Trails and Troubles with NWNx4 & wine

[kuscotopia]

I am hoping that someone here will be able to speak to running NWNx4 + nwn2server within wine.

For starters I am wondering if this is even possible with the latest version of NWNx4. Does anyone know of any success stories with regards to hosting a PW via wine?

I have been working on this for the past couple days, and I am able to host built-in campaigns just fine. However when I go to load my custom module NWNx4 decides that nwn2server is taking too long to load and starts kicking up new server processes even though the original spawned server returns with the module loaded. Is there a parameter I can set within nwnx.ini that will cause it to wait longer?

I tried Skywing's branch in hopes that it would solve my problems but I only saw a bunch of the following messages displayed in the nwnx4_gui window:

[Zebranky]

I have no first-hand experience with NWN2/NWNX4/wine, but my understanding is "it can be made to work, but you should really just use Windows or a Windows VM". Yorlik is the authority on that, IIRC, so maybe he'll chime in here.

As to your error, it looks to me like an issue with your wine setup -- it sounds like wine is supposed to provide NtContinue in ntdll.dll, but isn't for some reason, so it's exploding. Unless you're using a very old version of wine, I don't know why that would be.
_________________
Win32 SVN builds: http://www.mercuric.net/nwn/nwnx/

<Fluffy-Kooshy> NWNx plugin is to this as nuclear warheads are to getting rid of fire ants.

<ThriWork> whenever I hear nwn extender, I think what does NWN need a penis extender for?

[kuscotopia]

The version of wine I am using is 1.0. Not the latest, but the latest available for Ubuntu 8.04 LTS standard repositories .

[Skywing]

Sounds like the WINE folks haven't implemented things all the way.

It would be possible to rework the injector code I posted to not use NtContinue. The change I would make would be to write out into the jump trampoline mov instructions to set all of the registers, followed by a jmp rel32 to branch to the old eip value in lieu of calling NtContinue. (Standard longjmp semantics not used due to a need to preserve volatile registers.)

The current nwnx4 release uses Detours and I've no idea what the state of WINE compatibility of that is.

[kuscotopia]

I will have to take your word for it Skywing. I am not much of a windows developer as I prefer to stay as close to gcc as possible.

Detours SEEMED to work. I am only basing that on not seeing any error messages fly-by (or in log files) while the module was loaded. A good test would be to load the demo module, just to see if database access is possible.

I won't be able to fire up the demo module tonight, but I will post any findings as soon as I have them.

[kuscotopia]

I have to rescind my previous statement. The detours version does not work at all . For the run I cloned a known good nwnx4 configuration and ran the server within both a windows VM and wine.

The windows version gave a success message regarding plugins whereas the wine emulated version continued to display timeout errors and spawning new instances of nwn2server. Although I was able to connect to and login to both hostings the windows hosted demo module was the only one able to manipulate data within the DB.

I am guessing this means detours flavor is failing to inject the dlls properly.

Skywing would you have the time/motivation to make the changes you mentioned and give your branch a go at wine emulation?

[mckillroy]

The Nwnx4/Detours version runs for me on wine for Solaris.
It ~should~ run on Linux, but I heared varying success/failure.
The problem is to get the hook dll loaded on runtime.
Usually this is done by nwnx4.
Since it doesn't work natively the workaround I use is to edit
the import table of the so called PE-header of the server executable.
The executable then "thinks" it needs the hook dll natively and thus the dll
gets loaded by the run-time linker of windows / wine and not by nwnx4.

You can use the CFF Explorer program to do this edit.

http://www.ntcore.com/exsuite.php

You must import the "dummy" function exported by the hook dll
==>by ordinal into the import table.

Once you have done this the nwn2server executable will require the hook dll in any case. So be sure to have a backup of the server exe.


Good luck !!
_________________
/McKillroy
a.k.a Yorlik

[kuscotopia]

Well...maybe it is me...but importing NWNX4_Hook.dll into nwn2server.exe via CFF Explorer didn't change any of my results.

For a control, I have cloned my setup (with "improved" executable) in a windows VM and under wine.

