Difference between revisions of "Safely running untrusted Haskell code"

From HaskellWiki
Jump to navigation Jump to search
(mention outdatedness)
(EvilIx)
Line 69: Line 69:
 
* large array allocations can fill memory
 
* large array allocations can fill memory
 
* very large array allocations can integer overflow the storage manager, allowing arbitrary memory access (this appears to be fixed in GHC 6.8.x)
 
* very large array allocations can integer overflow the storage manager, allowing arbitrary memory access (this appears to be fixed in GHC 6.8.x)
* creating class instances that violate assumed laws (cf EvilIx)
+
* creating class instances that violate assumed laws (cf [http://www.haskell.org/pipermail/haskell-cafe/2006-December/019994.html EvilIx])
 
* various literal strings that print IRC protocol commands could be printed using exceptions.
 
* various literal strings that print IRC protocol commands could be printed using exceptions.
 
* if a user guesses the top level identifier the expression is bound to, it can be used to print a silly string
 
* if a user guesses the top level identifier the expression is bound to, it can be used to print a silly string

Revision as of 10:41, 20 November 2012

Obviously, don't run code in the IO monad, just show pure results (or possibly make your own monad that is a restricted subset of IO). But it's a lot more complicated than that...

Verifying safety : lambdabot's approach

Since 2004, lambdabot has executed arbitrary strings of Haskell provided by user's of various IRC channels, in particular, the Haskell channel. In order to do this, a particular security policy is required. The policy, and its implementation, is described here.

The policy

Only allow execution of pure Haskell expressions.

The implementation

Note: This section refers to the old Lambdabot evaluator; as of 2009, lambdabot calls out to mueval, which while it uses many of the same techniques, is structured differently.

The evaluator is essentially a function, eval :: String -> IO String, which takes a random Haskell string, verifies it, compiles it, and evaluates the result, returning a String representing the result, back over the network.

This function is implemented as two separate processes:

The driver reads a String from the network, and then subjects it to a simple test:

  • The expression is parsed as a Haskell 98 expression, hopefully preventing code injection (is this true? and can any string that can parse as a valid Haskell expression become something more sinister when put in a particular context?)

If the string parses as a Haskell 98 expression, the 'runplugs' process is then forked to evaluate the string, and the following checks are put in place:

  • Only a trusted module set is imported, avoiding unsafePerformIO and unsafeIOtoST and such like.
  • Module imports are disallowed
  • Time and space limitations on the runplugs process are set by the OS 'rlimit' facility
  • The expression type checked, enforcing lack of memory errors
  • Because the user code is not at the beginning of the file, malicious {-# LANGUAGE #-} and {-# OPTIONS #-} flags are ignored
  • Only -fextended-default-rules are allowed, as language extensions over H98.
  • The resulting .o file is dynamically linked only into the throw-away runplugs instance
  • Even if all went well, the first 2048 characters of the shown string are returned to the caller (no infinite output DoS)

A few other niceties are provided:

  • The expression is bound to a random identifier (harmless to guess), in order to allow nice line error messages with line pragmas.
  • The expression is wrapped in 'show'.
  • A catch-all instance of Show in terms of Typable is provided, to display non-displayable objects in a more useful way (e.g. putStrLn --> <[Char] -> IO ()>)
  • It is compiled to native code with -fasm for speed (compilation time is neglible compared to IRC lag)
  • The value is evaluated inside an exception handler; if an exception is thrown, the first 1024 characters of the exception string are returned.

Exploits

A variety of interesting exploits have been found, or thought of, over the years. Those we remember are listed below:

  • using newtype recursion to have the inliner not terminate
  • using pathological type inference cases to have the type checker not terminate
  • code injection of code fragments that aren't Haskell expressions
  • Template Haskell used to run IO actions during type checking
  • stToIO to convert a safe ST action, into an IO action that is run
  • large strings returned in exceptions
  • unsafePerformIO, of course
  • unsafeCoerce#
  • throwing a piece of code as an exception, which is evaluated when the exception is shown
  • non-terminating code, in a tight loop that doesn't allocate, can't use GHC's threadDelay/scheduler (let f () = f () in f ()) to timeout (must use OS resource limits).
  • large array allocations can fill memory
  • very large array allocations can integer overflow the storage manager, allowing arbitrary memory access (this appears to be fixed in GHC 6.8.x)
  • creating class instances that violate assumed laws (cf EvilIx)
  • various literal strings that print IRC protocol commands could be printed using exceptions.
  • if a user guesses the top level identifier the expression is bound to, it can be used to print a silly string
  • zombies could be created by multiple runplugs calls, leading to blocking on endless output. the resulting zombies accumulate, eventually leading to DOS. (if waitForProcess was broken)

Template Haskell

We believe that Template Haskell can be made safe for users by hiding runIO and reify.

See also