Re: [PATCH v4 1/3] virtio: add dma-buf support for exported objects

[David Stevens <stevensd@chromium.org>]

On Mon, Jun 8, 2020 at 3:00 PM Michael S. Tsirkin <mst@redhat.com> wrote:
>
> On Mon, Jun 08, 2020 at 10:33:09AM +0900, David Stevens wrote:
> > On Sun, Jun 7, 2020 at 5:04 AM Michael S. Tsirkin <mst@redhat.com> wrote:
> > >
> > > On Fri, Jun 05, 2020 at 10:28:42AM +0900, David Stevens wrote:
> > > > On Fri, Jun 5, 2020 at 4:05 AM Michael S. Tsirkin <mst@redhat.com> wrote:
> > > > >
> > > > > On Tue, May 26, 2020 at 07:58:09PM +0900, David Stevens wrote:
> > > > > > This change adds a new flavor of dma-bufs that can be used by virtio
> > > > > > drivers to share exported objects. A virtio dma-buf can be queried by
> > > > > > virtio drivers to obtain the UUID which identifies the underlying
> > > > > > exported object.
> > > > > >
> > > > > > Signed-off-by: David Stevens <stevensd@chromium.org>
> > > > >
> > > > > Is this just for graphics? If yes I'd rather we put it in the graphics
> > > > > driver. We can always move it later ...
> > > >
> > > > As stated in the cover letter, this will be used by virtio-video.
> > > >
> > > > The proposed virtio-video patches: https://markmail.org/thread/p5d3k566srtdtute
> > > > The patch which imports these dma-bufs (slightly out of data, uses v3
> > > > of this patch set): https://markmail.org/thread/j4xlqaaim266qpks
> > > >
> > > > > > ---
> > > > > >  drivers/virtio/Makefile         |  2 +-
> > > > > >  drivers/virtio/virtio.c         |  6 +++
> > > > > >  drivers/virtio/virtio_dma_buf.c | 89 +++++++++++++++++++++++++++++++++
> > > > > >  include/linux/virtio.h          |  1 +
> > > > > >  include/linux/virtio_dma_buf.h  | 58 +++++++++++++++++++++
> > > > > >  5 files changed, 155 insertions(+), 1 deletion(-)
> > > > > >  create mode 100644 drivers/virtio/virtio_dma_buf.c
> > > > > >  create mode 100644 include/linux/virtio_dma_buf.h
> > > > > >
> > > > > > diff --git a/drivers/virtio/Makefile b/drivers/virtio/Makefile
> > > > > > index 29a1386ecc03..ecdae5b596de 100644
> > > > > > --- a/drivers/virtio/Makefile
> > > > > > +++ b/drivers/virtio/Makefile
> > > > > > @@ -1,5 +1,5 @@
> > > > > >  # SPDX-License-Identifier: GPL-2.0
> > > > > > -obj-$(CONFIG_VIRTIO) += virtio.o virtio_ring.o
> > > > > > +obj-$(CONFIG_VIRTIO) += virtio.o virtio_ring.o virtio_dma_buf.o
> > > > > >  obj-$(CONFIG_VIRTIO_MMIO) += virtio_mmio.o
> > > > > >  obj-$(CONFIG_VIRTIO_PCI) += virtio_pci.o
> > > > > >  virtio_pci-y := virtio_pci_modern.o virtio_pci_common.o
> > > > > > diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c
> > > > > > index a977e32a88f2..5d46f0ded92d 100644
> > > > > > --- a/drivers/virtio/virtio.c
> > > > > > +++ b/drivers/virtio/virtio.c
> > > > > > @@ -357,6 +357,12 @@ int register_virtio_device(struct virtio_device *dev)
> > > > > >  }
> > > > > >  EXPORT_SYMBOL_GPL(register_virtio_device);
> > > > > >
> > > > > > +bool is_virtio_device(struct device *dev)
> > > > > > +{
> > > > > > +     return dev->bus == &virtio_bus;
> > > > > > +}
> > > > > > +EXPORT_SYMBOL_GPL(is_virtio_device);
> > > > > > +
> > > > > >  void unregister_virtio_device(struct virtio_device *dev)
> > > > > >  {
> > > > > >       int index = dev->index; /* save for after device release */
> > > > > > diff --git a/drivers/virtio/virtio_dma_buf.c b/drivers/virtio/virtio_dma_buf.c
> > > > > > new file mode 100644
> > > > > > index 000000000000..23e3399b11ed
> > > > > > --- /dev/null
> > > > > > +++ b/drivers/virtio/virtio_dma_buf.c
> > > > > > @@ -0,0 +1,89 @@
> > > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > > +/*
> > > > > > + * dma-bufs for virtio exported objects
> > > > > > + *
> > > > > > + * Copyright (C) 2020 Google, Inc.
> > > > > > + */
> > > > > > +
> > > > > > +#include <linux/virtio_dma_buf.h>
> > > > > > +
> > > > > > +/**
> > > > > > + * virtio_dma_buf_export - Creates a new dma-buf for a virtio exported object
> > > > > > + *
