Re: [dpdk-dev] [RFC v2] ethdev: add flow metadata

[Yongseok Koh <yskoh@mellanox.com>]

> On Aug 20, 2018, at 8:45 AM, Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> 
> On 6/26/2018 9:54 AM, Xueming Li wrote:
>> Currently, rte_flow pattern only match packet header fields.
>> This patch adds additional data to match the packet.
>> 
>> For example, in egress direction, to do an action depending on the VM
>> id, the application needs to configure rte_flow rule with the new
>> metadata pattern:
>>    pattern meta data is {vm} / end action encap …
>> Then the PMD will send VM id as metadata associated in mbuf to NIC,
>> then egress flow on NIC match metadata as other regular packet headers,
>> the appropriate encapsulation is done according to the VM id metadata.
>> 
>> Metadata could be used on ingress as well to save useful info before
>> flow modification (not defined yet) or decapsulation action. PMD is
>> responsible to save metadata into mbuf field. The application must get
>> metadata from the mbuf.
>> 
>> Cc: Thomas Monjalon <thomas@monjalon.net>
>> Cc: Olivier Matz <olivier.matz@6wind.com>
>> Cc: Shahaf Shuler <shahafs@mellanox.com>
>> Signed-off-by: Xueming Li <xuemingl@mellanox.com>
>> ---
>> doc/guides/prog_guide/rte_flow.rst |  7 +++++++
>> lib/librte_ethdev/rte_flow.c       |  1 +
>> lib/librte_ethdev/rte_flow.h       | 28 ++++++++++++++++++++++++++++
>> 3 files changed, 36 insertions(+)
>> 
>> diff --git a/doc/guides/prog_guide/rte_flow.rst b/doc/guides/prog_guide/rte_flow.rst
>> index b305a72a5..7989e5856 100644
>> --- a/doc/guides/prog_guide/rte_flow.rst
>> +++ b/doc/guides/prog_guide/rte_flow.rst
>> @@ -1191,6 +1191,13 @@ Normally preceded by any of:
>> - `Item: ICMP6_ND_NS`_
>> - `Item: ICMP6_ND_OPT`_
>> 
>> +Item: ``META``
>> +^^^^^^^^^^^^^^
>> +
>> +Matches a metadata variable.
>> +
>> +- ``data``: 64 bit value.
>> +
> 
> Isn't this "user data" more than metadata?
> 
> And there is only one userdata can be provided for a flow.
> Can there be a need to provide multiple userdata? Should we support it?
> For your example, on egress what if you want to do specific action on a flow for
> both VM id and process id, you can only pass single user data?
> What about passing a key-value pair?

It could be matter of how device supports metadata. I assume that most of devices
would support one single metadata. Then, we can leave it as user's choice. I mean
user can partition 64bit according to its need. E.g. (key << 32) | value.

Thanks,
Yongseok