On Thu, Feb 27, 2014 at 10:32:51AM -0800, Chuck Lever wrote:
> Iâm running xfstests to validate functionality of a particular
> NFS client/transport/server set up.
> There are a couple of tests that may not always be appropriate for
> FSTYP=nfs. We werenât able to identify a bugzilla where we can report
> these issues.
Just send the to the xfs list. (and cap lines after 75 chars like for all
> generic/192 tests atime changes. The Linux NFS client (and probably
> others) typically optimizes away atime changes. Thus this test fails
> for NFS mounts on Linux. Iâm wondering if we can treat NFS as always
> a noatime filesystem.
I guess there's two components to this if we want to do it properly:
- make sure nfs always reports noatime in /proc/self/mounts instead
of incorrectly claiming it supports relatime by default
- _notrun 192 if it detects the filesystems reports noatime.
> generic/237 tries setfacl. On NFS mounts, the operation always returns
> âOperation not supported,â but the test is looking for âOperation not
> permitted.â The former is an acceptable pass, IMO. I considered
> disabling this test for NFS, but I would bet some NFS client
> implementations actually do support setfacl().
The NFSv3 client support ACLs perfectly, it's just NFSv4 that's messed up.
237 does a _require_acls before ding thÃ test, which checks for
ENOSYS or EOPNOTSUPP, which in theory should catch NFSv4 in theory, but
if this doesn't happen try to debug it by looking at the _require_acls shell
function, the chacl implementation, the NFSv4 ACL stubs and anything else