On Tue, Dec 16, 2008 at 06:42:01PM +0100, Frank Batschulat (Home) wrote: > On Tue, 16 Dec 2008 18:28:02 +0100, Nicolas Williams <Nicolas.Williams at > sun.com> wrote: > > > On Tue, Dec 16, 2008 at 02:22:20PM +0100, Pavel Filipensky wrote: > >> this is a fix for "6778894 umountall -l still not really local yet". > >> webrev: http://cr.opensolaris.org/~pavelf/6778894/ > >> > >> Background: My recent fix for "6675447 NFSv4 client hangs > >> on shutdown if server is down beforehand" is built on "umountall -l", > >> but as noted above umountall -l still not really local yet. > >> Unless umountall -l is fixed, the NFSv4 client can still hang > >> on shutdown if server is down beforehand. > > > > So df -n doesn't hang when NFS servers are non-responsive? > > outch! Thanks ! > > Yes, of course, df uses statvfs() and statvfs() will go OTW for NFS.
Does it? truss doesn't show it doing that when the -n option is used, it only fstat64()s stdout. If Pavel has tested this fix correctly then there may not be a hang at all. > probably better to parse the 'mount -p' output instead. Now that I've trussed df -n I am certain that it won't hang. Of course, it's important to make sure that df -n will never be modified in such a way that it could hang, but I think such a change is extremely unlikely. A block comment in df wouldn't hurt though. Nico --
