Re: [PATCH] 9p: disconnect channel when PCI device is removed

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

Sasha Levin <levinsasha928@gmail.com> writes:

> On Mon, Apr 16, 2012 at 12:59 PM, Aneesh Kumar K.V
> <aneesh.kumar@linux.vnet.ibm.com> wrote:
>> Sasha Levin <levinsasha928@gmail.com> writes:
>>
>>> On Sun, Apr 15, 2012 at 2:27 PM, Aneesh Kumar K.V
>>> <aneesh.kumar@linux.vnet.ibm.com> wrote:
>>>> Sasha Levin <levinsasha928@gmail.com> writes:
>>>>
>>>>> When a virtio_9p pci device is being removed, we should close down any
>>>>> active channels and free up resources, we're not supposed to BUG() if there's
>>>>> still an open channel since it's a valid case when removing the PCI device.
>>>>>
>>>>> Otherwise, removing the PCI device with an open channel would cause the
>>>>> following BUG():
>>>>>
>>>> ...
>>>>
>>>>> diff --git a/net/9p/trans_virtio.c b/net/9p/trans_virtio.c
>>>>> index 3d43206..5af18d1 100644
>>>>> --- a/net/9p/trans_virtio.c
>>>>> +++ b/net/9p/trans_virtio.c
>>>>> @@ -615,7 +615,8 @@ static void p9_virtio_remove(struct virtio_device *vdev)
>>>>>  {
>>>>>       struct virtio_chan *chan = vdev->priv;
>>>>>
>>>>> -     BUG_ON(chan->inuse);
>>>>> +     if (chan->inuse)
>>>>> +             p9_virtio_close(chan->client);
>>>>>       vdev->config->del_vqs(vdev);
>>>>>
>>>>>       mutex_lock(&virtio_9p_lock);
>>>>
>>>> But an umount should have resulted in p9_virtio_close ? How are you
>>>> removing the device ? Are you removing the device with file system
>>>> mounted  ?. In that case may be we should return EBUSY ?
>>>
>>> I signal the underlying PCI device to remove (echo 1 >
>>> /sys/devices/pci0000\:00/[...]/remove), we can't really prevent that
>>> thing so we must clean up ourselves.
>>
>> What does that mean for the mounted file system ? What would happen to
>> the pending fs operations in that case ?
>
> I'm guessing that all of them should be canceled.

Pending operation we can cancel, but what about dirty pages in cached
mode ? Also how does virtio-blk handle this ? 

> virtio-pci simulates
> a PCI device, if the PCI device is unplugged there's not much to do
> about the filesystem or pending requests.

Ideal thing to do would be to make remove return -EBUSY; let the user
umount. But looking at the code, I guess there is no easy way to return
error from remove callback ?

-aneesh