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 <green@xxxxxxxxxxxxxx>
> > > 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 TEST_STATEID to figure out which stateids have been revoked, so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=<optimized
> out>,
> calldata=0xffff8800a7253b80)
> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:966
> #2 0xffffffff8185c639 in rpc_prepare_task (task=<optimized out>)
> 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=<optimized
> out>,
> server=<optimized out>, msg=<optimized out>, args=<optimized
> out>,
> res=<optimized out>) at /home/green/bk/linux-
> test/fs/nfs/nfs4proc.c:1051
> #7 0xffffffff813b4645 in nfs4_call_sync (cache_reply=<optimized
> out>,
> res=<optimized out>, args=<optimized out>, msg=<optimized out>,
> server=<optimized out>, clnt=<optimized out>)
> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:1069
> #8 _nfs4_do_setattr (state=<optimized out>, cred=<optimized out>,
> res=<optimized out>, arg=<optimized out>, inode=<optimized out>)
> ---Type <return> to continue, or q <return> to quit---
> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:2916
> #9 nfs4_do_setattr (inode=0xffff880079b152a8, cred=<optimized out>,
> fattr=<optimized out>, sattr=<optimized out>,
> state=0xffff880060588e00,
> ilabel=<optimized out>, olabel=0x0 <irq_stack_union>)
> at /home/green/bk/linux-test/fs/nfs/nfs4proc.c:2955
> #10 0xffffffff813b4a16 in nfs4_proc_setattr (dentry=<optimized 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?
8<------------------------------------------------------------
>From 98ddf32a99cfe00e9ae108044e2be67522987511 Mon Sep 17 00:00:00 2001
From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
Date: Sat, 17 Sep 2016 15:27:10 -0400
Subject: [PATCH] NFS: Don't assume a stateid represents a delegation in
nfs4_do_handle_exception
If the stateid being passed to the error handler is not a delegation
stateid, we want to mark the locks/open_state it does represent for
recovery.
Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
---
fs/nfs/nfs4proc.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 7cecb1d7a217..acc572c51735 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -397,6 +397,10 @@ static int nfs4_do_handle_exception(struct nfs_server *server,
exception->delay = 0;
exception->recovering = 0;
exception->retry = 0;
+
+ if (stateid == NULL && state != NULL)
+ stateid = &state->stateid;
+
switch(errorcode) {
case 0:
return 0;
@@ -405,7 +409,7 @@ static int nfs4_do_handle_exception(struct nfs_server *server,
case -NFS4ERR_EXPIRED:
case -NFS4ERR_BAD_STATEID:
if (inode != NULL && stateid != NULL) {
- nfs_inode_find_delegation_state_and_recover(inode,
+ nfs_inode_find_state_and_recover(inode,
stateid);
goto wait_on_recovery;
}
--
2.7.4��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
