Feature #9005: Tails Installer without a splash screen
Make Tails Installer's isohybrid detection code robust
I'm a tiny bit worried about the new isohybrid detection code (commit:b7dcb936a58bb0016981645c9224caee4493112a): I've seen very weird stuff happening when the kernel detects these devices, kinda lacking determinism (sometimes it would be recognized in a way, sometimes in the other) which is not surprising given such devices are really two different things at the same time. Let's not bother for now but keep it in mind if we get related bug reports.
I think I've found a bug:
- dd a Tails ISO to a USB stick
- leave that USB stick plugged
- start Tails Installer
- /dev/sdc1 is automatically selected
- the "Upgrade" button is visible (weird) but disabled (OK)
- the "Reinstall (delete all data)" button is enabled, which is inconsistent wrt. the "Upgrade" button (but if I click on it I'm told "No ISO image selected" as expected); that's another bug, please handle it separately
- once I select an ISO image the "Upgrade" button is enabled
- clicking "Upgrade" fails with "Unknown GLib exception while trying to mount device: udisks-error-quark: GDBus.Error:org.freedesktop.UDisks2.Error.Failed: Error mounting /dev/sdb1 at /media/intrigeri/Tails2: Wrong fs type, /dev/sdb1 has an invalid superblock or missing helper program. (0)"
- unplug / re-plug the USB stick => I see an "Install" button, which works.
That's a regression because in 4.4.18+dfsg-1 when clicking "Upgrade" I was told:
It is impossible to upgrade the device TOSHIBA TransMemory (7.8 GB) - /dev/sdb1 because it was not created using Tails Installer. You should instead use "Install from ISO" to upgrade Tails on this device.
… and when clicking "Install" the correct device (
/dev/sdb) is selected.
Granted, it's a corner case and thankfully our doc does not instructs anyone to do that. But it makes the developer experience a bit painful, and (worse) confusing when working on the code that handles isohybrid'ed devices. Also, I had doubts wrt. the robustness of some changes made in our isohybrid detection code on this ticket, and this experience decreases a bit my level of confidence.
And here is the response to the above from #8860#note-67:
So this is exactly what we do in the
Writing a Tails isohybrid to a USB drive [...] scenario, except that the
dd:ing doesn't happen during the same session we start
tails-installer. I'm hypothesizing that the flakiness you've seen re: "isohybrid detection" only occurs during the session the isohybrid is
dd:ed to the disk, due to a race between the writing of the partition table to the disk, and the scanning of said partition table, possibly in an intermediate state of the write.
- Assignee changed from anonym to kurono
kurono, perhaps you would like to work on this? Ideally it should be done for 3.3 but don't worry too much: let's first focus on polishing (if needed) and merging the work you've already submitted for QA; this one can wait a bit more.
- Priority changed from Normal to Elevated
I'm raising priority because we see lots of failures in our test suite when installing with Tails Installer on a device that had an ISO dd'ed to, but this time it's happening in a different session: the computer was rebooted in the meantime, which cancels anonym's initial hypothesis. See the full debug log on #15031. I don't know if this failure is related to the problem that made us open this ticket, but we already have #10912 to track a different kind of failures in similar situations, and I'd rather not create a 3rd ticket about this, so let's say this one is about the regressions wrt. isohybrid'ed sticks handling introduced in Tails Installer 5.x.
kurono, is this something you think you can work on in the next few weeks? Ideally something would be ready to review early in January, so we have time to review/polish/etc. before the 3.5 release.