Installation fails on s390x due to unexpected input for Archive Mirror

[Frank Heimes]

I got aware by mail that an installation on s390x fails and ended up in a crash during disk configuration. The crash files are attached, but I also asked for the entire /var/log content.

I've found in the crash file that fdasd got accidentally executed against a device-mapper device (/dev/dm-0) rather than a DASD ECKD device:

"
Running command ['fdasd', '--table', '/dev/dm-0'] with allowed return codes [0] (capture=False)

 fdasd error: Unsupported disk type
 Disk geometry does not match a DASD device of type 3390.
"
I found the same in the second crash file.

Not sure yet why a wrong device is used.
I think the crash files don't allow much more analysis at that time (at least not for me) and we may need to wait for the installer logs (from /var/log)...

[Frank Heimes (fheimes)]

/var/log from ext4 installation on 20.04

[Frank Heimes (fheimes)]

Interestingly there are no DASDs attached to this DPM LPAR (as the attached screenshots show).
Wondering why fdasd is called ...
(this is more curtin related, hence adding curtin to the bug)

[Frank Heimes (fheimes)]

output of 'probert --storage'

[Michael Hudson-Doyle (mwhudson)]

We try to call fdasd --table in a bunch of situations where it's not expected to work (on all architectures!), that's not what is causing the failure.

This is user error, they entered a URL to a an ISO in the mirror screen:

2021-09-23 14:51:40,822 DEBUG subiquity.client.controllers.mirror:43 MirrorController.done next_screen mirror=https://cdimage.ubuntu.com/releases/20.04/release/ubuntu-20.04.3-live-server-s390x.iso

We should think about the wording of the screen so users are less likely to do this and I've been meaning for a long time to do some checks of the value the user enters here.

[Frank Heimes (fheimes)]

Thx for identifying the root cause @mwhudson
(I fall for this 'speculative' execution too often, like with lsusb, too.)

+1 on adding some checks (to make this more robust against wrong entries)

Also, adjusting(correcting the bug title...