Modify vdev_elevator_switch() to use elevator_change()

As of Linux 2.6.36 an elevator_change() interface was added.
This commit updates vdev_elevator_switch() to use this interface
when available, otherwise it falls back to the usermodehelper
method.

Original-patch-by: foobarz <sysop@xeon.(none)>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #906

3 files changed, 49 insertions, 18 deletions

diff --git a/config/kernel-elevator-change.m4 b/config/kernel-elevator-change.m4
new file mode 100644
index 000000000..90ab51b2e
--- /dev/null
+++ b/config/kernel-elevator-change.m4
@@ -0,0 +1,25 @@
+dnl #
+dnl # 2.6.36 API change
+dnl # Verify the elevator_change() symbol is available.
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_ELEVATOR_CHANGE], [
+	AC_MSG_CHECKING([whether elevator_change() is available])
+	tmp_flags="$EXTRA_KCFLAGS"
+	EXTRA_KCFLAGS="-Wno-unused-but-set-variable"
+	ZFS_LINUX_TRY_COMPILE([
+		#include <linux/blkdev.h>
+		#include <linux/elevator.h>
+	],[
+		int ret;
+		struct request_queue *q = NULL;
+		char *elevator = NULL;
+		ret = elevator_change(q, elevator);
+	],[
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_ELEVATOR_CHANGE, 1,
+			  [elevator_change() is available])
+	],[
+		AC_MSG_RESULT(no)
+	])
+	EXTRA_KCFLAGS="$tmp_flags"
+])
diff --git a/config/kernel.m4 b/config/kernel.m4
index 262dc7fcb..89b0a9806 100644
--- a/config/kernel.m4
+++ b/config/kernel.m4
@@ -67,6 +67,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [
 	ZFS_AC_KERNEL_BDI
 	ZFS_AC_KERNEL_BDI_SETUP_AND_REGISTER
 	ZFS_AC_KERNEL_SET_NLINK
+	ZFS_AC_KERNEL_ELEVATOR_CHANGE
 
 	AS_IF([test "$LINUX_OBJ" != "$LINUX"], [
 		KERNELMAKE_PARAMS="$KERNELMAKE_PARAMS O=$LINUX_OBJ"
diff --git a/module/zfs/vdev_disk.c b/module/zfs/vdev_disk.c
index ffb2980d2..c56225693 100644
--- a/module/zfs/vdev_disk.c
+++ b/module/zfs/vdev_disk.c
@@ -111,19 +111,7 @@ vdev_disk_error(zio_t *zio)
  * elevator to do the maximum front/back merging allowed by the
  * physical device.  This yields the largest possible requests for
  * the device with the lowest total overhead.
- *
- * Unfortunately we cannot directly call the elevator_switch() function
- * because it is not exported from the block layer.  This means we have
- * to use the sysfs interface and a user space upcall.  Pools will be
- * automatically imported on module load so we must do this at device
- * open time from the kernel.
  */
-#define SET_SCHEDULER_CMD \
-	"exec 0</dev/null " \
-	"     1>/sys/block/%s/queue/scheduler " \
-	"     2>/dev/null; " \
-	"echo %s"
-
 static int
 vdev_elevator_switch(vdev_t *v, char *elevator)
 {
@@ -131,8 +119,6 @@ vdev_elevator_switch(vdev_t *v, char *elevator)
 	struct block_device *bdev = vd->vd_bdev;
 	struct request_queue *q = bdev_get_queue(bdev);
 	char *device = bdev->bd_disk->disk_name;
-	char *argv[] = { "/bin/sh", "-c", NULL, NULL };
-	char *envp[] = { NULL };
 	int error;
 
 	/* Skip devices which are not whole disks (partitions) */
@@ -147,14 +133,33 @@ vdev_elevator_switch(vdev_t *v, char *elevator)
 	if (!strncmp(elevator, "none", 4) && (strlen(elevator) == 4))
 		return (0);
 
-	argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator);
-	error = call_usermodehelper(argv[0], argv, envp, 1);
+#ifdef HAVE_ELEVATOR_CHANGE
+	error = elevator_change(q, elevator);
+#else
+	/* For pre-2.6.36 kernels elevator_change() is not available.
+	 * Therefore we fall back to using a usermodehelper to echo the
+	 * elevator into sysfs;  This requires /bin/echo and sysfs to be
+	 * mounted which may not be true early in the boot process.
+	 */
+# define SET_SCHEDULER_CMD \
+	"exec 0</dev/null " \
+	"     1>/sys/block/%s/queue/scheduler " \
+	"     2>/dev/null; " \
+	"echo %s"
+
+	{
+		char *argv[] = { "/bin/sh", "-c", NULL, NULL };
+		char *envp[] = { NULL };
+
+		argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator);
+		error = call_usermodehelper(argv[0], argv, envp, 1);
+		strfree(argv[2]);
+	}
+#endif /* HAVE_ELEVATOR_CHANGE */
 	if (error)
 		printk("ZFS: Unable to set \"%s\" scheduler for %s (%s): %d\n",
 		       elevator, v->vdev_path, device, error);
 
-	strfree(argv[2]);
-
 	return (error);
 }