opencryptoki postinst fails when pkcsslotd is already running

[Thomas Bushnell, BSG]

Version 2.3.1+dfsg-3ubuntu4.

opencryptoki postinst does
  invoke-rc.d opencryptoki start || exit $?

In turn, /etc/init.d/opencryptoki [start] does
  start-stop-daemon --start --quiet --pidfile "/var/lib/opencryptoki/.slotid --exec /usr/sbin/pkcsslotd --

But if pkcsslotd is already running, this fails.

# /etc/init.d/opencryptoki stop
Stopping PKCS#11 slot daemon: pkcsslotd.
# echo $?
0
# /etc/init.d/opencryptoki start
Starting PKCS#11 slot daemon: # echo $?
1

Note as well that when it fails, it fails badly (no newline or error message).

Here is the error from pkcsslotd:

# ps augxw | grep pkcsslotd
root 6351 0.0 0.0 20164 896 pts/0 S+ 16:35 0:00 grep --color=auto pkcsslotd
# /usr/sbin/pkcsslotd
# echo $?
0
# /usr/sbin/pkcsslotd
  ERROR pkcsslotd[6395.755480320]: Shared memory creation failed (0x11)
  ERROR pkcsslotd[6395.755480320]: perform ipcrm -M 0x62000E67
# echo $?
1

[Thomas Bushnell, BSG (tbushnell)]

It is possible that this is the root cause underlying https://bugs.launchpad.net/ubuntu/+source/opencryptoki/+bug/963533.

[Thomas Bushnell, BSG (tbushnell)]

Whoops, error in editing. the transcript of the failing init.d script should look like this:

# /etc/init.d/opencryptoki stop
Stopping PKCS#11 slot daemon: pkcsslotd.
# echo $?
0
# /etc/init.d/opencryptoki start
Starting PKCS#11 slot daemon: pkcsslotd.
# echo $?
0
# /etc/init.d/opencryptoki start
Starting PKCS#11 slot daemon: # echo $?
1

[Stéphane Graber (stgraber)]

The question is why was pkcsslotd still running... I'd expect the prerm to stop it and postinst to start it, making the only failing case happen if someone manually starts it at the middle of the upgrade.

Is that a case where "stop" doesn't actually stop pkcsslotd?
I did a few tests here and it seems to always stop/start reliably and return 0 as expected (unless you start it twice).

An easy workaround is to make the init script return 0 if the daemon is already running but I'm not sure returning 1 in that case is really wrong.

[Thomas Bushnell BSG (tb-becket)]

Are you sure that every old version of the package correctly stopped the daemon in the prerm?

(Hey, that might explain why the other bug I noted was happening.)