Tcl Built-In Commands - interp manual page

This command makes it possible to create one or more new Tcl interpreters that co-exist with the creating interpreter in the same application. The creating interpreter is called the master and the new interpreter is called a slave. A master can create any number of slaves, and each slave can itself create additional slaves for which it is master, resulting in a hierarchy of interpreters.

Each interpreter is independent from the others: it has its own name space for commands, procedures, and global variables. A master interpreter may create connections between its slaves and itself using a mechanism called an alias. An alias is a command in a slave interpreter which, when invoked, causes a command to be invoked in its master interpreter or in another slave interpreter. The only other connections between interpreters are through environment variables (the env variable), which are normally shared among all interpreters in the application. Note that the name space for files (such as the names returned by the open command) is no longer shared between interpreters. Explicit commands are provided to share files and to transfer references to open files from one interpreter to another.

The interp command also provides support for safe interpreters. A safe interpreter is a slave whose functions have been greatly restricted, so that it is safe to execute untrusted scripts without fear of them damaging other interpreters or the application's environment. For example, all IO channel creation commands and subprocess creation commands are made inaccessible to safe interpreters. See SAFE INTERPRETERS below for more information on what features are present in a safe interpreter. The dangerous functionality is not removed from the safe interpreter; instead, it is hidden, so that only trusted interpreters can obtain access to it. For a detailed explanation of hidden commands, see HIDDEN COMMANDS, below. The alias mechanism can be used for protected communication (analogous to a kernel call) between a slave interpreter and its master. See ALIAS INVOCATION, below, for more details on how the alias mechanism works.

A qualified interpreter name is a proper Tcl lists containing a subset of its ancestors in the interpreter hierarchy, terminated by the string naming the interpreter in its immediate master. Interpreter names are relative to the interpreter in which they are used. For example, if a is a slave of the current interpreter and it has a slave a1, which in turn has a slave a11, the qualified name of a11 in a is the list a1 a11.

The interp command, described below, accepts qualified interpreter names as arguments; the interpreter in which the command is being evaluated can always be referred to as {} (the empty list or string). Note that it is impossible to refer to a master (ancestor) interpreter by name in a slave interpreter except through aliases. Also, there is no global name by which one can refer to the first interpreter created in an application. Both restrictions are motivated by safety concerns.

THE INTERP COMMAND

The interp command is used to create, delete, and manipulate slave interpreters, and to share or transfer channels between interpreters. It can have any of several forms, depending on the option argument:

