Merge with 4.0.14

BitKeeper/etc/logging_ok:
  auto-union
scripts/make_win_src_distribution.old:
  Merge rename: scripts/make_win_src_distribution.sh -> scripts/make_win_src_distribution.old
BUILD/compile-pentium-debug-max:
  Auto merged
BitKeeper/deleted/.del-sel000001.result~383913ae4505ec86:
  Auto merged
BitKeeper/deleted/.del-sel000001.test~9567c1646058cc:
  Auto merged
Build-tools/Bootstrap:
  Auto merged
Build-tools/Do-compile:
  Auto merged
Docs/Makefile.am:
  Auto merged
client/get_password.c:
  Auto merged
client/mysql.cc:
  Auto merged
client/mysqltest.c:
  Auto merged
extra/perror.c:
  Auto merged
include/config-win.h:
  Auto merged
include/my_sys.h:
  Auto merged
innobase/btr/btr0cur.c:
  Auto merged
innobase/btr/btr0pcur.c:
  Auto merged
innobase/buf/buf0buf.c:
  Auto merged
innobase/buf/buf0flu.c:
  Auto merged
innobase/dict/dict0dict.c:
  Auto merged
innobase/dict/dict0load.c:
  Auto merged
innobase/include/buf0buf.h:
  Auto merged
innobase/include/log0recv.h:
  Auto merged
innobase/include/row0sel.h:
  Auto merged
innobase/include/srv0srv.h:
  Auto merged
innobase/include/ut0mem.h:
  Auto merged
innobase/lock/lock0lock.c:
  Auto merged
innobase/log/log0log.c:
  Auto merged
innobase/mem/mem0pool.c:
  Auto merged
innobase/os/os0file.c:
  Auto merged
innobase/row/row0mysql.c:
  Auto merged
innobase/row/row0sel.c:
  Auto merged
innobase/srv/srv0srv.c:
  Auto merged
innobase/srv/srv0start.c:
  Auto merged
innobase/trx/trx0sys.c:
  Auto merged
innobase/trx/trx0trx.c:
  Auto merged
innobase/ut/ut0mem.c:
  Auto merged
innobase/ut/ut0ut.c:
  Auto merged
myisam/ft_boolean_search.c:
  Auto merged
myisam/mi_check.c:
  Auto merged
myisam/mi_extra.c:
  Auto merged
myisam/mi_key.c:
  Auto merged
myisam/myisamdef.h:
  Auto merged
myisammrg/myrg_queue.c:
  Auto merged
mysql-test/mysql-test-run.sh:
  Auto merged
mysql-test/r/ctype_latin1_de.result:
  Auto merged
mysql-test/r/flush.result:
  Auto merged
mysql-test/r/func_time.result:
  Auto merged
mysql-test/r/grant_cache.result:
  Auto merged
mysql-test/r/join.result:
  Auto merged
mysql-test/r/join_outer.result:
  Auto merged
mysql-test/r/range.result:
  Auto merged
mysql-test/r/rpl000018.result:
  Auto merged
mysql-test/r/rpl_insert_id.result:
  Auto merged
mysql-test/r/rpl_master_pos_wait.result:
  Auto merged
mysql-test/r/rpl_relayspace.result:
  Auto merged
mysql-test/r/select_safe.result:
  Auto merged
mysql-test/r/symlink.result:
  Auto merged
mysql-test/r/type_date.result:
  Auto merged
mysql-test/r/type_datetime.result:
  Auto merged
mysql-test/t/alias.test:
  Auto merged
mysql-test/t/ctype_latin1_de.test:
  Auto merged
mysql-test/t/fulltext_left_join.test:
  Auto merged
mysql-test/t/func_time.test:
  Auto merged
mysql-test/t/handler.test:
  Auto merged
mysql-test/t/heap.test:
  Auto merged
mysql-test/t/join.test:
  Auto merged
mysql-test/t/join_outer.test:
  Auto merged
mysql-test/t/order_by.test:
  Auto merged
mysql-test/t/range.test:
  Auto merged
mysql-test/t/rpl000001.test:
  Auto merged
mysql-test/t/rpl000018.test:
  Auto merged
mysql-test/t/rpl_insert_id.test:
  Auto merged
mysql-test/t/sel000100.test:
  Auto merged
mysql-test/t/select_safe.test:
  Auto merged
mysql-test/t/type_date.test:
  Auto merged
mysql-test/t/type_datetime.test:
  Auto merged
mysql-test/t/user_var.test:
  Auto merged
mysys/default.c:
  Auto merged
mysys/mf_format.c:
  Auto merged
mysys/my_getopt.c:
  Auto merged
mysys/thr_lock.c:
  Auto merged
mysys/tree.c:
  Auto merged
scripts/Makefile.am:
  Auto merged
scripts/mysql_install_db.sh:
  Auto merged
scripts/mysqld_safe.sh:
  Auto merged
sql/Makefile.am:
  Auto merged
sql/field_conv.cc:
  Auto merged
sql/ha_innodb.h:
  Auto merged
