summaryrefslogtreecommitdiffstats
path: root/zebra/zebra_mlag.c
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--zebra/zebra_mlag.c1213
1 files changed, 1213 insertions, 0 deletions
diff --git a/zebra/zebra_mlag.c b/zebra/zebra_mlag.c
new file mode 100644
index 0000000..2042f94
--- /dev/null
+++ b/zebra/zebra_mlag.c
@@ -0,0 +1,1213 @@
+/* Zebra Mlag Code.
+ * Copyright (C) 2018 Cumulus Networks, Inc.
+ * Donald Sharp
+ *
+ * This file is part of FRR.
+ *
+ * FRR is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the
+ * Free Software Foundation; either version 2, or (at your option) any
+ * later version.
+ *
+ * FRR is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with FRR; see the file COPYING. If not, write to the Free
+ * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
+ * 02111-1307, USA.
+ */
+#include "zebra.h"
+
+#include "command.h"
+#include "hook.h"
+#include "frr_pthread.h"
+#include "mlag.h"
+
+#include "zebra/zebra_mlag.h"
+#include "zebra/zebra_mlag_vty.h"
+#include "zebra/zebra_router.h"
+#include "zebra/zapi_msg.h"
+#include "zebra/debug.h"
+
+#ifdef HAVE_PROTOBUF_VERSION_3
+#include "mlag/mlag.pb-c.h"
+#endif
+
+DEFINE_HOOK(zebra_mlag_private_write_data,
+ (uint8_t *data, uint32_t len), (data, len));
+DEFINE_HOOK(zebra_mlag_private_monitor_state, (), ());
+DEFINE_HOOK(zebra_mlag_private_open_channel, (), ());
+DEFINE_HOOK(zebra_mlag_private_close_channel, (), ());
+DEFINE_HOOK(zebra_mlag_private_cleanup_data, (), ());
+
+#define ZEBRA_MLAG_METADATA_LEN 4
+#define ZEBRA_MLAG_MSG_BCAST 0xFFFFFFFF
+
+uint8_t mlag_wr_buffer[ZEBRA_MLAG_BUF_LIMIT];
+uint8_t mlag_rd_buffer[ZEBRA_MLAG_BUF_LIMIT];
+
+static bool test_mlag_in_progress;
+
+static int zebra_mlag_signal_write_thread(void);
+static void zebra_mlag_terminate_pthread(struct thread *event);
+static void zebra_mlag_post_data_from_main_thread(struct thread *thread);
+static void zebra_mlag_publish_process_state(struct zserv *client,
+ zebra_message_types_t msg_type);
+
+/**********************MLAG Interaction***************************************/
+
+/*
+ * API to post the Registration to MLAGD
+ * MLAG will not process any messages with out the registration
+ */
+void zebra_mlag_send_register(void)
+{
+ struct stream *s = NULL;
+
+ s = stream_new(sizeof(struct mlag_msg));
+
+ stream_putl(s, MLAG_REGISTER);
+ stream_putw(s, MLAG_MSG_NULL_PAYLOAD);
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ stream_fifo_push_safe(zrouter.mlag_info.mlag_fifo, s);
+ zebra_mlag_signal_write_thread();
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Enqueued MLAG Register to MLAG Thread ",
+ __func__);
+}
+
+/*
+ * API to post the De-Registration to MLAGD
+ * MLAG will not process any messages after the de-registration
+ */
+void zebra_mlag_send_deregister(void)
+{
+ struct stream *s = NULL;
+
+ s = stream_new(sizeof(struct mlag_msg));
+
+ stream_putl(s, MLAG_DEREGISTER);
+ stream_putw(s, MLAG_MSG_NULL_PAYLOAD);
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ stream_fifo_push_safe(zrouter.mlag_info.mlag_fifo, s);
+ zebra_mlag_signal_write_thread();
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Enqueued MLAG De-Register to MLAG Thread ",
+ __func__);
+}
+
+/*
+ * API To handle MLAG Received data
+ * Decodes the data using protobuf and enqueue to main thread
+ * main thread publish this to clients based on client subscription
+ */
+void zebra_mlag_process_mlag_data(uint8_t *data, uint32_t len)
+{
+ struct stream *s = NULL;
+ int msg_type = 0;
+
+ s = stream_new(ZEBRA_MLAG_BUF_LIMIT);
+ /*
+ * Place holder we need the message type first
+ */
+ stream_putl(s, msg_type);
+ msg_type = zebra_mlag_protobuf_decode_message(s, data, len);
+
+ if (msg_type <= 0) {
+ /* Something went wrong in decoding */
+ stream_free(s);
+ zlog_err("%s: failed to process mlag data-%d, %u", __func__,
+ msg_type, len);
+ return;
+ }
+
+ /*
+ * additional four bytes are for message type
+ */
+ stream_putl_at(s, 0, msg_type);
+ thread_add_event(zrouter.master, zebra_mlag_post_data_from_main_thread,
+ s, 0, NULL);
+}
+
+/**********************End of MLAG Interaction********************************/
+
+/************************MLAG Thread Processing*******************************/
+
+/*
+ * after posting every 'ZEBRA_MLAG_POST_LIMIT' packets, MLAG Thread will be
+ * yielded to give CPU for other threads
+ */
+#define ZEBRA_MLAG_POST_LIMIT 100
+
+/*
+ * This thread reads the clients data from the Global queue and encodes with
+ * protobuf and pass on to the MLAG socket.
