Here's a wierd one for ya' (JNI - JVM Maladies!)

Question

I'm at my wit's end, sorry for the abstract manner in which I'm going to describe this, I don't have time to write a novel about the intricacies of what the C code described herein does and I'm sure you won't have time to read it.
I'm using JNI to wrap some C that's about a billion lines deep and at the very core of our product (it would take ten years to replace with an army of engineers).
In the back end, we start a process that allows other processes to connect to its shared memory block (it was designed to be connected to by other C executables). I've wrapped the C client that connects to the server process in JNI. To find the hooks the connecting process uses, it (the C client) reads environment variables that list the locations of files and the address at which the shared memory starts. The connecting process also grabs a semaphore created by the server process.
Okay, that said, the problem I'm having is less complicated. My JNI wrapper works perfectly every time I invoke it from the command line. The JVM comes up, loads the shared library, the C code talks to the already running process and does the work it needs to. THIS NEVER, EVER FAILS! However, when I take the very same java code and put it in a servlet, the C code I'm calling via JNI cannot find the hooks it needs (the error I'm getting has something to do with a semop()). The JNI side is working fine, I don't have a loadLibrary or class locator problem here. For some reason however, the webserver's JVM is hiding shared resources from the native code it is calling.
My question is, what is the @#$!&%#!!! difference between the two JAVA processes calling the same code? Why doesn't a servlet running in Tomcat have access to the already running process when the command line process that does have access and Tomcat have been invoked from the same environment?
I've got some theories about the webserver process having different access permissions or it's use of threads (the command line process that always works is single threaded) and child process spawning, but I don't know enough about the inner workings of the JVM to know where I should look. How do these types of permissions propagate? If the webserver process spawns a thread, does the thread have any knowledge of the environment in which its parent process was created? What about a child process? Any insights?
I've been able to break this in the same way on Solaris and HP-UX (believe it or not, NT doesn't care, I think it might have something to do with using a mutex instead of a semaphore). By the way, I'm using root to spawn all of these processes.
If you can give me a clue as to how to fix this, YOU ARE A GOD!
Anyway, your help would be appreciated!

Answer 1

When I'm running it from the command line, I'm usually logged in as root. I have attempted successfully to run it as other users from the command line. I'm also starting the webserver process as root.
I'm not familiar with a way to set platform specific environment variables in Tomcat, I'm sure I could use the -D option in starting the JVM. However, I have written a native stub that uses putenv() to set all of the environment variables present at the command line when I'm able to successfully invoke my JNI wrapper. This doesn't seem to change anything.
I've thought about this being a permissions propagation issue within the webserver. Any clues as to how I can confirm it?
Thanks for your help!