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 <[email protected]>
---
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 <[email protected]>
> ---
> 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