Hey Pavel,
My only concern w/removing the hold/rele pair is that you
cannot guarantee that the nfs4_server_t doesn't disappear
from underneath you, since you don't explicitly hold it. Is
this a concern ? (Maybe you've already thought about that)
rick
On Fri, Jul 17, 2009 at 12:13:04PM +0200, Pavel Filipensky wrote:
| Hi,
|
| can I get a code review for 6861594 NFSv4 client Deadlock: cycle in
| blocking chain at nfs4_move_mi
|
| Webrev:
|
| http://cr.opensolaris.org/~pavelf/6861594/
|
|
| Background:
|
| If there are two threads doing a failover for differnet mounts at the
| same time, both of them need to grab
| the s_lock for the new and the old nfs server. To prevent a deadlock
| they must grab the s_locks it in the same order.
| nfs4_move_mi() is not safe since it does not guarantee the same order
| and allows the deadlock to happen.
|
|
|
| Thanks,
| Pavel
|
|
| _______________________________________________
| nfs-discuss mailing list
| nfs-discuss at opensolaris.org
--
