Re: [dpdk-dev] [PATCH] ethdev: add VLAN attributes to ETH and VLAN items

[Maxime Leroy <maxime.leroy@6wind.com>]

Hi Dekel,

On Thu, Oct 1, 2020 at 8:49 PM Dekel Peled <dekelp@nvidia.com> wrote:
>
> From: Dekel Peled <dekelp@mellanox.com>
>
> This patch implements the change proposes in RFC [1], adding dedicated
> fields to ETH and VLAN items structs, to clearly define the required
> characteristic of a packet, and enable precise match criteria.
>
> [1] https://mails.dpdk.org/archives/dev/2020-August/177536.html
>
> Signed-off-by: Dekel Peled <dekelp@mellanox.com>
> ---
>  doc/guides/rel_notes/release_20_11.rst |  7 +++++++
>  lib/librte_ethdev/rte_flow.h           | 16 +++++++++++++---
>  2 files changed, 20 insertions(+), 3 deletions(-)
>
> diff --git a/doc/guides/rel_notes/release_20_11.rst b/doc/guides/rel_notes/release_20_11.rst
> index 7f9d0dd..199c60b 100644
> --- a/doc/guides/rel_notes/release_20_11.rst
> +++ b/doc/guides/rel_notes/release_20_11.rst
> @@ -173,6 +173,13 @@ API Changes
>    * ``_rte_eth_dev_callback_process()`` -> ``rte_eth_dev_callback_process()``
>    * ``_rte_eth_dev_reset`` -> ``rte_eth_dev_internal_reset()``
>
> +* ethdev: Added new field ``vlan_exist`` to structure ``rte_flow_item_eth``,
> +  indicating that at least one VLAN exists in the packet header.
> +
> +* ethdev: Added new field ``more_vlans_exist`` to structure
> +  ``rte_flow_item_vlan``, indicating that at least one more VLAN exists in
> +  packet header, following this VLAN.
> +
>  * rawdev: Added a structure size parameter to the functions
>    ``rte_rawdev_queue_setup()``, ``rte_rawdev_queue_conf_get()``,
>    ``rte_rawdev_info_get()`` and ``rte_rawdev_configure()``,
> diff --git a/lib/librte_ethdev/rte_flow.h b/lib/librte_ethdev/rte_flow.h
> index da8bfa5..39d04ef 100644
> --- a/lib/librte_ethdev/rte_flow.h
> +++ b/lib/librte_ethdev/rte_flow.h
> @@ -723,14 +723,18 @@ struct rte_flow_item_raw {
>   * If the @p type field contains a TPID value, then only tagged packets with the
>   * specified TPID will match the pattern.
>   * Otherwise, only untagged packets will match the pattern.
> - * If the @p ETH item is the only item in the pattern, and the @p type field
> - * is not specified, then both tagged and untagged packets will match the
> - * pattern.
> + * The field @p vlan_exist can be used to match specific packet types, instead
> + * of using the @p type field.
> + * This can be used to match any type of tagged packets.
> + * If the @p type and @p vlan_exist fields are not specified, then both tagged
> + * and untagged packets will match the pattern.
>   */
>  struct rte_flow_item_eth {
>         struct rte_ether_addr dst; /**< Destination MAC. */
>         struct rte_ether_addr src; /**< Source MAC. */
>         rte_be16_t type; /**< EtherType or TPID. */
> +       uint32_t vlan_exist:1; /**< At least one VLAN exist in header. */
> +       uint32_t reserved:31; /**< Reserved, must be zero. */
>  };

To resume:
- type and vlan_exists fields not specified:  tag and untagged matched
- with vlan_exists, match only tag or untagged
- with type matching specific ethernet type
- vlan_exists and type should not setted at the same time ?

With this new specification, I think you address all the use cases.
That's great !

>
>  /** Default mask for RTE_FLOW_ITEM_TYPE_ETH. */
> @@ -752,10 +756,16 @@ struct rte_flow_item_eth {
>   * the preceding pattern item.
>   * If a @p VLAN item is present in the pattern, then only tagged packets will
>   * match the pattern.
> + * The field @p more_vlans_exist can be used to match specific packet types,
> + * instead of using the @p inner_type field.
> + * This can be used to match any type of tagged packets.
>   */

Could you please specify what the expected behavior when inner_type
and more_vlans_exist are not specified .
What is the default behavior ?

>  struct rte_flow_item_vlan {
>         rte_be16_t tci; /**< Tag control information. */
>         rte_be16_t inner_type; /**< Inner EtherType or TPID. */
> +       uint32_t more_vlans_exist:1;
> +       /**< At least one more VLAN exist in header, following this VLAN. */
> +       uint32_t reserved:31; /**< Reserved, must be zero. */
>  };
>
>  /** Default mask for RTE_FLOW_ITEM_TYPE_VLAN. */
> --
> 1.8.3.1
>

I am still wondering, why not using a new item 'NOT' for example to
match only eth packet not tagged ?
example: eth / not vlan. It's a more generic solution.

Here in this commit, we add a reference on VLAN fields on ethernet header.
But tomorrow, we could do the same for mpls by adding mpls_exists in
the eth item and so on.

In fact, we  have the same needs for IPv6 options. To match for
example, ipv6 packet with no fragment option.
With a NOT field, it can be easily done: > eth / ipv6 / no ipv6_frag.

Adding new fields 'item'_exists into eth and ipv6 do the jobs, but
having a NOT attribute is a more generic solution.

It could address many other use cases like matching any udp packets
that are not vxlan ( eth / ipv4 / vxlan / not udp),

Let me know what you think about that.

Regards,

Maxime