2010-08-26 22:45:02 +04:00
|
|
|
/*
|
|
|
|
* CDDL HEADER START
|
|
|
|
*
|
|
|
|
* The contents of this file are subject to the terms of the
|
|
|
|
* Common Development and Distribution License (the "License").
|
|
|
|
* You may not use this file except in compliance with the License.
|
|
|
|
*
|
|
|
|
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
|
|
|
|
* or http://www.opensolaris.org/os/licensing.
|
|
|
|
* See the License for the specific language governing permissions
|
|
|
|
* and limitations under the License.
|
|
|
|
*
|
|
|
|
* When distributing Covered Code, include this CDDL HEADER in each
|
|
|
|
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
|
|
|
* If applicable, add the following below this CDDL HEADER, with the
|
|
|
|
* fields enclosed by brackets "[]" replaced with your own identifying
|
|
|
|
* information: Portions Copyright [yyyy] [name of copyright owner]
|
|
|
|
*
|
|
|
|
* CDDL HEADER END
|
|
|
|
*/
|
2011-02-22 23:15:13 +03:00
|
|
|
|
2010-08-26 22:45:02 +04:00
|
|
|
/*
|
2011-02-22 23:15:13 +03:00
|
|
|
* Copyright (C) 2011 Lawrence Livermore National Security, LLC.
|
2010-08-26 22:45:02 +04:00
|
|
|
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
|
|
|
|
* Written by Brian Behlendorf <behlendorf1@llnl.gov>.
|
|
|
|
* LLNL-CODE-403049.
|
|
|
|
*/
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
#ifndef _ZFS_BLKDEV_H
|
2013-11-01 23:26:11 +04:00
|
|
|
#define _ZFS_BLKDEV_H
|
2010-08-26 22:45:02 +04:00
|
|
|
|
|
|
|
#include <linux/blkdev.h>
|
|
|
|
#include <linux/elevator.h>
|
|
|
|
|
|
|
|
#ifndef HAVE_FMODE_T
|
|
|
|
typedef unsigned __bitwise__ fmode_t;
|
|
|
|
#endif /* HAVE_FMODE_T */
|
|
|
|
|
2011-09-05 13:11:38 +04:00
|
|
|
/*
|
|
|
|
* 2.6.36 API change,
|
|
|
|
* The blk_queue_flush() interface has replaced blk_queue_ordered()
|
|
|
|
* interface. However, while the old interface was available to all the
|
|
|
|
* new one is GPL-only. Thus if the GPL-only version is detected we
|
|
|
|
* implement our own trivial helper compatibility funcion. The hope is
|
|
|
|
* that long term this function will be opened up.
|
|
|
|
*/
|
|
|
|
#if defined(HAVE_BLK_QUEUE_FLUSH) && defined(HAVE_BLK_QUEUE_FLUSH_GPL_ONLY)
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_queue_flush __blk_queue_flush
|
2011-09-05 13:11:38 +04:00
|
|
|
static inline void
|
|
|
|
__blk_queue_flush(struct request_queue *q, unsigned int flags)
|
|
|
|
{
|
|
|
|
q->flush_flags = flags & (REQ_FLUSH | REQ_FUA);
|
|
|
|
}
|
|
|
|
#endif /* HAVE_BLK_QUEUE_FLUSH && HAVE_BLK_QUEUE_FLUSH_GPL_ONLY */
|
2010-11-11 03:38:14 +03:00
|
|
|
/*
|
|
|
|
* Most of the blk_* macros were removed in 2.6.36. Ostensibly this was
|
|
|
|
* done to improve readability and allow easier grepping. However, from
|
|
|
|
* a portability stand point the macros are helpful. Therefore the needed
|
|
|
|
* macros are redefined here if they are missing from the kernel.
|
|
|
|
*/
|
|
|
|
#ifndef blk_fs_request
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_fs_request(rq) ((rq)->cmd_type == REQ_TYPE_FS)
|
2010-11-11 03:38:14 +03:00
|
|
|
#endif
|
|
|
|
|
2011-03-22 19:26:38 +03:00
|
|
|
/*
|
|
|
|
* 2.6.27 API change,
|
|
|
|
* The blk_queue_stackable() queue flag was added in 2.6.27 to handle dm
|
|
|
|
* stacking drivers. Prior to this request stacking drivers were detected
|
|
|
|
* by checking (q->request_fn == NULL), for earlier kernels we revert to
|
|
|
|
* this legacy behavior.
