[ovs-dev] [PATCH 14/14] datapath-windows: Fix potential crash in OvsInitConfiguredSwitchNics: virtPort

Samuel Ghinet sghinet at cloudbasesolutions.com
Tue Sep 30 16:16:06 UTC 2014


Nithin,

basically, any userspace code that used flows was commented out. And code that was trying to set qos to the netdev was commented out (because it was retrying forever).
And netlink functions that used rtnetlink (were used for qos and netdev, if I am not mistaken) were commented out, so they do not get called.

So in principle, the functions in userspace that call into the kernel to add port, set port and delete port were reached, and thus were tested.
I used only one bridge (in patch 00 I explained about problems with multiple bridges / multiple internal ports) - basically, I used only one, because otherwise it periodically re-tries the creation of the failed internal vport.
I added manually the VM ports (used the WMI script as well). The "internal" port is created automatically by the userspace.
I had tested the creation and destruction only. I.e. I did not test vxlan tunneling itself (I think this would be impossible at  this stage), only that the vports are created and destroyed as they should.

Regards,
Sam
________________________________________
From: Alin Serdean
Sent: Tuesday, September 30, 2014 6:23 PM
To: Nithin Raju; Samuel Ghinet
Cc: dev at openvswitch.org; Eitan Eliahu; Ankur Sharma
Subject: RE: [PATCH 14/14] datapath-windows: Fix potential crash in OvsInitConfiguredSwitchNics: virtPort

Hi Nithin,

We used a modified version of netdev-windows.

Alin.

-----Mesaj original-----
De la: Nithin Raju [mailto:nithin at vmware.com]
Trimis: Tuesday, September 30, 2014 6:09 PM
Către: Samuel Ghinet
Cc: dev at openvswitch.org; Alin Serdean; Eitan Eliahu; Ankur Sharma
Subiect: Re: [PATCH 14/14] datapath-windows: Fix potential crash in OvsInitConfiguredSwitchNics: virtPort


On Sep 30, 2014, at 8:03 AM, Samuel Ghinet <sghinet at cloudbasesolutions.com>
 wrote:

> If the hyper-v switch port type is external, then the function
> OvsInitConfiguredSwitchNics allocates a vport, and if allocation
> succeeds, it procedes with the initialization. However, at this point,
> virtPort may happen to be null, but check against virtPort was not
> made - this means that if virtPort was null, later on its fields would
> try to be accessed.
>
> This patch adds a check for virtPort as well, so that the fields of
> virtPort will not be accessed if virtPort is NULL.
>
> Signed-off-by: Samuel Ghinet <sghinet at cloudbasesolutions.com>

hi Samuel,
Thanks for the patch series. Can you pls. add some notes about validation performed. Did you make any userspace changes?

thanks,
Nithin



More information about the dev mailing list