is an implicit boolean and available by concatenating :if commands, as shown in the second example above.


Super offers variables to make it easier to handle entries that are duplicated or are constructed out of other entries. Variables are defined by typing

:define VariableName VariableDefinition

or they may be imported from the environment by using

:getenv [EnvVarName...]

If you use the :getenv command, then the values of any imported environment variables may only contain the following ``safe'' characters: -/:+._a-zA-Z0-9. If ``bad'' characters are found in a value, the entire value is replaced with an empty string. Note that these imported variables do not enter the environment of any executed command; they simply become part of the variable set.

The VariableName should be made up only of letters, digits, and/or underscore. (You can actually use any characters you wish, but super doesn't promise to work correctly if you use characters outside the standard set.)

The VariableDefinition begins at the first non-whitespace character after the VariableName and continues up to but not including the final newline. Comments embedded on the variable definition line(s) are deleted before the variable definition is stored. A variable definition may be continued across multiple lines by preceding each newline with a backslash, and indenting the continuation line with whitespace. Just as for regular control lines, the backslash-newline-whitespace sequence is treated as whitespace if it follows a letter, digit, or underscore; otherwise, it is eliminated. For example,

:define Users user1,\                  user2,\                  user3
:define Users user1,user2,user3

are equivalent definitions.
  Unlike Makefiles, the variable definitions are not scanned first and then the file re-scanned. Instead, variables take effect at the point they are defined, and remain in effect until they are re-defined or end of file --- thus variables definitions must precede their first use in the file.

Variables may contain other variables (which must have already been defined). Variable substitution is done when a line is first read. A line is never re-scanned after variable substitution.

Variables are used in a file by typing




The special variable $$ is replaced by a single ``$''. Any other name $X (where X is not a letter, digit, or underscore) is an error.

Because a line is never re-scanned after variable substitution, the following sequence:

:define A $$ :define B A :define C $B $$B $C


defines C to be simply ``A $B''.

Variables can be helpful in grouping users or hosts together. For example, you might restrict access to a command so that it can't be run from a public-access workstation:

:define Room103_WS hosta,hostb,hostc,hostd :define Room105_WS hoste,hostf,hostg,hosth :define Room106_WS +nonprivate :define PublicWorkstations $Room103_WS,$Room105_WS,$Room106_WS SomeCmd FullPathHere !@$(PublicWorkstations)


In the above example, we have taken advantage of the implied braces that are always placed around any pattern, so that the comma-separated list of workstations is brace-expanded into !@hosta !@hostb ... !@hoste !@hostf ... !@+nonprivate.

Some variables are automatically defined by super.

After super determines whether it is processing a per-user .supertab (see super(1)), it defines IS_USERTAB to be ``yes'' if super is processing a per-user .supertab file, and ``no'' otherwise. The allows the super.init to act differently depending on how it is being invoked.

When the top-level file is opened, SUPER_OWNER is set to the login name of the owner, and SUPER_HOME is set to the home directory of the owner. This can be useful in per-user .supertab files, especially when they include files shared among several accounts. For example, each person's .supertab file could be simply

:include /opt/proj/common.supertab owner=cristy

Then, the /opt/proj/common.supertab file can use entries like the following:

:global logfile=$SUPER_HOME/.superlog
* /project/tools/$SUPER_OWNER/* :toolgroup

Because the SUPER_HOME and SUPER_OWNER variables apply to the top-level per-user files, they always refer to per-user locations.

Super defines the built-in variable CALLER to be the the login name of the of account invoking super, and CALLER_HOME is the home directory of $CALLER. Sample use:

 sam  /usr/sbin/sam  group~operator  uid=0 \
                     env=DISPLAY \
Here, the operator group can execute sam as root, and the GUI will display at the caller's display (due to the use of env=DISPLAY). Since the XAUTHORITY environment variable is set to the caller's .Xauthority file, this will give the caller access to the same display(s) to which s/he already has access.

The following variables are defined when super starts up. They can be useful in conditionally-included lines (:if lines). If your host doesn't supply these functions, or doesn't support some of the values that super tries to fetch, the corresponding variable will be initialized to an empty string. (Use super -b to print the names and values of all builtin variables. This makes it simple to see what variable values to check in :if lines.)

From the gethostname() or sysinfo() function: • HOSTNAME            Hostname, possibly canonicalized.

• HOST                Hostname, short (unqualified).

From the getdomainname() function: • NIS_DOMAIN          domain set for NIS purposes.