|
|
|
|
*/
|
|
|
|
#ifndef blk_queue_stackable
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_queue_stackable(q) ((q)->request_fn == NULL)
|
2011-03-22 19:26:38 +03:00
|
|
|
#endif
|
|
|
|
|
2011-09-05 17:15:45 +04:00
|
|
|
/*
|
|
|
|
* 2.6.34 API change,
|
|
|
|
* The blk_queue_max_hw_sectors() function replaces blk_queue_max_sectors().
|
|
|
|
*/
|
|
|
|
#ifndef HAVE_BLK_QUEUE_MAX_HW_SECTORS
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_queue_max_hw_sectors __blk_queue_max_hw_sectors
|
2011-09-05 17:15:45 +04:00
|
|
|
static inline void
|
|
|
|
__blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_sectors)
|
|
|
|
{
|
|
|
|
blk_queue_max_sectors(q, max_hw_sectors);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/*
|
|
|
|
* 2.6.34 API change,
|
|
|
|
* The blk_queue_max_segments() function consolidates
|
|
|
|
* blk_queue_max_hw_segments() and blk_queue_max_phys_segments().
|
|
|
|
*/
|
|
|
|
#ifndef HAVE_BLK_QUEUE_MAX_SEGMENTS
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_queue_max_segments __blk_queue_max_segments
|
2011-09-05 17:15:45 +04:00
|
|
|
static inline void
|
|
|
|
__blk_queue_max_segments(struct request_queue *q, unsigned short max_segments)
|
|
|
|
{
|
|
|
|
blk_queue_max_phys_segments(q, max_segments);
|
|
|
|
blk_queue_max_hw_segments(q, max_segments);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2010-08-26 22:45:02 +04:00
|
|
|
#ifndef HAVE_GET_DISK_RO
|
|
|
|
static inline int
|
|
|
|
get_disk_ro(struct gendisk *disk)
|
|
|
|
{
|
|
|
|
int policy = 0;
|
|
|
|
|
|
|
|
if (disk->part[0])
|
|
|
|
policy = disk->part[0]->policy;
|
|
|
|
|
2013-11-01 23:26:11 +04:00
|
|
|
return (policy);
|
2010-08-26 22:45:02 +04:00
|
|
|
}
|
|
|
|
#endif /* HAVE_GET_DISK_RO */
|
|
|
|
|
2014-03-28 11:08:21 +04:00
|
|
|
#ifdef HAVE_BIO_BVEC_ITER
|
|
|
|
#define BIO_BI_SECTOR(bio) (bio)->bi_iter.bi_sector
|
|
|
|
#define BIO_BI_SIZE(bio) (bio)->bi_iter.bi_size
|
|
|
|
#define BIO_BI_IDX(bio) (bio)->bi_iter.bi_idx
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#define bio_for_each_segment4(bv, bvp, b, i) \
|
|
|
|
bio_for_each_segment((bv), (b), (i))
|
|
|
|
typedef struct bvec_iter bvec_iterator_t;
|
2014-03-28 11:08:21 +04:00
|
|
|
#else
|
|
|
|
#define BIO_BI_SECTOR(bio) (bio)->bi_sector
|
|
|
|
#define BIO_BI_SIZE(bio) (bio)->bi_size
|
|
|
|
#define BIO_BI_IDX(bio) (bio)->bi_idx
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#define bio_for_each_segment4(bv, bvp, b, i) \
|
|
|
|
bio_for_each_segment((bvp), (b), (i))
|
|
|
|
typedef int bvec_iterator_t;
|
2014-03-28 11:08:21 +04:00
|
|
|
#endif
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
/*
|
|
|
|
* Portable helper for correctly setting the FAILFAST flags. The
|
|
|
|
* correct usage has changed 3 times from 2.6.12 to 2.6.38.