interp alias srcPath srcToken: Returns a Tcl list whose elements are the targetCmd and args associated with the alias represented by srcToken (this is the value returned when the alias was created; it is possible that the name of the source command in the slave is different from srcToken).
interp alias srcPath srcToken {}: Deletes the alias for srcToken in the slave interpreter identified by srcPath. srcToken refers to the value returned when the alias was created; if the source command has been renamed, the renamed command will be deleted.
interp alias srcPath srcCmd targetPath targetCmd ?arg arg ...?: This command creates an alias between one slave and another (see the alias slave command below for creating aliases between a slave and its master). In this command, either of the slave interpreters may be anywhere in the hierarchy of interpreters under the interpreter invoking the command. SrcPath and srcCmd identify the source of the alias. SrcPath is a Tcl list whose elements select a particular interpreter. For example, ``a b'' identifies an interpreter b, which is a slave of interpreter a, which is a slave of the invoking interpreter. An empty list specifies the interpreter invoking the command. srcCmd gives the name of a new command, which will be created in the source interpreter. TargetPath and targetCmd specify a target interpreter and command, and the arg arguments, if any, specify additional arguments to targetCmd which are prepended to any arguments specified in the invocation of srcCmd. TargetCmd may be undefined at the time of this call, or it may already exist; it is not created by this command. The alias arranges for the given target command to be invoked in the target interpreter whenever the given source command is invoked in the source interpreter. See ALIAS INVOCATION below for more details. The command returns a token that uniquely identifies the command created srcCmd, even if the command is renamed afterwards. The token may but does not have to be equal to srcCmd.
interp aliases ?path?: This command returns a Tcl list of the tokens of all the source commands for aliases defined in the interpreter identified by path. The tokens correspond to the values returned when the aliases were created (which may not be the same as the current names of the commands).
interp create ?-safe? ?- -? ?path?: Creates a slave interpreter identified by path and a new command, called a slave command. The name of the slave command is the last component of path. The new slave interpreter and the slave command are created in the interpreter identified by the path obtained by removing the last component from path. For example, if path is a b c then a new slave interpreter and slave command named c are created in the interpreter identified by the path a b. The slave command may be used to manipulate the new interpreter as described below. If path is omitted, Tcl creates a unique name of the form interpx, where x is an integer, and uses it for the interpreter and the slave command. If the -safe switch is specified (or if the master interpreter is a safe interpreter), the new slave interpreter will be created as a safe interpreter with limited functionality; otherwise the slave will include the full set of Tcl built-in commands and variables. The - - switch can be used to mark the end of switches; it may be needed if path is an unusual value such as -safe. The result of the command is the name of the new interpreter. The name of a slave interpreter must be unique among all the slaves for its master; an error occurs if a slave interpreter by the given name already exists in this master. The initial recursion limit of the slave interpreter is set to the current recursion limit of its parent interpreter.
interp delete ?path ...?: Deletes zero or more interpreters given by the optional path arguments, and for each interpreter, it also deletes its slaves. The command also deletes the slave command for each interpreter deleted. For each path argument, if no interpreter by that name exists, the command raises an error.
interp eval path arg ?arg ...?: This command concatenates all of the arg arguments in the same fashion as the concat command, then evaluates the resulting string as a Tcl script in the slave interpreter identified by path. The result of this evaluation (including error information such as the errorInfo and errorCode variables, if an error occurs) is returned to the invoking interpreter. Note that the script will be executed in the current context stack frame of the path interpreter; this is so that the implementations (in a master interpreter) of aliases in a slave interpreter can execute scripts in the slave that find out information about the slave's current state and stack frame.
interp exists path: Returns 1 if a slave interpreter by the specified path exists in this master, 0 otherwise. If path is omitted, the invoking interpreter is used.
interp expose path hiddenName ?exposedCmdName?: Makes the hidden command hiddenName exposed, eventually bringing it back under a new exposedCmdName name (this name is currently accepted only if it is a valid global name space name without any ::), in the interpreter denoted by path. If an exposed command with the targeted name already exists, this command fails. Hidden commands are explained in more detail in HIDDEN COMMANDS, below.
interp hide path exposedCmdName ?hiddenCmdName?: Makes the exposed command exposedCmdName hidden, renaming it to the hidden command hiddenCmdName, or keeping the same name if hiddenCmdName is not given, in the interpreter denoted by path. If a hidden command with the targeted name already exists, this command fails. Currently both exposedCmdName and hiddenCmdName can not contain namespace qualifiers, or an error is raised. Commands to be hidden by interp hide are looked up in the global namespace even if the current namespace is not the global one. This prevents slaves from fooling a master interpreter into hiding the wrong command, by making the current namespace be different from the global one. Hidden commands are explained in more detail in HIDDEN COMMANDS, below.
interp hidden path: Returns a list of the names of all hidden commands in the interpreter identified by path.
interp invokehidden path ?-global? hiddenCmdName ?arg ...?: Invokes the hidden command hiddenCmdName with the arguments supplied in the interpreter denoted by path. No substitutions or evaluation are applied to the arguments. If the -global flag is present, the hidden command is invoked at the global level in the target interpreter; otherwise it is invoked at the current call frame and can access local variables in that and outer call frames. Hidden commands are explained in more detail in HIDDEN COMMANDS, below.
interp issafe ?path?: Returns 1 if the interpreter identified by the specified path is safe, 0 otherwise.
interp marktrusted path: Marks the interpreter identified by path as trusted. Does not expose the hidden commands. This command can only be invoked from a trusted interpreter. The command has no effect if the interpreter identified by path is already trusted.
interp recursionlimit path ?newlimit?: Returns the maximum allowable nesting depth for the interpreter specified by path. If newlimit is specified, the interpreter recursion limit will be set so that nesting of more than newlimit calls to Tcl_Eval() and related procedures in that interpreter will return an error. The newlimit value is also returned. The newlimit value must be a positive integer between 1 and the maximum value of a non-long integer on the platform.
The command sets the maximum size of the Tcl call stack only. It cannot by itself prevent stack overflows on the C stack being used by the application. If your machine has a limit on the size of the C stack, you may get stack overflows before reaching the limit set by the command. If this happens, see if there is a mechanism in your system for increasing the maximum size of the C stack.
interp share srcPath channelId destPath: Causes the IO channel identified by channelId to become shared between the interpreter identified by srcPath and the interpreter identified by destPath. Both interpreters have the same permissions on the IO channel. Both interpreters must close it to close the underlying IO channel; IO channels accessible in an interpreter are automatically closed when an interpreter is destroyed.
interp slaves ?path?: Returns a Tcl list of the names of all the slave interpreters associated with the interpreter identified by path. If path is omitted, the invoking interpreter is used.
interp target path alias: Returns a Tcl list describing the target interpreter for an alias. The alias is specified with an interpreter path and source command name, just as in interp alias above. The name of the target interpreter is returned as an interpreter path, relative to the invoking interpreter. If the target interpreter for the alias is the invoking interpreter then an empty list is returned. If the target interpreter for the alias is not the invoking interpreter or one of its descendants then an error is generated. The target command does not have to be defined at the time of this invocation.
interp transfer srcPath channelId destPath: Causes the IO channel identified by channelId to become available in the interpreter identified by destPath and unavailable in the interpreter identified by srcPath.

