MDEV-15895 : make Innodb merge temp tables use pfs_os_file_t for

file IO, rather than int.

On Windows, it is suboptimal to depend on C runtime, as it has limited
number of file descriptors. This change eliminates
os_file_read_no_error_handling_int_fd(), os_file_write_int_fd(),
OS_FILE_FROM_FD() macro.

storage/innobase/handler/ha_innodb.cc

@@ -2410,19 +2410,18 @@ static int mysql_tmpfile_path(const char *path, const char *prefix)
 path. If the path is NULL, then it will be created in tmpdir.
 @param[in]	path	location for creating temporary file
 @return temporary file descriptor, or < 0 on error */
-int
+os_file_t
 innobase_mysql_tmpfile(
 	const char*	path)
 {
 #ifdef WITH_INNODB_DISALLOW_WRITES
 	os_event_wait(srv_allow_writes_event);
 #endif /* WITH_INNODB_DISALLOW_WRITES */
-	int	fd2 = -1;
 	File	fd;
 
 	DBUG_EXECUTE_IF(
 		"innobase_tmpfile_creation_failure",
-		return(-1);
+		return(OS_FILE_CLOSED);
 	);
 
 	if (path == NULL) {
@@ -2431,50 +2430,55 @@ innobase_mysql_tmpfile(
 		fd = mysql_tmpfile_path(path, "ib");
 	}
 
-	if (fd >= 0) {
-		/* Copy the file descriptor, so that the additional resources
-		allocated by create_temp_file() can be freed by invoking
-		my_close().
+	if (fd < 0)
+		return OS_FILE_CLOSED;
+
+	/* Copy the file descriptor, so that the additional resources
+	allocated by create_temp_file() can be freed by invoking
+	my_close().
 
-		Because the file descriptor returned by this function
-		will be passed to fdopen(), it will be closed by invoking
-		fclose(), which in turn will invoke close() instead of
-		my_close(). */
+	Because the file descriptor returned by this function
+	will be passed to fdopen(), it will be closed by invoking
+	fclose(), which in turn will invoke close() instead of
+	my_close(). */
 
 #ifdef _WIN32
-		/* Note that on Windows, the integer returned by mysql_tmpfile
-		has no relation to C runtime file descriptor. Here, we need
-		to call my_get_osfhandle to get the HANDLE and then convert it
-		to C runtime filedescriptor. */
-		{
-			HANDLE hFile = my_get_osfhandle(fd);
-			HANDLE hDup;
-			BOOL bOK = DuplicateHandle(
-					GetCurrentProcess(),
-					hFile, GetCurrentProcess(),
-					&hDup, 0, FALSE, DUPLICATE_SAME_ACCESS);
-			if (bOK) {
-				fd2 = _open_osfhandle((intptr_t) hDup, 0);
-			} else {
-				my_osmaperr(GetLastError());
-				fd2 = -1;
-			}
-		}
+	/* Note that on Windows, the integer returned by mysql_tmpfile
+	has no relation to C runtime file descriptor. Here, we need
+	to call my_get_osfhandle to get the HANDLE and then convert it
+	to C runtime filedescriptor. */
+
+	HANDLE hFile = my_get_osfhandle(fd);
+	HANDLE hDup;
+	BOOL bOK = DuplicateHandle(
+			GetCurrentProcess(),
+			hFile, GetCurrentProcess(),
+			&hDup, 0, FALSE, DUPLICATE_SAME_ACCESS);
+	my_close(fd, MYF(MY_WME));
+
+	if (!bOK) {
+		my_osmaperr(GetLastError());
+		goto error;
+	}
+	return hDup;
 #else
-		fd2 = dup(fd);
+	int fd2 = dup(fd);
+	my_close(fd, MYF(MY_WME));
+	if (fd2 < 0) {
+		set_my_errno(errno);
+		goto error;
+	}
+	return fd2;
 #endif
-		if (fd2 < 0) {
-			char errbuf[MYSYS_STRERROR_SIZE];
-			DBUG_PRINT("error",("Got error %d on dup",fd2));
-			set_my_errno(errno);
-			my_error(EE_OUT_OF_FILERESOURCES,
-				 MYF(0),
-				 "ib*", errno,
-				  my_strerror(errbuf, sizeof(errbuf), errno));
-		}
-		my_close(fd, MYF(MY_WME));
-	}
-	return(fd2);
+
+error:
+	char errbuf[MYSYS_STRERROR_SIZE];
+
+	my_error(EE_OUT_OF_FILERESOURCES,
+		MYF(0),
+		"ib*", errno,
+		my_strerror(errbuf, sizeof(errbuf), errno));
+	return (OS_FILE_CLOSED);
 }
 
