reactos-kernel

[ros-kernel] NtProtectVirtualMemory() is needed to run MSVC++ apps

[Eric K. <ek...@rz...>]

Hi!

I just found out why application compiled with MSVC++ don't run on ReactOS.

These apps have a slighly different layout compared to apps compiled with
MinGW. MSVC++ stores the IAT (Import Address Table) inside of the .rdata
section, MinGW stores the IAT inside of the .data section. Now the problem
is that the .rdata setion is read-only. To fix-up the IAT of an application
the offsets stored in the IAT must be overwritten by the real import
addresses. Obviously this wont work if the IAT is inside of a read-only
section.

To fix this bug I added calls to NtProtectVirtualMemory() to the fixup
routine to change the IAT's protection to read-write before it is modified
and reset its protection after the fixup is done.

When I had a look at NtProtectVirtualMemory() I saw that its core
functionality is disabled. What's wrong?


Regards,
Eric

Eric

Re: [ros-kernel] NtProtectVirtualMemory() is needed to run MSVC++ apps

[Steven E. <ste...@ya...>]

This rocks dude. If you want me to help with testing/debugging this I will reinstall MS_VC and
compile some simple applications once you figure out more of whats going on. 

Thanks
Steven

> I just found out why application compiled with MSVC++ don't run on ReactOS.
> These apps have a slighly different layout compared to apps compiled with
> MinGW. MSVC++ stores the IAT (Import Address Table) inside of the .rdata
> section, MinGW stores the IAT inside of the .data section. Now the problem
> is that the .rdata setion is read-only. To fix-up the IAT of an application
> the offsets stored in the IAT must be overwritten by the real import
> addresses. Obviously this wont work if the IAT is inside of a read-only
> section.
> 
> To fix this bug I added calls to NtProtectVirtualMemory() to the fixup
> routine to change the IAT's protection to read-write before it is modified
> and reset its protection after the fixup is done.
> 
> When I had a look at NtProtectVirtualMemory() I saw that its core
> functionality is disabled. What's wrong?


__________________________________________________
Do You Yahoo!?
Yahoo! Health - Feel better, live better
http://health.yahoo.com

Re: [ros-kernel] NtProtectVirtualMemory() is needed to run MSVC++ apps

[Robert K. <ro...@ko...>]

Congratulations. Let's hope this is the only thing that has to be fixed 
in order to run
MSVC compilend pgms.   It's at least an impotand step.

Whose mistake is it ?     MSVC, Mingw, NT or ROS?

Eric Kohl schrieb:

>Hi!
>
>I just found out why application compiled with MSVC++ don't run on ReactOS.
>
>These apps have a slighly different layout compared to apps compiled with
>MinGW. MSVC++ stores the IAT (Import Address Table) inside of the .rdata
>section, MinGW stores the IAT inside of the .data section. Now the problem
>is that the .rdata setion is read-only. To fix-up the IAT of an application
>the offsets stored in the IAT must be overwritten by the real import
>addresses. Obviously this wont work if the IAT is inside of a read-only
>section.
>
>To fix this bug I added calls to NtProtectVirtualMemory() to the fixup
>routine to change the IAT's protection to read-write before it is modified
>and reset its protection after the fixup is done.
>
>When I had a look at NtProtectVirtualMemory() I saw that its core
>functionality is disabled. What's wrong?
>
>
>Regards,
>Eric
>
>Eric
>
>
>
>-------------------------------------------------------
>This sf.net email is sponsored by:ThinkGeek
>Welcome to geek heaven.
>http://thinkgeek.com/sf
>_______________________________________________
>reactos-kernel mailing list
>rea...@li...
>https://lists.sourceforge.net/lists/listinfo/reactos-kernel
>  
>

Re: [ros-kernel] NtProtectVirtualMemory() is needed to run MSVC++ apps

[Eric K. <ek...@rz...>]

"Robert K." <ro...@ko...> wrote:

> Congratulations. Let's hope this is the only thing that has to be fixed
> in order to run
> MSVC compilend pgms.   It's at least an impotand step.

No, it seems there are more bugs ahead!
I changed the code in NtMapViewOfSection() to make all parts of a mapped
image section writable. The simple 'hello world!' example crashed upon
initialization. This is propably due to missing or buggy SEH-support.

> Whose mistake is it ?     MSVC, Mingw, NT or ROS?
It is not a mistake! MSVC executables and MinGW executables have a slightly
different layout. That's all. Of course, M$ could use this to get rid of
'evil, criminal, cancerous <insert more buzzwords here> open source
applications' by adding some checks to the loader.

Eric