[v2,2/4] net/nfp: add infrastructure for conntrack flow merge

Message ID 20231004093602.1278126-3-chaoyong.he@corigine.com (mailing list archive)
State Accepted, archived
Delegated to: Ferruh Yigit
Headers
Series support offload of simple conntrack flow rules |

Checks

Context Check Description
ci/checkpatch success coding style OK

Commit Message

Chaoyong He Oct. 4, 2023, 9:36 a.m. UTC
  Add the logic to process the merge of items and actions of
pre_ct and post_ct flow. The result will be stored in a field
of merged flow.

Signed-off-by: Chaoyong He <chaoyong.he@corigine.com>
---
 doc/guides/nics/features/nfp.ini       |    2 +
 drivers/net/nfp/flower/nfp_conntrack.c | 1616 ++++++++++++++++++++++++
 drivers/net/nfp/flower/nfp_conntrack.h |   32 +
 drivers/net/nfp/meson.build            |    1 +
 drivers/net/nfp/nfp_flow.c             |   56 +
 drivers/net/nfp/nfp_flow.h             |    4 +
 6 files changed, 1711 insertions(+)
 create mode 100644 drivers/net/nfp/flower/nfp_conntrack.c
 create mode 100644 drivers/net/nfp/flower/nfp_conntrack.h
  

Patch

diff --git a/doc/guides/nics/features/nfp.ini b/doc/guides/nics/features/nfp.ini
index 4264943f05..b53af7b60a 100644
--- a/doc/guides/nics/features/nfp.ini
+++ b/doc/guides/nics/features/nfp.ini
@@ -28,6 +28,7 @@  x86-64               = Y
 Usage doc            = Y
 
 [rte_flow items]
+conntrack            = Y
 eth                  = Y
 geneve               = Y
 gre                  = Y
@@ -42,6 +43,7 @@  vlan                 = Y
 vxlan                = Y
 