sql/ha_myisam.cc:
  Auto merged
sql/ha_myisammrg.h:
  Auto merged
sql/handler.cc:
  Auto merged
sql/handler.h:
  Auto merged
sql/item.h:
  Auto merged
sql/item_func.cc:
  Auto merged
sql/item_timefunc.cc:
  Auto merged
sql/net_serv.cc:
  Auto merged
sql/nt_servc.cc:
  Auto merged
sql/opt_range.cc:
  Auto merged
sql/sql_base.cc:
  Auto merged
sql/sql_cache.h:
  Auto merged
sql/sql_db.cc:
  Auto merged
sql/sql_delete.cc:
  Auto merged
sql/sql_insert.cc:
  Auto merged
sql/sql_list.h:
  Auto merged
sql/sql_load.cc:
  Auto merged
sql/sql_rename.cc:
  Auto merged
sql/sql_repl.h:
  Auto merged
sql/sql_update.cc:
  Auto merged
sql/table.cc:
  Auto merged
sql/table.h:
  Auto merged
sql/uniques.cc:
  Auto merged
support-files/mysql.spec.sh:
  Auto merged
vio/viosocket.c:
  Auto merged
BitKeeper/deleted/.del-ctype-latin1_de.c~c5d8f9208bceb98e:
  merge
BitKeeper/deleted/.del-mini_client.cc~8677895ec8169183:
  merge
acinclude.m4:
  Merge with 4.0 (openssl patch)
client/mysqlbinlog.cc:
  Merge with 4.0 in which we had added code from 4.1
  (We are basicly only using the 4.1 code here)
configure.in:
  Keep 4.1 file
heap/hp_open.c:
  merge with 4.0
include/my_base.h:
  merge with 4.0
include/my_global.h:
  merge with 4.0
include/mysqld_error.h:
  merge with 4.0
innobase/ha/ha0ha.c:
  merge with 4.0
  (Code should be same but we use indentaion from 4.0)
innobase/log/log0recv.c:
  merge with 4.0
libmysql/libmysql.c:
  Remove with 4.0 code that was ported from 4.1
libmysqld/lib_sql.cc:
  merge with 4.0
myisam/mi_open.c:
  Remove 4.0 specific code
myisam/myisamchk.c:
  merge with 4.0
myisammrg/myrg_rkey.c:
  merge with 4.0
mysql-test/r/alter_table.result:
  May need to be fixed after merge
mysql-test/r/create.result:
  May need to be fixed after merge
mysql-test/r/distinct.result:
  May need to be fixed after merge
mysql-test/r/drop.result:
  May need to be fixed after merge
mysql-test/r/fulltext.result:
  May need to be fixed after merge
mysql-test/r/func_set.result:
  May need to be fixed after merge
mysql-test/r/func_str.result:
  May need to be fixed after merge
mysql-test/r/func_test.result:
  May need to be fixed after merge
mysql-test/r/grant.result:
  May need to be fixed after merge
mysql-test/r/group_by.result:
  May need to be fixed after merge
mysql-test/r/handler.result:
  May need to be fixed after merge
mysql-test/r/heap.result:
  May need to be fixed after merge
mysql-test/r/innodb.result:
  May need to be fixed after merge
mysql-test/r/insert.result:
  May need to be fixed after merge
mysql-test/r/insert_select.result:
  May need to be fixed after merge
mysql-test/r/key_diff.result:
  May need to be fixed after merge
mysql-test/r/merge.result:
  May need to be fixed after merge
mysql-test/r/myisam.result:
  May need to be fixed after merge
mysql-test/r/order_by.result:
  May need to be fixed after merge
mysql-test/r/query_cache.result:
  May need to be fixed after merge
mysql-test/r/rpl_flush_log_loop.result:
  May need to be fixed after merge
mysql-test/r/rpl_loaddata.result:
  May need to be fixed after merge
mysql-test/r/rpl_log.result:
  May need to be fixed after merge
mysql-test/r/rpl_log_pos.result:
  May need to be fixed after merge
mysql-test/r/rpl_rotate_logs.result:
  May need to be fixed after merge
mysql-test/r/select.result:
  May need to be fixed after merge
mysql-test/r/union.result:
  May need to be fixed after merge
mysql-test/r/user_var.result:
  May need to be fixed after merge
mysql-test/t/alter_table.test:
  merge with 4.0
mysql-test/t/create.test:
  merge with 4.0
mysql-test/t/distinct.test:
  merge with 4.0
mysql-test/t/drop.test:
  merge with 4.0
mysql-test/t/flush.test:
  merge with 4.0
mysql-test/t/fulltext.test:
