[dpdk-dev] [PATCH v2] net/mlx5: zero out UDP csum for IPv6 encap headers

[dpdk-dev] [PATCH v2] net/mlx5: zero out UDP csum for IPv6 encap headers

[Viacheslav Ovsiienko]

From: Eli Britstein <elibr@mellanox.com>

Mellanox NICs do not support UDP checksum hardware tx offload over IPv6.
This limitation becomes critical for UDP based tunnels like VXLAN.
Beside the UDP checksum validity is required by IPv6 there is an option
in Linux to allow accepting UDP zero sum (see udp6zerocsumrx in iproute2
package).

This patch zeroes out the UDP checksum field for encapsulation headers
in raw encap action.

Signed-off-by: Eli Britstein <elibr@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
---
 drivers/net/mlx5/mlx5_flow_dv.c | 46 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 46 insertions(+)

diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 933ad0b..0e6d17d 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -1362,6 +1362,50 @@ struct field_modify_info modify_tcp[] = {
 	return 0;
 }
 
+static int
+flow_dv_zero_encap_udp_csum(void *data, struct rte_flow_error *error)
+{
+	struct rte_ether_hdr *eth = NULL;
+	struct rte_vlan_hdr *vlan = NULL;
+	struct rte_ipv6_hdr *ipv6 = NULL;
+	struct rte_udp_hdr *udp = NULL;
+	char *next_hdr;
+	uint16_t proto;
+
+	eth = (struct rte_ether_hdr *)data;
+	next_hdr = (char *)(eth + 1);
+	proto = RTE_BE16(eth->ether_type);
+
+	/* VLAN skipping */
+	while (proto == RTE_ETHER_TYPE_VLAN || proto == RTE_ETHER_TYPE_QINQ) {
+		next_hdr += sizeof(struct rte_vlan_hdr);
+		vlan = (struct rte_vlan_hdr *)next_hdr;
+		proto = RTE_BE16(vlan->eth_proto);
+	}
+
+	/* HW calculates IPv4 csum. no need to proceed */
+	if (proto == RTE_ETHER_TYPE_IPV4)
+		return 0;
+
+	/* non IPv4/IPv6 header. not supported */
+	if (proto != RTE_ETHER_TYPE_IPV6) {
+		return rte_flow_error_set(error, ENOTSUP,
+					  RTE_FLOW_ERROR_TYPE_ACTION,
+					  NULL, "Cannot offload non IPv4/IPv6");
+	}
+
+	ipv6 = (struct rte_ipv6_hdr *)next_hdr;
+
+	/* ignore non UDP */
+	if (ipv6->proto != IPPROTO_UDP)
+		return 0;
+
+	udp = (struct rte_udp_hdr *)(ipv6 + 1);
+	udp->dgram_cksum = 0;
+
+	return 0;
+}
+
 /**
  * Convert L2 encap action to DV specification.
  *
@@ -1400,6 +1444,8 @@ struct field_modify_info modify_tcp[] = {
 			(const struct rte_flow_action_raw_encap *)action->conf;
 		res.size = raw_encap_data->size;
 		memcpy(res.buf, raw_encap_data->data, res.size);
+		if (flow_dv_zero_encap_udp_csum(res.buf, error))
+			return -rte_errno;
 	} else {
 		if (action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP)
 			encap_data =
-- 
1.8.3.1

Re: [dpdk-dev] [PATCH v2] net/mlx5: zero out UDP csum for IPv6 encap headers

[Raslan Darawsheh]

Hi,

> -----Original Message-----
> From: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> Sent: Thursday, July 4, 2019 2:16 PM
> To: dev@dpdk.org
> Cc: Raslan Darawsheh <rasland@mellanox.com>; Eli Britstein
> <elibr@mellanox.com>
> Subject: [PATCH v2] net/mlx5: zero out UDP csum for IPv6 encap headers
> 
> From: Eli Britstein <elibr@mellanox.com>
> 
> Mellanox NICs do not support UDP checksum hardware tx offload over IPv6.
> This limitation becomes critical for UDP based tunnels like VXLAN.
> Beside the UDP checksum validity is required by IPv6 there is an option
> in Linux to allow accepting UDP zero sum (see udp6zerocsumrx in iproute2
> package).
> 
> This patch zeroes out the UDP checksum field for encapsulation headers
> in raw encap action.
> 
> Signed-off-by: Eli Britstein <elibr@mellanox.com>
> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> ---
>  drivers/net/mlx5/mlx5_flow_dv.c | 46
> +++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 46 insertions(+)
> 

Patch applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh