[0/4] vsock/virtio: several fixes in the .probe() and .remove()

Message

Stefano Garzarella May 28, 2019, 10:56 a.m. UTC

During the review of "[PATCH] vsock/virtio: Initialize core virtio vsock
before registering the driver", Stefan pointed out some possible issues
in the .probe() and .remove() callbacks of the virtio-vsock driver.

This series tries to solve these issues:
- Patch 1 postpones the 'the_virtio_vsock' assignment at the end of the
  .probe() to avoid that some sockets queue works when the initialization
  is not finished.
- Patches 2 and 3 stop workers before to call vdev->config->reset(vdev) to
  be sure that no one is accessing the device, and adds another flush at the
  end of the .remove() to avoid use after free.
- Patch 4 free also used buffers in the virtqueues during the .remove().

Stefano Garzarella (4):
  vsock/virtio: fix locking around 'the_virtio_vsock'
  vsock/virtio: stop workers during the .remove()
  vsock/virtio: fix flush of works during the .remove()
  vsock/virtio: free used buffers during the .remove()

 net/vmw_vsock/virtio_transport.c | 105 ++++++++++++++++++++++++++-----
 1 file changed, 90 insertions(+), 15 deletions(-)

Comments

Stefan Hajnoczi June 10, 2019, 1:09 p.m. UTC | #1

On Tue, May 28, 2019 at 12:56:19PM +0200, Stefano Garzarella wrote:
> During the review of "[PATCH] vsock/virtio: Initialize core virtio vsock
> before registering the driver", Stefan pointed out some possible issues
> in the .probe() and .remove() callbacks of the virtio-vsock driver.
> 
> This series tries to solve these issues:
> - Patch 1 postpones the 'the_virtio_vsock' assignment at the end of the
>   .probe() to avoid that some sockets queue works when the initialization
>   is not finished.
> - Patches 2 and 3 stop workers before to call vdev->config->reset(vdev) to
>   be sure that no one is accessing the device, and adds another flush at the
>   end of the .remove() to avoid use after free.
> - Patch 4 free also used buffers in the virtqueues during the .remove().
> 
> Stefano Garzarella (4):
>   vsock/virtio: fix locking around 'the_virtio_vsock'
>   vsock/virtio: stop workers during the .remove()
>   vsock/virtio: fix flush of works during the .remove()
>   vsock/virtio: free used buffers during the .remove()
> 
>  net/vmw_vsock/virtio_transport.c | 105 ++++++++++++++++++++++++++-----
>  1 file changed, 90 insertions(+), 15 deletions(-)

Looking forward to v2.  I took a look at the discussion and I'll review
v2 from scratch.  Just keep in mind that the mutex is used more for
mutual exclusion of the init/exit code than to protect the_virtio_vsock,
so we'll still need protection of init/exit code even with RCU.

Stefan

Stefano Garzarella June 27, 2019, 10:05 a.m. UTC | #2

On Mon, Jun 10, 2019 at 02:09:45PM +0100, Stefan Hajnoczi wrote:
> On Tue, May 28, 2019 at 12:56:19PM +0200, Stefano Garzarella wrote:
> > During the review of "[PATCH] vsock/virtio: Initialize core virtio vsock
> > before registering the driver", Stefan pointed out some possible issues
> > in the .probe() and .remove() callbacks of the virtio-vsock driver.
> > 
> > This series tries to solve these issues:
> > - Patch 1 postpones the 'the_virtio_vsock' assignment at the end of the
> >   .probe() to avoid that some sockets queue works when the initialization
> >   is not finished.
> > - Patches 2 and 3 stop workers before to call vdev->config->reset(vdev) to
> >   be sure that no one is accessing the device, and adds another flush at the
> >   end of the .remove() to avoid use after free.
> > - Patch 4 free also used buffers in the virtqueues during the .remove().
> > 
> > Stefano Garzarella (4):
> >   vsock/virtio: fix locking around 'the_virtio_vsock'
> >   vsock/virtio: stop workers during the .remove()
> >   vsock/virtio: fix flush of works during the .remove()
> >   vsock/virtio: free used buffers during the .remove()
> > 
> >  net/vmw_vsock/virtio_transport.c | 105 ++++++++++++++++++++++++++-----
> >  1 file changed, 90 insertions(+), 15 deletions(-)
> 
> Looking forward to v2.  I took a look at the discussion and I'll review
> v2 from scratch.  Just keep in mind that the mutex is used more for
> mutual exclusion of the init/exit code than to protect the_virtio_vsock,
> so we'll still need protection of init/exit code even with RCU.

Thanks for the advice! I'll send the v2 ASAP.

Thanks,
Stefano