Hi
On Thu, Jun 4, 2020 at 10:54 AM Christian Ehrhardt
<christian.ehrha...@canonical.com> wrote:
>
> Hi,
> while debugging a report I got in Ubuntu I found that since qemu 4.0 the
> guest agent shutdown feature works (guest is shutting down) but crashes when
> doing so each time. This can be a big red herring when debugging other things
> as well as people start to get "an application crashed, do you want to
> report" pop-ups if they have set up automatic crash reports.
>
> If you boot the guest after starting it again and check the guest-agent
> status you will see in journal:
> -- Logs begin at Tue 2020-06-02 07:41:32 UTC, end at Thu 2020-06-04 08:07:37
> UTC. --
> Jun 02 07:47:58 focal systemd[1]: Started QEMU Guest Agent.
> Jun 02 07:49:03 focal qemu-ga[1984]: info: guest-shutdown called, mode: (null)
> Jun 02 07:49:03 focal qemu-ga[1984]: **
> Jun 02 07:49:03 focal qemu-ga[1984]:
> ERROR:/build/qemu-7aKH5L/qemu-4.2/qga/main.c:532:send_response: assertion
> failed: (rsp && s->channel)
> Jun 02 07:49:03 focal qemu-ga[1984]: Bail out!
> ERROR:/build/qemu-7aKH5L/qemu-4.2/qga/main.c:532:send_response: assertion
> failed: (rsp && s->channel)
> Jun 02 07:49:04 focal systemd[1]: Stopping QEMU Guest Agent...
> Jun 02 07:49:04 focal systemd[1]: qemu-guest-agent.service: Succeeded.
> Jun 02 07:49:04 focal systemd[1]: Stopped QEMU Guest Agent.
>
> The actual assert is from "forever" [3] (v0.15) which is the initial addition
> of qemu guest agent in 2011. That was later restructured in [1] (v1.1) and
> [2] (v4.0).
>
> In a check through Ubuntu releases I got
> 1) Host: Q 2.11 L 4.0 (Bionic) - G 2.11 (Bionic)
> 2) Host: Q 4.0 L 5.4 (Eoan) - G 2.11 (Bionic)
> 3) Host: Q 4.2 L 6.0 (Focal) - G 2.11 (Bionic)
> 4) Host: Q 2.11 L 4.0 (Bionic) - G 4.0 (Eoan)
> 5) Host: Q 4.0 L 5.4 (Eoan) - G 4.0 (Eoan)
> 6) Host: Q 4.2 L 6.0 (Focal) - G 4.0 (Eoan)
> 7) Host: Q 2.11 L 4.0 (Bionic) - G 4.2 (Focal)
> 8) Host: Q 4.0 L 5.4 (Eoan) - G 4.2 (Focal)
> 9) Host: Q 4.2 L 6.0 (Focal) - G 4.2 (Focal)
>
> So it seemed to be the qemu-guest-agent portion since >=4.0.
> I did a build with [2] reverted and the crash is gone.
>
> I see from the host:
> $ virsh qemu-agent-command focal '{"execute": "guest-shutdown"}'
> "error: Guest agent is not responding: Guest agent disappeared while
> executing command"
>
> I'm not sure which part of the communication breaks first, but it could try
> to send on a dying connection, the old code had:
>
> rsp = qmp_dispatch(&ga_commands, QOBJECT(req), false);
> if (rsp) {
> ret = send_response(s, rsp)
>
> While the new code is like:
>
> rsp = qmp_dispatch(&ga_commands, obj, false);
> end:
> ret = send_response(s, rsp);
>
> Maybe it runs send_response despite qmp_dispatch failing now?
>
> I didn't stare at it long enough to have a solution yet, but wanted to make
> the maintainer of qga and the Author aware.
>
My bad, "guest-shutdown" is a QCO_NO_SUCCESS_RESP command, and in this
case qmp_dispatch() returns NULL. I'll send a patch.
thanks
> [1]: https://git.qemu.org/?p=qemu.git;a=commit;h=125b310e1d62
> [2]: https://git.qemu.org/?p=qemu.git;a=commit;h=781f2b3d1e5e
> [3]: https://git.qemu.org/?p=qemu.git;a=commit;h=48ff7a625b36
>
> --
> Christian Ehrhardt
> Staff Engineer, Ubuntu Server
> Canonical Ltd
