summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorNathan Skrzypczak <nathan.skrzypczak@gmail.com>2019-11-22 11:26:19 +0100
committerDave Wallace <dwallacelf@gmail.com>2019-11-22 21:08:19 +0000
commitb2fce36c8ebd19bd9a60b06988ce753d93e59d81 (patch)
treede175c7b7201eed8db53986d7224700e2616d1e5 /src
parent952a7b8b74be7fd237e6c0ccfe423876aa7e7b3e (diff)
hsa: Improve for mq-intensive
Type: feature * Remove the retry mechanism in case of failed connect * Limit the number of simultaneous connects (default to mqsize / 2) Change-Id: I7a0ed2e349ee3e8ca71639c2a2ec4cdf84f5c13e Signed-off-by: Nathan Skrzypczak <nathan.skrzypczak@gmail.com>
Diffstat (limited to 'src')
-rw-r--r--src/plugins/hs_apps/sapi/vpp_echo.c38
-rw-r--r--src/plugins/hs_apps/sapi/vpp_echo_bapi.c4
-rw-r--r--src/plugins/hs_apps/sapi/vpp_echo_common.c4
-rw-r--r--src/plugins/hs_apps/sapi/vpp_echo_common.h5
-rw-r--r--src/plugins/hs_apps/sapi/vpp_echo_proto_quic.c26
5 files changed, 31 insertions, 46 deletions
diff --git a/src/plugins/hs_apps/sapi/vpp_echo.c b/src/plugins/hs_apps/sapi/vpp_echo.c
index 143551f5848..cfe1a3d37be 100644
--- a/src/plugins/hs_apps/sapi/vpp_echo.c
+++ b/src/plugins/hs_apps/sapi/vpp_echo.c
@@ -587,6 +587,8 @@ session_connected_handler (session_connected_msg_t * mp)
u32 listener_index = htonl (mp->context);
svm_fifo_t *rx_fifo, *tx_fifo;
+ clib_atomic_add_fetch (&em->max_sim_connects, 1);
+
if (mp->retval)
{
if (em->proto_cb_vft->connected_cb)
@@ -738,23 +740,6 @@ echo_process_rpcs (echo_main_t * em)
}
}
-static inline int
-echo_mq_dequeue_batch (svm_msg_q_t * mq, svm_msg_q_msg_t * msg_vec,
- u32 n_max_msg)
-{
- svm_msg_q_msg_t *msg;
- u32 n_msgs;
- int i;
-
- n_msgs = clib_min (svm_msg_q_size (mq), n_max_msg);
- for (i = 0; i < n_msgs; i++)
- {
- vec_add2 (msg_vec, msg, 1);
- svm_msg_q_sub_w_lock (mq, msg);
- }
- return n_msgs;
-}
-
static void *
echo_mq_thread_fn (void *arg)
{
@@ -784,7 +769,11 @@ echo_mq_thread_fn (void *arg)
svm_msg_q_unlock (mq);
continue;
}
- echo_mq_dequeue_batch (mq, msg_vec, ~0);
+ for (i = 0; i < svm_msg_q_size (mq); i++)
+ {
+ vec_add2 (msg_vec, msg, 1);
+ svm_msg_q_sub_w_lock (mq, msg);
+ }
svm_msg_q_unlock (mq);
for (i = 0; i < vec_len (msg_vec); i++)
@@ -839,13 +828,14 @@ print_usage_and_exit (void)
int i;
fprintf (stderr,
"Usage: vpp_echo [socket-name SOCKET] [client|server] [uri URI] [OPTIONS]\n"
- "Generates traffic and assert correct teardown of the QUIC hoststack\n"
+ "Generates traffic and assert correct teardown of the hoststack\n"
"\n"
" socket-name PATH Specify the binary socket path to connect to VPP\n"
" use-svm-api Use SVM API to connect to VPP\n"
" test-bytes[:assert] Check data correctness when receiving (assert fails on first error)\n"
