Handle sync and service unload correctly

When sync started and service is unloaded in meantime, it can happen that
sync will call sync_* functions on unloaded service.

Signed-off-by: Jan Friesse <jfriesse@redhat.com>
Reviewed-by: Fabio M. Di Nitto <fdinitto@redhat.com>
(backported from commit 6f6988afff632c6c5068becc855aa4a37a656183)

exec/evil.c

@@ -167,7 +167,10 @@ extern int evil_callbacks_load (int sync_id,
 	int sync_dummy_found = 1;
 	int callbacks_init = 1;
 
-	memset (callbacks, 0, sizeof (struct sync_callbacks));
+	if (callbacks != NULL) {
+		memset (callbacks, 0, sizeof (struct sync_callbacks));
+	}
+
 	switch (sync_id) {
 		case EVS_SERVICE:
 			callbacks_init = 0;
@@ -176,20 +179,28 @@ extern int evil_callbacks_load (int sync_id,
 			/*
 			 * ugh
 			 */
-			memcpy (callbacks, &clm_sync_operations, sizeof (struct sync_callbacks));
+			if (callbacks != NULL) {
+				memcpy (callbacks, &clm_sync_operations, sizeof (struct sync_callbacks));
+			}
 			callbacks_init = 0;
 			break;
 		case AMF_SERVICE:
-			callbacks->name = "dummy AMF service";
+			if (callbacks != NULL) {
+				callbacks->name = "dummy AMF service";
+			}
 			break;
 		case CKPT_SERVICE:
-			callbacks->name = "dummy CKPT service";
+			if (callbacks != NULL) {
+				callbacks->name = "dummy CKPT service";
+			}
 			break;
 		case EVT_SERVICE:
 			/*
 			 * double ugh
 			 */
-			memcpy (callbacks, &evt_sync_operations, sizeof (struct sync_callbacks));
+			if (callbacks != NULL) {
+				memcpy (callbacks, &evt_sync_operations, sizeof (struct sync_callbacks));
+			}
 			callbacks_init = 0;
 			break;
 		case LCK_SERVICE:
@@ -202,7 +213,11 @@ extern int evil_callbacks_load (int sync_id,
 			callbacks_init = 0;
 			break;
 		case CPG_SERVICE:
-			callbacks->name = "dummy CPG service";
+			if (callbacks != NULL) {
+				callbacks->name = "dummy CPG service";
+			} else {
+				return (-1);
+			}
 			break;
 		case CONFDB_SERVICE:
 			callbacks_init = 0;
@@ -213,7 +228,7 @@ extern int evil_callbacks_load (int sync_id,
 			break;
 
 	}
-	if (callbacks_init) {
+	if (callbacks_init && callbacks != NULL) {
 		callbacks->sync_init_api.sync_init_v1 = sync_dummy_init;
 		callbacks->sync_process = sync_dummy_process;
 		callbacks->sync_activate = sync_dummy_activate;

exec/main.c

@@ -300,6 +300,11 @@ static int corosync_sync_callbacks_retrieve (int sync_id,
 		res = evil_callbacks_load (sync_id, callbacks);
 		return (res);
 	}
+
+	if (callbacks == NULL) {
+		return (0);
+	}
+
 	callbacks->name = ais_service[ais_service_index]->name;
 	callbacks->sync_init_api.sync_init_v1 = ais_service[ais_service_index]->sync_init;
 	callbacks->api_version = 1;
@@ -333,6 +338,10 @@ static int corosync_sync_v2_callbacks_retrieve (
 		return (-1);
 	}
 
+	if (callbacks == NULL) {
+		return (0);
+	}
+
 	callbacks->name = ais_service[service_id]->name;
 
 	callbacks->api_version = 1;

exec/sync.c

@@ -89,6 +89,8 @@ static void (*sync_next_start) (
 
 static int sync_recovery_index = 0;
 
+static int my_processing_idx = 0;
+
 static void *sync_callback_token_handle = 0;
 
 static struct barrier_data barrier_data_confchg[PROCESSOR_COUNT_MAX];
@@ -186,12 +188,16 @@ static void sync_start_init (const struct memb_ring_id *ring_id)
 static void sync_service_init (struct memb_ring_id *ring_id)
 {
 	if (sync_callbacks.api_version == 1) {
-		sync_callbacks.sync_init_api.sync_init_v1 (my_member_list,
-			my_member_list_entries, ring_id);
+		if (sync_callbacks_retrieve(my_processing_idx, NULL) != -1) {
+			sync_callbacks.sync_init_api.sync_init_v1 (my_member_list,
+				my_member_list_entries, ring_id);
+		}
 	} else {
-		sync_callbacks.sync_init_api.sync_init_v2 (my_trans_list,
-			my_trans_list_entries,
-			my_member_list, my_member_list_entries, ring_id);
+		if (sync_callbacks_retrieve(my_processing_idx, NULL) != -1) {
+			sync_callbacks.sync_init_api.sync_init_v2 (my_trans_list,
+				my_trans_list_entries,
+				my_member_list, my_member_list_entries, ring_id);
+		}
 	}
 	totempg_callback_token_destroy (&sync_callback_token_handle);
 