 /*********************************************************************//**

storage/innobase/include/os0file.h

@@ -68,10 +68,6 @@ the OS actually supports it: Win 95 does not, NT does. */
 /** File handle */
 typedef HANDLE os_file_t;
 
-/** Convert a C file descriptor to a native file handle
-@param fd file descriptor
-@return native file handle */
-# define OS_FILE_FROM_FD(fd) (HANDLE) _get_osfhandle(fd)
 
 #else /* _WIN32 */
 
@@ -80,14 +76,9 @@ typedef DIR*	os_file_dir_t;	/*!< directory stream */
 /** File handle */
 typedef int	os_file_t;
 
-/** Convert a C file descriptor to a native file handle
-@param fd file descriptor
-@return native file handle */
-# define OS_FILE_FROM_FD(fd) fd
-
 #endif /* _WIN32 */
 
-static const os_file_t OS_FILE_CLOSED = os_file_t(~0);
+static const os_file_t OS_FILE_CLOSED = IF_WIN(os_file_t(INVALID_HANDLE_VALUE),-1);
 
 /** File descriptor with optional PERFORMANCE_SCHEMA instrumentation */
 struct pfs_os_file_t
@@ -843,18 +834,10 @@ The wrapper functions have the prefix of "innodb_". */
 	pfs_os_file_read_no_error_handling_func(			\
 		type, file, buf, offset, n, o, __FILE__, __LINE__)
 
-# define os_file_read_no_error_handling_int_fd(type, file, buf, offset, n) \
-	pfs_os_file_read_no_error_handling_int_fd_func(			\
-		type, file, buf, offset, n, __FILE__, __LINE__)
-
 # define os_file_write(type, name, file, buf, offset, n)	\
 	pfs_os_file_write_func(type, name, file, buf, offset,	\
 			       n, __FILE__, __LINE__)
 
-# define os_file_write_int_fd(type, name, file, buf, offset, n)		\
-	pfs_os_file_write_int_fd_func(type, name, file, buf, offset,	\
-		n, __FILE__, __LINE__)
-
 # define os_file_flush(file)					\
 	pfs_os_file_flush_func(file, __FILE__, __LINE__)
 
@@ -1564,7 +1547,7 @@ path. If the path is NULL then it will be created on --tmpdir location.
 This function is defined in ha_innodb.cc.
 @param[in]	path	location for creating temporary file
 @return temporary file descriptor, or < 0 on error */
-int
+os_file_t
 innobase_mysql_tmpfile(
 	const char*	path);
 

storage/innobase/include/os0file.ic

@@ -340,49 +340,6 @@ pfs_os_file_read_no_error_handling_func(
 	return(result);
 }
 
-/** NOTE! Please use the corresponding macro
-os_file_read_no_error_handling_int_fd() to request
-a synchronous read operation.
-@param[in]	type		read request
-@param[in]      file            file handle
-@param[out]     buf             buffer where to read
-@param[in]      offset          file offset where to read
-@param[in]      n               number of bytes to read
-@param[in]      src_file        caller file name
-@param[in]      src_line        caller line number
-@return	whether the request was successful */
-UNIV_INLINE
-bool
-pfs_os_file_read_no_error_handling_int_fd_func(
-	const IORequest&	type,
-	int			file,
-	void*			buf,
-	os_offset_t		offset,
-	ulint			n,
-	const char*		src_file,
-	uint			src_line)
-{
-	PSI_file_locker_state	state;
-
-	PSI_file_locker* locker = PSI_FILE_CALL(
-		get_thread_file_descriptor_locker)(
-			&state, file, PSI_FILE_READ);
-	if (locker != NULL) {
-		PSI_FILE_CALL(start_file_wait)(
-			locker, n,
-			__FILE__, __LINE__);
-	}
-
-	bool success = DB_SUCCESS == os_file_read_no_error_handling_func(
-		type, OS_FILE_FROM_FD(file), buf, offset, n, NULL);
-
-	if (locker != NULL) {
-		PSI_FILE_CALL(end_file_wait)(locker, n);
-	}
-
-	return(success);
-}
-
 /** NOTE! Please use the corresponding macro os_file_write(), not directly
 this function!
 This is the performance schema instrumented wrapper function for
@@ -425,51 +382,6 @@ pfs_os_file_write_func(
 	return(result);
 }
 