 [rte_flow actions]
+conntrack            = Y
 count                = Y
 drop                 = Y
 jump                 = Y
diff --git a/drivers/net/nfp/flower/nfp_conntrack.c b/drivers/net/nfp/flower/nfp_conntrack.c
new file mode 100644
index 0000000000..4fa6fdff99
--- /dev/null
+++ b/drivers/net/nfp/flower/nfp_conntrack.c
@@ -0,0 +1,1616 @@ 
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2023 Corigine, Inc.
+ * All rights reserved.
+ */
+
+#include "nfp_conntrack.h"
+
+#include <rte_malloc.h>
+#include <rte_hash.h>
+#include <rte_jhash.h>
+
+#include "../nfp_flow.h"
+#include "../nfp_logs.h"
+#include "nfp_flower_representor.h"
+
+struct ct_data {
+	uint8_t  ct_state;        /* Connection state. */
+	uint16_t ct_zone;         /* Connection zone. */
+};
+
+enum ct_entry_type {
+	CT_TYPE_PRE_CT,
+	CT_TYPE_POST_CT,
+};
+
+struct nfp_initial_flow {
+	struct rte_flow_item *items;
+	struct rte_flow_action *actions;
+	uint8_t items_cnt;
+	uint8_t actions_cnt;
+};
+
+struct nfp_ct_flow_entry {
+	uint64_t cookie;
+	LIST_ENTRY(nfp_ct_flow_entry) pre_ct_list;
+	LIST_ENTRY(nfp_ct_flow_entry) post_ct_list;
+	LIST_HEAD(, nfp_ct_merge_entry) children;
+	enum ct_entry_type type;
+	struct nfp_flower_representor *repr;
+	struct nfp_ct_zone_entry *ze;
+	struct nfp_initial_flow rule;
+};
+
+struct nfp_ct_map_entry {
+	uint64_t cookie;
+	struct nfp_ct_flow_entry *fe;
+};
+
+struct nfp_ct_zone_entry {
+	uint32_t zone;
+	struct nfp_flow_priv *priv;
+	LIST_HEAD(, nfp_ct_flow_entry) pre_ct_list;
+	LIST_HEAD(, nfp_ct_flow_entry) post_ct_list;
+	struct rte_hash *ct_merge_table;
+};
+
+struct nfp_ct_merge_entry {
+	uint64_t cookie[2];
+	LIST_ENTRY(nfp_ct_merge_entry) pre_ct_list;
+	LIST_ENTRY(nfp_ct_merge_entry) post_ct_list;
+	struct nfp_initial_flow rule;
+	struct nfp_ct_zone_entry *ze;
+	struct nfp_ct_flow_entry *pre_ct_parent;
+	struct nfp_ct_flow_entry *post_ct_parent;
+};
+
+/* OVS_KEY_ATTR_CT_STATE flags */
+#define OVS_CS_F_NEW            0x01 /* Beginning of a new connection. */
+#define OVS_CS_F_ESTABLISHED    0x02 /* Part of an existing connection. */
+#define OVS_CS_F_RELATED        0x04 /* Related to an established connection. */
+#define OVS_CS_F_REPLY_DIR      0x08 /* Flow is in the reply direction. */
+#define OVS_CS_F_INVALID        0x10 /* Could not track connection. */
+#define OVS_CS_F_TRACKED        0x20 /* Conntrack has occurred. */
+#define OVS_CS_F_SRC_NAT        0x40 /* Packet's source address/port was mangled by NAT. */
+#define OVS_CS_F_DST_NAT        0x80 /* Packet's destination address/port was mangled by NAT. */
+
+typedef void (*nfp_action_free_fn)(void *field);
+typedef bool (*nfp_action_copy_fn)(const void *src, void *dst);
+
+static bool
+is_pre_ct_flow(const struct ct_data *ct,
+		const struct rte_flow_action *actions)
+{
+	const struct rte_flow_action *action;
+
+	if (ct == NULL)
+		return false;
+
+	for (action = actions; action->type != RTE_FLOW_ACTION_TYPE_END; ++action) {
+		if (action->type == RTE_FLOW_ACTION_TYPE_CONNTRACK)
+			return true;
+	}
+
+	return false;
+}
+
+static bool
+is_post_ct_flow(const struct ct_data *ct)
+{
+	if (ct == NULL)
+		return false;
+
+	if ((ct->ct_state & OVS_CS_F_ESTABLISHED) != 0)
+		return true;
+
+	return false;
+}
+
+static bool
+is_ct_commit_flow(const struct ct_data *ct)
+{
+	if (ct == NULL)
+		return false;
+
+	if ((ct->ct_state & OVS_CS_F_NEW) != 0)
+		return true;
+
+	return false;
+}
+
+static struct nfp_ct_merge_entry *
+nfp_ct_merge_table_search(struct nfp_ct_zone_entry *ze,
+		char *hash_data,
+		uint32_t hash_len)
+{
+	int index;
+	uint32_t hash_key;
+	struct nfp_ct_merge_entry *m_ent;
+
+	hash_key = rte_jhash(hash_data, hash_len, ze->priv->hash_seed);
+	index = rte_hash_lookup_data(ze->ct_merge_table, &hash_key, (void **)&m_ent);
+	if (index < 0) {
+		PMD_DRV_LOG(DEBUG, "Data NOT found in the ct_merge table");
+		return NULL;
+	}
+
+	return m_ent;
+}
+
+static bool
+nfp_ct_merge_table_add(struct nfp_ct_zone_entry *ze,
+		struct nfp_ct_merge_entry *merge_entry)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(merge_entry, sizeof(uint64_t) * 2, ze->priv->hash_seed);
+	ret = rte_hash_add_key_data(ze->ct_merge_table, &hash_key, merge_entry);
+	if (ret != 0) {
+		PMD_DRV_LOG(ERR, "Add to ct_merge table failed");
+		return false;
+	}
+
+	return true;
+}
+
+static void
+nfp_ct_merge_table_delete(struct nfp_ct_zone_entry *ze,
+		struct nfp_ct_merge_entry *m_ent)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(m_ent, sizeof(uint64_t) * 2, ze->priv->hash_seed);
+	ret = rte_hash_del_key(ze->ct_merge_table, &hash_key);
+	if (ret < 0)
+		PMD_DRV_LOG(ERR, "Delete from ct_merge table failed, ret=%d", ret);
+}
+
+static void
+nfp_ct_merge_entry_destroy(struct nfp_ct_merge_entry *m_ent)
+{
+	struct nfp_ct_zone_entry *ze;
+
+	ze = m_ent->ze;
+	nfp_ct_merge_table_delete(ze, m_ent);
+
+	rte_free(m_ent->rule.actions);
+	rte_free(m_ent->rule.items);
+	LIST_REMOVE(m_ent, pre_ct_list);
+	LIST_REMOVE(m_ent, post_ct_list);
+	rte_free(m_ent);
+}
+
+struct nfp_ct_map_entry *
+nfp_ct_map_table_search(struct nfp_flow_priv *priv,
+		char *hash_data,
+		uint32_t hash_len)
+{
+	int index;
+	uint32_t hash_key;
+	struct nfp_ct_map_entry *me;
+
+	hash_key = rte_jhash(hash_data, hash_len, priv->hash_seed);
+	index = rte_hash_lookup_data(priv->ct_map_table, &hash_key, (void **)&me);
+	if (index < 0) {
+		PMD_DRV_LOG(DEBUG, "Data NOT found in the ct_map table");
+		return NULL;
+	}
+
+	return me;
+}
+
+static bool
+nfp_ct_map_table_add(struct nfp_flow_priv *priv,
+		struct nfp_ct_map_entry *me)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(me, sizeof(uint64_t), priv->hash_seed);
+	ret = rte_hash_add_key_data(priv->ct_map_table, &hash_key, me);
+	if (ret != 0) {
+		PMD_DRV_LOG(ERR, "Add to ct_map table failed");
+		return false;
+	}
+
+	return true;
+}
+
+static void
+nfp_ct_map_table_delete(struct nfp_flow_priv *priv,
+		struct nfp_ct_map_entry *me)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(me, sizeof(uint64_t), priv->hash_seed);
+	ret = rte_hash_del_key(priv->ct_map_table, &hash_key);
+	if (ret < 0)
+		PMD_DRV_LOG(ERR, "Delete form ct_map table failed");
+}
+
+static void
+nfp_ct_map_entry_destroy(struct nfp_ct_map_entry *me)
+{
+	rte_free(me);
+}
+
+static void
+nfp_ct_flow_item_free_real(void *field,
+		enum rte_flow_item_type type)
+{
+	switch (type) {
+	case RTE_FLOW_ITEM_TYPE_VOID:
+		break;
+	case RTE_FLOW_ITEM_TYPE_ETH:        /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_VLAN:       /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_IPV4:       /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_IPV6:       /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_TCP:        /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_UDP:        /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_SCTP:       /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_VXLAN:      /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_GRE:        /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_GRE_KEY:    /* FALLTHROUGH */
+	case RTE_FLOW_ITEM_TYPE_GENEVE:
+		rte_free(field);
+		break;
+	default:
+		break;
+	}
+}
+
+static void
+nfp_ct_flow_item_free(struct rte_flow_item *item)
+{
+	if (item->spec != NULL)
+		nfp_ct_flow_item_free_real((void *)(ptrdiff_t)item->spec, item->type);
+
+	if (item->mask != NULL)
+		nfp_ct_flow_item_free_real((void *)(ptrdiff_t)item->mask, item->type);
+
+	if (item->last != NULL)
+		nfp_ct_flow_item_free_real((void *)(ptrdiff_t)item->last, item->type);
+}
+
+static void
+nfp_ct_flow_items_free(struct rte_flow_item *items,
+		uint8_t item_cnt)
+{
+	uint8_t loop;
+
+	for (loop = 0; loop < item_cnt; ++loop)
+		nfp_ct_flow_item_free(items + loop);
+}
+
+static bool
+nfp_flow_item_conf_size_get(enum rte_flow_item_type type,
+		size_t *size)
+{
+	size_t len = 0;
+
+	switch (type) {
+	case RTE_FLOW_ITEM_TYPE_VOID:
+		break;
+	case RTE_FLOW_ITEM_TYPE_ETH:
+		len = sizeof(struct rte_flow_item_eth);
+		break;
+	case RTE_FLOW_ITEM_TYPE_VLAN:
+		len = sizeof(struct rte_flow_item_vlan);
+		break;
+	case RTE_FLOW_ITEM_TYPE_IPV4:
+		len = sizeof(struct rte_flow_item_ipv4);
+		break;
+	case RTE_FLOW_ITEM_TYPE_IPV6:
+		len = sizeof(struct rte_flow_item_ipv6);
+		break;
+	case RTE_FLOW_ITEM_TYPE_TCP:
+		len = sizeof(struct rte_flow_item_tcp);
+		break;
+	case RTE_FLOW_ITEM_TYPE_UDP:
+		len = sizeof(struct rte_flow_item_udp);
+		break;
+	case RTE_FLOW_ITEM_TYPE_SCTP:
+		len = sizeof(struct rte_flow_item_sctp);
+		break;
+	case RTE_FLOW_ITEM_TYPE_VXLAN:
+		len = sizeof(struct rte_flow_item_vxlan);
+		break;
+	case RTE_FLOW_ITEM_TYPE_GRE:
+		len = sizeof(struct rte_flow_item_gre);
+		break;
+	case RTE_FLOW_ITEM_TYPE_GRE_KEY:
+		len = sizeof(rte_be32_t);