" fifo-size N[K|M|G] Use N[K|M|G] fifos\n"
- " mq-size N Use N event slots for vpp_echo <-> vpp events\n"
+ " mq-size N Use mq with N slots for [vpp_echo->vpp] communication\n"
+ " max-sim-connects N Do not allow more than N mq events inflight\n"
" rx-buf N[K|M|G] Use N[Kb|Mb|GB] RX buffer\n"
" tx-buf N[K|M|G] Use N[Kb|Mb|GB] TX test buffer\n"
" appns NAMESPACE Use the namespace NAMESPACE\n"
@@ -1008,6 +998,8 @@ echo_process_opts (int argc, char **argv)
echo_unformat_timing_event, &em->timing.start_event,
echo_unformat_timing_event, &em->timing.end_event))
;
+ else if (unformat (a, "max-sim-connects %d", &em->max_sim_connects))
+ ;
else
print_usage_and_exit ();
}
@@ -1040,6 +1032,9 @@ echo_process_opts (int argc, char **argv)
if (em->send_stream_disconnects == ECHO_CLOSE_F_INVALID)
em->send_stream_disconnects = default_f_active;
+ if (em->max_sim_connects == 0)
+ em->max_sim_connects = em->evt_q_size >> 1;
+
if (em->wait_for_gdb)
{
volatile u64 nop = 0;
@@ -1091,7 +1086,7 @@ main (int argc, char **argv)
char *app_name;
u64 i;
svm_msg_q_cfg_t _cfg, *cfg = &_cfg;
- u32 rpc_queue_size = 64 << 10;
+ u32 rpc_queue_size = 256 << 10;
em->session_index_by_vpp_handles = hash_create (0, sizeof (uword));
clib_spinlock_init (&em->sid_vpp_handles_lock);
@@ -1118,6 +1113,7 @@ main (int argc, char **argv)
em->tx_buf_size = 1 << 20;
em->data_source = ECHO_INVALID_DATA_SOURCE;
em->uri = format (0, "%s%c", "tcp://0.0.0.0/1234", 0);
+ em->max_sim_connects = 0;
em->crypto_engine = CRYPTO_ENGINE_NONE;
echo_set_each_proto_defaults_before_opts (em);
echo_process_opts (argc, argv);
diff --git a/src/plugins/hs_apps/sapi/vpp_echo_bapi.c b/src/plugins/hs_apps/sapi/vpp_echo_bapi.c
index 3cfe6a78ec5..2bba9fc0fe2 100644
--- a/src/plugins/hs_apps/sapi/vpp_echo_bapi.c
+++ b/src/plugins/hs_apps/sapi/vpp_echo_bapi.c
@@ -149,6 +149,10 @@ echo_send_connect (u64 parent_session_handle, u32 opaque)
session_connect_msg_t *mp;
svm_msg_q_t *mq = em->ctrl_mq;
+ clib_atomic_sub_fetch (&em->max_sim_connects, 1);
+ while (em->max_sim_connects <= 0)
+ ;
+
app_alloc_ctrl_evt_to_vpp (mq, app_evt, SESSION_CTRL_EVT_CONNECT);
mp = (session_connect_msg_t *) app_evt->evt->data;
memset (mp, 0, sizeof (*mp));
diff --git a/src/plugins/hs_apps/sapi/vpp_echo_common.c b/src/plugins/hs_apps/sapi/vpp_echo_common.c
index 7c58dad99ea..b91203a7bb3 100644
--- a/src/plugins/hs_apps/sapi/vpp_echo_common.c
+++ b/src/plugins/hs_apps/sapi/vpp_echo_common.c
@@ -530,13 +530,13 @@ echo_send_rpc (echo_main_t * em, void *fp, void *arg, u32 opaque)
echo_rpc_msg_t *evt;
if (PREDICT_FALSE (svm_msg_q_lock (em->rpc_msq_queue)))
{
- ECHO_LOG (1, "RPC lock failed");
+ ECHO_FAIL (ECHO_FAIL_RPC_SIZE, "RPC lock failed");
return -1;
}
if (PREDICT_FALSE (svm_msg_q_ring_is_full (em->rpc_msq_queue, 0)))
