2014-11-20 09:03:34

by Jason Wang

Subject: [PATCH net V5] virtio-net: validate features during probe

We currently trigger BUG when VIRTIO_NET_F_CTRL_VQ
is not set but one of features depending on it is.
That's not a friendly way to report errors to
hypervisors.
Let's check, and fail probe instead.

Cc: Rusty Russell <[email protected]>
Cc: Cornelia Huck <[email protected]>
Cc: Wanlong Gao <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
---
Changes from V1:
- Drop NETIF_F_*_UFO from checklist
Changes from V2:
- only check the features for ctrl vq (this fix the real bug)
- better error message and simplify API
Changes from V3:
- pass dbit directly and even better error message
- typo fix
Changes from v4:
- add "device" before "advertise"
---
drivers/net/virtio_net.c | 37 +++++++++++++++++++++++++++++++++++++
1 file changed, 37 insertions(+)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index ec2a8b4..b0bc8ea 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1673,6 +1673,40 @@ static const struct attribute_group virtio_net_mrg_rx_group = {
};
#endif

+static bool virtnet_fail_on_feature(struct virtio_device *vdev,
+ unsigned int fbit,
+ const char *fname, const char *dname)
+{
+ if (!virtio_has_feature(vdev, fbit))
+ return false;
+
+ dev_err(&vdev->dev, "device advertises feature %s but not %s",
+ fname, dname);
+
+ return true;
+}
+
+#define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
+ virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
+
+static bool virtnet_validate_features(struct virtio_device *vdev)
+{
+ if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
+ (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
+ "VIRTIO_NET_F_CTRL_VQ") ||
+ VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
+ "VIRTIO_NET_F_CTRL_VQ") ||
+ VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
+ "VIRTIO_NET_F_CTRL_VQ") ||
+ VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
+ VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
+ "VIRTIO_NET_F_CTRL_VQ"))) {
+ return false;
+ }
+
+ return true;
+}
+
static int virtnet_probe(struct virtio_device *vdev)
{
int i, err;
@@ -1680,6 +1714,9 @@ static int virtnet_probe(struct virtio_device *vdev)
struct virtnet_info *vi;
u16 max_queue_pairs;

+ if (!virtnet_validate_features(vdev))
+ return -EINVAL;
+
/* Find if host supports multiqueue virtio_net device */
err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
struct virtio_net_config,
--
1.9.1

2014-11-20 09:48:39

by Cornelia Huck

Subject: Re: [PATCH net V5] virtio-net: validate features during probe

On Thu, 20 Nov 2014 17:03:05 +0800
Jason Wang <[email protected]> wrote:

> We currently trigger BUG when VIRTIO_NET_F_CTRL_VQ
> is not set but one of features depending on it is.
> That's not a friendly way to report errors to
> hypervisors.
> Let's check, and fail probe instead.
>
> Cc: Rusty Russell <[email protected]>
> Cc: Cornelia Huck <[email protected]>
> Cc: Wanlong Gao <[email protected]>
> Signed-off-by: Michael S. Tsirkin <[email protected]>
> Signed-off-by: Jason Wang <[email protected]>

> +static bool virtnet_validate_features(struct virtio_device *vdev)
> +{
> + if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
> + (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
> + "VIRTIO_NET_F_CTRL_VQ"))) {
> + return false;
> + }
> +
> + return true;
> +}

I still think it may make sense to print all incorrectly set bits, but
let's just fix the BUG() trigger for now.

