Looking for brave testers ...

[sam]

Max Laier wrote:

>Hi all,
>
>I am looking for brave women and men to test the unexplored depth of a shared
>lock for pf. This will allow to have "even more parallelism" in the network
>code. As a small benefit you also get rid of a LOR that is suspected to cause
>deadlocks.
>
>So what must be tested?
>http://people.freebsd.org/~mlaier/MEGA_DIFF_WITH_IPFW_AND_PF_SHARED_LOCK.diff
>
>That monster of a patch is a joint work of Christian S.J. Peron (csjp@) and
>myself. It modifies pfil_hooks to pass a struct inpcb. This is used to
>forward the *inp - that is handed to ip_output() for local sends - to the
>pfil consumers. This avoids a lookup and a LOR (caused by the lookup). It
>also changes IPFW and PF to use a shared/exclusive lock for the "rules". This
>allows more than one packet in the ruleset evaluation at a time ...
>
>How to test:
>1) Get the patch, apply and install a kernel with it. Note that this breaks
>ABI for pfctl and friends. So make sure you rebuild and -install pfctl(8) at
>least.
>2) Run the patched kernel on:
> - SMP hardware (p4 HT is fine)
> - with debug.mpsafenet=1
> - with MPSAFE NICs
> - with a bunch of user/group rules.
>
>This is very, very ... very raw yet! IPFW is not ready in this version.
>Christian will post a version on freebsd-ipfw@, I guess.
>
>Nontheless, please test the pf part and give me feedback. I have it running
>fine on my router/gateway with ftp-proxy etc. ... so it can't be too bad. If
>you crash please try to get as much information as possible. Make sure you
>have WITNESS in the kernel.
>
>THANKS IN ADVANCE!
>
>  
>
I assumed execution of  "make release ...." in /usr/src/release 
directory after installed this patch in my own source tree (eg. 
/bdata/ncvs/src) should makes a new release contains this change. I need 
to test this in another machine or in a VMware machine to avoid 
disturbing my production work.

Thanks for your great work.
Sam