Safe mode

From Rosetta Code
Safe mode is a draft programming task. It is not yet considered ready to be promoted as a complete task, for reasons that should be found in its talk page.

Does the language implementation allow for a "safer mode" of execution? Usually termed Safe mode, a more realistic view is probably Safer mode or restricted mode. It is one thing to place restrictions on execution, and another thing entirely to allow execution of scripts from untrusted sources and assume nothing untoward will happen.

Along with a simple yes/no answer, describe what features are restricted when running in safe mode.

6502 Assembly

The 6502 has no safe mode.

68000 Assembly

There is a "Supervisor Mode", however there's really nothing stopping you from setting the supervisor flag to true if you can execute arbitrary code.

8080 Assembly

The 8080 has no safe mode.

8086 Assembly

The 8086 has no safe mode.

AWK

# syntax: GAWK --sandbox -f SAFE_MODE.AWK
#
# Per The GNU Awk User’s Guide, edition 4.2.1
# https://www.gnu.org/software/gawk/manual/gawk.html
#   --sandbox or -S
#   Disable the system() function, input redirections with getline, output
#   redirections with print and printf, and dynamic extensions.  This is
#   particularly useful when you want to run awk scripts from questionable
#   sources and need to make sure the scripts can't access your system (other
#   than the specified input data file).
#
# Error message when running in sandbox mode:
# gawk: SAFE_MODE.AWK:16: fatal: redirection not allowed in sandbox mode
#
BEGIN {
    print("hello world") >"A.TMP"
    exit(0)
}

C3

The C3 compiler has a `--fast` and a `--safe` mode respectively. The latter, intended for development, enables a wide range of checks from out-of-bounds checks and null dereference checks to runtime contracts with full stacktraces.

Frink

Frink has an extensive and fine-grained security manager architecture which allows the language to disallow dangerous operations by everyone, or allow very specific operations for specific users.

The easiest way to test this is to add the --sandbox option when starting Frink. This enforces the strictest sandboxing mode. Similarly, when creating a Frink interpreter from Java code, the most restrictive security can be enabled by calling its Frink.setRestrictiveSecurity(true) method.

frink.parser.Frink interp = new frink.parser.Frink();
interp.setRestrictiveSecurity(true);

Below are some operations that can be allowed/disallowed from a custom security manager. For most of these, the permission can be restricted to allow/disallow a particular file, URL, or class, or method:

  • Read a file or URL
  • Call unsafeEval
  • Import another file using use
  • Construct a new Java object
  • Call a static method on a Java class or object
  • Access a static Java field
  • Call a method on a Java object
  • Write a field on a Java object
  • Define a function
  • Set a global flag
  • Print to a printer
  • Write a file
  • Open a graphics window
  • Construct an expression from an expression type and argument list
  • Transform an expression
  • Create a transformation rule
  • Set a class-level variable

All of these operations are disallowed when the most restrictive security is enabled.

Go

Any code written in Go is considered to be 'safe' unless it uses one or more of the following features:

  • The 'unsafe' package.
  • The 'reflect' package.
  • cgo.


Although 'normal' Go code uses pointers, arithmetic on them is not permitted and so they cannot be made to point to arbitrary locations in memory. However, the 'unsafe' package contains features which do allow one to perform pointer arithmetic with all the risks this entails.

The 'reflect' package allows one to inspect and manipulate objects of arbitrary types and exposes internal data structures such as string and slice headers. This can result in fragile code where mistakes which would normally be identified at compile time will instead manifest themselves as runtime panics.

'cgo' is Go's bridge to using C code. As such it is just as unsafe as writing C code directly.

J

The security level (default: 0) can be increased to 1 by executing:

(9!:25) 1

Afterwards, all verbs able to alter the environment outside J are prohibited. See the J Community Wiki for details regarding the restrictions.

Jsish

The jsish interpreter allows a -s, --safe command line switch to restrict access to the file system.

For example, given safer.jsi:

File.write('/tmp/safer-mode.txt', 'data line');
Output:
prompt$ jsish safer.jsi
prompt$ jsish -s safer.jsi
/home/btiffin/lang/jsish/safer.jsi:2: error: write access denied by safe interp: /tmp/safer-mode.txt    (at or near "data line")

ERROR

The Jsish implementation borrows many ideas from Tcl, and also includes an Interp module. These sub interpreters can also be set to run in a safer mode.

prompt$ jsish
# var si = new Interp({isSafe:true});
variable
# si.source('safer.jsi');
error: read access denied: /home/btiffin/lang/jsish/safer.jsi
ERROR

Some control is allowed over the restrictions provided by safer mode.

var interp1 = new Interp({isSafe:true, safeWriteDirs:['/tmp'], , safeReadDirs:['/tmp']});

Julia

Julia does not have a "sandbox" mode that restricts access to operating system resources such as files, since this is considered to be the province of the underlying operating system. Julia does have functions that handle underlying OS memory resources similar to C type pointers. Such functions, including
 unsafe_wrap unsafe_read unsafe_load unsafe_write unsafe_trunc unsafe_string unsafe_store! unsafe_copyto! 
are prefixed with "unsafe_" to indicate that a memory access fault could be generated if arguments to those functions are in error.

