Virtualization - Apr 2019 - [PATCH] virtio_console: remove vq buf while unpluging port

A bug can be easily reproduced:
Host# cat guest-agent.xml
<channel type="unix">
  <source mode="bind"
path="/var/lib/libvirt/qemu/stretch.agent"/>
  <target type="virtio" name="org.qemu.guest_agent.0"
state="connected"/>
</channel>
Host# virsh attach-device instance guest-agent.xml
Host# virsh detach-device instance guest-agent.xml
Host# virsh attach-device instance guest-agent.xml

and guest report: virtio-ports vport0p1: Error allocating inbufs

The reason is that the port is unplugged and the vq buf still remained.
So, fix two cases in this patch:
1, fix memory leak with attach-device/detach-device.
2, fix logic bug with attach-device/detach-device/attach-device.

Signed-off-by: zhenwei pi <pizhenwei at bytedance.com>
---
 drivers/char/virtio_console.c | 21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
index fbeb719..f6e37f4 100644
--- a/drivers/char/virtio_console.c
+++ b/drivers/char/virtio_console.c
@@ -251,6 +251,7 @@ struct port {
 
 /* This is the very early arch-specified put chars function. */
 static int (*early_put_chars)(u32, const char *, int);
+static void remove_vq(struct virtqueue *vq);
 
 static struct port *find_port_by_vtermno(u32 vtermno)
 {
@@ -1550,6 +1551,9 @@ static void unplug_port(struct port *port)
 	}
 
 	remove_port_data(port);
+	spin_lock_irq(&port->inbuf_lock);
+	remove_vq(port->in_vq);
+	spin_unlock_irq(&port->inbuf_lock);
 
 	/*
 	 * We should just assume the device itself has gone off --
@@ -1945,17 +1949,22 @@ static const struct file_operations portdev_fops = {
 	.owner = THIS_MODULE,
 };
 
+static void remove_vq(struct virtqueue *vq)
+{
+	struct port_buffer *buf;
+
+	flush_bufs(vq, true);
+	while ((buf = virtqueue_detach_unused_buf(vq)))
+		free_buf(buf, true);
+}
+
 static void remove_vqs(struct ports_device *portdev)
 {
 	struct virtqueue *vq;
 
-	virtio_device_for_each_vq(portdev->vdev, vq) {
-		struct port_buffer *buf;
+	virtio_device_for_each_vq(portdev->vdev, vq)
+		remove_vq(vq);
 
-		flush_bufs(vq, true);
-		while ((buf = virtqueue_detach_unused_buf(vq)))
-			free_buf(buf, true);
-	}
 	portdev->vdev->config->del_vqs(portdev->vdev);
 	kfree(portdev->in_vqs);
 	kfree(portdev->out_vqs);
-- 
2.7.4

On Sun, Apr 28, 2019 at 09:50:04AM +0800, zhenwei pi
wrote:
> A bug can be easily reproduced:
> Host# cat guest-agent.xml
> <channel type="unix">
>   <source mode="bind"
path="/var/lib/libvirt/qemu/stretch.agent"/>
>   <target type="virtio"
name="org.qemu.guest_agent.0" state="connected"/>
> </channel>
> Host# virsh attach-device instance guest-agent.xml
> Host# virsh detach-device instance guest-agent.xml
> Host# virsh attach-device instance guest-agent.xml
> 
> and guest report: virtio-ports vport0p1: Error allocating inbufs
> 
> The reason is that the port is unplugged and the vq buf still remained.
> So, fix two cases in this patch:
> 1, fix memory leak with attach-device/detach-device.
> 2, fix logic bug with attach-device/detach-device/attach-device.
> 
> Signed-off-by: zhenwei pi <pizhenwei at bytedance.com>
> ---
>  drivers/char/virtio_console.c | 21 +++++++++++++++------
>  1 file changed, 15 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
> index fbeb719..f6e37f4 100644
> --- a/drivers/char/virtio_console.c
> +++ b/drivers/char/virtio_console.c
> @@ -251,6 +251,7 @@ struct port {
>  
>  /* This is the very early arch-specified put chars function. */
>  static int (*early_put_chars)(u32, const char *, int);
> +static void remove_vq(struct virtqueue *vq);
>  
>  static struct port *find_port_by_vtermno(u32 vtermno)
>  {
> @@ -1550,6 +1551,9 @@ static void unplug_port(struct port *port)
>  	}
>  
>  	remove_port_data(port);
> +	spin_lock_irq(&port->inbuf_lock);
> +	remove_vq(port->in_vq);
> +	spin_unlock_irq(&port->inbuf_lock);
>  
>  	/*
>  	 * We should just assume the device itself has gone off --
> @@ -1945,17 +1949,22 @@ static const struct file_operations portdev_fops =
{
>  	.owner = THIS_MODULE,
>  };
>  
> +static void remove_vq(struct virtqueue *vq)
> +{
> +	struct port_buffer *buf;
> +
> +	flush_bufs(vq, true);
> +	while ((buf = virtqueue_detach_unused_buf(vq)))
> +		free_buf(buf, true);
> +}
> +
>  static void remove_vqs(struct ports_device *portdev)
>  {
>  	struct virtqueue *vq;
>  
> -	virtio_device_for_each_vq(portdev->vdev, vq) {
> -		struct port_buffer *buf;
> +	virtio_device_for_each_vq(portdev->vdev, vq)
> +		remove_vq(vq);
>  
> -		flush_bufs(vq, true);
> -		while ((buf = virtqueue_detach_unused_buf(vq)))
> -			free_buf(buf, true);
> -	}
>  	portdev->vdev->config->del_vqs(portdev->vdev);
>  	kfree(portdev->in_vqs);
>  	kfree(portdev->out_vqs);
> -- 
> 2.7.4

Amit, any ideas if this is valid or not and if this should be applied?

thanks,

greg k-h

On Fri, 2019-05-24 at 20:51 +0200, Greg KH wrote:
> On Sun, Apr 28, 2019 at 09:50:04AM +0800, zhenwei pi wrote:
> > A bug can be easily reproduced:
> > Host# cat guest-agent.xml
> > <channel type="unix">
> >   <source mode="bind"
path="/var/lib/libvirt/qemu/stretch.agent"/>
> >   <target type="virtio"
name="org.qemu.guest_agent.0"
> > state="connected"/>
> > </channel>
> > Host# virsh attach-device instance guest-agent.xml
> > Host# virsh detach-device instance guest-agent.xml
> > Host# virsh attach-device instance guest-agent.xml
> > 
> > and guest report: virtio-ports vport0p1: Error allocating inbufs
> > 
> > The reason is that the port is unplugged and the vq buf still
> > remained.
> > So, fix two cases in this patch:
> > 1, fix memory leak with attach-device/detach-device.
> > 2, fix logic bug with attach-device/detach-device/attach-device.
The "leak" happens because the host-side of the connection is still
connected.  This is by design -- if a guest has written data before
being unplugged, the port isn't released till the host connection goes
down to ensure a host process reads all the data out of the port.

Can you try similar, but also disconnecting the host side and see if
that fixes things?
> Amit, any ideas if this is valid or not and if this should be
> applied?
This had indeed been missed, thanks!


			Amit