[systemd-devel] journal corruption on SIGKILL and mostly full disk: Bad message

[Marti Raudsepp]

Hi list,

As usual, attempting to debug one problem inevitably leads to another
bug. I found out that after filling my disk and SIGKILLing journald a
few times, it seems to corrupt the journal archive. Seems to be
reproducible every time.
journalctl reports: Failed to iterate through journal: Bad message

Running systemd-journald from git HEAD.

# Allocate large file to mostly fill the disk (770 MB free out of 32 GB)
fallocate --length=12500000000 /filler
rm -rf /var/log/journal/*
killall -9 systemd-journald
journalctl -n10

May 30 22:38:37 newn systemd-journal[28656]: Journal started
May 30 22:38:37 newn systemd[1]: systemd-journald.service: main
process exited, code=killed, status=9

killall -9 systemd-journald
journalctl -n10

May 30 22:38:37 newn systemd-journal[28656]: Journal started
May 30 22:38:37 newn systemd[1]: systemd-journald.service: main
process exited, code=killed, status=9

killall -9 systemd-journald
journalctl -n10

Failed to iterate through journal: Bad message

ls -lA /var/log/journal/dd9596f699f65b18472a55a24aa17111
-rw-r----- 1 root adm  36864 May 30 22:46
system at e82653abcaa74d8e9bb413325f98813b-0000000000000002-0000000000000000.journal
-rw-r----- 1 root root 36864 May 30 22:46 system.journal

The two broken files (mostly consisting of zeroes) attached.

Regards,
Marti