What every C Programmer should know about undefined behavior #2/3 [LWN.net]

What every C Programmer should know about undefined behavior #2/3

Posted May 16, 2011 22:56 UTC (Mon) by cmccabe (guest, #60281)
In reply to: What every C Programmer should know about undefined behavior #2/3 by njs
Parent article: What every C Programmer should know about undefined behavior #2/3

> Or another example: C++'s "private:" keyword is useful not because it
> gives some kind of 'security' or something (like many documents about it
> seem to think), but because it lets me know for certain (i.e., prove) that
> I only have to look at a certain bounded amount of code if I want to see
> how certain variables are modified

Actually, the "private" keyword in C++ allows you to prove no such thing. I can simply typecast the class to a byte buffer and modify to my heart's content.

If I want to be even more evil, I can add:
#define public private
to the beginning of my .cc file and include the header file for your class. Then nobody will stop me from doing whatever I want with your private data-and methods-- not the compiler, and certainly not the linker.

Java also has a way for "unauthorized" classes to get access to private data. I think you can use the Reflect package to get at it.

Private data in Java and C++ is not and was not intended to provide the kind of guarantees a proof checker would need.

On the other hand, if the goal is to allow the programmer to have a reasonable mental model, they work pretty well!

> I just want better tools for *non-heuristic* reasoning about programs.

Too bad. Artificial intelligence is about at where chemistry was in 1000 AD. Theorem provers are good for proving theorems, but bad at real-world reasoning.

On the other hand, what I can offer you is sandboxed programming langauges and model checkers. It's amazing how much more productive you can be when you have a garbage colector and a good type system.

to post comments

What every C Programmer should know about undefined behavior #2/3

Posted May 16, 2011 22:58 UTC (Mon) by cmccabe (guest, #60281) [Link] (7 responses)

That should have been
#define private public
#define protected public

What every C Programmer should know about undefined behavior #2/3

Posted May 16, 2011 23:12 UTC (Mon) by tialaramex (subscriber, #21167) [Link] (5 responses)

I think a compliant JVM's byte code verifier + security policy ensures that you can't use Reflection to poke the private members of somebody else's code. I think you'll get a SecurityException at the line where you attempt to violate policy if you go about it in the natural way. And if you try to write raw byte code to sidestep, the verifier will reject your .class as invalid.

Of course in a debug mode, or in a non-compliant JVM, or if there's a bug, you may make this work. But in /theory/ at least they've thought of this, so it would be fair for a Java programmer (unlike a C++ programmer) to treat private members as genuinely private.

What every C Programmer should know about undefined behavior #2/3

Posted May 16, 2011 23:30 UTC (Mon) by cmccabe (guest, #60281) [Link] (4 responses)

I can get the private data of any class that implements Serializable pretty easily. I can always get access to protected data members by defining my own subclass (unless the class is final, but putting protected data members in a final class is an odd choice).

I have a pretty good hunch that this API gives me a hole in "private" big enough to drive my truck through.
http://download.oracle.com/javase/1.5.0/docs/api/java/lan...

I haven't tried it, though.

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 1:07 UTC (Tue) by foom (subscriber, #14868) [Link]

If there's no security manager policy in the way, you certainly can access private members with the obvious reflection APIs. If there is a security manager policy, it also prevents you from doing the other Evil Things that would let you access the private members.

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 8:54 UTC (Tue) by tialaramex (subscriber, #21167) [Link] (1 responses)

Not only might a class choose not to implement Serializable if it doesn't want you looking at its internals, it might also choose to implement Serializable in a way that's completely unhelpful while obeying the spec. In some cases this might be the most natural approach e.g. their magic internal values could change in a future otherwise compatible replacement, so they reconstruct them when de-serialising, rather than including them in the serialised output.

In other cases it would be very deliberate e.g. we can imagine a system where untrusted code runs in a Java sandbox on a remote system, and has access to certain serialisable objects which are sensitive, so their serialisations are encrypted, versioned and signed with keys not available to the untrusted code.

Even if you're allowed to make the subclass (security policy again) - your subclass doesn't get to look at protected data members from other instances, so this will often be useless. Remember this is Java, so type restrictions are enforced at runtime.

Basically this goes on and on, unlike in C++ the designers actually intended this to be enforced, not just a vague guideline to help those willing to help themselves. So even if you find a crack in the wall, someone will fix it. There really aren't any gaps "big enough to drive a truck through" as you imagine and as is the case in something like C++. If you want to drive a truck in, you need someone to conveniently open the truck-sized gate from the other side by disabling the relevant security policy.

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 9:01 UTC (Tue) by tialaramex (subscriber, #21167) [Link]

"your subclass doesn't get to look at protected data members from other instances"

Actually this bit might be wrong. You could be able to just pass in a suitable instance and have the code inside your imposter subclass poke around in its protected internals. But again the security policy gets to decide whether you're allowed to make this subclass at all (unlike the 'final' keyword this places no such restriction on the author of the rest of the system who may very well operate under a different policy).

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 17:21 UTC (Tue) by jeremiah (subscriber, #1221) [Link]

Just as an FYI the easiest way to get access is to use the reflection API, and the setAccessible() method. Things get a little deep in the reflection stuff, but being able to gain access to anything and modify comes in handy sometimes. I generally use it when writing serialization libraries.

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 8:39 UTC (Tue) by chad.netzer (subscriber, #4257) [Link]

You also might need:

#define class struct

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 1:28 UTC (Tue) by foom (subscriber, #14868) [Link]

> If I want to be even more evil, I can add:
> #define public private
> to the beginning of my .cc file and include the header file for your class. Then nobody will stop me from doing whatever I want with your private data-and methods-- not the compiler, and certainly not the linker.

The compiler/linker will prevent that if you're using MS VC++: on that platform, the mangling for public and private methods is different! Forces you to be more evil than just #define to get your way. :)

What every C Programmer should know about undefined behavior #2/3

Posted May 17, 2011 14:35 UTC (Tue) by dgm (subscriber, #49227) [Link]

> It's amazing how much more productive you can be when you have a garbage colector and a good type system.

<flame>Oh, yes! You can churn buggy *and s... l... o... w* code much faster (and in greater quantities!) with that. The new guy in the corner just does this all day long. Thanks C#!</flame>

What every C Programmer should know about undefined behavior #2/3

Posted May 19, 2011 23:57 UTC (Thu) by marcH (subscriber, #57642) [Link]

> I can simply typecast the class to a byte buffer and modify to my heart's content. If I want to be even more evil, I can...

To demonstrate that code verification tools do not work, you hit them with a sledgehammer. Interesting and conclusive.