Safe manual page - Tcl Tcl Built-In Commands

::safe::interpCreate ?child? ?options...?
::safe::interpInit child ?options...?
::safe::interpConfigure child ?options...?
::safe::interpDelete child
::safe::interpAddToAccessPath child directory
::safe::interpFindInAccessPath child directory
::safe::setLogCmd ?cmd arg...?

OPTIONS

?-accessPath pathList? ?-statics boolean? ?-noStatics? ?-nested boolean? ?-nestedLoadOk? ?-deleteHook script?

DESCRIPTION

Safe Tcl is a mechanism for executing untrusted Tcl scripts safely and for providing mediated access by such scripts to potentially dangerous functionality.

Safe Tcl ensures that untrusted Tcl scripts cannot harm the hosting application. It prevents integrity and privacy attacks. Untrusted Tcl scripts are prevented from corrupting the state of the hosting application or computer. Untrusted scripts are also prevented from disclosing information stored on the hosting computer or in the hosting application to any party.

Safe Tcl allows a parent interpreter to create safe, restricted interpreters that contain a set of predefined aliases for the source, load, file, encoding, and exit commands and are able to use the auto-loading and package mechanisms.

No knowledge of the file system structure is leaked to the safe interpreter, because it has access only to a virtualized path containing tokens. When the safe interpreter requests to source a file, it uses the token in the virtual path as part of the file name to source; the parent interpreter transparently translates the token into a real directory name and executes the requested operation (see the section SECURITY below for details). Different levels of security can be selected by using the optional flags of the commands described below.

All commands provided in the parent interpreter by Safe Tcl reside in the safe namespace.

COMMANDS

The following commands are provided in the parent interpreter:

::safe::interpCreate ?child? ?options...?

Creates a safe interpreter, installs the aliases described in the section ALIASES and initializes the auto-loading and package mechanism as specified by the supplied options. See the OPTIONS section below for a description of the optional arguments. If the child argument is omitted, a name will be generated. ::safe::interpCreate always returns the interpreter name.

The interpreter name child may include namespace separators, but may not have leading or trailing namespace separators, or excess colon characters in namespace separators. The interpreter name is qualified relative to the global namespace ::, not the namespace in which the ::safe::interpCreate command is evaluated.

::safe::interpInit child ?options...?

This command is similar to interpCreate except it that does not create the safe interpreter. child must have been created by some other means, like interp create -safe. The interpreter name child may include namespace separators, subject to the same restrictions as for interpCreate.

::safe::interpConfigure child ?options...?

If no options are given, returns the settings for all options for the named safe interpreter as a list of options and their current values for that child. If a single additional argument is provided, it will return a list of 2 elements name and value where name is the full name of that option and value the current value for that option and the child. If more than two additional arguments are provided, it will reconfigure the safe interpreter and change each and only the provided options. See the section on OPTIONS below for options description. Example of use:

# Create new interp with the same configuration as "$i0":
set i1 [safe::interpCreate {*}[safe::interpConfigure $i0]]

# Get the current deleteHook
set dh [safe::interpConfigure $i0  -del]

# Change (only) the statics loading ok attribute of an
# interp and its deleteHook (leaving the rest unchanged):
safe::interpConfigure $i0  -delete {foo bar} -statics 0

::safe::interpDelete child

Deletes the safe interpreter and cleans up the corresponding parent interpreter data structures. If a deleteHook script was specified for this interpreter it is evaluated before the interpreter is deleted, with the name of the interpreter as an additional argument.

::safe::interpFindInAccessPath child directory

This command finds and returns the token for the real directory directory in the safe interpreter's current virtual access path. It generates an error if the directory is not found. Example of use:

$child eval [list set tk_library \
      [::safe::interpFindInAccessPath $name $tk_library]]

::safe::interpAddToAccessPath child directory

This command adds directory to the virtual path maintained for the safe interpreter in the parent, and returns the token that can be used in the safe interpreter to obtain access to files in that directory. If the directory is already in the virtual path, it only returns the token without adding the directory to the virtual path again. Example of use:

$child eval [list set tk_library \
      [::safe::interpAddToAccessPath $name $tk_library]]

::safe::setLogCmd ?cmd arg...?

This command installs a script that will be called when interesting life cycle events occur for a safe interpreter. When called with no arguments, it returns the currently installed script. When called with one argument, an empty string, the currently installed script is removed and logging is turned off. The script will be invoked with one additional argument, a string describing the event of interest. The main purpose is to help in debugging safe interpreters. Using this facility you can get complete error messages while the safe interpreter gets only generic error messages. This prevents a safe interpreter from seeing messages about failures and other events that might contain sensitive information such as real directory names.

Example of use:

::safe::setLogCmd puts stderr

Below is the output of a sample session in which a safe interpreter attempted to source a file not found in its virtual access path. Note that the safe interpreter only received an error message saying that the file was not found:

NOTICE for child interp10 : Created
NOTICE for child interp10 : Setting accessPath=(/foo/bar) staticsok=1 nestedok=0 deletehook=()
NOTICE for child interp10 : auto_path in interp10 has been set to {$p(:0:)}
ERROR for child interp10 : /foo/bar/init.tcl: no such file or directory

OPTIONS

The following options are common to ::safe::interpCreate, ::safe::interpInit, and ::safe::interpConfigure. Any option name can be abbreviated to its minimal non-ambiguous name. Option names are not case sensitive.

