LTTng user space tracing

Hi,

I looked more closely at mmap() and how it behaves upon process 
termination with Rebecca, and the kernel seems to safely unmap the 
region when a process is killed.

It would therefore make sense to use this approach. However, two 
questions arise :

- It will make the tracing impact more difficult to identify than with a 
separate process : with my current approach, all the I/O and the waiting 
is done on behalf of the "companion" process rather than in the traced 
process.

- Is there a way to do the same for network transport ? The companion 
process approach can be more or less trivially changed into a network 
client, but the file-backed mmap does not seem to have this capability. 
Ideas ?

Mathieu