+ */
+static void zebra_mlag_client_msg_handler(struct thread *event)
+{
+ struct stream *s;
+ uint32_t wr_count = 0;
+ uint32_t msg_type = 0;
+ uint32_t max_count = 0;
+ int len = 0;
+
+ wr_count = stream_fifo_count_safe(zrouter.mlag_info.mlag_fifo);
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(":%s: Processing MLAG write, %u messages in queue",
+ __func__, wr_count);
+
+ max_count = MIN(wr_count, ZEBRA_MLAG_POST_LIMIT);
+
+ for (wr_count = 0; wr_count < max_count; wr_count++) {
+ s = stream_fifo_pop_safe(zrouter.mlag_info.mlag_fifo);
+ if (!s) {
+ zlog_debug(":%s: Got a NULL Messages, some thing wrong",
+ __func__);
+ break;
+ }
+
+ /*
+ * Encode the data now
+ */
+ len = zebra_mlag_protobuf_encode_client_data(s, &msg_type);
+
+ /*
+ * write to MCLAGD
+ */
+ if (len > 0) {
+ hook_call(zebra_mlag_private_write_data,
+ mlag_wr_buffer, len);
+
+ /*
+ * If message type is De-register, send a signal to main
+ * thread, so that necessary cleanup will be done by
+ * main thread.
+ */
+ if (msg_type == MLAG_DEREGISTER) {
+ thread_add_event(zrouter.master,
+ zebra_mlag_terminate_pthread,
+ NULL, 0, NULL);
+ }
+ }
+
+ stream_free(s);
+ }
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(":%s: Posted %d messages to MLAGD", __func__,
+ wr_count);
+ /*
+ * Currently there is only message write task is enqueued to this
+ * thread, yielding was added for future purpose, so that this thread
+ * can server other tasks also and in case FIFO is empty, this task will
+ * be schedule when main thread adds some messages
+ */
+ if (wr_count >= ZEBRA_MLAG_POST_LIMIT)
+ zebra_mlag_signal_write_thread();
+}
+
+/*
+ * API to handle the process state.
+ * In case of Down, Zebra keep monitoring the MLAG state.
+ * all the state Notifications will be published to clients
+ */
+void zebra_mlag_handle_process_state(enum zebra_mlag_state state)
+{
+ if (state == MLAG_UP) {
+ zrouter.mlag_info.connected = true;
+ zebra_mlag_publish_process_state(NULL, ZEBRA_MLAG_PROCESS_UP);
+ zebra_mlag_send_register();
+ } else if (state == MLAG_DOWN) {
+ zrouter.mlag_info.connected = false;
+ zebra_mlag_publish_process_state(NULL, ZEBRA_MLAG_PROCESS_DOWN);
+ hook_call(zebra_mlag_private_monitor_state);
+ }
+}
+
+/***********************End of MLAG Thread processing*************************/
+
+/*************************Multi-entratnt Api's********************************/
+
+/*
+ * Provider api to signal that work/events are available
+ * for the Zebra MLAG Write pthread.
+ * This API is called from 2 pthreads..
+ * 1) by main thread when client posts a MLAG Message
+ * 2) by MLAG Thread, in case of yield
+ * though this api, is called from two threads we don't need any locking
+ * because Thread task enqueue is thread safe means internally it had
+ * necessary protection
+ */
+static int zebra_mlag_signal_write_thread(void)
+{
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(":%s: Scheduling MLAG write", __func__);
+ /*
+ * This api will be called from Both main & MLAG Threads.
+ * main thread writes, "zrouter.mlag_info.th_master" only
+ * during Zebra Init/after MLAG thread is destroyed.
+ * so it is safe to use without any locking
+ */
+ thread_add_event(zrouter.mlag_info.th_master,
+ zebra_mlag_client_msg_handler, NULL, 0,
+ &zrouter.mlag_info.t_write);
+ return 0;
+}
+
+/*
+ * API will be used to publish the MLAG state to interested clients
+ * In case client is passed, state is posted only for that client,
+ * otherwise to all interested clients
+ * this api can be called from two threads.
+ * 1) from main thread: when client is passed
+ * 2) from MLAG Thread: when client is NULL
+ *
+ * In second case, to avoid global data access data will be post to Main
+ * thread, so that actual posting to clients will happen from Main thread.
+ */
+static void zebra_mlag_publish_process_state(struct zserv *client,
+ zebra_message_types_t msg_type)
+{
+ struct stream *s;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Publishing MLAG process state:%s to %s Client",
+ __func__,
+ (msg_type == ZEBRA_MLAG_PROCESS_UP) ? "UP" : "DOWN",
+ (client) ? "one" : "all");
+
+ if (client) {
+ s = stream_new(ZEBRA_HEADER_SIZE);
+ zclient_create_header(s, msg_type, VRF_DEFAULT);
+ zserv_send_message(client, s);
+ return;
+ }
+
+
+ /*
+ * additional four bytes are for mesasge type
+ */
+ s = stream_new(ZEBRA_HEADER_SIZE + ZEBRA_MLAG_METADATA_LEN);
+ stream_putl(s, ZEBRA_MLAG_MSG_BCAST);
+ zclient_create_header(s, msg_type, VRF_DEFAULT);
+ thread_add_event(zrouter.master, zebra_mlag_post_data_from_main_thread,
+ s, 0, NULL);
+}
+
+/**************************End of Multi-entrant Apis**************************/
+
+/***********************Zebra Main thread processing**************************/
+
+/*
+ * To avoid data corruption, messages will be post to clients only from
+ * main thread, because for that access was needed for clients list.