  merge with 4.0
mysql-test/t/func_set.test:
  merge with 4.0
mysql-test/t/func_str.test:
  merge with 4.0
mysql-test/t/func_test.test:
  merge with 4.0
mysql-test/t/grant.test:
  merge with 4.0
mysql-test/t/grant_cache.test:
  merge with 4.0
mysql-test/t/innodb.test:
  Add back EXPLAIN and SHOW KEYS statements, but make them independent of number of rows returned by InnoDB
mysql-test/t/insert.test:
  merge with 4.0
mysql-test/t/insert_select.test:
  merge with 4.0
mysql-test/t/merge.test:
  merge with 4.0
mysql-test/t/query_cache.test:
  merge with 4.0
mysql-test/t/rpl_flush_log_loop.test:
  merge with 4.0
mysql-test/t/rpl_loaddata.test:
  merge with 4.0
mysql-test/t/rpl_rotate_logs.test:
  merge with 4.0
mysql-test/t/select.test:
  merge with 4.0
mysql-test/t/symlink.test:
  merge with 4.0
mysql-test/t/union.test:
  merge with 4.0
mysys/charset.c:
  merge with 4.0
scripts/mysql_fix_privilege_tables.sh:
  merge with 4.0 (Add quoting for some variables)
sql/field.h:
  merge with 4.0
sql/ha_innodb.cc:
  merge with 4.0
sql/item_cmpfunc.cc:
  merge with 4.0
sql/item_cmpfunc.h:
  merge with 4.0
sql/item_func.h:
  merge with 4.0
sql/item_strfunc.cc:
  merge with 4.0
  Fixed null handling with ELT()
sql/item_timefunc.h:
  merge with 4.0
sql/lex.h:
  merge with 4.0
sql/log.cc:
  merge with 4.0
sql/log_event.cc:
  Merge with 4.0
  Cleanups:
  - Indentation
  - #endif comments
  - Replace strmov() with *pos++= for two byte strings
  - Moved variable declarations to start of functions
  - Merged identical code (LOAD_EVENT)
  - Added casts when subtracting pointers
  Did a full diff between this and 4.0 to ensure that the file is correct after merge.
sql/log_event.h:
  merge with 4.0
sql/mysql_priv.h:
  merge with 4.0
sql/mysqld.cc:
  merge with 4.0
sql/repl_failsafe.cc:
  merge with 4.0
sql/set_var.cc:
  merge with 4.0
sql/set_var.h:
  merge with 4.0
sql/share/czech/errmsg.txt:
  merge with 4.0
sql/share/danish/errmsg.txt:
  merge with 4.0
sql/share/dutch/errmsg.txt:
  merge with 4.0
sql/share/english/errmsg.txt:
  merge with 4.0
sql/share/estonian/errmsg.txt:
  merge with 4.0
sql/share/french/errmsg.txt:
  merge with 4.0
sql/share/german/errmsg.txt:
  merge with 4.0
sql/share/greek/errmsg.txt:
  merge with 4.0
sql/share/hungarian/errmsg.txt:
  merge with 4.0
sql/share/italian/errmsg.txt:
  merge with 4.0
sql/share/japanese/errmsg.txt:
  merge with 4.0
sql/share/korean/errmsg.txt:
  merge with 4.0
sql/share/norwegian-ny/errmsg.txt:
  merge with 4.0
sql/share/norwegian/errmsg.txt:
  merge with 4.0
sql/share/polish/errmsg.txt:
  merge with 4.0
sql/share/portuguese/errmsg.txt:
  merge with 4.0
sql/share/romanian/errmsg.txt:
  merge with 4.0
sql/share/russian/errmsg.txt:
  merge with 4.0
sql/share/slovak/errmsg.txt:
  merge with 4.0
sql/share/spanish/errmsg.txt:
  merge with 4.0
sql/share/swedish/errmsg.txt:
  merge with 4.0
sql/share/ukrainian/errmsg.txt:
  merge with 4.0
sql/slave.cc:
  Merge + some indentation fixes
sql/slave.h:
  merge with 4.0
sql/sql_acl.cc:
  merge with 4.0
  Some end space removal to make it easier to do future merges
sql/sql_acl.h:
  merge with 4.0
sql/sql_cache.cc:
  merge with 4.0
sql/sql_class.h:
  merge with 4.0
sql/sql_handler.cc:
  merge with 4.0
sql/sql_lex.cc:
  merge with 4.0
sql/sql_lex.h:
  merge with 4.0
sql/sql_parse.cc:
  merge with 4.0
sql/sql_repl.cc:
  merge with 4.0
sql/sql_select.cc:
  merge with 4.0
sql/sql_table.cc:
  merge with 4.0
sql/sql_union.cc:
  Merge with 4.0
  Note that I couldn't find out how to merge OPTION_FOUND_ROWS handling so this has to be fixed later
sql/sql_yacc.yy:
  merge with 4.0
  Removed end space to make merge easier
vio/Makefile.am:
  merge with 4.0

1 files changed, 330 insertions, 53 deletions

diff --git a/innobase/os/os0file.c b/innobase/os/os0file.c
index e401503c4e3..e31fd1d9efe 100644
--- a/innobase/os/os0file.c
+++ b/innobase/os/os0file.c
@@ -60,6 +60,7 @@ struct os_aio_slot_struct{
 	ulint		pos;		/* index of the slot in the aio
 					array */
 	ibool		reserved;	/* TRUE if this slot is reserved */
+	time_t		reservation_time;/* time when reserved */
 	ulint		len;		/* length of the block to read or
 					write */
 	byte*		buf;		/* buffer used in i/o */
@@ -147,6 +148,12 @@ time_t	os_last_printout;
 
 ibool	os_has_said_disk_full	= FALSE;
 
+/* The mutex protecting the following counts of pending pread and pwrite
+operations */
+os_mutex_t os_file_count_mutex;
+ulint	os_file_n_pending_preads  = 0;
+ulint	os_file_n_pending_pwrites = 0;
+
 
 /***************************************************************************
 Gets the operating system version. Currently works only on Windows. */
@@ -204,9 +211,9 @@ os_file_get_last_error(void)
 	if (err != ERROR_DISK_FULL && err != ERROR_FILE_EXISTS) {
 		ut_print_timestamp(stderr);
 	     	fprintf(stderr,
-  "  InnoDB: Operating system error number %li in a file operation.\n"
+  "  InnoDB: Operating system error number %lu in a file operation.\n"
   "InnoDB: See http://www.innodb.com/ibman.html for installation help.\n",
-		(long) err);
+		err);
 
 		if (err == ERROR_PATH_NOT_FOUND) {
 		         fprintf(stderr,
@@ -248,9 +255,9 @@ os_file_get_last_error(void)
 		ut_print_timestamp(stderr);
 
 	     	fprintf(stderr,
-  "  InnoDB: Operating system error number %li in a file operation.\n"
+  "  InnoDB: Operating system error number %lu in a file operation.\n"
   "InnoDB: See http://www.innodb.com/ibman.html for installation help.\n",
-		(long) err);
+		err);
 
 		if (err == ENOENT) {
 		         fprintf(stderr,
@@ -344,7 +351,8 @@ os_file_handle_error(
 	                fprintf(stderr, "InnoDB: File name %s\n", name);
 	        }
 	  
-		fprintf(stderr, "InnoDB: System call %s.\n", operation);
+		fprintf(stderr, "InnoDB: File operation call: '%s'.\n",
+							       operation);
 		fprintf(stderr, "InnoDB: Cannot continue operation.\n");
 
 		fflush(stderr);
@@ -364,6 +372,8 @@ os_io_init_simple(void)
 {
 	ulint	i;
 
+	os_file_count_mutex = os_mutex_create(NULL);
+
 	for (i = 0; i < OS_FILE_N_SEEK_MUTEXES; i++) {
 		os_file_seek_mutexes[i] = os_mutex_create(NULL);
 	}
@@ -415,9 +425,8 @@ try_again:
 
 	file = CreateFile(name,
 			access,
-			FILE_SHARE_READ | FILE_SHARE_WRITE,
-					/* file can be read and written
-					also by other processes */
+			FILE_SHARE_READ,/* file can be read also by other
+					processes */
 			NULL,	/* default security attributes */
 			create_flag,
 			attributes,
@@ -481,6 +490,101 @@ try_again:
 	return(file);	
 #endif
 }
+
+/********************************************************************
+A simple function to open or create a file. */
+
+os_file_t
+os_file_create_simple_no_error_handling(
+/*====================================*/
+			/* out, own: handle to the file, not defined if error,
+			error number can be retrieved with os_get_last_error */
+	char*	name,	/* in: name of the file or path as a null-terminated
+			string */
+	ulint	create_mode,/* in: OS_FILE_OPEN if an existing file is opened
+			(if does not exist, error), or OS_FILE_CREATE if a new
+			file is created (if exists, error) */
+	ulint	access_type,/* in: OS_FILE_READ_ONLY or OS_FILE_READ_WRITE */
+	ibool*	success)/* out: TRUE if succeed, FALSE if error */
+{
+#ifdef __WIN__
+	os_file_t	file;
+	DWORD		create_flag;
+	DWORD		access;
+	DWORD		attributes	= 0;
+	
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		create_flag = OPEN_EXISTING;
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = CREATE_NEW;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (access_type == OS_FILE_READ_ONLY) {
+		access = GENERIC_READ;
+	} else if (access_type == OS_FILE_READ_WRITE) {
+		access = GENERIC_READ | GENERIC_WRITE;
+	} else {
+		access = 0;
+		ut_error;
+	}
+
+	file = CreateFile(name,
+			access,
+			FILE_SHARE_READ,/* file can be read also by other
+					processes */
+			NULL,	/* default security attributes */
+			create_flag,
+			attributes,
+			NULL);	/* no template file */
+
+	if (file == INVALID_HANDLE_VALUE) {
+		*success = FALSE;
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);
+#else
+	os_file_t	file;
+	int		create_flag;
+	
+	ut_a(name);
+
+	if (create_mode == OS_FILE_OPEN) {
+		if (access_type == OS_FILE_READ_ONLY) {
+			create_flag = O_RDONLY;
+		} else {
+			create_flag = O_RDWR;
+		}
+	} else if (create_mode == OS_FILE_CREATE) {
+		create_flag = O_RDWR | O_CREAT | O_EXCL;
+	} else {
+		create_flag = 0;
+		ut_error;
+	}
+
+	if (create_mode == OS_FILE_CREATE) {
+	        file = open(name, create_flag, S_IRUSR | S_IWUSR
+						| S_IRGRP | S_IWGRP);
+        } else {
+                file = open(name, create_flag);
+        }
+	
+	if (file == -1) {
+		*success = FALSE;
+	} else {
+		*success = TRUE;
+	}
+
+	return(file);	
+#endif
+}
+
 /********************************************************************
 Opens an existing file or creates a new. */
 
@@ -496,7 +600,11 @@ os_file_create(
 			file is created (if exists, error), OS_FILE_OVERWRITE
 			if a new is created or an old overwritten */
 	ulint	purpose,/* in: OS_FILE_AIO, if asynchronous, non-buffered i/o
-			is desired, OS_FILE_NORMAL, if any normal file */
+			is desired, OS_FILE_NORMAL, if any normal file;
+			NOTE that it also depends on type, os_aio_.. and srv_..
+			variables whether we really use async i/o or
+			unbuffered i/o: look in the function source code for
+			the exact rules */
 	ulint	type,	/* in: OS_DATA_FILE or OS_LOG_FILE */
 	ibool*	success)/* out: TRUE if succeed, FALSE if error */
 {
@@ -521,8 +629,8 @@ try_again:
 	}
 
 	if (purpose == OS_FILE_AIO) {
-		/* use asynchronous (overlapped) io and no buffering
-		of writes in the OS */
+		/* If specified, use asynchronous (overlapped) io and no
+		buffering of writes in the OS */
 		attributes = 0;
 #ifdef WIN_ASYNC_IO
 		if (os_aio_use_native_aio) {
@@ -530,17 +638,13 @@ try_again:
 		}
 #endif			
 #ifdef UNIV_NON_BUFFERED_IO
-		if (type == OS_LOG_FILE) {
+		if (type == OS_LOG_FILE && srv_flush_log_at_trx_commit == 2) {
 		        /* Do not use unbuffered i/o to log files because
-		        to allow group commit to work when MySQL binlogging
-			is used we must separate log file write and log
-			file flush to disk. */
-		} else {
-			if (srv_win_file_flush_method ==
-					SRV_WIN_IO_UNBUFFERED) {
-		        	attributes = attributes
-						| FILE_FLAG_NO_BUFFERING;
-			}
+		        value 2 denotes that we do not flush the log at every
+		        commit, but only once per second */
+		} else if (srv_win_file_flush_method ==
+						      SRV_WIN_IO_UNBUFFERED) {
+		        attributes = attributes | FILE_FLAG_NO_BUFFERING;
 		}
 #endif
 	} else if (purpose == OS_FILE_NORMAL) {
@@ -550,12 +654,9 @@ try_again:
 		        /* Do not use unbuffered i/o to log files because
 		        value 2 denotes that we do not flush the log at every
 		        commit, but only once per second */
-		} else {
-			if (srv_win_file_flush_method ==
-					SRV_WIN_IO_UNBUFFERED) {
-		        	attributes = attributes
-						| FILE_FLAG_NO_BUFFERING;
-			}
+		} else if (srv_win_file_flush_method ==
+						      SRV_WIN_IO_UNBUFFERED) {
+		        attributes = attributes | FILE_FLAG_NO_BUFFERING;
 		}
 #endif
 	} else {
@@ -566,9 +667,14 @@ try_again:
 	file = CreateFile(name,
 			GENERIC_READ | GENERIC_WRITE, /* read and write
 							access */
-			FILE_SHARE_READ | FILE_SHARE_WRITE,
-					/* file can be read and written
-					also by other processes */
+			FILE_SHARE_READ,/* File can be read also by other
+					processes; we must give the read
+					permission because of ibbackup. We do
+					not give the write permission to
+					others because if one would succeed to
+					start 2 instances of mysqld on the
+					SAME files, that could cause severe
+					database corruption! */
 			NULL,	/* default security attributes */
 			create_flag,
 			attributes,
@@ -592,30 +698,70 @@ try_again:
 	os_file_t	file;
 	int		create_flag;
 	ibool		retry;
+	const char*	mode_str	= NULL;
+	const char*	type_str	= NULL;
+	const char*	purpose_str	= NULL;
 	
 try_again:	
 	ut_a(name);
 
 	if (create_mode == OS_FILE_OPEN) {
+		mode_str = "OPEN";
+
 		create_flag = O_RDWR;
 	} else if (create_mode == OS_FILE_CREATE) {
+		mode_str = "CREATE";
+
 		create_flag = O_RDWR | O_CREAT | O_EXCL;
 	} else if (create_mode == OS_FILE_OVERWRITE) {
+		mode_str = "OVERWRITE";
+
 		create_flag = O_RDWR | O_CREAT | O_TRUNC;
 	} else {
 		create_flag = 0;
 		ut_error;
 	}
 
-	UT_NOT_USED(purpose);
+	if (type == OS_LOG_FILE) {
+		type_str = "LOG";
+	} else if (type == OS_DATA_FILE) {
+		type_str = "DATA";
+	} else {
+	        ut_a(0);
+	}
+	  
+	if (purpose == OS_FILE_AIO) {
+		purpose_str = "AIO";
+	} else if (purpose == OS_FILE_NORMAL) {
+		purpose_str = "NORMAL";
+	} else {
+	        ut_a(0);
+	}
 
+/*	printf("Opening file %s, mode %s, type %s, purpose %s\n",
+			       name, mode_str, type_str, purpose_str); */
 #ifdef O_SYNC
-	if ((!srv_use_doublewrite_buf || type != OS_DATA_FILE) 
+        /* We let O_SYNC only affect log files; note that we map O_DSYNC to
+	O_SYNC because the datasync options seemed to corrupt files in 2001
+	in both Linux and Solaris */
+	if (type == OS_LOG_FILE
 	    && srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
 
+/*		printf("Using O_SYNC for file %s\n", name); */
+
 	        create_flag = create_flag | O_SYNC;
 	}
 #endif
+#ifdef O_DIRECT
+        /* We let O_DIRECT only affect data files */
+	if (type != OS_LOG_FILE
+	    && srv_unix_file_flush_method == SRV_UNIX_O_DIRECT) {
+
+/*		printf("Using O_DIRECT for file %s\n", name); */
+
+	        create_flag = create_flag | O_DIRECT;
+	}
+#endif
 	if (create_mode == OS_FILE_CREATE) {
 	        file = open(name, create_flag, os_innodb_umask);
         } else {
@@ -677,6 +823,41 @@ os_file_close(
 }
 
 /***************************************************************************
+Closes a file handle. */
+
+ibool
+os_file_close_no_error_handling(
+/*============================*/
+				/* out: TRUE if success */
+	os_file_t	file)	/* in, own: handle to a file */
+{
+#ifdef __WIN__
+	BOOL	ret;
+
+	ut_a(file);
+
+	ret = CloseHandle(file);
+
+	if (ret) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+#else
+	int	ret;
+
+	ret = close(file);
+
+	if (ret == -1) {
+
+		return(FALSE);
+	}
+
+	return(TRUE);
+#endif
+}
+
+/***************************************************************************
 Gets a file size. */
 
 ibool
@@ -849,6 +1030,7 @@ os_file_flush(
 #ifdef HAVE_FDATASYNC
 	ret = fdatasync(file);
 #else
+/*	printf("Flushing to file %lu\n", (ulint)file); */
 	ret = fsync(file);
 #endif
 	os_n_fsyncs++;
@@ -896,6 +1078,7 @@ os_file_pread(
 				offset */
 {
         off_t	offs;
+	ssize_t	n_bytes;
 
 	ut_a((offset & 0xFFFFFFFF) == offset);
         
@@ -917,7 +1100,17 @@ os_file_pread(
 	os_n_file_reads++;
 
 #ifdef HAVE_PREAD
-	return(pread(file, buf, n, offs));
+        os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_preads++;
+        os_mutex_exit(os_file_count_mutex);
+
+        n_bytes = pread(file, buf, n, offs);
+
+        os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_preads--;
+        os_mutex_exit(os_file_count_mutex);
+
+	return(n_bytes);
 #else
 	{
 	ssize_t	ret;
@@ -982,8 +1175,16 @@ os_file_pwrite(
 	os_n_file_writes++;
 
 #ifdef HAVE_PWRITE
+        os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_pwrites++;
+        os_mutex_exit(os_file_count_mutex);
+
 	ret = pwrite(file, buf, n, offs);
 
+        os_mutex_enter(os_file_count_mutex);
+	os_file_n_pending_pwrites--;
+        os_mutex_exit(os_file_count_mutex);
+
 	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
 	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
 	    && !os_do_not_call_flush_at_each_write) {
@@ -1157,6 +1358,7 @@ os_file_write(
 	DWORD		high;
 	ulint		i;
 	ulint		n_retries	= 0;
+	ulint		err;
 
 	ut_a((offset & 0xFFFFFFFF) == offset);
 
@@ -1224,18 +1426,27 @@ retry:
 	
 	if (!os_has_said_disk_full) {
 	
+		err = (ulint)GetLastError();
+
 		ut_print_timestamp(stderr);
 
 		fprintf(stderr,
 "  InnoDB: Error: Write to file %s failed at offset %lu %lu.\n"
 "InnoDB: %lu bytes should have been written, only %lu were written.\n"
 "InnoDB: Operating system error number %lu.\n"
-"InnoDB: Look from section 13.2 at http://www.innodb.com/ibman.html\n"
-"InnoDB: what the error number means.\n"
 "InnoDB: Check that your OS and file system support files of this size.\n"
 "InnoDB: Check also that the disk is not full or a disk quota exceeded.\n",
 			name, offset_high, offset, n, (ulint)len,
-			(ulint)GetLastError());
+			err);
+
+		if (strerror((int)err) != NULL) {
+			fprintf(stderr,
+"InnoDB: Error number %lu means '%s'.\n", err, strerror((int)err));
+		}
+
+		fprintf(stderr,
+"InnoDB: See also section 13.2 at http://www.innodb.com/ibman.html\n"
+"InnoDB: about operating system error numbers.\n");
 
 		os_has_said_disk_full = TRUE;
 	}
@@ -1259,12 +1470,19 @@ retry:
 "  InnoDB: Error: Write to file %s failed at offset %lu %lu.\n"
 "InnoDB: %lu bytes should have been written, only %ld were written.\n"
 "InnoDB: Operating system error number %lu.\n"
-"InnoDB: Look from section 13.2 at http://www.innodb.com/ibman.html\n"
-"InnoDB: what the error number means or use the perror program of MySQL.\n"
 "InnoDB: Check that your OS and file system support files of this size.\n"
 "InnoDB: Check also that the disk is not full or a disk quota exceeded.\n",
 			name, offset_high, offset, n, (long int)ret,
 							(ulint)errno);
+		if (strerror(errno) != NULL) {
+			fprintf(stderr,
+"InnoDB: Error number %lu means '%s'.\n", (ulint)errno, strerror(errno));
+		}
+
+		fprintf(stderr,
+"InnoDB: See also section 13.2 at http://www.innodb.com/ibman.html\n"
+"InnoDB: about operating system error numbers.\n");
+
 		os_has_said_disk_full = TRUE;
 	}
 
@@ -1372,19 +1590,35 @@ os_aio_init(
 
 	os_io_init_simple();
 
+	for (i = 0; i < n_segments; i++) {
+	        srv_io_thread_op_info[i] = (char*)"not started yet";
+	}
+
 	n_per_seg = n / n_segments;
 	n_write_segs = (n_segments - 2) / 2;
 	n_read_segs = n_segments - 2 - n_write_segs;
 	
 	/* printf("Array n per seg %lu\n", n_per_seg); */
 
+	os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1);
+
+	srv_io_thread_function[0] = (char*)"insert buffer thread";
+
+	os_aio_log_array = os_aio_array_create(n_per_seg, 1);
+
+	srv_io_thread_function[1] = (char*)"log thread";
+
 	os_aio_read_array = os_aio_array_create(n_read_segs * n_per_seg,
 							n_read_segs);
+	for (i = 2; i < 2 + n_read_segs; i++) {
+	        srv_io_thread_function[i] = (char*)"read thread";
+	}
+
 	os_aio_write_array = os_aio_array_create(n_write_segs * n_per_seg,
 							n_write_segs);
-	os_aio_ibuf_array = os_aio_array_create(n_per_seg, 1);
-
-	os_aio_log_array = os_aio_array_create(n_per_seg, 1);
+	for (i = 2 + n_read_segs; i < n_segments; i++) {
+	        srv_io_thread_function[i] = (char*)"write thread";
+	}
 
 	os_aio_sync_array = os_aio_array_create(n_slots_sync, 1);
 
@@ -1677,6 +1911,7 @@ loop:
 	}
 	
 	slot->reserved = TRUE;
+	slot->reservation_time = time(NULL);
 	slot->message1 = message1;
 	slot->message2 = message2;
 	slot->file     = file;
@@ -2249,6 +2484,8 @@ os_aio_simulated_handle(
 	ulint		total_len;
 	ulint		offs;
 	ulint		lowest_offset;
+	ulint		biggest_age;
+	ulint		age;
 	byte*		combined_buf;
 	byte*		combined_buf2;
 	ibool		ret;
@@ -2301,22 +2538,55 @@ restart:
 
 	n_consecutive = 0;
 
-	/* Look for an i/o request at the lowest offset in the array
-	(we ignore the high 32 bits of the offset in these heuristics) */
+	/* If there are at least 2 seconds old requests, then pick the oldest
+	one to prevent starvation. If several requests have the same age,
+	then pick the one at the lowest offset. */
 
+	biggest_age = 0;
 	lowest_offset = ULINT_MAX;
-	
+
 	for (i = 0; i < n; i++) {
 		slot = os_aio_array_get_nth_slot(array, i + segment * n);
 
-		if (slot->reserved && slot->offset < lowest_offset) {
+		if (slot->reserved) {
+		        age = (ulint)difftime(time(NULL),
+						slot->reservation_time);
 
-			/* Found an i/o request */
-			consecutive_ios[0] = slot;
+			if ((age >= 2 && age > biggest_age)
+			    || (age >= 2 && age == biggest_age
+			        && slot->offset < lowest_offset)) {
 
-			n_consecutive = 1;
+			        /* Found an i/o request */
+				consecutive_ios[0] = slot;
 
-			lowest_offset = slot->offset;
+				n_consecutive = 1;
+
+				biggest_age = age;
+				lowest_offset = slot->offset;
+			}
+		}
+	}
+
+	if (n_consecutive == 0) {
+	        /* There were no old requests. Look for an i/o request at the
+		lowest offset in the array (we ignore the high 32 bits of the
+		offset in these heuristics) */
+
+		lowest_offset = ULINT_MAX;
+	
+		for (i = 0; i < n; i++) {
+		        slot = os_aio_array_get_nth_slot(array,
+							i + segment * n);
+
+			if (slot->reserved && slot->offset < lowest_offset) {
+
+			        /* Found an i/o request */
+				consecutive_ios[0] = slot;
+
+				n_consecutive = 1;
+
+				lowest_offset = slot->offset;
+			}
 		}
 	}
 
@@ -2422,7 +2692,7 @@ consecutive_loop:
 						+ FIL_PAGE_LSN + 4)
 				    != mach_read_from_4(combined_buf + len2
 				    		+ UNIV_PAGE_SIZE
-				    		- FIL_PAGE_END_LSN + 4)) {
+				    	- FIL_PAGE_END_LSN_OLD_CHKSUM + 4)) {
 				    	ut_print_timestamp(stderr);
 				    	fprintf(stderr,
 "  InnoDB: ERROR: The page to be written seems corrupt!\n");
@@ -2583,14 +2853,15 @@ os_aio_print(
 	double		avg_bytes_read;
 	ulint		i;
 
-	if (buf_end - buf < 1000) {
+	if (buf_end - buf < 1200) {
 
 		return;
 	}
 
 	for (i = 0; i < srv_n_file_io_threads; i++) {
-		buf += sprintf(buf, "I/O thread %lu state: %s\n", i,
-					srv_io_thread_op_info[i]);
+		buf += sprintf(buf, "I/O thread %lu state: %s (%s)\n", i,
+					srv_io_thread_op_info[i],
+					srv_io_thread_function[i]);
 	}
 
 	buf += sprintf(buf, "Pending normal aio reads:");
@@ -2665,6 +2936,12 @@ loop:
 		"%lu OS file reads, %lu OS file writes, %lu OS fsyncs\n",
 		os_n_file_reads, os_n_file_writes, os_n_fsyncs);
 
+	if (os_file_n_pending_preads != 0 || os_file_n_pending_pwrites != 0) {
+	        buf += sprintf(buf,
+		    "%lu pending preads, %lu pending pwrites\n",
+		    os_file_n_pending_preads, os_file_n_pending_pwrites);
+	}
+
 	if (os_n_file_reads == os_n_file_reads_old) {
 		avg_bytes_read = 0.0;
 	} else {