Coda Distributed File System, version 5.3.1
There is a new snapshot of the Coda development tree, which has been
built for the following platforms:
FreeBSD-2.2.5
FreeBSD-3.1
Linux (RedHat-5.2)
NetBSD-1.3
(all i386)
There are lots of bugs fixed, but also many new features, so it needs
more testing to find out if there are still problems. Several bugs that
were causing datalosses have been fixed, and we are hoping to upgrade
our fileservers to this new version next week. We really would appreciate
if the code has seen some serious pounding before that ;)
There have been several changes to the network protocol, which make it
impossible for non-5.3.1 clients to communicate with 5.3.1 servers (and
the other way around). This change is very inconvenient, but was necessary.
To emphasize this a bit more, Coda 5.3.1 clients and servers cannot
communicate with older versions of Coda due to network protocol changes.
All clients and servers need to be upgraded around the same time. New
clients cannot start when talking to old servers. New servers will block
old clients from reintegrating any logged changes.
Summary of the differences since 5.2.7
Bugfixes
- Clients that didn't complete connection setup were not garbage
collected by the server.
- Truncation of an uncached file while (write-)disconnected created a CML
entry without associated container file.
- Avoiding NULL-pointer dereferences when `unexpectedly' losing clients
or servers.
- Directory refcounting went wrong when a repair session failed.
- Some big/little endian-issues, and several other LinuxPPC fixes.
- pdbtool messed up the maxids field, and removing a group didn't clean
up the group-members correctly.
- Some fixes related to the fetch continuation code.
- Testing on a simulated ad-hoc network turned up several problems in
the RPC2/SFTP code.
- Improved stability of weak-reintegration and resolutions.
- RPC2 suffered from an overflow after 49 days... It didn't crash, but
became dreadfully slow.
- Partial reintegration could switch between servers, and as a result
only store part of a file.
- CML entries were sometimes discarded after repairing a local-global
conflict.
- Volumes would get in a deadlock state when a thread was waiting for
a resolve to finish.
New Features
- Writeback caching. Not well tested yet, still waiting for some more
changes.
- Filled in many of the blanks in the ViceGetStatistics rpc, the
source tarball also has a modified cmon program to pull the data
from the servers to create graphs of the server/resource usage.
- cfs forcereintegrate, blocks until changes have been sent to the server.
- cfs listvolume, gives additional information about # of CML entries.
- Added -realm and -keytab to kauth2.
- More pdbtool commands: maxids, export, import. Also enhanced useability
a little bit by accepting usernames as well as userids.
Information about ongoing porting projects
Windows95
Clients are currently still broken, last working version seems to be
something like 5.2.3pre1, working on getting it working..
WindowsNT
It will likely to take quite some time before the current kernel-code
will lead to a stable client. Anybody with experience in NT-kernel
programming interested in helping things along?
Solaris
The server is up and running. The client is still waiting for some kernel
code, things are looking promising.
Ofcourse, this all wouldn't have been here without the help of
Troy Benjegerdes, Peter Braam, Larry Greenfield, Yui-wah Lee (Clement),
Phil Nelson, Greg Troxel, Shafeeq Sinnamohideen, Nickolai Zeldovich
(email me if I missed anyone).
Compatibility with previous versions:
- network protocol: Not compatible
- Coda clients: clients probably need to be reinitialized
- Coda servers: RVM/vicepa pdb databases
no changes 5.2.3+
Have fun,
Jan
Received on 1999-09-01 18:52:49