+ * so instead of forcing the locks, messages will be posted from main thread.
+ */
+static void zebra_mlag_post_data_from_main_thread(struct thread *thread)
+{
+ struct stream *s = THREAD_ARG(thread);
+ struct stream *zebra_s = NULL;
+ struct listnode *node;
+ struct zserv *client;
+ uint32_t msg_type = 0;
+ uint32_t msg_len = 0;
+
+ if (!s)
+ return;
+
+ STREAM_GETL(s, msg_type);
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "%s: Posting MLAG data for msg_type:0x%x to interested clients",
+ __func__, msg_type);
+
+ msg_len = s->endp - ZEBRA_MLAG_METADATA_LEN;
+ for (ALL_LIST_ELEMENTS_RO(zrouter.client_list, node, client)) {
+ if (client->mlag_updates_interested == true) {
+ if (msg_type != ZEBRA_MLAG_MSG_BCAST
+ && !CHECK_FLAG(client->mlag_reg_mask1,
+ (1 << msg_type))) {
+ continue;
+ }
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "%s: Posting MLAG data of length-%d to client:%d ",
+ __func__, msg_len, client->proto);
+
+ zebra_s = stream_new(msg_len);
+ STREAM_GET(zebra_s->data, s, msg_len);
+ zebra_s->endp = msg_len;
+ stream_putw_at(zebra_s, 0, msg_len);
+
+ /*
+ * This stream will be enqueued to client_obuf, it will
+ * be freed after posting to client socket.
+ */
+ zserv_send_message(client, zebra_s);
+ zebra_s = NULL;
+ }
+ }
+
+ stream_free(s);
+ return;
+stream_failure:
+ stream_free(s);
+ if (zebra_s)
+ stream_free(zebra_s);
+}
+
+/*
+ * Start the MLAG Thread, this will be used to write client data on to
+ * MLAG Process and to read the data from MLAG and post to clients.
+ * when all clients are un-registered, this Thread will be
+ * suspended.
+ */
+static void zebra_mlag_spawn_pthread(void)
+{
+ /* Start MLAG write pthread */
+
+ struct frr_pthread_attr pattr = {.start =
+ frr_pthread_attr_default.start,
+ .stop = frr_pthread_attr_default.stop};
+
+ zrouter.mlag_info.zebra_pth_mlag =
+ frr_pthread_new(&pattr, "Zebra MLAG thread", "Zebra MLAG");
+
+ zrouter.mlag_info.th_master = zrouter.mlag_info.zebra_pth_mlag->master;
+
+
+ /* Enqueue an initial event to the Newly spawn MLAG pthread */
+ zebra_mlag_signal_write_thread();
+
+ frr_pthread_run(zrouter.mlag_info.zebra_pth_mlag, NULL);
+}
+
+/*
+ * all clients are un-registered for MLAG Updates, terminate the
+ * MLAG write thread
+ */
+static void zebra_mlag_terminate_pthread(struct thread *event)
+{
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Zebra MLAG write thread terminate called");
+
+ if (zrouter.mlag_info.clients_interested_cnt) {
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "Zebra MLAG: still some clients are interested");
+ return;
+ }
+
+ frr_pthread_stop(zrouter.mlag_info.zebra_pth_mlag, NULL);
+
+ /* Destroy pthread */
+ frr_pthread_destroy(zrouter.mlag_info.zebra_pth_mlag);
+ zrouter.mlag_info.zebra_pth_mlag = NULL;
+ zrouter.mlag_info.th_master = NULL;
+ zrouter.mlag_info.t_read = NULL;
+ zrouter.mlag_info.t_write = NULL;
+
+ /*
+ * Send Notification to clean private data
+ */
+ hook_call(zebra_mlag_private_cleanup_data);
+}
+
+/*
+ * API to register zebra client for MLAG Updates
+ */
+void zebra_mlag_client_register(ZAPI_HANDLER_ARGS)
+{
+ struct stream *s;
+ uint32_t reg_mask = 0;
+ int rc = 0;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Received MLAG Registration from client-proto:%d",
+ client->proto);
+
+
+ /* Get input stream. */
+ s = msg;
+
+ /* Get data. */
+ STREAM_GETL(s, reg_mask);
+
+ if (client->mlag_updates_interested == true) {
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "Client is registered, existing mask: 0x%x, new mask: 0x%x",
+ client->mlag_reg_mask1, reg_mask);
+ if (client->mlag_reg_mask1 != reg_mask)
+ client->mlag_reg_mask1 = reg_mask;
+ /*
+ * Client might missed MLAG-UP Notification, post-it again
+ */
+ zebra_mlag_publish_process_state(client, ZEBRA_MLAG_PROCESS_UP);
+ return;
+ }
+
+
+ client->mlag_updates_interested = true;
+ client->mlag_reg_mask1 = reg_mask;
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Registering for MLAG Updates with mask: 0x%x, ",
+ client->mlag_reg_mask1);
+
+ zrouter.mlag_info.clients_interested_cnt++;
+
+ if (zrouter.mlag_info.clients_interested_cnt == 1) {
+ /*
+ * First-client for MLAG Updates,open the communication channel
+ * with MLAG
+ */
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "First client, opening the channel with MLAG");
+
+ zebra_mlag_spawn_pthread();
+ rc = hook_call(zebra_mlag_private_open_channel);
+ if (rc < 0) {
+ /*
+ * For some reason, zebra not able to open the
+ * comm-channel with MLAG, so post MLAG-DOWN to client.
