Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 9 Mar 2021 15:40:21 -0500
From:   Scott Mayhew <smayhew@redhat.com>
To:     "J. Bruce Fields" <bfields@redhat.com>
Cc:     Olga Kornievskaia <olga.kornievskaia@gmail.com>,
        chuck.lever@oracle.com, linux-nfs@vger.kernel.org
Subject: Re: [PATCH 1/1] NFSD: fix error handling in callbacks
Message-ID: <20210309204021.GG2597609@aion.usersys.redhat.com>
References: <20210309144127.57833-1-olga.kornievskaia@gmail.com>
 <YEeWK+gs4c8O7k0u@pick.fieldses.org>
 <20210309194519.GF2597609@aion.usersys.redhat.com>
 <YEfWlkKaIr2lz896@pick.fieldses.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YEfWlkKaIr2lz896@pick.fieldses.org>
Precedence: bulk

On Tue, 09 Mar 2021, J. Bruce Fields wrote:

> On Tue, Mar 09, 2021 at 02:45:19PM -0500, Scott Mayhew wrote:
> > On Tue, 09 Mar 2021, J. Bruce Fields wrote:
> > 
> > > On Tue, Mar 09, 2021 at 09:41:27AM -0500, Olga Kornievskaia wrote:
> > > > From: Olga Kornievskaia <kolga@netapp.com>
> > > > 
> > > > When the server tries to do a callback and a client fails it due to
> > > > authentication problems, we need the server to set callback down
> > > > flag in RENEW so that client can recover.
> > > 
> > > I was looking at this.  It looks to me like this should really be just:
> > > 
> > > 	case 1:
> > > 		if (task->tk_status)
> > > 			nfsd4_mark_cb_down(clp, task->tk_status);
> > > 
> > > If tk_status showed an error, and the ->done method doesn't return 0 to
> > > tell us it something worth retrying, then the callback failed
> > > permanently, so we should mark the callback path down, regardless of the
> > > exact error.
> > 
> > That switch was added because the server was erroneously setting
> > SEQ4_STATUS_CB_PATH_DOWN in the event that a client returned an
> > NFS-level error for a CB_RECALL that the client had already done a
> > FREE_STATEID for.  Removing the switch will cause the server to go back
> > to that behaviour, won't it?
> 
> Oh, thanks for the history.
> 
> My knee-jerk reaction is: that sounds like a recall-specific issue, so
> maybe that logic should be in nfsd4_cb_recall_done?
> 
> I guess I'm OK continuing instead to enumerate transport-layer errors in
> nfsd4_cb_done, but do we know that EACCES makes it a complete list?

No idea.  I'm pretty sure EIO and ETIMEDOUT were the two errors that I
was seeing at the time.  I don't remember the exact test case but I
think I was hammering the server trying to reproduce seq misordered &
false retry errors.  And EACCES is what Olga's seeing when the server
uses the wrong principal for the callback cred.

-Scott
> 
> --b.
> 
> > 
> > -Scott
> > > 
> > > --b.
> > > 
> > > > 
> > > > Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
> > > > ---
> > > >  fs/nfsd/nfs4callback.c | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > > 
> > > > diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
> > > > index 052be5bf9ef5..7325592b456e 100644
> > > > --- a/fs/nfsd/nfs4callback.c
> > > > +++ b/fs/nfsd/nfs4callback.c
> > > > @@ -1189,6 +1189,7 @@ static void nfsd4_cb_done(struct rpc_task *task, void *calldata)
> > > >  		switch (task->tk_status) {
> > > >  		case -EIO:
> > > >  		case -ETIMEDOUT:
> > > > +		case -EACCES:
> > > >  			nfsd4_mark_cb_down(clp, task->tk_status);
> > > >  		}
> > > >  		break;
> > > > -- 
> > > > 2.27.0
> > > > 
> > >