-/** NOTE! Please use the corresponding macro os_file_write_int_fd(),
-not directly this function!
-This is the performance schema instrumented wrapper function for
-os_file_write_int_fd() which requests a synchronous write operation.
-@param[in]	type		write request
-@param[in]	name		file name
-@param[in]	file		file handle
-@param[in]	buf		buffer to write
-@param[in]	offset		file offset
-@param[in]	n		number of bytes
-@param[in]	src_file	file name where func invoked
-@param[in]	src_line	line where the func invoked
-@return	whether the request was successful */
-UNIV_INLINE
-bool
-pfs_os_file_write_int_fd_func(
-	const IORequest&	type,
-	const char*		name,
-	int			file,
-	const void*		buf,
-	os_offset_t		offset,
-	ulint			n,
-	const char*		src_file,
-	uint			src_line)
-{
-	PSI_file_locker_state   state;
-	struct PSI_file_locker* locker;
-
-	locker = PSI_FILE_CALL(get_thread_file_descriptor_locker)(
-		&state, file, PSI_FILE_WRITE);
-	if (locker != NULL) {
-                PSI_FILE_CALL(start_file_wait)(
-			locker, n,
-			__FILE__, __LINE__);
-	}
-
-        bool success = DB_SUCCESS == os_file_write_func(
-		type, name, OS_FILE_FROM_FD(file), buf, offset, n);
-
-        if (locker != NULL) {
-                PSI_FILE_CALL(end_file_wait)(locker, n);
-        }
-
-        return(success);
-}
 
 /** NOTE! Please use the corresponding macro os_file_flush(), not directly
 this function!

storage/innobase/include/row0merge.h

@@ -101,7 +101,7 @@ struct row_merge_buf_t {
 
 /** Information about temporary files used in merge sort */
 struct merge_file_t {
-	int		fd;		/*!< file descriptor */
+	pfs_os_file_t	fd;		/*!< file descriptor */
 	ulint		offset;		/*!< file offset (end of file) */
 	ib_uint64_t	n_rec;		/*!< number of records in the file */
 };
@@ -193,7 +193,7 @@ row_merge_drop_temp_indexes(void);
 UNIV_PFS_IO defined, register the file descriptor with Performance Schema.
 @param[in]	path	location for creating temporary merge files, or NULL
 @return File descriptor */
-int
+pfs_os_file_t
 row_merge_file_create_low(
 	const char*	path)
 	MY_ATTRIBUTE((warn_unused_result));
@@ -203,7 +203,7 @@ if UNIV_PFS_IO is defined. */
 void
 row_merge_file_destroy_low(
 /*=======================*/
-	int		fd);	/*!< in: merge file descriptor */
+	const pfs_os_file_t&	fd);	/*!< in: merge file descriptor */
 
 /*********************************************************************//**
 Provide a new pathname for a table that is being renamed if it belongs to
@@ -372,7 +372,7 @@ UNIV_INTERN
 bool
 row_merge_write(
 /*============*/
-	int		fd,	/*!< in: file descriptor */
+	const pfs_os_file_t&	fd,	/*!< in: file descriptor */
 	ulint		offset,	/*!< in: offset where to write,
 				in number of row_merge_block_t elements */
 	const void*	buf,	/*!< in: data */
@@ -393,7 +393,7 @@ row_merge_buf_empty(
 @param[out]	merge_file	merge file structure
 @param[in]	path		location for creating temporary file, or NULL
 @return file descriptor, or -1 on failure */
-int
+pfs_os_file_t
 row_merge_file_create(
 	merge_file_t*	merge_file,
 	const char*	path)
@@ -421,7 +421,7 @@ row_merge_sort(
 	const row_merge_dup_t*	dup,
 	merge_file_t*		file,
 	row_merge_block_t*	block,
-	int*			tmpfd,
+	pfs_os_file_t*		tmpfd,
 	const bool		update_progress,
 	const double	pct_progress,
 	const double	pct_cost,
@@ -460,7 +460,7 @@ row_merge_file_destroy(
 bool
 row_merge_read(
 /*===========*/
-	int			fd,	/*!< in: file descriptor */
+	const pfs_os_file_t&	fd,	/*!< in: file descriptor */
 	ulint			offset,	/*!< in: offset where to read
 					in number of row_merge_block_t
 					elements */
@@ -479,7 +479,7 @@ row_merge_read_rec(
 	mrec_buf_t*		buf,	/*!< in/out: secondary buffer */
 	const byte*		b,	/*!< in: pointer to record */
 	const dict_index_t*	index,	/*!< in: index of the record */
-	int			fd,	/*!< in: file descriptor */
+	const pfs_os_file_t&	fd,	/*!< in: file descriptor */
 	ulint*			foffs,	/*!< in/out: file offset */
 	const mrec_t**		mrec,	/*!< out: pointer to merge record,
 					or NULL on end of list

