Re: [PATCH 1/3] net: add MACsec header

[Olivier Matz <olivier.matz@6wind.com>]

Hi Akhil,

Few comments below.

On Mon, Aug 15, 2022 at 12:16:18AM +0530, Akhil Goyal wrote:
> Added MACsec protocol header to be used for supporting
> MACsec protocol offload in hardware or directly in the application.
> 
> Signed-off-by: Akhil Goyal <gakhil@marvell.com>
> ---
>  doc/api/doxy-api-index.md |  3 ++-
>  lib/net/meson.build       |  1 +
>  lib/net/rte_macsec.h      | 56 +++++++++++++++++++++++++++++++++++++++
>  3 files changed, 59 insertions(+), 1 deletion(-)
>  create mode 100644 lib/net/rte_macsec.h
> 
> diff --git a/doc/api/doxy-api-index.md b/doc/api/doxy-api-index.md
> index 186a258be4..99e49340d3 100644
> --- a/doc/api/doxy-api-index.md
> +++ b/doc/api/doxy-api-index.md
> @@ -126,7 +126,8 @@ The public API headers are grouped by topics:
>    [Geneve](@ref rte_geneve.h),
>    [eCPRI](@ref rte_ecpri.h),
>    [L2TPv2](@ref rte_l2tpv2.h),
> -  [PPP](@ref rte_ppp.h)
> +  [PPP](@ref rte_ppp.h),
> +  [MACsec](@ref rte_macsec.h)
>  
>  - **QoS**:
>    [metering](@ref rte_meter.h),
> diff --git a/lib/net/meson.build b/lib/net/meson.build
> index e899846578..3e63abaca8 100644
> --- a/lib/net/meson.build
> +++ b/lib/net/meson.build
> @@ -21,6 +21,7 @@ headers = files(
>          'rte_geneve.h',
>          'rte_l2tpv2.h',
>          'rte_ppp.h',
> +        'rte_macsec.h',
>  )
>  
>  sources = files(
> diff --git a/lib/net/rte_macsec.h b/lib/net/rte_macsec.h
> new file mode 100644
> index 0000000000..f1b59253f6
> --- /dev/null
> +++ b/lib/net/rte_macsec.h
> @@ -0,0 +1,56 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(C) 2022 Marvell.
> + */
> +
> +#ifndef _RTE_MACSEC_H_
> +#define _RTE_MACSEC_H_
> +
> +/**
> + * @file
> + *
> + * MACsec-related defines
> + */
> +
> +#include <rte_byteorder.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +
> +/* SecTAG length = macsec ether header without the optional SCI */
> +#define RTE_MACSEC_TAG_LEN 6

Use a doxygen-like comment.

Is this define required? In my understanding, it is the same as
sizeof(struct rte_macsec_hdr).

> +#define RTE_MACSEC_SCI_LEN 8

Missing doxygen doc.

> +
> +#define RTE_MACSEC_TCI_VERSION	0x80 /**< Version mask for MACsec. Should be 0. */
> +#define RTE_MACSEC_TCI_ES	0x40 /**< End station - SCI is not valid */
> +#define RTE_MACSEC_TCI_SC	0x20 /**< SCI present */
> +#define RTE_MACSEC_TCI_SCB	0x10 /**< Secure channel support EPON single copy broadcast */

support -> supports?

> +#define RTE_MACSEC_TCI_E	0x08 /**< User data is encrypted */
> +#define RTE_MACSEC_TCI_C	0x04 /**< User data was changed (because of encryption) */


In [1], I can read the following, which is not similar:

E and C bits used to determine if packet is encrypted
• E, C = 1, 1 – Encrypted
• E, C = 0, 0 – Authenticated-Only

Is there any reference paper for macsec header?

[1] https://www.marvell.com/content/dam/marvell/en/public-collateral/automotive-solutions/marvell-macsec-security-in-ethernet-based-vehicle-white-paper.pdf


> +#define RTE_MACSEC_AN_MASK	0x03 /**< Association number mask in tci_an */

nit: all comments should end with a dot.


>
> +#define RTE_MACSEC_NUM_AN	4    /**< 2 bits for the association number */

I don't get how this defined is used. Can the comment be clarified?

> +#define RTE_MACSEC_SALT_LEN	12   /**< Salt length for MACsec SA */

Same here.

> +
> +/**
> + * MACsec Header
> + */
> +struct rte_macsec_hdr {
> +	/* SecTAG */

Is the SecTAG comment required? Or maybe it should be moved
above the struct?

> +	uint8_t  tci_an;	/**< Tag control information and Association number of SC */

nit: duplicated spaces after uint8_t

Can we use a bitfield here for tci and an, like you did for short_length?

Like this:

	uint8_t tci:6;
	uint8_t an:2;

Or:

	uint8_t tci_version:1;
	uint8_t tci_es:1;
	uint8_t tci_sc:1;
	uint8_t tci_scb:1;
	uint8_t tci_e:1;
	uint8_t tci_c:1;
	uint8_t an:2;

I think the 2nd one is easier to use.

> +#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
> +	uint8_t short_length : 6; /**< Short Length */
> +	uint8_t unused : 2;

nit: no spaces around ':'

> +#elif RTE_BYTE_ORDER == RTE_BIG_ENDIAN
> +	uint8_t unused : 2;
> +	uint8_t short_length : 6;
> +#endif
> +	rte_be32_t packet_number; /**< Packet number to support replay protection */
> +	uint8_t secure_channel_id[8]; /* optional */

8 -> RTE_MACSEC_SCI_LEN ?

I think it would be more convenient to have another struct
for the secure_channel_id.

For instance, this pseudo code:

	struct struct rte_macsec_hdr *hdr = NULL;
	struct struct rte_macsec_sci_hdr *hdr_sci = NULL;

	if (seg_len(mbuf) < sizeof(*hdr))
		return -1;
	if (hdr.tci_sc) {
		if (seg_len(mbuf) < sizeof(*hdr_sci))
			return -1;
		hdr_sci = hdr;
	}

With only one struct, it is difficult to properly do the length check.


> +} __rte_packed;
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif /* RTE_MACSEC_H_ */
> -- 
> 2.25.1
> 

Thanks,
Olivier