Merging upstream version 1.44.3.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

6 files changed, 2684 insertions, 0 deletions

diff --git a/fluent-bit/plugins/out_forward/CMakeLists.txt b/fluent-bit/plugins/out_forward/CMakeLists.txt
new file mode 100644
index 000000000..fd639c2eb
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/CMakeLists.txt
@@ -0,0 +1,6 @@
+set(src
+  forward.c
+  forward_format.c
+  )
+
+FLB_PLUGIN(out_forward "${src}" "")
diff --git a/fluent-bit/plugins/out_forward/README.md b/fluent-bit/plugins/out_forward/README.md
new file mode 100644
index 000000000..f08b45619
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/README.md
@@ -0,0 +1,12 @@
+# Fluentd Forward Protocol Implementation
+
+This plugin is based in Fluentd Forward Protocol Spec v1 available here:
+
+- https://github.com/fluent/fluentd/wiki/Forward-Protocol-Specification-v1
+
+The following Event modes are implemented:
+
+- Message Mode
+- Forward Mode
+
+Depending of the configuration, the plugin will decide to go with Message Mode or Forward Mode.
diff --git a/fluent-bit/plugins/out_forward/forward.c b/fluent-bit/plugins/out_forward/forward.c
new file mode 100644
index 000000000..8cc2ca2cc
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/forward.c
@@ -0,0 +1,1832 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  Fluent Bit
+ *  ==========
+ *  Copyright (C) 2015-2022 The Fluent Bit Authors
+ *
+ *  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.
+ */
+
+#include <fluent-bit/flb_output_plugin.h>
+#include <fluent-bit/flb_utils.h>
+#include <fluent-bit/flb_network.h>
+#include <fluent-bit/flb_time.h>
+#include <fluent-bit/flb_upstream.h>
+#include <fluent-bit/flb_upstream_ha.h>
+#include <fluent-bit/flb_hash.h>
+#include <fluent-bit/flb_crypto.h>
+#include <fluent-bit/flb_config_map.h>
+#include <fluent-bit/flb_random.h>
+#include <fluent-bit/flb_gzip.h>
+#include <fluent-bit/flb_log_event.h>
+#include <msgpack.h>
+
+#include "forward.h"
+#include "forward_format.h"
+
+#ifdef FLB_HAVE_UNIX_SOCKET
+#include <sys/socket.h>
+#include <sys/un.h>
+#endif
+
+#define SECURED_BY "Fluent Bit"
+
+pthread_once_t uds_connection_tls_slot_init_once_control = PTHREAD_ONCE_INIT;
+FLB_TLS_DEFINE(struct flb_forward_uds_connection, uds_connection);
+
+void initialize_uds_connection_tls_slot()
+{
+    FLB_TLS_INIT(uds_connection);
+}
+
+#ifdef FLB_HAVE_UNIX_SOCKET
+static flb_sockfd_t forward_unix_connect(struct flb_forward_config *config,
+                                         struct flb_forward *ctx)
+{
+    flb_sockfd_t fd = -1;
+    struct sockaddr_un address;
+
+    if (sizeof(address.sun_path) <= flb_sds_len(config->unix_path)) {
+        flb_plg_error(ctx->ins, "unix_path is too long");
+        return -1;
+    }
+
+    memset(&address, 0, sizeof(struct sockaddr_un));
+
+    fd = flb_net_socket_create(AF_UNIX, FLB_FALSE);
+    if (fd < 0) {
+        flb_plg_error(ctx->ins, "flb_net_socket_create error");
+        return -1;
+    }
+
+    address.sun_family = AF_UNIX;
+    strncpy(address.sun_path, config->unix_path, flb_sds_len(config->unix_path));
+
+    if(connect(fd, (const struct sockaddr*) &address, sizeof(address)) < 0) {
+        flb_errno();
+        close(fd);
+
+        return -1;
+    }
+
+    return fd;
+}
+
+static flb_sockfd_t forward_uds_get_conn(struct flb_forward_config *config,
+                                         struct flb_forward *ctx)
+{
+    struct flb_forward_uds_connection *connection_entry;
+    flb_sockfd_t                       connection;
+
+    connection_entry = FLB_TLS_GET(uds_connection);
+
+    /* We need to allow the code to try to get the value from the TLS
+     * regardless of if it's provided with a config and context because
+     * when we establish the connection we do have both of them but those
+     * are not passed along to the functions in charge of doing IO.
+     */
+
+    if (connection_entry == NULL) {
+        if (config == NULL ||
+            ctx == NULL) {
+            return -1;
+        }
+
+        connection_entry = flb_calloc(1, sizeof(struct flb_forward_uds_connection));
+
+        if (connection_entry == NULL) {
+            flb_errno();
+
+            return -1;
+        }
+
+        connection = forward_unix_connect(config, ctx);
+
+        if (connection == -1) {
+            flb_free(connection_entry);
+
+            return -1;
+        }
+
+        connection_entry->descriptor = connection;
+
+        pthread_mutex_lock(&ctx->uds_connection_list_mutex);
+
+        cfl_list_add(&connection_entry->_head, &ctx->uds_connection_list);
+
+        pthread_mutex_unlock(&ctx->uds_connection_list_mutex);
+
+        FLB_TLS_SET(uds_connection, connection_entry);
+    }
+
+    return connection_entry->descriptor;
+}
+
+static void forward_uds_drop_conn(struct flb_forward *ctx,
+                                  flb_sockfd_t connection)
+{
+    struct flb_forward_uds_connection *connection_entry;
+
+    if (ctx != NULL) {
+        connection_entry = FLB_TLS_GET(uds_connection);
+
+        if (connection_entry != NULL) {
+            pthread_mutex_lock(&ctx->uds_connection_list_mutex);
+
+            if (connection == connection_entry->descriptor) {
+                close(connection);
+
+                if (!cfl_list_entry_is_orphan(&connection_entry->_head)) {
+                    cfl_list_del(&connection_entry->_head);
+                }
+
+                free(connection_entry);
+
+                FLB_TLS_SET(uds_connection, NULL);
+            }
+
+            pthread_mutex_unlock(&ctx->uds_connection_list_mutex);
+        }
+    }
+}
+
+static void forward_uds_drop_all(struct flb_forward *ctx)
+{
+    struct flb_forward_uds_connection *connection_entry;
+    struct cfl_list                   *head;
+    struct cfl_list                   *tmp;
+
+    if (ctx != NULL) {
+        pthread_mutex_lock(&ctx->uds_connection_list_mutex);
+
+        cfl_list_foreach_safe(head, tmp, &ctx->uds_connection_list) {
+            connection_entry = cfl_list_entry(head,
+                                              struct flb_forward_uds_connection,
+                                              _head);
+
+            if (connection_entry->descriptor != -1) {
+                close(connection_entry->descriptor);
+
+                connection_entry->descriptor = -1;
+            }
+
+            if (!cfl_list_entry_is_orphan(&connection_entry->_head)) {
+                cfl_list_del(&connection_entry->_head);
+            }
+
+            free(connection_entry);
+        }
+
+        pthread_mutex_unlock(&ctx->uds_connection_list_mutex);
+    }
+}
+
+/* In these functions forward_uds_get_conn
+ * should not return -1 because it should have been
+ * called earlier with a proper context and it should
+ * have saved a file descriptor to the TLS.
+ */
+
+static int io_unix_write(struct flb_connection *unused, int deprecated_fd, const void* data,
+                         size_t len, size_t *out_len)
+{
+    flb_sockfd_t uds_conn;
+
+    uds_conn = forward_uds_get_conn(NULL, NULL);
+
+    return flb_io_fd_write(uds_conn, data, len, out_len);
+}
+
+static int io_unix_read(struct flb_connection *unused, int deprecated_fd, void* buf,size_t len)
+{
+    flb_sockfd_t uds_conn;
+
+    uds_conn = forward_uds_get_conn(NULL, NULL);
+
+    return flb_io_fd_read(uds_conn, buf, len);
+}
+
+#else
+
+static flb_sockfd_t forward_uds_get_conn(struct flb_forward_config *config,
+                                         struct flb_forward *ctx)
+{
+    (void) config;
+    (void) ctx;
+
+    return -1;
+}
+
+static void forward_uds_drop_conn(struct flb_forward *ctx,
+                                  flb_sockfd_t connection)
+{
+    (void) ctx;
+    (void) connection;
+}
+
+static void forward_uds_drop_all(struct flb_forward *ctx)
+{
+    (void) ctx;
+}
+
+#endif
+
+#ifdef FLB_HAVE_TLS
+
+static int io_net_write(struct flb_connection *conn, int unused_fd,
+                        const void* data, size_t len, size_t *out_len)
+{
+    return flb_io_net_write(conn, data, len, out_len);
+}
+
+static int io_net_read(struct flb_connection *conn, int unused_fd,
+                       void* buf, size_t len)
+{
+    return flb_io_net_read(conn, buf, len);
+}
+
+static int secure_forward_init(struct flb_forward *ctx,
+                               struct flb_forward_config *fc)
+{
+    return 0;
+}
+
+#endif
+
+static inline void print_msgpack_status(struct flb_forward *ctx,
+                                        int ret, char *context)
+{
+    switch (ret) {
+    case MSGPACK_UNPACK_EXTRA_BYTES:
+        flb_plg_error(ctx->ins, "%s MSGPACK_UNPACK_EXTRA_BYTES", context);
+        break;
+    case MSGPACK_UNPACK_CONTINUE:
+        flb_plg_trace(ctx->ins, "%s MSGPACK_UNPACK_CONTINUE", context);
+        break;
+    case MSGPACK_UNPACK_PARSE_ERROR:
+        flb_plg_error(ctx->ins, "%s MSGPACK_UNPACK_PARSE_ERROR", context);
+        break;
+    case MSGPACK_UNPACK_NOMEM_ERROR:
+        flb_plg_error(ctx->ins, "%s MSGPACK_UNPACK_NOMEM_ERROR", context);
+        break;
+    }
+}
+
+/* Read a secure forward msgpack message */
+static int secure_forward_read(struct flb_forward *ctx,
+                               struct flb_connection *u_conn,
+                               struct flb_forward_config *fc,
+                               char *buf, size_t size, size_t *out_len)
+{
+    int ret;
+    size_t off;
+    size_t avail;
+    size_t buf_off = 0;
+    msgpack_unpacked result;
+
+    msgpack_unpacked_init(&result);
+    while (1) {
+        avail = size - buf_off;
+        if (avail < 1) {
+            goto error;
+        }
+
+        /* Read the message */
+        ret = fc->io_read(u_conn, fc->unix_fd, buf + buf_off, size - buf_off);
+        if (ret <= 0) {
+            goto error;
+        }
+        buf_off += ret;
+
+        /* Validate */
+        off = 0;
+        ret = msgpack_unpack_next(&result, buf, buf_off, &off);
+        switch (ret) {
+        case MSGPACK_UNPACK_SUCCESS:
+            msgpack_unpacked_destroy(&result);
+            *out_len = buf_off;
+            return 0;
+        default:
+            print_msgpack_status(ctx, ret, "handshake");
+            goto error;
+        };
+    }
+
+ error:
+    msgpack_unpacked_destroy(&result);
+    return -1;
+}