{
svm_msg_q_unlock (em->rpc_msq_queue);
- ECHO_LOG (1, "RPC ring is full");
+ ECHO_FAIL (ECHO_FAIL_RPC_SIZE, "RPC ring is full");
return -2;
}
msg = svm_msg_q_alloc_msg_w_ring (em->rpc_msq_queue, 0);
diff --git a/src/plugins/hs_apps/sapi/vpp_echo_common.h b/src/plugins/hs_apps/sapi/vpp_echo_common.h
index 5ddd8fb12dd..7d134d9a2c2 100644
--- a/src/plugins/hs_apps/sapi/vpp_echo_common.h
+++ b/src/plugins/hs_apps/sapi/vpp_echo_common.h
@@ -95,7 +95,8 @@
_(ECHO_FAIL_TEST_ASSERT_RX_TOTAL, "ECHO_FAIL_TEST_ASSERT_RX_TOTAL") \
_(ECHO_FAIL_TEST_ASSERT_TX_TOTAL, "ECHO_FAIL_TEST_ASSERT_TX_TOTAL") \
_(ECHO_FAIL_TEST_ASSERT_ALL_SESSIONS_CLOSED, \
- "ECHO_FAIL_TEST_ASSERT_ALL_SESSIONS_CLOSED")
+ "ECHO_FAIL_TEST_ASSERT_ALL_SESSIONS_CLOSED") \
+ _(ECHO_FAIL_RPC_SIZE, "ECHO_FAIL_RPC_SIZE")
typedef enum
{
@@ -328,6 +329,8 @@ typedef struct
volatile u32 n_clients_connected; /* Number of STREAM sessions connected */
volatile u32 nxt_available_sidx; /* next unused prealloced session_index */
+ volatile int max_sim_connects;
+
/* VNET_API_ERROR_FOO -> "Foo" hash table */
uword *error_string_by_error_number;
echo_proto_cb_vft_t *available_proto_cb_vft[TRANSPORT_N_PROTO];
diff --git a/src/plugins/hs_apps/sapi/vpp_echo_proto_quic.c b/src/plugins/hs_apps/sapi/vpp_echo_proto_quic.c
index bd7903bb771..d20bacb2a5d 100644
--- a/src/plugins/hs_apps/sapi/vpp_echo_proto_quic.c
+++ b/src/plugins/hs_apps/sapi/vpp_echo_proto_quic.c
@@ -299,32 +299,14 @@ quic_echo_on_connected (session_connected_msg_t * mp, u32 session_index)
}
static void
-quic_echo_retry_connect (u32 session_index)
-{
- /* retry connect */
- echo_session_t *session;
- echo_main_t *em = &echo_main;
- if (session_index == SESSION_INVALID_INDEX)
- {
- ECHO_LOG (1, "Retrying Qsession connect");
- echo_send_rpc (em, echo_send_connect, (void *) SESSION_INVALID_HANDLE,
- SESSION_INVALID_INDEX);
- }
- else
- {
- session = pool_elt_at_index (em->sessions, session_index);
- ECHO_LOG (1, "Retrying connect %U", echo_format_session, session);
- echo_send_rpc (em, echo_send_connect,
- (void *) session->vpp_session_handle, session_index);
- }
-}
-
-static void
quic_echo_connected_cb (session_connected_bundled_msg_t * mp,
u32 session_index, u8 is_failed)
{
if (is_failed)
- return quic_echo_retry_connect (session_index);
+ {
+ ECHO_FAIL (ECHO_FAIL_QUIC_WRONG_CONNECT, "Echo connect failed");
+ return;
+ }
return quic_echo_on_connected ((session_connected_msg_t *) mp,
session_index);
}
8 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 
/*
 * Copyright (c) 2017 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.
 */
/**
 * @file
 * @brief IPv6 to IPv4 translation
 */
#ifndef __included_ip6_to_ip4_h__
#define __included_ip6_to_ip4_h__

#include <vnet/ip/ip.h>

/**
 * IPv6 to IPv4 set call back function type
 */
typedef int (*ip6_to_ip4_set_fn_t) (ip6_header_t * ip6, ip4_header_t * ip4,
				    void *ctx);

/* *INDENT-OFF* */
static u8 icmp6_to_icmp_updater_pointer_table[] =
  { 0, 1, ~0, ~0,
    2, 2, 9, 8,
    12, 12, 12, 12,
    12, 12, 12, 12,
    12, 12, 12, 12,
    12, 12, 12, 12,
    24, 24, 24, 24,
    24, 24, 24, 24,
    24, 24, 24, 24,
    24, 24, 24, 24
  };
/* *INDENT-ON* */

#define frag_id_6to4(id) ((id) ^ ((id) >> 16))

/**
 * @brief Parse some useful information from IPv6 header.
 *
 * @param ip6             IPv6 header.
 * @param buff_len        Buffer length.
 * @param l4_protocol     L4 protocol number.
 * @param l4_offset       L4 header offset.
 * @param frag_hdr_offset Fragment header offset if present, 0 otherwise.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
static_always_inline int
ip6_parse (const ip6_header_t * ip6, u32 buff_len,
	   u8 * l4_protocol, u16 * l4_offset, u16 * frag_hdr_offset)
{
  if (ip6->protocol == IP_PROTOCOL_IPV6_FRAGMENTATION)
    {
      *l4_protocol = ((ip6_frag_hdr_t *) (ip6 + 1))->next_hdr;
      *frag_hdr_offset = sizeof (*ip6);
      *l4_offset = sizeof (*ip6) + sizeof (ip6_frag_hdr_t);
    }
  else
    {
      *l4_protocol = ip6->protocol;
      *frag_hdr_offset = 0;
      *l4_offset = sizeof (*ip6);
    }

  return (buff_len < (*l4_offset + 4)) ||
    (clib_net_to_host_u16 (ip6->payload_length) <
     (*l4_offset + 4 - sizeof (*ip6)));
}

/**
 * @brief Get TCP/UDP port number or ICMP id from IPv6 packet.
 *
 * @param ip6        IPv6 header.
 * @param sender     1 get sender port, 0 get receiver port.
 * @param buffer_len Buffer length.
 *
 * @returns Port number on success, 0 otherwise.
 */
always_inline u16
ip6_get_port (ip6_header_t * ip6, u8 sender, u16 buffer_len)
{
  u8 l4_protocol;
  u16 l4_offset;
  u16 frag_offset;
  u8 *l4;

  if (ip6_parse (ip6, buffer_len, &l4_protocol, &l4_offset, &frag_offset))
    return 0;

  if (frag_offset &&
      ip6_frag_hdr_offset (((ip6_frag_hdr_t *)
			    u8_ptr_add (ip6, frag_offset))))
    return 0;			//Can't deal with non-first fragment for now

  l4 = u8_ptr_add (ip6, l4_offset);
  if (l4_protocol == IP_PROTOCOL_TCP || l4_protocol == IP_PROTOCOL_UDP)
    {
      return (sender) ? ((udp_header_t *) (l4))->src_port : ((udp_header_t
							      *)
							     (l4))->dst_port;
    }
  else if (l4_protocol == IP_PROTOCOL_ICMP6)
    {
      icmp46_header_t *icmp = (icmp46_header_t *) (l4);
      if (icmp->type == ICMP6_echo_request)
	{
	  return (sender) ? ((u16 *) (icmp))[2] : -1;
	}
      else if (icmp->type == ICMP6_echo_reply)
	{
	  return (sender) ? -1 : ((u16 *) (icmp))[2];
	}
    }
  return 0;
}

/**
 * @brief Convert type and code value from ICMP6 to ICMP4.
 *
 * @param icmp      ICMP header.
 * @param inner_ip6 Inner IPv6 header if present, 0 otherwise.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
static_always_inline int
icmp6_to_icmp_header (icmp46_header_t * icmp, ip6_header_t ** inner_ip6)
{
  *inner_ip6 = NULL;
  switch (icmp->type)
    {
    case ICMP6_echo_request:
      icmp->type = ICMP4_echo_request;
      break;
    case ICMP6_echo_reply:
      icmp->type = ICMP4_echo_reply;
      break;
    case ICMP6_destination_unreachable:
      *inner_ip6 = (ip6_header_t *) u8_ptr_add (icmp, 8);

      switch (icmp->code)
	{
	case ICMP6_destination_unreachable_no_route_to_destination:	//0
	case ICMP6_destination_unreachable_beyond_scope_of_source_address:	//2
	case ICMP6_destination_unreachable_address_unreachable:	//3
	  icmp->type = ICMP4_destination_unreachable;
	  icmp->code =
	    ICMP4_destination_unreachable_destination_unreachable_host;
	  break;
	case ICMP6_destination_unreachable_destination_administratively_prohibited:	//1
	  icmp->type =
	    ICMP4_destination_unreachable;
	  icmp->code =
	    ICMP4_destination_unreachable_communication_administratively_prohibited;
	  break;
	case ICMP6_destination_unreachable_port_unreachable:
	  icmp->type = ICMP4_destination_unreachable;
	  icmp->code = ICMP4_destination_unreachable_port_unreachable;
	  break;
	default:
	  return -1;
	}
      break;
    case ICMP6_packet_too_big:
      *inner_ip6 = (ip6_header_t *) u8_ptr_add (icmp, 8);

      icmp->type = ICMP4_destination_unreachable;
      icmp->code = 4;
      {
	u32 advertised_mtu = clib_net_to_host_u32 (*((u32 *) (icmp + 1)));
	advertised_mtu -= 20;
	//FIXME: = minimum(advertised MTU-20, MTU_of_IPv4_nexthop, (MTU_of_IPv6_nexthop)-20)
	((u16 *) (icmp))[3] = clib_host_to_net_u16 (advertised_mtu);
      }
      break;

    case ICMP6_time_exceeded:
      *inner_ip6 = (ip6_header_t *) u8_ptr_add (icmp, 8);

      icmp->type = ICMP4_time_exceeded;
      break;

    case ICMP6_parameter_problem:
      *inner_ip6 = (ip6_header_t *) u8_ptr_add (icmp, 8);

      switch (icmp->code)
	{
	case ICMP6_parameter_problem_erroneous_header_field:
	  icmp->type = ICMP4_parameter_problem;
	  icmp->code = ICMP4_parameter_problem_pointer_indicates_error;
	  u32 pointer = clib_net_to_host_u32 (*((u32 *) (icmp + 1)));
	  if (pointer >= 40)
	    return -1;

	  ((u8 *) (icmp + 1))[0] =
	    icmp6_to_icmp_updater_pointer_table[pointer];
	  break;
	case ICMP6_parameter_problem_unrecognized_next_header:
	  icmp->type = ICMP4_destination_unreachable;
	  icmp->code = ICMP4_destination_unreachable_port_unreachable;
	  break;
	case ICMP6_parameter_problem_unrecognized_option:
	default:
	  return -1;
	}
      break;
    default:
      return -1;
      break;
    }
  return 0;
}

/**
 * @brief Translate TOS value from IPv6 to IPv4.
 *
 * @param ip6 IPv6 header.
 *
 * @returns IPv4 TOS value.
 */
static_always_inline u8
ip6_translate_tos (const ip6_header_t * ip6)
{
  return (clib_net_to_host_u32 (ip6->ip_version_traffic_class_and_flow_label)
	  & 0x0ff00000) >> 20;
}

/**
 * @brief Translate ICMP6 packet to ICMP4.
 *
 * @param p         Buffer to translate.
 * @param fn        The function to translate outer header.
 * @param ctx       A context passed in the outer header translate function.
 * @param inner_fn  The function to translate inner header.
 * @param inner_ctx A context passed in the inner header translate function.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
always_inline int
icmp6_to_icmp (vlib_buffer_t * p, ip6_to_ip4_set_fn_t fn, void *ctx,
	       ip6_to_ip4_set_fn_t inner_fn, void *inner_ctx)
{
  ip6_header_t *ip6, *inner_ip6;
  ip4_header_t *ip4, *inner_ip4;
  u32 ip6_pay_len;
  icmp46_header_t *icmp;
  ip_csum_t csum;
  int rv;

  ip6 = vlib_buffer_get_current (p);
  ip6_pay_len = clib_net_to_host_u16 (ip6->payload_length);
  icmp = (icmp46_header_t *) (ip6 + 1);
  ASSERT (ip6_pay_len + sizeof (*ip6) <= p->current_length);

  //No extensions headers allowed here
  if (ip6->protocol != IP_PROTOCOL_ICMP6)
    return -1;

  //There are no fragmented ICMP messages, so no extension header for now
  if (icmp6_to_icmp_header (icmp, &inner_ip6))
    return -1;

  if (inner_ip6)
    {
      u16 *inner_L4_checksum, inner_l4_offset, inner_frag_offset,
	inner_frag_id;
      u8 *inner_l4, inner_protocol;

      //We have two headers to translate
      //   FROM
      //   [   IPv6   ]<- ext ->[IC][   IPv6   ]<- ext ->[L4 header ...
      // Handled cases:
      //                     [   IPv6   ][IC][   IPv6   ][L4 header ...
      //                 [   IPv6   ][IC][   IPv6   ][Fr][L4 header ...
      //    TO
      //                               [ IPv4][IC][ IPv4][L4 header ...

      if (ip6_parse (inner_ip6, ip6_pay_len - 8,
		     &inner_protocol, &inner_l4_offset, &inner_frag_offset))
	return -1;

      inner_l4 = u8_ptr_add (inner_ip6, inner_l4_offset);
      inner_ip4 =
	(ip4_header_t *) u8_ptr_add (inner_l4, -sizeof (*inner_ip4));
      if (inner_frag_offset)
	{
	  ip6_frag_hdr_t *inner_frag =
	    (ip6_frag_hdr_t *) u8_ptr_add (inner_ip6, inner_frag_offset);
	  inner_frag_id = frag_id_6to4 (inner_frag->identification);
	}
      else
	{
	  inner_frag_id = 0;
	}

      //Do the translation of the inner packet
      if (inner_protocol == IP_PROTOCOL_TCP)
	{
	  inner_L4_checksum = (u16 *) u8_ptr_add (inner_l4, 16);
	}
      else if (inner_protocol == IP_PROTOCOL_UDP)
	{
	  inner_L4_checksum = (u16 *) u8_ptr_add (inner_l4, 6);
	}
      else if (inner_protocol == IP_PROTOCOL_ICMP6)
	{
	  icmp46_header_t *inner_icmp = (icmp46_header_t *) inner_l4;
	  //It cannot be of a different type as ip6_icmp_to_icmp6_in_place succeeded
	  inner_icmp->type = (inner_icmp->type == ICMP6_echo_request) ?
	    ICMP4_echo_request : ICMP4_echo_reply;
	  inner_protocol = IP_PROTOCOL_ICMP;	//Will be copied to ip6 later
	  inner_L4_checksum = &inner_icmp->checksum;
	}
      else
	{
	  return -1;
	}

      csum = *inner_L4_checksum;
      csum = ip_csum_sub_even (csum, inner_ip6->src_address.as_u64[0]);
      csum = ip_csum_sub_even (csum, inner_ip6->src_address.as_u64[1]);
      csum = ip_csum_sub_even (csum, inner_ip6->dst_address.as_u64[0]);
      csum = ip_csum_sub_even (csum, inner_ip6->dst_address.as_u64[1]);
      *inner_L4_checksum = ip_csum_fold (csum);

      if ((rv = inner_fn (inner_ip6, inner_ip4, inner_ctx)) != 0)
	return rv;

      inner_ip4->ip_version_and_header_length =
	IP4_VERSION_AND_HEADER_LENGTH_NO_OPTIONS;
      inner_ip4->tos = ip6_translate_tos (inner_ip6);
      inner_ip4->length =
	u16_net_add (inner_ip6->payload_length,
		     sizeof (*ip4) + sizeof (*ip6) - inner_l4_offset);
      inner_ip4->fragment_id = inner_frag_id;
      inner_ip4->flags_and_fragment_offset =
	clib_host_to_net_u16 (IP4_HEADER_FLAG_MORE_FRAGMENTS);
      inner_ip4->ttl = inner_ip6->hop_limit;
      inner_ip4->protocol = inner_protocol;
      inner_ip4->checksum = ip4_header_checksum (inner_ip4);

      if (inner_ip4->protocol == IP_PROTOCOL_ICMP)
	{
	  //Recompute ICMP checksum
	  icmp46_header_t *inner_icmp = (icmp46_header_t *) inner_l4;
	  inner_icmp->checksum = 0;
	  csum =
	    ip_incremental_checksum (0, inner_icmp,
				     clib_net_to_host_u16 (inner_ip4->length)
				     - sizeof (*inner_ip4));
	  inner_icmp->checksum = ~ip_csum_fold (csum);
	}
      else
	{
	  //Update to new pseudo-header
	  csum = *inner_L4_checksum;
	  csum = ip_csum_add_even (csum, inner_ip4->src_address.as_u32);
	  csum = ip_csum_add_even (csum, inner_ip4->dst_address.as_u32);
	  *inner_L4_checksum = ip_csum_fold (csum);
	}

      //Move up icmp header
      ip4 = (ip4_header_t *) u8_ptr_add (inner_l4, -2 * sizeof (*ip4) - 8);
      clib_memcpy (u8_ptr_add (inner_l4, -sizeof (*ip4) - 8), icmp, 8);
      icmp = (icmp46_header_t *) u8_ptr_add (inner_l4, -sizeof (*ip4) - 8);
    }
  else
    {
      //Only one header to translate
      ip4 = (ip4_header_t *) u8_ptr_add (ip6, sizeof (*ip6) - sizeof (*ip4));
    }

  vlib_buffer_advance (p, (u32) (((u8 *) ip4) - ((u8 *) ip6)));

  if ((rv = fn (ip6, ip4, ctx)) != 0)
    return rv;

  ip4->ip_version_and_header_length =
    IP4_VERSION_AND_HEADER_LENGTH_NO_OPTIONS;
  ip4->tos = ip6_translate_tos (ip6);
  ip4->fragment_id = 0;
  ip4->flags_and_fragment_offset = 0;
  ip4->ttl = ip6->hop_limit;
  ip4->protocol = IP_PROTOCOL_ICMP;
  //TODO fix the length depending on offset length
  ip4->length = u16_net_add (ip6->payload_length,
			     (inner_ip6 ==
			      NULL) ? sizeof (*ip4) : (2 * sizeof (*ip4) -
						       sizeof (*ip6)));
  ip4->checksum = ip4_header_checksum (ip4);

