[Mnj] AIPS problems with certain Linux kernels

[Eric Greisen]

Our system manager writes:
---------------------------

    I'm fairly positive we've found the problem and solution to the
AIPS bug where open() would erroneously return ENOENT for existing
files.

     The problem appeared in the RHEL 2.6.9-55 kernel which inherited
from the 2.6.19 mainline kernel.

     The problem appears to be a race condition in the automounter
when a filesystem umounts via the timeout and then is immediately
remounted due to the open().   The includes loopback mounts of a
machines own data area.

     It's been identified and patched in 2.6.23.1 (the latest mainline
kernel) and should appear in the RHEL 2.6.9-61 kernel to be released
soon (TM).

      For users that absolutely positively need it fixed now you can
tell them to keep the filesystem mounted by cd'ing to it in a shell.

      The only real reason to have the RHEL 2.6.9-55 kernel is for
Intel ICH-8 SATA controller support.

------------------------------------

Please be aware of this when updating your operating systems.  I have
defended AIPS 31DEC07 against the worst (most common) occurence of
this problem, but it would still arise on other files such as message
files, image catalogs, header files, etc.  It is a statistical thing
that does not happen often, but enough to stop most Y2K tests from
completing (without the new defense), for example.

Eric Greisen