+ * later when the channel is open, zebra will send
+ * MLAG-UP
+ */
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "Fail to open channel with MLAG,rc:%d, post Proto-down",
+ rc);
+ zebra_mlag_publish_process_state(
+ client, ZEBRA_MLAG_PROCESS_DOWN);
+ }
+ }
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Client Registered successfully for MLAG Updates");
+
+ if (zrouter.mlag_info.connected == true)
+ zebra_mlag_publish_process_state(client, ZEBRA_MLAG_PROCESS_UP);
+stream_failure:
+ return;
+}
+
+/*
+ * API to un-register for MLAG Updates
+ */
+void zebra_mlag_client_unregister(ZAPI_HANDLER_ARGS)
+{
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Received MLAG De-Registration from client-proto:%d",
+ client->proto);
+
+ if (client->mlag_updates_interested == false)
+ /* Unexpected */
+ return;
+
+ client->mlag_updates_interested = false;
+ client->mlag_reg_mask1 = 0;
+ zrouter.mlag_info.clients_interested_cnt--;
+
+ if (zrouter.mlag_info.clients_interested_cnt == 0) {
+ /*
+ * No-client is interested for MLAG Updates,close the
+ * communication channel with MLAG
+ */
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Last client for MLAG, close the channel ");
+
+ /*
+ * Clean up flow:
+ * =============
+ * 1) main thread calls socket close which posts De-register
+ * to MLAG write thread
+ * 2) after MLAG write thread posts De-register it sends a
+ * signal back to main thread to do the thread cleanup
+ * this was mainly to make sure De-register is posted to MCLAGD.
+ */
+ hook_call(zebra_mlag_private_close_channel);
+ }
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "Client De-Registered successfully for MLAG Updates");
+}
+
+/*
+ * Does following things.
+ * 1) allocated new local stream, and copies the client data and enqueue
+ * to MLAG Thread
+ * 2) MLAG Thread after dequeing, encode the client data using protobuf
+ * and write on to MLAG
+ */
+void zebra_mlag_forward_client_msg(ZAPI_HANDLER_ARGS)
+{
+ struct stream *zebra_s;
+ struct stream *mlag_s;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Received Client MLAG Data from client-proto:%d",
+ client->proto);
+
+ /* Get input stream. */
+ zebra_s = msg;
+ mlag_s = stream_new(zebra_s->endp);
+
+ /*
+ * Client data is | Zebra Header + MLAG Data |
+ * we need to enqueue only the MLAG data, skipping Zebra Header
+ */
+ stream_put(mlag_s, zebra_s->data + zebra_s->getp,
+ STREAM_READABLE(zebra_s));
+ stream_fifo_push_safe(zrouter.mlag_info.mlag_fifo, mlag_s);
+ zebra_mlag_signal_write_thread();
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Enqueued Client:%d data to MLAG Thread ",
+ __func__, client->proto);
+}
+
+/***********************End of Zebra Main thread processing*************/
+
+enum mlag_role zebra_mlag_get_role(void)
+{
+ return zrouter.mlag_info.role;
+}
+
+int32_t zebra_mlag_test_mlag_internal(const char *none, const char *primary,
+ const char *secondary)
+{
+ enum mlag_role orig = zrouter.mlag_info.role;
+ char buf1[MLAG_ROLE_STRSIZE], buf2[MLAG_ROLE_STRSIZE];
+
+ if (none)
+ zrouter.mlag_info.role = MLAG_ROLE_NONE;
+ if (primary)
+ zrouter.mlag_info.role = MLAG_ROLE_PRIMARY;
+ if (secondary)
+ zrouter.mlag_info.role = MLAG_ROLE_SECONDARY;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("Test: Changing role from %s to %s",
+ mlag_role2str(orig, buf1, sizeof(buf1)),
+ mlag_role2str(orig, buf2, sizeof(buf2)));
+
+ if (orig != zrouter.mlag_info.role) {
+ zsend_capabilities_all_clients();
+ if (zrouter.mlag_info.role != MLAG_ROLE_NONE) {
+ if (zrouter.mlag_info.clients_interested_cnt == 0
+ && !test_mlag_in_progress) {
+ if (zrouter.mlag_info.zebra_pth_mlag == NULL)
+ zebra_mlag_spawn_pthread();
+ zrouter.mlag_info.clients_interested_cnt++;
+ test_mlag_in_progress = true;
+ hook_call(zebra_mlag_private_open_channel);
+ }
+ } else {
+ if (test_mlag_in_progress) {
+ test_mlag_in_progress = false;
+ zrouter.mlag_info.clients_interested_cnt--;
+ hook_call(zebra_mlag_private_close_channel);
+ }
+ }
+ }
+
+ return CMD_SUCCESS;
+}
+
+void zebra_mlag_init(void)