storage/innobase/os/os0file.cc

@@ -1253,10 +1253,23 @@ os_file_create_tmpfile()
 {
 	FILE*	file	= NULL;
 	WAIT_ALLOW_WRITES();
-	int	fd	= innobase_mysql_tmpfile(NULL);
+	os_file_t	fd	= innobase_mysql_tmpfile(NULL);
 
-	if (fd >= 0) {
+	if (fd != OS_FILE_CLOSED) {
+#ifdef _WIN32
+		int crt_fd = _open_osfhandle((intptr_t)HANDLE(fd), 0);
+		if (crt_fd != -1) {
+			file = fdopen(crt_fd, "w+b");
+			if (!file) {
+				close(crt_fd);
+			}
+		}
+#else
 		file = fdopen(fd, "w+b");
+		if (!file) {
+			close(fd);
+		}
+#endif
 	}
 
 	if (file == NULL) {
@@ -1264,10 +1277,6 @@ os_file_create_tmpfile()
 		ib::error()
 			<< "Unable to create temporary file; errno: "
 			<< errno;
-
-		if (fd >= 0) {
-			close(fd);
-		}
 	}
 
 	return(file);

storage/innobase/row/row0ftsort.cc

@@ -255,7 +255,7 @@ row_fts_psort_info_init(
 				dup->index);
 
 			if (row_merge_file_create(psort_info[j].merge_file[i],
-						  path) < 0) {
+						  path) == OS_FILE_CLOSED) {
 				goto func_exit;
 			}
 
@@ -776,7 +776,7 @@ DECLARE_THREAD(fts_parallel_tokenization)(
 	merge_file_t**		merge_file;
 	row_merge_block_t**	block;
 	row_merge_block_t**	crypt_block;
-	int			tmpfd[FTS_NUM_AUX_INDEX];
+	pfs_os_file_t		tmpfd[FTS_NUM_AUX_INDEX];
 	ulint			mycount[FTS_NUM_AUX_INDEX];
 	ib_uint64_t		total_rec = 0;
 	ulint			num_doc_processed = 0;
@@ -1029,7 +1029,7 @@ DECLARE_THREAD(fts_parallel_tokenization)(
 		}
 
 		tmpfd[i] = row_merge_file_create_low(path);
-		if (tmpfd[i] < 0) {
+		if (tmpfd[i] == OS_FILE_CLOSED) {
 			error = DB_OUT_OF_MEMORY;
 			goto func_exit;
 		}
@@ -1041,12 +1041,12 @@ DECLARE_THREAD(fts_parallel_tokenization)(
 				       crypt_block[i], table->space->id);
 
 		if (error != DB_SUCCESS) {
-			close(tmpfd[i]);
+			os_file_close(tmpfd[i]);
 			goto func_exit;
 		}
 
 		total_rec += merge_file[i]->n_rec;
-		close(tmpfd[i]);
+		os_file_close(tmpfd[i]);
 	}
 
 func_exit:
@@ -1570,7 +1570,7 @@ row_fts_merge_insert(
 	ib_alloc_t*		heap_alloc;
 	ulint			i;
 	mrec_buf_t**		buf;
-	int*			fd;
+	pfs_os_file_t*			fd;
 	byte**			block;
 	byte**			crypt_block;
 	const mrec_t**		mrec;
@@ -1612,7 +1612,7 @@ row_fts_merge_insert(
 		heap, sizeof(*offsets) * fts_sort_pll_degree);
 	buf = (mrec_buf_t**) mem_heap_alloc(
 		heap, sizeof(*buf) * fts_sort_pll_degree);
-	fd = (int*) mem_heap_alloc(heap, sizeof(*fd) * fts_sort_pll_degree);
+	fd = (pfs_os_file_t*) mem_heap_alloc(heap, sizeof(*fd) * fts_sort_pll_degree);
 	block = (byte**) mem_heap_alloc(
 		heap, sizeof(*block) * fts_sort_pll_degree);
 	crypt_block = (byte**) mem_heap_alloc(