  //Recompute ICMP checksum
  icmp->checksum = 0;
  csum =
    ip_incremental_checksum (0, icmp,
			     clib_net_to_host_u16 (ip4->length) -
			     sizeof (*ip4));
  icmp->checksum = ~ip_csum_fold (csum);

  return 0;
}

/**
 * @brief Translate IPv6 fragmented packet to IPv4.
 *
 * @param p   Buffer to translate.
 * @param fn  The function to translate header.
 * @param ctx A context passed in the header translate function.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
always_inline int
ip6_to_ip4_fragmented (vlib_buffer_t * p, ip6_to_ip4_set_fn_t fn, void *ctx)
{
  ip6_header_t *ip6;
  ip6_frag_hdr_t *frag;
  ip4_header_t *ip4;
  u16 frag_id;
  u8 frag_more;
  u16 frag_offset;
  u8 l4_protocol;
  u16 l4_offset;
  int rv;

  ip6 = vlib_buffer_get_current (p);

  if (ip6_parse
      (ip6, p->current_length, &l4_protocol, &l4_offset, &frag_offset))
    return -1;

  frag = (ip6_frag_hdr_t *) u8_ptr_add (ip6, frag_offset);
  ip4 = (ip4_header_t *) u8_ptr_add (ip6, l4_offset - sizeof (*ip4));
  vlib_buffer_advance (p, l4_offset - sizeof (*ip4));

  frag_id = frag_id_6to4 (frag->identification);
  frag_more = ip6_frag_hdr_more (frag);
  frag_offset = ip6_frag_hdr_offset (frag);

  if ((rv = fn (ip6, ip4, ctx)) != 0)
    return rv;

  ip4->ip_version_and_header_length =
    IP4_VERSION_AND_HEADER_LENGTH_NO_OPTIONS;
  ip4->tos = ip6_translate_tos (ip6);
  ip4->length = u16_net_add (ip6->payload_length,
			     sizeof (*ip4) - l4_offset + sizeof (*ip6));
  ip4->fragment_id = frag_id;
  ip4->flags_and_fragment_offset =
    clib_host_to_net_u16 (frag_offset |
			  (frag_more ? IP4_HEADER_FLAG_MORE_FRAGMENTS : 0));
  ip4->ttl = ip6->hop_limit;
  ip4->protocol =
    (l4_protocol == IP_PROTOCOL_ICMP6) ? IP_PROTOCOL_ICMP : l4_protocol;
  ip4->checksum = ip4_header_checksum (ip4);

  return 0;
}

/**
 * @brief Translate IPv6 UDP/TCP packet to IPv4.
 *
 * @param p   Buffer to translate.
 * @param fn  The function to translate header.
 * @param ctx A context passed in the header translate function.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
always_inline int
ip6_to_ip4_tcp_udp (vlib_buffer_t * p, ip6_to_ip4_set_fn_t fn, void *ctx,
		    u8 udp_checksum)
{
  ip6_header_t *ip6;
  u16 *checksum;
  ip_csum_t csum = 0;
  ip4_header_t *ip4;
  u16 fragment_id;
  u16 flags;
  u16 frag_offset;
  u8 l4_protocol;
  u16 l4_offset;
  int rv;

  ip6 = vlib_buffer_get_current (p);

  if (ip6_parse
      (ip6, p->current_length, &l4_protocol, &l4_offset, &frag_offset))
    return -1;

  if (l4_protocol == IP_PROTOCOL_TCP)
    {
      tcp_header_t *tcp = ip6_next_header (ip6);
      checksum = &tcp->checksum;
    }
  else
    {
      udp_header_t *udp = ip6_next_header (ip6);
      checksum = &udp->checksum;
      //UDP checksum is optional over IPv4
      if (!udp_checksum)
	goto no_csum;
    }