+
+static void secure_forward_set_ping(struct flb_forward_ping *ping,
+                                    msgpack_object *map)
+{
+    int i;
+    msgpack_object key;
+    msgpack_object val;
+    const char *ptr;
+    int len;
+
+    memset(ping, 0, sizeof(struct flb_forward_ping));
+    ping->keepalive = 1; /* default, as per spec */
+
+    for (i = 0; i < map->via.map.size; i++) {
+        key = map->via.map.ptr[i].key;
+        val = map->via.map.ptr[i].val;
+
+        ptr = key.via.str.ptr;
+        len = key.via.str.size;
+
+        if (len == 5 && memcmp(ptr, "nonce", len) == 0) {
+            ping->nonce = val.via.bin.ptr;
+            ping->nonce_len = val.via.bin.size;
+        }
+        else if (len == 4 && memcmp(ptr, "auth", len) == 0) {
+            ping->auth = val.via.bin.ptr;
+            ping->auth_len = val.via.bin.size;
+        }
+        else if (len == 9 && memcmp(ptr, "keepalive", len) == 0) {
+            ping->keepalive = val.via.boolean;
+        }
+    }
+}
+
+static int secure_forward_hash_shared_key(struct flb_forward_config *fc,
+                                          struct flb_forward_ping *ping,
+                                          char *buf, int buflen)
+{
+    size_t             length_entries[4];
+    unsigned char     *data_entries[4];
+    uint8_t            hash[64];
+    int                result;
+
+    if (buflen < 128) {
+        return -1;
+    }
+
+    data_entries[0]   = (unsigned char *) fc->shared_key_salt;
+    length_entries[0] = 16;
+
+    data_entries[1]   = (unsigned char *) fc->self_hostname;
+    length_entries[1] = strlen(fc->self_hostname);
+
+    data_entries[2]   = (unsigned char *) ping->nonce;
+    length_entries[2] = ping->nonce_len;
+
+    data_entries[3]   = (unsigned char *) fc->shared_key;
+    length_entries[3] = strlen(fc->shared_key);
+
+    result = flb_hash_simple_batch(FLB_HASH_SHA512,
+                                   4,
+                                   data_entries,
+                                   length_entries,
+                                   hash,
+                                   sizeof(hash));
+
+    if (result != FLB_CRYPTO_SUCCESS) {
+        return -1;
+    }
+
+    flb_forward_format_bin_to_hex(hash, 64, buf);
+
+    return 0;
+}
+
+static int secure_forward_hash_password(struct flb_forward_config *fc,
+                                        struct flb_forward_ping *ping,
+                                        char *buf, int buflen)
+{
+    size_t             length_entries[3];
+    unsigned char     *data_entries[3];
+    uint8_t            hash[64];
+    int                result;
+
+    if (buflen < 128) {
+        return -1;
+    }
+
+    data_entries[0]   = (unsigned char *) ping->auth;
+    length_entries[0] = ping->auth_len;
+
+    data_entries[1]   = (unsigned char *) fc->username;
+    length_entries[1] = strlen(fc->username);
+
+    data_entries[2]   = (unsigned char *) fc->password;
+    length_entries[2] = strlen(fc->password);
+
+    result = flb_hash_simple_batch(FLB_HASH_SHA512,
+                                   3,
+                                   data_entries,
+                                   length_entries,
+                                   hash,
+                                   sizeof(hash));
+
+    if (result != FLB_CRYPTO_SUCCESS) {
+        return -1;
+    }
+
+    flb_forward_format_bin_to_hex(hash, 64, buf);
+
+    return 0;
+}
+
+static int secure_forward_ping(struct flb_connection *u_conn,
+                               msgpack_object map,
+                               struct flb_forward_config *fc,
+                               struct flb_forward *ctx)
+{
+    int ret;
+    size_t bytes_sent;
+    char shared_key_hexdigest[128];
+    char password_hexdigest[128];
+    msgpack_sbuffer mp_sbuf;
+    msgpack_packer mp_pck;
+    struct flb_forward_ping ping;
+
+    secure_forward_set_ping(&ping, &map);
+
+    if (ping.nonce == NULL) {
+        flb_plg_error(ctx->ins, "nonce not found");
+        return -1;
+    }
+
+    if (secure_forward_hash_shared_key(fc, &ping, shared_key_hexdigest, 128)) {
+        flb_plg_error(ctx->ins, "failed to hash shared_key");
+        return -1;
+    }
+
+    if (ping.auth != NULL) {
+        if (secure_forward_hash_password(fc, &ping, password_hexdigest, 128)) {
+            flb_plg_error(ctx->ins, "failed to hash password");
+            return -1;
+        }
+    }
+
+    /* Prepare outgoing msgpack PING */
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+    msgpack_pack_array(&mp_pck, 6);
+
+    /* [0] PING */
+    msgpack_pack_str(&mp_pck, 4);
+    msgpack_pack_str_body(&mp_pck, "PING", 4);
+
+    /* [1] Hostname */
+    msgpack_pack_str(&mp_pck, flb_sds_len(fc->self_hostname));
+    msgpack_pack_str_body(&mp_pck, fc->self_hostname,
+                          flb_sds_len(fc->self_hostname));
+
+    /* [2] Shared key salt */
+    msgpack_pack_str(&mp_pck, 16);
+    msgpack_pack_str_body(&mp_pck, fc->shared_key_salt, 16);
+
+    /* [3] Shared key in Hexdigest format */
+    msgpack_pack_str(&mp_pck, 128);
+    msgpack_pack_str_body(&mp_pck, shared_key_hexdigest, 128);
+
+    /* [4] Username and password (optional) */
+    if (ping.auth != NULL) {
+        msgpack_pack_str(&mp_pck, strlen(fc->username));
+        msgpack_pack_str_body(&mp_pck, fc->username, strlen(fc->username));
+        msgpack_pack_str(&mp_pck, 128);
+        msgpack_pack_str_body(&mp_pck, password_hexdigest, 128);
+    }
+    else {
+        msgpack_pack_str(&mp_pck, 0);
+        msgpack_pack_str_body(&mp_pck, "", 0);
+        msgpack_pack_str(&mp_pck, 0);
+        msgpack_pack_str_body(&mp_pck, "", 0);
+    }
+
+    ret = fc->io_write(u_conn, fc->unix_fd, mp_sbuf.data, mp_sbuf.size, &bytes_sent);
+    flb_plg_debug(ctx->ins, "PING sent: ret=%i bytes sent=%lu", ret, bytes_sent);
+
+    msgpack_sbuffer_destroy(&mp_sbuf);
+
+    if (ret > -1 && bytes_sent > 0) {
+        return 0;
+    }
+
+    return -1;
+}
+
+static int secure_forward_pong(struct flb_forward *ctx, char *buf, int buf_size)
+{
+    int ret;
+    char msg[32] = {0};
+    size_t off = 0;
+    msgpack_unpacked result;
+    msgpack_object root;
+    msgpack_object o;
+
+    msgpack_unpacked_init(&result);
+    ret = msgpack_unpack_next(&result, buf, buf_size, &off);
+    if (ret != MSGPACK_UNPACK_SUCCESS) {
+        return -1;
+    }
+
+    root = result.data;
+    if (root.type != MSGPACK_OBJECT_ARRAY) {
+        goto error;
+    }
+
+    if (root.via.array.size < 4) {
+        goto error;
+    }
+
+    o = root.via.array.ptr[0];
+    if (o.type != MSGPACK_OBJECT_STR) {
+        goto error;
+    }
+
+    if (strncmp(o.via.str.ptr, "PONG", 4) != 0 || o.via.str.size != 4) {
+        goto error;
+    }
+
+    o = root.via.array.ptr[1];
+    if (o.type != MSGPACK_OBJECT_BOOLEAN) {
+        goto error;
+    }
+
+    if (o.via.boolean) {
+        msgpack_unpacked_destroy(&result);
+        return 0;
+    }
+    else {
+        o = root.via.array.ptr[2];
+        memcpy(msg, o.via.str.ptr, o.via.str.size);
+        flb_plg_error(ctx->ins, "failed authorization: %s", msg);
+    }
+
+ error:
+    msgpack_unpacked_destroy(&result);
+    return -1;
+}
+
+static int secure_forward_handshake(struct flb_connection *u_conn,
+                                    struct flb_forward_config *fc,
+                                    struct flb_forward *ctx)
+{
+    int ret;
+    char buf[1024];
+    size_t out_len;
+    size_t off;
+    msgpack_unpacked result;
+    msgpack_object root;
+    msgpack_object o;
+
+    /* Wait for server HELO */
+    ret = secure_forward_read(ctx, u_conn, fc, buf, sizeof(buf) - 1, &out_len);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "handshake error expecting HELO");
+        return -1;
+    }
+
+    /* Unpack message and validate */
+    off = 0;
+    msgpack_unpacked_init(&result);
+    ret = msgpack_unpack_next(&result, buf, out_len, &off);
+    if (ret != MSGPACK_UNPACK_SUCCESS) {
+        print_msgpack_status(ctx, ret, "HELO");
+        return -1;
+    }
+
+    /* Parse HELO message */
+    root = result.data;
+    if (root.via.array.size < 2) {
+        flb_plg_error(ctx->ins, "Invalid HELO message");
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    o = root.via.array.ptr[0];
+    if (o.type != MSGPACK_OBJECT_STR) {
+        flb_plg_error(ctx->ins, "Invalid HELO type message");
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    if (strncmp(o.via.str.ptr, "HELO", 4) != 0 || o.via.str.size != 4) {
+        flb_plg_error(ctx->ins, "Invalid HELO content message");
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    flb_plg_debug(ctx->ins, "protocol: received HELO");
+
+    /* Compose and send PING message */
+    o = root.via.array.ptr[1];
+    ret = secure_forward_ping(u_conn, o, fc, ctx);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "Failed PING");
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    /* Expect a PONG */
+    ret = secure_forward_read(ctx, u_conn, fc, buf, sizeof(buf) - 1, &out_len);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "handshake error expecting HELO");
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    /* Process PONG */
+    ret = secure_forward_pong(ctx, buf, out_len);
+    if (ret == -1) {
+        msgpack_unpacked_destroy(&result);
+        return -1;
+    }
+
+    msgpack_unpacked_destroy(&result);
+    return 0;
+}
+
+static int forward_read_ack(struct flb_forward *ctx,
+                            struct flb_forward_config *fc,
+                            struct flb_connection *u_conn,
+                            char *chunk, int chunk_len)
+{
+    int ret;
+    int i;
+    size_t out_len;
+    size_t off;
+    const char *ack;
+    size_t ack_len;
+    msgpack_unpacked result;
+    msgpack_object root;
+    msgpack_object_map map;
+    msgpack_object key;
+    msgpack_object val;
+    char buf[512];  /* ack should never be bigger */
+
+    flb_plg_trace(ctx->ins, "wait ACK (%.*s)", chunk_len, chunk);
+
+    /* Wait for server ACK */
+    ret = secure_forward_read(ctx, u_conn, fc, buf, sizeof(buf) - 1, &out_len);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "cannot get ack");
+        return -1;
+    }
+
+    /* Unpack message and validate */
+    off = 0;
