Return-Path: Received: from linuxhacker.ru ([217.76.32.60]:46926 "EHLO fiona.linuxhacker.ru" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755333AbcIRXbE (ORCPT ); Sun, 18 Sep 2016 19:31:04 -0400 Subject: Re: [PATCH v5 00/25] Fix delegation behaviour when server revokes some state Mime-Version: 1.0 (Apple Message framework v1283) Content-Type: text/plain; charset=windows-1252 From: Oleg Drokin In-Reply-To: Date: Sun, 18 Sep 2016 19:30:47 -0400 Cc: Schumaker Anna , List Linux NFS Mailing Message-Id: <88B2C3E0-3E6D-437E-9723-E1BBADF6F8D9@linuxhacker.ru> References: <1474089213-104014-1-git-send-email-trond.myklebust@primarydata.com> <838E325D-2DA1-4229-A046-76316302F813@linuxhacker.ru> <1474140727.7526.1.camel@primarydata.com> <1474148731.7526.3.camel@primarydata.com> <96CBC4AF-B0EA-4AFA-8068-E93CBACD1B0B@linuxhacker.ru> <8E0CD004-8BAE-4EC0-BAA2-ED7B83F85610@linuxhacker.ru> <2FBB0CA9-F6B4-4D80-B9C5-387DE4D267DD@primarydata.com> <3C3FEBB3-302A-4E22-ADD9-26DA1E6FAE3C@linuxhacker.ru> To: Trond Myklebust Sender: linux-nfs-owner@vger.kernel.org List-ID: On Sep 18, 2016, at 7:08 PM, Trond Myklebust wrote: > >> On Sep 18, 2016, at 18:44, Oleg Drokin wrote: >> >> >> On Sep 18, 2016, at 6:38 PM, Trond Myklebust wrote: >> >>> >>>> On Sep 18, 2016, at 12:23, Oleg Drokin wrote: >>>> >>>> >>>> On Sep 17, 2016, at 9:19 PM, Oleg Drokin wrote: >>>> >>>>> >>>>> On Sep 17, 2016, at 5:59 PM, Trond Myklebust wrote: >>>>> >>>>>> >>>>>>> On Sep 17, 2016, at 17:55, Oleg Drokin wrote: >>>>>>> >>>>>>> >>>>>>> On Sep 17, 2016, at 5:45 PM, Trond Myklebust wrote: >>>>>>> >>>>>>>> On Sat, 2016-09-17 at 15:32 -0400, Trond Myklebust wrote: >>>>>>>>> On Sat, 2016-09-17 at 15:16 -0400, Oleg Drokin wrote: >>>>>>>>>> >>>>>>>>>> On Sep 17, 2016, at 2:18 PM, Trond Myklebust wrote: >>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> On Sep 17, 2016, at 14:04, Oleg Drokin >>>>>>>>>>>> wrote: >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> On Sep 17, 2016, at 1:13 AM, Trond Myklebust wrote: >>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> According to RFC5661, if any of the SEQUENCE status bits >>>>>>>>>>>>> SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED, >>>>>>>>>>>>> SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED, >>>>>>>>>>>>> SEQ4_STATUS_ADMIN_STATE_REVOKED, >>>>>>>>>>>>> or SEQ4_STATUS_RECALLABLE_STATE_REVOKED are set, then we need >>>>>>>>>>>>> to use >>>>>>>>>>>>> TEST_STATEID to figure out which stateids have been revoked, >>>>>>>>>>>>> so >>>>>>>>>>>>> we >>>>>>>>>>>>> can acknowledge the loss of state using FREE_STATEID. >>>>>>>>>>>>> >>>>>>>>>>>>> While we already do this for open and lock state, we have not >>>>>>>>>>>>> been doing >>>>>>>>>>>>> so for all the delegations. >>>>>>>>>>>>> >>>>>>>>>>>>> v2: nfs_v4_2_minor_ops needs to set .test_and_free_expired >>>>>>>>>>>>> too >>>>>>>>>>>>> v3: Now with added lock revoke fixes and >>>>>>>>>>>>> close/delegreturn/locku fixes >>>>>>>>>>>>> v4: Close a bunch of corner cases >>>>>>>>>>>>> v5: Report revoked delegations as invalid in >>>>>>>>>>>>> nfs_have_delegation() >>>>>>>>>>>>> Fix