diff mbox

[dpdk-dev] net/virtio-user: fix cannot get initialized

Message ID 1491551860-15005-1-git-send-email-jianfeng.tan@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Yuanhan Liu
Headers

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation fail Compilation issues

Commit Message

Jianfeng Tan April 7, 2017, 7:57 a.m. UTC 
  After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
by default. However, virtio-user vtpci does not support to get
MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
Plus, struct virtio_net_config is defined as a uninitialized
variable, and could be different values in
virtio_negotiate_features() and virtio_init_device().

In some cases, it passes the check in virtio_negotiate_features()
but fails the check in virtio_init_device(). As a result,
virtio-user canno be initialized.

To fix it, (1) accessing uninitialized variable is not a good
practice, so initialize it as zero; (2) explicitly disable MTU
feature in virtio-user.

Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>

Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
---
 drivers/net/virtio/virtio_ethdev.c               | 4 ++--
 drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
 2 files changed, 8 insertions(+), 2 deletions(-)

Comments

Yuanhan Liu April 13, 2017, 1:58 a.m. UTC | #1 
On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
> After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
> by default. However, virtio-user vtpci does not support to get
> MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
> Plus, struct virtio_net_config is defined as a uninitialized
> variable, and could be different values in
> virtio_negotiate_features() and virtio_init_device().
> 
> In some cases, it passes the check in virtio_negotiate_features()
> but fails the check in virtio_init_device(). As a result,
> virtio-user canno be initialized.
> 
> To fix it, (1) accessing uninitialized variable is not a good
> practice, so initialize it as zero; (2) explicitly disable MTU
> feature in virtio-user.
> 
> Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> 
> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> ---
>  drivers/net/virtio/virtio_ethdev.c               | 4 ++--
>  drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
>  2 files changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> index 78cb3e8..4c43784 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -1163,7 +1163,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
>  
>  	/* If supported, ensure MTU value is valid before acknowledging it. */
>  	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
> -		struct virtio_net_config config;
> +		struct virtio_net_config config = {0};

virtio-user does not support the MTU feature, this patch should not be
reached. The virtio-user feature negotiation should be broken.

	--yliu

>  
>  		vtpci_read_dev_config(hw,
>  			offsetof(struct virtio_net_config, mtu),
> @@ -1332,7 +1332,7 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
>  {
>  	struct virtio_hw *hw = eth_dev->data->dev_private;
>  	struct virtio_net_config *config;
> -	struct virtio_net_config local_config;
> +	struct virtio_net_config local_config = {0};
>  	struct rte_pci_device *pci_dev = NULL;
>  	int ret;
>  
> diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> index 6871cd4..529b3d7 100644
> --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> @@ -362,6 +362,12 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
>  	/* The backend will not report this feature, we add it explicitly */
>  	dev->device_features |= (1ull << VIRTIO_NET_F_STATUS);
>  
> +	/* TODO: VIRTIO_NET_F_MTU is for QEMU to advertise MTU to both frontend
> +	 * and backend driver. For virtio-user, disable it for now, until we
> +	 * have a parameter to specify the MTU.
> +	 */
> +	dev->device_features &= ~(1ull << VIRTIO_NET_F_MTU);
> +
>  	return 0;
>  }
>  
> -- 
> 2.7.4
Jianfeng Tan April 13, 2017, 2:18 a.m. UTC | #2 
On 4/13/2017 9:58 AM, Yuanhan Liu wrote:
> On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
>> After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
>> by default. However, virtio-user vtpci does not support to get
>> MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
>> Plus, struct virtio_net_config is defined as a uninitialized
>> variable, and could be different values in
>> virtio_negotiate_features() and virtio_init_device().
>>
>> In some cases, it passes the check in virtio_negotiate_features()
>> but fails the check in virtio_init_device(). As a result,
>> virtio-user canno be initialized.
>>
>> To fix it, (1) accessing uninitialized variable is not a good
>> practice, so initialize it as zero; (2) explicitly disable MTU
>> feature in virtio-user.
>>
>> Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
>> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
>>
>> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
>> ---
>>   drivers/net/virtio/virtio_ethdev.c               | 4 ++--
>>   drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
>>   2 files changed, 8 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>> index 78cb3e8..4c43784 100644
>> --- a/drivers/net/virtio/virtio_ethdev.c
>> +++ b/drivers/net/virtio/virtio_ethdev.c
>> @@ -1163,7 +1163,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
>>   
>>   	/* If supported, ensure MTU value is valid before acknowledging it. */
>>   	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
>> -		struct virtio_net_config config;
>> +		struct virtio_net_config config = {0};
> virtio-user does not support the MTU feature, this patch should not be
> reached. The virtio-user feature negotiation should be broken.
>
> 	--yliu
>

Yes, it will not come to here anyway. But some static code analyzer 
might report this as an error: there's chance to read uninitialized 
variable.

