[ovs-dev] [PATCHv2] ofproto-dpif-sflow: Add snaplen for sample action and sFlow.

[William Tu]

Hi Ben,

Thanks, I've fixed it and submitted new version.

Regards,
William


On Fri, Jul 1, 2016 at 9:28 PM, Ben Pfaff <blp at ovn.org> wrote:
> On Fri, Jul 01, 2016 at 12:43:02PM -0700, William Tu wrote:
>> This patch adds a 'snaplen' field in sample action (nx_action_sample2).
>> Currently, sample action is used by sFlow and IPFIX. For IPFIX, nothing
>> is changed.  For sFlow configuration, the patch translates header=N to
>> a sample action with snaplen=N, then the snaplen=N translates to trunc(N)
>> in kernel datapath.  Thus, only N bytes instead of full-packet size will
>> be copied from kernel to userspace, saving the copying overhead.
>>
>> Also, the patch parses OVS_PACKET_ATTR_LEN from nlattr to upcall related
>> structures so the sFlow receiver knows the original packet size before it
>> is truncated.
>>
>> Tested-at: https://travis-ci.org/williamtu/ovs-travis/builds/141704344
>> Signed-off-by: William Tu <u9012063 at gmail.com>
>> ---
>> v1->v2
>> http://openvswitch.org/pipermail/dev/2016-June/072404.html
>> - Add snaplen to nx_action_sample2
>> - Since truncate then userspace action is supported in userspace datapath,
>>   ,commit aaca4fe0ce9e (ofp-actions: Add truncate action.), remove
>>   datapath-specific implementation and testcase.
>
> This changes the OpenFlow sample action but it doesn't document the new
> feature in ovs-ofctl(8).
>
> The new features should be mentioned in NEWS.
>
> In encode_SAMPLE(), NXAST_RAW_SAMPLE2 should also be selected if
> nas->snaplen != 0.
>
> In format_SAMPLE(), I would only output snaplen if it is nonzero (and
> not UINT16_MAX?).
>
> The datapath will reject a snaplen less than 14, but I don't see
> anything here that prevents userspace from trying such a snaplen.
>
> Thanks,
>
> Ben.