Return-Path: Received: from mail-oi0-f41.google.com ([209.85.218.41]:35647 "EHLO mail-oi0-f41.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751175AbbIQNes convert rfc822-to-8bit (ORCPT ); Thu, 17 Sep 2015 09:34:48 -0400 Received: by oiww128 with SMTP id w128so9381646oiw.2 for ; Thu, 17 Sep 2015 06:34:48 -0700 (PDT) MIME-Version: 1.0 In-Reply-To: References: Date: Thu, 17 Sep 2015 09:34:47 -0400 Message-ID: Subject: Re: Failing to send a CLOSE if file is opened WRONLY and server reboots on a 4.x mount From: Trond Myklebust To: Olga Kornievskaia Cc: linux-nfs Content-Type: text/plain; charset=UTF-8 Sender: linux-nfs-owner@vger.kernel.org List-ID: On Tue, Sep 15, 2015 at 12:52 PM, Olga Kornievskaia wrote: > On Tue, Sep 15, 2015 at 11:49 AM, Trond Myklebust > wrote: >> On Tue, Sep 15, 2015 at 10:27 AM, Olga Kornievskaia wrote: >>> On Tue, Sep 15, 2015 at 9:39 AM, Trond Myklebust >>> wrote: >>>> On Mon, Sep 14, 2015 at 7:54 PM, Olga Kornievskaia wrote: >>>>> A test case is as the description says: >>>>> open(foobar, O_WRONLY); >>>>> sleep() --> reboot the server >>>>> close(foobar) >>>>> >>>>> The bug is because in nfs4state.c in nfs4_reclaim_open_state() a few >>>>> line before going to restart, there is >>>>> clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &state->flags). >>>>> >>>>> NFS4CLNT_RECLAIM_NOGRACE is a flag for the client states not open >>>>> owner states. Value of NFS4CLNT_RECLAIM_NOGRACE is 4 which is the >>>>> value of NFS_O_WRONLY_STATE in nfs4_state->flags. So clearing it wipes >>>>> out state and when we go to close it, “call_close” doesn’t get set as >>>>> state flag is not set and CLOSE doesn’t go on the wire. >>>>> >>>>> That line was introduced to fix an infinite loop for OPEN recovery >>>>> upon receiving a BAD_STATEID error: commit e8d975e73. I have tested >>>>> injecting BAD_STATEID error using the patch below and the code >>>>> recovers without problems. However, I'm not sure the clearing of the >>>>> bit is needed any more. I have tested for infinite loop by reverting >>>>> the patch and didn't hit the infinite loop. >>>>> >>>>> diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c >>>>> index da73bc4..5db3246 100644 >>>>> --- a/fs/nfs/nfs4state.c >>>>> +++ b/fs/nfs/nfs4state.c >>>>> @@ -1481,7 +1481,7 @@ restart: >>>>> spin_unlock(&state->state_lock); >>>>> } >>>>> nfs4_put_open_state(state); >>>>> - clear_bit(NFS4CLNT_RECLAIM_NOGRACE, >>>>> + clear_bit(NFS_STATE_RECLAIM_NOGRACE, >>>>> &state->flags); >>>>> spin_lock(&sp->so_lock); >>>>> goto restart; >>>> >>>> That's an obvious typo. Thanks for spotting it! >>>> >>>> As for whether or not the bit clear is needed at all, I think it is >>>> for NFSv4 on older kernels. On newer kernels, we do have the NFSv4 >>>> state recovery drain the slot table (just like we've always done for >>>> NFSv4.1) and so I agree that those kernels probably won't be >>>> afflicted. >>>> >>> >>> Thanks Trond. Do you need me to resubmit it without the last paragraph >>> or is the patch ok as is? >>> >> >> I can easily remove that paragraph when applying the patch, if you >> agree that it is superfluous. > > Thanks. Works for me. > May I also add a signed-off-by line from you? I can't really apply this (or any other patches) without it. Cheers Trond