+    msgpack_unpacked_init(&result);
+    ret = msgpack_unpack_next(&result, buf, out_len, &off);
+    if (ret != MSGPACK_UNPACK_SUCCESS) {
+        print_msgpack_status(ctx, ret, "ACK");
+        goto error;
+    }
+
+    /* Parse ACK message */
+    root = result.data;
+    if (root.type != MSGPACK_OBJECT_MAP) {
+        flb_plg_error(ctx->ins, "ACK response not MAP (type:%d)", root.type);
+        goto error;
+    }
+
+    map = root.via.map;
+    ack = NULL;
+    /* Lookup ack field */
+    for (i = 0; i < map.size; i++) {
+        key = map.ptr[i].key;
+        if (key.via.str.size == 3 && strncmp(key.via.str.ptr, "ack", 3) == 0) {
+            val     = map.ptr[i].val;
+            ack_len = val.via.str.size;
+            ack     = val.via.str.ptr;
+            break;
+        }
+    }
+
+    if (!ack) {
+        flb_plg_error(ctx->ins, "ack: ack not found");
+        goto error;
+    }
+
+    if (ack_len != chunk_len) {
+        flb_plg_error(ctx->ins,
+                      "ack: ack len does not match ack(%ld)(%.*s) chunk(%d)(%.*s)",
+                      ack_len, (int) ack_len, ack,
+                      chunk_len, (int) chunk_len, chunk);
+        goto error;
+    }
+
+    if (strncmp(ack, chunk, ack_len) != 0) {
+        flb_plg_error(ctx->ins, "ACK: mismatch received=%s, expected=(%.*s)",
+                      ack, chunk_len, chunk);
+        goto error;
+    }
+
+    flb_plg_debug(ctx->ins, "protocol: received ACK %.*s", (int)ack_len, ack);
+    msgpack_unpacked_destroy(&result);
+    return 0;
+
+ error:
+    msgpack_unpacked_destroy(&result);
+    return -1;
+}
+
+
+static int forward_config_init(struct flb_forward_config *fc,
+                               struct flb_forward *ctx)
+{
+    if (fc->io_read == NULL || fc->io_write == NULL) {
+        flb_plg_error(ctx->ins, "io_read/io_write is NULL");
+        return -1;
+    }
+
+#ifdef FLB_HAVE_TLS
+    /* Initialize Secure Forward mode */
+    if (fc->secured == FLB_TRUE) {
+        secure_forward_init(ctx, fc);
+    }
+#endif
+
+    /* Generate the shared key salt */
+    if (flb_random_bytes(fc->shared_key_salt, 16)) {
+        flb_plg_error(ctx->ins, "cannot generate shared key salt");
+        return -1;
+    }
+
+    mk_list_add(&fc->_head, &ctx->configs);
+    return 0;
+}
+
+static flb_sds_t config_get_property(char *prop,
+                                     struct flb_upstream_node *node,
+                                     struct flb_forward *ctx)
+{
+    if (node) {
+        return (flb_sds_t) flb_upstream_node_get_property(prop, node);
+    }
+    else {
+        return (flb_sds_t) flb_output_get_property(prop, ctx->ins);
+    }
+}
+
+static int config_set_properties(struct flb_upstream_node *node,
+                                 struct flb_forward_config *fc,
+                                 struct flb_forward *ctx)
+{
+    flb_sds_t tmp;
+
+    /* Shared Key */
+    tmp = config_get_property("empty_shared_key", node, ctx);
+    if (tmp && flb_utils_bool(tmp)) {
+        fc->empty_shared_key = FLB_TRUE;
+    }
+    else {
+        fc->empty_shared_key = FLB_FALSE;
+    }
+
+    tmp = config_get_property("shared_key", node, ctx);
+    if (fc->empty_shared_key) {
+        fc->shared_key = flb_sds_create("");
+    }
+    else if (tmp) {
+        fc->shared_key = flb_sds_create(tmp);
+    }
+    else {
+        fc->shared_key = NULL;
+    }
+
+    tmp = config_get_property("username", node, ctx);
+    if (tmp) {
+        fc->username = tmp;
+    }
+    else {
+        fc->username = "";
+    }
+
+    tmp = config_get_property("password", node, ctx);
+    if (tmp) {
+        fc->password = tmp;
+    }
+    else {
+        fc->password = "";
+    }
+
+    /* Self Hostname */
+    tmp = config_get_property("self_hostname", node, ctx);
+    if (tmp) {
+        fc->self_hostname = flb_sds_create(tmp);
+    }
+    else {
+        fc->self_hostname = flb_sds_create("localhost");
+    }
+
+    /* Backward compatible timing mode */
+    tmp = config_get_property("time_as_integer", node, ctx);
+    if (tmp) {
+        fc->time_as_integer = flb_utils_bool(tmp);
+    }
+    else {
+        fc->time_as_integer = FLB_FALSE;
+    }
+
+    /* send always options (with size) */
+    tmp = config_get_property("send_options", node, ctx);
+    if (tmp) {
+        fc->send_options = flb_utils_bool(tmp);
+    }
+
+    /* add_option -> extra_options: if the user has defined 'add_option'
+     * we need to enable the 'send_options' flag
+     */
+    if (fc->extra_options && mk_list_size(fc->extra_options) > 0) {
+        fc->send_options = FLB_TRUE;
+    }
+
+    /* require ack response  (implies send_options) */
+    tmp = config_get_property("require_ack_response", node, ctx);
+    if (tmp) {
+        fc->require_ack_response = flb_utils_bool(tmp);
+        if (fc->require_ack_response) {
+            fc->send_options = FLB_TRUE;
+        }
+    }
+
+    /* Tag Overwrite */
+    tmp = config_get_property("tag", node, ctx);
+    if (tmp) {
+        /* Set the tag */
+        fc->tag = flb_sds_create(tmp);
+        if (!fc->tag) {
+            flb_plg_error(ctx->ins, "cannot allocate tag");
+            return -1;
+        }
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+        /* Record Accessor */
+        fc->ra_tag = flb_ra_create(fc->tag, FLB_TRUE);
+        if (!fc->ra_tag) {
+            flb_plg_error(ctx->ins, "cannot create record accessor for tag: %s",
+                          fc->tag);
+            return -1;
+        }
+
+        /* Static record accessor ? (no dynamic values from map) */
+        fc->ra_static = flb_ra_is_static(fc->ra_tag);
+#endif
+    }
+    else {
+        fc->tag = NULL;
+
+    }
+
+    /* compress (implies send_options) */
+    tmp = config_get_property("compress", node, ctx);
+    if (tmp) {
+        if (!strcasecmp(tmp, "text")) {
+            fc->compress = COMPRESS_NONE;
+        }
+        else if (!strcasecmp(tmp, "gzip")) {
+            fc->compress = COMPRESS_GZIP;
+            fc->send_options = FLB_TRUE;
+        }
+        else {
+            flb_plg_error(ctx->ins, "invalid compress mode: %s", tmp);
+            return -1;
+        }
+    }
+    else {
+        fc->compress = COMPRESS_NONE;
+    }
+
+    if (fc->compress != COMPRESS_NONE && fc->time_as_integer == FLB_TRUE) {
+        flb_plg_error(ctx->ins, "compress mode %s is incompatible with "
+                      "time_as_integer", tmp);
+        return -1;
+    }
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    if (fc->compress != COMPRESS_NONE &&
+        (fc->ra_tag && fc->ra_static == FLB_FALSE) ) {
+        flb_plg_error(ctx->ins, "compress mode %s is incompatible with dynamic "
+                      "tags", tmp);
+        return -1;
+    }
+#endif
+
+    return 0;
+}
+
+static void forward_config_destroy(struct flb_forward_config *fc)
+{
+    flb_sds_destroy(fc->shared_key);
+    flb_sds_destroy(fc->self_hostname);
+    flb_sds_destroy(fc->tag);
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    if (fc->ra_tag) {
+        flb_ra_destroy(fc->ra_tag);
+    }
+#endif
+
+    flb_free(fc);
+}
+
+/* Configure in HA mode */
+static int forward_config_ha(const char *upstream_file,
+                             struct flb_forward *ctx,
+                             struct flb_config *config)
+{
+    int ret;
+    struct mk_list *head;
+    struct flb_upstream_node *node;
+    struct flb_forward_config *fc = NULL;
+
+    ctx->ha_mode = FLB_TRUE;
+    ctx->ha = flb_upstream_ha_from_file(upstream_file, config);
+    if (!ctx->ha) {
+        flb_plg_error(ctx->ins, "cannot load Upstream file");
+        return -1;
+    }
+
+    /* Iterate nodes and create a forward_config context */
+    mk_list_foreach(head, &ctx->ha->nodes) {
+        node = mk_list_entry(head, struct flb_upstream_node, _head);
+
+        /* create forward_config context */
+        fc = flb_calloc(1, sizeof(struct flb_forward_config));
+        if (!fc) {
+            flb_errno();
+            flb_plg_error(ctx->ins, "failed config allocation");
+            continue;
+        }
+        fc->unix_fd = -1;
+        fc->secured = FLB_FALSE;
+        fc->io_write = io_net_write;
+        fc->io_read  = io_net_read;
+
+        /* Is TLS enabled ? */
+        if (node->tls_enabled == FLB_TRUE) {
+            fc->secured = FLB_TRUE;
+        }
+
+        /* Read properties into 'fc' context */
+        config_set_properties(node, fc, ctx);
+
+        /* Initialize and validate forward_config context */
+        ret = forward_config_init(fc, ctx);
+        if (ret == -1) {
+            if (fc) {
+                forward_config_destroy(fc);
+            }
+            return -1;
+        }
+
+        /* Set our forward_config context into the node */
+        flb_upstream_node_set_data(fc, node);
+    }
+
+    flb_output_upstream_ha_set(ctx->ha, ctx->ins);
+
+    return 0;
+}
+
+static int forward_config_simple(struct flb_forward *ctx,
+                                 struct flb_output_instance *ins,
+                                 struct flb_config *config)
+{
+    int ret;
+    int io_flags;
+    struct flb_forward_config *fc = NULL;
+    struct flb_upstream *upstream;
+
+    /* Set default network configuration if not set */
+    flb_output_net_default("127.0.0.1", 24224, ins);
+
+    /* Configuration context */
+    fc = flb_calloc(1, sizeof(struct flb_forward_config));
+    if (!fc) {
+        flb_errno();
+        return -1;
+    }
+    fc->unix_fd = -1;
+    fc->secured = FLB_FALSE;
+    fc->io_write = NULL;
+    fc->io_read  = NULL;
+
+    /* Set default values */
+    ret = flb_output_config_map_set(ins, fc);
+    if (ret == -1) {
+        flb_free(fc);
+        return -1;
+    }
+
+    /* Check if TLS is enabled */
+#ifdef FLB_HAVE_TLS
+    if (ins->use_tls == FLB_TRUE) {
+        io_flags = FLB_IO_TLS;
+        fc->secured = FLB_TRUE;
+    }
+    else {
+        io_flags = FLB_IO_TCP;
+    }
+#else
+    io_flags = FLB_IO_TCP;
+#endif
+
+    if (ins->host.ipv6 == FLB_TRUE) {
+        io_flags |= FLB_IO_IPV6;
+    }
+
+    if (fc->unix_path) {
+#ifdef FLB_HAVE_UNIX_SOCKET
+        /* In older versions if the UDS server was not up
+         * at this point fluent-bit would fail because it
+         * would not be able to establish the conntection.
