Archive-Team
/
zfs
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 103 Wiki Activity

Remove zfs_vdev_elevator module option 

As described in commit f81d5ef6 the zfs_vdev_elevator module
option is being removed.  Users who require this functionality
should update their systems to set the disk scheduler using a
udev rule.

Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #8664
Closes #9417
Closes #9609
This commit is contained in:
Brian Behlendorf 2019-11-27 10:35:49 -08:00 committed by GitHub
parent 3c819a2c7d
commit 9e17e6f254
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 19 additions and 161 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
config/kernel-elevator-change.m4
View File

@ -1,26 +0,0 @@
dnl #
dnl # 2.6.36 API, exported elevator_change() symbol
dnl # 4.12 API, removed elevator_change() symbol
dnl #
AC_DEFUN([ZFS_AC_KERNEL_SRC_ELEVATOR_CHANGE], [
ZFS_LINUX_TEST_SRC([elevator_change], [
#include <linux/blkdev.h>
#include <linux/elevator.h>
],[
struct request_queue *q = NULL;
char *elevator = NULL;
int error __attribute__ ((unused)) =
elevator_change(q, elevator);
])
])
AC_DEFUN([ZFS_AC_KERNEL_ELEVATOR_CHANGE], [
AC_MSG_CHECKING([whether elevator_change() is available])
ZFS_LINUX_TEST_RESULT([elevator_change], [
AC_MSG_RESULT(yes)
AC_DEFINE(HAVE_ELEVATOR_CHANGE, 1,
[elevator_change() is available])
],[
AC_MSG_RESULT(no)
])
])

2
config/kernel.m4
View File

@ -92,7 +92,6 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
ZFS_AC_KERNEL_SRC_FST_MOUNT ZFS_AC_KERNEL_SRC_FST_MOUNT
ZFS_AC_KERNEL_SRC_BDI ZFS_AC_KERNEL_SRC_BDI
ZFS_AC_KERNEL_SRC_SET_NLINK ZFS_AC_KERNEL_SRC_SET_NLINK
ZFS_AC_KERNEL_SRC_ELEVATOR_CHANGE
ZFS_AC_KERNEL_SRC_SGET ZFS_AC_KERNEL_SRC_SGET
ZFS_AC_KERNEL_SRC_LSEEK_EXECUTE ZFS_AC_KERNEL_SRC_LSEEK_EXECUTE
ZFS_AC_KERNEL_SRC_VFS_GETATTR ZFS_AC_KERNEL_SRC_VFS_GETATTR
@ -186,7 +185,6 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
ZFS_AC_KERNEL_FST_MOUNT ZFS_AC_KERNEL_FST_MOUNT
ZFS_AC_KERNEL_BDI ZFS_AC_KERNEL_BDI
ZFS_AC_KERNEL_SET_NLINK ZFS_AC_KERNEL_SET_NLINK
ZFS_AC_KERNEL_ELEVATOR_CHANGE
ZFS_AC_KERNEL_SGET ZFS_AC_KERNEL_SGET
ZFS_AC_KERNEL_LSEEK_EXECUTE ZFS_AC_KERNEL_LSEEK_EXECUTE
ZFS_AC_KERNEL_VFS_GETATTR ZFS_AC_KERNEL_VFS_GETATTR

8
include/os/linux/kernel/linux/blkdev_compat.h
View File

@ -456,14 +456,6 @@ blk_queue_discard_secure(struct request_queue *q)
#endif #endif
} }
/*
* Default Linux IO Scheduler,
* Setting the scheduler to noop will allow the Linux IO scheduler to
* still perform front and back merging, while leaving the request
* ordering and prioritization to the ZFS IO scheduler.
*/
#define VDEV_SCHEDULER "noop"
/* /*
* A common holder for vdev_bdev_open() is used to relax the exclusive open * A common holder for vdev_bdev_open() is used to relax the exclusive open
* semantics slightly. Internal vdev disk callers may pass VDEV_HOLDER to * semantics slightly. Internal vdev disk callers may pass VDEV_HOLDER to

14
man/man5/zfs-module-parameters.5
View File

@ -3182,20 +3182,6 @@ threshold.
Default value: \fB32,768\fR. Default value: \fB32,768\fR.
.RE .RE
.sp
.ne 2
.na
\fBzfs_vdev_scheduler\fR (charp)
.ad
.RS 12n
Set the Linux I/O scheduler on whole disk vdevs to this scheduler. This
option has been deprecated and will be removed in a future release. The
standard \fB/sys/block/<block>/queue/scheduler\fR interface should be used
to set a block device scheduler.
.sp
Default value: \fBnoop\fR.
.RE
.sp .sp
.ne 2 .ne 2
.na .na

