[PATCH] net/mlx5/hws: fix ESP header match in strict mode

[PATCH] net/mlx5/hws: fix ESP header match in strict mode

[Viacheslav Ovsiienko]

The pattern like "eth / ipv6 / esp / end" matched on any IPv6
packet in strict mode, because there was no impicit match on the
IP.proto forced.

This patch adds the implicit match on IP.proto with value 50 (ESP)
and adds implicit match on UDP.dport with value 4500 for the case
ESP over UDP.

Fixes: 18ca4a4ec73a ("net/mlx5: support ESP SPI match and RSS hash")
Cc: stable@dpdk.org

Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
 drivers/net/mlx5/hws/mlx5dr_definer.c | 38 ++++++++++++++++++++++++---
 1 file changed, 35 insertions(+), 3 deletions(-)

diff --git a/drivers/net/mlx5/hws/mlx5dr_definer.c b/drivers/net/mlx5/hws/mlx5dr_definer.c
index 7464d95373..113feae291 100644
--- a/drivers/net/mlx5/hws/mlx5dr_definer.c
+++ b/drivers/net/mlx5/hws/mlx5dr_definer.c
@@ -14,6 +14,7 @@
 #define UDP_VXLAN_PORT	4789
 #define UDP_VXLAN_GPE_PORT	4790
 #define UDP_GTPU_PORT	2152
+#define UDP_ESP_PORT	4500
 #define UDP_PORT_MPLS	6635
 #define UDP_GENEVE_PORT 6081
 #define UDP_ROCEV2_PORT	4791
@@ -231,6 +232,8 @@ struct mlx5dr_definer_conv_data {
 	X(SET_BE16,	nvgre_protocol,		v->protocol,		rte_flow_item_nvgre) \
 	X(SET_BE32P,	nvgre_dw1,		&v->tni[0],		rte_flow_item_nvgre) \
 	X(SET,		meter_color,		rte_col_2_mlx5_col(v->color),	rte_flow_item_meter_color) \
+	X(SET,		ipsec_protocol,		IPPROTO_ESP,		rte_flow_item_esp) \
+	X(SET,		ipsec_udp_port,		UDP_ESP_PORT,		rte_flow_item_esp) \
 	X(SET_BE32,     ipsec_spi,              v->hdr.spi,             rte_flow_item_esp) \
 	X(SET_BE32,     ipsec_sequence_number,  v->hdr.seq,             rte_flow_item_esp) \
 	X(SET,		ib_l4_udp_port,		UDP_ROCEV2_PORT,	rte_flow_item_ib_bth) \
@@ -2481,7 +2484,9 @@ mlx5dr_definer_conv_item_meter_color(struct mlx5dr_definer_conv_data *cd,
 }
 
 static struct mlx5dr_definer_fc *
-mlx5dr_definer_get_flex_parser_fc(struct mlx5dr_definer_conv_data *cd, uint32_t byte_off)
+mlx5dr_definer_get_flex_parser_fc(struct mlx5dr_definer_conv_data *cd,
+				  uint32_t byte_off,
+				  int item_idx)
 {
 	uint32_t byte_off_fp7 = MLX5_BYTE_OFF(definer_hl, flex_parser.flex_parser_7);
 	uint32_t byte_off_fp0 = MLX5_BYTE_OFF(definer_hl, flex_parser.flex_parser_0);
@@ -2493,6 +2498,33 @@ mlx5dr_definer_get_flex_parser_fc(struct mlx5dr_definer_conv_data *cd, uint32_t
 		rte_errno = EINVAL;
 		return NULL;
 	}
+
+	/* To match on ESP we must match on ip_protocol and optionally on l4_dport */
+	if (!cd->relaxed) {
+		bool over_udp;
+
+		fc = &cd->fc[DR_CALC_FNAME(IP_PROTOCOL, false)];
+		over_udp = fc->tag_set == &mlx5dr_definer_udp_protocol_set;
+
+		if (over_udp) {
+			fc = &cd->fc[DR_CALC_FNAME(L4_DPORT, false)];
+			if (!fc->tag_set) {
+				fc->item_idx = item_idx;
+				fc->tag_mask_set = &mlx5dr_definer_ones_set;
+				fc->tag_set = &mlx5dr_definer_ipsec_udp_port_set;
+				DR_CALC_SET(fc, eth_l4, destination_port, false);
+			}
+		} else {
+			fc = &cd->fc[DR_CALC_FNAME(IP_PROTOCOL, false)];
+			if (!fc->tag_set) {
+				fc->item_idx = item_idx;
+				fc->tag_set = &mlx5dr_definer_ipsec_protocol_set;
+				fc->tag_mask_set = &mlx5dr_definer_ones_set;
+				DR_CALC_SET(fc, eth_l3, protocol_next_header, false);
+			}
+		}
+	}
+
 	idx = (byte_off_fp0 - byte_off) / (sizeof(uint32_t));
 	fname += (enum mlx5dr_definer_fname)idx;
 	fc = &cd->fc[fname];
@@ -2544,7 +2576,7 @@ mlx5dr_definer_conv_item_ipv6_routing_ext(struct mlx5dr_definer_conv_data *cd,
 
