Re: [Qemu-devel] [PATCH 3/3] qga: set file descriptor in qmp_guest_file_open non-blocking on Win32

["Denis V. Lunev" <den@openvz.org>]

On 10/27/2015 10:49 PM, Michael Roth wrote:
> Quoting Denis V. Lunev (2015-10-27 14:13:57)
>> On 10/27/2015 10:11 PM, Michael Roth wrote:
>>> Quoting Denis V. Lunev (2015-10-27 12:48:43)
>>>> From: Olga Krishtal <okrishtal@parallels.com>
>>>>
>>>> Set fd non-blocking to avoid common use cases (like reading from a
>>>> named pipe) from hanging the agent. This was missed in the original
>>>> code.
>>>>
>>>> The patch introduces analog of qemu_set_non/block for HANDLES.
>>>> The usage of handles in qemu_set_non/block is impossible, because for
>>>> win32 there is a difference between file discriptors and file handles,
>>>> and all file ops are made via Win32 api.
>>> If this is specific to HANDLEs, why do we need to cast back and forth
>>> between int64_t and HANDLE? I haven't build tested, but it seems like
>>> this would break for 32-bit mingw builds.
>>>
>>> I would define these as qemu_set_*_by_handle(HANDLE fh, ...) instead
>>> and make them win32 only. If someone wants to introduce a FILE*
>>> variant for posix they can introduce it as
>>> qemu_set_*_by_handle(FILE *fh, ...) rather than us needing to
>>> abstract away the handle type.
>> may be it would be better to add static function for this in QGA for now?
> I'd be fine with either approach. It could be generally useful for
> other w32 users. But if we're thinking about dropping the QGA
> use case soon then maybe having it live in QGA is best.
>
>> I am eager to drop this code at once for Posix and Windows and
>> switch to GLIB like was done for guest exec.
> You mean switching all the guest-file-* interfaces to glib? I
> took a stab at it once for w32 guest-file-* implementation, but
> one issue I hit was that I couldn't figure out how to implement
> guest-file-seek to report back the absolute position in the
> file, or whether or not we'd hit EOF. You can set position
> via g_io_channel_seek_position(), but if they hit EOF, or are
> using relative offsets via G_SEEK_CUR, you don't really know
> the position and glib doesn't seem to provide a way to query
> that. We could maybe work around it by tracking it manually
> via guest-file-* calls but that sounds terrible.
>
> Hopefully I just missed something though. Also couldn't figure
> out how you can get glib to report that you'd already seeked
> to EOF. I had some comments about it in my WIP:
>
> https://github.com/mdroth/qemu/commit/8b2e5c69266bb48e492af9826122c2aaa4a82197#diff-7f29c3e51a7b387cc7717e7be4f6e205R525

I see. Then we can have platform specific open code and use FILE*
interface for the rest. According to Olga HANDLE -> fd -> FILE*
transition is possible.

At least we will try :)

Den