> > > > > > + * This wraps dma_buf_export() to allow virtio drivers to create a dma-buf
> > > > > > + * for an virtio exported object that can be queried by other virtio drivers
> > > > > > + * for the object's UUID.
> > > > > > + */
> > > > > > +struct dma_buf *virtio_dma_buf_export(
> > > > > > +             const struct virtio_dma_buf_export_info *virtio_exp_info)
> > > > > > +{
> > > > > > +     struct dma_buf_export_info exp_info;
> > > > > > +
> > > > > > +     if (!virtio_exp_info->ops
> > > > > > +             || virtio_exp_info->ops->ops.attach != &virtio_dma_buf_attach
> > > > > > +             || !virtio_exp_info->ops->get_uuid) {
> > > > > > +             return ERR_PTR(-EINVAL);
> > > > > > +     }
> > > > > > +
> > > > > > +     exp_info.exp_name = virtio_exp_info->exp_name;
> > > > > > +     exp_info.owner = virtio_exp_info->owner;
> > > > > > +     exp_info.ops = &virtio_exp_info->ops->ops;
> > > > > > +     exp_info.size = virtio_exp_info->size;
> > > > > > +     exp_info.flags = virtio_exp_info->flags;
> > > > > > +     exp_info.resv = virtio_exp_info->resv;
> > > > > > +     exp_info.priv = virtio_exp_info->priv;
> > > > > > +     BUILD_BUG_ON(sizeof(struct virtio_dma_buf_export_info)
> > > > > > +                  != sizeof(struct dma_buf_export_info));
> > > > >
> > > > > This is the only part that gives me pause. Why do we need this hack?
> > > > > What's wrong with just using dma_buf_export_info directly,
> > > > > and if you want the virtio ops, just using container_off?
> > > >
> > > > This approach provides a more explicit type signature and a little
> > > > more type safety, I think. If others don't think it's a worthwhile
> > > > tradeoff, I can remove it.
> > > >
> > > > -David
> > >
> > > The cost is that if dma_buf_export_info changes even slightly, we get
> > > weird crashes.
> >
> > I'm not sure I understand what types of changes you're referring to.
> > As this is written, virtio-dma-buf is just another client of the
> > dma-buf API. If this were rewritten to use dma-buf directly, then
> > whatever code calls virtio_dma_buf_export would become a client of the
> > dma-buf API. If the semantics of existing fields in the dma-buf API
> > were changed and virtio-dma-buf wasn't updated, then yes, you could
> > get weird crashes from virtio-dma-buf.
> > However, the same problem would
> > exist if virtio_dma_buf_export used dma-buf directly - changes to
> > dma-buf's semantics could cause weird crashes if the caller of
> > virtio_dma_buf_export wasn't updated properly. The only potential
> > source of problems I see is if virtio_dma_buf_export_info wasn't
> > updated properly, but virtio_dma_buf_export_info is dead simple, so I
> > don't know if that's really a problem.
> >
> > -David
>
> I think you can get weird crashes if fields in dma buf are reordered, or
> if a field size changes.  You have a build bug catching overall struct
> size changes but that can remain the same due do compiler padding or
> such.

Since it's manually copying the fields instead of trying something
clever like memcpy, I don't see how reordering the fields or changing
the size of the fields would cause problems. Right now,
virtio_dma_buf_export is just a regular client of dma_buf_export, no
different than any of the other call sites in the kernel.

Overall, I don't really think that this is a problem. If someone makes
breaking changes to the semantics of dma-buf, then they will need to
update this call site, just like they will need to update all of the
other call sites in the kernel. If someone adds new functionality to
dma-buf and adds another field to dma_buf_export_info, the build bug
is a reminder to add it to virtio_dma_buf_export_info. However, if the
struct padding happens to work out such that the build bug doesn't
trigger, that doesn't really matter - it just means that the new
dma-buf feature won't be exposed by virito-dma-buf until someone needs
it and notices that the new field is missing.

-David