[mod.computers.vax] Miscellaneous VMS 4.5 bugs reported from Australia

EVERHART%ARISIA@rca.com.UUCP (02/19/87)

I received a phone call from Paul Rapacholi who is working in
Wollengong, Australia (tel. 61-42-745011 x 5444 for the REAL
adventurous) and reports a number of bugs in VMS 4.5, one of which
can silently trash thousands of files on your disks.
	The disk corruption problem occurs on index file extension on
cluster mounted disks and has hit Paul a couple times so far. It appears
that when two processors are creating files on a volume at the same
time, the index file gets two extensions, the second of which goes
up the creek whenever a cluster member reboots. Somehow it seems
one of the processors uses the wrong extension or loses track of
what extension to use and when a cluster node crashes or is rebooted
and the cluster resynchronizes, the disk shows huge amounts of
garbage. Typically this starts at some file ID which is the first
file of an index file extension. The index file extension is totally
corrupted, and you don't see the damage until there's a reboot. You
can get error messages like "unsupported file structure" then.
	According to a DEC Australia guy, there was a major change in
the XQP to lock manager change in VMS 4.5 which may have something to
do with the appearance of the problem for the 4.5 version.
	A workaround is to always INIT a disk with a big index file (and
use VMS Backup a lot, to tape.)

	The other bugs he reports are less serious...
	For example, if you have an account with a quota of 1000 blocks
or so and write into it with a process with EXQUOTA, you get device
full errors after about 25000 blocks (even though the device is NOT full).
	Another one was that if your account password is "password" then
the access via
  nodename"account password"::
fails, where
  nodename"account PASSWORD"::
succeeds. VMS reports "invalid node" in the first case... could be some
code that checks against "password" since VMS replaces the real password
with that string in some reports. Paul hasn't checked to see if a password
of "..." also fails (RSX uses that as its default report)...
	He asked me to post this info to warn others of the dangers...
		Glenn Everhart