+{
+ zebra_mlag_vty_init();
+
+ /*
+ * Intialiaze the MLAG Global variables
+ * write thread will be created during actual registration with MCLAG
+ */
+ zrouter.mlag_info.clients_interested_cnt = 0;
+ zrouter.mlag_info.connected = false;
+ zrouter.mlag_info.timer_running = false;
+ zrouter.mlag_info.mlag_fifo = stream_fifo_new();
+ zrouter.mlag_info.zebra_pth_mlag = NULL;
+ zrouter.mlag_info.th_master = NULL;
+ zrouter.mlag_info.t_read = NULL;
+ zrouter.mlag_info.t_write = NULL;
+ test_mlag_in_progress = false;
+ zebra_mlag_reset_read_buffer();
+}
+
+void zebra_mlag_terminate(void)
+{
+}
+
+
+/*
+ *
+ * ProtoBuf Encoding APIs
+ */
+
+#ifdef HAVE_PROTOBUF_VERSION_3
+
+DEFINE_MTYPE_STATIC(ZEBRA, MLAG_PBUF, "ZEBRA MLAG PROTOBUF");
+
+int zebra_mlag_protobuf_encode_client_data(struct stream *s, uint32_t *msg_type)
+{
+ ZebraMlagHeader hdr = ZEBRA_MLAG__HEADER__INIT;
+ struct mlag_msg mlag_msg;
+ uint8_t tmp_buf[ZEBRA_MLAG_BUF_LIMIT];
+ int len = 0;
+ int n_len = 0;
+ int rc = 0;
+ char buf[ZLOG_FILTER_LENGTH_MAX];
+ size_t length;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Entering..", __func__);
+
+ rc = mlag_lib_decode_mlag_hdr(s, &mlag_msg, &length);
+ if (rc)
+ return rc;
+
+ memset(tmp_buf, 0, ZEBRA_MLAG_BUF_LIMIT);
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Mlag ProtoBuf encoding of message:%s, len:%d",
+ __func__,
+ mlag_lib_msgid_to_str(mlag_msg.msg_type, buf,
+ sizeof(buf)),
+ mlag_msg.data_len);
+ *msg_type = mlag_msg.msg_type;
+ switch (mlag_msg.msg_type) {
+ case MLAG_MROUTE_ADD: {
+ struct mlag_mroute_add msg;
+ ZebraMlagMrouteAdd pay_load = ZEBRA_MLAG_MROUTE_ADD__INIT;
+ uint32_t vrf_name_len = 0;
+
+ rc = mlag_lib_decode_mroute_add(s, &msg, &length);
+ if (rc)
+ return rc;
+
+ vrf_name_len = strlen(msg.vrf_name) + 1;
+ pay_load.vrf_name = XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+ strlcpy(pay_load.vrf_name, msg.vrf_name, vrf_name_len);
+ pay_load.source_ip = msg.source_ip;
+ pay_load.group_ip = msg.group_ip;
+ pay_load.cost_to_rp = msg.cost_to_rp;
+ pay_load.owner_id = msg.owner_id;
+ pay_load.am_i_dr = msg.am_i_dr;
+ pay_load.am_i_dual_active = msg.am_i_dual_active;
+ pay_load.vrf_id = msg.vrf_id;
+
+ if (msg.owner_id == MLAG_OWNER_INTERFACE) {
+ vrf_name_len = strlen(msg.intf_name) + 1;
+ pay_load.intf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+ strlcpy(pay_load.intf_name, msg.intf_name,
+ vrf_name_len);
+ }
+
+ len = zebra_mlag_mroute_add__pack(&pay_load, tmp_buf);
+ XFREE(MTYPE_MLAG_PBUF, pay_load.vrf_name);
+ if (msg.owner_id == MLAG_OWNER_INTERFACE)
+ XFREE(MTYPE_MLAG_PBUF, pay_load.intf_name);
+ } break;
+ case MLAG_MROUTE_DEL: {
+ struct mlag_mroute_del msg;
+ ZebraMlagMrouteDel pay_load = ZEBRA_MLAG_MROUTE_DEL__INIT;
+ uint32_t vrf_name_len = 0;
+
+ rc = mlag_lib_decode_mroute_del(s, &msg, &length);
+ if (rc)
+ return rc;
+ vrf_name_len = strlen(msg.vrf_name) + 1;
+ pay_load.vrf_name = XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+ strlcpy(pay_load.vrf_name, msg.vrf_name, vrf_name_len);
+ pay_load.source_ip = msg.source_ip;
+ pay_load.group_ip = msg.group_ip;
+ pay_load.owner_id = msg.owner_id;
+ pay_load.vrf_id = msg.vrf_id;
+
+ if (msg.owner_id == MLAG_OWNER_INTERFACE) {
+ vrf_name_len = strlen(msg.intf_name) + 1;
+ pay_load.intf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+ strlcpy(pay_load.intf_name, msg.intf_name,
+ vrf_name_len);
+ }
+
+ len = zebra_mlag_mroute_del__pack(&pay_load, tmp_buf);
+ XFREE(MTYPE_MLAG_PBUF, pay_load.vrf_name);
+ if (msg.owner_id == MLAG_OWNER_INTERFACE)
+ XFREE(MTYPE_MLAG_PBUF, pay_load.intf_name);
+ } break;
+ case MLAG_MROUTE_ADD_BULK: {
+ struct mlag_mroute_add msg;
+ ZebraMlagMrouteAddBulk Bulk_msg =
+ ZEBRA_MLAG_MROUTE_ADD_BULK__INIT;
+ ZebraMlagMrouteAdd **pay_load = NULL;
+ bool cleanup = false;
+ uint32_t i, actual;
+
+ Bulk_msg.n_mroute_add = mlag_msg.msg_cnt;
+ pay_load = XMALLOC(MTYPE_MLAG_PBUF, sizeof(ZebraMlagMrouteAdd *)
+ * mlag_msg.msg_cnt);
+
+ for (i = 0, actual = 0; i < mlag_msg.msg_cnt; i++, actual++) {
+
+ uint32_t vrf_name_len = 0;