+         *
+         * With the concurrency fixes we moved the connection
+         * to a later stage which will cause fluent-bit to
+         * properly launch but if the UDS server is not
+         * available at flush time then an error similar to
+         * the one we would get for a network based output
+         * plugin will be logged and FLB_RETRY will be returned.
+         */
+
+        fc->io_write = io_unix_write;
+        fc->io_read  = io_unix_read;
+#else
+        flb_plg_error(ctx->ins, "unix_path is not supported");
+        flb_free(fc);
+        flb_free(ctx);
+        return -1;
+#endif /* FLB_HAVE_UNIX_SOCKET */
+    }
+    else {
+        /* Prepare an upstream handler */
+        upstream = flb_upstream_create(config,
+                                       ins->host.name,
+                                       ins->host.port,
+                                       io_flags, ins->tls);
+        if (!upstream) {
+            flb_free(fc);
+            flb_free(ctx);
+            return -1;
+        }
+        fc->io_write = io_net_write;
+        fc->io_read  = io_net_read;
+        ctx->u = upstream;
+        flb_output_upstream_set(ctx->u, ins);
+    }
+    /* Read properties into 'fc' context */
+    config_set_properties(NULL, fc, ctx);
+
+    /* Initialize and validate forward_config context */
+    ret = forward_config_init(fc, ctx);
+    if (ret == -1) {
+        if (fc) {
+            forward_config_destroy(fc);
+        }
+        return -1;
+    }
+
+    return 0;
+}
+
+static int cb_forward_init(struct flb_output_instance *ins,
+                           struct flb_config *config, void *data)
+{
+    int ret;
+    const char *tmp;
+    struct flb_forward *ctx;
+    (void) data;
+
+    ctx = flb_calloc(1, sizeof(struct flb_forward));
+    if (!ctx) {
+        flb_errno();
+        return -1;
+    }
+
+    ret = pthread_once(&uds_connection_tls_slot_init_once_control,
+                       initialize_uds_connection_tls_slot);
+
+    if (ret != 0) {
+        flb_errno();
+        flb_free(ctx);
+
+        return -1;
+    }
+
+    ret = pthread_mutex_init(&ctx->uds_connection_list_mutex, NULL);
+
+    if (ret != 0) {
+        flb_errno();
+        flb_free(ctx);
+
+        return -1;
+    }
+
+    cfl_list_init(&ctx->uds_connection_list);
+
+    ctx->ins = ins;
+    mk_list_init(&ctx->configs);
+    flb_output_set_context(ins, ctx);
+
+
+    /* Configure HA or simple mode ? */
+    tmp = flb_output_get_property("upstream", ins);
+    if (tmp) {
+        ret = forward_config_ha(tmp, ctx, config);
+    }
+    else {
+        ret = forward_config_simple(ctx, ins, config);
+    }
+
+    return ret;
+}
+
+struct flb_forward_config *flb_forward_target(struct flb_forward *ctx,
+                                              struct flb_upstream_node **node)
+{
+    struct flb_forward_config *fc = NULL;
+    struct flb_upstream_node *f_node;
+
+    if (ctx->ha_mode == FLB_TRUE) {
+        f_node = flb_upstream_ha_node_get(ctx->ha);
+        if (!f_node) {
+            return NULL;
+        }
+
+        /* Get forward_config stored in node opaque data */
+        fc = flb_upstream_node_get_data(f_node);
+        *node = f_node;
+    }
+    else {
+        fc = mk_list_entry_first(&ctx->configs,
+                                 struct flb_forward_config,
+                                 _head);
+        *node = NULL;
+    }
+    return fc;
+}
+
+static int flush_message_mode(struct flb_forward *ctx,
+                              struct flb_forward_config *fc,
+                              struct flb_connection *u_conn,
+                              char *buf, size_t size)
+{
+    int ret;
+    int ok = MSGPACK_UNPACK_SUCCESS;
+    size_t sent = 0;
+    size_t rec_size;
+    size_t pre = 0;
+    size_t off = 0;
+    msgpack_object root;
+    msgpack_object options;
+    msgpack_object chunk;
+    msgpack_unpacked result;
+
+    /* If the sender requires 'ack' from the remote end-point */
+    if (fc->require_ack_response) {
+        msgpack_unpacked_init(&result);
+        while (msgpack_unpack_next(&result, buf, size, &off) == ok) {
+            /* get the record size */
+            rec_size = off - pre;
+
+            /* write single message */
+            ret = fc->io_write(u_conn,fc->unix_fd,
+                                   buf + pre, rec_size, &sent);
+            pre = off;
+
+            if (ret == -1) {
+                /*
+                 * FIXME: we might take advantage of 'flush_ctx' and store the
+                 * message that failed it delivery, we could have retries but with
+                 * the flush context.
+                 */
+                flb_plg_error(ctx->ins, "message_mode: error sending message");
+                msgpack_unpacked_destroy(&result);
+                return FLB_RETRY;
+            }
+
+            /* Sucessful delivery, now get message 'chunk' and wait for it */
+            root = result.data;
+            options = root.via.array.ptr[3];
+            chunk = options.via.map.ptr[0].val;
+
+            /* Read ACK */
+            ret = forward_read_ack(ctx, fc, u_conn,
+                                   (char *) chunk.via.str.ptr, chunk.via.str.size);
+            if (ret == -1) {
+                msgpack_unpacked_destroy(&result);
+                return FLB_RETRY;
+            }
+        }
+
+        /* All good */
+        msgpack_unpacked_destroy(&result);
+        return FLB_OK;
+    }
+
+    /* Normal data write */
+    ret = fc->io_write(u_conn, fc->unix_fd, buf, size, &sent);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "message_mode: error sending data");
+        return FLB_RETRY;
+    }
+
+    return FLB_OK;
+}
+
+/* pack payloads of cmetrics or ctraces with Fluentd compat format */
+static int pack_metricses_payload(msgpack_packer *mp_pck, const void *data, size_t bytes) {
+    int entries;
+    struct flb_time tm;
+
+    /* Format with event stream format of entries: [[time, [{entries map}]]] */
+    msgpack_pack_array(mp_pck, 1);
+    msgpack_pack_array(mp_pck, 2);
+    flb_time_get(&tm);
+    flb_time_append_to_msgpack(&tm, mp_pck, 0);
+    entries = flb_mp_count(data, bytes);
+    msgpack_pack_array(mp_pck, entries);
+
+    return 0;
+}
+
+#include <fluent-bit/flb_pack.h>
+/*
+ * Forward Mode: this is the generic mechanism used in Fluent Bit, it takes
+ * advantage of the internal data representation and avoid re-formatting data,
+ * it only sends a msgpack header, pre-existent 'data' records and options.
+ *
+ * note: if the user has enabled time_as_integer (compat mode for Fluentd <= 0.12),
+ * the 'flush_forward_compat_mode' is used instead.
+ */
+static int flush_forward_mode(struct flb_forward *ctx,
+                              struct flb_forward_config *fc,
+                              struct flb_connection *u_conn,
+                              int event_type,
+                              const char *tag, int tag_len,
+                              const void *data, size_t bytes,
+                              char *opts_buf, size_t opts_size)
+{
+    int ret;
+    int entries;
+    int send_options;
+    size_t off = 0;
+    size_t bytes_sent;
+    msgpack_object root;
+    msgpack_object chunk;
+    msgpack_unpacked result;
+    msgpack_sbuffer mp_sbuf;
+    msgpack_packer mp_pck;
+    void *final_data;
+    size_t final_bytes;
+    char *transcoded_buffer;
+    size_t transcoded_length;
+
+    transcoded_buffer = NULL;
+    transcoded_length = 0;
+
+    /* Pack message header */
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+
+    send_options = fc->send_options;
+    if (event_type == FLB_EVENT_TYPE_METRICS || event_type == FLB_EVENT_TYPE_TRACES) {
+        send_options = FLB_TRUE;
+    }
+    msgpack_pack_array(&mp_pck, send_options ? 3 : 2);
+
+    /* Tag */
+    flb_forward_format_append_tag(ctx, fc, &mp_pck, NULL, tag, tag_len);
+
+    if (!fc->fwd_retain_metadata && event_type == FLB_EVENT_TYPE_LOGS) {
+        ret = flb_forward_format_transcode(ctx, FLB_LOG_EVENT_FORMAT_FORWARD,
+                                           (char *) data, bytes,
+                                           &transcoded_buffer,
+                                           &transcoded_length);
+
+        if (ret != 0) {
+            flb_plg_error(ctx->ins, "could not transcode entries");
+            msgpack_sbuffer_destroy(&mp_sbuf);
+            return FLB_RETRY;
+        }
+    }
+
+    if (fc->compress == COMPRESS_GZIP) {
+        /* When compress is set, we switch from using Forward mode to using
+         * CompressedPackedForward mode.