From the sysinfo() function: • SI_SYSNAME          Name of operating system.

• SI_HOSTNAME         Name of node.

• SI_RELEASE          Release of operating system.

• SI_VERSION          Version field of utsname.

• SI_MACHINE          Kind of machine.

• SI_ARCHITECTURE     Instruction set arch.

• SI_HW_SERIAL        Hardware serial number.

• SI_HW_PROVIDER      Hardware manufacturer.

• SI_SRPC_DOMAIN      Secure RPC domain.

From the uname() function: • UNAME_SYSNAME       Operating system name.

• UNAME_NODENAME      The nodename.

• UNAME_RELEASE       Operating system release.

• UNAME_VERSION       Operating system version.

• UNAME_MACHINE       Machine hardware name (class).


The configuration file can specify a wide variety of options, such as requiring the user's password before executing some commands, or restricting the command-line arguments to match certain patterns.

Options are handled very differently from conditions (conditions include user, group, host, and time). If a control line's conditions aren't met, super falls through and tries the next control line in the file. After finding an acceptable control line, super will execute the command if the options are satisfied; otherwise, it stops and doesn't search the file any further.

Options can be divided into (a) local options, which are defined on a regular control line, and apply only to that control line; and (b) global options, which are defined on a :global or :global_options line, take effect immediately after the line, and are valid until the end of the input or until there is a countermanding global option or pattern.

All options are orthogonal to each other. It doesn't matter in what order they appear on a control line.

Some options can be given as either local or global options. If both are used, the local setting overrides the global one.

Two special names can be used with any of the options that take user or group ids as arguments: owner=xxx, uid=xxx, euid=xxx, gid=xxx, egid=xxx, u+g=xxx, groups=xxx, addgroups=xxx. These names are <owner>, meaning the owner of the file to be executed (or the owner's group, whichever is appropriate in the context); and <caller>, meaning the owner or group of the user calling super. The angle brackets are literally part of the name. These have the same values as the built-in variables CALLER and OWNER (see the Variables section, above). For example, the options

gid=Foo uid=<caller>
would change the group to Foo, but leave the uid unchanged.

The recognized options are:

Group 1. Options Affecting How Superfiles Are Read and Processed.

Specifies the pattern-matching type for conditions and options. The string xxx must be one of:
posix --- patterns are POSIX
regular expressions. You can use ``posix/extended'' for extended regular expressions; ``posix/icase'' for case-insensitive regular expressions; or ``posix/extended/icase'' for both. See your regular-expression man pages for details.
regex --- patterns are ed-style
regular expressions, using the rules embodied in the BSD 4.x routines re_comp()/re_exec(), with the addition of csh-style brace expansion. This is the default for historical reasons, but most people prefer to use shell-style patterns here, and it is recommended that you put patterns=shell (see below) in your global options list.
shell --- patterns are approximately Bourne-shell style, with the addition
of csh-style brace expansion and the special [[chars]] pattern. The patterns are formed from:
\x  force x to be a plain character;
?   matches any single character;
*   matches any sequence of 0 or more chars;
[chars]  matches any single character in the set;
[^chars] matches any single char NOT in the set;
[[chars]]     When the pattern begins with
and ends with ]], then each and every character in the string must match the ordinary square-bracket pattern [chars] (or [^chars]).
^pat     inverts the sense of the match ---
the string must NOT match the pattern.
This option sets the language used for weekdays (in time conditions). Here, zzz may be any locale available on your host. For example, lang=de would typically cause super to use German names. The default is the C locale, hence English names.
If y, FullPathNames can be relative instead of absolute. By default this is disallowed, because it is almost always a very foolish (unsafe) thing to do.
If y, group names can contain a slash. By default this is disallowed, so that super can catch certain typos in the file. (Namely, super can catch errors in which an entry is of the form Cmd:File instead of the required Cmd::File. The trouble is that Cmd:File looks syntactically like user:group, and can therefore be mistaken for a valid part of a control line. But the filename will contain a slash --- unless you have unwisely enabled the relative_path option --- so disallowing slashes allows super to flag the line as syntactically invalid.)
Enables or disables the use of gethostbyname() to find the fully-qualified domain name (see the discussion in the Permitted Users section, which describes the security issues associated with enabling this option.) Default: enabled (if your host supports gethostbyname()).

Group 2. Logging Options.

