Many files:

  Fix hang introduced by selective deadlock resolution
srv0srv.c, row0mysql.c:
  Fix hang introduced by selective deadlock resolution + corruption caused by lock timeout or sel deadl res in ON DELETE CASCADE


innobase/include/que0que.h:
  Fix hang introduced by selective deadlock resolution
innobase/include/trx0trx.h:
  Fix hang introduced by selective deadlock resolution
innobase/include/ut0ut.h:
  Fix hang introduced by selective deadlock resolution
innobase/lock/lock0lock.c:
  Fix hang introduced by selective deadlock resolution
innobase/log/log0log.c:
  Fix hang introduced by selective deadlock resolution
innobase/que/que0que.c:
  Fix hang introduced by selective deadlock resolution
innobase/row/row0mysql.c:
  Fix hang introduced by selective deadlock resolution + corruption caused by lock timeout or sel deadl res in ON DELETE CASCADE
innobase/srv/srv0srv.c:
  Fix hang introduced by selective deadlock resolution + corruption caused by lock timeout or sel deadl res in ON DELETE CASCADE
innobase/trx/trx0sys.c:
  Fix hang introduced by selective deadlock resolution
innobase/trx/trx0trx.c:
  Fix hang introduced by selective deadlock resolution

innobase/include/que0que.h

@@ -117,6 +117,7 @@ que_thr_stop(
 /**************************************************************************
 Moves a thread from another state to the QUE_THR_RUNNING state. Increments
 the n_active_thrs counters of the query graph and transaction. */
+
 void
 que_thr_move_to_run_state_for_mysql(
 /*================================*/
@@ -125,14 +126,17 @@ que_thr_move_to_run_state_for_mysql(
 /**************************************************************************
 A patch for MySQL used to 'stop' a dummy query thread used in MySQL
 select, when there is no error or lock wait. */
+
 void
 que_thr_stop_for_mysql_no_error(
 /*============================*/
 	que_thr_t*	thr,	/* in: query thread */
 	trx_t*		trx);	/* in: transaction */
 /**************************************************************************
-A patch for MySQL used to 'stop' a dummy query thread used in MySQL
-select. */
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL. The
+query thread is stopped and made inactive, except in the case where
+it was put to the lock wait state in lock0lock.c, but the lock has already
+been granted or the transaction chosen as a victim in deadlock resolution. */
 
 void
 que_thr_stop_for_mysql(

innobase/include/trx0trx.h

@@ -429,7 +429,10 @@ struct trx_struct{
 					MySQL */
 	/*------------------------------*/
 	ulint		error_state;	/* 0 if no error, otherwise error
-					number */
+					number; NOTE That ONLY the thread
+					doing the transaction is allowed to
+					set this field: this is NOT protected
+					by the kernel mutex */
 	void*		error_info;	/* if the error number indicates a
 					duplicate key error, a pointer to
 					the problematic index is stored here */
@@ -466,6 +469,12 @@ struct trx_struct{
 					TRX_QUE_LOCK_WAIT, this points to
 					the lock request, otherwise this is
 					NULL */
+	ibool		was_chosen_as_deadlock_victim;
+					/* when the transaction decides to wait
+					for a lock, this it sets this to FALSE;
+					if another transaction chooses this
+					transaction as a victim in deadlock
+					resolution, it sets this to TRUE */
 	time_t          wait_started;   /* lock wait started at this time */
 	UT_LIST_BASE_NODE_T(que_thr_t)
 			wait_thrs;	/* query threads belonging to this

innobase/include/ut0ut.h

@@ -10,7 +10,6 @@ Created 1/20/1994 Heikki Tuuri
 #define ut0ut_h
 
 #include "univ.i"
-#include <string.h>
 #include <time.h>
 #ifndef MYSQL_SERVER
 #include <ctype.h>

innobase/lock/lock0lock.c

@@ -1727,6 +1727,7 @@ index->table_name);
 	}
 
 	trx->que_state = TRX_QUE_LOCK_WAIT;
+	trx->was_chosen_as_deadlock_victim = FALSE;
 	trx->wait_started = time(NULL);
 
 	ut_a(que_thr_stop(thr));
@@ -3173,7 +3174,8 @@ lock_deadlock_recursive(
 				err_buf += sprintf(err_buf,
 				"*** WE ROLL BACK TRANSACTION (1)\n");
 				
-				wait_lock->trx->error_state = DB_DEADLOCK;
+				wait_lock->trx->was_chosen_as_deadlock_victim
+								= TRUE;
 				
 				lock_cancel_waiting_and_release(wait_lock);
 
@@ -3353,6 +3355,7 @@ table->name);
 	}
 	
 	trx->que_state = TRX_QUE_LOCK_WAIT;
+	trx->was_chosen_as_deadlock_victim = FALSE;
 	trx->wait_started = time(NULL);
 
 	ut_a(que_thr_stop(thr));

innobase/log/log0log.c

@@ -1654,8 +1654,8 @@ log_reset_first_header_and_checkpoint(
 	lsn = ut_dulint_add(start, LOG_BLOCK_HDR_SIZE);
 
 	/* Write the label of ibbackup --restore */
-	sprintf((char*) hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP, "ibbackup ");
-	ut_sprintf_timestamp((char*) hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP
+	sprintf(hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP, "ibbackup ");
+	ut_sprintf_timestamp(hdr_buf + LOG_FILE_WAS_CREATED_BY_HOT_BACKUP
 						+ strlen("ibbackup "));
 	buf = hdr_buf + LOG_CHECKPOINT_1;
 	

innobase/que/que0que.c

@@ -1046,14 +1046,16 @@ que_thr_stop(
 }
 
 /**************************************************************************
-A patch for MySQL used to 'stop' a dummy query thread used in MySQL. */
+A patch for MySQL used to 'stop' a dummy query thread used in MySQL. The
+query thread is stopped and made inactive, except in the case where
+it was put to the lock wait state in lock0lock.c, but the lock has already
+been granted or the transaction chosen as a victim in deadlock resolution. */
 
 void
 que_thr_stop_for_mysql(
 /*===================*/
 	que_thr_t*	thr)	/* in: query thread */
 {
-	ibool	stopped 	= FALSE;
 	trx_t*	trx;
 
 	trx = thr_get_trx(thr);
@@ -1067,13 +1069,10 @@ que_thr_stop_for_mysql(
 
 			/* Error handling built for the MySQL interface */
 			thr->state = QUE_THR_COMPLETED;
-
-			stopped = TRUE;
-		}
-		
-		if (!stopped) {
-			/* It must have been a lock wait but the
-			lock was already released */
+		} else {
+			/* It must have been a lock wait but the lock was
+			already released, or this transaction was chosen
+			as a victim in selective deadlock resolution */
 
 			mutex_exit(&kernel_mutex);
 
@@ -1081,6 +1080,10 @@ que_thr_stop_for_mysql(
 		}
 	}
 		
+	ut_ad(thr->is_active == TRUE);
+	ut_ad(trx->n_active_thrs == 1);
+	ut_ad(thr->graph->n_active_thrs == 1);
+
 	thr->is_active = FALSE;
 	(thr->graph)->n_active_thrs--;
 
@@ -1132,6 +1135,9 @@ que_thr_stop_for_mysql_no_error(
 	trx_t*		trx)	/* in: transaction */
 {
 	ut_ad(thr->state == QUE_THR_RUNNING);
+	ut_ad(thr->is_active == TRUE);
+	ut_ad(trx->n_active_thrs == 1);
+	ut_ad(thr->graph->n_active_thrs == 1);
 		
 	if (thr->magic_n != QUE_THR_MAGIC_N) {
 		fprintf(stderr,

innobase/row/row0mysql.c

@@ -198,8 +198,9 @@ row_mysql_handle_errors(
 				/* out: TRUE if it was a lock wait and
 				we should continue running the query thread */
 	ulint*		new_err,/* out: possible new error encountered in
-				rollback, or the old error which was
-				during the function entry */
+				lock wait, or if no new error, the value
+				of trx->error_state at the entry of this
+				function */
 	trx_t*		trx,	/* in: transaction */
 	que_thr_t*	thr,	/* in: query thread */
 	trx_savept_t*	savept)	/* in: savepoint or NULL */
@@ -998,8 +999,8 @@ row_update_cascade_for_mysql(
 				or set null operation */
 	dict_table_t*	table)	/* in: table where we do the operation */
 {
-	ulint		err;
-	trx_t*		trx;
+	ulint	err;
+	trx_t*	trx;
 
 	trx = thr_get_trx(thr);
 run_again:
@@ -1010,11 +1011,28 @@ row_update_cascade_for_mysql(
 
 	err = trx->error_state;
 
+	/* Note that the cascade node is a subnode of another InnoDB
+	query graph node. We do a normal lock wait in this node, but
+	all errors are handled by the parent node. */
+
 	if (err == DB_LOCK_WAIT) {
-		que_thr_stop_for_mysql(thr);
+		/* Handle lock wait here */
 	
-		row_mysql_handle_errors(&err, trx, thr, NULL);
+		que_thr_stop_for_mysql(thr);
 
+		srv_suspend_mysql_thread(thr);
+
+		/* Note that a lock wait may also end in a lock wait timeout,
+		or this transaction is picked as a victim in selective
+		deadlock resolution */
+
+		if (trx->error_state != DB_SUCCESS) {
+
+			return(trx->error_state);
+		}
+
+		/* Retry operation after a normal lock wait */
+		
 		goto run_again;
 	}
 

innobase/srv/srv0srv.c

@@ -2082,13 +2082,24 @@ srv_suspend_mysql_thread(
 
 	if (thr->state == QUE_THR_RUNNING) {
 
-		/* The lock has already been released: no need to suspend */
+		ut_ad(thr->is_active == TRUE);
+	
+		/* The lock has already been released or this transaction
+		was chosen as a deadlock victim: no need to suspend */
+
+		if (trx->was_chosen_as_deadlock_victim) {
+
+			trx->error_state = DB_DEADLOCK;
+			trx->was_chosen_as_deadlock_victim = FALSE;
+		}
 
 		mutex_exit(&kernel_mutex);
 
 		return;
 	}
 	
+	ut_ad(thr->is_active == FALSE);
+
 	slot = srv_table_reserve_slot_for_mysql();
 
 	event = slot->event;
@@ -2142,6 +2153,12 @@ srv_suspend_mysql_thread(
 
 	wait_time = ut_difftime(ut_time(), slot->suspend_time);
 	
+	if (trx->was_chosen_as_deadlock_victim) {
+
+		trx->error_state = DB_DEADLOCK;
+		trx->was_chosen_as_deadlock_victim = FALSE;
+	}
+
 	mutex_exit(&kernel_mutex);
 
 	if (srv_lock_wait_timeout < 100000000 && 

innobase/trx/trx0sys.c

@@ -474,7 +474,7 @@ trx_sys_update_mysql_binlog_offset(
 
 		mlog_write_string(sys_header + field
 					+ TRX_SYS_MYSQL_LOG_NAME,
-			(byte*) file_name, 1 + ut_strlen(file_name), mtr);
+			file_name, 1 + ut_strlen(file_name), mtr);
 	}
 
 	if (mach_read_from_4(sys_header + field

innobase/trx/trx0trx.c

@@ -129,6 +129,7 @@ trx_create(
 	trx->graph = NULL;
 
 	trx->wait_lock = NULL;
+	trx->was_chosen_as_deadlock_victim = FALSE;
 	UT_LIST_INIT(trx->wait_thrs);
 
 	trx->lock_heap = mem_heap_create_in_buffer(256);