+         */
+
+        if (transcoded_buffer != NULL) {
+            ret = flb_gzip_compress((void *) transcoded_buffer,
+                                    transcoded_length,
+                                    &final_data,
+                                    &final_bytes);
+        }
+        else {
+            ret = flb_gzip_compress((void *) data, bytes, &final_data, &final_bytes);
+        }
+
+        if (ret == -1) {
+            flb_plg_error(ctx->ins, "could not compress entries");
+            msgpack_sbuffer_destroy(&mp_sbuf);
+
+            if (transcoded_buffer != NULL) {
+                flb_free(transcoded_buffer);
+            }
+
+            return FLB_RETRY;
+        }
+
+        msgpack_pack_bin(&mp_pck, final_bytes);
+    }
+    else {
+        if (transcoded_buffer != NULL) {
+            final_data = (void *) transcoded_buffer;
+            final_bytes = transcoded_length;
+        }
+        else {
+            final_data = (void *) data;
+            final_bytes = bytes;
+        }
+
+        if (event_type == FLB_EVENT_TYPE_LOGS) {
+            /* for log events we create an array for the serialized messages */
+            entries = flb_mp_count(data, bytes);
+            msgpack_pack_array(&mp_pck, entries);
+        }
+        else {
+            /* FLB_EVENT_TYPE_METRICS and FLB_EVENT_TYPE_TRACES */
+            if (fc->fluentd_compat) {
+                pack_metricses_payload(&mp_pck, data, bytes);
+            }
+            else {
+                msgpack_pack_bin(&mp_pck, final_bytes);
+            }
+        }
+    }
+
+    /* Write message header */
+    ret = fc->io_write(u_conn, fc->unix_fd, mp_sbuf.data, mp_sbuf.size, &bytes_sent);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "could not write forward header");
+        msgpack_sbuffer_destroy(&mp_sbuf);
+        if (fc->compress == COMPRESS_GZIP) {
+            flb_free(final_data);
+        }
+
+        if (transcoded_buffer != NULL) {
+            flb_free(transcoded_buffer);
+        }
+
+        return FLB_RETRY;
+    }
+    msgpack_sbuffer_destroy(&mp_sbuf);
+
+    /* Write msgpack content / entries */
+    ret = fc->io_write(u_conn, fc->unix_fd, final_data, final_bytes, &bytes_sent);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "could not write forward entries");
+        if (fc->compress == COMPRESS_GZIP) {
+            flb_free(final_data);
+        }
+
+        if (transcoded_buffer != NULL) {
+            flb_free(transcoded_buffer);
+        }
+
+        return FLB_RETRY;
+    }
+
+    if (fc->compress == COMPRESS_GZIP) {
+        flb_free(final_data);
+    }
+
+    if (transcoded_buffer != NULL) {
+        flb_free(transcoded_buffer);
+    }
+
+    /* Write options */
+    if (send_options == FLB_TRUE) {
+        ret = fc->io_write(u_conn, fc->unix_fd, opts_buf, opts_size, &bytes_sent);
+        if (ret == -1) {
+            flb_plg_error(ctx->ins, "could not write forward options");
+            return FLB_RETRY;
+        }
+    }
+
+    /* If the sender requires 'ack' from the remote end-point */
+    if (fc->require_ack_response) {
+        msgpack_unpacked_init(&result);
+        ret = msgpack_unpack_next(&result, opts_buf, opts_size, &off);
+        if (ret != MSGPACK_UNPACK_SUCCESS) {
+            msgpack_unpacked_destroy(&result);
+            return -1;
+        }
+
+        /* Sucessful delivery, now get message 'chunk' and wait for it */
+        root = result.data;
+
+        /* 'chunk' is always in the first key of the map */
+        chunk = root.via.map.ptr[0].val;
+
+        /* Read ACK */
+        ret = forward_read_ack(ctx, fc, u_conn,
+                               (char *) chunk.via.str.ptr, chunk.via.str.size);
+        if (ret == -1) {
+            msgpack_unpacked_destroy(&result);
+            return FLB_RETRY;
+        }
+
+        /* All good */
+        msgpack_unpacked_destroy(&result);
+        return FLB_OK;
+    }
+
+    return FLB_OK;
+}
+
+/*
+ * Forward Mode Compat: data is packaged in Forward mode but the timestamps are
+ * integers (compat mode for Fluentd <= 0.12).
+ */
+static int flush_forward_compat_mode(struct flb_forward *ctx,
+                                     struct flb_forward_config *fc,
+                                     struct flb_connection *u_conn,
+                                     const char *tag, int tag_len,
+                                     const void *data, size_t bytes)
+{
+    int ret;
+    size_t off = 0;
+    size_t bytes_sent;
+    msgpack_object root;
+    msgpack_object chunk;
+    msgpack_object map; /* dummy parameter */
+    msgpack_unpacked result;
+
+    /* Write message header */
+    ret = fc->io_write(u_conn, fc->unix_fd, data, bytes, &bytes_sent);
+    if (ret == -1) {
+        flb_plg_error(ctx->ins, "could not write forward compat mode records");
+        return FLB_RETRY;
+    }
+
+    /* If the sender requires 'ack' from the remote end-point */
+    if (fc->require_ack_response) {
+        msgpack_unpacked_init(&result);
+        ret = msgpack_unpack_next(&result, data, bytes, &off);
+        if (ret != MSGPACK_UNPACK_SUCCESS) {
+            msgpack_unpacked_destroy(&result);
+            return -1;
+        }
+
+        /* Sucessful delivery, now get message 'chunk' and wait for it */
+        root = result.data;
+
+        map = root.via.array.ptr[2];
+
+        /* 'chunk' is always in the first key of the map */
+        chunk = map.via.map.ptr[0].val;
+
+        /* Read ACK */
+        ret = forward_read_ack(ctx, fc, u_conn,
+                               (char *) chunk.via.str.ptr, chunk.via.str.size);
+        if (ret == -1) {
+            msgpack_unpacked_destroy(&result);
+            return FLB_RETRY;
+        }
+
+        /* All good */
+        msgpack_unpacked_destroy(&result);
+        return FLB_OK;
+    }
+
+    return FLB_OK;
+}
+
+static void cb_forward_flush(struct flb_event_chunk *event_chunk,
+                             struct flb_output_flush *out_flush,
+                             struct flb_input_instance *i_ins,
+                             void *out_context,
+                             struct flb_config *config)
+{
+    int ret = -1;
+    int mode;
+    msgpack_packer   mp_pck;
+    msgpack_sbuffer  mp_sbuf;
+    void *out_buf = NULL;
+    size_t out_size = 0;
+    struct flb_forward *ctx = out_context;
+    struct flb_forward_config *fc = NULL;
+    struct flb_connection *u_conn = NULL;
+    struct flb_upstream_node *node = NULL;
+    struct flb_forward_flush *flush_ctx;
+    flb_sockfd_t uds_conn;
+
+    (void) i_ins;
+    (void) config;
+
+    fc = flb_forward_target(ctx, &node);
+    if (!fc) {
+        FLB_OUTPUT_RETURN(FLB_RETRY);
+    }
+
+    flb_plg_debug(ctx->ins, "request %lu bytes to flush",
+                  event_chunk->size);
+
+    /* Initialize packager */
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+
+    /*
+     * Flush context: structure used to pass custom information to the
+     * formatter function.
+     */
+    flush_ctx = flb_calloc(1, sizeof(struct flb_forward_flush));
+    if (!flush_ctx) {
+        flb_errno();
+        msgpack_sbuffer_destroy(&mp_sbuf);
+        FLB_OUTPUT_RETURN(FLB_RETRY);
+    }
+    flush_ctx->fc = fc;
+
+    /* Format the right payload and retrieve the 'forward mode' used */
+    mode = flb_forward_format(config, i_ins, ctx, flush_ctx,
+                              event_chunk->type,
+                              event_chunk->tag, flb_sds_len(event_chunk->tag),
+                              event_chunk->data, event_chunk->size,
+                              &out_buf, &out_size);
+
+    /* Get a TCP connection instance */
+    if (fc->unix_path == NULL) {
+        if (ctx->ha_mode == FLB_TRUE) {
+            u_conn = flb_upstream_conn_get(node->u);
+        }
+        else {
+            u_conn = flb_upstream_conn_get(ctx->u);
+        }
+
+        if (!u_conn) {
+            flb_plg_error(ctx->ins, "no upstream connections available");
+            msgpack_sbuffer_destroy(&mp_sbuf);
+            if (fc->time_as_integer == FLB_TRUE) {
+                flb_free(out_buf);
+            }
+            flb_free(flush_ctx);
+            FLB_OUTPUT_RETURN(FLB_RETRY);
+        }
+
+        uds_conn = -1;
+    }
+    else {
+        uds_conn = forward_uds_get_conn(fc, ctx);
+
+        if (uds_conn == -1) {
+            flb_plg_error(ctx->ins, "no unix socket connection available");
+
+            msgpack_sbuffer_destroy(&mp_sbuf);
+            if (fc->time_as_integer == FLB_TRUE) {
+                flb_free(out_buf);
+            }
+            flb_free(flush_ctx);
+            FLB_OUTPUT_RETURN(FLB_RETRY);
+        }
+
+        /* This is a hack, because the rest of the code is written to use
+         * the shared forward config unix_fd field so at this point we need
+         * to ensure that we either have a working connection or we can
+         * establish one regardless of not passing it along.
+         *
+         * Later on we will get the file descriptor from the TLS.
+        */
+    }
+
+    /*
+     * Shared Key: if ka_count > 0 it means the handshake has already been done lately
+     */
+    if (fc->shared_key && u_conn->ka_count == 0) {
+        ret = secure_forward_handshake(u_conn, fc, ctx);
+        flb_plg_debug(ctx->ins, "handshake status = %i", ret);
+        if (ret == -1) {
+            if (u_conn) {
+                flb_upstream_conn_release(u_conn);
+            }
+
+            if (uds_conn != -1) {
+                forward_uds_drop_conn(ctx, uds_conn);
+            }
+
+            msgpack_sbuffer_destroy(&mp_sbuf);
+            if (fc->time_as_integer == FLB_TRUE) {
+                flb_free(out_buf);
+            }
+            flb_free(flush_ctx);
+            FLB_OUTPUT_RETURN(FLB_RETRY);
+        }
+    }
+
+    /*
+     * Note about the mode used for different type of events/messages:
+     *
+     * - Logs can be send either by using MODE_MESSAGE, MODE_FORWARD
+     *   OR MODE_FORWARD_COMPAT.
+     *
+     * - Metrics and Traces uses MODE_FORWARD only.
+     */
+
+    if (mode == MODE_MESSAGE) {
+        ret = flush_message_mode(ctx, fc, u_conn, out_buf, out_size);
+        flb_free(out_buf);
+    }
+    else if (mode == MODE_FORWARD) {
+        ret = flush_forward_mode(ctx, fc, u_conn,
+                                 event_chunk->type,
+                                 event_chunk->tag, flb_sds_len(event_chunk->tag),
+                                 event_chunk->data, event_chunk->size,
+                                 out_buf, out_size);
+        flb_free(out_buf);
+    }
+    else if (mode == MODE_FORWARD_COMPAT) {
+        ret = flush_forward_compat_mode(ctx, fc, u_conn,
+                                        event_chunk->tag,
+                                        flb_sds_len(event_chunk->tag),
+                                        out_buf, out_size);
+        flb_free(out_buf);
+    }
+
+    if (u_conn) {
+        flb_upstream_conn_release(u_conn);
+    }
+
+    if (ret != FLB_OK) {
+        /* Since UDS connections have been used as permanent
+         * connections up to this point we only release the
+         * connection in case of error.