+
+ rc = mlag_lib_decode_mroute_add(s, &msg, &length);
+ if (rc) {
+ cleanup = true;
+ break;
+ }
+ pay_load[i] = XMALLOC(MTYPE_MLAG_PBUF,
+ sizeof(ZebraMlagMrouteAdd));
+ zebra_mlag_mroute_add__init(pay_load[i]);
+
+ vrf_name_len = strlen(msg.vrf_name) + 1;
+ pay_load[i]->vrf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+ strlcpy(pay_load[i]->vrf_name, msg.vrf_name,
+ vrf_name_len);
+ pay_load[i]->source_ip = msg.source_ip;
+ pay_load[i]->group_ip = msg.group_ip;
+ pay_load[i]->cost_to_rp = msg.cost_to_rp;
+ pay_load[i]->owner_id = msg.owner_id;
+ pay_load[i]->am_i_dr = msg.am_i_dr;
+ pay_load[i]->am_i_dual_active = msg.am_i_dual_active;
+ pay_load[i]->vrf_id = msg.vrf_id;
+ if (msg.owner_id == MLAG_OWNER_INTERFACE) {
+ vrf_name_len = strlen(msg.intf_name) + 1;
+ pay_load[i]->intf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+
+ strlcpy(pay_load[i]->intf_name, msg.intf_name,
+ vrf_name_len);
+ }
+ }
+ if (!cleanup) {
+ Bulk_msg.mroute_add = pay_load;
+ len = zebra_mlag_mroute_add_bulk__pack(&Bulk_msg,
+ tmp_buf);
+ }
+
+ for (i = 0; i < actual; i++) {
+ /*
+ * The mlag_lib_decode_mroute_add can
+ * fail to properly decode and cause nothing
+ * to be allocated. Prevent a crash
+ */
+ if (!pay_load[i])
+ continue;
+
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]->vrf_name);
+ if (pay_load[i]->owner_id == MLAG_OWNER_INTERFACE
+ && pay_load[i]->intf_name)
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]->intf_name);
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]);
+ }
+ XFREE(MTYPE_MLAG_PBUF, pay_load);
+ if (cleanup)
+ return -1;
+ } break;
+ case MLAG_MROUTE_DEL_BULK: {
+ struct mlag_mroute_del msg;
+ ZebraMlagMrouteDelBulk Bulk_msg =
+ ZEBRA_MLAG_MROUTE_DEL_BULK__INIT;
+ ZebraMlagMrouteDel **pay_load = NULL;
+ bool cleanup = false;
+ uint32_t i, actual;
+
+ Bulk_msg.n_mroute_del = mlag_msg.msg_cnt;
+ pay_load = XMALLOC(MTYPE_MLAG_PBUF, sizeof(ZebraMlagMrouteDel *)
+ * mlag_msg.msg_cnt);
+
+ for (i = 0, actual = 0; i < mlag_msg.msg_cnt; i++, actual++) {
+
+ uint32_t vrf_name_len = 0;
+
+ rc = mlag_lib_decode_mroute_del(s, &msg, &length);
+ if (rc) {
+ cleanup = true;
+ break;
+ }
+
+ pay_load[i] = XMALLOC(MTYPE_MLAG_PBUF,
+ sizeof(ZebraMlagMrouteDel));
+ zebra_mlag_mroute_del__init(pay_load[i]);
+
+ vrf_name_len = strlen(msg.vrf_name) + 1;
+ pay_load[i]->vrf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+
+ strlcpy(pay_load[i]->vrf_name, msg.vrf_name,
+ vrf_name_len);
+ pay_load[i]->source_ip = msg.source_ip;
+ pay_load[i]->group_ip = msg.group_ip;
+ pay_load[i]->owner_id = msg.owner_id;
+ pay_load[i]->vrf_id = msg.vrf_id;
+ if (msg.owner_id == MLAG_OWNER_INTERFACE) {
+ vrf_name_len = strlen(msg.intf_name) + 1;
+ pay_load[i]->intf_name =
+ XMALLOC(MTYPE_MLAG_PBUF, vrf_name_len);
+
+ strlcpy(pay_load[i]->intf_name, msg.intf_name,
+ vrf_name_len);
+ }
+ }
+ if (!cleanup) {
+ Bulk_msg.mroute_del = pay_load;
+ len = zebra_mlag_mroute_del_bulk__pack(&Bulk_msg,
+ tmp_buf);
+ }
+
+ for (i = 0; i < actual; i++) {
+ /*
+ * The mlag_lib_decode_mroute_add can
+ * fail to properly decode and cause nothing
+ * to be allocated. Prevent a crash
+ */
+ if (!pay_load[i])
+ continue;
+
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]->vrf_name);
+ if (pay_load[i]->owner_id == MLAG_OWNER_INTERFACE
+ && pay_load[i]->intf_name)
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]->intf_name);
+ XFREE(MTYPE_MLAG_PBUF, pay_load[i]);
+ }
+ XFREE(MTYPE_MLAG_PBUF, pay_load);
+ if (cleanup)
+ return -1;
+ } break;
+ default:
+ break;
+ }
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: length of Mlag ProtoBuf encoded message:%s, %d",
+ __func__,
+ mlag_lib_msgid_to_str(mlag_msg.msg_type, buf,
+ sizeof(buf)),
+ len);
+ hdr.type = (ZebraMlagHeader__MessageType)mlag_msg.msg_type;
+ if (len != 0) {
+ hdr.data.len = len;
+ hdr.data.data = XMALLOC(MTYPE_MLAG_PBUF, len);
+ memcpy(hdr.data.data, tmp_buf, len);
+ }
+
+ /*
+ * ProtoBuf Infra will not support to demarc the pointers whem multiple
+ * messages are posted inside a single Buffer.
