On Wed, 12 May 2004, Jeff Moyer wrote:
> I had a user report an issue which is summed up in the subject.  Basically,
> he disallowed UDP traffic to his NFS server.  Because the autofs code uses
> UDP in the rpc_ping function to determine if a server is alive, it would
> conclude that the server was not alive.  He was, of course, using NFSv4
> over TCP.

Would it be possible to change the paradigm: attempt the mount from one
server, and if the filesystem doesn't appear promptly, rename the mount
point (later attempting to unmount it), and attempt the mount from another
server.  That way, someone else takes complete responsibility for _how_ the
mount is done.  Likely the needed timeouts are about the same for an actual
mount as for rpc_ping.  The reason for renaming is in case the mount
eventually succeeds.  Maybe it would be better to always mount on a
temporary name and then always rename the successful mount point (if any)
to the requested name, so the mount point name doesn't change at random 
times while other kernel threads are trying to use it.  You would use 
"mount --move olddir newdir" (kernel 2.5.x and above only).

An alternative that will never happen: have a "no-op" option to mount,
which determines if the mount could be done (for NFS, this is the rpc_ping
thing) but doesn't actually mount anything.

James F. Carter          Voice 310 825 2897    FAX 310 206 6673
UCLA-Mathnet;  6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA  90095-1555
Email: [EMAIL PROTECTED]    http://www.math.ucla.edu/~jimc (q.v. for PGP key)

_______________________________________________
autofs mailing list
[EMAIL PROTECTED]
http://linux.kernel.org/mailman/listinfo/autofs

Reply via email to