+         *
+         * There could be a logical error in here but what
+         * I think at the moment is, if something goes wrong
+         * we can just drop the connection and let the worker
+         * establish a new one the next time a flush happens.
+         */
+
+        if (uds_conn != -1) {
+            forward_uds_drop_conn(ctx, uds_conn);
+        }
+    }
+
+    flb_free(flush_ctx);
+    FLB_OUTPUT_RETURN(ret);
+}
+
+static int cb_forward_exit(void *data, struct flb_config *config)
+{
+    struct flb_forward *ctx = data;
+    struct flb_forward_config *fc;
+    struct mk_list *head;
+    struct mk_list *tmp;
+    (void) config;
+
+    if (!ctx) {
+        return 0;
+    }
+
+    /* Destroy forward_config contexts */
+    mk_list_foreach_safe(head, tmp, &ctx->configs) {
+        fc = mk_list_entry(head, struct flb_forward_config, _head);
+
+        mk_list_del(&fc->_head);
+        forward_config_destroy(fc);
+    }
+
+    forward_uds_drop_all(ctx);
+
+    if (ctx->ha_mode == FLB_TRUE) {
+        if (ctx->ha) {
+            flb_upstream_ha_destroy(ctx->ha);
+        }
+    }
+    else {
+        if (ctx->u) {
+            flb_upstream_destroy(ctx->u);
+        }
+    }
+
+    pthread_mutex_destroy(&ctx->uds_connection_list_mutex);
+
+    flb_free(ctx);
+
+    return 0;
+}
+
+static struct flb_config_map config_map[] = {
+    {
+     FLB_CONFIG_MAP_BOOL, "time_as_integer", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, time_as_integer),
+     "Set timestamp in integer format (compat mode for old Fluentd v0.12)"
+    },
+    {
+     FLB_CONFIG_MAP_BOOL, "retain_metadata_in_forward_mode", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, fwd_retain_metadata),
+     "Retain metadata when operating in forward mode"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "shared_key", NULL,
+     0, FLB_FALSE, 0,
+     "Shared key for authentication"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "self_hostname", NULL,
+     0, FLB_FALSE, 0,
+     "Hostname"
+    },
+    {
+     FLB_CONFIG_MAP_BOOL, "empty_shared_key", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, empty_shared_key),
+     "Set an empty shared key for authentication"
+    },
+    {
+     FLB_CONFIG_MAP_BOOL, "send_options", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, send_options),
+     "Send 'forward protocol options' to remote endpoint"
+    },
+    {
+     FLB_CONFIG_MAP_BOOL, "require_ack_response", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, require_ack_response),
+     "Require that remote endpoint confirms data reception"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "username", "",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, username),
+     "Username for authentication"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "password", "",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, password),
+     "Password for authentication"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "unix_path", NULL,
+     0, FLB_TRUE, offsetof(struct flb_forward_config, unix_path),
+     "Path to unix socket. It is ignored when 'upstream' property is set"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "upstream", NULL,
+     0, FLB_FALSE, 0,
+     "Path to 'upstream' configuration file (define multiple nodes)"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "tag", NULL,
+     0, FLB_FALSE, 0,
+     "Set a custom Tag for the outgoing records"
+    },
+    {
+     FLB_CONFIG_MAP_STR, "compress", NULL,
+     0, FLB_FALSE, 0,
+     "Compression mode"
+    },
+    {
+     FLB_CONFIG_MAP_BOOL, "fluentd_compat", "false",
+     0, FLB_TRUE, offsetof(struct flb_forward_config, fluentd_compat),
+     "Send metrics and traces with Fluentd compatible format"
+    },
+
+    {
+     FLB_CONFIG_MAP_SLIST_2, "add_option", NULL,
+     FLB_CONFIG_MAP_MULT, FLB_TRUE, offsetof(struct flb_forward_config, extra_options),
+     "Set an extra Forward protocol option. This is an advance feature, use it only for "
+     "very specific use-cases."
+    },
+
+    /* EOF */
+    {0}
+};
+
+/* Plugin reference */
+struct flb_output_plugin out_forward_plugin = {
+    .name         = "forward",
+    .description  = "Forward (Fluentd protocol)",
+
+    /* Callbacks */
+    .cb_init      = cb_forward_init,
+    .cb_pre_run   = NULL,
+    .cb_flush     = cb_forward_flush,
+    .cb_exit      = cb_forward_exit,
+    .workers      = 2,
+
+    /* Config map validator */
+    .config_map   = config_map,
+
+    /* Test */
+    .test_formatter.callback = flb_forward_format,
+
+    /* Flags */
+    .flags        = FLB_OUTPUT_NET | FLB_IO_OPT_TLS,
+
+    /* Event types */
+    .event_type   = FLB_OUTPUT_LOGS | FLB_OUTPUT_METRICS | FLB_OUTPUT_TRACES
+};
diff --git a/fluent-bit/plugins/out_forward/forward.h b/fluent-bit/plugins/out_forward/forward.h
new file mode 100644
index 000000000..8e77e6e11
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/forward.h
@@ -0,0 +1,146 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  Fluent Bit
+ *  ==========
+ *  Copyright (C) 2015-2022 The Fluent Bit Authors
+ *
+ *  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 FLB_OUT_FORWARD
+#define FLB_OUT_FORWARD
+
+#include <fluent-bit/flb_info.h>
+#include <fluent-bit/flb_sds.h>
+#include <fluent-bit/flb_upstream_ha.h>
+#include <fluent-bit/flb_record_accessor.h>
+#include <fluent-bit/flb_connection.h>
+#include <fluent-bit/flb_pthread.h>
+#include <cfl/cfl_list.h>
+
+/*
+ * Forward modes
+ * =============
+ */
+
+/*
+ * Message mode
+ * ------------
+ * https://github.com/fluent/fluentd/wiki/Forward-Protocol-Specification-v1#message-modes
+ */
+#define MODE_MESSAGE               0
+
+/*
+ * Forward mode
+ * ------------
+ * https://github.com/fluent/fluentd/wiki/Forward-Protocol-Specification-v1#forward-mode
+ */
+#define MODE_FORWARD               1
+
+/*
+ * Forward Compat: similar to MODE_FORWARD, but it sends the timestamps as unsigned
+ * integers for compatibility with very old versions of Fluentd that don't have timestamps
+ * with nanoseconds. This mode only applies for Logs.
+ */
+#define MODE_FORWARD_COMPAT        3
+
+/* Compression options */
+#define COMPRESS_NONE              0
+#define COMPRESS_GZIP              1
+
+/*
+ * Configuration: we put this separate from the main
+ * context so every Upstream Node can have it own configuration
+ * reference and pass it smoothly to the required caller.
+ *
+ * On simple mode (no HA), the structure is referenced
+ * by flb_forward->config. In HA mode the structure is referenced
+ * by the Upstream node context as an opaque data type.
+ */
+struct flb_forward_config {
+    int secured;              /* Using Secure Forward mode ?  */
+    int compress;             /* Using compression ? */
+    int time_as_integer;      /* Use backward compatible timestamp ? */
+    int fluentd_compat;       /* Use Fluentd compatible payload for
+                               * metrics and ctraces */
+
+    /* add extra options to the Forward payload (advanced) */
+    struct mk_list *extra_options;
+
+    int fwd_retain_metadata;  /* Do not drop metadata in forward mode */
+
+    /* config */
+    flb_sds_t shared_key;        /* shared key                   */
+    flb_sds_t self_hostname;     /* hostname used in certificate  */
+    flb_sds_t tag;               /* Overwrite tag on forward */
+    int empty_shared_key;        /* use an empty string as shared key */
+    int require_ack_response;    /* Require acknowledge for "chunk" */
+    int send_options;            /* send options in messages */
+    flb_sds_t unix_path;         /* unix socket path */
+    int       unix_fd;
+
+    const char *username;
+    const char *password;
+
+    /* mbedTLS specifics */
+    unsigned char shared_key_salt[16];
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    struct flb_record_accessor *ra_tag; /* Tag Record accessor */
+    int ra_static;                      /* Is the record accessor static ? */
+#endif
+    int (*io_write)(struct flb_connection* conn, int fd, const void* data,
+                        size_t len, size_t *out_len);
+    int (*io_read)(struct flb_connection* conn, int fd, void* buf, size_t len);
+    struct mk_list _head;     /* Link to list flb_forward->configs */
+};
+
+struct flb_forward_uds_connection {
+    flb_sockfd_t    descriptor;
+    struct cfl_list _head;     /* Link to list flb_forward->uds_connnection_list */
+};
+
+/* Plugin Context */
+struct flb_forward {
+    /* if HA mode is enabled */
+    int ha_mode;              /* High Availability mode enabled ? */
+    char *ha_upstream;        /* Upstream configuration file      */
+    struct flb_upstream_ha *ha;
+
+    struct cfl_list uds_connection_list;
+    pthread_mutex_t uds_connection_list_mutex;
+
+    /* Upstream handler and config context for single mode (no HA) */
+    struct flb_upstream *u;
+    struct mk_list configs;
+    struct flb_output_instance *ins;
+};
+
+struct flb_forward_ping {
+    const char *nonce;
+    int nonce_len;
+    const char *auth;
+    int auth_len;
+    int keepalive;
+};
+
+/* Flush callback context */
+struct flb_forward_flush {
+    struct flb_forward_config *fc;
+    char checksum_hex[33];
+};
+
+struct flb_forward_config *flb_forward_target(struct flb_forward *ctx,
+                                              struct flb_upstream_node **node);
+
+#endif
diff --git a/fluent-bit/plugins/out_forward/forward_format.c b/fluent-bit/plugins/out_forward/forward_format.c
new file mode 100644
index 000000000..48dedd862
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/forward_format.c
@@ -0,0 +1,640 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  Fluent Bit
+ *  ==========
+ *  Copyright (C) 2015-2022 The Fluent Bit Authors
+ *
+ *  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.