+		break;
+	case RTE_FLOW_ITEM_TYPE_GENEVE:
+		len = sizeof(struct rte_flow_item_geneve);
+		break;
+	default:
+		PMD_DRV_LOG(ERR, "Unsupported item type: %d", type);
+		return false;
+	}
+
+	*size = len;
+
+	return true;
+}
+
+static void *
+nfp_ct_flow_item_copy_real(const void *src,
+		enum rte_flow_item_type type)
+{
+	bool ret;
+	void *dst;
+	size_t len;
+
+	ret = nfp_flow_item_conf_size_get(type, &len);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Get flow item conf size failed");
+		return NULL;
+	}
+
+	dst = rte_zmalloc("flow_item", len, 0);
+	if (dst == NULL) {
+		PMD_DRV_LOG(ERR, "Malloc memory for ct item failed");
+		return NULL;
+	}
+
+	rte_memcpy(dst, src, len);
+
+	return dst;
+}
+
+static bool
+nfp_ct_flow_item_copy(const struct rte_flow_item *src,
+		struct rte_flow_item *dst)
+{
+	dst->type = src->type;
+
+	if (src->spec != NULL) {
+		dst->spec = nfp_ct_flow_item_copy_real(src->spec, src->type);
+		if (dst->spec == NULL) {
+			PMD_DRV_LOG(ERR, "Copy spec of ct item failed");
+			goto end;
+		}
+	}
+
+	if (src->mask != NULL) {
+		dst->mask = nfp_ct_flow_item_copy_real(src->mask, src->type);
+		if (dst->mask == NULL) {
+			PMD_DRV_LOG(ERR, "Copy mask of ct item failed");
+			goto free_spec;
+		}
+	}
+
+	if (src->last != NULL) {
+		dst->last = nfp_ct_flow_item_copy_real(src->last, src->type);
+		if (dst->last == NULL) {
+			PMD_DRV_LOG(ERR, "Copy last of ct item failed");
+			goto free_mask;
+		}
+	}
+
+	return true;
+
+free_mask:
+	nfp_ct_flow_item_free_real((void *)(ptrdiff_t)dst->mask, dst->type);
+free_spec:
+	nfp_ct_flow_item_free_real((void *)(ptrdiff_t)dst->spec, dst->type);
+end:
+	return false;
+}
+
+static bool
+nfp_ct_flow_items_copy(const struct rte_flow_item *src,
+		struct rte_flow_item *dst,
+		uint8_t item_cnt)
+{
+	bool ret;
+	uint8_t loop;
+
+	for (loop = 0; loop < item_cnt; ++loop) {
+		ret = nfp_ct_flow_item_copy(src + loop, dst + loop);
+		if (!ret) {
+			PMD_DRV_LOG(ERR, "Copy ct item failed");
+			nfp_ct_flow_items_free(dst, loop);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static void
+nfp_ct_flow_action_free_real(void *field,
+		nfp_action_free_fn func)
+{
+	if (func != NULL)
+		func(field);
+
+	rte_free(field);
+}
+
+static void
+nfp_ct_flow_action_free_vxlan(void *field)
+{
+	struct vxlan_data *vxlan = field;
+
+	nfp_ct_flow_items_free(vxlan->items, ACTION_VXLAN_ENCAP_ITEMS_NUM);
+}
+
+static void
+nfp_ct_flow_action_free_raw(void *field)
+{
+	struct rte_flow_action_raw_encap *raw_encap = field;
+
+	rte_free(raw_encap->data);
+}
+
+static void
+nfp_ct_flow_action_free(struct rte_flow_action *action)
+{
+	nfp_action_free_fn func = NULL;
+
+	if (action->conf == NULL)
+		return;
+
+	switch (action->type) {
+	case RTE_FLOW_ACTION_TYPE_VOID:           /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_DROP:           /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_COUNT:          /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_JUMP:           /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
+		return;
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_PORT_ID:        /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_TTL:        /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:     /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
+		break;
+	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+		func = nfp_ct_flow_action_free_vxlan;
+		break;
+	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
+		func = nfp_ct_flow_action_free_raw;
+		break;
+	default:
+		PMD_DRV_LOG(ERR, "Unsupported action type: %d", action->type);
+		break;
+	}
+
+	nfp_ct_flow_action_free_real((void *)(ptrdiff_t)action->conf, func);
+}
+
+static void
+nfp_ct_flow_actions_free(struct rte_flow_action *actions,
+		uint8_t action_cnt)
+{
+	uint8_t loop;
+
+	for (loop = 0; loop < action_cnt; ++loop)
+		nfp_ct_flow_action_free(actions + loop);
+}
+
+static void *
+nfp_ct_flow_action_copy_real(const void *src,
+		size_t len,
+		nfp_action_copy_fn func)
+{
+	bool ret;
+	void *dst;
+
+	dst = rte_zmalloc("flow_action", len, 0);
+	if (dst == NULL) {
+		PMD_DRV_LOG(ERR, "Malloc memory for ct action failed");
+		return NULL;
+	}
+
+	if (func != NULL) {
+		ret = func(src, dst);
+		if (!ret) {
+			PMD_DRV_LOG(ERR, "Copy ct action failed");
+			return NULL;
+		}
+
+		return dst;
+	}
+
+	rte_memcpy(dst, src, len);
+
+	return dst;
+}
+
+static bool
+nfp_ct_flow_action_copy_vxlan(const void *src,
+		void *dst)
+{
+	struct vxlan_data *vxlan_dst = dst;
+	const struct vxlan_data *vxlan_src = src;
+
+	vxlan_dst->conf.definition = vxlan_dst->items;
+	return nfp_ct_flow_items_copy(vxlan_src->items, vxlan_dst->items,
+			ACTION_VXLAN_ENCAP_ITEMS_NUM);
+}
+
+static bool
+nfp_ct_flow_action_copy_raw(const void *src,
+		void *dst)
+{
+	struct rte_flow_action_raw_encap *raw_dst = dst;
+	const struct rte_flow_action_raw_encap *raw_src = src;
+
+	raw_dst->size = raw_src->size;
+	raw_dst->data = nfp_ct_flow_action_copy_real(raw_src->data,
+			raw_src->size, NULL);
+	if (raw_dst->data == NULL) {
+		PMD_DRV_LOG(ERR, "Copy ct action process failed");
+		return false;
+	}
+
+	return true;
+}
+
+static bool
+nfp_ct_flow_action_copy(const struct rte_flow_action *src,
+		struct rte_flow_action *dst)
+{
+	size_t len;
+	nfp_action_copy_fn func = NULL;
+
+	dst->type = src->type;
+
+	if (src->conf == NULL)
+		return true;
+
+	switch (src->type) {
+	case RTE_FLOW_ACTION_TYPE_VOID:         /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_DROP:         /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_COUNT:        /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_JUMP:         /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
+		return true;
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
+		len = sizeof(struct rte_flow_action_set_mac);
+		break;
+	case RTE_FLOW_ACTION_TYPE_PORT_ID:
+		len = sizeof(struct rte_flow_action_port_id);
+		break;
+	case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
+		len = sizeof(struct rte_flow_action_of_push_vlan);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
+		len = sizeof(struct rte_flow_action_set_ipv4);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
+		len = sizeof(struct rte_flow_action_set_dscp);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
+		len = sizeof(struct rte_flow_action_set_ipv6);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_TTL:
+		len = sizeof(struct rte_flow_action_set_ttl);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:  /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
+		len = sizeof(struct rte_flow_action_set_tp);
+		break;
+	case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+		len = sizeof(struct vxlan_data);
+		func = nfp_ct_flow_action_copy_vxlan;
+		break;
+	case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
+		len = sizeof(struct rte_flow_action_raw_encap);
+		func = nfp_ct_flow_action_copy_raw;
+		break;
+	default:
+		PMD_DRV_LOG(DEBUG, "Unsupported action type: %d", src->type);
+		return false;
+	}
+
+	dst->conf = nfp_ct_flow_action_copy_real(src->conf, len, func);
+	if (dst->conf == NULL) {
+		PMD_DRV_LOG(DEBUG, "Copy ct action process failed");
+		return false;
+	}
+
+	return true;
+}
+
+static bool
+nfp_ct_flow_actions_copy(const struct rte_flow_action *src,
+		struct rte_flow_action *dst,
+		uint8_t action_cnt)
+{
+	bool ret;
+	uint8_t loop;
+
+	for (loop = 0; loop < action_cnt; ++loop) {
+		ret = nfp_ct_flow_action_copy(src + loop, dst + loop);
+		if (!ret) {
+			PMD_DRV_LOG(DEBUG, "Copy ct action failed");
+			nfp_ct_flow_actions_free(dst, loop);
+			return false;
+		}
+	}
+
+	return true;
+}
+
+static struct nfp_ct_flow_entry *
+nfp_ct_flow_entry_get(struct nfp_ct_zone_entry *ze,
+		struct nfp_flower_representor *repr,
+		const struct rte_flow_item items[],
+		const struct rte_flow_action actions[],
+		uint64_t cookie)
+{
+	bool ret;
+	uint8_t loop;
+	uint8_t item_cnt = 1;      /* the RTE_FLOW_ITEM_TYPE_END */
+	uint8_t action_cnt = 1;    /* the RTE_FLOW_ACTION_TYPE_END */
+	struct nfp_flow_priv *priv;
+	struct nfp_ct_map_entry *me;
+	struct nfp_ct_flow_entry *fe;
+
+	fe = rte_zmalloc("ct_flow_entry", sizeof(*fe), 0);
+	if (fe == NULL) {
+		PMD_DRV_LOG(ERR, "Could not alloc ct_flow entry");
+		return NULL;
+	}
+
+	fe->ze = ze;
+	fe->repr = repr;
+	fe->cookie = cookie;
+	LIST_INIT(&fe->children);
+
+	for (loop = 0; (items + loop)->type != RTE_FLOW_ITEM_TYPE_END; loop++)
+		item_cnt++;
+	for (loop = 0; (actions + loop)->type != RTE_FLOW_ACTION_TYPE_END; loop++)
+		action_cnt++;
+
+	fe->rule.items = rte_zmalloc("ct_flow_item",
+			sizeof(struct rte_flow_item) * item_cnt, 0);
+	if (fe->rule.items == NULL) {
+		PMD_DRV_LOG(ERR, "Could not alloc ct flow items");
