buchanan@cs.ubc.ca (John (juancho) buchanan) (01/16/91)
In an effort to provide a reasonably reliable environment I had set up our user file system on an external disk. This disk was mounted on a host referenced by a host alias. Under 3.2.1 and 3.1 I was able to self mount the nfs filesystem even on the file server, this has the advantage that the /etc/fstab file on all my machines is then identical. I suppose that there is a slight performance hit taken when a file server accesses her own files but this is something I am quite happy with. The greatest advantage of this method is that in order to relocate the disk for whatever reason all that has to be done is to umount the filesystem on all the clients, move the disk, change the host alias on the yp server, and mount the filesystem on all the clients. I hope that that answers the question, why would you want to self mount an nfs filesystem? Now the real problem is that this is broken under 3.3.1 When a write is attempted to a self mounted nfs filesytem the system hangs so bad that only a belly button reset has any effect. The SYSLOG file contains these messages (which may or may not be related) Jan 15 14:23:01 chong mountd[193]: svcudp_create - cannot getsockname: Bad file number Jan 15 14:23:01 chong mountd[193]: couldn't create SGI's mount UDP transport Jan 15 14:23:01 chong mountd[193]: svctcp_create - cannot getsockname or listen: Socket operation on non-socket Jan 15 14:23:01 chong mountd[193]: couldn't create SGI's mount TCP transport I have a call into the 1 800 number but I thought that this was a serious enough problem to merit a post to the net. ========================================================================= | |===============================| | John Buchanan (juancho) | buchanan@cs.ubc.ca | | Imager Manager |===============================| | Imager | (604) 228-2218 | | Department of Computer Science |===============================| | University of British Columbia | Standard disclaimer | | Vancouver, BC, Canada | included in this | | | box, right here. | ========================================================================= | This from ;login: (The USENIX Association Newsletter), Sep/Oct 1990, | | in a report on a Usenix Standards BOF session: | | | | An overheard conversation: | | | | -"Mach was the greatest intellectual fraud in the last ten years." | | | | -"What about X?" | | | | -"I said `intellectual' " | | | =========================================================================
dana@tread.sgi.com (Dana Treadwell) (01/17/91)
Self-mounting (a.k.a., loopback mounting) is not supported over NFS due to deadlocks (immediate on MP systems, and upon I/O on SP systems). You might look to automount(1M) to solve your /etc/fstab maintenance woes. The mountd complaints in syslog point to a problem with YP/NIS (most likely, an out of date /etc/rpc database or domain name confusion). Dana dana@sgi.com