an infinite loop in nfs_reap_expired_delegations. >>>>>>>>>>>>> Fixes for other looping behaviour >>>>>>>>>>>> >>>>>>>>>>>> This time around the loop seems to be more tight, >>>>>>>>>>>> in userspace process: >>>>>>>>>>>> >>>>>>>>>>>> [ 9197.256571] --> nfs41_call_sync_prepare data->seq_server >>>>>>>>>>>> ffff8800a73ce000 >>>>>>>>>>>> [ 9197.256572] --> nfs41_setup_sequence >>>>>>>>>>>> [ 9197.256573] --> nfs4_alloc_slot used_slots=0000 >>>>>>>>>>>> highest_used=4294967295 max_slots=31 >>>>>>>>>>>> [ 9197.256574] <-- nfs4_alloc_slot used_slots=0001 >>>>>>>>>>>> highest_used=0 >>>>>>>>>>>> slotid=0 >>>>>>>>>>>> [ 9197.256574] <-- nfs41_setup_sequence slotid=0 seqid=14013800 >>>>>>>>>>>> [ 9197.256582] encode_sequence: sessionid=1474126170:1:2:0 >>>>>>>>>>>> seqid=14013800 slotid=0 max_slotid=0 cache_this=1 >>>>>>>>>>>> [ 9197.256755] --> nfs4_alloc_slot used_slots=0001 >>>>>>>>>>>> highest_used=0 >>>>>>>>>>>> max_slots=31 >>>>>>>>>>>> [ 9197.256756] <-- nfs4_alloc_slot used_slots=0003 >>>>>>>>>>>> highest_used=1 >>>>>>>>>>>> slotid=1 >>>>>>>>>>>> [ 9197.256757] nfs4_free_slot: slotid 1 highest_used_slotid 0 >>>>>>>>>>>> [ 9197.256758] nfs41_sequence_process: Error 0 free the slot >>>>>>>>>>>> [ 9197.256760] nfs4_free_slot: slotid 0 highest_used_slotid >>>>>>>>>>>> 4294967295 >>>>>>>>>>>> [ 9197.256779] --> nfs_put_client({2}) >>>>>>>>>>> >>>>>>>>>>> What operation is the userspace process hanging on? Do you have a >>>>>>>>>>> stack trace for it? >>>>>>>>>> >>>>>>>>>> seems to be open_create->truncate->ssetattr coming from: >>>>>>>>>> cp /bin/sleep /mnt/nfs2/racer/12 >>>>>>>>>> >>>>>>>>>> (gdb) bt >>>>>>>>>> #0 nfs41_setup_sequence (session=0xffff88005a853800, >>>>>>>>>> args=0xffff8800a7253b80, >>>>>>>>>> res=0xffff8800a7253b48, task=0xffff8800b0eb0f00) >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:876 >>>>>>>>>> #1 0xffffffff813a751c in nfs41_call_sync_prepare (task=>>>>>>>>> out>, >>>>>>>>>> calldata=0xffff8800a7253b80) >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:966 >>>>>>>>>> #2 0xffffffff8185c639 in rpc_prepare_task (task=) >>>>>>>>>> at /home/green/bk/linux-test/net/sunrpc/sched.c:683 >>>>>>>>>> #3 0xffffffff8185f12b in __rpc_execute (task=0xffff88005a853800) >>>>>>>>>> at /home/green/bk/linux-test/net/sunrpc/sched.c:775 >>>>>>>>>> #4 0xffffffff818617b4 in rpc_execute (task=0xffff88005a853800) >>>>>>>>>> at /home/green/bk/linux-test/net/sunrpc/sched.c:843 >>>>>>>>>> #5 0xffffffff818539b9 in rpc_run_task >>>>>>>>>> (task_setup_data=0xffff8800a7253a50) >>>>>>>>>> at /home/green/bk/linux-test/net/sunrpc/clnt.c:1052 >>>>>>>>>> #6 0xffffffff813a75e3 in nfs4_call_sync_sequence (clnt=>>>>>>>>> out>, >>>>>>>>>> server=, msg=, args=>>>>>>>>> out>, >>>>>>>>>> res=) at /home/green/bk/linux- >>>>>>>>>> test/fs/nfs/nfs4proc.c:1051 >>>>>>>>>> #7 0xffffffff813b4645 in nfs4_call_sync (cache_reply=>>>>>>>>> out>, >>>>>>>>>> res=, args=, msg=>>>>>>>>> out>, >>>>>>>>>> server=, clnt=) >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:1069 >>>>>>>>>> #8 _nfs4_do_setattr (state=, cred=, >>>>>>>>>> res=, arg=, inode=>>>>>>>>> out>) >>>>>>>>>> ---Type