Messages in this thread | | | Date | Tue, 11 Oct 2011 09:14:04 -0400 | From | Jeff Layton <> | Subject | Re: [PATCH 3/4] sunrpc: make rpc_wait_bit_killable handle freeze events |
| |
On Tue, 11 Oct 2011 05:52:57 -0700 "Myklebust, Trond" <Trond.Myklebust@netapp.com> wrote:
> > -----Original Message----- > > From: Jeff Layton [mailto:jlayton@redhat.com] > > Sent: Tuesday, October 11, 2011 6:12 AM > > To: Pavel Machek > > Cc: Myklebust, Trond; smfrench@gmail.com; rjw@sisk.pl; linux- > > pm@lists.linux-foundation.org; linux-cifs@vger.kernel.org; linux- > > nfs@vger.kernel.org; john@Calva.COM; linux-kernel@vger.kernel.org > > Subject: Re: [PATCH 3/4] sunrpc: make rpc_wait_bit_killable handle > freeze > > events > > > > On Tue, 11 Oct 2011 08:19:23 +0200 > > Pavel Machek <pavel@ucw.cz> wrote: > > > > > On Wed 2011-09-28 07:52:40, Jeff Layton wrote: > > > > Allow the wait_on_bit_killable sleeps in SUNRPC layer to respect > the > > > > freezer. This should allow suspend and hibernate events to occur, > > > > even when there are RPC's pending on the wire. > > > > > > Will the RPC protocols used handle that correctly? What will happen > > > during resume? > > > > > > > That depends on the state of the socket during resume. If the > > suspend/resume is quick enough, then the socket may still be > connected, or > > if we're using UDP then we might just get the reply and carry on > successfully. > > Otherwise, the call will eventually time out, or will be cancelled > when the > > kernel finds that the socket has been closed. > > > > Either way, this should do the right thing. > > Well... The problem when this sort of thing happens is with the replay > cache. If the RPC in question was a mkdir, for instance, then replaying > the RPC call when you wake up can be problematic because chances are > that the server will have forgotten who created the directory, and so > will reply with EEXIST instead of OK. > However this is a generic problem when the client is unable to talk to > the server for a while, and is not particular to suspend. >
Yeah, that's always a problem. You can hit the same thing if you just unplug the cable from the network interface at an inopportune time.
I think this patch is really the best we can do under those circumstances. Eventually we'll all be on 4.1 with sessions and this problem will go away, right? ;)
-- Jeff Layton <jlayton@redhat.com>
| |