Nim

Nim doesn’t provide safe mode, but it make a distinction between safe and unsafe features. Safe features are those which cannot corrupt memory integrity while unsafe ones can.

There is currently no restrictions for using unsafe features, but a programmer should be aware that they must be used with care.

Here are some unsafe features:

  • The ones dealing with raw memory and especially those using pointers. Note that Nim makes a difference between pointers which allow access to raw (untraced) memory and references which allow access to traced memory.
  • Type casting which, contrary to type conversion, is a simple assignment of a new type without any conversion to make the value fully compatible with the new type.
  • Using cstring variables as no index checking is performed when accessing an element.
  • Inserting assembly instructions with the asm statement.

Perl

The only built-in 'safer' way to run Perl is to invoke it in 'taint mode' with the command line option -T. While in this mode input from the user, and all variables derived from it, cannot be used in certain contexts until 'sanitized' by being passed through a regular expression.

There is a CPAN module Safe that purports to allow Perl to be compiled and executed in restricted compartments, isolated namespaces with limits on allowed operators. However, it has not been updated in over 6 years, and so not tested with recent releases of Perl. Further, it does not appear to work at all on BSD-derived versions of UNIX.

There's really no switch to flip to make Perl code more secure. It is up to the programmer to follow security best-practices, such as employing the strict and warnings pragmas, using 3-argument form of open for filehandles, being careful about the contents of $ENV{PATH}, and so forth. The CPAN module Perl::Critic can be helpful in this regard. Read further on this topic in the language documentation on Perl security

Phix

See Untrusted_environment#Phix

Raku

(formerly Perl 6)

Mostly a cut-n-paste from the Untrusted environment task.

Raku doesn't really provide a high security mode for untrusted environments. By default, Raku is sort of a walled garden. It is difficult to access memory directly, especially in locations not controlled by the Raku interpreter, so unauthorized memory access is unlikely to be a threat with default Raku commands and capabilities.

It is possible (and quite easy) to run Raku with a restricted setting which will disable many IO commands that can be used to access or modify things outside of the Raku interpreter. However, a determined bad actor could theoretically work around the restrictions, especially if the nativecall interface is available. The nativecall interface allows directly calling in to and executing code from C libraries so anything possible in C is now possible in Raku. This is great for all of the power it provides, but along with that comes the responsibility and inherent security risk. The same issue arises with unrestricted loading of modules. If modules can be loaded, especially from arbitrary locations, then any and all restriction imposed by the setting can be worked around.

The restricted setting is modifiable, but by default places restrictions on or completely disables the following things:

User Subroutines (disabled)
  • sub chmod() modify filesystem permissions
  • sub copy() copy a file
  • sub link() create a link to a file
  • sub mkdir() make a filesystem directory
  • sub open() open a filesystem location / file
  • sub pipe() open a pipe
  • sub QX() execute arbitrary code
  • sub rename() rename a file
  • sub rmdir() remove a directory
  • sub run() run arbitrary code
  • sub shell() execute code in a shell
  • sub socket() open a socket
  • sub spurt() write a file
  • sub symlink() create a symbolic link to a location
  • sub unlink() delete a file
Internal Subroutines (disabled)
  • sub CHANGE-DIRECTORY change directory
  • sub CHMOD-PATH change permissions
  • sub COPY-FILE copy a file
  • sub MAKE-DIR make a directory
  • sub REMOVE-DIR remove a directory
  • sub RENAME-PATH rename a directory
  • sub SYMLINK-PATH create a symbolic link
  • sub UNLINK-PATH delete a file
Classes (disabled)
  • class IO::CatHandle streaming file handle
  • class IO::Handle file handle
  • class IO::Path filesystem path
  • class IO::Pipe OS pipe
  • class IO::Socket OS socket
  • class IO::Socket::INET Network socket
  • class NativeCall Nativecall interface to foreign code (C mostly)
  • class Proc OS Process
  • class Proc::Async Asynchronous OS Process
Method Mixins / Roles (locked down so can't be overridden)
  • method FALLBACK() handle unknown method calls
  • method new() create a new instance
  • method gist() display method

Really, if you want to lock down a Raku instance so it is "safe" for unauthenticated, untrusted, general access, you are better off running it in some kind of locked down virtual machine or sandbox managed by the operating system rather than trying to build an ad hoc "safe" environment.

REXX

For running REXX on IBM mainframes,   REXX supports the option   Scan   for the   trace   statement.

This allows the program to be processed (and be checked for syntax errors),   but commands to the "host system" won't be executed.

However, not all REXXes support this option.

Regina REXX supports a --restricted command-line option, and embedded interpreters can also be set to run restricted. Many commands are disabled in this mode, including most access to hosted services. The intrinsic FUNCTION REXX() extension in GnuCOBOL defaults to restricted mode, and programmers must explicitly use FUNCTION REXX-UNRESTRICTED(script, args...) for access to the full REXX programming environment from that COBOL implementation.

       identification division.
       program-id. rexxtrial.

       environment division.
       configuration section.
       repository.
           function all intrinsic.

       data division.
       working-storage section.

       procedure division.
      *> First attempt fails and return statement does not execute
       display rexx("ADDRESS SYSTEM; 'ls rexxtrial.cob'; return 'fail'")
       display "Exception: " exception-status

      *> Second is allowed and succeeds
       display "Try with rexx-unrestricted"
       display rexx-unrestricted(
           "ADDRESS SYSTEM; 'ls -l rexxtrial.cob'; return 'success'")
       display "No exception raised: " exception-status
       goback.
       end program rexxtrial.
Output:
$ cobc -xj rexxtrial.cob
     1 +++ 'ls rexxtrial.cob'
Error 95 running "gnucobol", line 1: [Restricted feature used in "safe" mode]
Error 95.5: [Running external commands invalid in "safe" mode]

Exception: EC-IMP-SCRIPT
Try with rexx-unrestricted
-rw-rw-r--. 1 btiffin btiffin 727 Feb 19 04:26 rexxtrial.cob
success
No exception raised:

Rust

While Rust compiles to native code and does not provide any kind of runtime sandbox, it does implement a compile-time enforced distinction between "safe" and "unsafe" code, intended to improve the maintainability of complex codebases by confining sources of certain types of difficult-to-debug problems to small, clearly marked subsets of the code which can be audited more intensely.

Safe code, which is the default, cannot cause memory unsafety or data races as long the unsafe code it depends on upholds the invariants expected of it.

Unsafe code, enabled by marking a function, block, or trait (interface) with the unsafe keyword, enables the use of four additional language capabilities which the compiler cannot verify correct use of and which are intended for building safe abstractions, such as the standard library's Mutex and reference-counted pointers.

Those four capabilities are:

  • Dereferencing raw pointers (Rust's name for C-style pointers)
  • Calling unsafe functions (All foreign functions, as well native APIs with safety invariants that are impossible or impractical to encode in the type system)
  • Interacting with mutable static variables (the idiomatic solution is to use "interior mutability" via a wrapper type like Mutex or RWLock which allows a mutable value to be stored in an "immutable" static variable.)
  • Implementing an unsafe trait (interface)

To further the goal of improving maintainability in large codebases, the Rust compiler can also be configured to warn or error out if unsafe code is encountered within a given scope.

(An example of this would be an enterprise project where the coders most experienced in low-level work are responsible for the module where unsafe is allowed, while the majority of the codebase lives in modules which depend on the unsafe-containing module, but are configured to forbid the use of unsafe within their own code.)

Scala

Actually, with a high-level programming language as Scala, it's a bad idea to flag or unflag for a "Safe mode".. This should be a task for the target system.

Wren

Wren code is considered to be 'safe' in itself. Method/function calls are dynamically checked and generate runtime errors which can be caught and handled. A fiber's stack grows if it gets close to overflowing. There is no support for pointers nor reflection and memory is managed automatically by the runtime.

However, when Wren is embedded in a host application, one needs to deal with the embedding API (or a wrapper thereof) to pass data between Wren and the host. The embedding API is written in C and, as such, is intrinsically unsafe. It is generally up to the programmer to ensure that API functions are passed the correct number of arguments and those arguments are of the correct types.

Z80 Assembly

Has no safe mode.

zkl

zkl is unsafe. Any program can access any method and many methods access the system. Additionally, any program can load a program or eval (compile and run) text.

Zig

Zig provides compilation mode settings for safety (`Debug` and `ReleaseSafe`) and code-block annotations (`@setRuntimeSafety`) to allow the user tight control of optimization and safety. The list of checked safety properties and tooling to debug existing memory problems is extensive. Computations at compiletime are unconditionally checked for those errors and memory problems.

Zig is optimized for compilation performance and thus will not include advanced shape analysis like Rust's borrow checker into the compilation phase or compromise on compilation performance for the necessary data. Thus temporal memory safety and data race safety are not covered during compilation time analysis and must be tested, ie with thread sanitizer, Valgrind and test allocator. There are plans to offer upper bound stack- and uninitialized memory analysis.