Re: [Qemu-devel] [PATCH v5 10/11] qemu-ga: Install Windows VSS provider on `qemu-ga -s install'

[Laszlo Ersek]

On 07/08/13 16:16, Tomoki Sekiyama wrote:
> On 7/8/13 9:58 , "Laszlo Ersek" <address@hidden> wrote:
> 
>> On 07/05/13 19:06, Tomoki Sekiyama wrote:
>>> On 7/4/13 8:54 , "Paolo Bonzini" <address@hidden> wrote:
>>>
>>>> Il 03/07/2013 18:19, Tomoki Sekiyama ha scritto:
>>>>> On 7/3/13 11:58 , "Paolo Bonzini" <address@hidden> wrote:
>>>>>
>>>>>> Il 03/07/2013 17:49, Tomoki Sekiyama ha scritto:
>>>>>>> -                return ga_install_service(path, log_filepath,
>>>>>>> fixed_state_dir);
>>>>>>> +                if (ga_install_vss_provider()) {
>>>>>>> +                    return EXIT_FAILURE;
>>>>>>> +                }
>>>>>>> +                if (ga_install_service(path, log_filepath,
>>>>>>> fixed_state_dir)) {
>>>>>>> +                    ga_uninstall_vss_provider();
>>>>>>> +                    return EXIT_FAILURE;
>>>>>>> +                }
>>>>>>> +                return 0;
>>>>>>>              } else if (strcmp(service, "uninstall") == 0) {
>>>>>>> +                ga_uninstall_vss_provider();
>>>>>>>                  return ga_uninstall_service();
>>>>>>
>>>>>> I think this shouldn't be a hard failure.  Only the freeze/thaw
>>>>>> commands
>>>>>> should fail.
>>>>>>
>>>>>> Paolo
>>>>>
>>>>> Do you mean that qemu-ga should work without qga-provider.dll etc.
>>>>> even if it is configured --with-vss-sdk ?
>>>>
>>>> Yes, and I'm even wondering if we should move all VSS code to a DLL
>>>> (provider and requestor---they are very tied to each other anyway
>>>> because of hEventFrozen/hEventThaw), and have qemu-ga simply look for
>>>> qga-provider.dll dropped into the executable directory.
>>>>
>>>> Then qemu-ga can look for it even if it is not configured
>>>> --with-vss-sdk.
>>>
>>> Hm, that sounds reasonable.
>>>
>>> I will try on moving the requestor into qga-provider.dll at next
>>> iteration. 
>>
>> Is it OK with you if I skip reviewing v5 then? If the v5->v6
>> reorganization is going to be intrusive (ie. I won't be able to do a
>> patch-by-patch interdiff), then I'd prefer skipping v5.
> 
> Yes. I appreciate your review for the next version.
> 
> It actually requires some rework especially in "Error" handling in
> the requester, because I'd not like to link glib and some other dlls
> to VSS system service.... May be need to give function pointer of
> error_set_win32() to the dll, or something like that.

I haven't given much thought yet to extracting the requestor side into a
DLL as well (I did read your discussion with Paolo, but that was it).

Should one DLL contain both the requestor and the provider? I think that
would not be very pure, the DLL registered as VSS provider would carry a
lot of cruft. Vice versa, when you load the same DLL into qemu-ga for
the requestor role, then the provider parts are cruft.

OTOH two DLLs would be overly complex...

Hopefully windows loads the DLL with an mmap()-like mechanism, and the
two instances of the one common-purpose DLL will be shared im memory.

Laszlo