to continue, or q to quit--- >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:2916 >>>>>>>>>> #9 nfs4_do_setattr (inode=0xffff880079b152a8, cred=>>>>>>>>> out>, >>>>>>>>>> fattr=, sattr=, >>>>>>>>>> state=0xffff880060588e00, >>>>>>>>>> ilabel=, olabel=0x0 ) >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:2955 >>>>>>>>>> #10 0xffffffff813b4a16 in nfs4_proc_setattr (dentry=>>>>>>>>> out>, >>>>>>>>>> fattr=0xffff8800a7253b80, sattr=0xffff8800a7253b48) >>>>>>>>>> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:3684 >>>>>>>>>> #11 0xffffffff8138f1cb in nfs_setattr (dentry=0xffff8800740c1000, >>>>>>>>>> >>>>>>>>> >>>>>>>>> Cool! Does the following help? >>>>>>>> >>>>>>>> Grrr... There is another bug there? >>>>>>> >>>>>>> Is this in addition to the previous patch or instead of? >>>>>> >>>>>> It can apply on top of it. The two patches fix different bugs. >>>>> >>>>> Ok, it does not seem to have helped: >>>> >>>> Of the 10 nodes where I run this, I had some more failures overnight, >>>> some of them different. >>>> >>>> The other one I got was (I have 3 nodes in this state, >>>> only one in the one reported yesterday.): >>> >>> >>> I?ve been trying desperately to reproduce the issue, but I can?t make knfsd return or revoke the delegations with the ?racer? script that you pointed me to (I?m using the stock knfsd from Linux 4.7.3). Is there anything else special about your setup that I should know about? >> >> >> There is not really anything special here. >> I have a full debug enabled kernel (I checking latest Linus' master before >> applying your whole patchset, but it reproduces the same on earlier ones too). >> >> In my particular case I run in VMs with overcommitted CPUs (that really helps to >> promote races) and with HT enabled (that serves the same purpose). > > Interesting. I can try that. > >> >> I also boot my VMs from nfsroot, but I doubt it plays any role here. >> > > Do they all have different hostnames (and/or are they set up with unique nfs.nfs4_unique_id module parameters) , or is there a chance that they might be using the same client-id and thus be stepping on each others state? They have uniqueue hostnames now, but that is not important for the reproducer, and it's a relatively recent change. If you remember my test script, I start a local ext4 on every client, then export that with nfs and then mount it twice on the same node locally, and then run 3 copies of racer - once per mountpoint. >> Then I just ssh into the VMs and run the script and that is all. >> >> I can show you my kernel config if you think it would help and I can also >> provide you with access to my testbed if you think you want to take a closer >> look at my setup. > > Let me see if I can come up with some extra debugging dprintk()s that might help understand that loop. The current set are apparently not sufficient to clue me in (sorry.. :-( ) I can attach a gdb to fetch some more data with no debug printks and I have several nodes sitting in this state now. Just tell me what it is you want me to look for?