130
module/os/linux/zfs/vdev_disk.c
View File

@ -37,7 +37,6 @@
#include <linux/msdos_fs.h> #include <linux/msdos_fs.h>
#include <linux/vfs_compat.h> #include <linux/vfs_compat.h>
char *zfs_vdev_scheduler = VDEV_SCHEDULER;
static void *zfs_vdev_holder = VDEV_HOLDER; static void *zfs_vdev_holder = VDEV_HOLDER;
/* size of the "reserved" partition, in blocks */ /* size of the "reserved" partition, in blocks */
@ -141,75 +140,6 @@ vdev_disk_error(zio_t *zio)
zio->io_flags); zio->io_flags);
} }
/*
* Use the Linux 'noop' elevator for zfs managed block devices. This
* strikes the ideal balance by allowing the zfs elevator to do all
* request ordering and prioritization. While allowing the Linux
* 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.
*/
static void
vdev_elevator_switch(vdev_t *v, char *elevator)
{
vdev_disk_t *vd = v->vdev_tsd;
struct request_queue *q;
char *device;
int error;
for (int c = 0; c < v->vdev_children; c++)
vdev_elevator_switch(v->vdev_child[c], elevator);
if (!v->vdev_ops->vdev_op_leaf || vd->vd_bdev == NULL)
return;
q = bdev_get_queue(vd->vd_bdev);
device = vd->vd_bdev->bd_disk->disk_name;
/*
* Skip devices which are not whole disks (partitions).
* Device-mapper devices are excepted since they may be whole
* disks despite the vdev_wholedisk flag, in which case we can
* and should switch the elevator. If the device-mapper device
* does not have an elevator (i.e. dm-raid, dm-crypt, etc.) the
* "Skip devices without schedulers" check below will fail.
*/
if (!v->vdev_wholedisk && strncmp(device, "dm-", 3) != 0)
return;
/* Leave existing scheduler when set to "none" */
if ((strncmp(elevator, "none", 4) == 0) && (strlen(elevator) == 4))
return;
/*
* The elevator_change() function was available in kernels from
* 2.6.36 to 4.11. When not available fall back to using the user
* mode helper functionality to set the elevator via sysfs. This
* requires /bin/echo and sysfs to be mounted which may not be true
* early in the boot process.
*/
#ifdef HAVE_ELEVATOR_CHANGE
error = elevator_change(q, elevator);
#else
#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, UMH_NO_WAIT);
kmem_strfree(argv[2]);
#endif /* HAVE_ELEVATOR_CHANGE */
if (error) {
zfs_dbgmsg("Unable to set \"%s\" scheduler for %s (%s): %d",
elevator, v->vdev_path, device, error);
}
}
static int static int
vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize, vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift) uint64_t *ashift)
@ -343,9 +273,6 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
/* Based on the minimum sector size set the block size */ /* Based on the minimum sector size set the block size */
*ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1; *ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
/* Try to set the io scheduler elevator algorithm */
(void) vdev_elevator_switch(v, zfs_vdev_scheduler);
return (0); return (0);
} }
@ -857,44 +784,6 @@ vdev_disk_rele(vdev_t *vd)
/* XXX: Implement me as a vnode rele for the device */ /* XXX: Implement me as a vnode rele for the device */
} }
static int
param_set_vdev_scheduler(const char *val, zfs_kernel_param_t *kp)
{
spa_t *spa = NULL;
char *p;
if (val == NULL)
return (SET_ERROR(-EINVAL));
if ((p = strchr(val, '\n')) != NULL)
*p = '\0';
if (spa_mode_global != SPA_MODE_UNINIT) {
mutex_enter(&spa_namespace_lock);
while ((spa = spa_next(spa)) != NULL) {
if (spa_state(spa) != POOL_STATE_ACTIVE ||
!spa_writeable(spa) || spa_suspended(spa))
continue;
spa_open_ref(spa, FTAG);
mutex_exit(&spa_namespace_lock);
vdev_elevator_switch(spa->spa_root_vdev, (char *)val);
mutex_enter(&spa_namespace_lock);
spa_close(spa, FTAG);
}
mutex_exit(&spa_namespace_lock);
}
int error = param_set_charp(val, kp);
if (error == 0) {
printk(KERN_INFO "The 'zfs_vdev_scheduler' module option "
"will be removed in a future release.\n");
}
return (error);
}
vdev_ops_t vdev_disk_ops = { vdev_ops_t vdev_disk_ops = {
.vdev_op_open = vdev_disk_open, .vdev_op_open = vdev_disk_open,
.vdev_op_close = vdev_disk_close, .vdev_op_close = vdev_disk_close,
@ -911,6 +800,25 @@ vdev_ops_t vdev_disk_ops = {
.vdev_op_leaf = B_TRUE /* leaf vdev */ .vdev_op_leaf = B_TRUE /* leaf vdev */
}; };
/*
* The zfs_vdev_scheduler module option has been deprecated. Setting this
* value no longer has any effect. It has not yet been entirely removed
* to allow the module to be loaded if this option is specified in the
* /etc/modprobe.d/zfs.conf file. The following warning will be logged.
*/
static int
param_set_vdev_scheduler(const char *val, zfs_kernel_param_t *kp)
{
int error = param_set_charp(val, kp);
if (error == 0) {
printk(KERN_INFO "The 'zfs_vdev_scheduler' module option "
"is not supported.\n");
}
return (error);
}
char *zfs_vdev_scheduler = "unused";
module_param_call(zfs_vdev_scheduler, param_set_vdev_scheduler, module_param_call(zfs_vdev_scheduler, param_set_vdev_scheduler,
param_get_charp, &zfs_vdev_scheduler, 0644); param_get_charp, &zfs_vdev_scheduler, 0644);
MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler"); MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler");