[804] in SIPB-AFS-requests
idec() problem
daemon@ATHENA.MIT.EDU (Richard Basch)
Sat Oct 3 22:09:38 1992
Date: Sat, 3 Oct 92 22:08:14 -0400
To: op@Athena.MIT.EDU, vice-squad@Athena.MIT.EDU, afsreq@Athena.MIT.EDU,
Cc: wesommer@Athena.MIT.EDU, afshelp@transarc.com, bug-afs@Athena.MIT.EDU
From: "Richard Basch" <basch@MIT.EDU>
I have brought to everyone's attention, in the past, the existence of a
potential problem where the inode reference counts may not be maintained
properly. This was my theory for explaining the occassional message in
the SalvageLog:
vnode ###: corresponding inode ### missing; deleted.
This has also been seen by the fileserver, during a DeleteTarget()
operation - where a file has already disappeared at the time it is
trying to remove the file, thus causing the fileserver to coredump. At
first, I thought there might be a race between a couple fileserver
threads. However, later evidence suggested that this is NOT the case.
There have been other times when a volume cannot be dumped or cloned
because of this problem. The problem has now struck some of the Athena
source volumes (the BSD sources, specifically), and since this has NOT
been modified recently, all hints that this may be related to the
fileserver can now be dispelled.
The idec() problem can occur solely with volume cloning!!!
The problem has been observed primarily on DECstation servers, but there
is currently no evidence that this should be the only architecture
affected.
Transarc:
We are still running 3.1b, but with many of the patches that exist in
3.2; the remaining components of 3.2 do not seem to suggest that this
bug has been fixed.
-R
