Dear Maintainer,

If LUNs are accessed through infiniband they will not be available until the opensm
daemon is properly running resulting in failed login and attachment of LUN's. In
worst case if root is installed on such a LUN the system will fail to boot.

How to fix: Either opensm should start i runlevel S with a lower number than open-iscsi
or open-iscsi should have a dependency to opensm.
----

This mail was virus scanned and spam checked before delivery.
This mail is also DKIM signed. See header dkim-signature.

It may not be that simple.

First, if you have a modern init (like systemd), the correct set of
dependencies there can help solve the ordering problem. It is not the
ordering that is the problem in most cases, but about how devices are
discovered. For example, if there are devices that are discovered by IB
after the iSCSI daemon start, in traditional method, they'd be left
over. Modern inits like systemd solve that problem.

As for root installed on a LUN, there's a lot of manual work in initrd,
that you'll have to add for opensm.

open-iscsi needs to be started at early-boot, otherwise filesystems on
iSCSI will not be available for services that use them. Therefore,
opensm should also be started at early-boot, thus reassigning this bug
to opensm, because open-iscsi is started at the right time during the
boot process.

Note that open-iscsi is started so early that /usr might not be
available if sysvinit is used as init system (for systemd it's mounted
in initrd), so the next version of open-iscsi will have its binaries in
/ instead of /usr - this will probably be necessary for opensm as well
if it's moved to early-boot.

Also note that in the future (for Stretch) a native systemd service
needs to be provided for init scripts that run in rcS - so simply
changing the init script of opensm to be started in runlevel S will not
be enough - a systemd service will be needed as well.

Christian

Link to bug report:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=615864

Sorry about that - I was triaging some bugs and made a stupid C&P error.
(But at least I reassigned the right bug. Phew.) Proper bug report:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=782177

Sorry about the noise. :/

Christian