Enables logging to a local file. Each invocation of super (aside from ones for help) generates an entry in file fname.
If logging is enabled with logfile=fname, the logfile will be opened for writing using uid=xxx (can be either a username or numeric uid). This option allows you to have the file created/opened under another uid that does have cross-host access, such as the uid of a system manager. (See timestampuid=xxx for additional comments). Default: loguid=0.
Notices of super failures are piped to the shell command mail-command. This is independent of the setting of the logfile and syslog options. For instance, mail="mail -s Super joeblow" will cause error messages to be mailed to user joeblow (on some systems you may need to substitute mailx for mail). Note: the mail-command is executed by passing it as an argument to popen(3). This is safe to execute because of the clean environment assured by super.
This is identical to the mail option, except that mailany sends notification of successful invocations as well as errors.
Tells super which host's syslog daemon is to receive log messages when option syslog=y is enabled. The option has no effect if used after the first message is logged: once the logger has been opened, it is not re-opened if the rlog_host is changed. Default: the local host. Note: One could instead configure the syslog.conf file to forward the messages to a central machine.
Logging information is passed to the logs maintained by the syslogd daemon. This is independent of the setting of the logfile option (above). Error messages are by default logged at priority LOG_ERR and successful attempts to run programs are logged at priority LOG_INFO. (See options syslog_error and syslog_success to change these levels.)
If syslog is enabled (see the syslog option), then by default super logs error messages using syslog(3) code LOG_ERR. This option changes the code to xxx, where xxx is any of the usual syslog(3) priority and/or facility codes, such as LOG_WARNING or LOG_LOCAL7|LOG_ERR. The LOG_xxx words can be separated by whitespace, dot, and/or ``|'', but of course you must use quotes if whitespace is included. The leading "LOG_" is optional, and the value is case-insensitive. For example, LOG_LOCAL7|LOG_ERR can alternatively be written as local7.err. Super doesn't know what are sensible codes --- it's up to the writer to choose meaningful values. For instance, if you put
      syslog_error=LOG_INFO | LOG_ERR (bad!)
then you will get both those values or'd together and passed to syslog().
This option is just like syslog_error, except that it applies to successful-execution messages instead of error messages. Default: LOG_INFO .

Group 3. Extra Help Information for Users.

The string xxx is printed when giving help to users. It should be set to a helpful one-line description of the command.

Group 4. Password and Other Restrictions Before Approval.

