gso: packet coalesce library

Type: feature

Change-Id: Ia19d3611e596d9ec47509889b34e8fe793a0ccc3
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
diff --git a/src/vnet/gso/FEATURE.yaml b/src/vnet/gso/FEATURE.yaml
index 79b506d..d3db0cc 100644
--- a/src/vnet/gso/FEATURE.yaml
+++ b/src/vnet/gso/FEATURE.yaml
@@ -8,6 +8,8 @@
   - GSO for IP-IP tunnel
   - GSO for IPSec tunnel  
   - Provide inline function to get header offsets
+  - Basic GRO support
+  - Implements flow table support
 description: "Generic Segmentation Offload"
 missing:
   - Thorough Testing, GRE, Geneve
diff --git a/src/vnet/gso/gro.h b/src/vnet/gso/gro.h
new file mode 100644
index 0000000..bfa5920
--- /dev/null
+++ b/src/vnet/gso/gro.h
@@ -0,0 +1,292 @@
+/*
+ * Copyright (c) 2020 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef included_gro_h
+#define included_gro_h
+
+#include <vlib/vlib.h>
+#include <vppinfra/error.h>
+#include <vnet/ip/ip46_address.h>
+
+#define GRO_FLOW_TABLE_MAX_SIZE 16
+#define GRO_FLOW_TABLE_FLUSH 1e-5
+#define GRO_FLOW_N_BUFFERS 64
+#define GRO_FLOW_TIMEOUT 1e-5	/* 10 micro-seconds */
+#define GRO_TO_VECTOR_SIZE(X)   (X + GRO_FLOW_TABLE_MAX_SIZE)
+
+typedef union
+{
+  struct
+  {
+    u32 sw_if_index[VLIB_N_RX_TX];
+    ip46_address_t src_address;
+    ip46_address_t dst_address;
+    u16 src_port;
+    u16 dst_port;
+  };
+
+  u64 flow_data[5];
+  u32 flow_data_u32;
+} gro_flow_key_t;
+
+typedef struct
+{
+  gro_flow_key_t flow_key;
+  f64 next_timeout_ts;
+  u32 last_ack_number;
+  u32 buffer_index;
+  u16 n_buffers;
+} gro_flow_t;
+
+typedef struct
+{
+  f64 timeout_ts;
+  u64 total_vectors;
+  u32 n_vectors;
+  u32 node_index;
+  u8 is_enable;
+  u8 is_l2;
+  u8 flow_table_size;
+  gro_flow_t gro_flow[GRO_FLOW_TABLE_MAX_SIZE];
+} gro_flow_table_t;
+
+static_always_inline void
+gro_flow_set_flow_key (gro_flow_t * to, gro_flow_key_t * from)
+{
+  to->flow_key.flow_data[0] = from->flow_data[0];
+  to->flow_key.flow_data[1] = from->flow_data[1];
+  to->flow_key.flow_data[2] = from->flow_data[2];
+  to->flow_key.flow_data[3] = from->flow_data[3];
+  to->flow_key.flow_data[4] = from->flow_data[4];
+  to->flow_key.flow_data_u32 = from->flow_data_u32;
+}
+
+static_always_inline u8
+gro_flow_is_equal (gro_flow_key_t * first, gro_flow_key_t * second)
+{
+  if (first->flow_data[0] == second->flow_data[0] &&
+      first->flow_data[1] == second->flow_data[1] &&
+      first->flow_data[2] == second->flow_data[2] &&
+      first->flow_data[3] == second->flow_data[3] &&
+      first->flow_data[4] == second->flow_data[4] &&
+      first->flow_data_u32 == second->flow_data_u32)
+    return 1;
+
+  return 0;
+}
+
+/**
+ * timeout_expire is in between 3 to 10 microseconds
+ * 3e-6 1e-5
+ */
+static_always_inline void
+gro_flow_set_timeout (vlib_main_t * vm, gro_flow_t * gro_flow,
+		      f64 timeout_expire)
+{
+  gro_flow->next_timeout_ts = vlib_time_now (vm) + timeout_expire;
+}
+
+static_always_inline u8
+gro_flow_is_timeout (vlib_main_t * vm, gro_flow_t * gro_flow)
+{
+  if (gro_flow->next_timeout_ts < vlib_time_now (vm))
