An error occured handling 'dm_crypt-0': TypeError - join() argument must be str or bytes, not 'NoneType'

[Andreas Hasenack]

Crash file attached. This was an attempt at installing on /dev/sdb, which is an external sata drive. I selected lvm with encryption, but no customization.

[Paride Legovini (paride)]

This can be reproduced with subiquity 20.03.2 (currently candidate). No need to have a separate disk: it is sufficient to add a partition and configure it as an encrypted LVM volume.

I'm attaching the full /var/log of the installer system.

[Paride Legovini (paride)]

From curtin's install.log:

Running command ['blockdev', '--rereadpt', '/dev/vda'] with allowed return codes [0] (capture=True)
Error rescanning devices, possibly known issue LP: #1489521
cmd: ['blockdev', '--rereadpt', '/dev/vda']
stdout:''
stderr:blockdev: ioctl error on BLKRRPART: Device or resource busy
exit_code:1

[Łukasz Zemczak (sil2100)]

Confirmed on a KVM, the test case is really trivial. This is a release critical bug.

[Ryan Harper (raharper)]

Can you provide the virtual machine configuration, either libvirt xml or qemu command line?

[Łukasz Zemczak (sil2100)]

I was running subiquity via this, as I run most of my quick tests:
`kvm -drive if=virtio,file=/path/to/hdd.qcow2 -m 3072 -boot d -smp 2 --cdrom focal-live-server-amd64.iso`

[Ryan Harper (raharper)]

- id: dm_crypt-0
        key: ubuntu
        preserve: false
        type: dm_crypt
        volume: partition-2

The block schema for dm_crypt types requires 'dm_name' field, this is not present. The error message is not great, so I;ll sort out curtin.

Subiquity has a bug here in not emitting a dm_name in the type: dm_crypt config dictionary.

[Brian Murray (brian-murray)]

add_dm_crypt() in subiquity/model/filesystems.py isn't passing dm_name and it looks like that defaults to None.

[Łukasz Zemczak (sil2100)]

So it's as Brian commented, but it seems that this part of the code is the same since May 2019, and we have confirmed that encrypted LVM worked fine with 19.10. What I do find weird is that even with an empty dm_name things should still work in curtin, no? Per:

        dm_name = vol.get('dm_name')
        if not dm_name:
            dm_name = vol.get('id')
        volume_path = os.path.join("/dev", "mapper", dm_name)

from curtin/commands/block_meta.py. The id attribute seems to be set correctly, so I wouldn't expect it to suddenly be None? Though the logs seem to indicate that.

[Łukasz Zemczak (sil2100)]

Ok, wrong thing I checked, so I see the issue. So on the 2020-03-10 with commit bf03e4f78 the "dmcrypt_dev = os.path.join("/dev", "mapper", dm_name)" line has been added to dm_crypt_handler() which is now crashing. As per the snippet above, in other places we are defaulting to using 'id' when 'dm_name' is not available, so I'd recommend doing the same here. Is there a reason for not following the same procedures in both places?

[Łukasz Zemczak (sil2100)]

Ok, attached MP for a quick fix that I hope might help? Let me try hacking that live to see if it helps.

[Server Team CI bot (server-team-bot)]

This bug is fixed with commit 2f226905 to curtin on branch master.
To view that commit see the following URL:
https://git.launchpad.net/curtin/commit/?id=2f226905

[Paride Legovini (paride)]

The fix works works perfectly, thanks!

[Ryan Harper (raharper)]

This bug is believed to be fixed in curtin in version 20.1. If this is still a problem for you, please make a comment and set the state back to New

Thank you.