[Qemu-devel] virtio-9p: hot-plug/unplug about virtio-9p device

[Qemu-devel] virtio-9p: hot-plug/unplug about virtio-9p device

[sochin.jiang]

     Hi, guys.

    I'm looking for the hot-plug/unplug features of virtio-9p device recently, and found there's a lack of support.

    I am wondering why ? Is there a reason. Actually, I write a qmp command to support fsdev_add, then a device_add qmp will    

    successfully add a virtio-9p device(just like virtio-blk).

   Whether there is some concerns that we haven't support this ? hope for a reply, thanks.


sochin.

Re: [Qemu-devel] virtio-9p: hot-plug/unplug about virtio-9p device

[Greg Kurz]

On Tue, 19 Dec 2017 13:41:12 +0800
sochin.jiang <sochin.jiang@huawei.com> wrote:

>      Hi, guys.
> 
>     I'm looking for the hot-plug/unplug features of virtio-9p device recently, and found there's a lack of support.
> 
>     I am wondering why ? Is there a reason. Actually, I write a qmp command to support fsdev_add, then a device_add qmp will    
> 
>     successfully add a virtio-9p device(just like virtio-blk).
> 
>    Whether there is some concerns that we haven't support this ? hope for a reply, thanks.
> 
> 
> sochin.
> 

Hi Sochin,

I've just discovered this mail by chance. Please note I'm the only active
maintainer for virtio-9p. You really should check the content of MAINTAINERS
and Cc the appropriate people. :)

Now, back to your question. Yes, there's only some partial support for
hot-plug/unplug. Mostly because nobody cared to work on it I guess.

So, indeed, we don't have fsdev_add/fsdev_del, ie, we can only rely on
shared directories specified on the QEMU command line.

On the virtio-9p device side, hotplug is supported, ie, device_add virtio-9p
works as expected.

Hot-unplug is different as it requires some coordination with the guest. The
current status is that it requires the 9p shared directory to be unmounted
in the guest: if you do device_del while the directory is mounted in a linux
guest, you'll get these messages in the guest syslog:

kernel:9pnet_virtio virtio2: p9_virtio_remove: waiting for device in use.

If the 9p directory is unmounted at some point, then the hot unplug
sequence will eventually succeed.

But this shouldn't be done like this: the guest should cancel inflight
requests and cause any new I/O requests in the guest to fail right away.
I have a tentative patch for the linux driver I can share if you want.

Cheers,

--
Greg

Re: [Qemu-devel] virtio-9p: hot-plug/unplug about virtio-9p device

[sochin.jiang]

On 2018/1/8 18:10, Greg Kurz wrote:
> On Tue, 19 Dec 2017 13:41:12 +0800
> sochin.jiang <sochin.jiang@huawei.com> wrote:
>
>>      Hi, guys.
>>
>>     I'm looking for the hot-plug/unplug features of virtio-9p device recently, and found there's a lack of support.
>>
>>     I am wondering why ? Is there a reason. Actually, I write a qmp command to support fsdev_add, then a device_add qmp will    
>>
>>     successfully add a virtio-9p device(just like virtio-blk).
>>
>>    Whether there is some concerns that we haven't support this ? hope for a reply, thanks.
>>
>>
>> sochin.
>>
> Hi Sochin,
>
> I've just discovered this mail by chance. Please note I'm the only active
> maintainer for virtio-9p. You really should check the content of MAINTAINERS
> and Cc the appropriate people. :)
>
> Now, back to your question. Yes, there's only some partial support for
> hot-plug/unplug. Mostly because nobody cared to work on it I guess.
>
> So, indeed, we don't have fsdev_add/fsdev_del, ie, we can only rely on
> shared directories specified on the QEMU command line.
>
> On the virtio-9p device side, hotplug is supported, ie, device_add virtio-9p
> works as expected.
>
> Hot-unplug is different as it requires some coordination with the guest. The
> current status is that it requires the 9p shared directory to be unmounted
> in the guest: if you do device_del while the directory is mounted in a linux
> guest, you'll get these messages in the guest syslog:
>
> kernel:9pnet_virtio virtio2: p9_virtio_remove: waiting for device in use.
>
> If the 9p directory is unmounted at some point, then the hot unplug
> sequence will eventually succeed.
>
> But this shouldn't be done like this: the guest should cancel inflight
> requests and cause any new I/O requests in the guest to fail right away.
> I have a tentative patch for the linux driver I can share if you want.
>
> Cheers,
>
> --
> Greg
>
> .
>


Thanks, Greg.

Indeed, I got those error messages while trying to do device_del with
shared directory mounted in the guest. I really would like to see you patch for linux driver.

About virtio-9p, we are actually considering to use it for hypervisor-based container, 
a support of hotplug/unplug will be better, also I believe there will be more people to work on it.


Sochin.

Re: [Qemu-devel] virtio-9p: hot-plug/unplug about virtio-9p device

[Greg Kurz]

On Thu, 11 Jan 2018 10:23:32 +0800
sochin.jiang <sochin.jiang@huawei.com> wrote:

> On 2018/1/8 18:10, Greg Kurz wrote:
> > On Tue, 19 Dec 2017 13:41:12 +0800
> > sochin.jiang <sochin.jiang@huawei.com> wrote:
> >  
> >>      Hi, guys.
> >>
> >>     I'm looking for the hot-plug/unplug features of virtio-9p device recently, and found there's a lack of support.
> >>
> >>     I am wondering why ? Is there a reason. Actually, I write a qmp command to support fsdev_add, then a device_add qmp will    
> >>
> >>     successfully add a virtio-9p device(just like virtio-blk).
> >>
> >>    Whether there is some concerns that we haven't support this ? hope for a reply, thanks.
> >>
> >>
> >> sochin.
> >>  
> > Hi Sochin,
> >
> > I've just discovered this mail by chance. Please note I'm the only active
> > maintainer for virtio-9p. You really should check the content of MAINTAINERS
> > and Cc the appropriate people. :)
> >
> > Now, back to your question. Yes, there's only some partial support for
> > hot-plug/unplug. Mostly because nobody cared to work on it I guess.
> >
> > So, indeed, we don't have fsdev_add/fsdev_del, ie, we can only rely on
> > shared directories specified on the QEMU command line.
> >
> > On the virtio-9p device side, hotplug is supported, ie, device_add virtio-9p
> > works as expected.
> >
> > Hot-unplug is different as it requires some coordination with the guest. The
> > current status is that it requires the 9p shared directory to be unmounted
> > in the guest: if you do device_del while the directory is mounted in a linux
> > guest, you'll get these messages in the guest syslog:
> >
> > kernel:9pnet_virtio virtio2: p9_virtio_remove: waiting for device in use.
> >
> > If the 9p directory is unmounted at some point, then the hot unplug
> > sequence will eventually succeed.
> >
> > But this shouldn't be done like this: the guest should cancel inflight
> > requests and cause any new I/O requests in the guest to fail right away.
> > I have a tentative patch for the linux driver I can share if you want.
> >
> > Cheers,
> >
> > --
> > Greg
> >
> > .
> >  
> 
> 
> Thanks, Greg.
> 
> Indeed, I got those error messages while trying to do device_del with
> shared directory mounted in the guest. I really would like to see you patch for linux driver.
> 

I'll try to send it soon.

> About virtio-9p, we are actually considering to use it for hypervisor-based container, 

Something like Intel's Clear Containers ?

> a support of hotplug/unplug will be better, also I believe there will be more people to work on it.

That's good news !

Cheers,

--
Greg

> 
> 
> Sochin.
> 
> 
>