+    return 1;
+  return 0;
+}
+
+static_always_inline void
+gro_flow_store_packet (gro_flow_t * gro_flow, u32 bi0)
+{
+  if (gro_flow->n_buffers == 0)
+    {
+      gro_flow->buffer_index = bi0;
+    }
+  gro_flow->n_buffers++;
+}
+
+static_always_inline u32
+gro_flow_table_init (gro_flow_table_t ** flow_table, u8 is_l2, u32 node_index)
+{
+  if (*flow_table)
+    return 0;
+
+  gro_flow_table_t *flow_table_temp = 0;
+  flow_table_temp =
+    (gro_flow_table_t *) clib_mem_alloc (sizeof (gro_flow_table_t));
+  if (!flow_table_temp)
+    return 0;
+  clib_memset (flow_table_temp, 0, sizeof (gro_flow_table_t));
+  flow_table_temp->node_index = node_index;
+  flow_table_temp->is_enable = 1;
+  flow_table_temp->is_l2 = is_l2;
+  *flow_table = flow_table_temp;
+  return 1;
+}
+
+static_always_inline void
+gro_flow_table_set_timeout (vlib_main_t * vm, gro_flow_table_t * flow_table,
+			    f64 timeout_expire)
+{
+  if (flow_table)
+    flow_table->timeout_ts = vlib_time_now (vm) + timeout_expire;
+}
+
+static_always_inline u8
+gro_flow_table_is_timeout (vlib_main_t * vm, gro_flow_table_t * flow_table)
+{
+  if (flow_table && (flow_table->timeout_ts < vlib_time_now (vm)))
+    return 1;
+  return 0;
+}
+
+static_always_inline u8
+gro_flow_table_is_enable (gro_flow_table_t * flow_table)
+{
+  if (flow_table)
+    return flow_table->is_enable;
+
+  return 0;
+}
+
+static_always_inline void
+gro_flow_table_set_is_enable (gro_flow_table_t * flow_table, u8 is_enable)
+{
+  if (flow_table)
+    {
+      if (is_enable)
+	{
+	  flow_table->is_enable = 1;
+	}
+      else
+	{
+	  flow_table->is_enable = 0;
+	}
+    }
+}
+
+static_always_inline void
+gro_flow_table_free (gro_flow_table_t * flow_table)
+{
+  if (flow_table)
+    clib_mem_free (flow_table);
+}
+
+static_always_inline void
+gro_flow_table_set_node_index (gro_flow_table_t * flow_table, u32 node_index)
+{
+  if (flow_table)
+    flow_table->node_index = node_index;
+}
+
+static_always_inline gro_flow_t *
+gro_flow_table_new_flow (gro_flow_table_t * flow_table)
+{
+  if (PREDICT_TRUE (flow_table->flow_table_size < GRO_FLOW_TABLE_MAX_SIZE))
+    {
+      gro_flow_t *gro_flow;
+      u32 i = 0;
+      while (i < GRO_FLOW_TABLE_MAX_SIZE)
+	{
+	  gro_flow = &flow_table->gro_flow[i];
+	  if (gro_flow->n_buffers == 0)
+	    {
+	      flow_table->flow_table_size++;
+	      return gro_flow;
+	    }
+	  i++;
+	}
+    }
+
+  return (0);
+}
+
+static_always_inline gro_flow_t *
+gro_flow_table_get_flow (gro_flow_table_t * flow_table,
+			 gro_flow_key_t * flow_key)
+{
+  gro_flow_t *gro_flow = 0;
+  u32 i = 0;
+  while (i < GRO_FLOW_TABLE_MAX_SIZE)
+    {
+      gro_flow = &flow_table->gro_flow[i];
+      if (gro_flow_is_equal (flow_key, &gro_flow->flow_key))
+	return gro_flow;
+      i++;
+    }
+  return (0);
+}
+
+static_always_inline gro_flow_t *
+gro_flow_table_find_or_add_flow (gro_flow_table_t * flow_table,
+				 gro_flow_key_t * flow_key)
+{
+  gro_flow_t *gro_flow = 0;
+
+  gro_flow = gro_flow_table_get_flow (flow_table, flow_key);
+  if (gro_flow)
+    return gro_flow;
+