Each argument must be no more than mmm characters long (including the terminating null), and the sum length of all arguments must not exceed nnn characters. A negative value means that no limit is applied. The defaults for mmm and nnn are set by the compile-time manifest constants MAXLEN1ARG and MAXLENARGS, which are usually 1000 and 10,000 characters, respectively.
The user is required to enter mmm - nnn arguments to the command. If just nnn is given, the user must enter exactly nnn arguments. These arguments are in addition to any arguments entered in the command part of the file. The default is to allow the user to enter any number of arguments.
This means that the nnn'th or mmm-nnn'th arguments must match pattern sss. (Arguments are numbered from 1.) The pattern must be enclosed in quotes if it contains whitespace. Note that this option does not require that there be mmm-nnn arguments; it only says what those arguments must look like, if entered. You can use this option several times on one line, with different mmm-nnn values each time, to apply different patterns to different arguments. If more than one pattern applies to a given argument, all of those patterns must be matched. An empty pattern ("") is special: it has the effect of unsetting (removing) any previous patterns for the matching [mmm-]nnn. This can be useful if you want to change :global settings, instead of adding to them. If there are local arg[mmm-]nnn option(s), they completely replace all global arg[mmm-]nnn values, even if the local mmm-nnn values do not overlap with the global values.
This option specifies that the FullPath (after asterisk-substitution) must be owned by user (or uid) xxx, or else it won't be executed.
If y, user authentication is required before the command is executed. The default authentication method is Unix password authentication. See also the options authprompt, authtype, authuser, timeout, and renewtime, and be sure to read the warning under timestampbyhost.
Specifies the prompt used when authenticating the user (usually the default prompt is fine). Variable substitution is done on the prompt before printing.
Specifies the type of authentication used when auth=y. The type can be password or pam (if PAM is supported on your system). The default is password. If PAM authentication is used, super uses the service name ``super'' when looking for authentication in your system PAM configuration files.
Specifies the user whose authentication is required when auth=y. The default is the password (or other authentication) for the user who invoked super.
This is a deprecated option; it has been replaced by the auth and authtype options. Password=y is equivalent to auth=y authtype=password ; and Password=n is equivalent to auth=n.
If y, the user's timestamp file is updated to the current time whenever an authentication-requiring command is executed. The result is that a user who frequently executes authentication-requiring commands won't need to re-authenticate until more than timeout minutes elapse since the last such command. Otherwise, the user will need to re-authenticate timeout minutes after last entering the password. The default is n.
User authentication is good for m minutes; that is, the command may be executed without re-authenticating for m minutes after the previous authentication (for any command). After m minutes, user authentication will be required again before the command can be executed. If timeout is zero or negative, authentication is required every time the command is used. The timestamp for user usr is recorded in the file TIMESTAMP_DIR directory (see timestampbyhost and the FILES section, below).
If y (default), the timestamp files are given names that are unique on each host. For instance, jouser@somehost will be given a timestamp file named TIMESTAMP_DIR/somehost/jouser, where TIMESTAMP_DIR is defined in the FILES section. If timestmapbyhost=n, the timestamp files are given names that are unique to each user, but not unique per host. For instance, jouser on any host will be given a timestamp file named TIMESTAMP_DIR/jouser.
WARNING: The hostname used is that from gethostname(). Note that this is not necessarily unique across internet domains, since it is frequently not a fully-qualified domain name. Therefore you should not share timestamp directories with hosts outside the local domain. (Generally such connections don't exist, but one could crossmount the timestamp directory disk...)
If a password is required, the time at which it was entered is recorded as the mtime of a timestamp file. The timestamp file is normally created with owner=root; however, this option causes it to be created/modified using uid=xxx (xxx can be either a username or numeric uid). This option is useful when a network of hosts are sharing a cross-mounted timestamp directory. Note that networks are typically configured to not allow root on one host to have root access to files on another host, which will forbid root on other hosts from creating the timestamp file unless it's world-writable. This option allows you to have the file created/opened under another uid that does have cross-host access, such as the uid of a system manager. Default: timestampuid=0.
Each name in the comma-separated list is a variable which the user must enter at a prompt from super. For example, you might have a super shutdown command which halts the computer. If you execute this on the wrong host there may be some very annoyed users! So, you can include checkvar=HOST, and the user will have to type the correct hostname in response to a prompt from super.

Group 5. Modifications to Environment Before Executing Command.

Sets the real uid to xxx just before executing the command. If option euid isn't used, also sets the effective uid to xxx. The uid xxx is first tried as a login name and then as a number. If the options uid=xxx and u+g=yyy (see below) are used together, then the u+g option only sets the group id, and not the user id.
Sets the effective uid to xxx just before executing the command. The uid xxx is first tried as a login name and then as a number.
Sets the real gid to yyy just before executing the command. If option egid isn't used, also sets the effective gid to yyy. The gid yyy is first tried as a group name and then as a number.
Sets the effective gid to yyy just before executing the command. The gid yyy is first tried as a group name and then as a number.
Sets the real uid to zzz, the real gid to zzz's login gid, and the supplementary groups list to zzz's supplementary groups just before executing the command. If the euid and/or egid option aren't given, the effective uid and/or gid are also set. The options u+g and gid=yyy conflict with each other, and may not be used together.
By default, the user's supplementary groups list is deleted before executing a command (unless the option u+g is used). This option instead sets the group list to name[,...]
This option adds the listed groups to the supplementary groups list. Since the default is to provide an empty supplementary groups list, this option usually has the same effect as the plain groups option. However, if the options u+g=foo addgroups=a,b,c are used, then the supplementary groups list is composed of user foo's supplementary groups plus a, b, and c.
Execute the command will execute with its first argument (that is, the argument conventionally denoted as argv[0]), set to name. As a convenient shorthand, the value <path> (the angle brackets are literally part of the name) means to use the FullPath specified in the file. By default, argv[0] is set to Cmd, the name of the super command invoked by the user, regardless of the actual path being invoked. However, some programs will not run properly unless argv[0] has a particular value. For example, suppose you want to permit users to safely mount zip disks, and you use something like:
      zipmount "/etc/mount -o nosuid /dev/xz10 /zip"
This command will fail if /etc/mount requires that it be invoked with argv[0] set to [.../]mount, because super will use the name zipmount. However, you can put argv0=<path> into your file, and then the mount command will work properly.
Each name in the comma-separated list is an environment variable which should not be deleted before executing the Cmd; these variables are in addition to the normal variables created or passed by super (TERM, IFS, PATH, USER, LOGNAME, HOME, ORIG_USER, ORIG_LOGNAME, ORIG_HOME, LINES, COLUMNS, SUPERCMD). Be careful here; environment variables can sometimes be abused to create security holes. If you use the option more than once, the later instance overrides the earlier one, instead of adding to it. Similarly, using it as a local option completely overrides any global setting.
Specifies the maximum length of an environment variable definition (including name, equal sign, value, and trailing null character). The default is given by the compile-time manifest constant MAXENVLEN, usually 1000 characters. A negative value means no limit.
Just before executing the command, super changes the working directory to dir.
The environment variable var is defined to have the value xxx, and is passed on when executing the Cmd. You can add several environment variable definitions by using the option more than once.
Each file descriptor n in the comma-separated list should not be closed before executing the Cmd. These descriptors are added to the usual set of descriptors kept open, namely 0, 1, 2.
changes the ``nice'' value of the executed command by an amount n from the default level. (Positive increments reduce the command's priority; negative increments increase it.)
sets the umask of any executed command to n. A leading 0x or 0X in nnn means a hexadecimal value; otherwise, a leading 0 means octal; otherwise it's decimal.

Group 6. Other Options

If the rest of the line is matched, then super prints the specified message just before executing the command.
If the rest of the line is matched, then super does variable-substitution on the specified message message, prints it, and exits. This lets you conveniently set up some ``stop'' conditions, and use the die option to prevent super from looking at any line past the stop conditions. Otherwise, you'd have to individually attach the stop conditions to every control line.

Include Files

A file can include other files by means of an entry like

:include filename [ owner=xxx ] [ group=yyy  ]
:optinclude file [ owner=xxx ] [ group=yyy  ]
If the file isn't an absolute path, it is taken to be relative to the directory containing the file. Include files may be nested up to the system limit on the number of simultaneously-open file descriptors.

The owner=xxx option specifies that the file must be owned by user xxx; the group=yyy options specifies that the file must belong to group yyy. If group=yyy is specified, then the file can be group-writable; by default, the file must be writable only by owner. This can be useful for a collection of accounts that are operated together as part of a single project --- the several of accounts can share .supertab files by :include-ing files belonging to the trusted user xxx and/or group yyy. Notes:

1. The regular root-owned file can also use the owner= or group= constructs, but it's not a good idea. Don't do it.
2. Beware of the transitive nature of this trust: the file owned by xxx can in turn include a file owned by yet another user. You might wind up trusting a user you didn't intend to trust!

The difference between :include and :optinclude is that the former generates an error if the named file doesn't exist, whereas the latter (optional-include) silently ignores files that don't exist.

WARNING: You should use :optinclude with great caution, and be sure not to depend on that file being present. It is easy to imagine a scenario in which an administrator carelessly changes an entry so that the wrong permission is granted if an :optinclude'd file was missing.


The location of the file on your system.
The location of the super.init file on your system.
Default location of the file whose mtime is used as the timestamp for the last time the user entered his or her password for password-requiring commands. Check your installation for the directory used on your system.


Example 1. The control line
 doit /usr/local/bin/doit                \
           me                            \
           you@{h1,h32}                  \
           ja.*:ok_j                     \

allows /usr/local/bin/doit to be run setuid-root by • user me on any host,
• user you on hosts h1 and h32;
• any users named ja.* in group ok_j;
• and anybody in group goodguys.

Example 2. The pair of control lines
 doit /usr/local/bin/doit                \
           u+g=smith env=TZ,TAPE         \
           password=y timeout=0          \
 doit /usr/local/bin/doit                \
           u+g=smith env=TZ,TAPE         \

allows user jo to run /usr/local/bin/doit with uid = smith, gid = smith's login gid, and keeping the environment variables TZ and TAPE in addition to the standard set. If user jo is at PublicWorkstation, the first entry will match, requiring jo's password every time the command is used; otherwise, super will match at the second entry, and no password is needed to run the command.

Example 3. Here is an entry restricting CD-ROM mounting on different hosts: tas is the only user who may mount CD's on elgar; anybody in group xyz may mount CD's on alpha or delta; and anybody on a host in the netgroup india may mount a CD on the india hosts. However, user jo may never run cdmount, regardless of his or her group or host (assuming that there is no overriding global pattern that permits jo to use the command). Note that shell-style patterns are used, not regex-style patterns.
 cdmount /usr/local/bin/cdmount          \
           tas@elgar                     \
           :xyz@{alpha,delta}            \
           *@+india                      \