Under windows, everything loads great. I see success messages NWNX4_GUI, nwnx4 log files are generated, and the demo module works just great.

Under wine though I see the same problem I have encountered previously. NWNX4_GUI comes up starts the server process and although the server loads the demo module, NWNX4 reports a failure to load the module and begins starting new server processes. No log files are created for nwnx4 or error messages reported. Connecting to the module succeeds but no database access. The only difference I see is a message when I shutdown nwn2server which reports, "NWNx4 shutting down."

Am I missing a dependency for NWNx4 that is preventing proper loading of nwn2server of the xp_* plugins?

[mckillroy]

It is possible you need things not found in the native wine installation.
I suggest you look for the "winetricks" script to add some common stuff needed by many programs, like directX and such.
You should also try to run the hacked executable directly under your windows vm and see if it loads the hook dll even without nwnx4 (which should happen) and then check the demo module.

Good luck.
_________________
/McKillroy
a.k.a Yorlik

[kuscotopia]

After a long break..I have returned .

So, after a few months break..I have come around to trying again, sadly the same results as before.

To answer you previous post mckillroy, I have followed the standard install procedure for NWN2 as listed on WineHQ. Below are the winetricks I have installed:

vcrun2005, directx9, dotnet20

alternatively, I also tried this configuration:

vrun2005, directx9, mono-2.6.1

CFF Explore was a no-go for me, I can't even get the spliced server binary on widows to properly run the demo modules. Each time I log into the module I am met with warning messages about NWNX4 not being properly installed (unless of course I use NWNX4_GUI.exe).

*Sigh* So at this point I am still at a loss...mckillroy, what version of wine are your running? Could you perhaps share an install procedure using the latest NWNx4 1.09? Or perhaps, list which winetricks you used?

Thanks!

[Skywing]

My branch and the current one release use completely different methods of injecting the nwnx hook module -- did you end up trying both on the latest wine? Maybe wine has enough implemented for one of the mechanisms to work now.

[kuscotopia]

I have tried both Skywing...I just updated to the latest version of wine, 1.1.39.

I see the same output from NWNX4_GUI.exe from your branch as I did before:

[Skywing]

Yes, you could change the injected code to emit some machine code to restore the register context manually as immediate operands.

NtContinue is just used as essentially a longjmp to restore the original thread context here. It's probably best to replace the dependency on the undocumented API anyway, but I already had this code laying around and it works on (full) Windows.

For example, you could write out a series of instructions in the injected blob like so:

mov eax, 0x<oldeax>
mov ebx, 0x<oldebx>
mov esp, 0x<oldesp>
...
jmp <relative-32 delta to oldeip>

[kuscotopia]

I have very little experience with long-jumps and windows development, so I am sure the next few weeks of hackery should prove interesting. Skywing if you are feeling ambitious and would be willing to help me with the assembly generation, I certainly need the help in that department

I have checked out the source code and managed to find a copy of Visual Studio 2005 Express. The plan is to poke around the code a bit, just to see if I can get passed the timeout.

In the main-line code, the detours dll injection is not failing which makes me wonder why the IPC is borked. It looks to just be a shared memory structure, so I wonder if I am hitting the injected code at all. The lack of log files during execution certainly supports this theory...

I don't know at which step the plugins initialize themselves, but perhaps it is after the initialized boolean gets set to true? I think my first test will be to break the IPC and force a manual sleep. Perhaps everything is working properly and we just need to blow passed the handshake...

[Skywing]

The general sequence of events at a high level is:

- nwnx controller (service or gui) launches nwn2server.exe (and arranges for the nwnx4_hook.dll module to be injected into nwn2server.exe).
- nwnx controller now waits for nwnx4_hook to signal that initialization is complete
- nwn2server.exe process is initialized and the injected nwnx4_hook.dll initializes (in nwn2server.exe address space)
- nwnx4_hook.dll loads all plugin modules, performs various other initial setup tasks, then indicates to the nwnx controller that hook initialization has completed
- control within nwn2server.exe transfers to nwn2server's main program entry point
- nwnx controller receives notification that the hook dll has successfully spun up, and all is well