+  gro_flow = gro_flow_table_new_flow (flow_table);
+
+  if (gro_flow)
+    {
+      gro_flow_set_flow_key (gro_flow, flow_key);
+      return gro_flow;
+    }
+
+  return (0);
+}
+
+static_always_inline void
+gro_flow_table_reset_flow (gro_flow_table_t * flow_table,
+			   gro_flow_t * gro_flow)
+{
+  if (PREDICT_TRUE (flow_table->flow_table_size > 0))
+    {
+      clib_memset (gro_flow, 0, sizeof (gro_flow_t));
+      flow_table->flow_table_size--;
+    }
+}
+
+static_always_inline u8 *
+gro_flow_table_format (u8 * s, va_list * args)
+{
+  gro_flow_table_t *flow_table = va_arg (*args, gro_flow_table_t *);
+
+  s =
+    format (s,
+	    "flow-table: size %u gro-total-vectors %lu gro-n-vectors %u",
+	    flow_table->flow_table_size, flow_table->total_vectors,
+	    flow_table->n_vectors);
+  if (flow_table->n_vectors)
+    {
+      double average_rate =
+	(double) flow_table->total_vectors / (double) flow_table->n_vectors;
+      s = format (s, " gro-average-rate %.2f", average_rate);
+    }
+  else
+    s = format (s, " gro-average-rate 0.00");
+
+  return s;
+}
+#endif /* included_gro_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vnet/gso/gro_func.h b/src/vnet/gso/gro_func.h
new file mode 100644
index 0000000..a410a65
--- /dev/null
+++ b/src/vnet/gso/gro_func.h
@@ -0,0 +1,593 @@
+/*
+ * Copyright (c) 2020 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef included_gro_func_h
+#define included_gro_func_h
+
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/gso/gro.h>
+#include <vnet/gso/hdr_offset_parser.h>
+#include <vnet/udp/udp_packet.h>
+#include <vnet/tcp/tcp.h>
+#include <vnet/vnet.h>
+
+static_always_inline u8
+gro_is_bad_packet (vlib_buffer_t * b, u8 flags, i16 l234_sz)
+{
+  if (((b->current_length - l234_sz) <= 0) || ((flags &= ~TCP_FLAG_ACK) != 0))
+    return 1;
+  return 0;
+}
+
+static_always_inline void
+gro_get_ip4_flow_from_packet (u32 * sw_if_index,
+			      ip4_header_t * ip4, tcp_header_t * tcp,
+			      gro_flow_key_t * flow_key, int is_l2)
+{
+  flow_key->sw_if_index[VLIB_RX] = sw_if_index[VLIB_RX];
+  flow_key->sw_if_index[VLIB_TX] = sw_if_index[VLIB_TX];
+  ip46_address_set_ip4 (&flow_key->src_address, &ip4->src_address);
+  ip46_address_set_ip4 (&flow_key->dst_address, &ip4->dst_address);
+  flow_key->src_port = tcp->src_port;
+  flow_key->dst_port = tcp->dst_port;
+}
+
+static_always_inline void
+gro_get_ip6_flow_from_packet (u32 * sw_if_index,
+			      ip6_header_t * ip6, tcp_header_t * tcp,
+			      gro_flow_key_t * flow_key, int is_l2)
+{
+  flow_key->sw_if_index[VLIB_RX] = sw_if_index[VLIB_RX];
+  flow_key->sw_if_index[VLIB_TX] = sw_if_index[VLIB_TX];
+  ip46_address_set_ip6 (&flow_key->src_address, &ip6->src_address);
+  ip46_address_set_ip6 (&flow_key->dst_address, &ip6->dst_address);
+  flow_key->src_port = tcp->src_port;
+  flow_key->dst_port = tcp->dst_port;
+}
+
+static_always_inline u32
+gro_is_ip4_or_ip6_packet (vlib_buffer_t * b0, int is_l2)
+{
+  if (b0->flags & VNET_BUFFER_F_IS_IP4)
+    return VNET_BUFFER_F_IS_IP4;
+  if (b0->flags & VNET_BUFFER_F_IS_IP6)
+    return VNET_BUFFER_F_IS_IP6;
