incomplete emulation of fstenv under TCG

Steps to reproduce:

1) Install Windows (tried XP and 7) in qemu (tried qemu without kvm and qemu-kvm).

2) Get OllyDbg ( http://ollydbg.de/odbg200.zip ).

3) Use some Metasploit-encoded file, example included.

It is not a virus!

File was generated with Metasploit, command (if i remember it right): `msfpayload windows/exec cmd=notepad R | msfencode -e x86/shikata_ga_nai -t exe -o cmd_exec_notepad.shikata_ga_nai.exe`.

4) Launch the file under Windows in qemu, make sure it opens a notepad.

5) Open file under OllyDbg, run (F9) it there. It opens a notpad. Close OllyDbg.

6) Open file under OllyDbg, trace over (Ctrl+F12) it there. It fails with `Access violation when writing to [some address]`.
Command: 316A 13, XOR DWORD PTR DS:[EDX+13],EBP 

Under native Windows, the trace over command works fine.

Under VMware the trace works fine.
Under VirtualBox it also fails (checked in the spring).

$ qemu-kvm --version
QEMU PC emulator version 0.12.5 (qemu-kvm-0.12.5), Copyright (c) 2003-2008 Fabrice Bellard





http://imagebin.ca/view/zue0YNZ.html - This is VMware screenshot just before executing that command.

Looks like something is wrong with EDX register in OllyDbg under QEMU. That register was popped as a result of FSTENV command.

linux-user testcase:

extern void *x;

int main()
{
	int a;
	asm volatile ("x: fldz\n\
	     push %%edx\n\
	     .byte 0xd9,0x74,0x24,0xf4\n\
	     pop %%edx\n" : "=d" (a) : : "memory");
	printf ("%x %x\n", a, &x);
}

yakj:~ pbonzini$ ./a.out 
80483d9 80483d9
yakj:~ pbonzini$ qemu-i386 ./a.out 
0 80483d9


On Sat, Oct 16, 2010 at 3:24 PM, Paolo Bonzini <email address hidden> wrote:
> linux-user testcase:
>
> extern void *x;
>
> int main()
> {
>        int a;
>        asm volatile ("x: fldz\n\
>             push %%edx\n\
>             .byte 0xd9,0x74,0x24,0xf4\n\
>             pop %%edx\n" : "=d" (a) : : "memory");
>        printf ("%x %x\n", a, &x);
> }
>
> yakj:~ pbonzini$ ./a.out
> 80483d9 80483d9
> yakj:~ pbonzini$ qemu-i386 ./a.out
> 0 80483d9
>
>
> ** Summary changed:
>
> - Ollydbg under Windows in qemu does not work as it does under native Windows.
> + incomplete emulation of fstenv under TCG

Each FP instruction should store the needed data into new env fields,
including IP, CS and opcode. These are known at translation time. Data
pointers need to be saved at execution time.

The new env fields would be then used by FSTENV, FSAVE, FXSAVE (which
also suffer from the problem) etc.

Any news on this bug?

The full testcase:

#include <stdio.h>
extern void *x;
int main() {
   int a;
   asm volatile ("x: fldz\n\
   push %%edx\n\
   fnstenv -0xc(%%esp)\n\
   pop %%edx\n" : "=d" (a) : : "memory");
   printf ("%x %x\n", a, &x);
   return 0;
}

$ gcc -m32 test.c -o test
$ ./test
80483ae 80483ae
$ ./qemu/i386-linux-user/qemu-i386 ./test
0 80483ae

Cleaned up .byte row.

Example patch.
Works only for FLDZ and only in -singlestep mode.
Based on version 0.12.5.

This was just an example of how it could be done.

$ ./qemu-0.12.5/i386-linux-user/qemu-i386 -singlestep ./test
80483b4 80483b4


Ok. Here is a full patch to QEMU 0.13.

Works with and without -singlestep.
Works with all fpu instructions.

Should also work with fsave.

I've had quite some problems with this bug as well. It would be really nice if it could be fixed.

I have ported Chalkerx's patch to QEMU-1.5.0. The patch is attached.

// MOKI

The bug is still present in the newly released QEMU-1.5.1. I've ported Chalkerx's patch to this release as well. See attached patch.

Is there a problem with this patch, since it has not been committed?

// MOKI

Thanks for porting the patch.

This is the mailing thread I started back in 2010 with that patch:
http://lists.gnu.org/archive/html/qemu-devel/2010-11/msg02497.html
That thread has some problems noted.

Sadly, I did not have enough free time to investigate all the other
places that should be fixed for proper fpip/etc. support.

That patch seems to be related:
https://patchwork.kernel.org/patch/871232/, but is kvm only. You could
try contacting the person that did the patch for kvm.

Nikita Skovoroda.

2013/6/29 Morten Shearman Kirkegaard <email address hidden>:
> The bug is still present in the newly released QEMU-1.5.1. I've ported
> Chalkerx's patch to this release as well. See attached patch.
>
> Is there a problem with this patch, since it has not been committed?
>
> // MOKI
>
> ** Patch added: "patch-qemu-1.5.1-fpip.diff"
>    https://bugs.launchpad.net/qemu/+bug/661696/+attachment/3718352/+files/patch-qemu-1.5.1-fpip.diff


Thanks for the links Nikita.

I'll see if I can add the missing features (fpop, fpdp) to the patch. Since I don't depend on those features, it will be relatively low priority for me, and I will probably not get to it before late August.

// MOKI

The QEMU project is currently considering to move its bug tracking to
another system. For this we need to know which bugs are still valid
and which could be closed already. Thus we are setting older bugs to
"Incomplete" now.

If you still think this bug report here is valid, then please switch
the state back to "New" within the next 60 days, otherwise this report
will be marked as "Expired". Or please mark it as "Fix Released" if
the problem has been solved with a newer version of QEMU already.

Thank you and sorry for the inconvenience.


Test case in comment #7 still fails -- still a bug.



This is an automated cleanup. This bug report has been moved
to QEMU's new bug tracker on gitlab.com and thus gets marked
as 'expired' now. Please continue with the discussion here:

 https://gitlab.com/qemu-project/qemu/-/issues/67