user jobs cause kernel panic

This behavior occurs in both natty upstart package and 1.3 built from source. (DBus access rules taken from 1.3 in both cases)

$ cat $HOME/.init/testjob.conf
#### the test job ####
description "Crashes your system"

task

script
end script
#### end test job

$ initctl --system start testjob
initctl: Message did not receive a reply (timeout by message bus)

Init crashes only if the job executes any command or has "script - end script", even if it's empty.

Jobs starting on signals and with dbus-send like that also crash init.
$ dbus-send --system --print-reply --dest=com.ubuntu.Upstart /com/ubuntu/Upstart/jobs/1000/testjob com.ubuntu.Upstart0_6.Job.Stop array:string: boolean:true

Abstract jobs which do nothing (do not contain exec or script statements) start and stop as they should. Starting on manually emitted signals also work with those jobs.

Crash is caused by assertion failure in job_process.c:849 (job_process_error_read).

Whilst the bug is investigated, we strongly recommend user jobs are disabled by changing /etc/dbus-1/system.d/Upstart.conf to, for example, the version of this file shipped by Ubuntu which disables user jobs:

http://bazaar.launchpad.net/~ubuntu-branches/ubuntu/oneiric/upstart/oneiric/view/head:/dbus/Upstart.conf

The assertion error is raised because of unhandled err->type == 10 in job_process_error_read, with strerror output "Operation not permitted".

Seems that setuid is failing somehow in job_process.c:643

Uh, looked at a wrong line in job_process.h, assuming enum is still zero-based.
The error is JOB_PROCESS_ERROR_SETGID (job_process.c:648)

Seems this was a n00b bug with setting UID before GID, and since root priviliges were already lost, setgid failed.

The attached patch solves this problem.