+  if (is_l2)
+    {
+      ethernet_header_t *eh =
+	(ethernet_header_t *) vlib_buffer_get_current (b0);
+      u16 ethertype = clib_net_to_host_u16 (eh->type);
+
+      if (ethernet_frame_is_tagged (ethertype))
+	{
+	  ethernet_vlan_header_t *vlan = (ethernet_vlan_header_t *) (eh + 1);
+
+	  ethertype = clib_net_to_host_u16 (vlan->type);
+	  if (ethertype == ETHERNET_TYPE_VLAN)
+	    {
+	      vlan++;
+	      ethertype = clib_net_to_host_u16 (vlan->type);
+	    }
+	}
+      if (ethertype == ETHERNET_TYPE_IP4)
+	return VNET_BUFFER_F_IS_IP4;
+      if (ethertype == ETHERNET_TYPE_IP6)
+	return VNET_BUFFER_F_IS_IP6;
+    }
+  else
+    {
+      if ((((u8 *) vlib_buffer_get_current (b0))[0] & 0xf0) == 0x40)
+	return VNET_BUFFER_F_IS_IP4;
+      if ((((u8 *) vlib_buffer_get_current (b0))[0] & 0xf0) == 0x60)
+	return VNET_BUFFER_F_IS_IP6;
+    }
+
+  return 0;
+}
+
+typedef enum
+{
+  GRO_PACKET_ACTION_NONE = 0,
+  GRO_PACKET_ACTION_ENQUEUE = 1,
+  GRO_PACKET_ACTION_FLUSH = 2,
+} gro_packet_action_t;
+
+static_always_inline gro_packet_action_t
+gro_tcp_sequence_check (tcp_header_t * tcp0, tcp_header_t * tcp1,
+			u32 payload_len0)
+{
+  u32 next_tcp_seq0 = clib_net_to_host_u32 (tcp0->seq_number);
+  u32 next_tcp_seq1 = clib_net_to_host_u32 (tcp1->seq_number);
+
+  /* next packet, enqueue */
+  if (PREDICT_TRUE (next_tcp_seq0 + payload_len0 == next_tcp_seq1))
+    return GRO_PACKET_ACTION_ENQUEUE;
+  /* flush all packets */
+  else
+    return GRO_PACKET_ACTION_FLUSH;
+}
+
+static_always_inline void
+gro_merge_buffers (vlib_main_t * vm, vlib_buffer_t * b0,
+		   vlib_buffer_t * b1, u32 bi1, u32 payload_len1,
+		   u16 l234_sz1)
+{
+  vlib_buffer_t *pb = b0;
+
+  if (PREDICT_FALSE ((b0->flags & VLIB_BUFFER_NEXT_PRESENT) == 0))
+    b0->total_length_not_including_first_buffer = 0;
+
+  while (pb->flags & VLIB_BUFFER_NEXT_PRESENT)
+    pb = vlib_get_buffer (vm, pb->next_buffer);
+
+  vlib_buffer_advance (b1, l234_sz1);
+  pb->flags |= VLIB_BUFFER_NEXT_PRESENT;
+  pb->next_buffer = bi1;
+  b0->total_length_not_including_first_buffer += payload_len1;
+  b0->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
+}
+
+static_always_inline u32
+gro_get_packet_data (vlib_main_t * vm, vlib_buffer_t * b0,
+		     generic_header_offset_t * gho0,
+		     gro_flow_key_t * flow_key0, int is_l2)
+{
+  ip4_header_t *ip4_0 = 0;
+  ip6_header_t *ip6_0 = 0;
+  tcp_header_t *tcp0 = 0;
+  u32 pkt_len0 = 0;
+  u16 l234_sz0 = 0;
+  u32 sw_if_index0[VLIB_N_RX_TX] = { ~0 };
+
+  u32 is_ip0 = gro_is_ip4_or_ip6_packet (b0, is_l2);
+
+  if (is_ip0 & VNET_BUFFER_F_IS_IP4)
+    vnet_generic_header_offset_parser (b0, gho0, is_l2, 1 /* is_ip4 */ ,
+				       0 /* is_ip6 */ );
+  else if (is_ip0 & VNET_BUFFER_F_IS_IP6)
+    vnet_generic_header_offset_parser (b0, gho0, is_l2, 0 /* is_ip4 */ ,
+				       1 /* is_ip6 */ );
+  else
+    return 0;
+
+  if (PREDICT_FALSE ((gho0->gho_flags & GHO_F_TCP) == 0))
+    return 0;
+
+  ip4_0 =
+    (ip4_header_t *) (vlib_buffer_get_current (b0) + gho0->l3_hdr_offset);
+  ip6_0 =
+    (ip6_header_t *) (vlib_buffer_get_current (b0) + gho0->l3_hdr_offset);
+  tcp0 =
+    (tcp_header_t *) (vlib_buffer_get_current (b0) + gho0->l4_hdr_offset);
+
+  l234_sz0 = gho0->hdr_sz;
+  if (PREDICT_FALSE (gro_is_bad_packet (b0, tcp0->flags, l234_sz0)))
+    return 0;
+
+  sw_if_index0[VLIB_RX] = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+  sw_if_index0[VLIB_TX] = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+
+  if (gho0->gho_flags & GHO_F_IP4)
+    {
+      gro_get_ip4_flow_from_packet (sw_if_index0, ip4_0, tcp0, flow_key0,
+				    is_l2);
+    }
+  else if (gho0->gho_flags & GHO_F_IP6)
+    {
+      gro_get_ip6_flow_from_packet (sw_if_index0, ip6_0, tcp0, flow_key0,
+				    is_l2);
+    }
+  else
+    return 0;
+
+  pkt_len0 = vlib_buffer_length_in_chain (vm, b0);
+  if (PREDICT_FALSE (pkt_len0 >= TCP_MAX_GSO_SZ))
+    return 0;
+
+  return pkt_len0;
+}
+
+static_always_inline u32
+gro_coalesce_buffers (vlib_main_t * vm, vlib_buffer_t * b0,
+		      vlib_buffer_t * b1, u32 bi1, int is_l2)
+{
+  generic_header_offset_t gho0 = { 0 };
+  generic_header_offset_t gho1 = { 0 };
+  gro_flow_key_t flow_key0, flow_key1;
+  ip4_header_t *ip4_0, *ip4_1;
+  ip6_header_t *ip6_0, *ip6_1;
+  tcp_header_t *tcp0, *tcp1;
+  u16 l234_sz0, l234_sz1;
+  u32 pkt_len0, pkt_len1, payload_len0, payload_len1;
+  u32 sw_if_index0[VLIB_N_RX_TX] = { ~0 };
+  u32 sw_if_index1[VLIB_N_RX_TX] = { ~0 };
+
+  u32 is_ip0 = gro_is_ip4_or_ip6_packet (b0, is_l2);
+  u32 is_ip1 = gro_is_ip4_or_ip6_packet (b1, is_l2);
+
+  if (is_ip0 & VNET_BUFFER_F_IS_IP4)
+    vnet_generic_header_offset_parser (b0, &gho0, is_l2, 1 /* is_ip4 */ ,
+				       0 /* is_ip6 */ );
+  else if (is_ip0 & VNET_BUFFER_F_IS_IP6)
+    vnet_generic_header_offset_parser (b0, &gho0, is_l2, 0 /* is_ip4 */ ,
+				       1 /* is_ip6 */ );
+  else
+    return 0;
+
+  if (is_ip1 & VNET_BUFFER_F_IS_IP4)
+    vnet_generic_header_offset_parser (b1, &gho1, is_l2, 1 /* is_ip4 */ ,
+				       0 /* is_ip6 */ );
+  else if (is_ip1 & VNET_BUFFER_F_IS_IP6)
+    vnet_generic_header_offset_parser (b1, &gho1, is_l2, 0 /* is_ip4 */ ,
+				       1 /* is_ip6 */ );
+  else
+    return 0;
+
+  pkt_len0 = vlib_buffer_length_in_chain (vm, b0);
+  pkt_len1 = vlib_buffer_length_in_chain (vm, b1);
+
+  if (((gho0.gho_flags & GHO_F_TCP) == 0)
+      || ((gho1.gho_flags & GHO_F_TCP) == 0))
+    return 0;
+
+  ip4_0 =
+    (ip4_header_t *) (vlib_buffer_get_current (b0) + gho0.l3_hdr_offset);
+  ip4_1 =
+    (ip4_header_t *) (vlib_buffer_get_current (b1) + gho1.l3_hdr_offset);
+  ip6_0 =
+    (ip6_header_t *) (vlib_buffer_get_current (b0) + gho0.l3_hdr_offset);
+  ip6_1 =
+    (ip6_header_t *) (vlib_buffer_get_current (b1) + gho1.l3_hdr_offset);
+
+  tcp0 = (tcp_header_t *) (vlib_buffer_get_current (b0) + gho0.l4_hdr_offset);
+  tcp1 = (tcp_header_t *) (vlib_buffer_get_current (b1) + gho1.l4_hdr_offset);
+
+  l234_sz0 = gho0.hdr_sz;
+  l234_sz1 = gho1.hdr_sz;
+
+  if (gro_is_bad_packet (b0, tcp0->flags, l234_sz0)
+      || gro_is_bad_packet (b1, tcp1->flags, l234_sz1))
+    return 0;
+
+  sw_if_index0[VLIB_RX] = vnet_buffer (b0)->sw_if_index[VLIB_RX];