SLAVE COMMAND

For each slave interpreter created with the interp command, a new Tcl command is created in the master interpreter with the same name as the new interpreter. This command may be used to invoke various operations on the interpreter. It has the following general form:

slave command ?arg arg ...?

Slave is the name of the interpreter, and command and the args determine the exact behavior of the command. The valid forms of this command are:

slave aliases: Returns a Tcl list whose elements are the tokens of all the aliases in slave. The tokens correspond to the values returned when the aliases were created (which may not be the same as the current names of the commands).
slave alias srcToken: Returns a Tcl list whose elements are the targetCmd and args associated with the alias represented by srcToken (this is the value returned when the alias was created; it is possible that the actual source command in the slave is different from srcToken).
slave alias srcToken {}: Deletes the alias for srcToken in the slave interpreter. srcToken refers to the value returned when the alias was created; if the source command has been renamed, the renamed command will be deleted.
slave alias srcCmd targetCmd ?arg ..?: Creates an alias such that whenever srcCmd is invoked in slave, targetCmd is invoked in the master. The arg arguments will be passed to targetCmd as additional arguments, prepended before any arguments passed in the invocation of srcCmd. See ALIAS INVOCATION below for details. The command returns a token that uniquely identifies the command created srcCmd, even if the command is renamed afterwards. The token may but does not have to be equal to srcCmd.
slave eval arg ?arg ..?: This command concatenates all of the arg arguments in the same fashion as the concat command, then evaluates the resulting string as a Tcl script in slave. The result of this evaluation (including error information such as the errorInfo and errorCode variables, if an error occurs) is returned to the invoking interpreter. Note that the script will be executed in the current context stack frame of slave; this is so that the implementations (in a master interpreter) of aliases in a slave interpreter can execute scripts in the slave that find out information about the slave's current state and stack frame.
slave expose hiddenName ?exposedCmdName?: This command exposes the hidden command hiddenName, eventually bringing it back under a new exposedCmdName name (this name is currently accepted only if it is a valid global name space name without any ::), in slave. If an exposed command with the targeted name already exists, this command fails. For more details on hidden commands, see HIDDEN COMMANDS, below.
slave hide exposedCmdName ?hiddenCmdName?: This command hides the exposed command exposedCmdName, renaming it to the hidden command hiddenCmdName, or keeping the same name if the argument is not given, in the slave interpreter. If a hidden command with the targeted name already exists, this command fails. Currently both exposedCmdName and hiddenCmdName can not contain namespace qualifiers, or an error is raised. Commands to be hidden are looked up in the global namespace even if the current namespace is not the global one. This prevents slaves from fooling a master interpreter into hiding the wrong command, by making the current namespace be different from the global one. For more details on hidden commands, see HIDDEN COMMANDS, below.
slave hidden: Returns a list of the names of all hidden commands in slave.
slave invokehidden ?-global hiddenName ?arg ..?: This command invokes the hidden command hiddenName with the supplied arguments, in slave. No substitutions or evaluations are applied to the arguments. If the -global flag is given, the command is invoked at the global level in the slave; otherwise it is invoked at the current call frame and can access local variables in that or outer call frames. For more details on hidden commands, see HIDDEN COMMANDS, below.
slave issafe: Returns 1 if the slave interpreter is safe, 0 otherwise.
slave marktrusted: Marks the slave interpreter as trusted. Can only be invoked by a trusted interpreter. This command does not expose any hidden commands in the slave interpreter. The command has no effect if the slave is already trusted.
slave recursionlimit ?newlimit?: Returns the maximum allowable nesting depth for the slave interpreter. If newlimit is specified, the recursion limit in slave will be set so that nesting of more than newlimit calls to Tcl_Eval() and related procedures in slave will return an error. The newlimit value is also returned. The newlimit value must be a positive integer between 1 and the maximum value of a non-long integer on the platform.
The command sets the maximum size of the Tcl call stack only. It cannot by itself prevent stack overflows on the C stack being used by the application. If your machine has a limit on the size of the C stack, you may get stack overflows before reaching the limit set by the command. If this happens, see if there is a mechanism in your system for increasing the maximum size of the C stack.

