I've just modified the code for 3.4 to properly use mutexes and
"read-write" locks, doing away with the need for GIANT being held
while executing IPFilter code.  I've uploaded this as:

http://coombs.anu.edu.au/~avalon/ip_fil3.4next.tar.gz

I'm interested in hearing from people using FreeBSD 5.x about how
this goes, especially on multi-CPU systems, whether the system
falls in a heap or runs well.  In case I've frightened you, it should
all go well, modulo bugs in FreeBSD's locking and in multi-CPU systems,
result in better performance than what currently runs.  The only
difference between the locking for FreeBSD and Solaris (where it has
had to run in an SMP-friendly environment for years) is the interface
to kernel locking provided by the OS.

Cheers,
Darren

Reply via email to