+  sw_if_index0[VLIB_TX] = vnet_buffer (b0)->sw_if_index[VLIB_TX];
+
+  sw_if_index1[VLIB_RX] = vnet_buffer (b1)->sw_if_index[VLIB_RX];
+  sw_if_index1[VLIB_TX] = vnet_buffer (b1)->sw_if_index[VLIB_TX];
+
+  if ((gho0.gho_flags & GHO_F_IP4) && (gho1.gho_flags & GHO_F_IP4))
+    {
+      gro_get_ip4_flow_from_packet (sw_if_index0, ip4_0, tcp0, &flow_key0,
+				    is_l2);
+      gro_get_ip4_flow_from_packet (sw_if_index1, ip4_1, tcp1, &flow_key1,
+				    is_l2);
+    }
+  else if ((gho0.gho_flags & GHO_F_IP6) && (gho1.gho_flags & GHO_F_IP6))
+    {
+      gro_get_ip6_flow_from_packet (sw_if_index0, ip6_0, tcp0, &flow_key0,
+				    is_l2);
+      gro_get_ip6_flow_from_packet (sw_if_index1, ip6_1, tcp1, &flow_key1,
+				    is_l2);
+    }
+  else
+    return 0;
+
+  if (gro_flow_is_equal (&flow_key0, &flow_key1) == 0)
+    return 0;
+
+  payload_len0 = pkt_len0 - l234_sz0;
+  payload_len1 = pkt_len1 - l234_sz1;
+
+  if (pkt_len0 >= TCP_MAX_GSO_SZ || pkt_len1 >= TCP_MAX_GSO_SZ
+      || (pkt_len0 + payload_len1) >= TCP_MAX_GSO_SZ)
+    return 0;
+
+  if (gro_tcp_sequence_check (tcp0, tcp1, payload_len0) ==
+      GRO_PACKET_ACTION_ENQUEUE)
+    {
+      gro_merge_buffers (vm, b0, b1, bi1, payload_len1, l234_sz1);
+      return tcp1->ack_number;
+    }
+
+  return 0;
+}
+
+static_always_inline void
+gro_fixup_header (vlib_main_t * vm, vlib_buffer_t * b0, u32 ack_number,
+		  int is_l2)
+{
+  generic_header_offset_t gho0 = { 0 };
+
+  u32 is_ip0 = gro_is_ip4_or_ip6_packet (b0, is_l2);
+
+  if (is_ip0 & VNET_BUFFER_F_IS_IP4)
+    vnet_generic_header_offset_parser (b0, &gho0, is_l2, 1 /* is_ip4 */ ,
+				       0 /* is_ip6 */ );
+  else if (is_ip0 & VNET_BUFFER_F_IS_IP6)
+    vnet_generic_header_offset_parser (b0, &gho0, is_l2, 0 /* is_ip4 */ ,
+				       1 /* is_ip6 */ );
+
+  vnet_buffer2 (b0)->gso_size = b0->current_length - gho0.hdr_sz;
+
+  if (gho0.gho_flags & GHO_F_IP4)
+    {
+      ip4_header_t *ip4 =
+	(ip4_header_t *) (vlib_buffer_get_current (b0) + gho0.l3_hdr_offset);
+      ip4->length =
+	clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
+			      gho0.l3_hdr_offset);
+      b0->flags |=
+	(VNET_BUFFER_F_GSO | VNET_BUFFER_F_IS_IP4 |
+	 VNET_BUFFER_F_OFFLOAD_TCP_CKSUM | VNET_BUFFER_F_OFFLOAD_IP_CKSUM);
+    }
+  else if (gho0.gho_flags & GHO_F_IP6)
+    {
+      ip6_header_t *ip6 =
+	(ip6_header_t *) (vlib_buffer_get_current (b0) + gho0.l3_hdr_offset);
+      ip6->payload_length =
+	clib_host_to_net_u16 (vlib_buffer_length_in_chain (vm, b0) -
+			      gho0.l4_hdr_offset);
+      b0->flags |=
+	(VNET_BUFFER_F_GSO | VNET_BUFFER_F_IS_IP6 |
+	 VNET_BUFFER_F_OFFLOAD_TCP_CKSUM);
+    }
+
+  tcp_header_t *tcp0 =
+    (tcp_header_t *) (vlib_buffer_get_current (b0) + gho0.l4_hdr_offset);
+  tcp0->ack_number = ack_number;
+  b0->flags &= ~VLIB_BUFFER_IS_TRACED;
+}
+
+static_always_inline u32
+vnet_gro_flow_table_flush (vlib_main_t * vm, gro_flow_table_t * flow_table,
+			   u32 * to)
+{
+  if (flow_table->flow_table_size > 0)
+    {
+      gro_flow_t *gro_flow;
+      u32 i = 0, j = 0;
