1995-11-13 - Java’s verifier (was Re: Java insecurity - long - argumentative - you are warned.)

Header Data

From: Simon Spero <ses@tipper.oit.unc.edu>
To: “Dietrich J. Kappe” <goedel@tezcat.com>
Message Hash: 6ea81f0cb22933947078eeb7ef9353e91f37f62a98eef1bdabfd7b197e33d20a
Message ID: <Pine.SOL.3.91.951108225902.10149A-100000@chivalry>
Reply To: <v01510103acc4ce7ca31a@[206.1.161.4]>
UTC Datetime: 1995-11-13 10:57:34 UTC
Raw Date: Mon, 13 Nov 1995 18:57:34 +0800

Raw message

From: Simon Spero <ses@tipper.oit.unc.edu>
Date: Mon, 13 Nov 1995 18:57:34 +0800
To: "Dietrich J. Kappe" <goedel@tezcat.com>
Subject: Java's verifier (was Re: Java insecurity - long - argumentative - you are warned.)
In-Reply-To: <v01510103acc4ce7ca31a@[206.1.161.4]>
Message-ID: <Pine.SOL.3.91.951108225902.10149A-100000@chivalry>
MIME-Version: 1.0
Content-Type: text/plain


On Tue, 7 Nov 1995, Dietrich J. Kappe wrote:
> 
> This "checking," as any comp-sci undergrad will tell you, amounts to solving
> the halting problem for the java interpreter. While this is possible for a
[...]
> If you can write a checker that works in a reasonable amount of time, I'll
> write a turing machine simulator that'll do something nasty if the input
> machine halts. Then we'll split the fame and fortune for solving the 5 state

Yeah, but when you graduate, they let you in to the real secret- if a 
problem is NPC or Undecideable, either use some wild guesswork (oops, 
heuristic), or try solving enough of the problem to be usable. 

The java verifier not only terminates, but runs in time linear to the 
size of the program to be verified. This is because the verifier doesn't 
really calculate whether a program is safe or not; it determines whether 
it can prove that the program is safe or not. It's possible to generate 
sequences of bytecodes that do not perform unsafe accesses, yet which are 
still rejected by the verifier because they violate it's requirements.

The verifier can be considered to be an abstract interpretation over the 
depth and type-state of the operand stack. If the state is known before 
an instruction, it is always known after that instruction, and if there 
is more than one way to arrive at an instruction, each control path must 
arrive at that instruction with the same typestate.

Examples (not real JavaVM, but similar)

	load-int  <int>	== push an int onto the stack.
			before: ...
			after : ...,int

	load-float <flt>== push a float onto the tack
			before: ...
			after: ...,float
		
	add-int		== pop two ints off the stack, push sum onto stack
			before: ...,int,int
			after:  ...,int

	blt <val><add>  == pop an int off the stack, compare to val, and 
			      jump to address add if int is less than val

			before:  ...,int
			after:	 ...

	jmp <add>	== jump to adddress add
			before: ...
			after: ...

VALID
	load-int  1	; stack = (int)
	load-int  1	; stack = (int), (int)
	add-int		; stack = (int)

INVALID
	load-int 1	; stack = (int)
	load-float 1.0	; stack = (int) (float)
	add-int		; error, stack != (int), (int)

VALID
	load-int 2	;stack = (int)
	blt 1, a  	;stack = null
	load-int 3	; stack = (int)
	jmp b		; stack = (int)
    a	load-int 1	; stack = (int);
    b 	load-int 4	; stack = (int) (int)
	add-int		; stack = (int)

INVALID
	load-int 2	;stack = (int)
	blt 1, a  	;stack = null
	load-float 3	; stack = (float)
	jmp b		; stack = (float)	
    a	load-int 1	; stack = (int);	
    b 	load-int 4	; stack = ERROR (float || int)
	add-int		; stack = (int) ERROR

This last example is invlaid, even though it's possible in this case to show 
dynamically that the program will always arrive at b with an int on the 
stack; there are still two control paths that arive at b, one with an 
int, the other with a float.

I hope this makes sense

Simon





Thread