Acked-by: Cornelia Huck <[email protected]>

2014-11-20 09:53:14

by Jason Wang

Subject: Re: [PATCH net V5] virtio-net: validate features during probe

On 11/20/2014 05:48 PM, Cornelia Huck wrote:
> On Thu, 20 Nov 2014 17:03:05 +0800
> Jason Wang <[email protected]> wrote:
>
>> We currently trigger BUG when VIRTIO_NET_F_CTRL_VQ
>> is not set but one of features depending on it is.
>> That's not a friendly way to report errors to
>> hypervisors.
>> Let's check, and fail probe instead.
>>
>> Cc: Rusty Russell <[email protected]>
>> Cc: Cornelia Huck <[email protected]>
>> Cc: Wanlong Gao <[email protected]>
>> Signed-off-by: Michael S. Tsirkin <[email protected]>
>> Signed-off-by: Jason Wang <[email protected]>
>> +static bool virtnet_validate_features(struct virtio_device *vdev)
>> +{
>> + if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
>> + (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
>> + "VIRTIO_NET_F_CTRL_VQ") ||
>> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
>> + "VIRTIO_NET_F_CTRL_VQ") ||
>> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
>> + "VIRTIO_NET_F_CTRL_VQ") ||
>> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
>> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
>> + "VIRTIO_NET_F_CTRL_VQ"))) {
>> + return false;
>> + }
>> +
>> + return true;
>> +}
> I still think it may make sense to print all incorrectly set bits, but
> let's just fix the BUG() trigger for now.
>
> Acked-by: Cornelia Huck <[email protected]>
>

Right, will do it in the future.

Thanks

2014-11-20 12:45:15

by Michael S. Tsirkin

Subject: Re: [PATCH net V5] virtio-net: validate features during probe

On Thu, Nov 20, 2014 at 05:03:05PM +0800, Jason Wang wrote:
> We currently trigger BUG when VIRTIO_NET_F_CTRL_VQ
> is not set but one of features depending on it is.
> That's not a friendly way to report errors to
> hypervisors.
> Let's check, and fail probe instead.
>
> Cc: Rusty Russell <[email protected]>
> Cc: Cornelia Huck <[email protected]>
> Cc: Wanlong Gao <[email protected]>
> Signed-off-by: Michael S. Tsirkin <[email protected]>
> Signed-off-by: Jason Wang <[email protected]>


Acked-by: Michael S. Tsirkin <[email protected]>

Cc DaveM to pick up the patch.

> ---
> Changes from V1:
> - Drop NETIF_F_*_UFO from checklist
> Changes from V2:
> - only check the features for ctrl vq (this fix the real bug)
> - better error message and simplify API
> Changes from V3:
> - pass dbit directly and even better error message
> - typo fix
> Changes from v4:
> - add "device" before "advertise"
> ---
> drivers/net/virtio_net.c | 37 +++++++++++++++++++++++++++++++++++++
> 1 file changed, 37 insertions(+)
>
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index ec2a8b4..b0bc8ea 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -1673,6 +1673,40 @@ static const struct attribute_group virtio_net_mrg_rx_group = {
> };
> #endif
>
> +static bool virtnet_fail_on_feature(struct virtio_device *vdev,
> + unsigned int fbit,
> + const char *fname, const char *dname)
> +{
> + if (!virtio_has_feature(vdev, fbit))
> + return false;
> +
> + dev_err(&vdev->dev, "device advertises feature %s but not %s",
> + fname, dname);
> +
> + return true;
> +}
> +
> +#define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
> + virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
> +
> +static bool virtnet_validate_features(struct virtio_device *vdev)
> +{
> + if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
> + (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
> + "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
> + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
> + "VIRTIO_NET_F_CTRL_VQ"))) {
> + return false;
> + }
> +
> + return true;
> +}
> +
> static int virtnet_probe(struct virtio_device *vdev)
> {
> int i, err;
> @@ -1680,6 +1714,9 @@ static int virtnet_probe(struct virtio_device *vdev)
> struct virtnet_info *vi;
> u16 max_queue_pairs;
>
> + if (!virtnet_validate_features(vdev))
> + return -EINVAL;
> +
> /* Find if host supports multiqueue virtio_net device */
> err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
> struct virtio_net_config,
> --
> 1.9.1

2014-11-21 05:27:32

by David Miller

Subject: Re: [PATCH net V5] virtio-net: validate features during probe

From: Jason Wang <[email protected]>
Date: Thu, 20 Nov 2014 17:03:05 +0800

> We currently trigger BUG when VIRTIO_NET_F_CTRL_VQ
> is not set but one of features depending on it is.
> That's not a friendly way to report errors to
> hypervisors.
> Let's check, and fail probe instead.
>
> Cc: Rusty Russell <[email protected]>
> Cc: Cornelia Huck <[email protected]>
> Cc: Wanlong Gao <[email protected]>
> Signed-off-by: Michael S. Tsirkin <[email protected]>
> Signed-off-by: Jason Wang <[email protected]>

Applied, thanks.