MDEV-4338 - Support FusionIO/directFS atomic writes

5 files changed, 153 insertions, 0 deletions

diff --git a/storage/xtradb/fil/fil0fil.c b/storage/xtradb/fil/fil0fil.c
index af8353cdf45..397c4de4b6e 100644
--- a/storage/xtradb/fil/fil0fil.c
+++ b/storage/xtradb/fil/fil0fil.c
@@ -4865,6 +4865,24 @@ fil_extend_space_to_desired_size(
 	start_page_no = space->size;
 	file_start_page_no = space->size - node->size;
 
+#ifdef HAVE_POSIX_FALLOCATE
+	if (srv_use_posix_fallocate) {
+		offset_high = size_after_extend * page_size / (4ULL*1024*1024*1024);
+		offset_low = size_after_extend * page_size % (4ULL*1024*1024*1024);
+
+		mutex_exit(&fil_system->mutex);
+		success = os_file_set_size(node->name, node->handle,
+				offset_low, offset_high);
+		mutex_enter(&fil_system->mutex);
+		if (success) {
+			node->size += (size_after_extend - start_page_no);
+			space->size += (size_after_extend - start_page_no);
+			os_has_said_disk_full = FALSE;
+		}
+		goto complete_io;
+	}
+#endif
+
 	/* Extend at most 64 pages at a time */
 	buf_size = ut_min(64, size_after_extend - start_page_no) * page_size;
 	buf2 = mem_alloc(buf_size + page_size);
@@ -4921,6 +4939,10 @@ fil_extend_space_to_desired_size(
 
 	mem_free(buf2);
 
+#ifdef HAVE_POSIX_FALLOCATE
+complete_io:
+#endif
+
 	fil_node_complete_io(node, fil_system, OS_FILE_WRITE);
 
 	*actual_size = space->size;
diff --git a/storage/xtradb/handler/ha_innodb.cc b/storage/xtradb/handler/ha_innodb.cc
index 19ead6a5b29..c194dc38889 100644
--- a/storage/xtradb/handler/ha_innodb.cc
+++ b/storage/xtradb/handler/ha_innodb.cc
@@ -185,6 +185,8 @@ static my_bool	innobase_file_format_check		= TRUE;
 static my_bool	innobase_log_archive			= FALSE;
 static char*	innobase_log_arch_dir			= NULL;
 #endif /* UNIV_LOG_ARCHIVE */
+static my_bool	innobase_use_atomic_writes		= FALSE;
+static my_bool	innobase_use_fallocate			= TRUE;
 static my_bool	innobase_use_doublewrite		= TRUE;
 static my_bool	innobase_use_checksums			= TRUE;
 static my_bool	innobase_fast_checksum			= FALSE;
@@ -3057,6 +3059,38 @@ innobase_change_buffering_inited_ok:
 	srv_kill_idle_transaction = 0;
 #endif
 
+#ifdef HAVE_POSIX_FALLOCATE
+	srv_use_posix_fallocate = (ibool) innobase_use_fallocate;
+#endif
+	srv_use_atomic_writes = (ibool) innobase_use_atomic_writes;
+	if (innobase_use_atomic_writes) {
+		fprintf(stderr, "InnoDB: using atomic writes.\n");
+
+		/* Force doublewrite buffer off, atomic writes replace it. */
+		if (srv_use_doublewrite_buf) {
+			fprintf(stderr, "InnoDB: Switching off doublewrite buffer "
+				"because of atomic writes.\n");
+				innobase_use_doublewrite = srv_use_doublewrite_buf = FALSE;
+		}
+
+		/* Force O_DIRECT on Unixes (on Windows writes are always unbuffered)*/
+#ifndef _WIN32
+		if(!innobase_file_flush_method ||
+			!strstr(innobase_file_flush_method, "O_DIRECT")) {
+			innobase_file_flush_method = 
+				srv_file_flush_method_str = (char*)"O_DIRECT";
+			fprintf(stderr, "InnoDB: using O_DIRECT due to atomic writes.\n");
+		}
+#endif
+#ifdef HAVE_POSIX_FALLOCATE
+		/* Due to a bug in directFS, using atomics needs  
+		 * posix_fallocate to extend the file
+		 * pwrite()  past end of the file won't work
+		 */
+		srv_use_posix_fallocate = TRUE;
+#endif
+	}
+
 #ifdef HAVE_PSI_INTERFACE
 	/* Register keys with MySQL performance schema */
 	if (PSI_server) {
@@ -12615,6 +12649,20 @@ static MYSQL_SYSVAR_BOOL(doublewrite, innobase_use_doublewrite,
   "Disable with --skip-innodb-doublewrite.",
   NULL, NULL, TRUE);
 
+static MYSQL_SYSVAR_BOOL(use_atomic_writes, innobase_use_atomic_writes,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Prevent partial page writes, via atomic writes."
+  "The option is used to prevent partial writes in case of a crash/poweroff, "
+  "as faster alternative to doublewrite buffer."
+  "Currently this option works only "
+  "on Linux only with FusionIO device, and directFS filesystem.",
+  NULL, NULL, FALSE);
+
+static MYSQL_SYSVAR_BOOL(use_fallocate, innobase_use_fallocate,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Preallocate files fast, using operating system functionality. On POSIX systems, posix_fallocate system call is used.",
+  NULL, NULL, TRUE);
+
 static MYSQL_SYSVAR_ULONG(io_capacity, srv_io_capacity,
   PLUGIN_VAR_RQCMDARG,
   "Number of IOPs the server can do. Tunes the background IO rate",
@@ -13259,6 +13307,8 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(doublewrite_file),
   MYSQL_SYSVAR(data_home_dir),
   MYSQL_SYSVAR(doublewrite),
+  MYSQL_SYSVAR(use_atomic_writes),
+  MYSQL_SYSVAR(use_fallocate),
   MYSQL_SYSVAR(recovery_stats),
   MYSQL_SYSVAR(fast_shutdown),
   MYSQL_SYSVAR(file_io_threads),
diff --git a/storage/xtradb/include/srv0srv.h b/storage/xtradb/include/srv0srv.h
index a64e5ee607b..586c1e73879 100644
--- a/storage/xtradb/include/srv0srv.h
+++ b/storage/xtradb/include/srv0srv.h
@@ -249,6 +249,11 @@ extern ulong	srv_sys_stats_root_page;
 #endif
 