+ */
+
+#include <fluent-bit/flb_output_plugin.h>
+#include <fluent-bit/flb_time.h>
+#include <fluent-bit/flb_mp.h>
+#include <fluent-bit/flb_hash.h>
+#include <fluent-bit/flb_crypto.h>
+#include <fluent-bit/flb_record_accessor.h>
+#include <fluent-bit/flb_log_event_encoder.h>
+#include <fluent-bit/flb_log_event_decoder.h>
+
+#include "forward.h"
+
+void flb_forward_format_bin_to_hex(uint8_t *buf, size_t len, char *out)
+{
+    int i;
+    static char map[] = "0123456789abcdef";
+
+	for (i = 0; i < len; i++) {
+		out[i * 2]     = map[buf[i] >> 4];
+        out[i * 2 + 1] = map[buf[i] & 0x0f];
+	}
+}
+
+int flb_forward_format_append_tag(struct flb_forward *ctx,
+                                  struct flb_forward_config *fc,
+                                  msgpack_packer *mp_pck,
+                                  msgpack_object *map,
+                                  const char *tag, int tag_len)
+{
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    flb_sds_t tmp;
+    msgpack_object m;
+    
+    memset(&m, 0, sizeof(m));
+
+    if (!fc->ra_tag) {
+        msgpack_pack_str(mp_pck, tag_len);
+        msgpack_pack_str_body(mp_pck, tag, tag_len);
+        return 0;
+    }
+
+    if (map) {
+        m = *map;
+    }
+
+    /* Tag */
+    tmp = flb_ra_translate(fc->ra_tag, (char *) tag, tag_len, m, NULL);
+    if (!tmp) {
+        flb_plg_warn(ctx->ins, "Tag translation failed, using default Tag");
+        msgpack_pack_str(mp_pck, tag_len);
+        msgpack_pack_str_body(mp_pck, tag, tag_len);
+    }
+    else {
+        msgpack_pack_str(mp_pck, flb_sds_len(tmp));
+        msgpack_pack_str_body(mp_pck, tmp, flb_sds_len(tmp));
+        flb_sds_destroy(tmp);
+    }
+#else
+    msgpack_pack_str(mp_pck, tag_len);
+    msgpack_pack_str_body(mp_pck, tag, tag_len);
+
+#endif
+
+    return 0;
+}
+
+static int append_options(struct flb_forward *ctx,
+                          struct flb_forward_config *fc,
+                          int event_type,
+                          msgpack_packer *mp_pck,
+                          int entries, void *data, size_t bytes,
+                          msgpack_object *metadata,
+                          char *out_chunk)
+{
+    char *chunk = NULL;
+    uint8_t checksum[64];
+    int     result;
+    struct mk_list *head;
+    struct flb_config_map_val *mv;
+    struct flb_mp_map_header mh;
+    struct flb_slist_entry *eopt_key;
+    struct flb_slist_entry *eopt_val;
+
+    /* options is map, use the dynamic map type */
+    flb_mp_map_header_init(&mh, mp_pck);
+
+    if (fc->require_ack_response == FLB_TRUE) {
+        /*
+         * for ack we calculate  sha512 of context, take 16 bytes,
+         * make 32 byte hex string of it
+         */
+        result = flb_hash_simple(FLB_HASH_SHA512,
+                                 data, bytes,
+                                 checksum, sizeof(checksum));
+
+        if (result != FLB_CRYPTO_SUCCESS) {
+            return -1;
+        }
+
+        flb_forward_format_bin_to_hex(checksum, 16, out_chunk);
+
+        out_chunk[32] = '\0';
+        chunk = (char *) out_chunk;
+    }
+
+    /* "chunk": '<checksum-base-64>' */
+    if (chunk) {
+        flb_mp_map_header_append(&mh);
+        msgpack_pack_str(mp_pck, 5);
+        msgpack_pack_str_body(mp_pck, "chunk", 5);
+        msgpack_pack_str(mp_pck, 32);
+        msgpack_pack_str_body(mp_pck, out_chunk, 32);
+    }
+
+    /* "size": entries */
+    if (entries > 0) {
+        flb_mp_map_header_append(&mh);
+        msgpack_pack_str(mp_pck, 4);
+        msgpack_pack_str_body(mp_pck, "size", 4);
+        msgpack_pack_int64(mp_pck, entries);
+    }
+
+    /* "compressed": "gzip" */
+    if (entries > 0 &&                      /* not message mode */
+        fc->time_as_integer == FLB_FALSE && /* not compat mode */
+        fc->compress == COMPRESS_GZIP) {
+
+        flb_mp_map_header_append(&mh);
+        msgpack_pack_str(mp_pck, 10);
+        msgpack_pack_str_body(mp_pck, "compressed", 10);
+        msgpack_pack_str(mp_pck, 4);
+        msgpack_pack_str_body(mp_pck, "gzip", 4);
+    }
+
+    /* event type (FLB_EVENT_TYPE_LOGS, FLB_EVENT_TYPE_METRICS, FLB_EVENT_TYPE_TRACES) */
+    flb_mp_map_header_append(&mh);
+    msgpack_pack_str(mp_pck, 13);
+    msgpack_pack_str_body(mp_pck, "fluent_signal", 13);
+    msgpack_pack_int64(mp_pck, event_type);
+
+    /* process 'extra_option(s)' */
+    if (fc->extra_options) {
+        flb_config_map_foreach(head, mv, fc->extra_options) {
+            eopt_key = mk_list_entry_first(mv->val.list, struct flb_slist_entry, _head);
+            eopt_val = mk_list_entry_last(mv->val.list, struct flb_slist_entry, _head);
+
+            flb_mp_map_header_append(&mh);
+            msgpack_pack_str(mp_pck, flb_sds_len(eopt_key->str));
+            msgpack_pack_str_body(mp_pck, eopt_key->str, flb_sds_len(eopt_key->str));
+            msgpack_pack_str(mp_pck, flb_sds_len(eopt_val->str));
+            msgpack_pack_str_body(mp_pck, eopt_val->str, flb_sds_len(eopt_val->str));
+        }
+    }
+
+    if (metadata != NULL &&
+        metadata->type == MSGPACK_OBJECT_MAP &&
+        metadata->via.map.size > 0) {
+        flb_mp_map_header_append(&mh);
+        msgpack_pack_str_with_body(mp_pck, "metadata", 8);
+        msgpack_pack_object(mp_pck, *metadata);
+    }
+
+    flb_mp_map_header_end(&mh);
+
+    flb_plg_debug(ctx->ins,
+                  "send options records=%d chunk='%s'",
+                  entries, out_chunk ? out_chunk : "NULL");
+    return 0;
+}
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+/*
+ * Forward Protocol: Message Mode
+ * ------------------------------
+ * This mode is only used if the Tag is dynamically composed using some
+ * content of the records.
+ *
+ *  [
+ *    "TAG",
+ *    TIMESTAMP,
+ *    RECORD/MAP,
+ *    *OPTIONS*
+ *  ]
+ *
+ */
+static int flb_forward_format_message_mode(struct flb_forward *ctx,
+                                           struct flb_forward_config *fc,
+                                           struct flb_forward_flush *ff,
+                                           const char *tag, int tag_len,
+                                           const void *data, size_t bytes,
+                                           void **out_buf, size_t *out_size)
+{
+    int entries = 0;
+    size_t pre = 0;
+    size_t off = 0;
+    size_t record_size;
+    char *chunk;
+    char chunk_buf[33];
+    msgpack_packer   mp_pck;
+    msgpack_sbuffer  mp_sbuf;
+    struct flb_time tm;
+    struct flb_log_event_decoder log_decoder;
+    struct flb_log_event log_event;
+    int ret;
+
+    /*
+     * Our only reason to use Message Mode is because the user wants to generate
+     * dynamic Tags based on records content.
+     */
+    if (!fc->ra_tag) {
+        return -1;
+    }
+
+    /*
+     * if the case, we need to compose a new outgoing buffer instead
+     * of use the original one.
+     */
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+
+    ret = flb_log_event_decoder_init(&log_decoder, (char *) data, bytes);
+
+    if (ret != FLB_EVENT_DECODER_SUCCESS) {
+        flb_plg_error(ctx->ins,
+                      "Log event decoder initialization error : %d", ret);
+
+        return -1;
+    }
+
+    while ((ret = flb_log_event_decoder_next(
+                    &log_decoder,
+                    &log_event)) == FLB_EVENT_DECODER_SUCCESS) {
+        flb_time_copy(&tm, &log_event.timestamp);
+
+        /* Prepare main array: tag, timestamp and record/map */
+        msgpack_pack_array(&mp_pck, 4);
+
+        /* Generate dynamic Tag or use default one */
+        flb_forward_format_append_tag(ctx, fc, &mp_pck,
+                                      log_event.body,
+                                      tag, tag_len);
+
+        /* Pack timestamp */
+        if (fc->time_as_integer == FLB_TRUE) {
+            flb_time_append_to_msgpack(&log_event.timestamp,
+                                       &mp_pck,
+                                       FLB_TIME_ETFMT_INT);
+        }
+        else {
+            flb_time_append_to_msgpack(&log_event.timestamp,
+                                       &mp_pck,
+                                       FLB_TIME_ETFMT_V1_FIXEXT);
+        }
+
+        /* Pack records */
+        msgpack_pack_object(&mp_pck, *log_event.body);
+
+        record_size = off - pre;
+
+        if (ff) {
+            chunk = ff->checksum_hex;
+        }
+        else {
+            chunk = chunk_buf;
+        }
+
+        append_options(ctx, fc, FLB_EVENT_TYPE_LOGS, &mp_pck, 0,
+                       (char *) data + pre, record_size,
+                       log_event.metadata,
+                       chunk);
+
+        pre = off;
+        entries++;
+    }
+
+    flb_log_event_decoder_destroy(&log_decoder);
+
+    *out_buf  = mp_sbuf.data;
+    *out_size = mp_sbuf.size;
+
+    return entries;
+}
+#endif
+
+int flb_forward_format_transcode(
+        struct flb_forward *ctx, int format,
+        char *input_buffer, size_t input_length,
+        char **output_buffer, size_t *output_length)
+{
+    struct flb_log_event_encoder log_encoder;
+    struct flb_log_event_decoder log_decoder;
+    struct flb_log_event         log_event;
+    int                          result;
+
+    result = flb_log_event_decoder_init(&log_decoder, input_buffer, input_length);
+
+    if (result != FLB_EVENT_DECODER_SUCCESS) {
+        flb_plg_error(ctx->ins,
+                      "Log event decoder initialization error : %d", result);
+
+        return -1;
+    }
+
+    result = flb_log_event_encoder_init(&log_encoder, format);
+
+    if (result != FLB_EVENT_ENCODER_SUCCESS) {
+        flb_plg_error(ctx->ins,
+                      "Log event encoder initialization error : %d", result);
+
+        flb_log_event_decoder_destroy(&log_decoder);
+
+        return -1;
+    }
+
+    while ((result = flb_log_event_decoder_next(
+                        &log_decoder,
+                        &log_event)) == FLB_EVENT_DECODER_SUCCESS) {
+
+        result = flb_log_event_encoder_begin_record(&log_encoder);
+
+        if (result == FLB_EVENT_ENCODER_SUCCESS) {
+            result = flb_log_event_encoder_set_timestamp(
+                        &log_encoder, &log_event.timestamp);
+        }
+
+        if (result == FLB_EVENT_ENCODER_SUCCESS) {
+            result = flb_log_event_encoder_set_metadata_from_msgpack_object(
+                        &log_encoder,
+                        log_event.metadata);
+        }
+
+        if (result == FLB_EVENT_ENCODER_SUCCESS) {
+            result = flb_log_event_encoder_set_body_from_msgpack_object(
+                        &log_encoder,
+                        log_event.body);
+        }
+
+        if (result == FLB_EVENT_ENCODER_SUCCESS) {
+            result = flb_log_event_encoder_commit_record(&log_encoder);
+        }
+    }
+
+    if (log_encoder.output_length > 0) {
+        *output_buffer = log_encoder.output_buffer;
+        *output_length = log_encoder.output_length;
+
+        flb_log_event_encoder_claim_internal_buffer_ownership(&log_encoder);
+
+        result = 0;
+    }
+    else {
+        flb_plg_error(ctx->ins,
+                      "Log event encoder error : %d", result);
+
+        result = -1;
+    }
+
+    flb_log_event_decoder_destroy(&log_decoder);
+    flb_log_event_encoder_destroy(&log_encoder);
+
+    return result;
+}
+
+/*
+ * Forward Protocol: Forward Mode
+ * ------------------------------
+ * In forward mode we don't format the serialized entries. We just compose
+ * the outgoing 'options'.