Thanks,
Jianfeng
Yuanhan Liu April 13, 2017, 2:21 a.m. UTC | #3 
On Thu, Apr 13, 2017 at 10:18:43AM +0800, Tan, Jianfeng wrote:
> 
> 
> On 4/13/2017 9:58 AM, Yuanhan Liu wrote:
> >On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
> >>After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
> >>by default. However, virtio-user vtpci does not support to get
> >>MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
> >>Plus, struct virtio_net_config is defined as a uninitialized
> >>variable, and could be different values in
> >>virtio_negotiate_features() and virtio_init_device().
> >>
> >>In some cases, it passes the check in virtio_negotiate_features()
> >>but fails the check in virtio_init_device(). As a result,
> >>virtio-user canno be initialized.
> >>
> >>To fix it, (1) accessing uninitialized variable is not a good
> >>practice, so initialize it as zero; (2) explicitly disable MTU
> >>feature in virtio-user.
> >>
> >>Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
> >>Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> >>Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> >>
> >>Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> >>---
> >>  drivers/net/virtio/virtio_ethdev.c               | 4 ++--
> >>  drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
> >>  2 files changed, 8 insertions(+), 2 deletions(-)
> >>
> >>diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> >>index 78cb3e8..4c43784 100644
> >>--- a/drivers/net/virtio/virtio_ethdev.c
> >>+++ b/drivers/net/virtio/virtio_ethdev.c
> >>@@ -1163,7 +1163,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
> >>  	/* If supported, ensure MTU value is valid before acknowledging it. */
> >>  	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
> >>-		struct virtio_net_config config;
> >>+		struct virtio_net_config config = {0};
> >virtio-user does not support the MTU feature, this patch should not be
> >reached. The virtio-user feature negotiation should be broken.
> >
> >	--yliu
> >
> 
> Yes, it will not come to here anyway. But some static code analyzer might
> report this as an error: there's chance to read uninitialized variable.

If so, fix it in another patch, with the detailed (and real) errors in
the commit log.

	--yliu
Yuanhan Liu April 13, 2017, 2:36 a.m. UTC | #4 
On Thu, Apr 13, 2017 at 10:18:43AM +0800, Tan, Jianfeng wrote:
> 
> 
> On 4/13/2017 9:58 AM, Yuanhan Liu wrote:
> >On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
> >>After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
> >>by default.

Actually, that's my point. The vhost lib just claim he supports MTU
feature and MTU is not enabled by the virtio-user device.  Meaning,
the MTU feature bit should not be set after the negoitation.

You were seeing the bug because you found the bit is set, right?
If so, that means the virtio-user feature negotiation is broken.