 extern ibool	srv_use_doublewrite_buf;
+extern ibool	srv_use_atomic_writes;
+#ifdef HAVE_POSIX_FALLOCATE
+extern ibool	srv_use_posix_fallocate;
+#endif
+
 extern ibool	srv_use_checksums;
 extern ibool	srv_fast_checksum;
 
diff --git a/storage/xtradb/os/os0file.c b/storage/xtradb/os/os0file.c
index f4a6c33655f..8f1b3e46bb2 100644
--- a/storage/xtradb/os/os0file.c
+++ b/storage/xtradb/os/os0file.c
@@ -1454,6 +1454,43 @@ os_file_set_nocache(
 #endif
 }
 
+
+#ifdef __linux__
+#include <sys/ioctl.h>
+#ifndef DFS_IOCTL_ATOMIC_WRITE_SET 
+#define DFS_IOCTL_ATOMIC_WRITE_SET _IOW(0x95, 2, uint)
+#endif
+static int os_file_set_atomic_writes(os_file_t file, const char *name) 
+{
+	static int first_time = 1;
+	int atomic_option = 1;
+
+	int ret = ioctl (file, DFS_IOCTL_ATOMIC_WRITE_SET, &atomic_option);
+
+	if (ret) {
+		fprintf(stderr, 
+		"InnoDB : can't use atomic write on %s, errno %d\n",
+		name, errno);
+		return ret;
+	}
+	return ret;
+}
+#else 
+static int os_file_set_atomic_writes(os_file_t file, const char *name) 
+{
+	fprintf(stderr,
+	"InnoDB : can't use atomic writes on %s - not implemented on this platform."
+	"innodb_use_atomic_writes needs to be 0.\n", 
+	name);
+#ifdef _WIN32
+	SetLastError(ERROR_INVALID_FUNCTION);
+#else
+	errno = EINVAL;
+#endif
+	return -1;
+}
+#endif
+
 /****************************************************************//**
 NOTE! Use the corresponding macro os_file_create(), not directly
 this function!
@@ -1618,6 +1655,13 @@ try_again:
 		}
 	}
 
+	if (srv_use_atomic_writes && type == OS_DATA_FILE && 
+		os_file_set_atomic_writes(file, name)) {
+			 CloseHandle(file);
+			*success = FALSE;
+			file = INVALID_HANDLE_VALUE;
+	}
+
 	return(file);
 #else /* __WIN__ */
 	os_file_t	file;
@@ -1737,6 +1781,12 @@ try_again:
 		file = -1;
 	}
 #endif /* USE_FILE_LOCK */
+	if (srv_use_atomic_writes && type == OS_DATA_FILE 
+		&& os_file_set_atomic_writes(file, name)) {
+			close(file);
+			*success = FALSE;
+			file = -1;
+	}
 
 	return(file);
 #endif /* __WIN__ */
@@ -2081,6 +2131,28 @@ os_file_set_size(
 	current_size = 0;
 	desired_size = (ib_int64_t)size + (((ib_int64_t)size_high) << 32);
 
+#ifdef HAVE_POSIX_FALLOCATE
+        if (srv_use_posix_fallocate) {
+		if (posix_fallocate(file, current_size, desired_size) == -1) {
+			fprintf(stderr,
+		 	"InnoDB: Error: preallocating data for"
+			" file %s failed at\n"
+			"InnoDB: offset 0 size %lld %lld. Operating system"
+			" error number %llu.\n"
+			"InnoDB: Check that the disk is not full"
+			" or a disk quota exceeded.\n"
+			"InnoDB: Some operating system error numbers"
+			" are described at\n"
+			"InnoDB: "
+			REFMAN "operating-system-error-codes.html\n",
+			name,  (long long)size_high,  (long long)size, errno);
+
+			return (FALSE);
+		}
+		return (TRUE);
+	}
+#endif
+
 	/* Write up to 1 megabyte at a time. */
 	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
 		* UNIV_PAGE_SIZE;
diff --git a/storage/xtradb/srv/srv0srv.c b/storage/xtradb/srv/srv0srv.c
index 1de5523f62c..6edfbaa7755 100644
--- a/storage/xtradb/srv/srv0srv.c
+++ b/storage/xtradb/srv/srv0srv.c
@@ -409,6 +409,10 @@ UNIV_INTERN ulong	srv_sys_stats_root_page = 0;
 #endif
 
 UNIV_INTERN ibool	srv_use_doublewrite_buf	= TRUE;
+UNIV_INTERN ibool	srv_use_atomic_writes = FALSE;
+#ifdef HAVE_POSIX_FALLOCATE
+UNIV_INTERN ibool	srv_use_posix_fallocate = TRUE;
+#endif
 UNIV_INTERN ibool	srv_use_checksums = TRUE;
 UNIV_INTERN ibool	srv_fast_checksum = FALSE;