Best explained by giving reproducing instructions: 1) Do a rawhide install with /home or /srv on an iscsi target 2) Reboot into the fresh installation 3) iscsi service runs before NM, fails to attach to target (this is another bug, but even without this this bug will still happen) 4) netfs runs from upstart, does nothing as no network yet. 5) NetworkManager runs 6) NetworkManager is done bringing up the network and runs netfs, note that netfs no is run from NetworkManager not from upstart and thus has no console/tty 7) netfs tries to fsck network attached block devices such as iscsi, but those are not present due to 3 (if this error was not there, the following could still happen in case of fs damage) 8) netfs detects a fatal fsck errors and tries to start a root shell 9) starting the rootshell fails as no stdin to read the password from 10) netfs reboots the machine 11) machine starts rebooting while it had already continued booting way past NM (I actually briefly get a login screen) as NM was doing its stuff async. So it looks like we need some magic in netfs to not try to run the rootshell but instead log an error and exit when running from NM. netfs *must* exit when this happens and not continue as we do not want to mount the network block devices with fsck errors to avoid further damage.
http://git.fedorahosted.org/git/?p=initscripts.git;a=commitdiff;h=de71796ec6544630765462b2c0c339d45acc314a