Merge work:/home/bk/mysql-4.0

into serg.mysql.com:/usr/home/serg/Abk/mysql-4.0

innobase/include/srv0srv.h

@@ -25,6 +25,7 @@ extern char*	srv_arch_dir;
 extern ulint	srv_n_data_files;
 extern char**	srv_data_file_names;
 extern ulint*	srv_data_file_sizes;
+extern ulint*   srv_data_file_is_raw_partition;
 
 extern char**	srv_log_group_home_dirs;
 
@@ -103,13 +104,26 @@ typedef struct srv_sys_struct	srv_sys_t;
 /* The server system */
 extern srv_sys_t*	srv_sys;
 
-/* Alternatives for fiel flush option in Unix; see the InnoDB manual about
+/* Alternatives for file flush option in Unix; see the InnoDB manual about
 what these mean */
 #define SRV_UNIX_FDATASYNC   1
 #define SRV_UNIX_O_DSYNC     2
 #define SRV_UNIX_LITTLESYNC  3
 #define SRV_UNIX_NOSYNC      4
 
+/* Raw partition flags */
+#define SRV_OLD_RAW 1
+#define SRV_NEW_RAW 2
+
+void
+srv_mysql_thread_release(void);
+/*==========================*/
+os_event_t
+srv_mysql_thread_event_get(void);
+void
+srv_mysql_thread_slot_free(
+/*==========================*/
+			   os_event_t event);
 /*************************************************************************
 Boots Innobase server. */
 

innobase/row/row0sel.c

@@ -2537,7 +2537,10 @@ row_search_for_mysql(
 
 		unique_search_from_clust_index = TRUE;
 
-		if (trx->mysql_n_tables_locked == 0
+		/* Disable this optimization (hence FALSE below) until
+		the hang of Peter Zaitsev has been tracked down */
+
+		if (FALSE && trx->mysql_n_tables_locked == 0
 					&& !prebuilt->sql_stat_start) {
 
 			/* This is a SELECT query done as a consistent read,

innobase/srv/srv0srv.c

@@ -64,6 +64,8 @@ ulint	srv_n_data_files = 0;
 char**	srv_data_file_names = NULL;
 ulint*	srv_data_file_sizes = NULL;	/* size in database pages */ 
 
+ulint*  srv_data_file_is_raw_partition = NULL;
+
 char**	srv_log_group_home_dirs = NULL; 
 
 ulint	srv_n_log_groups	= ULINT_MAX;
@@ -1490,6 +1492,7 @@ srv_init(void)
 		slot = srv_mysql_table + i;
 		slot->in_use = FALSE;
 		slot->event = os_event_create(NULL);
+		slot->suspended = FALSE;
 		ut_a(slot->event);
 	}
 
@@ -1658,6 +1661,7 @@ srv_suspend_mysql_thread(
 	slot->thr = thr;
 
 	os_event_reset(event);	
+	slot->suspended = TRUE;
 
 	slot->suspend_time = ut_time();
 
@@ -1689,6 +1693,27 @@ srv_suspend_mysql_thread(
 	return(FALSE);
 }
 
+os_event_t
+srv_mysql_thread_event_get(void)
+{
+	srv_slot_t*	slot;
+	os_event_t	event;
+
+	mutex_enter(&kernel_mutex);
+	
+	slot = srv_table_reserve_slot_for_mysql();
+
+	event = slot->event;
+	
+	os_event_reset(event);	
+
+	slot->suspended = TRUE;
+
+	mutex_exit(&kernel_mutex);
+
+	return(event);
+}
+
 /************************************************************************
 Releases a MySQL OS thread waiting for a lock to be released, if the
 thread is already suspended. */
@@ -1712,6 +1737,7 @@ srv_release_mysql_thread_if_suspended(
 			/* Found */
 
 			os_event_set(slot->event);
+			slot->suspended = FALSE;
 
 			return;
 		}
@@ -1720,6 +1746,59 @@ srv_release_mysql_thread_if_suspended(
 	/* not found */
 }
 
+void
+srv_mysql_thread_release(void)
+/*==========================*/
+{
+	srv_slot_t*	slot;
+	ulint		i;
+	
+        mutex_enter(&kernel_mutex);
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = srv_mysql_table + i;
+
+		if (slot->in_use && slot->suspended) {
+			/* Found */
+			slot->suspended = FALSE;
+		        mutex_exit(&kernel_mutex);
+
+			os_event_set(slot->event);
+
+			return;
+		}
+	}
+
+	ut_a(0);
+}
+
+void
+srv_mysql_thread_slot_free(
+/*==========================*/
+			   os_event_t event)
+{
+	srv_slot_t*	slot;
+	ulint		i;
+	
+        mutex_enter(&kernel_mutex);
+
+	for (i = 0; i < OS_THREAD_MAX_N; i++) {
+
+		slot = srv_mysql_table + i;
+
+		if (slot->in_use && slot->event == event) {
+			/* Found */
+		        slot->in_use = FALSE;
+		        mutex_exit(&kernel_mutex);
+
+			return;
+		}
+	}
+
+	ut_a(0);
+}
+
 /*************************************************************************
 A thread which wakes up threads whose lock wait may have lasted too long. */
 
@@ -1907,6 +1986,11 @@ loop:
 	}
 
 background_loop:
+	/*
+	sync_array_print_info(sync_primary_wait_array);
+	os_aio_print();
+	buf_print_io();
+	*/
 	/* In this loop we run background operations while the server
 	is quiet */
 
@@ -1967,9 +2051,15 @@ background_loop:
 	}
 		
 /*	mem_print_new_info();
+ */
 
-	fsp_print(0);
+/*	fsp_print(0); */
+
+/*	fprintf(stderr, "Validating tablespace\n");
+	fsp_validate(0);
+	fprintf(stderr, "Validation ok\n");
 */
+
 #ifdef UNIV_SEARCH_PERF_STAT
 /*	btr_search_print_info(); */
 #endif

innobase/srv/srv0start.c

@@ -330,10 +330,28 @@ open_or_create_data_files(
 
 		sprintf(name, "%s%s", srv_data_home, srv_data_file_names[i]);
 	
+		if (srv_data_file_is_raw_partition[i] == 0) {
+
 		  files[i] = os_file_create(name, OS_FILE_CREATE,
 						OS_FILE_NORMAL, &ret);
+		} else if (srv_data_file_is_raw_partition[i] == SRV_OLD_RAW) {
+		  ret = FALSE;
+		} else if (srv_data_file_is_raw_partition[i] == SRV_NEW_RAW) {
+
+		  files[i] = os_file_create(
+				name, OS_FILE_OPEN, OS_FILE_NORMAL, &ret);
+
+		  if (!ret) {
+				fprintf(stderr,
+				"InnoDB: Error in opening %s\n", name);
+
+				return(DB_ERROR);
+		  }
+		}
+
 		if (ret == FALSE) {
-			if (os_file_get_last_error() !=
+			if (srv_data_file_is_raw_partition[i] == 0
+			    && os_file_get_last_error() !=
 						OS_FILE_ALREADY_EXISTS) {
 				fprintf(stderr,
 				"InnoDB: Error in creating or opening %s\n",
@@ -364,8 +382,10 @@ open_or_create_data_files(
 			ret = os_file_get_size(files[i], &size, &size_high);
 			ut_a(ret);
 		
-			if (size != UNIV_PAGE_SIZE * srv_data_file_sizes[i]
-		    					|| size_high != 0) {
+			if (srv_data_file_is_raw_partition[i] == 0 
+			    && (size != UNIV_PAGE_SIZE * srv_data_file_sizes[i]
+				|| size_high != 0)) {
+
 				fprintf(stderr,
 			"InnoDB: Error: data file %s is of different size\n"
 			"InnoDB: than specified in the .cnf file!\n", name);
@@ -722,6 +742,7 @@ innobase_start_or_create_for_mysql(void)
 		mutex_exit(&(log_sys->mutex));
 	}
 
+	/* 	mutex_create(&row_mysql_thread_mutex); */
 	sess_sys_init_at_db_start();
 
 	if (create_new_db) {

sql/ha_innobase.cc

@@ -287,6 +287,26 @@ innobase_parse_data_file_paths_and_sizes(void)
 		        str++;
 		}
 
+		if (size >= 4096) {
+		  fprintf(stderr,
+		     "InnoDB: error: data file size must not be >= 4096M\n");
+		  return(FALSE);
+		}
+
+	        if (strlen(str) >= 6
+			   && *str == 'n'
+			   && *(str + 1) == 'e' 
+		           && *(str + 2) == 'w') {
+		  str += 3;
+		}
+
+	        if (strlen(str) >= 3
+			   && *str == 'r'
+			   && *(str + 1) == 'a' 
+		           && *(str + 2) == 'w') {
+		  str += 3;
+		}
+
 		if (size == 0) {
 			return(FALSE);
 		}
@@ -301,8 +321,9 @@ innobase_parse_data_file_paths_and_sizes(void)
 		}
 	}
 
-	srv_data_file_names = (char**) ut_malloc(i * sizeof(void*));
+	srv_data_file_names = (char**)ut_malloc(i * sizeof(void*));
 	srv_data_file_sizes = (ulint*)ut_malloc(i * sizeof(ulint));
+	srv_data_file_is_raw_partition = (ulint*)ut_malloc(i * sizeof(ulint));
 
 	srv_n_data_files = i;
 
@@ -337,6 +358,27 @@ innobase_parse_data_file_paths_and_sizes(void)
 		        str++;
 		}
 
+		srv_data_file_is_raw_partition[i] = 0;
+
+	        if (strlen(str) >= 6
+			   && *str == 'n'
+			   && *(str + 1) == 'e' 
+		           && *(str + 2) == 'w') {
+		  str += 3;
+		  srv_data_file_is_raw_partition[i] = SRV_NEW_RAW;
+		}
+
+	        if (strlen(str) >= 3
+			   && *str == 'r'
+			   && *(str + 1) == 'a' 
+		           && *(str + 2) == 'w') {
+		  str += 3;
+		  
+		  if (srv_data_file_is_raw_partition[i] == 0) {
+		    srv_data_file_is_raw_partition[i] = SRV_OLD_RAW;
+		  }		  
+		}
+
 		srv_data_file_names[i] = path;
 		srv_data_file_sizes[i] = size;
 
@@ -464,6 +506,7 @@ innobase_init(void)
 	ret = innobase_parse_data_file_paths_and_sizes();
 
 	if (ret == FALSE) {
+	  fprintf(stderr, "InnoDB: syntax error in innodb_data_file_path\n");
 	  DBUG_RETURN(TRUE);
 	}