Install fails on ppc64: partman: *** stack smashing detected ***

[Paride Legovini]

Since March 6 the installer of the ubuntu-disco-server-ppc64el daily images consistently fails. The installer can't pass the "Partition disks" stage: the stage fails silently, without showing the classic d-i red error screen. A dig in the syslog shows:

partman: *** stack smashing detected ***: <unknown> terminated
partman: *** stack smashing detected ***: <unknown> terminated
partman-lvm: *** stack smashing detected ***: <unknown> terminated
debconf: --> PROGRESS STEP 1
debconf: <-- 0 OK
debconf: --> PROGRESS INFO partman/progress/init/parted
debconf: <-- 0 OK
debconf: --> GET partman/alignment
debconf: <-- 0 optimal
debconf: --> PROGRESS STOP
debconf: <-- 0 OK
main-menu[2181]: (process:10235): *** stack smashing detected ***: <unknown> terminated
main-menu[2181]: (process:10235): Aborted
debconf: --> GET debconf/priority
debconf: <-- 0 critical
main-menu[2181]: INFO: Menu item 'partman-base' succeeded but requested to be left unconfigured.

but the step clearly did not succeed. Trying again to perform the "Partition disks" step produces more

main-menu[2181]: (process:10235): *** stack smashing detected ***: <unknown> terminated

errors in the syslog, and indeed this is what happens when trying to run partman manually:

~ # partman
*** stack smashing detected ***: <unknown> terminated

This problem renders the current Ubuntu disco images uninstallable on ppc64. Let me know if you need to know any other information to better investigate it.

Paride

[Paride Legovini (paride)]

Attached to this message is the full syslog of the machine running the installer.

[Steve Langasek (vorlon)]

March 5 was the day the glibc 2.29 migration landed in the release pocket, bringing with it a large number of other dependent packages. This could be due to version skew of glibc itself between the d-i initrd and the image, or it could be a regression introduced by some other package updated in the image (merge of new upstream version partman-base migrated that day, so it's my second guess).

A rebuild of d-i against current glibc has just landed, so trying an image respin now to see if this alone is sufficient to fix the problem.

[Paride Legovini (paride)]

Fixed in the last d-i compiled against the current glibc. Thanks a lot for the prompt feedback!