+ * 2 -solutions exist to solve this
+ * 1. add Unenoced length at the beginning of every message, this will
+ * be used to point to next message in the buffer
+ * 2. another solution is defining all messages insides another message
+ * But this will permit only 32 messages. this can be extended with
+ * multiple levels.
+ * for simplicity we are going with solution-1.
+ */
+ len = zebra_mlag__header__pack(&hdr,
+ (mlag_wr_buffer + ZEBRA_MLAG_LEN_SIZE));
+ n_len = htonl(len);
+ memcpy(mlag_wr_buffer, &n_len, ZEBRA_MLAG_LEN_SIZE);
+ len += ZEBRA_MLAG_LEN_SIZE;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug(
+ "%s: length of Mlag ProtoBuf message:%s with Header %d",
+ __func__,
+ mlag_lib_msgid_to_str(mlag_msg.msg_type, buf,
+ sizeof(buf)),
+ len);
+ XFREE(MTYPE_MLAG_PBUF, hdr.data.data);
+
+ return len;
+}
+
+static void zebra_fill_protobuf_msg(struct stream *s, char *name, int len)
+{
+ int str_len = strlen(name) + 1;
+
+ stream_put(s, name, str_len);
+ /* Fill the rest with Null Character for aligning */
+ stream_put(s, NULL, len - str_len);
+}
+
+int zebra_mlag_protobuf_decode_message(struct stream *s, uint8_t *data,
+ uint32_t len)
+{
+ uint32_t msg_type;
+ ZebraMlagHeader *hdr;
+ char buf[80];
+
+ hdr = zebra_mlag__header__unpack(NULL, len, data);
+ if (hdr == NULL)
+ return -1;
+
+ /*
+ * ADD The MLAG Header
+ */
+ zclient_create_header(s, ZEBRA_MLAG_FORWARD_MSG, VRF_DEFAULT);
+
+ msg_type = hdr->type;
+
+ if (IS_ZEBRA_DEBUG_MLAG)
+ zlog_debug("%s: Mlag ProtoBuf decoding of message:%s", __func__,
+ mlag_lib_msgid_to_str(msg_type, buf, 80));
+
+ /*
+ * Internal MLAG Message-types & MLAG.proto message types should
+ * always match, otherwise there can be decoding errors
+ * To avoid exposing clients with Protobuf flags, using internal
+ * message-types
+ */
+ stream_putl(s, hdr->type);
+
+ if (hdr->data.len == 0) {
+ /* NULL Payload */
+ stream_putw(s, MLAG_MSG_NULL_PAYLOAD);
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ } else {
+ switch (msg_type) {
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_STATUS_UPDATE: {
+ ZebraMlagStatusUpdate *msg = NULL;
+
+ msg = zebra_mlag_status_update__unpack(
+ NULL, hdr->data.len, hdr->data.data);
+ if (msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, sizeof(struct mlag_status));
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ /* Actual Data */
+ zebra_fill_protobuf_msg(s, msg->peerlink,
+ INTERFACE_NAMSIZ);
+ stream_putl(s, msg->my_role);
+ stream_putl(s, msg->peer_state);
+ zebra_mlag_status_update__free_unpacked(msg, NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_VXLAN_UPDATE: {
+ ZebraMlagVxlanUpdate *msg = NULL;
+
+ msg = zebra_mlag_vxlan_update__unpack(
+ NULL, hdr->data.len, hdr->data.data);
+ if (msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, sizeof(struct mlag_vxlan));
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ /* Actual Data */
+ stream_putl(s, msg->anycast_ip);
+ stream_putl(s, msg->local_ip);
+ zebra_mlag_vxlan_update__free_unpacked(msg, NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_MROUTE_ADD: {
+ ZebraMlagMrouteAdd *msg = NULL;
+
+ msg = zebra_mlag_mroute_add__unpack(NULL, hdr->data.len,
+ hdr->data.data);
+ if (msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, sizeof(struct mlag_mroute_add));
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ /* Actual Data */
+ zebra_fill_protobuf_msg(s, msg->vrf_name, VRF_NAMSIZ);
+
+ stream_putl(s, msg->source_ip);
+ stream_putl(s, msg->group_ip);
+ stream_putl(s, msg->cost_to_rp);
+ stream_putl(s, msg->owner_id);
+ stream_putc(s, msg->am_i_dr);
+ stream_putc(s, msg->am_i_dual_active);
+ stream_putl(s, msg->vrf_id);
+ if (msg->owner_id == MLAG_OWNER_INTERFACE)
+ zebra_fill_protobuf_msg(s, msg->intf_name,
+ INTERFACE_NAMSIZ);
+ else
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+ zebra_mlag_mroute_add__free_unpacked(msg, NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_MROUTE_DEL: {
+ ZebraMlagMrouteDel *msg = NULL;
+
+ msg = zebra_mlag_mroute_del__unpack(NULL, hdr->data.len,
+ hdr->data.data);
