Bug#776276: unblock: open-iscsi/2.0.873+git0.3b4b4500-4

[Christian Seiler]

Am 27.01.2015 um 15:39 schrieb Michael Biebl:
> Am 27.01.2015 um 13:07 schrieb Christian Seiler:
>> Am 27.01.2015 um 06:19 schrieb Michael Biebl:
>>> - Drop $remote_fs from the open-isci LSB header and replace it with
>>> $local_fs. That looks wrong anyway.
>>
>> As I said before: that breaks sysvinit systems with /usr on NFS.
> 
> How likely is such a setup? I.e. using /usr-on-NFS + openisci in
> combination. Has anyone actually ever tested this?

No idea. I'm just wary of breaking potentially legitimate setups,
especially this late in the freeze. Currently, your proposed solution
differs from mine by just a single dependency on mountnfs-bootclean.

I'm not saying this shouldn't be reworked at a later point in time (I do
agree that it's quite ugly), but do you have any technical reason for
objecting to that? (mountnfs-bootclean.sh is part of initscripts btw.)

>> [Service]
>> # Replace stop actions, because calling umountiscsi.sh from
>> # within open-iscsi script won't work due to systemctl
>> # redirection (and umountiscsi.sh is never started)
>> # Note that we have to test for /etc/iscsi/iscsi.initramfs,
>> # since the init script also does this test. Use exec to
>> # make sure that umountiscsi.sh's PPID is 1, so we don't run
>> # into the redirect ourselves.
>> ExecStop=
>> ExecStop=/bin/sh -c "[ -f /etc/iscsi/iscsi.initramfs ] || exec
>> /etc/init.d/umountiscsi.sh stop"
>> ExecStop=/etc/init.d/open-iscsi stop
> 
> Why is this check needed? After all, on shutdown we do call
> "/etc/init.d/umountiscsi.sh stop" before "/etc/init.d/open-iscsi stop".
> I.e. wouldn't this break the shutdown case, assuming there is an
> /etc/iscsi/iscsi.initramfs file?

Hmmm, right, update-rc.d already inserts stop links... The interesting
thing is that for some reason the open-iscsi init script calls that
again, but only after a check for iscsi.initramfs. No idea, why.

The reason that check for iscsi.initramfs is there is that the iSCSI
sessions should not be deactivated if rootfs is on iSCSI - but that
isn't really an issue, I just tested it: if something's still active,
umountiscsi.sh will complain a bit, since it can't umount stuff, but is
otherwise harmless, so yeah, that's not needed.

However, exit code will be 1 in that case, so it should be:

ExecStop=-/etc/init.d/umountiscsi.sh stop

Christian