[PATCH v1 6/6] virtio_ring: Prevent bounds check bypass on descriptor index

Alexander Shishkin posted 6 patches 2 years, 7 months ago
[PATCH v1 6/6] virtio_ring: Prevent bounds check bypass on descriptor index
Posted by Alexander Shishkin 2 years, 7 months ago
The descriptor index in virtqueue_get_buf_ctx_split() comes from the
device/VMM.a Use array_index_nospec() to prevent the CPU from speculating
beyond the descriptor array bounds and providing a primitive for building
a side channel.

Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
---
 drivers/virtio/virtio_ring.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 2e7689bb933b..c42d070ab68d 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -9,6 +9,7 @@
 #include <linux/device.h>
 #include <linux/slab.h>
 #include <linux/module.h>
+#include <linux/nospec.h>
 #include <linux/hrtimer.h>
 #include <linux/dma-mapping.h>
 #include <linux/kmsan.h>
@@ -819,6 +820,8 @@ static void *virtqueue_get_buf_ctx_split(struct virtqueue *_vq,
 		BAD_RING(vq, "id %u out of range\n", i);
 		return NULL;
 	}
+
+	i = array_index_nospec(i, vq->split.vring.num);
 	if (unlikely(!vq->split.desc_state[i].data)) {
 		BAD_RING(vq, "id %u is not a head!\n", i);
 		return NULL;
-- 
2.39.0
Re: [PATCH v1 6/6] virtio_ring: Prevent bounds check bypass on descriptor index
Posted by Michael S. Tsirkin 2 years, 7 months ago
On Thu, Jan 19, 2023 at 03:57:21PM +0200, Alexander Shishkin wrote:
> The descriptor index in virtqueue_get_buf_ctx_split() comes from the
> device/VMM.a Use array_index_nospec() to prevent the CPU from speculating
> beyond the descriptor array bounds and providing a primitive for building
> a side channel.
> 
> Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
> ---
>  drivers/virtio/virtio_ring.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> index 2e7689bb933b..c42d070ab68d 100644
> --- a/drivers/virtio/virtio_ring.c
> +++ b/drivers/virtio/virtio_ring.c
> @@ -9,6 +9,7 @@
>  #include <linux/device.h>
>  #include <linux/slab.h>
>  #include <linux/module.h>
> +#include <linux/nospec.h>
>  #include <linux/hrtimer.h>
>  #include <linux/dma-mapping.h>
>  #include <linux/kmsan.h>
> @@ -819,6 +820,8 @@ static void *virtqueue_get_buf_ctx_split(struct virtqueue *_vq,
>  		BAD_RING(vq, "id %u out of range\n", i);
>  		return NULL;
>  	}
> +
> +	i = array_index_nospec(i, vq->split.vring.num);

I suspect plain
	 i &= split.vring.num - 1
is more efficient.

We know num is a power of two but compiler doesn't.
And pls add a comment explaining what's going on.

>  	if (unlikely(!vq->split.desc_state[i].data)) {
>  		BAD_RING(vq, "id %u is not a head!\n", i);
>  		return NULL;
> -- 
> 2.39.0