+      while (i < GRO_FLOW_TABLE_MAX_SIZE)
+	{
+	  gro_flow = &flow_table->gro_flow[i];
+	  if (gro_flow->n_buffers && gro_flow_is_timeout (vm, gro_flow))
+	    {
+	      // flush the packet
+	      vlib_buffer_t *b0 =
+		vlib_get_buffer (vm, gro_flow->buffer_index);
+	      gro_fixup_header (vm, b0, gro_flow->last_ack_number,
+				flow_table->is_l2);
+	      to[j] = gro_flow->buffer_index;
+	      gro_flow_table_reset_flow (flow_table, gro_flow);
+	      flow_table->n_vectors++;
+	      j++;
+	    }
+	  i++;
+	}
+
+      return j;
+    }
+  return 0;
+}
+
+static_always_inline void
+vnet_gro_flow_table_schedule_node_on_dispatcher (vlib_main_t * vm,
+						 gro_flow_table_t *
+						 flow_table)
+{
+  if (gro_flow_table_is_timeout (vm, flow_table))
+    {
+      u32 to[GRO_FLOW_TABLE_MAX_SIZE] = { 0 };
+      u32 n_to = vnet_gro_flow_table_flush (vm, flow_table, to);
+
+      if (n_to > 0)
+	{
+	  u32 node_index = flow_table->node_index;
+	  vlib_frame_t *f = vlib_get_frame_to_node (vm, node_index);
+	  u32 *f_to = vlib_frame_vector_args (f);
+	  u32 i = 0;
+
+	  while (i < n_to)
+	    {
+	      f_to[f->n_vectors] = to[i];
+	      i++;
+	      f->n_vectors++;
+	    }
+	  vlib_put_frame_to_node (vm, node_index, f);
+	}
+      gro_flow_table_set_timeout (vm, flow_table, GRO_FLOW_TABLE_FLUSH);
+    }
+}
+
+static_always_inline u32
+vnet_gro_flow_table_inline (vlib_main_t * vm, gro_flow_table_t * flow_table,
+			    u32 bi0, u32 * to)
+{
+  vlib_buffer_t *b0 = vlib_get_buffer (vm, bi0);
+  generic_header_offset_t gho0 = { 0 };
+  gro_flow_t *gro_flow = 0;
+  gro_flow_key_t flow_key0 = { };
+  tcp_header_t *tcp0 = 0;
+  u32 pkt_len0 = 0;
+  int is_l2 = flow_table->is_l2;
+
+  if (!gro_flow_table_is_enable (flow_table))
+    {
+      to[0] = bi0;
+      return 1;
+    }
+
+  if (PREDICT_FALSE (b0->flags & VNET_BUFFER_F_GSO))
+    {
+      to[0] = bi0;
+      return 1;
+    }
+
+  pkt_len0 = gro_get_packet_data (vm, b0, &gho0, &flow_key0, is_l2);
+  if (pkt_len0 == 0)
+    {
+      to[0] = bi0;
+      return 1;
+    }
+
+  gro_flow = gro_flow_table_find_or_add_flow (flow_table, &flow_key0);
+  if (!gro_flow)
+    {
+      to[0] = bi0;
+      return 1;
+    }
+
+  if (PREDICT_FALSE (gro_flow->n_buffers == 0))
+    {
+      flow_table->total_vectors++;
+      gro_flow_store_packet (gro_flow, bi0);
+      tcp0 =
+	(tcp_header_t *) (vlib_buffer_get_current (b0) + gho0.l4_hdr_offset);
+      gro_flow->last_ack_number = tcp0->ack_number;
+      gro_flow_set_timeout (vm, gro_flow, GRO_FLOW_TIMEOUT);
+      return 0;
+    }
+  else
+    {
+      tcp0 =
+	(tcp_header_t *) (vlib_buffer_get_current (b0) + gho0.l4_hdr_offset);
+      generic_header_offset_t gho_s = { 0 };
+      tcp_header_t *tcp_s;
+      u16 l234_sz0, l234_sz_s;
+      u32 pkt_len_s, payload_len0, payload_len_s;
+      u32 bi_s = gro_flow->buffer_index;
+
+      vlib_buffer_t *b_s = vlib_get_buffer (vm, bi_s);
+      u32 is_ip_s = gro_is_ip4_or_ip6_packet (b_s, is_l2);
+      if (is_ip_s & VNET_BUFFER_F_IS_IP4)
+	vnet_generic_header_offset_parser (b_s, &gho_s, is_l2,