+		goto free_flow_entry;
+	}
+
+	fe->rule.actions = rte_zmalloc("ct_flow_action",
+			sizeof(struct rte_flow_action) * action_cnt, 0);
+	if (fe->rule.actions == NULL) {
+		PMD_DRV_LOG(ERR, "Could not alloc ct flow actions");
+		goto free_flow_item;
+	}
+
+	/* Deep copy of items */
+	ret = nfp_ct_flow_items_copy(items, fe->rule.items, item_cnt);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Could not deep copy ct flow items");
+		goto free_flow_action;
+	}
+
+	/* Deep copy of actions */
+	ret = nfp_ct_flow_actions_copy(actions, fe->rule.actions, action_cnt);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Could not deep copy ct flow actions");
+		goto free_copied_items;
+	}
+
+	fe->rule.items_cnt = item_cnt;
+	fe->rule.actions_cnt = action_cnt;
+
+	/* Now add a ct map entry */
+	me = rte_zmalloc("ct_map_entry", sizeof(*me), 0);
+	if (me == NULL) {
+		PMD_DRV_LOG(ERR, "Malloc memory for ct map entry failed");
+		goto free_copied_actions;
+	}
+
+	me->cookie = fe->cookie;
+	me->fe = fe;
+
+	priv = repr->app_fw_flower->flow_priv;
+	ret = nfp_ct_map_table_add(priv, me);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Add into ct map table failed");
+		goto free_map_entry;
+	}
+
+	return fe;
+
+free_map_entry:
+	nfp_ct_map_entry_destroy(me);
+free_copied_actions:
+	nfp_ct_flow_actions_free(fe->rule.actions, action_cnt);
+free_copied_items:
+	nfp_ct_flow_items_free(fe->rule.items, item_cnt);
+free_flow_action:
+	rte_free(fe->rule.actions);
+free_flow_item:
+	rte_free(fe->rule.items);
+free_flow_entry:
+	rte_free(fe);
+
+	return NULL;
+}
+
+static void
+nfp_flow_children_merge_free(struct nfp_ct_flow_entry *fe)
+{
+	struct nfp_ct_merge_entry *m_ent;
+
+	switch (fe->type) {
+	case CT_TYPE_PRE_CT:
+		LIST_FOREACH(m_ent, &fe->children, pre_ct_list)
+			nfp_ct_merge_entry_destroy(m_ent);
+		break;
+	case CT_TYPE_POST_CT:
+		LIST_FOREACH(m_ent, &fe->children, post_ct_list)
+			nfp_ct_merge_entry_destroy(m_ent);
+		break;
+	default:
+		break;
+	}
+}
+
+static void
+nfp_ct_flow_entry_destroy_partly(struct nfp_ct_flow_entry *fe)
+{
+	struct nfp_ct_map_entry *me;
+
+	if (!LIST_EMPTY(&fe->children))
+		nfp_flow_children_merge_free(fe);
+
+	me = nfp_ct_map_table_search(fe->ze->priv, (char *)&fe->cookie, sizeof(uint64_t));
+	if (me != NULL) {
+		nfp_ct_map_table_delete(fe->ze->priv, me);
+		nfp_ct_map_entry_destroy(me);
+	}
+
+	nfp_ct_flow_actions_free(fe->rule.actions, fe->rule.actions_cnt);
+	nfp_ct_flow_items_free(fe->rule.items, fe->rule.items_cnt);
+	rte_free(fe->rule.actions);
+	rte_free(fe->rule.items);
+	rte_free(fe);
+}
+
+static void
+nfp_ct_flow_entry_destroy(struct nfp_ct_flow_entry *fe)
+{
+	LIST_REMOVE(fe, pre_ct_list);
+	LIST_REMOVE(fe, post_ct_list);
+
+	nfp_ct_flow_entry_destroy_partly(fe);
+}
+
+static struct nfp_ct_zone_entry *
+nfp_ct_zone_table_search(struct nfp_flow_priv *priv,
+		char *hash_data,
+		uint32_t hash_len)
+{
+	int index;
+	uint32_t hash_key;
+	struct nfp_ct_zone_entry *ze;
+
+	hash_key = rte_jhash(hash_data, hash_len, priv->hash_seed);
+	index = rte_hash_lookup_data(priv->ct_zone_table, &hash_key, (void **)&ze);
+	if (index < 0) {
+		PMD_DRV_LOG(DEBUG, "Data NOT found in the ct_zone table");
+		return NULL;
+	}
+
+	return ze;
+}
+
+static bool
+nfp_ct_zone_table_add(struct nfp_flow_priv *priv,
+		struct nfp_ct_zone_entry *ze)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(ze, sizeof(uint32_t), priv->hash_seed);
+	ret = rte_hash_add_key_data(priv->ct_zone_table, &hash_key, ze);
+	if (ret != 0) {
+		PMD_DRV_LOG(ERR, "Add to the ct_zone table failed");
+		return false;
+	}
+
+	return true;
+}
+
+static void
+nfp_ct_zone_table_delete(struct nfp_flow_priv *priv,
+		struct nfp_ct_zone_entry *ze)
+{
+	int ret;
+	uint32_t hash_key;
+
+	hash_key = rte_jhash(ze, sizeof(uint32_t), priv->hash_seed);
+	ret = rte_hash_del_key(priv->ct_zone_table, &hash_key);
+	if (ret < 0)
+		PMD_DRV_LOG(ERR, "Delete from the ct_zone table failed");
+}
+
+static bool
+nfp_ct_zone_entry_init(struct nfp_ct_zone_entry *ze,
+		struct nfp_flow_priv *priv,
+		uint32_t zone,
+		bool wildcard)
+{
+	char hash_name[RTE_HASH_NAMESIZE];
+	struct rte_hash_parameters ct_merge_hash_params = {
+		.entries    = 1000,
+		.hash_func  = rte_jhash,
+		.socket_id  = rte_socket_id(),
+		.key_len    = sizeof(uint32_t),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_RW_CONCURRENCY,
+	};
+
+	if (wildcard) {
+		ct_merge_hash_params.name = "ct_wc_merge_table";
+	} else {
+		snprintf(hash_name, sizeof(hash_name), "ct_%d_merge_table", ze->zone);
+		ct_merge_hash_params.name = hash_name;
+	}
+
+	ct_merge_hash_params.hash_func_init_val = priv->hash_seed;
+	ze->ct_merge_table = rte_hash_create(&ct_merge_hash_params);
+	if (ze->ct_merge_table == NULL) {
+		PMD_DRV_LOG(ERR, "ct merge table creation failed");
+		return false;
+	}
+
+	ze->zone = zone;
+	ze->priv = priv;
+	LIST_INIT(&ze->pre_ct_list);
+	LIST_INIT(&ze->post_ct_list);
+
+	return true;
+}
+
+static void
+nfp_ct_zone_entry_destroy(struct nfp_ct_zone_entry *ze)
+{
+	struct nfp_ct_flow_entry *fe;
+
+	if (ze == NULL)
+		return;
+
+	rte_hash_free(ze->ct_merge_table);
+
+	LIST_FOREACH(fe, &ze->pre_ct_list, pre_ct_list)
+		nfp_ct_flow_entry_destroy(fe);
+
+	LIST_FOREACH(fe, &ze->post_ct_list, post_ct_list)
+		nfp_ct_flow_entry_destroy(fe);
+
+	rte_free(ze);
+}
+
+static struct nfp_ct_zone_entry *
+nfp_ct_zone_entry_get(struct nfp_flow_priv *priv,
+		uint32_t zone,
+		bool wildcard)
+{
+	bool is_ok;
+	struct nfp_ct_zone_entry *ze;
+
+	if (wildcard) {
+		if (priv->ct_zone_wc != NULL)
+			return priv->ct_zone_wc;
+
+		ze = rte_zmalloc("ct_zone_wc", sizeof(*ze), 0);
+		if (ze == NULL) {
+			PMD_DRV_LOG(ERR, "Could not alloc ct_zone_wc entry");
+			return NULL;
+		}
+
+		is_ok = nfp_ct_zone_entry_init(ze, priv, zone, true);
+		if (!is_ok) {
+			PMD_DRV_LOG(ERR, "Init ct zone wc entry failed");
+			goto free_ct_zone_entry;
+		}
+
+		priv->ct_zone_wc = ze;
+	} else {
+		ze = nfp_ct_zone_table_search(priv, (char *)&zone, sizeof(uint32_t));
+		if (ze != NULL)
+			return ze;
+
+		ze = rte_zmalloc("ct_zone_entry", sizeof(*ze), 0);
+		if (ze == NULL) {
+			PMD_DRV_LOG(ERR, "Could not alloc ct_zone entry");
+			return NULL;
+		}
+
+		is_ok = nfp_ct_zone_entry_init(ze, priv, zone, false);
+		if (!is_ok) {
+			PMD_DRV_LOG(ERR, "Init ct zone entry failed");
+			goto free_ct_zone_entry;
+		}
+
+		is_ok = nfp_ct_zone_table_add(priv, ze);
+		if (!is_ok) {
+			PMD_DRV_LOG(ERR, "Add into ct zone table failed");
+			goto free_ct_zone_entry;
+		}
+	}
+
+	return ze;
+
+free_ct_zone_entry:
+	nfp_ct_zone_entry_destroy(ze);
+
+	return NULL;
+}
+
+static void
+nfp_ct_zone_entry_free(struct nfp_ct_zone_entry *ze,
+		bool wildcard)
+{
+	if (LIST_EMPTY(&ze->pre_ct_list) && LIST_EMPTY(&ze->post_ct_list)) {
+		if (!wildcard)
+			nfp_ct_zone_table_delete(ze->priv, ze);
+
+		nfp_ct_zone_entry_destroy(ze);
+	}
+}
+
+static inline bool
+is_item_check_pass(const struct rte_flow_item *item1,
+		const struct rte_flow_item *item2,
+		uint8_t *cnt_same)
+{
+	bool pass;
+	uint32_t i;
+	size_t size;
+	const char *key1 = item1->spec;
+	const char *key2 = item2->spec;
+	const char *mask1 = item1->mask;
+	const char *mask2 = item2->mask;
+
+	if (item1->type != item2->type)
+		return true;
+
+	pass = nfp_flow_item_conf_size_get(item1->type, &size);
+	if (!pass)
+		return false;
+
+	for (i = 0; i < size; i++) {
+		if ((key1[i] & mask1[i] & mask2[i]) ^ (key2[i] & mask1[i] & mask2[i]))
+			return false;
+	}
+
+	*cnt_same = *cnt_same + 1;
+
+	return true;
+}
+
+static bool
+nfp_ct_merge_items_check(struct rte_flow_item *items1,
+		struct rte_flow_item *items2,
+		uint8_t *cnt_same)
+{
+	bool pass;
+	bool is_tun_flow_1;
+	bool is_tun_flow_2;
+	const struct rte_flow_item *item1;
+	const struct rte_flow_item *item2;
+	const struct rte_flow_item *inner_item1 = NULL;
+	const struct rte_flow_item *inner_item2 = NULL;
+
+	is_tun_flow_1 = nfp_flow_inner_item_get(items1, &inner_item1);
+	is_tun_flow_2 = nfp_flow_inner_item_get(items2, &inner_item2);
+
+	if (is_tun_flow_1) {
+		if (is_tun_flow_2) {
+			/* Outer layer */
+			for (item1 = items1; item1 != inner_item1; item1++) {
+				for (item2 = items2; item2 != inner_item2; item2++) {
+					pass = is_item_check_pass(item1, item2, cnt_same);
+					if (!pass)
+						return false;
+				}
+			}
+			/* Inner layer */
+			for (item1 = inner_item1; item1->type != RTE_FLOW_ITEM_TYPE_END; item1++) {
+				for (item2 = inner_item2; item2->type != RTE_FLOW_ITEM_TYPE_END;
+						item2++) {
+					pass = is_item_check_pass(item1, item2, cnt_same);
+					if (!pass)
+						return false;
+				}
+			}
+		} else {
+			for (item1 = items1; item1 != inner_item1; item1++) {
+				for (item2 = items2; item2->type != RTE_FLOW_ITEM_TYPE_END;
+						item2++) {
+					pass = is_item_check_pass(item1, item2, cnt_same);
+					if (!pass)
+						return false;
+				}
+			}
+		}
+	} else {
+		if (is_tun_flow_2) {
+			for (item1 = items1; item1->type != RTE_FLOW_ITEM_TYPE_END; item1++) {
+				for (item2 = items2; item2 != inner_item2; item2++) {
+					pass = is_item_check_pass(item1, item2, cnt_same);
+					if (!pass)
+						return false;
+				}
+			}
+		} else {
+			for (item1 = items1; item1->type != RTE_FLOW_ITEM_TYPE_END; item1++) {
+				for (item2 = items2; item2->type != RTE_FLOW_ITEM_TYPE_END;
+						item2++) {
+					pass = is_item_check_pass(item1, item2, cnt_same);
+					if (!pass)
+						return false;
+				}
+			}
+		}
+	}
+
+	return true;
+}
+
+static inline bool
+is_action_pattern_check_pass(struct rte_flow_item *items,
+		enum rte_flow_item_type type)
+{
+	struct rte_flow_item *item;
+
+	for (item = items; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+		if (item->type == type)
+			return false;
+	}
+
+	return true;
+}
+
+static bool
+nfp_ct_merge_action_check(struct rte_flow_action *action,
+		struct rte_flow_item *items)
+{
+	bool pass = true;
+
+	switch (action->type) {
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
+		pass = is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_ETH);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:
+		pass = is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_IPV4);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:   /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:
+		pass = is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_IPV6);
+		break;
+	case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:    /* FALLTHROUGH */
+	case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
+		pass = is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_UDP);
+		pass |= is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_TCP);
+		pass |= is_action_pattern_check_pass(items, RTE_FLOW_ITEM_TYPE_SCTP);
+		break;
+	default:
+		break;
+	}
+
+	return pass;
+}
+
+static bool
+nfp_ct_merge_actions_check(struct rte_flow_action *actions,
+		struct rte_flow_item *items,
+		uint8_t *cnt_same)
+{
+	bool pass = true;
+	struct rte_flow_action *action;
+
+	for (action = actions; action->type != RTE_FLOW_ACTION_TYPE_END; action++) {
+		switch (action->type) {
+		case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:    /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:    /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:   /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:   /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV4_DSCP:  /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:   /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:   /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_IPV6_DSCP:  /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:     /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
+			pass = nfp_ct_merge_action_check(action, items);
+			break;
+		case RTE_FLOW_ACTION_TYPE_CONNTRACK: /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_JUMP:      /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_COUNT:     /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_DROP:      /* FALLTHROUGH */
+		case RTE_FLOW_ACTION_TYPE_VOID:
+			*cnt_same = *cnt_same + 1;
+			break;
+		default:
+			pass = false;
+			break;
+		}
+	}
+
+	return pass;
+}
+
+static void
+nfp_ct_merge_item_real(const struct rte_flow_item *item_src,
+		struct rte_flow_item *item_dst)
+{
+	uint32_t i;
+	size_t size;
+	char *key_dst;
+	char *mask_dst;
+	const char *key_src;
+	const char *mask_src;
+
+	key_src = item_src->spec;
+	mask_src = item_src->mask;
+	key_dst = (char *)(ptrdiff_t)item_dst->spec;
+	mask_dst = (char *)(ptrdiff_t)item_dst->mask;
+	nfp_flow_item_conf_size_get(item_src->type, &size);
+
+	for (i = 0; i < size; i++) {
+		key_dst[i] |= key_src[i];
+		mask_dst[i] |= mask_src[i];
+	}
+}
+
+static bool
+nfp_ct_merge_item(uint32_t index,
+		const struct rte_flow_item *item1,
+		const struct rte_flow_item *item2_start,
+		const struct rte_flow_item *item2_end,
+		struct nfp_ct_merge_entry *merge_entry)
+{
+	struct rte_flow_item *item;
+	const struct rte_flow_item *item2;
+
+	/* Copy to the merged items */
+	item = &merge_entry->rule.items[index];
+	*item = *item1;
+
+	item2 = item2_start;
+	if (item2_end != NULL) {
+		for (; item2 != item2_end; item2++) {
+			if (item1->type == item2->type) {
+				nfp_ct_merge_item_real(item2, item);
+				return true;
+			}
+		}
+	} else {
+		for (; item2->type != RTE_FLOW_ITEM_TYPE_END; item2++) {
+			if (item1->type == item2->type) {
+				nfp_ct_merge_item_real(item2, item);
+				return true;
+			}
+		}
+	}
+
+	return false;
+}
+
+static void
+nfp_ct_merge_items(struct nfp_ct_merge_entry *merge_entry)
+{
+	uint32_t index = 0;
+	bool is_tun_flow_1;
+	bool is_tun_flow_2;
+	struct rte_flow_item *items1;
+	struct rte_flow_item *items2;
+	struct rte_flow_item *merge_item;
+	const struct rte_flow_item *item;
+	const struct rte_flow_item *inner1 = NULL;
+	const struct rte_flow_item *inner2 = NULL;
+
+	items1 = merge_entry->pre_ct_parent->rule.items;
+	items2 = merge_entry->post_ct_parent->rule.items;
+	is_tun_flow_1 = nfp_flow_inner_item_get(items1, &inner1);
+	is_tun_flow_2 = nfp_flow_inner_item_get(items2, &inner2);
+
+	if (is_tun_flow_1) {
+		if (is_tun_flow_2) {
+			/* Outer layer */
+			for (item = items1; item != inner1; item++, index++) {
+				if (nfp_ct_merge_item(index, item, items2, inner2, merge_entry))
+					items2++;
+			}
+
+			/* Copy the remainning outer layer items */
+			for (item = items2; item != inner2; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+
+			/* Inner layer */
+			for (item = inner1; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				if (nfp_ct_merge_item(index, item, inner2, NULL, merge_entry))
+					items2++;
+			}
+
+			/* Copy the remainning inner layer items */
+			for (item = items2; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+		} else {
+			for (item = items1; item != inner1; item++, index++) {
+				if (nfp_ct_merge_item(index, item, items2, NULL, merge_entry))
+					items2++;
+			}
+
+			/* Copy the remainning items */
+			for (item = items2; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+
+			/* Copy the inner layer items */
+			for (item = inner1; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+		}
+	} else {
+		if (is_tun_flow_2) {
+			for (item = items1; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				if (nfp_ct_merge_item(index, item, items2, inner2, merge_entry))
+					items2++;
+			}
+
+			/* Copy the remainning items */
+			for (item = items2; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+		} else {
+			for (item = items1; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				if (nfp_ct_merge_item(index, item, items2, NULL, merge_entry))
+					items2++;
+			}
+
+			/* Copy the remainning items */
+			for (item = items2; item->type != RTE_FLOW_ITEM_TYPE_END; item++, index++) {
+				merge_item = &merge_entry->rule.items[index];
+				*merge_item = *item;
+			}
+		}
+	}
+}
+
+static void
+nfp_ct_merge_actions(struct nfp_ct_merge_entry *merge_entry)
+{
+	struct rte_flow_action *action;
+	struct rte_flow_action *merge_actions;
+
+	merge_actions = merge_entry->rule.actions;
+
+	action = merge_entry->pre_ct_parent->rule.actions;
+	for (; action->type != RTE_FLOW_ACTION_TYPE_END; action++) {
+		if (action->type == RTE_FLOW_ACTION_TYPE_CONNTRACK ||
+				action->type == RTE_FLOW_ACTION_TYPE_JUMP)
+			continue;
+
+		*merge_actions = *action;
+		merge_actions++;
+	}
+
+	action = merge_entry->post_ct_parent->rule.actions;
+	for (; action->type != RTE_FLOW_ACTION_TYPE_END; action++) {
+		*merge_actions = *action;
+		merge_actions++;
+	}
+}
+
+static bool
+nfp_ct_do_flow_merge(struct nfp_ct_zone_entry *ze,
+		struct nfp_ct_flow_entry *pre_ct_entry,
+		struct nfp_ct_flow_entry *post_ct_entry)
+{
+	bool ret;
+	uint64_t new_cookie[2];
+	uint8_t cnt_same_item = 0;
+	uint8_t cnt_same_action = 0;
+	struct nfp_ct_merge_entry *merge_entry;
+
+	if (pre_ct_entry->repr != post_ct_entry->repr)
+		return true;
+
+	ret = nfp_ct_merge_items_check(pre_ct_entry->rule.items,
+			post_ct_entry->rule.items, &cnt_same_item);
+	if (!ret)
+		return true;
+
+	ret = nfp_ct_merge_actions_check(pre_ct_entry->rule.actions,
+			post_ct_entry->rule.items, &cnt_same_action);
+	if (!ret)
+		return true;
+
+	new_cookie[0] = pre_ct_entry->cookie;
+	new_cookie[1] = post_ct_entry->cookie;
+	merge_entry = nfp_ct_merge_table_search(ze, (char *)&new_cookie, sizeof(uint64_t) * 2);
+	if (merge_entry != NULL)
+		return true;
+