|
|
|
|
*/
|
2010-10-01 21:57:56 +04:00
|
|
|
static inline void
|
|
|
|
bio_set_flags_failfast(struct block_device *bdev, int *flags)
|
|
|
|
{
|
2010-11-11 01:40:38 +03:00
|
|
|
#ifdef CONFIG_BUG
|
2010-10-01 21:57:56 +04:00
|
|
|
/*
|
2010-11-11 01:40:38 +03:00
|
|
|
* Disable FAILFAST for loopback devices because of the
|
|
|
|
* following incorrect BUG_ON() in loop_make_request().
|
2010-10-01 21:57:56 +04:00
|
|
|
* This support is also disabled for md devices because the
|
|
|
|
* test suite layers md devices on top of loopback devices.
|
|
|
|
* This may be removed when the loopback driver is fixed.
|
|
|
|
*
|
|
|
|
* BUG_ON(!lo || (rw != READ && rw != WRITE));
|
|
|
|
*/
|
|
|
|
if ((MAJOR(bdev->bd_dev) == LOOP_MAJOR) ||
|
|
|
|
(MAJOR(bdev->bd_dev) == MD_MAJOR))
|
|
|
|
return;
|
|
|
|
|
|
|
|
#ifdef BLOCK_EXT_MAJOR
|
|
|
|
if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
|
|
|
|
return;
|
|
|
|
#endif /* BLOCK_EXT_MAJOR */
|
|
|
|
#endif /* CONFIG_BUG */
|
2010-11-11 01:40:38 +03:00
|
|
|
|
2015-05-08 23:49:56 +03:00
|
|
|
#if defined(HAVE_BIO_RW_FAILFAST_DTD)
|
2010-11-11 01:40:38 +03:00
|
|
|
/* BIO_RW_FAILFAST_* preferred interface from 2.6.28 - 2.6.35 */
|
2013-11-01 23:26:11 +04:00
|
|
|
*flags |= (
|
|
|
|
(1 << BIO_RW_FAILFAST_DEV) |
|
|
|
|
(1 << BIO_RW_FAILFAST_TRANSPORT) |
|
|
|
|
(1 << BIO_RW_FAILFAST_DRIVER));
|
2015-05-08 23:49:56 +03:00
|
|
|
#elif defined(HAVE_REQ_FAILFAST_MASK)
|
2013-11-01 23:26:11 +04:00
|
|
|
/*
|
|
|
|
* REQ_FAILFAST_* preferred interface from 2.6.36 - 2.6.xx,
|
|
|
|
* the BIO_* and REQ_* flags were unified under REQ_* flags.
|
|
|
|
*/
|
2010-11-11 01:40:38 +03:00
|
|
|
*flags |= REQ_FAILFAST_MASK;
|
2015-05-08 23:49:56 +03:00
|
|
|
#else
|
|
|
|
#error "Undefined block IO FAILFAST interface."
|
|
|
|
#endif
|
2010-10-01 21:57:56 +04:00
|
|
|
}
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
/*
|
|
|
|
* Maximum disk label length, it may be undefined for some kernels.
|
|
|
|
*/
|
2010-08-26 22:45:02 +04:00
|
|
|
#ifndef DISK_NAME_LEN
|
2013-11-01 23:26:11 +04:00
|
|
|
#define DISK_NAME_LEN 32
|
2010-08-26 22:45:02 +04:00
|
|
|
#endif /* DISK_NAME_LEN */
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
/*
|
|
|
|
* 2.6.24 API change,
|
|
|
|
* The bio_end_io() prototype changed slightly. These are helper
|
|
|
|
* macro's to ensure the prototype and return value are handled.
|
|
|
|
*/
|
|
|
|
#ifdef HAVE_2ARGS_BIO_END_IO_T
|
2013-11-01 23:26:11 +04:00
|
|
|
#define BIO_END_IO_PROTO(fn, x, y, z) static void fn(struct bio *x, int z)
|
|
|
|
#define BIO_END_IO_RETURN(rc) return
|
2011-02-22 23:15:13 +03:00
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define BIO_END_IO_PROTO(fn, x, y, z) static int fn( \
|
|
|
|
struct bio *x, \
|
|
|
|
unsigned int y, \
|
|
|
|
int z)
|
|
|
|
#define BIO_END_IO_RETURN(rc) return rc
|
2011-02-22 23:15:13 +03:00
|
|
|
#endif /* HAVE_2ARGS_BIO_END_IO_T */
|
|
|
|
|
|
|
|
/*
|
2011-02-23 01:55:35 +03:00
|
|
|
* 2.6.38 - 2.6.x API,
|
|
|
|
* blkdev_get_by_path()
|
|
|
|
* blkdev_put()
|
|
|
|
*
|
|
|
|
* 2.6.28 - 2.6.37 API,
|
|
|
|
* open_bdev_exclusive()
|
|
|
|
* close_bdev_exclusive()
|
|
|
|
*
|
|
|
|
* 2.6.12 - 2.6.27 API,
|
|
|
|
* open_bdev_excl()
|
|
|
|
* close_bdev_excl()
|
|
|
|
*
|
2011-02-22 23:15:13 +03:00
|
|
|
* Used to exclusively open a block device from within the kernel.
