Browse Prior Art Database

Two-Phase Unmount for Globally Mounted File Systems in a Cluster Environment

IP.com Disclosure Number: IPCOM000117004D
Original Publication Date: 1995-Dec-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 76K

Publishing Venue

IBM

Related People

Curran, RJ: AUTHOR [+5]

Abstract

Disclosed is a scheme that provides standard single-system UNIX* semantics in a loosely-coupled distributed system for unmounting a file system. To appear as a single-system, a distributed system should unmount a file system if and only if all nodes of the distributed system can successfully perform the unmount operation, i.e., the unmount must be atomic. If any node has open files, the unmount should fail and the file system state should remain unchanged. Such a behavior for the mount operation (which is the complementary operation to unmount) is relatively easy to support and therefore not discussed here.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Two-Phase Unmount for Globally Mounted File Systems in a Cluster
Environment

      Disclosed is a scheme that provides standard single-system
UNIX* semantics in a loosely-coupled distributed system for
unmounting a file system.  To appear as a single-system, a
distributed system should unmount a file system if and only if all
nodes of the distributed system can successfully perform the unmount
operation, i.e., the unmount must be atomic.  If any node has open
files, the unmount should fail and the file system state should
remain unchanged.  Such a behavior for the mount operation (which is
the complementary operation to unmount) is relatively easy to support
and therefore not discussed here.

      The disclosed scheme consists of a 2-phase unmount protocol.
In the first phase, new accesses to the unmounting file system are
blocked and the file system is checked for active accesses.  Should
the first phase complete successfully (No active accesses) on a given
node the global unmount protocol is insured that a subsequent unmount
will succeed.

      Should the first phase fail on any node then the protocol must
"back-out" of the global unmount by means of initiating the second
phase in "failure" mode on all nodes that have successfully completed
phase 1.  As part of the "back-out" process, all processes blocked
while waiting for the unmount to complete are made runnable and will
continue usual.

      Should the first phase complete successfully on all unmounting
nodes, the second phase is initiated in "success" mode on all such
nodes.  The "success" mode phase-2 is a variant of the standard
virtual file system unmount operation (in the kernel file system
code) that also wakes any processes blocked while waiting for the
unmount to complete and causes them to return ESTALE.  Implementation
of the disclosed scheme is described in the following paragraphs.

      A distributed global mount/unmount subsystem keeps track of
which cluster n...