[dpdk-dev] [PATCH RFC V2 0/2] TCP flow classification using 4-tuple and flags

[dpdk-dev] [PATCH RFC V2 0/2] TCP flow classification using 4-tuple and flags

[Sowmini Varadhan]

V2 updates: checkpatch fixes, revert accidently spelling error
introduced in V1;

The problem space of TCP flow tracking and classification
based on TCP state requires the ability to classify TCP
flows on more packet properties than just the 4-tuple,
e.g., TCP flags. This patch-set investigates the set of
changes needed in the examples/flow_classify.c needed to
achieve this.

Patch 1 extends examples/flow_classify.c to allow constraints
on tcp flags. Patch 2 extends the ACL handling in  
librte_flow_classify to include keys on the properties in 
addition to the tcp 4-tuple.

Note that one particular part of this patch-set where feedback
is requested is in allocate_acl_ipv4_tcp_5tuple_rule():
we need to add a key for the 8 bit flags, but the multibit
trie lookup moves in steps of 4 bytes, so it took some hackery
to figure out what byte-ordering was expected, and there were
no documentation/examples to provide guidelines. Comments/suggestions
would be particularly helpful.

Sowmini Varadhan (2):
  Hooks to allow the setting of filters on tcp flags
  Allow the flow_classify example to add an ACL table for tcp.

 examples/flow_classify/flow_classify.c        | 121 +++++++++++++++---
 examples/flow_classify/ipv4_rules_file.txt    |  22 ++--
 lib/librte_flow_classify/rte_flow_classify.c  |  87 +++++++++++++
 lib/librte_flow_classify/rte_flow_classify.h  |  19 +++
 .../rte_flow_classify_parse.c                 |   8 +-
 5 files changed, 230 insertions(+), 27 deletions(-)

-- 
2.20.1

[dpdk-dev] [PATCH RFC V2 1/2] Hooks to allow the setting of filters on tcp flags

[Sowmini Varadhan]

The rte_eth_ntuple_filter allows tcp_flags which can check for things
like
    #define RTE_TCP_CWR_FLAG 0x80 /**< Congestion Window Reduced */
    #define RTE_TCP_ECE_FLAG 0x40 /**< ECN-Echo */
    #define RTE_TCP_URG_FLAG 0x20 /**< Urgent Pointer field significant */
    #define RTE_TCP_ACK_FLAG 0x10 /**< Acknowledgment field significant */
    #define RTE_TCP_PSH_FLAG 0x08 /**< Push Function */
    #define RTE_TCP_RST_FLAG 0x04 /**< Reset the connection */
    #define RTE_TCP_SYN_FLAG 0x02 /**< Synchronize sequence numbers */
    #define RTE_TCP_FIN_FLAG 0x01 /**< No more data from sender */
but there are no existing examples that demonstrate how to use this
feature.

This patch extends the exisiting classification support to allow
an optional flags in the input file. The flags string can be any
concatenation of characters from {C, E, U, A, P, R, S, F} and
"*" indicates "dont care". These flags are set in the ntuple_filter and
are used to construct the tcp_spec and tcp_mask sent to the driver

The rte_acl_field_def is updated to use the (u8) tcp flag as lookup key.
Note that, as per
  https://doc.dpdk.org/guides/prog_guide/packet_classif_access_ctrl.html
this field MUST be allocated fo4 4 bytes, thus it has sizeof(uint32_t).

However, also note the XXX in this commit: additional updates are
needed to the rte_flow_classify_table_entry_add() so that it does
not ignore any key fields other than the 5-tuple.

Signed-off-by: Sowmini Varadhan <sowmini05@gmail.com>
---
V2 : checkpatch fixes; revert accidental spelling mistake introduced in V1.

 examples/flow_classify/flow_classify.c     | 87 ++++++++++++++++++++--
 examples/flow_classify/ipv4_rules_file.txt | 22 +++---
 2 files changed, 91 insertions(+), 18 deletions(-)

diff --git a/examples/flow_classify/flow_classify.c b/examples/flow_classify/flow_classify.c
index 1c12bbb2f..e8a41741f 100644
--- a/examples/flow_classify/flow_classify.c
+++ b/examples/flow_classify/flow_classify.c
@@ -51,6 +51,7 @@ enum {
 	CB_FLD_DST_PORT_MASK,
 	CB_FLD_PROTO,
 	CB_FLD_PRIORITY,
+	CB_FLD_TCP_FLAGS,
 	CB_FLD_NUM,
 };
 
@@ -81,6 +82,7 @@ enum {
 	DST_FIELD_IPV4,
 	SRCP_FIELD_IPV4,
 	DSTP_FIELD_IPV4,
+	TCP_FLAGS_FIELD,
 	NUM_FIELDS_IPV4
 };
 
@@ -88,7 +90,8 @@ enum {
 	PROTO_INPUT_IPV4,
 	SRC_INPUT_IPV4,
 	DST_INPUT_IPV4,
-	SRCP_DESTP_INPUT_IPV4
+	SRCP_DESTP_INPUT_IPV4,
+	TCP_FLAGS_INDEX,
 };
 
 static struct rte_acl_field_def ipv4_defs[NUM_FIELDS_IPV4] = {
@@ -145,6 +148,17 @@ static struct rte_acl_field_def ipv4_defs[NUM_FIELDS_IPV4] = {
 			sizeof(struct rte_ipv4_hdr) +
 			offsetof(struct rte_tcp_hdr, dst_port),
 	},
+	/* next field must be 4 bytes, even though flags is only 1 byte */
+	{
+		/* rte_flags */
+		.type = RTE_ACL_FIELD_TYPE_BITMASK,
+		.size = sizeof(uint32_t),
+		.field_index = TCP_FLAGS_FIELD,
+		.input_index = TCP_FLAGS_INDEX,
+		.offset = sizeof(struct rte_ether_hdr) +
+			sizeof(struct rte_ipv4_hdr) +
+			offsetof(struct rte_tcp_hdr, tcp_flags),
+	},
 };
 
 /* flow classify data */
@@ -272,12 +286,14 @@ lcore_main(struct flow_classifier *cls_app)
 	int ret;
 	int i = 0;
 
-	ret = rte_flow_classify_table_entry_delete(cls_app->cls,
-			rules[7]);
-	if (ret)
-		printf("table_entry_delete failed [7] %d\n\n", ret);
-	else
-		printf("table_entry_delete succeeded [7]\n\n");
+	if (rules[7]) {
+		ret = rte_flow_classify_table_entry_delete(cls_app->cls,
+				rules[7]);
+		if (ret)
+			printf("table_entry_delete failed [7] %d\n\n", ret);
+		else
+			printf("table_entry_delete succeeded [7]\n\n");
+	}
 
 	/*
 	 * Check that the port is on the same NUMA node as the polling thread
@@ -395,6 +411,53 @@ parse_ipv4_net(char *in, uint32_t *addr, uint32_t *mask_len)
 	return 0;
 }
 
+static int
+get_tcp_flags(char *in, struct rte_eth_ntuple_filter *ntuple_filter)
+{
+	int len = strlen(in);
+	int i;
+	uint8_t flags = 0;
+
+	if (strcmp(in, "*") == 0) {
+		ntuple_filter->tcp_flags = 0;
+		return 0;
+	}
+
+	for (i = 0; i < len; i++) {
+		switch (in[i]) {
+		case 'S':
+			flags |= RTE_TCP_SYN_FLAG;
+			break;
+		case 'F':
+			flags |= RTE_TCP_FIN_FLAG;
+			break;
+		case 'R':
+			flags |= RTE_TCP_RST_FLAG;
+			break;
+		case 'P':
+			flags |= RTE_TCP_PSH_FLAG;
+			break;
+		case 'A':
+			flags |= RTE_TCP_ACK_FLAG;
+			break;
+		case 'U':
+			flags |= RTE_TCP_URG_FLAG;
+			break;
+		case 'E':
+			flags |= RTE_TCP_ECE_FLAG;
+			break;
+		case 'C':
+			flags |= RTE_TCP_CWR_FLAG;
+			break;
+		default:
+			fprintf(stderr, "unknown flag %c\n", in[i]);
+			return -1;
+		}
+	}
+	ntuple_filter->tcp_flags = flags;
+	return 0;
+}
+
 static int
 parse_ipv4_5tuple_rule(char *str, struct rte_eth_ntuple_filter *ntuple_filter)
 {
@@ -466,6 +529,8 @@ parse_ipv4_5tuple_rule(char *str, struct rte_eth_ntuple_filter *ntuple_filter)
 	ntuple_filter->priority = (uint16_t)temp;
 	if (ntuple_filter->priority > FLOW_CLASSIFY_MAX_PRIORITY)
 		ret = -EINVAL;
+	if (get_tcp_flags(in[CB_FLD_TCP_FLAGS], ntuple_filter))
+		return -EINVAL;
 
 	return ret;
 }
@@ -582,10 +647,13 @@ add_classify_rule(struct rte_eth_ntuple_filter *ntuple_filter,
 		memset(&tcp_spec, 0, sizeof(tcp_spec));
 		tcp_spec.hdr.src_port = ntuple_filter->src_port;
 		tcp_spec.hdr.dst_port = ntuple_filter->dst_port;
+		tcp_spec.hdr.tcp_flags = ntuple_filter->tcp_flags;
 
 		memset(&tcp_mask, 0, sizeof(tcp_mask));
 		tcp_mask.hdr.src_port = ntuple_filter->src_port_mask;
 		tcp_mask.hdr.dst_port = ntuple_filter->dst_port_mask;
+		if (tcp_spec.hdr.tcp_flags != 0)
+			tcp_mask.hdr.tcp_flags = 0xff;
 
 		tcp_item.type = RTE_FLOW_ITEM_TYPE_TCP;
 		tcp_item.spec = &tcp_spec;
@@ -640,6 +708,11 @@ add_classify_rule(struct rte_eth_ntuple_filter *ntuple_filter,
 		return ret;
 	}
 
+	/* XXX but this only adds table_type of
+	 * RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE
+	 * i.e., it only ever does allocate_acl_ipv4_5tuple_rule()
+	 * so the tcp_flags is ignored!
+	 */
 	rule = rte_flow_classify_table_entry_add(
 			cls_app->cls, &attr, pattern_ipv4_5tuple,
 			actions, &key_found, &error);
diff --git a/examples/flow_classify/ipv4_rules_file.txt b/examples/flow_classify/ipv4_rules_file.txt
index dfa0631fc..415573732 100644
--- a/examples/flow_classify/ipv4_rules_file.txt
+++ b/examples/flow_classify/ipv4_rules_file.txt
@@ -1,14 +1,14 @@
 #file format:
-#src_ip/masklen dst_ip/masklen src_port : mask dst_port : mask proto/mask priority
+#src_ip/masklen dst_ip/masklen src_port : mask dst_port : mask proto/mask priority tcp_flags
 #
-2.2.2.3/24 2.2.2.7/24 32 : 0xffff 33 : 0xffff 17/0xff 0
-9.9.9.3/24 9.9.9.7/24 32 : 0xffff 33 : 0xffff 17/0xff 1
-9.9.9.3/24 9.9.9.7/24 32 : 0xffff 33 : 0xffff 6/0xff 2
-9.9.8.3/24 9.9.8.7/24 32 : 0xffff 33 : 0xffff 6/0xff 3
-6.7.8.9/24 2.3.4.5/24 32 : 0x0000 33 : 0x0000 132/0xff 4
-6.7.8.9/32 192.168.0.36/32 10 : 0xffff 11 : 0xffff 6/0xfe 5
-6.7.8.9/24 192.168.0.36/24 10 : 0xffff 11 : 0xffff 6/0xfe 6
-6.7.8.9/16 192.168.0.36/16 10 : 0xffff 11 : 0xffff 6/0xfe 7
-6.7.8.9/8 192.168.0.36/8 10 : 0xffff 11 : 0xffff 6/0xfe 8
+2.2.2.3/24 2.2.2.7/24 32 : 0xffff 33 : 0xffff 17/0xff 0 *
+9.9.9.3/24 9.9.9.7/24 32 : 0xffff 33 : 0xffff 17/0xff 1 *
+9.9.9.3/24 9.9.9.7/24 32 : 0xffff 33 : 0xffff 6/0xff 2 *
+9.9.8.3/24 9.9.8.7/24 32 : 0xffff 33 : 0xffff 6/0xff 3 *
+6.7.8.9/24 2.3.4.5/24 32 : 0x0000 33 : 0x0000 132/0xff 4 *
+6.7.8.9/32 192.168.0.36/32 10 : 0xffff 11 : 0xffff 6/0xfe 5 *
+6.7.8.9/24 192.168.0.36/24 10 : 0xffff 11 : 0xffff 6/0xfe 6 *
+6.7.8.9/16 192.168.0.36/16 10 : 0xffff 11 : 0xffff 6/0xfe 7 *
+6.7.8.9/8 192.168.0.36/8 10 : 0xffff 11 : 0xffff 6/0xfe 8 *
 #error rules
-#9.8.7.6/8 192.168.0.36/8 10 : 0xffff 11 : 0xffff 6/0xfe 9
\ No newline at end of file
+#9.8.7.6/8 192.168.0.36/8 10 : 0xffff 11 : 0xffff 6/0xfe 9 *
-- 
2.20.1

[dpdk-dev] [PATCH RFC V2 2/2] Allow the flow_classify example to add an ACL table for tcp.

[Sowmini Varadhan]

The struct rte_flow_classifier can have upto RTE_FLOW_CLASSIFY_TABLE_MAX
(32) classifier tables, but the existing flow_classify examples only adds
a single table for the L4 5-tuple.

When dealing with tcp flows, we frequently want to add add ACLs and filters
to filter based on the state of the TCP connection, e.g., by looking at
the tcp flags field.

So we enhance flow_classify to add an additional acl table for
tcp 5-typles. If the input-file-parser detects a filter for a tcp flow with
a non-wildcard argument for tcp_flags, the IP4_TCP_5TUPLE table is used
by flow_classify.

Signed-off-by: Sowmini Varadhan <sowmini05@gmail.com>
---
V2 updates: checkpatch fixes

 examples/flow_classify/flow_classify.c        | 44 +++++++---
 lib/librte_flow_classify/rte_flow_classify.c  | 87 +++++++++++++++++++
 lib/librte_flow_classify/rte_flow_classify.h  | 19 ++++
 .../rte_flow_classify_parse.c                 |  8 +-
 4 files changed, 144 insertions(+), 14 deletions(-)

diff --git a/examples/flow_classify/flow_classify.c b/examples/flow_classify/flow_classify.c
index e8a41741f..a0d138eb5 100644
--- a/examples/flow_classify/flow_classify.c
+++ b/examples/flow_classify/flow_classify.c
@@ -708,11 +708,6 @@ add_classify_rule(struct rte_eth_ntuple_filter *ntuple_filter,
 		return ret;
 	}
 
-	/* XXX but this only adds table_type of
-	 * RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE
-	 * i.e., it only ever does allocate_acl_ipv4_5tuple_rule()
-	 * so the tcp_flags is ignored!
-	 */
 	rule = rte_flow_classify_table_entry_add(
 			cls_app->cls, &attr, pattern_ipv4_5tuple,
 			actions, &key_found, &error);
@@ -839,7 +834,8 @@ main(int argc, char *argv[])
 	int ret;
 	int socket_id;
 	struct rte_table_acl_params table_acl_params;
-	struct rte_flow_classify_table_params cls_table_params;
+	struct rte_table_acl_params table_acl_tcp_params;
+	struct rte_flow_classify_table_params cls_table_params[2];
 	struct flow_classifier *cls_app;
 	struct rte_flow_classifier_params cls_params;
 	uint32_t size;
@@ -902,20 +898,42 @@ main(int argc, char *argv[])
 	memcpy(table_acl_params.field_format, ipv4_defs, sizeof(ipv4_defs));
 
 	/* initialise table create params */
-	cls_table_params.ops = &rte_table_acl_ops;
-	cls_table_params.arg_create = &table_acl_params;
-	cls_table_params.type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE;
+	cls_table_params[0].ops = &rte_table_acl_ops;
+	cls_table_params[0].arg_create = &table_acl_params;
+	cls_table_params[0].type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE;
+
+	/* initialise ACL table params */
+	table_acl_tcp_params.name = "table_acl_ipv4_tcp_5tuple";
+	table_acl_tcp_params.n_rules = FLOW_CLASSIFY_MAX_RULE_NUM;
+	table_acl_tcp_params.n_rule_fields = RTE_DIM(ipv4_defs);
+	memcpy(table_acl_tcp_params.field_format, ipv4_defs, sizeof(ipv4_defs));
+
+	/* initialise table create params */
+	cls_table_params[1].ops = &rte_table_acl_ops;
+	cls_table_params[1].arg_create = &table_acl_tcp_params;
+	cls_table_params[1].type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_TCP_5TUPLE;
 
-	ret = rte_flow_classify_table_create(cls_app->cls, &cls_table_params);
+	ret = rte_flow_classify_table_create(cls_app->cls,
+					     &cls_table_params[0]);
 	if (ret) {
 		rte_flow_classifier_free(cls_app->cls);
 		rte_free(cls_app);
-		rte_exit(EXIT_FAILURE, "Failed to create classifier table\n");
+		rte_exit(EXIT_FAILURE,
+			 "Failed to create classifier table\n");
 	}
+	ret = rte_flow_classify_table_create(cls_app->cls,
+					     &cls_table_params[1]);
+	if (ret) {
+		rte_flow_classifier_free(cls_app->cls);
+		rte_free(cls_app);
+		rte_exit(EXIT_FAILURE,
+			 "Failed to create classifier table\n");
+	}
+
 
 	/* read file of IPv4 5 tuple rules and initialize parameters
-	 * for rte_flow_classify_validate and rte_flow_classify_table_entry_add
-	 * API's.
+	 * for rte_flow_classify_validate and
+	 * rte_flow_classify_table_entry_add  API's.
 	 */
 	if (add_rules(parm_config.rule_ipv4_name, cls_app)) {
 		rte_flow_classifier_free(cls_app->cls);
diff --git a/lib/librte_flow_classify/rte_flow_classify.c b/lib/librte_flow_classify/rte_flow_classify.c
index 5ff585803..cda1a6129 100644
--- a/lib/librte_flow_classify/rte_flow_classify.c
+++ b/lib/librte_flow_classify/rte_flow_classify.c
@@ -62,6 +62,7 @@ enum {
 	DST_FIELD_IPV4,
 	SRCP_FIELD_IPV4,
 	DSTP_FIELD_IPV4,
+	TCP_FLAGS_FIELD,
 	NUM_FIELDS_IPV4
 };
 
@@ -74,6 +75,7 @@ struct classify_rules {
 	enum rte_flow_classify_rule_type type;
 	union {
 		struct rte_flow_classify_ipv4_5tuple ipv4_5tuple;
+		struct rte_flow_classify_ipv4_tcp_5tuple ipv4_tcp_5tuple;
 	} u;
 };
 
@@ -482,6 +484,84 @@ allocate_acl_ipv4_5tuple_rule(struct rte_flow_classifier *cls)
 	return rule;
 }
 
+static struct rte_flow_classify_rule *
+allocate_acl_ipv4_tcp_5tuple_rule(struct rte_flow_classifier *cls)
+{
+	struct rte_flow_classify_rule *rule;
+	int log_level;
+
+	rule = malloc(sizeof(struct rte_flow_classify_rule));
+	if (!rule)
+		return rule;
+
+	memset(rule, 0, sizeof(struct rte_flow_classify_rule));
+	rule->id = unique_id++;
+	rule->rules.type = RTE_FLOW_CLASSIFY_RULE_TYPE_IPV4_TCP_5TUPLE;
+
+	/* key add values */
+	rule->u.key.key_add.priority = cls->ntuple_filter.priority;
+	rule->u.key.key_add.field_value[PROTO_FIELD_IPV4].mask_range.u8 =
+			cls->ntuple_filter.proto_mask;
+	rule->u.key.key_add.field_value[PROTO_FIELD_IPV4].value.u8 =
+			cls->ntuple_filter.proto;
+	rule->rules.u.ipv4_tcp_5tuple.proto = cls->ntuple_filter.proto;
+	rule->rules.u.ipv4_tcp_5tuple.proto_mask =
+			cls->ntuple_filter.proto_mask;
+
+	rule->u.key.key_add.field_value[SRC_FIELD_IPV4].mask_range.u32 =
+			cls->ntuple_filter.src_ip_mask;
+	rule->u.key.key_add.field_value[SRC_FIELD_IPV4].value.u32 =
+			cls->ntuple_filter.src_ip;
+	rule->rules.u.ipv4_tcp_5tuple.src_ip_mask =
+			cls->ntuple_filter.src_ip_mask;
+	rule->rules.u.ipv4_tcp_5tuple.src_ip = cls->ntuple_filter.src_ip;
+
+	rule->u.key.key_add.field_value[DST_FIELD_IPV4].mask_range.u32 =
+			cls->ntuple_filter.dst_ip_mask;
+	rule->u.key.key_add.field_value[DST_FIELD_IPV4].value.u32 =
+			cls->ntuple_filter.dst_ip;
+	rule->rules.u.ipv4_tcp_5tuple.dst_ip_mask =
+			cls->ntuple_filter.dst_ip_mask;
+	rule->rules.u.ipv4_tcp_5tuple.dst_ip = cls->ntuple_filter.dst_ip;
+
+	rule->u.key.key_add.field_value[SRCP_FIELD_IPV4].mask_range.u16 =
+			cls->ntuple_filter.src_port_mask;
+	rule->u.key.key_add.field_value[SRCP_FIELD_IPV4].value.u16 =
+			cls->ntuple_filter.src_port;
+	rule->rules.u.ipv4_tcp_5tuple.src_port_mask =
+			cls->ntuple_filter.src_port_mask;
+	rule->rules.u.ipv4_tcp_5tuple.src_port = cls->ntuple_filter.src_port;
+
+	rule->u.key.key_add.field_value[DSTP_FIELD_IPV4].mask_range.u16 =
+			cls->ntuple_filter.dst_port_mask;
+	rule->u.key.key_add.field_value[DSTP_FIELD_IPV4].value.u16 =
+			cls->ntuple_filter.dst_port;
+	rule->rules.u.ipv4_tcp_5tuple.dst_port_mask =
+			cls->ntuple_filter.dst_port_mask;
+	rule->rules.u.ipv4_tcp_5tuple.dst_port = cls->ntuple_filter.dst_port;
+
+	rule->u.key.key_add.field_value[TCP_FLAGS_FIELD].mask_range.u32 =
+			rte_be_to_cpu_32(0xff);
+	rule->u.key.key_add.field_value[TCP_FLAGS_FIELD].value.u32 =
+			rte_be_to_cpu_32(cls->ntuple_filter.tcp_flags);
+	rule->rules.u.ipv4_tcp_5tuple.tcp_flags = cls->ntuple_filter.tcp_flags;
+
+	log_level = rte_log_get_level(librte_flow_classify_logtype);
+
+	if (log_level == RTE_LOG_DEBUG)
+		print_acl_ipv4_key_add(&rule->u.key.key_add);
+
+	/* key delete values */
+	memcpy(&rule->u.key.key_del.field_value[PROTO_FIELD_IPV4],
+	       &rule->u.key.key_add.field_value[PROTO_FIELD_IPV4],
+	       NUM_FIELDS_IPV4 * sizeof(struct rte_acl_field));
+
+	if (log_level == RTE_LOG_DEBUG)
+		print_acl_ipv4_key_delete(&rule->u.key.key_del);
+
+	return rule;
+}
+
 struct rte_flow_classify_rule *
 rte_flow_classify_table_entry_add(struct rte_flow_classifier *cls,
 		const struct rte_flow_attr *attr,
@@ -519,6 +599,13 @@ rte_flow_classify_table_entry_add(struct rte_flow_classifier *cls,
 		rule->tbl_type = table_type;
 		cls->table_mask |= table_type;
 		break;
+	case RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_TCP_5TUPLE:
+		rule = allocate_acl_ipv4_tcp_5tuple_rule(cls);
+		if (!rule)
+			return NULL;
+		rule->tbl_type = table_type;
+		cls->table_mask |= table_type;
+		break;
 	default:
 		return NULL;
 	}
diff --git a/lib/librte_flow_classify/rte_flow_classify.h b/lib/librte_flow_classify/rte_flow_classify.h
index 74d1ecaf5..921277f90 100644
--- a/lib/librte_flow_classify/rte_flow_classify.h
+++ b/lib/librte_flow_classify/rte_flow_classify.h
@@ -78,6 +78,8 @@ enum rte_flow_classify_rule_type {
 	RTE_FLOW_CLASSIFY_RULE_TYPE_NONE,
 	/** IPv4 5tuple type */
 	RTE_FLOW_CLASSIFY_RULE_TYPE_IPV4_5TUPLE,
+	/** IPv4 TCP 5tuple type */
+	RTE_FLOW_CLASSIFY_RULE_TYPE_IPV4_TCP_5TUPLE,
 };
 
 /** Flow classify table type */
@@ -90,6 +92,8 @@ enum rte_flow_classify_table_type {
 	RTE_FLOW_CLASSIFY_TABLE_ACL_VLAN_IP4_5TUPLE = 1 << 2,
 	/** ACL QinQ IP4 5TUPLE */
 	RTE_FLOW_CLASSIFY_TABLE_ACL_QINQ_IP4_5TUPLE = 1 << 3,
+	/** ACL IP4 5TUPLE with tcp_flags */
+	RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_TCP_5TUPLE = 1 << 4,
 
 };
 
@@ -129,6 +133,21 @@ struct rte_flow_classify_ipv4_5tuple {
 	uint8_t proto_mask;      /**< Mask of L4 protocol. */
 };
 
+/** IPv4 5-tuple data with tcp flags*/
+struct rte_flow_classify_ipv4_tcp_5tuple {
+	uint32_t dst_ip;         /**< Destination IP address in big endian. */
+	uint32_t dst_ip_mask;    /**< Mask of destination IP address. */
+	uint32_t src_ip;         /**< Source IP address in big endian. */
+	uint32_t src_ip_mask;    /**< Mask of destination IP address. */
+	uint16_t dst_port;       /**< Destination port in big endian. */
+	uint16_t dst_port_mask;  /**< Mask of destination port. */
+	uint16_t src_port;       /**< Source Port in big endian. */
+	uint16_t src_port_mask;  /**< Mask of source port. */
+	uint8_t proto;           /**< L4 protocol. */
+	uint8_t proto_mask;      /**< Mask of L4 protocol. */
+	uint8_t tcp_flags;       /**< Tcp only */
+};
+
 /**
  * Flow stats
  *
diff --git a/lib/librte_flow_classify/rte_flow_classify_parse.c b/lib/librte_flow_classify/rte_flow_classify_parse.c
index 465330291..fe4ee05b6 100644
--- a/lib/librte_flow_classify/rte_flow_classify_parse.c
+++ b/lib/librte_flow_classify/rte_flow_classify_parse.c
@@ -216,6 +216,7 @@ classify_parse_ntuple_filter(const struct rte_flow_attr *attr,
 	const struct rte_flow_action_count *count;
 	const struct rte_flow_action_mark *mark_spec;
 	uint32_t index;
+	bool have_tcp = false;
 
 	/* parse pattern */
 	index = 0;
@@ -375,6 +376,8 @@ classify_parse_ntuple_filter(const struct rte_flow_attr *attr,
 		filter->dst_port  = tcp_spec->hdr.dst_port;
 		filter->src_port  = tcp_spec->hdr.src_port;
 		filter->tcp_flags = tcp_spec->hdr.tcp_flags;
+		if (filter->tcp_flags != 0)
+			have_tcp = true;
 	} else if (item->type == RTE_FLOW_ITEM_TYPE_UDP) {
 		udp_mask = item->mask;
 
@@ -434,7 +437,10 @@ classify_parse_ntuple_filter(const struct rte_flow_attr *attr,
 		return -EINVAL;
 	}
 
-	table_type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE;
+	if (have_tcp)
+		table_type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_TCP_5TUPLE;
+	else
+		table_type = RTE_FLOW_CLASSIFY_TABLE_ACL_IP4_5TUPLE;
 
 	/* parse attr */
 	/* must be input direction */
-- 
2.20.1

[dpdk-dev] [PATCH RFC V2 2/2] Allow the flow_classify example to add an ACL table for tcp.

[Bernard Iremonger]

Hi Sowmini,

Could you rebase this patchset to the latest DPDK-21.05-rc1  code.

Regards,

Bernard

Re: [dpdk-dev] [PATCH RFC V2 0/2] TCP flow classification using 4-tuple and flags

[Stephen Hemminger]

On Mon, 13 Jan 2020 18:05:28 +0000
Sowmini Varadhan <sowmini05@gmail.com> wrote:

> V2 updates: checkpatch fixes, revert accidently spelling error
> introduced in V1;
> 
> The problem space of TCP flow tracking and classification
> based on TCP state requires the ability to classify TCP
> flows on more packet properties than just the 4-tuple,
> e.g., TCP flags. This patch-set investigates the set of
> changes needed in the examples/flow_classify.c needed to
> achieve this.
> 
> Patch 1 extends examples/flow_classify.c to allow constraints
> on tcp flags. Patch 2 extends the ACL handling in  
> librte_flow_classify to include keys on the properties in 
> addition to the tcp 4-tuple.
> 
> Note that one particular part of this patch-set where feedback
> is requested is in allocate_acl_ipv4_tcp_5tuple_rule():
> we need to add a key for the 8 bit flags, but the multibit
> trie lookup moves in steps of 4 bytes, so it took some hackery
> to figure out what byte-ordering was expected, and there were
> no documentation/examples to provide guidelines. Comments/suggestions
> would be particularly helpful.
> 
> Sowmini Varadhan (2):
>   Hooks to allow the setting of filters on tcp flags
>   Allow the flow_classify example to add an ACL table for tcp.
> 
>  examples/flow_classify/flow_classify.c        | 121 +++++++++++++++---
>  examples/flow_classify/ipv4_rules_file.txt    |  22 ++--
>  lib/librte_flow_classify/rte_flow_classify.c  |  87 +++++++++++++
>  lib/librte_flow_classify/rte_flow_classify.h  |  19 +++
>  .../rte_flow_classify_parse.c                 |   8 +-
>  5 files changed, 230 insertions(+), 27 deletions(-)
> 

Is anyone still interested in this patch?
It would need work for DPDK 23.08 or later code base.
For now, marking it as "Changes Requested"