@@ -237,6 +243,7 @@ static void sync_callbacks_load (void)
 			sync_processing = 0;
 			break;
 		}
+		my_processing_idx = sync_recovery_index;
 		sync_recovery_index += 1;
 		if (sync_callbacks.sync_init_api.sync_init_v1) {
 			break;
@@ -263,7 +270,11 @@ static int sync_service_process (enum totem_callback_token_type type,
 	 * If process returns 0, then its time to activate
 	 * and start the next service's synchronization
 	 */
-	res = sync_callbacks.sync_process ();
+	if (sync_callbacks_retrieve(my_processing_idx, NULL) != -1) {
+		res = sync_callbacks.sync_process ();
+	} else {
+		res = 0;
+	}
 	if (res != 0) {
 		return (0);
 	}
@@ -341,6 +352,7 @@ static void sync_primary_callback_fn (
 	totempg_callback_token_destroy (&sync_callback_token_handle);
 
 	sync_recovery_index = 0;
+	my_processing_idx = 0;
 	memset (&barrier_data_confchg, 0, sizeof (barrier_data_confchg));
 	for (i = 0; i < view_list_entries; i++) {
 		barrier_data_confchg[i].nodeid = view_list[i];
@@ -426,7 +438,9 @@ static void sync_deliver_fn (
 	 * This sync is complete so activate and start next service sync
 	 */
 	if (barrier_completed && sync_callbacks.sync_activate) {
-		sync_callbacks.sync_activate ();
+		if (sync_callbacks_retrieve(my_processing_idx, NULL) != -1) {
+			sync_callbacks.sync_activate ();
+		}
 
 		log_printf (LOGSYS_LEVEL_DEBUG,
 			"Committing synchronization for (%s)\n",
@@ -481,7 +495,9 @@ static void sync_confchg_fn (
 
 	sync_aborted ();
 	if (sync_processing && sync_callbacks.sync_abort != NULL) {
-		sync_callbacks.sync_abort ();
+		if (sync_callbacks_retrieve(my_processing_idx, NULL) != -1) {
+			sync_callbacks.sync_abort ();
+		}
 		sync_callbacks.sync_activate = NULL;
 	}
 

exec/syncv2.c

@@ -227,8 +227,9 @@ static void sync_barrier_handler (unsigned int nodeid, const void *msg)
 		log_printf (LOGSYS_LEVEL_DEBUG, "Committing synchronization for %s\n",
 			my_service_list[my_processing_idx].name);
 		my_service_list[my_processing_idx].state = ACTIVATE;
-		my_service_list[my_processing_idx].sync_activate ();
-
+		if (my_sync_callbacks_retrieve(my_service_list[my_processing_idx].service_id, NULL) != -1) {
+			my_service_list[my_processing_idx].sync_activate ();
+		}
 		my_processing_idx += 1;
 		if (my_service_list_entries == my_processing_idx) {
 			my_memb_determine_list_entries = 0;
@@ -525,9 +526,11 @@ static int schedwrk_processor (const void *context)
 	if (my_service_list[my_processing_idx].state == INIT) {
 		my_service_list[my_processing_idx].state = PROCESS;
 		if (my_service_list[my_processing_idx].api_version == 1) {
-			my_service_list[my_processing_idx].sync_init_api.sync_init_v1 (my_member_list,
-				my_member_list_entries,
-				&my_ring_id);
+			if (my_sync_callbacks_retrieve(my_service_list[my_processing_idx].service_id, NULL) != -1) {
+				my_service_list[my_processing_idx].sync_init_api.sync_init_v1 (my_member_list,
+					my_member_list_entries,
+					&my_ring_id);
+			}
 		} else {
 			unsigned int old_trans_list[PROCESSOR_COUNT_MAX];
 			size_t old_trans_list_entries = 0;
@@ -548,15 +551,21 @@ static int schedwrk_processor (const void *context)
 				}
 			}
 
-			my_service_list[my_processing_idx].sync_init_api.sync_init_v2 (my_trans_list,
-				my_trans_list_entries, my_member_list,
-				my_member_list_entries,
-				&my_ring_id);
+			if (my_sync_callbacks_retrieve(my_service_list[my_processing_idx].service_id, NULL) != -1) {
+				my_service_list[my_processing_idx].sync_init_api.sync_init_v2 (my_trans_list,
+					my_trans_list_entries, my_member_list,
+					my_member_list_entries,
+					&my_ring_id);
+			}
 		}
 	}
 	if (my_service_list[my_processing_idx].state == PROCESS) {
 		my_service_list[my_processing_idx].state = PROCESS;
-		res = my_service_list[my_processing_idx].sync_process ();
+		if (my_sync_callbacks_retrieve(my_service_list[my_processing_idx].service_id, NULL) != -1) {
+			res = my_service_list[my_processing_idx].sync_process ();
+		} else {
+			res = 0;
+		}
 		if (res == 0) {
 			sync_barrier_enter();
 		} else {
@@ -598,7 +607,9 @@ void sync_v2_abort (void)
 {
 	if (my_state == SYNC_PROCESS) {
 		schedwrk_destroy (my_schedwrk_handle);
-		my_service_list[my_processing_idx].sync_abort ();
+		if (my_sync_callbacks_retrieve(my_service_list[my_processing_idx].service_id, NULL) != -1) {
+			my_service_list[my_processing_idx].sync_abort ();
+		}
 	}
 
 	/* this will cause any "old" barrier messages from causing