Re: [Qemu-devel] [PATCH 3/3] QMP: Introduce the BLOCK_MEDIA_EJECT event

[Kevin Wolf <kwolf@redhat.com>]

Am 30.05.2011 16:49, schrieb Markus Armbruster:
> Luiz Capitulino <lcapitulino@redhat.com> writes:
> 
>> On Sat, 28 May 2011 09:58:24 +0200
>> Markus Armbruster <armbru@redhat.com> wrote:
>>
>>> Luiz Capitulino <lcapitulino@redhat.com> writes:
>>>> diff --git a/block.h b/block.h
>>>> index 1f58eab..e4053dd 100644
>>>> --- a/block.h
>>>> +++ b/block.h
>>>> @@ -50,6 +50,7 @@ typedef enum {
>>>>      BDRV_ACTION_REPORT, BDRV_ACTION_IGNORE, BDRV_ACTION_STOP
>>>>  } BlockMonEventAction;
>>>>  
>>>> +void bdrv_eject_mon_event(const BlockDriverState *bdrv);
>>>>  void bdrv_error_mon_event(const BlockDriverState *bdrv,
>>>>                            BlockMonEventAction action, int is_read);
>>>>  void bdrv_info_print(Monitor *mon, const QObject *data);
>>>> diff --git a/blockdev.c b/blockdev.c
>>>> index 6e0eb83..5fd0043 100644
>>>> --- a/blockdev.c
>>>> +++ b/blockdev.c
>>>> @@ -661,6 +661,11 @@ static int eject_device(Monitor *mon, BlockDriverState *bs, int force)
>>>>              return -1;
>>>>          }
>>>>      }
>>>> +
>>>> +    if (bdrv_is_removable(bs) && bdrv_is_inserted(bs)) {
>>>> +        bdrv_eject_mon_event(bs);
>>>> +    }
>>>> +
>>>>      bdrv_close(bs);
>>>>      return 0;
>>>>  }
>>>
>>> This covers monitor-initiated eject (commands eject and change).
>>>
>>> The event is not suppressed when the tray is already open (previous
>>> guest-initiated eject), is it?.  Contradicts spec.
>>
>> That's a bug.
>>
>>> The event is suppressed when the tray is empty.
>>>
>>> "eject -f" on a non-removable drive does not trigger an event.  Why
>>> treat it specially?  I'm not saying you shouldn't, just wondering.
>>
>> Ejecting a non-removable drive is a qemu bug.
> 
> It's clearly intentional, so it's a (mis-)feature, not a bug.

Is there really a use case for it? The closest thing to a specification
that we have is the help text and it says:

        .help       = "eject a removable medium (use -f to force it)",

QMP describes it like this:

        Eject a removable medium.

So I start tending to agree that this whole trouble around the 'eject'
monitor command is in fact a long standing bug rather than overloaded
semantics. Nowhere is stated that it disconnects a BlockDriverState from
the image, and I can't imagine a use case for this semantics either.

Do we break anything if we make eject really eject the medium (we have a
virtual tray status now) instead of just closing the image? I think the
most visible change is that we'll eject the host medium when using
pass-through. I consider this an additional bugfix.

Kevin