[ovs-dev] [PATCH] linux: Signal datapath that unaligned Netlink message can be received

[Thomas Graf]

On 11/15/2013 10:32 AM, Jesse Gross wrote:
> On Wed, Nov 13, 2013 at 5:46 PM, Thomas Graf <tgraf at redhat.com> wrote:
>> On 11/13/2013 07:11 AM, Jesse Gross wrote:
>>>
>>> On Mon, Nov 11, 2013 at 11:53 PM, Thomas Graf <tgraf at redhat.com> wrote:
>>>>
>>>> On 11/11/2013 04:50 PM, Ben Pfaff wrote:
>>>>>
>>>>>
>>>>> On Mon, Nov 11, 2013 at 04:36:24PM +0100, Thomas Graf wrote:
>>>>>>
>>>>>>
>>>>>> Following commit (''netlink: Do not enforce alignment of last Netlink
>>>>>> attribute''), signal the ability to receive unaligned Netlink messages
>>>>>> to the datapath to enable utilization of zerocopy optimizations.
>>>>>>
>>>>>> Signed-off-by: Thomas Graf <tgraf at redhat.com>
>>>>>
>>>>>
>>>>>
>>>>> Seems OK from a userspace point of view.  I am a little concerned that
>>>>> downgrading userspace without deleting and re-creating the datapath
>>>>> (e.g. via "force-reload-kmod") will result in a totally broken setup
>>>>> since userspace will then drop every packet from the kernel.
>>>>
>>>>
>>>>
>>>> Is that something that occurs occasionally in installations? Utilizing
>>>> the version field in the genl header could be used to track this and
>>>> clear user_features.
>>>
>>>
>>> It's probably a good idea. I could see us having more of these
>>> features flags in the future (although obviously we should try to
>>> avoid them if possible) and, as Ben said, it would potentially lead to
>>> a bad state otherwise.
>>>
>>> I'm not sure exactly what you have in mind though, can you elaborate a
>>> little?
>>
>>
>> My initial thought was to use a version field to notice the replacement
>> of user space. On second thought that is not required, modifying user
>> space to provide the user features in OVS_DP_CMD_GET as well will
>> overwrite the features. Resetting user_features to 0 if not features are
>> provided will provide backwards compatibility to versions not aware of
>> user features yet. Thoughts?
>
> Hmm, it doesn't really seem ideal to make DP_CMD_GET change settings
> as it's probably not the expected behavior for most users of the
> command. One example of where this could be a problem is if it is
> called from both ovs-dpctl and ovs-vswitch. Usually these would be
> have the same capabilities but I'm not sure that it's strictly
> required in all cases. DP_CMD_SET seems like the ideal place to put it
> but I guess we don't use that at all on existing versions of OVS
> userspace.

Agreed. Ideal and clean is a DP_CMD_SET lookup-or-create that updates
the user features. That will leave one OVS version to be non backwards
compatible in terms of downgrading user features.

For that reason we can bump OVS_DATAPATH_VERSION to version 2 and reset
the user features if we receive a DP_CMD_GET version 1.

It's not a problem if we reset the user features unnecessarily in some
corner cases such as mismatching ovs-dpctl and vswitchd. The next
iteration of the patchset has support for memory mapped netlink i/o
which would still apply.

If you agree to this then I will cook this up.