-accessPath directoryList: This option sets the list of directories from which the safe interpreter can source and load files. If this option is not specified, or if it is given as the empty list, the safe interpreter will use the same directories as its parent for auto-loading. See the section SECURITY below for more detail about virtual paths, tokens and access control.
-statics boolean: This option specifies if the safe interpreter will be allowed to load statically linked packages (like load {} Tk). The default value is true : safe interpreters are allowed to load statically linked packages.
-noStatics: This option is a convenience shortcut for -statics false and thus specifies that the safe interpreter will not be allowed to load statically linked packages.
-nested boolean: This option specifies if the safe interpreter will be allowed to load packages into its own sub-interpreters. The default value is false : safe interpreters are not allowed to load packages into their own sub-interpreters.
-nestedLoadOk: This option is a convenience shortcut for -nested true and thus specifies the safe interpreter will be allowed to load packages into its own sub-interpreters.
-deleteHook script: When this option is given a non-empty script, it will be evaluated in the parent with the name of the safe interpreter as an additional argument just before actually deleting the safe interpreter. Giving an empty value removes any currently installed deletion hook script for that safe interpreter. The default value ({}) is not to have any deletion call back.

ALIASES

The following aliases are provided in a safe interpreter:

source fileName: The requested file, a Tcl source file, is sourced into the safe interpreter if it is found. The source alias can only source files from directories in the virtual path for the safe interpreter. The source alias requires the safe interpreter to use one of the token names in its virtual path to denote the directory in which the file to be sourced can be found. See the section on SECURITY for more discussion of restrictions on valid filenames.
load fileName: The requested file, a shared object file, is dynamically loaded into the safe interpreter if it is found. The filename must contain a token name mentioned in the virtual path for the safe interpreter for it to be found successfully. Additionally, the shared object file must contain a safe entry point; see the manual page for the load command for more details.
file ?subCmd args...?: The file alias provides access to a safe subset of the subcommands of the file command; it allows only dirname, join, extension, root, tail, pathname and split subcommands. For more details on what these subcommands do see the manual page for the file command.
encoding ?subCmd args...?: The encoding alias provides access to a safe subset of the subcommands of the encoding command; it disallows setting of the system encoding, but allows all other subcommands including system to check the current encoding.
exit: The calling interpreter is deleted and its computation is stopped, but the Tcl process in which this interpreter exists is not terminated.

SECURITY

Safe Tcl does not attempt to completely prevent annoyance and denial of service attacks. These forms of attack prevent the application or user from temporarily using the computer to perform useful work, for example by consuming all available CPU time or all available screen real estate. These attacks, while aggravating, are deemed to be of lesser importance in general than integrity and privacy attacks that Safe Tcl is to prevent.

The commands available in a safe interpreter, in addition to the safe set as defined in interp manual page, are mediated aliases for source, load, exit, and safe subsets of file and encoding. The safe interpreter can also auto-load code and it can request that packages be loaded.

Because some of these commands access the local file system, there is a potential for information leakage about its directory structure. To prevent this, commands that take file names as arguments in a safe interpreter use tokens instead of the real directory names. These tokens are translated to the real directory name while a request to, e.g., source a file is mediated by the parent interpreter. This virtual path system is maintained in the parent interpreter for each safe interpreter created by ::safe::interpCreate or initialized by ::safe::interpInit and the path maps tokens accessible in the safe interpreter into real path names on the local file system thus preventing safe interpreters from gaining knowledge about the structure of the file system of the host on which the interpreter is executing. The only valid file names arguments for the source and load aliases provided to the child are path in the form of [file join token filename] (i.e. when using the native file path formats: token/filename on Unix and token\filename on Windows), where token is representing one of the directories of the accessPath list and filename is one file in that directory (no sub directories access are allowed).

When a token is used in a safe interpreter in a request to source or load a file, the token is checked and translated to a real path name and the file to be sourced or loaded is located on the file system. The safe interpreter never gains knowledge of the actual path name under which the file is stored on the file system.

To further prevent potential information leakage from sensitive files that are accidentally included in the set of files that can be sourced by a safe interpreter, the source alias restricts access to files meeting the following constraints: the file name must fourteen characters or shorter, must not contain more than one dot (“.”), must end up with the extension (“.tcl”) or be called (“tclIndex”.)

Each element of the initial access path list will be assigned a token that will be set in the child auto_path and the first element of that list will be set as the tcl_library for that child.

If the access path argument is not given or is the empty list, the default behavior is to let the child access the same packages as the parent has access to (Or to be more precise: only packages written in Tcl (which by definition cannot be dangerous as they run in the child interpreter) and C extensions that provides a _SafeInit entry point). For that purpose, the parent's auto_path will be used to construct the child access path. In order that the child successfully loads the Tcl library files (which includes the auto-loading mechanism itself) the tcl_library will be added or moved to the first position if necessary, in the child access path, so the child tcl_library will be the same as the parent's (its real path will still be invisible to the child though). In order that auto-loading works the same for the child and the parent in this by default case, the first-level sub directories of each directory in the parent auto_path will also be added (if not already included) to the child access path. You can always specify a more restrictive path for which sub directories will never be searched by explicitly specifying your directory list with the -accessPath flag instead of relying on this default mechanism.

When the accessPath is changed after the first creation or initialization (i.e. through interpConfigure -accessPath list), an auto_reset is automatically evaluated in the safe interpreter to synchronize its auto_index with the new token list.

KEYWORDS

alias, auto-loading, auto_mkindex, load, parent interpreter, safe interpreter, child interpreter, source