+ if (msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, sizeof(struct mlag_mroute_del));
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ /* Actual Data */
+ zebra_fill_protobuf_msg(s, msg->vrf_name, VRF_NAMSIZ);
+
+ stream_putl(s, msg->source_ip);
+ stream_putl(s, msg->group_ip);
+ stream_putl(s, msg->owner_id);
+ stream_putl(s, msg->vrf_id);
+ if (msg->owner_id == MLAG_OWNER_INTERFACE)
+ zebra_fill_protobuf_msg(s, msg->intf_name,
+ INTERFACE_NAMSIZ);
+ else
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+ zebra_mlag_mroute_del__free_unpacked(msg, NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_MROUTE_ADD_BULK: {
+ ZebraMlagMrouteAddBulk *Bulk_msg = NULL;
+ ZebraMlagMrouteAdd *msg = NULL;
+ size_t i, length_spot;
+
+ Bulk_msg = zebra_mlag_mroute_add_bulk__unpack(
+ NULL, hdr->data.len, hdr->data.data);
+ if (Bulk_msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, (Bulk_msg->n_mroute_add
+ * sizeof(struct mlag_mroute_add)));
+ /* No. of msgs in Batch */
+ length_spot = stream_putw(s, Bulk_msg->n_mroute_add);
+
+ /* Actual Data */
+ for (i = 0; i < Bulk_msg->n_mroute_add; i++) {
+ if (STREAM_SIZE(s)
+ < VRF_NAMSIZ + 22 + INTERFACE_NAMSIZ) {
+ zlog_warn(
+ "We have received more messages than we can parse at this point in time: %zu",
+ Bulk_msg->n_mroute_add);
+ break;
+ }
+
+ msg = Bulk_msg->mroute_add[i];
+
+ zebra_fill_protobuf_msg(s, msg->vrf_name,
+ VRF_NAMSIZ);
+ stream_putl(s, msg->source_ip);
+ stream_putl(s, msg->group_ip);
+ stream_putl(s, msg->cost_to_rp);
+ stream_putl(s, msg->owner_id);
+ stream_putc(s, msg->am_i_dr);
+ stream_putc(s, msg->am_i_dual_active);
+ stream_putl(s, msg->vrf_id);
+ if (msg->owner_id == MLAG_OWNER_INTERFACE)
+ zebra_fill_protobuf_msg(
+ s, msg->intf_name,
+ INTERFACE_NAMSIZ);
+ else
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+ }
+
+ stream_putw_at(s, length_spot, i + 1);
+
+ zebra_mlag_mroute_add_bulk__free_unpacked(Bulk_msg,
+ NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_MROUTE_DEL_BULK: {
+ ZebraMlagMrouteDelBulk *Bulk_msg = NULL;
+ ZebraMlagMrouteDel *msg = NULL;
+ size_t i, length_spot;
+
+ Bulk_msg = zebra_mlag_mroute_del_bulk__unpack(
+ NULL, hdr->data.len, hdr->data.data);
+ if (Bulk_msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, (Bulk_msg->n_mroute_del
+ * sizeof(struct mlag_mroute_del)));
+ /* No. of msgs in Batch */
+ length_spot = stream_putw(s, Bulk_msg->n_mroute_del);
+
+ /* Actual Data */
+ for (i = 0; i < Bulk_msg->n_mroute_del; i++) {
+ if (STREAM_SIZE(s)
+ < VRF_NAMSIZ + 16 + INTERFACE_NAMSIZ) {
+ zlog_warn(
+ "We have received more messages than we can parse at this time");
+ break;
+ }
+
+ msg = Bulk_msg->mroute_del[i];
+
+ zebra_fill_protobuf_msg(s, msg->vrf_name,
+ VRF_NAMSIZ);
+ stream_putl(s, msg->source_ip);
+ stream_putl(s, msg->group_ip);
+ stream_putl(s, msg->owner_id);
+ stream_putl(s, msg->vrf_id);
+ if (msg->owner_id == MLAG_OWNER_INTERFACE)
+ zebra_fill_protobuf_msg(
+ s, msg->intf_name,
+ INTERFACE_NAMSIZ);
+ else
+ stream_put(s, NULL, INTERFACE_NAMSIZ);
+ }
+
+ stream_putw_at(s, length_spot, i + 1);
+
+ zebra_mlag_mroute_del_bulk__free_unpacked(Bulk_msg,
+ NULL);
+ } break;
+ case ZEBRA_MLAG__HEADER__MESSAGE_TYPE__ZEBRA_MLAG_ZEBRA_STATUS_UPDATE: {
+ ZebraMlagZebraStatusUpdate *msg = NULL;
+
+ msg = zebra_mlag_zebra_status_update__unpack(
+ NULL, hdr->data.len, hdr->data.data);
+ if (msg == NULL) {
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return -1;
+ }
+ /* Payload len */
+ stream_putw(s, sizeof(struct mlag_frr_status));
+ /* No Batching */
+ stream_putw(s, MLAG_MSG_NO_BATCH);
+ /* Actual Data */
+ stream_putl(s, msg->peer_frrstate);
+ zebra_mlag_zebra_status_update__free_unpacked(msg,
+ NULL);
+ } break;
+ default:
+ break;
+ }
+ }
+ zebra_mlag__header__free_unpacked(hdr, NULL);
+ return msg_type;
+}
+
+#else
+int zebra_mlag_protobuf_encode_client_data(struct stream *s, uint32_t *msg_type)
+{
+ return 0;
+}
+
+int zebra_mlag_protobuf_decode_message(struct stream *s, uint8_t *data,
+ uint32_t len)
+{
+ return 0;
+}
+#endif