|
|
|
|
*/
|
2011-02-23 01:55:35 +03:00
|
|
|
#if defined(HAVE_BLKDEV_GET_BY_PATH)
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_open(path, md, hld) blkdev_get_by_path(path, \
|
2011-02-23 01:55:35 +03:00
|
|
|
(md) | FMODE_EXCL, hld)
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_close(bdev, md) blkdev_put(bdev, (md) | FMODE_EXCL)
|
2011-02-23 01:55:35 +03:00
|
|
|
#elif defined(HAVE_OPEN_BDEV_EXCLUSIVE)
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_open(path, md, hld) open_bdev_exclusive(path, md, hld)
|
|
|
|
#define vdev_bdev_close(bdev, md) close_bdev_exclusive(bdev, md)
|
2011-02-22 23:15:13 +03:00
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_open(path, md, hld) open_bdev_excl(path, md, hld)
|
|
|
|
#define vdev_bdev_close(bdev, md) close_bdev_excl(bdev)
|
2011-02-23 01:55:35 +03:00
|
|
|
#endif /* HAVE_BLKDEV_GET_BY_PATH | HAVE_OPEN_BDEV_EXCLUSIVE */
|
2011-02-22 23:15:13 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* 2.6.22 API change
|
|
|
|
* The function invalidate_bdev() lost it's second argument because
|
|
|
|
* it was unused.
|
|
|
|
*/
|
|
|
|
#ifdef HAVE_1ARG_INVALIDATE_BDEV
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev)
|
2011-02-22 23:15:13 +03:00
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_invalidate(bdev) invalidate_bdev(bdev, 1)
|
2011-02-22 23:15:13 +03:00
|
|
|
#endif /* HAVE_1ARG_INVALIDATE_BDEV */
|
|
|
|
|
2013-01-29 02:15:39 +04:00
|
|
|
/*
|
|
|
|
* 2.6.27 API change
|
|
|
|
* The function was exported for use, prior to this it existed by the
|
|
|
|
* symbol was not exported.
|
|
|
|
*/
|
|
|
|
#ifndef HAVE_LOOKUP_BDEV
|
2013-11-01 23:26:11 +04:00
|
|
|
#define lookup_bdev(path) ERR_PTR(-ENOTSUP)
|
2013-01-29 02:15:39 +04:00
|
|
|
#endif
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
/*
|
|
|
|
* 2.6.30 API change
|
2012-09-03 03:34:12 +04:00
|
|
|
* To ensure good performance preferentially use the physical block size
|
|
|
|
* for proper alignment. The physical size is supposed to be the internal
|
|
|
|
* sector size used by the device. This is often 4096 byte for AF devices,
|
|
|
|
* while a smaller 512 byte logical size is supported for compatibility.
|
|
|
|
*
|
|
|
|
* Unfortunately, many drives still misreport their physical sector size.
|
|
|
|
* For devices which are known to lie you may need to manually set this
|
|
|
|
* at pool creation time with 'zpool create -o ashift=12 ...'.
|
|
|
|
*
|
|
|
|
* When the physical block size interface isn't available, we fall back to
|
|
|
|
* the logical block size interface and then the older hard sector size.