+ */
+static int flb_forward_format_forward_mode(struct flb_forward *ctx,
+                                           struct flb_forward_config *fc,
+                                           struct flb_forward_flush *ff,
+                                           int event_type,
+                                           const char *tag, int tag_len,
+                                           const void *data, size_t bytes,
+                                           void **out_buf, size_t *out_size)
+{
+    int result;
+    int entries = 0;
+    char *chunk;
+    char chunk_buf[33];
+    msgpack_packer   mp_pck;
+    msgpack_sbuffer  mp_sbuf;
+    char *transcoded_buffer;
+    size_t transcoded_length;
+
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+
+    if (ff) {
+        chunk = ff->checksum_hex;
+    }
+    else {
+        chunk = chunk_buf;
+    }
+
+    if (fc->send_options == FLB_TRUE || (event_type == FLB_EVENT_TYPE_METRICS || event_type == FLB_EVENT_TYPE_TRACES)) {
+        if (event_type == FLB_EVENT_TYPE_LOGS) {
+            entries = flb_mp_count(data, bytes);
+        }
+        else {
+            /* for non logs, we don't count the number of entries */
+            entries = 0;
+        }
+
+        if (!fc->fwd_retain_metadata && event_type == FLB_EVENT_TYPE_LOGS) {
+            result = flb_forward_format_transcode(ctx, FLB_LOG_EVENT_FORMAT_FORWARD,
+                                                  (char *) data, bytes,
+                                                  &transcoded_buffer,
+                                                  &transcoded_length);
+
+            if (result == 0) {
+                append_options(ctx, fc, event_type, &mp_pck, entries,
+                               transcoded_buffer,
+                               transcoded_length,
+                               NULL, chunk);
+
+                free(transcoded_buffer);
+            }
+        }
+        else {
+            append_options(ctx, fc, event_type, &mp_pck, entries, (char *) data, bytes, NULL, chunk);
+        }
+    }
+
+    *out_buf  = mp_sbuf.data;
+    *out_size = mp_sbuf.size;
+
+    return 0;
+}
+
+/*
+ * Forward Protocol: Forward Mode Compat (for Fluentd <= 0.12)
+ * -----------------------------------------------------------
+ * Use Forward mode but format the timestamp as integers
+ *
+ * note: yes, the function name it's a big long...
+ */
+static int flb_forward_format_forward_compat_mode(struct flb_forward *ctx,
+                                                  struct flb_forward_config *fc,
+                                                  struct flb_forward_flush *ff,
+                                                  const char *tag, int tag_len,
+                                                  const void *data, size_t bytes,
+                                                  void **out_buf, size_t *out_size)
+{
+    int entries = 0;
+    char *chunk;
+    char chunk_buf[33];
+    msgpack_packer   mp_pck;
+    msgpack_sbuffer  mp_sbuf;
+    struct flb_log_event_decoder log_decoder;
+    struct flb_log_event log_event;
+    int ret;
+
+    ret = flb_log_event_decoder_init(&log_decoder, (char *) data, bytes);
+
+    if (ret != FLB_EVENT_DECODER_SUCCESS) {
+        flb_plg_error(ctx->ins,
+                      "Log event decoder initialization error : %d", ret);
+
+        return -1;
+    }
+
+    msgpack_sbuffer_init(&mp_sbuf);
+    msgpack_packer_init(&mp_pck, &mp_sbuf, msgpack_sbuffer_write);
+
+    if (ff) {
+        chunk = ff->checksum_hex;
+    }
+    else {
+        chunk = chunk_buf;
+    }
+
+    msgpack_pack_array(&mp_pck, fc->send_options ? 3 : 2);
+
+    /* Tag */
+    flb_forward_format_append_tag(ctx, fc, &mp_pck,
+                                  NULL, tag, tag_len);
+
+    /* Entries */
+    entries = flb_mp_count(data, bytes);
+    msgpack_pack_array(&mp_pck, entries);
+
+    while ((ret = flb_log_event_decoder_next(
+                    &log_decoder,
+                    &log_event)) == FLB_EVENT_DECODER_SUCCESS) {
+        msgpack_pack_array(&mp_pck, 2);
+
+        /* Pack timestamp */
+        if (fc->time_as_integer == FLB_TRUE) {
+            flb_time_append_to_msgpack(&log_event.timestamp,
+                                       &mp_pck,
+                                       FLB_TIME_ETFMT_INT);
+        }
+        else {
+            flb_time_append_to_msgpack(&log_event.timestamp,
+                                       &mp_pck,
+                                       FLB_TIME_ETFMT_V1_FIXEXT);
+        }
+
+        /* Pack records */
+        msgpack_pack_object(&mp_pck, *log_event.body);
+    }
+
+    if (fc->send_options == FLB_TRUE) {
+        append_options(ctx, fc, FLB_EVENT_TYPE_LOGS, &mp_pck, entries,
+                       (char *) data, bytes, NULL, chunk);
+    }
+
+    flb_log_event_decoder_destroy(&log_decoder);
+
+    *out_buf  = mp_sbuf.data;
+    *out_size = mp_sbuf.size;
+
+    return 0;
+}
+
+int flb_forward_format(struct flb_config *config,
+                       struct flb_input_instance *ins,
+                       void *ins_ctx,
+                       void *flush_ctx,
+                       int event_type,
+                       const char *tag, int tag_len,
+                       const void *data, size_t bytes,
+                       void **out_buf, size_t *out_size)
+{
+    int ret = 0;
+    int mode = MODE_FORWARD;
+    struct flb_upstream_node *node = NULL;
+    struct flb_forward_config *fc;
+    struct flb_forward_flush *ff = flush_ctx;
+    struct flb_forward *ctx = ins_ctx;
+
+    if (!flush_ctx) {
+        fc = flb_forward_target(ctx, &node);
+    }
+    else {
+        fc = ff->fc;
+    }
+
+    if (!fc) {
+        flb_plg_error(ctx->ins, "cannot get an Upstream single or HA node");
+        return -1;
+    }
+
+    if (event_type == FLB_EVENT_TYPE_METRICS) {
+        mode = MODE_FORWARD;
+        goto do_formatting;
+    }
+    else if (event_type == FLB_EVENT_TYPE_TRACES) {
+        mode = MODE_FORWARD;
+        goto do_formatting;
+    }
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    /*
+     * Based in the configuration, decide the preferred protocol mode
+     */
+    if (fc->ra_tag && fc->ra_static == FLB_FALSE) {
+        /*
+         * Dynamic tag per records needs to include the Tag for every entry,
+         * if record accessor option has been enabled we jump into this
+         * mode.
+         */
+        mode = MODE_MESSAGE;
+    }
+    else {
+#endif
+        /* Forward Modes */
+        if (fc->time_as_integer == FLB_FALSE) {
+            /*
+             * In forward mode we optimize in memory allocation and we reuse the
+             * original msgpack buffer. So we don't compose the outgoing buffer
+             * and just let the caller handle it.
+             */
+            mode = MODE_FORWARD;
+        }
+        else if (fc->time_as_integer == FLB_TRUE) {
+            /*
+             * This option is similar to MODE_FORWARD but since we have to convert the
+             * timestamp to integer type, we need to format the buffer (in the previous
+             * case we avoid that step.
+             */
+            mode = MODE_FORWARD_COMPAT;
+        }
+
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+    }
+#endif
+
+
+do_formatting:
+
+    /* Message Mode: the user needs custom Tags */
+    if (mode == MODE_MESSAGE) {
+#ifdef FLB_HAVE_RECORD_ACCESSOR
+        ret = flb_forward_format_message_mode(ctx, fc, ff,
+                                              tag, tag_len,
+                                              data, bytes,
+                                              out_buf, out_size);
+#endif
+    }
+    else if (mode == MODE_FORWARD) {
+        ret = flb_forward_format_forward_mode(ctx, fc, ff,
+                                              event_type,
+                                              tag, tag_len,
+                                              data, bytes,
+                                              out_buf, out_size);
+    }
+    else if (mode == MODE_FORWARD_COMPAT) {
+        ret = flb_forward_format_forward_compat_mode(ctx, fc, ff,
+                                                     tag, tag_len,
+                                                     data, bytes,
+                                                     out_buf, out_size);
+    }
+
+    if (ret == -1) {
+        return -1;
+    }
+
+    return mode;
+}
diff --git a/fluent-bit/plugins/out_forward/forward_format.h b/fluent-bit/plugins/out_forward/forward_format.h
new file mode 100644
index 000000000..bc6c47349
--- /dev/null
+++ b/fluent-bit/plugins/out_forward/forward_format.h
@@ -0,0 +1,48 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  Fluent Bit
+ *  ==========
+ *  Copyright (C) 2015-2022 The Fluent Bit Authors
+ *
+ *  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 FLB_OUT_FORWARD_FORMAT_H
+#define FLB_OUT_FORWARD_FORMAT_H
+
+#include <fluent-bit/flb_output_plugin.h>
+#include "forward.h"
+
+void flb_forward_format_bin_to_hex(uint8_t *buf, size_t len, char *out);
+
+int flb_forward_format_append_tag(struct flb_forward *ctx,
+                                  struct flb_forward_config *fc,
+                                  msgpack_packer *mp_pck,
+                                  msgpack_object *map,
+                                  const char *tag, int tag_len);
+
+int flb_forward_format(struct flb_config *config,
+                       struct flb_input_instance *ins,
+                       void *ins_ctx,
+                       void *flush_ctx,
+                       int event_type,
+                       const char *tag, int tag_len,
+                       const void *data, size_t bytes,
+                       void **out_buf, size_t *out_size);
+
+int flb_forward_format_transcode(
+        struct flb_forward *ctx, int format,
+        char *input_buffer, size_t input_length,
+        char **output_buffer, size_t *output_length);
+
+#endif