+					   1 /* is_ip4 */ , 0 /* is_ip6 */ );
+      else if (is_ip_s & VNET_BUFFER_F_IS_IP6)
+	vnet_generic_header_offset_parser (b_s, &gho_s, is_l2,
+					   0 /* is_ip4 */ , 1 /* is_ip6 */ );
+
+      tcp_s =
+	(tcp_header_t *) (vlib_buffer_get_current (b_s) +
+			  gho_s.l4_hdr_offset);
+      pkt_len_s = vlib_buffer_length_in_chain (vm, b_s);
+      l234_sz0 = gho0.hdr_sz;
+      l234_sz_s = gho_s.hdr_sz;
+      payload_len0 = pkt_len0 - l234_sz0;
+      payload_len_s = pkt_len_s - l234_sz_s;
+      gro_packet_action_t action =
+	gro_tcp_sequence_check (tcp_s, tcp0, payload_len_s);
+
+      if (PREDICT_TRUE (action == GRO_PACKET_ACTION_ENQUEUE))
+	{
+	  if (PREDICT_TRUE ((pkt_len_s + payload_len0) < TCP_MAX_GSO_SZ))
+	    {
+	      flow_table->total_vectors++;
+	      gro_merge_buffers (vm, b_s, b0, bi0, payload_len0, l234_sz0);
+	      gro_flow_store_packet (gro_flow, bi0);
+	      gro_flow->last_ack_number = tcp0->ack_number;
+	      return 0;
+	    }
+	  else
+	    {
+	      // flush the stored GSO size packet and buffer the current packet
+	      flow_table->n_vectors++;
+	      flow_table->total_vectors++;
+	      gro_fixup_header (vm, b_s, gro_flow->last_ack_number, is_l2);
+	      gro_flow->n_buffers = 0;
+	      gro_flow_store_packet (gro_flow, bi0);
+	      gro_flow->last_ack_number = tcp0->ack_number;
+	      gro_flow_set_timeout (vm, gro_flow, GRO_FLOW_TIMEOUT);
+	      to[0] = bi_s;
+	      return 1;
+	    }
+	}
+      else
+	{
+	  // flush the all (current and stored) packets
+	  flow_table->n_vectors++;
+	  flow_table->total_vectors++;
+	  gro_fixup_header (vm, b_s, gro_flow->last_ack_number, is_l2);
+	  gro_flow->n_buffers = 0;
+	  gro_flow_table_reset_flow (flow_table, gro_flow);
+	  to[0] = bi_s;
+	  to[1] = bi0;
+	  return 2;
+	}
+    }
+}
+
+/**
+ * coalesce buffers with flow tables
+ */
+static_always_inline u32
+vnet_gro_inline (vlib_main_t * vm, gro_flow_table_t * flow_table, u32 * from,
+		 u16 n_left_from, u32 * to)
+{
+  u16 count = 0, i = 0;
+
+  for (i = 0; i < n_left_from; i++)
+    count += vnet_gro_flow_table_inline (vm, flow_table, from[i], &to[count]);
+
+  return count;
+}
+
+/**
+ * coalesce buffers in opportunistic way without flow tables
+ */
+static_always_inline u32
+vnet_gro_simple_inline (vlib_main_t * vm, u32 * from, u16 n_left_from,
+			int is_l2)
+{
+  vlib_buffer_t *bufs[VLIB_FRAME_SIZE], **b = bufs;
+  vlib_get_buffers (vm, from, b, n_left_from);
+  u32 bi = 1, ack_number = 0;
+  if (PREDICT_TRUE (((b[0]->flags & VNET_BUFFER_F_GSO) == 0)))
+    {
+      while (n_left_from > 1)
+	{
+	  if (PREDICT_TRUE (((b[bi]->flags & VNET_BUFFER_F_GSO) == 0)))
+	    {
+	      u32 ret;
+	      if ((ret =
+		   gro_coalesce_buffers (vm, b[0], b[bi], from[bi],
+					 is_l2)) != 0)
+		{
+		  n_left_from -= 1;
+		  bi += 1;
+		  ack_number = ret;
+		  continue;
+		}
+	      else
+		break;
+	    }
+	  else
+	    break;
+	}
+
+      if (bi >= 2)
+	{
+	  gro_fixup_header (vm, b[0], ack_number, is_l2);
+	}
+    }
+  return bi;
+}
+#endif /* included_gro_func_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */