This is interesting experimentation. I think what emerges is that clearly
CondSymLinks shouldn't be processed in the kernel; probably all the kernel
should need would be a new inode type for them, and code to pass their
processing out to kerneld. /proc/<pid>/env would be one way of giving kerneld
access to per-process env info; another possibility would be some mechanism
whereby kerneld could send a message to the user process requesting an env
lookup. This would require some weird libc hacks, but weird libc and kerneld
hacks might be better than more wired code in kernel-space.
----Jim " " \ ____/| / \ \ o.O| / \ =(_)= / \ U / \ | / \ #/ ACK! PHTHPHTH! "I am not a crook!"