+	merge_entry = rte_zmalloc("ct_merge_entry", sizeof(*merge_entry), 0);
+	if (merge_entry == NULL) {
+		PMD_DRV_LOG(ERR, "Malloc memory for ct merge entry failed");
+		return false;
+	}
+
+	merge_entry->ze = ze;
+	merge_entry->pre_ct_parent = pre_ct_entry;
+	merge_entry->post_ct_parent = post_ct_entry;
+	rte_memcpy(merge_entry->cookie, new_cookie, sizeof(new_cookie));
+	merge_entry->rule.items_cnt = pre_ct_entry->rule.items_cnt +
+			post_ct_entry->rule.items_cnt - cnt_same_item - 1;
+	merge_entry->rule.actions_cnt = pre_ct_entry->rule.actions_cnt +
+			post_ct_entry->rule.actions_cnt - cnt_same_action - 1;
+
+	merge_entry->rule.items = rte_zmalloc("ct_flow_item",
+			sizeof(struct rte_flow_item) * merge_entry->rule.items_cnt, 0);
+	if (merge_entry->rule.items == NULL) {
+		PMD_DRV_LOG(ERR, "Could not alloc items for merged flow");
+		goto merge_exit;
+	}
+
+	merge_entry->rule.actions = rte_zmalloc("ct_flow_action",
+			sizeof(struct rte_flow_action) * merge_entry->rule.actions_cnt, 0);
+	if (merge_entry->rule.actions == NULL) {
+		PMD_DRV_LOG(ERR, "Could not alloc actions for merged flow");
+		goto free_items;
+	}
+
+	nfp_ct_merge_items(merge_entry);
+	nfp_ct_merge_actions(merge_entry);
+
+	/* Add this entry to the pre_ct and post_ct lists */
+	LIST_INSERT_HEAD(&pre_ct_entry->children, merge_entry, pre_ct_list);
+	LIST_INSERT_HEAD(&post_ct_entry->children, merge_entry, post_ct_list);
+
+	ret = nfp_ct_merge_table_add(ze, merge_entry);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Add into ct merge table failed");
+		goto free_actions;
+	}
+
+	return true;
+
+free_actions:
+	rte_free(merge_entry->rule.actions);
+free_items:
+	rte_free(merge_entry->rule.items);
+merge_exit:
+	LIST_REMOVE(merge_entry, post_ct_list);
+	LIST_REMOVE(merge_entry, pre_ct_list);
+	rte_free(merge_entry);
+
+	return ret;
+}
+
+static bool
+nfp_ct_merge_flow_entries(struct nfp_ct_flow_entry *fe,
+		struct nfp_ct_zone_entry *ze_src,
+		struct nfp_ct_zone_entry *ze_dst)
+{
+	bool ret;
+	struct nfp_ct_flow_entry *fe_tmp;
+
+	if (fe->type == CT_TYPE_PRE_CT) {
+		LIST_FOREACH(fe_tmp, &ze_src->post_ct_list, post_ct_list) {
+			ret = nfp_ct_do_flow_merge(ze_dst, fe, fe_tmp);
+			if (!ret) {
+				PMD_DRV_LOG(ERR, "Merge for ct pre flow failed");
+				return false;
+			}
+		}
+	} else {
+		LIST_FOREACH(fe_tmp, &ze_src->pre_ct_list, pre_ct_list) {
+			ret = nfp_ct_do_flow_merge(ze_dst, fe_tmp, fe);
+			if (!ret) {
+				PMD_DRV_LOG(ERR, "Merge for ct post flow failed");
+				return false;
+			}
+		}
+	}
+
+	return true;
+}
+
+static bool
+nfp_flow_handle_pre_ct(const struct rte_flow_item *ct_item,
+		struct nfp_flower_representor *representor,
+		const struct rte_flow_item items[],
+		const struct rte_flow_action actions[],
+		uint64_t cookie)
+{
+	bool ret;
+	struct nfp_flow_priv *priv;
+	struct nfp_ct_zone_entry *ze;
+	struct nfp_ct_flow_entry *fe;
+	const struct ct_data *ct = ct_item->spec;
+
+	priv = representor->app_fw_flower->flow_priv;
+	ze = nfp_ct_zone_entry_get(priv, ct->ct_zone, false);
+	if (ze == NULL) {
+		PMD_DRV_LOG(ERR, "Could not get ct zone entry");
+		return false;
+	}
+
+	/* Add entry to pre_ct_list */
+	fe = nfp_ct_flow_entry_get(ze, representor, items, actions, cookie);
+	if (fe == NULL) {
+		PMD_DRV_LOG(ERR, "Could not get ct flow entry");
+		goto ct_zone_entry_free;
+	}
+
+	fe->type = CT_TYPE_PRE_CT;
+	LIST_INSERT_HEAD(&ze->pre_ct_list, fe, pre_ct_list);
+
+	ret = nfp_ct_merge_flow_entries(fe, ze, ze);
+	if (!ret) {
+		PMD_DRV_LOG(ERR, "Merge ct flow entries failed");
+		goto ct_flow_entry_free;
+	}
+
+	/* Need to check and merge with tables in the wc_zone as well */
+	if (priv->ct_zone_wc != NULL) {
+		ret = nfp_ct_merge_flow_entries(fe, priv->ct_zone_wc, ze);
+		if (!ret) {
+			PMD_DRV_LOG(ERR, "Merge ct flow entries wildcast failed");
+			goto ct_flow_entry_free;
+		}
+	}
+
+	/* The real offload logic comes in next commit, so here just return false for now */
+
+ct_flow_entry_free:
+	nfp_ct_flow_entry_destroy(fe);
+
+ct_zone_entry_free:
+	nfp_ct_zone_entry_free(ze, false);
+
+	return false;
+}
+
+static bool
+nfp_flow_handle_post_ct(const struct rte_flow_item *ct_item,
+		struct nfp_flower_representor *representor,
+		const struct rte_flow_item items[],
+		const struct rte_flow_action actions[],
+		uint64_t cookie)
+{
+	bool ret;
+	void *next_data;
+	uint32_t iter = 0;
+	const void *next_key;
+	bool wildcard = false;
+	struct nfp_flow_priv *priv;
+	struct nfp_ct_zone_entry *ze;
+	struct nfp_ct_flow_entry *fe;
+	const struct ct_data *ct = ct_item->spec;
+	const struct ct_data *ct_mask = ct_item->mask;
+
+	if (ct_mask->ct_zone == 0) {
+		wildcard = true;
+	} else if (ct_mask->ct_zone != UINT16_MAX) {
+		PMD_DRV_LOG(ERR, "Partially wildcard ct_zone is not supported");
+		return false;
+	}
+
+	priv = representor->app_fw_flower->flow_priv;
+	ze = nfp_ct_zone_entry_get(priv, ct->ct_zone, wildcard);
+	if (ze == NULL) {
+		PMD_DRV_LOG(ERR, "Could not get ct zone entry");
+		return false;
+	}
+
+	/* Add entry to post_ct_list */
+	fe = nfp_ct_flow_entry_get(ze, representor, items, actions, cookie);
+	if (fe == NULL) {
+		PMD_DRV_LOG(ERR, "Could not get ct flow entry");
+		goto ct_zone_entry_free;
+	}
+
+	fe->type = CT_TYPE_POST_CT;
+	LIST_INSERT_HEAD(&ze->post_ct_list, fe, post_ct_list);
+
+	if (wildcard) {
+		while (rte_hash_iterate(priv->ct_zone_table, &next_key, &next_data, &iter) >= 0) {
+			ze = (struct nfp_ct_zone_entry *)next_data;
+			ret = nfp_ct_merge_flow_entries(fe, ze, ze);
+			if (!ret) {
+				PMD_DRV_LOG(ERR, "Merge ct flow entries wildcast failed");
+				break;
+			}
+		}
+	} else {
+		ret = nfp_ct_merge_flow_entries(fe, ze, ze);
+	}
+
+	if (!ret)
+		goto ct_flow_entry_free;
+
+	/* The real offload logic comes in next commit, so here just return false for now */
+
+ct_flow_entry_free:
+	nfp_ct_flow_entry_destroy(fe);
+
+ct_zone_entry_free:
+	nfp_ct_zone_entry_free(ze, wildcard);
+
+	return false;
+}
+
+struct rte_flow *
+nfp_ct_flow_setup(struct nfp_flower_representor *representor,
+		const struct rte_flow_item items[],
+		const struct rte_flow_action actions[],
+		const struct rte_flow_item *ct_item,
+		bool validate_flag,
+		uint64_t cookie)
+{
+	const struct ct_data *ct;
+
+	if (ct_item == NULL)
+		return NULL;
+
+	ct = ct_item->spec;
+
+	if (is_ct_commit_flow(ct)) {
+		return nfp_flow_process(representor, &items[1], actions,
+				validate_flag, cookie, false);
+	}
+
+	if (is_post_ct_flow(ct)) {
+		if (nfp_flow_handle_post_ct(ct_item, representor, &items[1],
+				actions, cookie)) {
+			return nfp_flow_process(representor, &items[1], actions,
+					validate_flag, cookie, false);
+		}
+
+		PMD_DRV_LOG(ERR, "Handle nfp post ct flow failed.");
+		return NULL;
+	}
+
+	if (is_pre_ct_flow(ct, actions)) {
+		if (nfp_flow_handle_pre_ct(ct_item, representor, &items[1],
+				actions, cookie)) {
+			return nfp_flow_process(representor, &items[1], actions,
+					validate_flag, cookie, false);
+		}
+
+		PMD_DRV_LOG(ERR, "Handle nfp pre ct flow failed.");
+		return NULL;
+	}
+
+	PMD_DRV_LOG(ERR, "Unsupported ct flow type.");
+	return NULL;
+}
diff --git a/drivers/net/nfp/flower/nfp_conntrack.h b/drivers/net/nfp/flower/nfp_conntrack.h
new file mode 100644
index 0000000000..149a3eb040
--- /dev/null
+++ b/drivers/net/nfp/flower/nfp_conntrack.h
@@ -0,0 +1,32 @@ 
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2023 Corigine, Inc.
+ * All rights reserved.
+ */
+
+#ifndef __NFP_CONNTRACK_H__
+#define __NFP_CONNTRACK_H__
+
+#include <stdbool.h>
+
+#include <rte_flow.h>
+
+#include "../nfp_flow.h"
+
+struct nfp_ct_map_entry;
+
+struct nfp_ct_zone_entry;
+
+struct nfp_ct_merge_entry;
+
+struct nfp_ct_map_entry *nfp_ct_map_table_search(struct nfp_flow_priv *priv,
+		char *hash_data,
+		uint32_t hash_len);
+
+struct rte_flow *nfp_ct_flow_setup(struct nfp_flower_representor *representor,
+		const struct rte_flow_item items[],
+		const struct rte_flow_action actions[],
+		const struct rte_flow_item *ct_item,
+		bool validate_flag,
+		uint64_t cookie);
+
+#endif /* __NFP_CONNTRACK_H__ */
diff --git a/drivers/net/nfp/meson.build b/drivers/net/nfp/meson.build
index 3912566134..7627c3e3f1 100644
--- a/drivers/net/nfp/meson.build
+++ b/drivers/net/nfp/meson.build
@@ -6,6 +6,7 @@  if not is_linux or not dpdk_conf.get('RTE_ARCH_64')
     reason = 'only supported on 64-bit Linux'
 endif
 sources = files(
+        'flower/nfp_conntrack.c',
         'flower/nfp_flower.c',
         'flower/nfp_flower_cmsg.c',
         'flower/nfp_flower_ctrl.c',
diff --git a/drivers/net/nfp/nfp_flow.c b/drivers/net/nfp/nfp_flow.c
index 1bb93bcfb5..16a5c7e055 100644
--- a/drivers/net/nfp/nfp_flow.c
+++ b/drivers/net/nfp/nfp_flow.c
@@ -10,6 +10,7 @@ 
 #include <rte_jhash.h>
 #include <rte_malloc.h>
 
+#include "flower/nfp_conntrack.h"
 #include "flower/nfp_flower_representor.h"
 #include "nfpcore/nfp_rtsym.h"
 #include "nfp_logs.h"
@@ -3748,6 +3749,8 @@  nfp_flow_setup(struct nfp_flower_representor *representor,
 		bool validate_flag)
 {
 	uint64_t cookie;
+	const struct rte_flow_item *item;
+	const struct rte_flow_item *ct_item = NULL;
 
 	if (attr->group != 0)
 		PMD_DRV_LOG(INFO, "Pretend we support group attribute.");
@@ -3758,8 +3761,19 @@  nfp_flow_setup(struct nfp_flower_representor *representor,
 	if (attr->transfer != 0)
 		PMD_DRV_LOG(INFO, "Pretend we support transfer attribute.");
 
+	for (item = items; item->type != RTE_FLOW_ITEM_TYPE_END; ++item) {
+		if (item->type == RTE_FLOW_ITEM_TYPE_CONNTRACK) {
+			ct_item = item;
+			break;
+		}
+	}
+
 	cookie = rte_rand();
 
+	if (ct_item != NULL)
+		return nfp_ct_flow_setup(representor, items, actions,
+				ct_item, validate_flag, cookie);
+
 	return nfp_flow_process(representor, items, actions, validate_flag, cookie, true);
 }
 
@@ -4235,6 +4249,23 @@  nfp_flow_priv_init(struct nfp_pf_dev *pf_dev)
 		.extra_flag = RTE_HASH_EXTRA_FLAGS_RW_CONCURRENCY,
 	};
 
+	struct rte_hash_parameters ct_zone_hash_params = {
+		.name       = "ct_zone_table",
+		.entries    = 65536,
+		.hash_func  = rte_jhash,
+		.socket_id  = rte_socket_id(),
+		.key_len    = sizeof(uint32_t),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_RW_CONCURRENCY,
+	};
+
+	struct rte_hash_parameters ct_map_hash_params = {
+		.name       = "ct_map_table",
+		.hash_func  = rte_jhash,
+		.socket_id  = rte_socket_id(),
+		.key_len    = sizeof(uint32_t),
+		.extra_flag = RTE_HASH_EXTRA_FLAGS_RW_CONCURRENCY,
+	};
+
 	ctx_count = nfp_rtsym_read_le(pf_dev->sym_tbl,
 			"CONFIG_FC_HOST_CTX_COUNT", &ret);
 	if (ret < 0) {
@@ -4325,6 +4356,25 @@  nfp_flow_priv_init(struct nfp_pf_dev *pf_dev)
 		goto free_flow_table;
 	}
 
+	/* ct zone table */
+	ct_zone_hash_params.hash_func_init_val = priv->hash_seed;
+	priv->ct_zone_table = rte_hash_create(&ct_zone_hash_params);
+	if (priv->ct_zone_table == NULL) {
+		PMD_INIT_LOG(ERR, "ct zone table creation failed");
+		ret = -ENOMEM;
+		goto free_pre_tnl_table;
+	}
+
+	/* ct map table */
+	ct_map_hash_params.hash_func_init_val = priv->hash_seed;
+	ct_map_hash_params.entries = ctx_count;
+	priv->ct_map_table = rte_hash_create(&ct_map_hash_params);
+	if (priv->ct_map_table == NULL) {
+		PMD_INIT_LOG(ERR, "ct map table creation failed");
+		ret = -ENOMEM;
+		goto free_ct_zone_table;
+	}
+
 	/* ipv4 off list */
 	rte_spinlock_init(&priv->ipv4_off_lock);
 	LIST_INIT(&priv->ipv4_off_list);
@@ -4338,6 +4388,10 @@  nfp_flow_priv_init(struct nfp_pf_dev *pf_dev)
 
 	return 0;
 
+free_ct_zone_table:
+	rte_hash_free(priv->ct_zone_table);
+free_pre_tnl_table:
+	rte_hash_free(priv->pre_tun_table);
 free_flow_table:
 	rte_hash_free(priv->flow_table);
 free_mask_table:
@@ -4363,6 +4417,8 @@  nfp_flow_priv_uninit(struct nfp_pf_dev *pf_dev)
 	app_fw_flower = NFP_PRIV_TO_APP_FW_FLOWER(pf_dev->app_fw_priv);
 	priv = app_fw_flower->flow_priv;
 
+	rte_hash_free(priv->ct_map_table);
+	rte_hash_free(priv->ct_zone_table);
 	rte_hash_free(priv->pre_tun_table);
 	rte_hash_free(priv->flow_table);
 	rte_hash_free(priv->mask_table);
diff --git a/drivers/net/nfp/nfp_flow.h b/drivers/net/nfp/nfp_flow.h
index 817eaecba2..df16cab8b5 100644
--- a/drivers/net/nfp/nfp_flow.h
+++ b/drivers/net/nfp/nfp_flow.h
@@ -150,6 +150,10 @@  struct nfp_flow_priv {
 	rte_spinlock_t ipv6_off_lock; /**< Lock the ipv6 off list */
 	/* neighbor next */
 	LIST_HEAD(, nfp_fl_tun)nn_list; /**< Store nn entry */
+	/* Conntrack */
+	struct rte_hash *ct_zone_table; /**< Hash table to store ct zone entry */
+	struct nfp_ct_zone_entry *ct_zone_wc; /**< The wildcard ct zone entry */
+	struct rte_hash *ct_map_table; /**< Hash table to store ct map entry */
 };
 
 struct rte_flow {