Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Thu, 14 Dec 2017 13:51:15 +0100
From: harrim4n <harrim4n@...rim4n.com>
To: john-users@...ts.openwall.com
Subject: Re: Compiling John as a shared object

On 08.12.2017 20:44, Solar Designer wrote:
> On Tue, Nov 28, 2017 at 03:17:43PM +0100, harrim4n wrote:
>> I'm trying to compile John as a shared object to use in python bindings
>> for a project I'm working on. I'm using the latest version from
>> https://github.com/magnumripper/JohnTheRipper/tree/bleeding-jumbo.
> Great.
>
>> However I'm running into some issues doing so, namely the x86-64.S file.
>> As this file is assembly, I can't simply recompile it with -fPIC.
>>
>> This causes the following error when trying to create a "libjohn.so" file:
>>
>> /usr/bin/ld: x86-64.o: relocation R_X86_64_PC32 against symbol
>> `nt_buffer8x' can not be used when making a shared object; recompile
>> with -fPIC
>>
>> However, to me, the assembly already looks to be position independent.
> Unfortunately, it is not.  For PIC code, the RIP-relative addressing
> should be used to access the GOT entry, through which the actual
> "non-local" variables (not in the same translation unit) are accessed.
> Accessing them directly via RIP-relative addressing is not enough.
>
>> Does anyone have pointers as to how I could either modify the assembly
>> to be PI
> Here's the kind of changes needed:
>
> $ cat x.c
> extern int x;
> int f(void) { return x; }
> $ gcc -S -O2 x.c -o x1.s
> $ gcc -S -O2 x.c -o x2.s -fPIC
> $ diff -U0 x1.s x2.s
> [...]
> -	movl	x(%rip), %eax
> +	movq	x@...PCREL(%rip), %rax
> +	movl	(%rax), %eax

Ah, I see.

>
>> or any other way I could create python bindings for John?
> You could drop the assembly files altogether.  We have equivalent (and
> sometimes superior) C code for everything contained in there, except for
> the runtime CPU detection.  You just need to set the corresponding *_ASM
> defines in arch.h (a file that's generated when you ./configure, or is
> symlinked to e.g. x86-64.h) to 0 and "#undef NT_X86_64" to tell the rest
> of JtR not to expect assembly code for any of the crypto.

This seems like the best option to me. I'll give this a shot when I get
time and report back if I get any errors.

>
> Yet another option might be to look at and reuse some experience from
> this project:
>
> https://github.com/L0phtCrack/jtrdll
>
> This is how JtR is integrated with LC7, so it focuses on Windows builds
> and is not usable for you directly.
>
>> I am aware that I could rebuild the python interpreter and use static
>> linking, however I would like to avoid this option because of obvious
>> maintainability issues. I also tried simply using a statically built lib
>> in the setup.py file, but then I get
>>
>> ImportError: /usr/lib/python2.7/site-packages/pyJohn.so: undefined
>> symbol: fmt_nsec3
> I suppose you merely include samples of these errors, and you actually
> got a lot more of similar errors?  It's hard to debug this via a mailing
> list.  I guess you'd need to investigate it on your own.
>
> Alexander
Actually, this was the only error I got on import. But I suspect that
the python interpreter aborts on the first error it encounters, so if
this error was "fixed" a new one would pop up.

Thanks for your help!

harrim4n


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.