]> git.ipfire.org Git - thirdparty/systemd.git/commit
test: avoid using external commands in trap handlers
authorFrantisek Sumsal <frantisek@sumsal.cz>
Tue, 21 Apr 2026 10:07:02 +0000 (12:07 +0200)
committerLuca Boccassi <luca.boccassi@gmail.com>
Tue, 21 Apr 2026 15:19:47 +0000 (16:19 +0100)
commit4d92c72b819bd6d54dfbbb12e4cd25de5053714c
tree682a3e09fbe01e09bf5c22de4621394a699b0fee
parent812aa57d2cbcb037219552c48d4ec1a6754892aa
test: avoid using external commands in trap handlers

In #39675 the reported fail was as follows:

5580s [  247.559994] TEST-13-NSPAWN.sh[1858]: Exported 93%.
5580s [  247.659002] TEST-13-NSPAWN.sh[1858]: Exported 95%.
5580s [  247.785893] TEST-13-NSPAWN.sh[1858]: Operation completed successfully.
5580s [  247.923727] TEST-13-NSPAWN.sh[1858]: Exiting.
5580s [  258.300406] TEST-13-NSPAWN.sh[1074]: + machinectl import-raw /var/tmp/container-export.raw container-raw-reimport
5580s [  258.323328] TEST-13-NSPAWN.sh[1884]: The 'machinectl import-raw' command has been replaced by 'importctl -m import-raw'. Redirecting invocation.
5580s [  258.659982] TEST-13-NSPAWN.sh[1884]: Failed to transfer image: Remote peer disconnected
5580s [  258.734218] TEST-13-NSPAWN.sh[1074]: + at_exit

Turns out that the real reason behind this fail is that the machine was
under heavy load due to a busy-loop from the stub init. The cause of
this is a bug in bash, where running commands that fork (i.e. not
built-ins) can cause a permanent busy-loop due to a desync in trap
handling if you send the signals to the bash process _just right_:

[   90.855318] TEST-13-NSPAWN.sh[1074]: + machinectl poweroff long-running long-running long-running
[   90.855318] TEST-13-NSPAWN.sh[1074]: + machinectl reboot long-running long-running long-running
[   90.928980] systemd-nspawn[1679]: ++ touch /poweroff
[   90.928980] systemd-nspawn[1679]: +++ touch /reboot
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + wait
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + wait
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + :
[   90.928980] systemd-nspawn[1679]: + wait
...

$ journalctl --file TEST-13-NSPAWN-1.journal -o short-monotonic --no-hostname --grep "^\+ wait$" | wc -l
349734

So the stub-init was hammering the machine in a tight endless loop,
which then caused systemd-importd to timeout when talking to D-Bus:

[  258.300096] TEST-13-NSPAWN.sh[1074]: + machinectl import-raw /var/tmp/container-export.raw container-raw-reimport
...
[  258.415319] systemd-importd[1859]: Unable to request name, failing connection: Method call timed out
[  258.483662] systemd-importd[1859]: Bus n/a: changing state RUNNING → CLOSING
[  258.605442] systemd-importd[1859]: Bus n/a: changing state CLOSING → CLOSED
[  258.659958] TEST-13-NSPAWN.sh[1884]: Failed to transfer image: Remote peer disconnected

Given this is not our issue, let's work around it by using just
built-ins from the trap handlers, which are not susceptible to this bug.

Resolves: #39675
test/units/TEST-13-NSPAWN.machined.sh