Red Hat Bugzilla – Bug 161874
lvm suffers from "chicken-and-egg" problem in a startup
Last modified: 2007-11-30 17:11:09 EST
Description of problem:
When booting a system where /var is mounted from a separate file system,
which you may want to do if your / is mounted read-only as one of many
possibilities, you will see:
/var/lock/lvm: mkdir failed: Read-only file system
The problem is that at this moment /var indeed is yet read-only but lvm
does not care. Well, /var could be located conceivably on some LVM volume.
Interestingly enough this message is not showing up with lvm2-2.01.12-1.0
from the current rawhide but this does not say anything that /var/lock/lvm
will be created and I do not know if existing volume groups would be
found if such indeed would exist. All of this despite that
'--ignorelockingfailure' flag was given.
Removing lvm2, even if it not really used and despite of
"if [ -x /sbin/lvm.static ]; then ..." in rc.sysconfing, is not really an
option. 'yum remove lvm2' comes back with a proposition to basically
remove the whole system. :-) The only possible hack would be to
rename /sbin/lvm.static which would have to be traced through every update.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
The easiest way to reproduce the condition is to do
mount -o ro -t tmpfs none /var/lock
/sbin/lvm.static vgchange -a y --ignorelockingfailure
On FC4 this responds with "/var/lock/lvm: mkdir failed: Read-only file system".
I guess that the easiest thing would be to "2>/dev/null" in a corresponding
place of rc.sysinit but I have no idea if this is really a correct course of
Having /var be a separate partition is by no means a rare configuration.
The error message should be suppressed nowadays. (Please reopen if there are