SAFE INTERPRETERS

A safe interpreter is one with restricted functionality, so that is safe to execute an arbitrary script from your worst enemy without fear of that script damaging the enclosing application or the rest of your computing environment. In order to make an interpreter safe, certain commands and variables are removed from the interpreter. For example, commands to create files on disk are removed, and the exec command is removed, since it could be used to cause damage through subprocesses. Limited access to these facilities can be provided, by creating aliases to the master interpreter which check their arguments carefully and provide restricted access to a safe subset of facilities. For example, file creation might be allowed in a particular subdirectory and subprocess invocation might be allowed for a carefully selected and fixed set of programs.

A safe interpreter is created by specifying the -safe switch to the interp create command. Furthermore, any slave created by a safe interpreter will also be safe.

A safe interpreter is created with exactly the following set of built-in commands:

after   append  array   binary
break   case    catch   clock
close   concat  continue    eof
error   eval    expr    fblocked
fcopy   fileevent   flush   for
foreach format  gets    global
if  incr    info    interp
join    lappend lindex  linsert
list    llength lrange  lreplace
lsearch lsort   namespace   package
pid proc    puts    read
regexp  regsub  rename  return
scan    seek    set split
string  subst   switch  tell
time    trace   unset   update
uplevel upvar   variable    vwait
while

The following commands are hidden by interp create when it creates a safe interpreter:

cd  encoding    exec    exit
fconfigure  file    glob    load
open    pwd socket  source

These commands can be recreated later as Tcl procedures or aliases, or re-exposed by interp expose.

The following commands from Tcl's library of support procedures are not present in a safe interpreter:

auto_exec_ok    auto_import auto_load
auto_load_index auto_qualify    unknown

Note in particular that safe interpreters have no default unknown command, so Tcl's default autoloading facilities are not available. Autoload access to Tcl's commands that are normally autoloaded:

auto_mkindex    auto_mkindex_old
auto_reset  history
parray  pkg_mkIndex
::pkg::create   ::safe::interpAddToAccessPath
::safe::interpCreate    ::safe::interpConfigure
::safe::interpDelete    ::safe::interpFindInAccessPath
::safe::interpInit  ::safe::setLogCmd
tcl_endOfWord   tcl_findLibrary
tcl_startOfNextWord tcl_startOfPreviousWord
tcl_wordBreakAfter  tcl_wordBreakBefore

can only be provided by explicit definition of an unknown command in the safe interpreter. This will involve exposing the source command. This is most easily accomplished by creating the safe interpreter with Tcl's Safe-Tcl mechanism. Safe-Tcl provides safe versions of source, load, and other Tcl commands needed to support autoloading of commands and the loading of packages.

In addition, the env variable is not present in a safe interpreter, so it cannot share environment variables with other interpreters. The env variable poses a security risk, because users can store sensitive information in an environment variable. For example, the PGP manual recommends storing the PGP private key protection password in the environment variable PGPPASS. Making this variable available to untrusted code executing in a safe interpreter would incur a security risk.

If extensions are loaded into a safe interpreter, they may also restrict their own functionality to eliminate unsafe commands. For a discussion of management of extensions for safety see the manual entries for Safe-Tcl and the load Tcl command.

A safe interpreter may not alter the recursion limit of any interpreter, including itself.

ALIAS INVOCATION

The alias mechanism has been carefully designed so that it can be used safely when an untrusted script is executing in a safe slave and the target of the alias is a trusted master. The most important thing in guaranteeing safety is to ensure that information passed from the slave to the master is never evaluated or substituted in the master; if this were to occur, it would enable an evil script in the slave to invoke arbitrary functions in the master, which would compromise security.

