Re: [zones-discuss] Need code review: 6558857 ZSD callback locking cause deadlocks

2007-09-04 Thread Erik Nordmark 
Nicolas Williams wrote:
 On Wed, Aug 29, 2007 at 03:47:22PM -0700, Erik Nordmark wrote:
 Comments?
 
  - $SRC/uts/common/sys/zone.h:483
  - $SRC/uts/common/sys/netstack.h:90
 
s/looks/locks/
 
 
  - $SRC/uts/common/os/zone.c:321
 
s/ZONE_EVENT_UNINITIALIZED/ZONE_EVENT_INITIALIZED/
 
(i.e., s/UN//)
 
 
 This is just a quick glance.  The rest of the changes is too much for me
 to review right now or anytime soon, thus these comments do not
 represent a commitment to review the rest.

Thanks.

I've updated the webrev with the above fixes.

Erik
___
zones-discuss mailing list
zones-discuss@opensolaris.org

Re: [zones-discuss] Need code review: 6558857 ZSD callback locking cause deadlocks

2007-08-29 Thread Nicolas Williams 
On Wed, Aug 29, 2007 at 03:52:49PM -0700, Dan Price wrote:
 On Wed 29 Aug 2007 at 03:47PM, Erik Nordmark wrote:
  
  Fixing the above CR requires changing how locking is done across the ZSD 
  callbacks. The new design is to determine what callbacks are needed 
  while holding the usual locks, but not hold any locks across the actual 
  callbacks.
  
  This holds up under extreme stress testing where zones come and go at 
  the same time as zone_key_create and zone_key_destroy are being called.
  
  The code changes are at:
  
  http://cr.grommit.com/~nordmark/6558857/webrev/
  
  Comments?
 
 Wow, it's not trivial.  I'll try to dig in over the next couple of
 days.

Hmmm, someone commented (in the VSD case) that maybe we ought to have a
common object-specific key facility.  Here we have fixes to two
different object-specific key facilities.  I think a heads up to the RE
for CR 6588504 is in order.  And maybe we should think about having a
generic OSD.

Nico
-- 
___
zones-discuss mailing list
zones-discuss@opensolaris.org

Re: [zones-discuss] Need code review: 6558857 ZSD callback locking cause deadlocks

2007-08-29 Thread Erik Nordmark 
Nicolas Williams wrote:

 Hmmm, someone commented (in the VSD case) that maybe we ought to have a
 common object-specific key facility.  Here we have fixes to two
 different object-specific key facilities.  I think a heads up to the RE
 for CR 6588504 is in order.  And maybe we should think about having a
 generic OSD.

Commonality is good but requires some care.

For instance, just looking at zsd and the netstack there are quite 
different ordering requirements with the netstacks needing things to be 
destroyed in the reverse order of which they were created.

   Erik


___
zones-discuss mailing list
zones-discuss@opensolaris.org