[libvirt] [PATCH] qemu: fix one reboot scenario

[Maxim Nestratov]

07-Nov-16 20:15, Michal Privoznik пишет:

> On 07.11.2016 17:19, Michal Privoznik wrote:
>> On 03.11.2016 19:11, Maxim Nestratov wrote:
>>> Both qemuDomainReboot and qemuDomainShutdownFlags do the following if
>>> they were called to reboot:
>>> 1. use agent and call qemuAgentShutdown
>>> 2. then if the above function doesn't succeed, try qemuMonitorSystemPowerdown
>>>
>>> When the first step is called, it resets fakeReboot flag, while the second one,
>>> opposite to that, sets it. Thus, in case we tried to use agent to reboot a
>>> guest and failed for some reason, we end up with fakeReboot flag set. After that,
>>> as qemuMonitorSystemPowerdown function was called, libvirt is notified with POWERDOWN.
>>> The problem is that there is no callback routine set for it. The lack of monitor event
>>> reaction leads to incorrect logic and guest doesn't restart or reboot correctly.
>>> The patch simply sets domainPowerdown monitor callback to qemuProcessHandleShutdown as
>>> powerdown event processing is actually equal to shutdown.
>>>
>>> Signed-off-by: Maxim Nestratov <mnestratov at virtuozzo.com>
>>> ---
>>>   src/qemu/qemu_process.c | 1 +
>>>   1 file changed, 1 insertion(+)
>>>
>>> diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c
>>> index 04b25fe..0de9fa5 100644
>>> --- a/src/qemu/qemu_process.c
>>> +++ b/src/qemu/qemu_process.c
>>> @@ -1630,6 +1630,7 @@ static qemuMonitorCallbacks monitorCallbacks = {
>>>       .diskSecretLookup = qemuProcessFindVolumeQcowPassphrase,
>>>       .domainEvent = qemuProcessHandleEvent,
>>>       .domainShutdown = qemuProcessHandleShutdown,
>>> +    .domainPowerdown = qemuProcessHandleShutdown,
>>>       .domainStop = qemuProcessHandleStop,
>>>       .domainResume = qemuProcessHandleResume,
>>>       .domainReset = qemuProcessHandleReset,
>>>
>> Huh, this event was never ever handled. How did we even manage that?
> Oh, now looking into the qemu code, maybe we don't want this to be
> handled after all. POWERDOWN event is emitted on ACPI powerdown event,
> which means that guest is still running and hence we don't want to
> mangle our internal state (i.e. set it to SHUTDOWN).
>
>> ACK
> I'm sorry, but I have to discard my own ACK. We need a different fix for
> the scenario you're seeing.
>
> Michal

Yeah, you are absolutely right. Please disregard the patch. I'll send another one.

Maxim