storage/innobase/row/row0log.cc

@@ -171,7 +171,7 @@ When head.blocks == tail.blocks, the reader will access tail.block
 directly. When also head.bytes == tail.bytes, both counts will be
 reset to 0 and the file will be truncated. */
 struct row_log_t {
-	int		fd;	/*!< file descriptor */
+	pfs_os_file_t	fd;	/*!< file descriptor */
 	ib_mutex_t	mutex;	/*!< mutex protecting error,
 				max_trx and tail */
 	page_no_map*	blobs;	/*!< map of page numbers of off-page columns
@@ -226,18 +226,18 @@ struct row_log_t {
 @param[in,out] log     online rebuild log
 @return true if success, false if not */
 static MY_ATTRIBUTE((warn_unused_result))
-int
+pfs_os_file_t
 row_log_tmpfile(
 	row_log_t*	log)
 {
 	DBUG_ENTER("row_log_tmpfile");
-	if (log->fd < 0) {
+	if (log->fd == OS_FILE_CLOSED) {
 		log->fd = row_merge_file_create_low(log->path);
 		DBUG_EXECUTE_IF("row_log_tmpfile_fail",
-				if (log->fd > 0)
+				if (log->fd != OS_FILE_CLOSED)
 					row_merge_file_destroy_low(log->fd);
-				log->fd = -1;);
-		if (log->fd >= 0) {
+				log->fd = OS_FILE_CLOSED;);
+		if (log->fd != OS_FILE_CLOSED) {
 			MONITOR_ATOMIC_INC(MONITOR_ALTER_TABLE_LOG_FILES);
 		}
 	}
@@ -392,7 +392,7 @@ row_log_online_op(
 
 		UNIV_MEM_ASSERT_RW(buf, srv_sort_buf_size);
 
-		if (row_log_tmpfile(log) < 0) {
+		if (row_log_tmpfile(log) == OS_FILE_CLOSED) {
 			log->error = DB_OUT_OF_MEMORY;
 			goto err_exit;
 		}
@@ -413,7 +413,7 @@ row_log_online_op(
 		}
 
 		log->tail.blocks++;
-		if (!os_file_write_int_fd(
+		if (!os_file_write(
 			    request,
 			    "(modification log)",
 			    log->fd,
@@ -528,7 +528,7 @@ row_log_table_close_func(
 
 		UNIV_MEM_ASSERT_RW(buf, srv_sort_buf_size);
 
-		if (row_log_tmpfile(log) < 0) {
+		if (row_log_tmpfile(log) == OS_FILE_CLOSED) {
 			log->error = DB_OUT_OF_MEMORY;
 			goto err_exit;
 		}
@@ -549,7 +549,7 @@ row_log_table_close_func(
 		}
 
 		log->tail.blocks++;
-		if (!os_file_write_int_fd(
+		if (!os_file_write(
 			    request,
 			    "(modification log)",
 			    log->fd,
@@ -2865,9 +2865,9 @@ row_log_table_apply_ops(
 		IORequest		request(IORequest::READ);
 		byte*			buf = index->online_log->head.block;
 
-		if (!os_file_read_no_error_handling_int_fd(
+		if (!os_file_read_no_error_handling(
 			    request, index->online_log->fd,
-			    buf, ofs, srv_sort_buf_size)) {
+			    buf, ofs, srv_sort_buf_size, 0)) {
 			ib::error()
 				<< "Unable to read temporary file"
 				" for table " << index->table->name;
@@ -3173,7 +3173,7 @@ row_log_allocate(
 		DBUG_RETURN(false);
 	}
 
-	log->fd = -1;
+	log->fd = OS_FILE_CLOSED;
 	mutex_create(LATCH_ID_INDEX_ONLINE_LOG, &log->mutex);
 
 	log->blobs = NULL;
@@ -3732,9 +3732,9 @@ row_log_apply_ops(
 
 		byte*	buf = index->online_log->head.block;
 
-		if (!os_file_read_no_error_handling_int_fd(
+		if (!os_file_read_no_error_handling(
 			    request, index->online_log->fd,
-			    buf, ofs, srv_sort_buf_size)) {
+			    buf, ofs, srv_sort_buf_size, 0)) {
 			ib::error()
 				<< "Unable to read temporary file"
 				" for index " << index->name;

storage/innobase/srv/srv0start.cc

@@ -756,7 +756,7 @@ dberr_t
 srv_check_undo_redo_logs_exists()
 {
 	bool		ret;
-	os_file_t	fh;
+	pfs_os_file_t	fh;
 	char	name[OS_FILE_MAX_PATH];
 
 	/* Check if any undo tablespaces exist */