When the source for an alias is invoked in the slave interpreter, the usual Tcl substitutions are performed when parsing that command. These substitutions are carried out in the source interpreter just as they would be for any other command invoked in that interpreter. The command procedure for the source command takes its arguments and merges them with the targetCmd and args for the alias to create a new array of arguments. If the words of srcCmd were ``srcCmd arg1 arg2 ... argN'', the new set of words will be ``targetCmd arg arg ... arg arg1 arg2 ... argN'', where targetCmd and args are the values supplied when the alias was created. TargetCmd is then used to locate a command procedure in the target interpreter, and that command procedure is invoked with the new set of arguments. An error occurs if there is no command named targetCmd in the target interpreter. No additional substitutions are performed on the words: the target command procedure is invoked directly, without going through the normal Tcl evaluation mechanism. Substitutions are thus performed on each word exactly once: targetCmd and args were substituted when parsing the command that created the alias, and arg1 - argN are substituted when the alias's source command is parsed in the source interpreter.

When writing the targetCmds for aliases in safe interpreters, it is very important that the arguments to that command never be evaluated or substituted, since this would provide an escape mechanism whereby the slave interpreter could execute arbitrary code in the master. This in turn would compromise the security of the system.

HIDDEN COMMANDS

Safe interpreters greatly restrict the functionality available to Tcl programs executing within them. Allowing the untrusted Tcl program to have direct access to this functionality is unsafe, because it can be used for a variety of attacks on the environment. However, there are times when there is a legitimate need to use the dangerous functionality in the context of the safe interpreter. For example, sometimes a program must be sourced into the interpreter. Another example is Tk, where windows are bound to the hierarchy of windows for a specific interpreter; some potentially dangerous functions, e.g. window management, must be performed on these windows within the interpreter context.

The interp command provides a solution to this problem in the form of hidden commands. Instead of removing the dangerous commands entirely from a safe interpreter, these commands are hidden so they become unavailable to Tcl scripts executing in the interpreter. However, such hidden commands can be invoked by any trusted ancestor of the safe interpreter, in the context of the safe interpreter, using interp invoke. Hidden commands and exposed commands reside in separate name spaces. It is possible to define a hidden command and an exposed command by the same name within one interpreter.

Hidden commands in a slave interpreter can be invoked in the body of procedures called in the master during alias invocation. For example, an alias for source could be created in a slave interpreter. When it is invoked in the slave interpreter, a procedure is called in the master interpreter to check that the operation is allowable (e.g. it asks to source a file that the slave interpreter is allowed to access). The procedure then it invokes the hidden source command in the slave interpreter to actually source in the contents of the file. Note that two commands named source exist in the slave interpreter: the alias, and the hidden command.

Because a master interpreter may invoke a hidden command as part of handling an alias invocation, great care must be taken to avoid evaluating any arguments passed in through the alias invocation. Otherwise, malicious slave interpreters could cause a trusted master interpreter to execute dangerous commands on their behalf. See the section on ALIAS INVOCATION for a more complete discussion of this topic. To help avoid this problem, no substitutions or evaluations are applied to arguments of interp invokehidden.

Safe interpreters are not allowed to invoke hidden commands in themselves or in their descendants. This prevents safe slaves from gaining access to hidden functionality in themselves or their descendants.

The set of hidden commands in an interpreter can be manipulated by a trusted interpreter using interp expose and interp hide. The interp expose command moves a hidden command to the set of exposed commands in the interpreter identified by path, potentially renaming the command in the process. If an exposed command by the targeted name already exists, the operation fails. Similarly, interp hide moves an exposed command to the set of hidden commands in that interpreter. Safe interpreters are not allowed to move commands between the set of hidden and exposed commands, in either themselves or their descendants.

Currently, the names of hidden commands cannot contain namespace qualifiers, and you must first rename a command in a namespace to the global namespace before you can hide it. Commands to be hidden by interp hide are looked up in the global namespace even if the current namespace is not the global one. This prevents slaves from fooling a master interpreter into hiding the wrong command, by making the current namespace be different from the global one.

CREDITS

This mechanism is based on the Safe-Tcl prototype implemented by Nathaniel Borenstein and Marshall Rose.

EXAMPLES

Creating and using an alias for a command in the current interpreter:

interp alias {} getIndex {} lsearch {alpha beta gamma delta}
set idx [getIndex delta]

Executing an arbitrary command in a safe interpreter where every invokation of lappend is logged:

set i [interp create -safe]
interp hide $i lappend
interp alias $i lappend {} loggedLappend $i
proc loggedLappend {i args} {
   puts "logged invokation of lappend $args"
   # Be extremely careful about command construction
   eval [linsert $args 0 \
         interp invokehidden $i lappend]
}
interp eval $i $someUntrustedScript

KEYWORDS

alias, master interpreter, safe interpreter, slave interpreter

Copyright © 1995-1996 Sun Microsystems, Inc.
Copyright © 1995-1997 Roger E. Critchlow Jr.