svn commit: r276747 - head/sys/netpfil/pf

[Gleb Smirnoff]

On Wed, Jan 07, 2015 at 11:46:31PM +0300, Gleb Smirnoff wrote:
T> On Tue, Jan 06, 2015 at 09:03:04AM +0000, Craig Rodrigues wrote:
T> C> Author: rodrigc
T> C> Date: Tue Jan  6 09:03:03 2015
T> C> New Revision: 276747
T> C> URL: https://svnweb.freebsd.org/changeset/base/276747
T> C> 
T> C> Log:
T> C>   Instead of creating a purge thread for every vnet, create
T> C>   a single purge thread and clean up all vnets from this thread.
T> C>   
T> C>   PR:                     194515
T> C>   Differential Revision:  D1315
T> C>   Submitted by:           Nikos Vassiliadis <nvass at gmx.com>
T> 
T> I am not sure that this is a good idea. The core idea of VNETs
T> is that they are isolated from each other. If we serialize purging,
T> then vnets are strongly affecting each other.
T> 
T> AFAIU, from the PR there is some panic fixed. What is the actual bug
T> and why couldn't it be fixed with having per-vnet thread?

So, after closer inspection, this commit is a completely messed up.

You blindly remove kproc_exit(). What do you think would happen on
'kldunload -f pf'?

You removed PF_RULES_RLOCK(). Cool! Now the purging thread doesn't
acquire the pf lock.

You substitute rw_sleep() with tsleep(). And the latter requires
Giant to be held. If you tried your change with INVARIANTS, it
would panic immediately.

-- 
Totus tuus, Glebius.