Coda File System

Finally, 5.3.17...

From: Jan Harkes <>
Date: Tue, 6 Nov 2001 17:28:15 -0500
I finally have a feeling that the latest 'stable' snapshot isn't going
to crash and burn within 10 minutes after installation. (which the
5.3.16 release did that was up for ftp for about 48 hours before I
pulled it).

So for your enjoyment, new stuff is on

The latest versions are,

    coda-5.3.17 (binaries & sources)
    rpc2-1.13   (binaries & sources)
    lwp-1.9	(source only, just some minor compilation fixes)


Here is a overview of changes/bugfixes/improvements,

Coda client changes,
- venus-setup created NOT_REALLY_CODA in / instead of in /coda (Ivan Popov).
- Avoid killing clients when receiving truncated directory data.
- Block when accessing an uncached object during resolution instead of
  failing. This is especially useful during weak-reintegration.
- Added -local flag to cfs lv to avoid client-server rpc2 traffic.
- Fetch data as well as status in write-disconnected mode, the dirty
  status-only object would prevent subsequent data-fetches until successful
  reintegration completed.
- Added venus.conf option 'validateattrs' to limit piggybacked fids during
  validate attrs to help reducing IP layer fragmentation.
- Another new venus.conf option 'serverprobe' to increase server probe
  intervals (keeps firewalls from dropping masqueraded connections).
- venus.conf.ex file now defaults to turning on both masquerading and using
  private mmap's for RVM.
- Disallow unlinking special names with 'cfs rmm'
- Optimized local-global fid replacement.
- Removed /dev/cfs0 from RPM package to allow unpacking the RPM on devfs
  enabled systems, we don't fix up the devfs autoloading stuff yet.
- Moved /usr/bin/replay to /usr/bin/coda_replay to avoid name conflicts.
- Added getvolinfo program to be useful when debugging volume mounting

Coda server changes,
- Added the missing fflush back in the server logging.
- Fixed createvol_rep to disable resolution logging when creating singly
  replicated volumes.
- Fixed possible errors during runt-resolution.
- Added assert to the server when not all hostnames in a VSG can be resolved.
- Avoid servers getting stuck in file-creation when there are pre-existing
  files in /vicepa.
- Always pre-allocate fids with the maximum possible stride, will become
  useful when we allowing modifications to a volume's replication group.
  However, any replicated volume over a mix of old and new servers has a
  slightly higher probability on getting unwanted fid-collissions.
- Reduced server memory usage during reintegration.

RPC2 changes,
- Increased RPC2 rtt lower bound to 300ms to improve reliability over slow
- Removed Linux specific hack that allowed servers to appear to respond faster
  than they really did. This affected both RTT estimates and bandwidth based
  decisions in the clients.
- Fixed a bad pointer dereference in SFTP that caused occasional server
- Fixed masquerading SFTP acks in when server does backfetches from client
  (many thanks to Greg Troxel for hunting down this one).
- 'Quick-acking' slow server operations with RPC2_BUSY to avoid request

Build fixes,
- gcc-3.0 compilation fixes.
- Lots of cygwin/win32 patches (Phil Nelson).
- Fixed the test for readline library version (Phil Nelson).
- Fixed detecting libdb3 with db 1.85 'compatibile' API (Russ Mennie).
- OpenBSD patches (Jeff Bachtel).
Received on 2001-11-06 17:28:37