How do you mount ZFS drives when dual booting? (Solved. ish.)

[Tater]

Hello,

I have set up a mirrored pair of 4tb drives with ZFS to store data. I run LM 20 Cinnamon and dual boot POP_OS which my daughter uses. ZFS works fine for one or the other but requires commands to import the pool and mount to data sets if you boot into the other OS.

I'm trying to figure out how to make the process of unmounting/exporting and importing/mounting automatic. I have scripts to do these things but they required sudo. I assume that means they would would need to be run in the root part of the boot/shutdown processes to be automatic.

My questions:

1) Am I missing something? I would have thought this use case would be addressed in ZFS but I can't find it explained anywhere. Is it so blindingly obvious that nobody ever had to ask the question? (This possibility seems likely to me.)

2) If my assumptions are correct about having to add the commands (or my scripts) to the startup/shutdown routines, where do I insert the zpool import and zfs mount / zfs unmount and zpool export commands so they are run with proper privileges?

Thank you for any thoughts or advice.

[AndyMH]

Can't comment on most of your question, but

I have scripts to do these things but they required sudo

You can define a policy kit for your scripts. An example here:
viewtopic.php?f=47&t=317804&p=1800030&h ... d#p1800030

[Tater]

Thank you for that information, Andy. I have not had a reason, so far, to dive into policy kit.

My scripts are simply zpool and zfs commands that require sudo. I should think I would be able to drop them into a startup file under root somewhere without need for a script or changing permissions.

All my searches find are tutorials for installation on zfs, which is a whole different kettle of fish.

I'm trying not to roll my counter down below to the next number.

[AndyMH]

Define a polkit for your script that gives it admin privileges and run it with pkexec myscript. Polkits are the systemd replacement for sudo. It is not that difficult.

If you want to stick with sudo then you will have to modify your sudoers file.

[Tater]

Thanks again, Andy. I will start down the polkit rabbit hole and see what's for tea.

[AndyMH]

It's not that difficult, once you grasp the basic principles.

Create a text file in /usr/share/polkit-1/actions/, e.g. org.you.myscript.policy. You will need to be root to do this and I use xed (default cinnamon text editor).
Copy/paste the following into it:

Code: Select all

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE policyconfig PUBLIC "-//freedesktop//DTD PolicyKit Policy Configuration 1.0//EN"
 "http://www.freedesktop.org/standards/PolicyKit/1/policyconfig.dtd">
<policyconfig>

    <vendor>My zfs Project</vendor>
    <vendor_url>http://myzfsproject.org</vendor_url>
    <icon_name>zfsproject</icon_name>
    <action id="org.you.myscript">
 
        <description gettext-domain="myscript">Run myscript as root</description>
        <message gettext-domain="myscript">Authentication is required to run myscript as root</message>
        <defaults>
            <allow_any>yes</allow_any>                                                             
            <allow_inactive>yes</allow_inactive>                                                   
            <allow_active>yes</allow_active>                                                       
        </defaults>
        <annotate key="org.freedesktop.policykit.exec.path">/usr/bin/myscript</annotate>
        <annotate key="org.freedesktop.policykit.exec.allow_gui">false</annotate>
    </action>
</policyconfig>

Most of the entries are descriptive, you can change them to what makes sense to you, the important ones are:

Code: Select all

    <action id="org.you.myscript">

still descriptive, suggest you use your username, e.g. fred, and the script name

And

Code: Select all

            <allow_any>yes</allow_any>                                                            
            <allow_inactive>yes</allow_inactive>                                                  
            <allow_active>yes</allow_active>                                                       

Normally instead of yes it would say auth_admin which would mean ask for a password, yes means run without asking.

Code: Select all

        <annotate key="org.freedesktop.policykit.exec.path">/usr/bin/myscript</annotate>
        <annotate key="org.freedesktop.policykit.exec.allow_gui">false</annotate>

You must specify the full pathname to your script and the script must be executable. Don't think it makes much difference if you say true or false to allow gui, but I'm assuming a script so no gui.

Once done, test with pkexec myscript - runs without a pwd = success, prompt for a pwd = you did something wrong. I did find when developing foxclone* (which is why I had to learn about polkits) that in spite of running foxclone as root, some commands it ran, specifically mount, did not behave. I ended up defining a polkit for every terminal command that foxclone uses (all in one file).

I found this was a useful article:
https://wiki.archlinux.org/title/Polkit ... %20policy.
and there is lots of other stuff on the web. There are also some terminal commands like pkaction that can give you more info, man pkaction for details.

I did try modifying sudoers initially, but found polkits better.

Any questions, ask, but I'm only a couple of steps ahead of you in understanding

* linux image backup utility.

[Tater]

Andy,

That is very helpful. I will start messing around with it.

I am very much a fan of foxclone, I have used it a few times and it sits on my Ventoy usb drive in case of need. Thank you for making it.

[AndyMH]

Good luck, if it helps, as a further example, I can send you the foxclone polkit - it has multiple entries in it and is a 38kb file.

[Tater]

To follow up on this thread, I did "solve" the mounting issue.

The issue was that the system showed the mount point for the pool as being occupied upon boot. The error was "mount point not empty". I am sure there is a proper and elegant way to address this, but what I found was that there is an overlay setting which will ignore the the "mount point not empty" result and mount the ZFS pool anyway.

My ZFS pool is named pool1 in this example.

Check whether overlay is enabled:

Code: Select all

zfs get overlay pool1
NAME     PROPERTY  VALUE	SOURCE
pool1  overlay	off		default

turn on the overlay:

Code: Select all

zfs set overlay=on pool1

Check:

Code: Select all

zfs get overlay pool1
NAME      PROPERTY  VALUE    SOURCE
pool1  overlay   on       local

Reboot and the pool should mount and show your datasets. I assume there might be some horrible possible side effect of doing this in this way, but I haven't had any issues so far.

Hopefully this will help someone in the future.

[Larry78723]

Please open your original post, click on the pencil, and add (Solved) to the Subject. It may help someone in the future.

[Tater]

Please open your original post, click on the pencil, and add (Solved) to the Subject. It may help someone in the future.

I'm not absolutely sure it's the best answer, but you are right.