  csum = ip_csum_sub_even (*checksum, ip6->src_address.as_u64[0]);
  csum = ip_csum_sub_even (csum, ip6->src_address.as_u64[1]);
  csum = ip_csum_sub_even (csum, ip6->dst_address.as_u64[0]);
  csum = ip_csum_sub_even (csum, ip6->dst_address.as_u64[1]);
  *checksum = ip_csum_fold (csum);

no_csum:
  ip4 = (ip4_header_t *) u8_ptr_add (ip6, l4_offset - sizeof (*ip4));

  vlib_buffer_advance (p, l4_offset - sizeof (*ip4));

  if (PREDICT_FALSE (frag_offset))
    {
      //Only the first fragment
      ip6_frag_hdr_t *hdr = (ip6_frag_hdr_t *) u8_ptr_add (ip6, frag_offset);
      fragment_id = frag_id_6to4 (hdr->identification);
      flags = clib_host_to_net_u16 (IP4_HEADER_FLAG_MORE_FRAGMENTS);
    }
  else
    {
      fragment_id = 0;
      flags = 0;
    }

  if ((rv = fn (ip6, ip4, ctx)) != 0)
    return rv;

  ip4->ip_version_and_header_length =
    IP4_VERSION_AND_HEADER_LENGTH_NO_OPTIONS;
  ip4->tos = ip6_translate_tos (ip6);
  ip4->length = u16_net_add (ip6->payload_length,
			     sizeof (*ip4) + sizeof (*ip6) - l4_offset);
  ip4->fragment_id = fragment_id;
  ip4->flags_and_fragment_offset = flags;
  ip4->ttl = ip6->hop_limit;
  ip4->protocol = l4_protocol;
  ip4->checksum = ip4_header_checksum (ip4);

  //UDP checksum is optional over IPv4
  if (!udp_checksum && l4_protocol == IP_PROTOCOL_UDP)
    {
      *checksum = 0;
    }
  else
    {
      csum = ip_csum_add_even (*checksum, ip4->dst_address.as_u32);
      csum = ip_csum_add_even (csum, ip4->src_address.as_u32);
      *checksum = ip_csum_fold (csum);
    }

  return 0;
}

/**
 * @brief Translate IPv6 packet to IPv4 (IP header only).
 *
 * @param p   Buffer to translate.
 * @param fn  The function to translate header.
 * @param ctx A context passed in the header translate function.
 *
 * @returns 0 on success, non-zero value otherwise.
 */
always_inline int
ip6_to_ip4 (vlib_buffer_t * p, ip6_to_ip4_set_fn_t fn, void *ctx)
{
  ip6_header_t *ip6;
  ip4_header_t *ip4;
  u16 fragment_id;
  u16 flags;
  u16 frag_offset;
  u8 l4_protocol;
  u16 l4_offset;
  int rv;

  ip6 = vlib_buffer_get_current (p);

  if (ip6_parse
      (ip6, p->current_length, &l4_protocol, &l4_offset, &frag_offset))
    return -1;

  ip4 = (ip4_header_t *) u8_ptr_add (ip6, l4_offset - sizeof (*ip4));

  vlib_buffer_advance (p, l4_offset - sizeof (*ip4));

  if (PREDICT_FALSE (frag_offset))
    {
      //Only the first fragment
      ip6_frag_hdr_t *hdr = (ip6_frag_hdr_t *) u8_ptr_add (ip6, frag_offset);
      fragment_id = frag_id_6to4 (hdr->identification);
      flags = clib_host_to_net_u16 (IP4_HEADER_FLAG_MORE_FRAGMENTS);
    }
  else
    {
      fragment_id = 0;
      flags = 0;
    }

  if ((rv = fn (ip6, ip4, ctx)) != 0)
    return rv;

  ip4->ip_version_and_header_length =
    IP4_VERSION_AND_HEADER_LENGTH_NO_OPTIONS;
  ip4->tos = ip6_translate_tos (ip6);
  ip4->length = u16_net_add (ip6->payload_length,
			     sizeof (*ip4) + sizeof (*ip6) - l4_offset);
  ip4->fragment_id = fragment_id;
  ip4->flags_and_fragment_offset = flags;
  ip4->ttl = ip6->hop_limit;
  ip4->protocol = l4_protocol;
  ip4->checksum = ip4_header_checksum (ip4);

  return 0;
}

#endif /* __included_ip6_to_ip4_h__ */

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */