net/enic: flow API for Legacy NICs
authorJohn Daley <johndale@cisco.com>
Wed, 17 May 2017 22:38:09 +0000 (15:38 -0700)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 12 Jun 2017 09:41:26 +0000 (10:41 +0100)
5-tuple exact Flow support for 1200 series adapters. This allows:
Attributes: ingress
Items: ipv4, ipv6, udp, tcp (must exactly match src/dst IP
       addresses and ports and all must be specified).
Actions: queue and void
Selectors: 'is'

Signed-off-by: John Daley <johndale@cisco.com>
Reviewed-by: Nelson Escobar <neescoba@cisco.com>
drivers/net/enic/enic_flow.c

index a32e25e..c20ff86 100644 (file)
@@ -89,6 +89,9 @@ struct enic_action_cap {
 };
 
 /* Forward declarations */
+static enic_copy_item_fn enic_copy_item_ipv4_v1;
+static enic_copy_item_fn enic_copy_item_udp_v1;
+static enic_copy_item_fn enic_copy_item_tcp_v1;
 static enic_copy_item_fn enic_copy_item_eth_v2;
 static enic_copy_item_fn enic_copy_item_vlan_v2;
 static enic_copy_item_fn enic_copy_item_ipv4_v2;
@@ -101,6 +104,36 @@ static enic_copy_item_fn enic_copy_item_vxlan_v2;
 static copy_action_fn enic_copy_action_v1;
 static copy_action_fn enic_copy_action_v2;
 
+/**
+ * Legacy NICs or NICs with outdated firmware. Only 5-tuple perfect match
+ * is supported.
+ */
+static const struct enic_items enic_items_v1[] = {
+       [RTE_FLOW_ITEM_TYPE_IPV4] = {
+               .copy_item = enic_copy_item_ipv4_v1,
+               .valid_start_item = 1,
+               .prev_items = (const enum rte_flow_item_type[]) {
+                              RTE_FLOW_ITEM_TYPE_END,
+               },
+       },
+       [RTE_FLOW_ITEM_TYPE_UDP] = {
+               .copy_item = enic_copy_item_udp_v1,
+               .valid_start_item = 0,
+               .prev_items = (const enum rte_flow_item_type[]) {
+                              RTE_FLOW_ITEM_TYPE_IPV4,
+                              RTE_FLOW_ITEM_TYPE_END,
+               },
+       },
+       [RTE_FLOW_ITEM_TYPE_TCP] = {
+               .copy_item = enic_copy_item_tcp_v1,
+               .valid_start_item = 0,
+               .prev_items = (const enum rte_flow_item_type[]) {
+                              RTE_FLOW_ITEM_TYPE_IPV4,
+                              RTE_FLOW_ITEM_TYPE_END,
+               },
+       },
+};
+
 /**
  * NICs have Advanced Filters capability but they are disabled. This means
  * that layer 3 must be specified.
@@ -252,6 +285,9 @@ static const struct enic_items enic_items_v3[] = {
 
 /** Filtering capabilities indexed this NICs supported filter type. */
 static const struct enic_filter_cap enic_filter_cap[] = {
+       [FILTER_IPV4_5TUPLE] = {
+               .item_info = enic_items_v1,
+       },
        [FILTER_USNIC_IP] = {
                .item_info = enic_items_v2,
        },
@@ -285,6 +321,171 @@ static const struct enic_action_cap enic_action_cap[] = {
                .copy_fn = enic_copy_action_v2,
        },
 };
+
+static int
+mask_exact_match(const u8 *supported, const u8 *supplied,
+                unsigned int size)
+{
+       unsigned int i;
+       for (i = 0; i < size; i++) {
+               if (supported[i] != supplied[i])
+                       return 0;
+       }
+       return 1;
+}
+
+/**
+ * Copy IPv4 item into version 1 NIC filter.
+ *
+ * @param item[in]
+ *   Item specification.
+ * @param enic_filter[out]
+ *   Partially filled in NIC filter structure.
+ * @param inner_ofst[in]
+ *   Should always be 0 for version 1.
+ */
+static int
+enic_copy_item_ipv4_v1(const struct rte_flow_item *item,
+                      struct filter_v2 *enic_filter, u8 *inner_ofst)
+{
+       const struct rte_flow_item_ipv4 *spec = item->spec;
+       const struct rte_flow_item_ipv4 *mask = item->mask;
+       struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
+       struct ipv4_hdr supported_mask = {
+               .src_addr = 0xffffffff,
+               .dst_addr = 0xffffffff,
+       };
+
+       FLOW_TRACE();
+
+       if (*inner_ofst)
+               return ENOTSUP;
+
+       if (!mask)
+               mask = &rte_flow_item_ipv4_mask;
+
+       /* This is an exact match filter, both fields must be set */
+       if (!spec || !spec->hdr.src_addr || !spec->hdr.dst_addr) {
+               FLOW_LOG(ERR, "IPv4 exact match src/dst addr");
+               return ENOTSUP;
+       }
+
+       /* check that the suppied mask exactly matches capabilty */
+       if (!mask_exact_match((const u8 *)&supported_mask,
+                             (const u8 *)item->mask, sizeof(*mask))) {
+               FLOW_LOG(ERR, "IPv4 exact match mask");
+               return ENOTSUP;
+       }
+
+       enic_filter->u.ipv4.flags = FILTER_FIELDS_IPV4_5TUPLE;
+       enic_5tup->src_addr = spec->hdr.src_addr;
+       enic_5tup->dst_addr = spec->hdr.dst_addr;
+
+       return 0;
+}
+
+/**
+ * Copy UDP item into version 1 NIC filter.
+ *
+ * @param item[in]
+ *   Item specification.
+ * @param enic_filter[out]
+ *   Partially filled in NIC filter structure.
+ * @param inner_ofst[in]
+ *   Should always be 0 for version 1.
+ */
+static int
+enic_copy_item_udp_v1(const struct rte_flow_item *item,
+                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+{
+       const struct rte_flow_item_udp *spec = item->spec;
+       const struct rte_flow_item_udp *mask = item->mask;
+       struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
+       struct udp_hdr supported_mask = {
+               .src_port = 0xffff,
+               .dst_port = 0xffff,
+       };
+
+       FLOW_TRACE();
+
+       if (*inner_ofst)
+               return ENOTSUP;
+
+       if (!mask)
+               mask = &rte_flow_item_udp_mask;
+
+       /* This is an exact match filter, both ports must be set */
+       if (!spec || !spec->hdr.src_port || !spec->hdr.dst_port) {
+               FLOW_LOG(ERR, "UDP exact match src/dst addr");
+               return ENOTSUP;
+       }
+
+       /* check that the suppied mask exactly matches capabilty */
+       if (!mask_exact_match((const u8 *)&supported_mask,
+                             (const u8 *)item->mask, sizeof(*mask))) {
+               FLOW_LOG(ERR, "UDP exact match mask");
+               return ENOTSUP;
+       }
+
+       enic_filter->u.ipv4.flags = FILTER_FIELDS_IPV4_5TUPLE;
+       enic_5tup->src_port = spec->hdr.src_port;
+       enic_5tup->dst_port = spec->hdr.dst_port;
+       enic_5tup->protocol = PROTO_UDP;
+
+       return 0;
+}
+
+/**
+ * Copy TCP item into version 1 NIC filter.
+ *
+ * @param item[in]
+ *   Item specification.
+ * @param enic_filter[out]
+ *   Partially filled in NIC filter structure.
+ * @param inner_ofst[in]
+ *   Should always be 0 for version 1.
+ */
+static int
+enic_copy_item_tcp_v1(const struct rte_flow_item *item,
+                     struct filter_v2 *enic_filter, u8 *inner_ofst)
+{
+       const struct rte_flow_item_tcp *spec = item->spec;
+       const struct rte_flow_item_tcp *mask = item->mask;
+       struct filter_ipv4_5tuple *enic_5tup = &enic_filter->u.ipv4;
+       struct tcp_hdr supported_mask = {
+               .src_port = 0xffff,
+               .dst_port = 0xffff,
+       };
+
+       FLOW_TRACE();
+
+       if (*inner_ofst)
+               return ENOTSUP;
+
+       if (!mask)
+               mask = &rte_flow_item_tcp_mask;
+
+       /* This is an exact match filter, both ports must be set */
+       if (!spec || !spec->hdr.src_port || !spec->hdr.dst_port) {
+               FLOW_LOG(ERR, "TCPIPv4 exact match src/dst addr");
+               return ENOTSUP;
+       }
+
+       /* check that the suppied mask exactly matches capabilty */
+       if (!mask_exact_match((const u8 *)&supported_mask,
+                            (const u8 *)item->mask, sizeof(*mask))) {
+               FLOW_LOG(ERR, "TCP exact match mask");
+               return ENOTSUP;
+       }
+
+       enic_filter->u.ipv4.flags = FILTER_FIELDS_IPV4_5TUPLE;
+       enic_5tup->src_port = spec->hdr.src_port;
+       enic_5tup->dst_port = spec->hdr.dst_port;
+       enic_5tup->protocol = PROTO_TCP;
+
+       return 0;
+}
+
 /**
  * Copy ETH item into version 2 NIC filter.
  *
@@ -878,11 +1079,6 @@ enic_match_action(const struct rte_flow_action *action,
 static const struct enic_filter_cap *
 enic_get_filter_cap(struct enic *enic)
 {
-       /* FIXME: only support advanced filters for now */
-       if ((enic->flow_filter_mode != FILTER_DPDK_1) &&
-          (enic->flow_filter_mode != FILTER_USNIC_IP))
-               return (const struct enic_filter_cap *)NULL;
-
        if (enic->flow_filter_mode)
                return &enic_filter_cap[enic->flow_filter_mode];