 	if (m->hdr.next_hdr || m->hdr.type || m->hdr.segments_left) {
 		byte_off = flow_hw_get_srh_flex_parser_byte_off_from_ctx(cd->ctx);
-		fc = mlx5dr_definer_get_flex_parser_fc(cd, byte_off);
+		fc = mlx5dr_definer_get_flex_parser_fc(cd, byte_off, item_idx);
 		if (!fc)
 			return rte_errno;
 
@@ -2666,7 +2698,7 @@ mlx5dr_definer_conv_item_ecpri(struct mlx5dr_definer_conv_data *cd,
 		if (!mask)
 			continue;
 		mask = htobe32(mask);
-		fc = mlx5dr_definer_get_flex_parser_fc(cd, byte_off[i]);
+		fc = mlx5dr_definer_get_flex_parser_fc(cd, byte_off[i], item_idx);
 		if (!fc)
 			return rte_errno;
 
-- 
2.34.1

Re: [PATCH] net/mlx5/hws: fix ESP header match in strict mode

[Dariusz Sosnowski]

On Mon, Aug 04, 2025 at 08:05:14AM +0300, Viacheslav Ovsiienko wrote:
> The pattern like "eth / ipv6 / esp / end" matched on any IPv6
> packet in strict mode, because there was no impicit match on the
> IP.proto forced.
> 
> This patch adds the implicit match on IP.proto with value 50 (ESP)
> and adds implicit match on UDP.dport with value 4500 for the case
> ESP over UDP.
> 
> Fixes: 18ca4a4ec73a ("net/mlx5: support ESP SPI match and RSS hash")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>

Acked-by: Dariusz Sosnowski <dsosnowski@nvidia.com>

Best regards,
Dariusz Sosnowski

Re: [PATCH] net/mlx5/hws: fix ESP header match in strict mode

[Raslan Darawsheh]

Hi,


On 04/08/2025 8:05 AM, Viacheslav Ovsiienko wrote:
> The pattern like "eth / ipv6 / esp / end" matched on any IPv6
> packet in strict mode, because there was no impicit match on the
> IP.proto forced.
> 
> This patch adds the implicit match on IP.proto with value 50 (ESP)
> and adds implicit match on UDP.dport with value 4500 for the case
> ESP over UDP.
> 
> Fixes: 18ca4a4ec73a ("net/mlx5: support ESP SPI match and RSS hash")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>

Patch applied to next-net-mlx,

Kindest regards
Raslan Darawsheh

RE: [PATCH] net/mlx5/hws: fix ESP header match in strict mode

[Maayan Kashani]

Please drop this patch, 
the merge was incorrect and the code is incorrect.

Regards,
Maayan Kashani

> -----Original Message-----
> From: Raslan Darawsheh <rasland@nvidia.com>
> Sent: Monday, 18 August 2025 9:33
> To: Slava Ovsiienko <viacheslavo@nvidia.com>; dev@dpdk.org
> Cc: Matan Azrad <matan@nvidia.com>; Suanming Mou
> <suanmingm@nvidia.com>; Dariusz Sosnowski <dsosnowski@nvidia.com>;
> stable@dpdk.org
> Subject: Re: [PATCH] net/mlx5/hws: fix ESP header match in strict mode
> 
> External email: Use caution opening links or attachments
> 
> 
> Hi,
> 
> 
> On 04/08/2025 8:05 AM, Viacheslav Ovsiienko wrote:
> > The pattern like "eth / ipv6 / esp / end" matched on any IPv6 packet
> > in strict mode, because there was no impicit match on the IP.proto
> > forced.
> >
> > This patch adds the implicit match on IP.proto with value 50 (ESP) and
> > adds implicit match on UDP.dport with value 4500 for the case ESP over
> > UDP.
> >
> > Fixes: 18ca4a4ec73a ("net/mlx5: support ESP SPI match and RSS hash")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
> 
> Patch applied to next-net-mlx,
> 
> Kindest regards
> Raslan Darawsheh

Re: [PATCH] net/mlx5/hws: fix ESP header match in strict mode

[Raslan Darawsheh]

On 08/09/2025 3:12 PM, Maayan Kashani wrote:
> Please drop this patch,
> the merge was incorrect and the code is incorrect.
there isn't anything wrong in the merge, the patch was merged as is.
I'll drop this version of this patch and you can send a V2 for it.

Kindest regards
Raslan Darawsheh