	--yliu

> However, virtio-user vtpci does not support to get
> >>MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
> >>Plus, struct virtio_net_config is defined as a uninitialized
> >>variable, and could be different values in
> >>virtio_negotiate_features() and virtio_init_device().
> >>
> >>In some cases, it passes the check in virtio_negotiate_features()
> >>but fails the check in virtio_init_device(). As a result,
> >>virtio-user canno be initialized.
> >>
> >>To fix it, (1) accessing uninitialized variable is not a good
> >>practice, so initialize it as zero; (2) explicitly disable MTU
> >>feature in virtio-user.
> >>
> >>Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
> >>Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> >>Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> >>
> >>Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
> >>---
> >>  drivers/net/virtio/virtio_ethdev.c               | 4 ++--
> >>  drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
> >>  2 files changed, 8 insertions(+), 2 deletions(-)
> >>
> >>diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> >>index 78cb3e8..4c43784 100644
> >>--- a/drivers/net/virtio/virtio_ethdev.c
> >>+++ b/drivers/net/virtio/virtio_ethdev.c
> >>@@ -1163,7 +1163,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
> >>  	/* If supported, ensure MTU value is valid before acknowledging it. */
> >>  	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
> >>-		struct virtio_net_config config;
> >>+		struct virtio_net_config config = {0};
> >virtio-user does not support the MTU feature, this patch should not be
> >reached. The virtio-user feature negotiation should be broken.
> >
> >	--yliu
> >
> 
> Yes, it will not come to here anyway. But some static code analyzer might
> report this as an error: there's chance to read uninitialized variable.
> 
> Thanks,
> Jianfeng
Jianfeng Tan April 13, 2017, 7:37 a.m. UTC | #5 
On 4/13/2017 10:21 AM, Yuanhan Liu wrote:
> On Thu, Apr 13, 2017 at 10:18:43AM +0800, Tan, Jianfeng wrote:
>>
>> On 4/13/2017 9:58 AM, Yuanhan Liu wrote:
>>> On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
>>>> After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
>>>> by default. However, virtio-user vtpci does not support to get
>>>> MTU from device yet, i.e., vtpci_read_dev_config(MTU) fails.
>>>> Plus, struct virtio_net_config is defined as a uninitialized
>>>> variable, and could be different values in
>>>> virtio_negotiate_features() and virtio_init_device().
>>>>
>>>> In some cases, it passes the check in virtio_negotiate_features()
>>>> but fails the check in virtio_init_device(). As a result,
>>>> virtio-user canno be initialized.
>>>>
>>>> To fix it, (1) accessing uninitialized variable is not a good
>>>> practice, so initialize it as zero; (2) explicitly disable MTU
>>>> feature in virtio-user.
>>>>
>>>> Fixes: 49d26d9e3f47 ("net/virtio: support MTU feature")
>>>> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
>>>> Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
>>>>
>>>> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
>>>> ---
>>>>   drivers/net/virtio/virtio_ethdev.c               | 4 ++--
>>>>   drivers/net/virtio/virtio_user/virtio_user_dev.c | 6 ++++++
>>>>   2 files changed, 8 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>>>> index 78cb3e8..4c43784 100644
>>>> --- a/drivers/net/virtio/virtio_ethdev.c
>>>> +++ b/drivers/net/virtio/virtio_ethdev.c
>>>> @@ -1163,7 +1163,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
>>>>   	/* If supported, ensure MTU value is valid before acknowledging it. */
>>>>   	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
>>>> -		struct virtio_net_config config;
>>>> +		struct virtio_net_config config = {0};
>>> virtio-user does not support the MTU feature, this patch should not be
>>> reached. The virtio-user feature negotiation should be broken.
>>>
>>> 	--yliu
>>>
>> Yes, it will not come to here anyway. But some static code analyzer might
>> report this as an error: there's chance to read uninitialized variable.
> If so, fix it in another patch, with the detailed (and real) errors in
> the commit log.
>

OK, I'll drop this fix from this patch.
Jianfeng Tan April 13, 2017, 7:39 a.m. UTC | #6 
On 4/13/2017 10:36 AM, Yuanhan Liu wrote:
> On Thu, Apr 13, 2017 at 10:18:43AM +0800, Tan, Jianfeng wrote:
>>
>> On 4/13/2017 9:58 AM, Yuanhan Liu wrote:
>>> On Fri, Apr 07, 2017 at 07:57:40AM +0000, Jianfeng Tan wrote:
>>>> After the introduction of vhost MTU, VIRTIO_NET_F_MTU is enabled
>>>> by default.
> Actually, that's my point. The vhost lib just claim he supports MTU
> feature and MTU is not enabled by the virtio-user device.  Meaning,
> the MTU feature bit should not be set after the negoitation.
>
> You were seeing the bug because you found the bit is set, right?
> If so, that means the virtio-user feature negotiation is broken.

Make sense. Then I'll use a macro to define all supported features by 
virtio-user as a filter, so that any new features will be disabled in 
virtio-user.

Thanks,
Jianfeng
diff mbox

Patch

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index 78cb3e8..4c43784 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1163,7 +1163,7 @@  virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
 
 	/* If supported, ensure MTU value is valid before acknowledging it. */
 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
-		struct virtio_net_config config;
+		struct virtio_net_config config = {0};
 
 		vtpci_read_dev_config(hw,
 			offsetof(struct virtio_net_config, mtu),
@@ -1332,7 +1332,7 @@  virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
 {
 	struct virtio_hw *hw = eth_dev->data->dev_private;
 	struct virtio_net_config *config;
-	struct virtio_net_config local_config;
+	struct virtio_net_config local_config = {0};
 	struct rte_pci_device *pci_dev = NULL;
 	int ret;
 
diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c b/drivers/net/virtio/virtio_user/virtio_user_dev.c
index 6871cd4..529b3d7 100644
--- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
+++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
@@ -362,6 +362,12 @@  virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
 	/* The backend will not report this feature, we add it explicitly */
 	dev->device_features |= (1ull << VIRTIO_NET_F_STATUS);
 
+	/* TODO: VIRTIO_NET_F_MTU is for QEMU to advertise MTU to both frontend
+	 * and backend driver. For virtio-user, disable it for now, until we
+	 * have a parameter to specify the MTU.
+	 */
+	dev->device_features &= ~(1ull << VIRTIO_NET_F_MTU);
+
 	return 0;
 }