Re: [PATCH - v2] NFSv4: handle EINVAL from EXCHANGE_ID better.

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Apr 03, 2018 at 10:41:56AM +1000, NeilBrown wrote:
> On Tue, Mar 20 2018, J. Bruce Fields wrote:
> 
> > On Fri, Mar 16, 2018 at 10:44:23AM +1100, NeilBrown wrote:
> >> 
> >> nfs4_proc_exchange_id() can return -EINVAL if the server
> >> reported NFS4INVAL (which I have seen in a packet trace),
> >
> > Can you say which server this was?  (One we can fix?)
> 
> It was Linux 2.6.32 (on armv5), and presumably anything before
> Commit: 357f54d6b382 ("NFS fix the setting of exchange id flag")
> which landed in 2.6.38.

The kernel defaulted to keeping 4.1 off at that point, probably for good
reason--that default didn't change for another 4 or 5 years.  I suspect
this is only the first of multiple 4.1 bugs you'd run into in a server
of that vintage.  So the solution is probably just to keep 4.1 off.

--b.

> 
> These servers return NFS4ERR_INVAL when EXCHGID4_FLAG_BIND_PRINC_STATEID
> is set in the exchange_id flags, which Linux has done since
> Commit: 4f0b429df104 ("NFSv4.1: Enable state protection")
> in 3.11.
> 
> Maybe NFS should retry without that flag if it gets EINVAL??  Or it could
> just say that NFSv4.1 isn't supported.  I still don't think EIO is justified.
> 
> Thanks,
> NeilBrown