[systemd-devel] loginctl hangs when run at boot time

[Jeff Solomon]

Thanks Lennart, I hear what you're saying but I'm still confused.

I can run any commands in the "User Data" section. And when I run
"systemctl" immediately prior to attempting to run loginctl (which hangs),
I see the following (snipped):

    dbus.service                                                  loaded
active     running         D-Bus System Message Bus
    dbus.socket                                                  loaded
active     running         D-Bus System Message Bus Socket
    systemd-logind.service                                   loaded
active     running         Login Service

which makes me think that even though cloud-init and these two services all
run after basic.target, the two needed to loginctl to run successfully are
already running.

Additionally, the following command works:

    systemd-run loginctl ...

I understand what you're saying about dead-lock, but when I run
systemd-run, all I'm really doing is changing the execution environment,
not the flow of my entire script, and loginctl works in that case.

I guess I still don't understand how your explanation makes sense given the
behavior I observe.

On Thu, Apr 5, 2018 at 1:42 AM, Lennart Poettering <lennart at poettering.net>
wrote:

> On Mi, 04.04.18 23:19, Jeff Solomon (jsolomon8080 at gmail.com) wrote:
>
> > Hi,
> >
> > On the CentOS7 AWS with the following systemd installed:
> >
> >     systemd-219-42.el7_4.4.x86_64
> >
> > Any loginctl command that we try to run at boot time (during the AWS
> "user
> > data" section of cloud-init), will hang and then timeout.
> >
> > The same loginctl command run after you ssh into the box and become root,
> > works fine.
> >
> > I can currently get around this problem by running:
> >
> >     systemd-run loginctl ....
> >
> > which implies that something about the boot time execution environment
> > doesn't allow loginctl to run.
> >
> > When I strace the hanging command, I can see that it's waiting on
> > /run/dbus/system_bus_socket.
> >
> > I don't think I have a unit ordering problem because systemd-run is able
> to
> > run the command. It must be the environment.
> >
> > Sorry for what must be a dumb question, but what is different about the
> > execution environment at boot time that is preventing loginctl from
> running?
>
> loginctl is a client tool for systemd-logind, the communicatoin
> between both happens through dbus, hence both are client to
> dbus-daemon. Both systemd-logind and dbus-daemon are run during late
> boot (i.e. are ordered after basic.target). systemd-logind is bus
> activated, and dbus-daemon is socket activated, which means that as
> soon as dbus-daemon's socket unit is up (i.e. dbus.socket is started)
> any client (such as loginctl) connecting to it will implicitly wait
> first for dbus-daemon to be fully started, and then until logind to be
> fully started.
>
> Now, if you invoke loginctl from something that itself delays the boot
> process, then you might create a deadlock: you are waiting for
> dbus.service/systemd-logind.service to start-up but are at the same
> time blocking it from being started.
>
> Lennart
>
> --
> Lennart Poettering, Red Hat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/systemd-devel/attachments/20180405/e1dd3b6a/attachment.html>