|
2011-02-22 23:15:13 +03:00
|
|
|
*/
|
2012-09-03 03:34:12 +04:00
|
|
|
#ifdef HAVE_BDEV_PHYSICAL_BLOCK_SIZE
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_block_size(bdev) bdev_physical_block_size(bdev)
|
|
|
|
#else
|
|
|
|
#ifdef HAVE_BDEV_LOGICAL_BLOCK_SIZE
|
|
|
|
#define vdev_bdev_block_size(bdev) bdev_logical_block_size(bdev)
|
2011-02-22 23:15:13 +03:00
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define vdev_bdev_block_size(bdev) bdev_hardsect_size(bdev)
|
|
|
|
#endif /* HAVE_BDEV_LOGICAL_BLOCK_SIZE */
|
2012-09-03 03:34:12 +04:00
|
|
|
#endif /* HAVE_BDEV_PHYSICAL_BLOCK_SIZE */
|
2011-02-22 23:15:13 +03:00
|
|
|
|
2011-06-16 22:20:22 +04:00
|
|
|
/*
|
|
|
|
* 2.6.37 API change
|
|
|
|
* The WRITE_FLUSH, WRITE_FUA, and WRITE_FLUSH_FUA flags have been
|
|
|
|
* introduced as a replacement for WRITE_BARRIER. This was done to
|
|
|
|
* allow richer semantics to be expressed to the block layer. It is
|
|
|
|
* the block layers responsibility to choose the correct way to
|
|
|
|
* implement these semantics.
|
2011-09-05 13:11:38 +04:00
|
|
|
*
|
|
|
|
* The existence of these flags implies that REQ_FLUSH an REQ_FUA are
|
|
|
|
* defined. Thus we can safely define VDEV_REQ_FLUSH and VDEV_REQ_FUA
|
|
|
|
* compatibility macros.
|
2011-06-16 22:20:22 +04:00
|
|
|
*/
|
|
|
|
#ifdef WRITE_FLUSH_FUA
|
2013-11-01 23:26:11 +04:00
|
|
|
#define VDEV_WRITE_FLUSH_FUA WRITE_FLUSH_FUA
|
|
|
|
#define VDEV_REQ_FLUSH REQ_FLUSH
|
|
|
|
#define VDEV_REQ_FUA REQ_FUA
|
2011-06-16 22:20:22 +04:00
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define VDEV_WRITE_FLUSH_FUA WRITE_BARRIER
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#ifdef HAVE_BIO_RW_BARRIER
|
|
|
|
#define VDEV_REQ_FLUSH (1 << BIO_RW_BARRIER)
|
|
|
|
#define VDEV_REQ_FUA (1 << BIO_RW_BARRIER)
|
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define VDEV_REQ_FLUSH REQ_HARDBARRIER
|
2015-08-31 18:36:32 +03:00
|
|
|
#define VDEV_REQ_FUA REQ_FUA
|
2011-06-16 22:20:22 +04:00
|
|
|
#endif
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#endif
|
2011-06-16 22:20:22 +04:00
|
|
|
|
2011-09-02 17:23:12 +04:00
|
|
|
/*
|
|
|
|
* 2.6.32 API change
|
|
|
|
* Use the normal I/O patch for discards.
|
|
|
|
*/
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#ifdef QUEUE_FLAG_DISCARD
|
|
|
|
#ifdef HAVE_BIO_RW_DISCARD
|
|
|
|
#define VDEV_REQ_DISCARD (1 << BIO_RW_DISCARD)
|
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define VDEV_REQ_DISCARD REQ_DISCARD
|
2011-09-02 17:23:12 +04:00
|
|
|
#endif
|
zvol processing should use struct bio
Internally, zvols are files exposed through the block device API. This
is intended to reduce overhead when things require block devices.
However, the ZoL zvol code emulates a traditional block device in that
it has a top half and a bottom half. This is an unnecessary source of
overhead that does not exist on any other OpenZFS platform does this.
This patch removes it. Early users of this patch reported double digit
performance gains in IOPS on zvols in the range of 50% to 80%.
Comments in the code suggest that the current implementation was done to
obtain IO merging from Linux's IO elevator. However, the DMU already
does write merging while arc_read() should implicitly merge read IOs
because only 1 thread is permitted to fetch the buffer into ARC. In
addition, commercial ZFSOnLinux distributions report that regular files
are more performant than zvols under the current implementation, and the
main consumers of zvols are VMs and iSCSI targets, which have their own
elevators to merge IOs.
Some minor refactoring allows us to register zfs_request() as our
->make_request() handler in place of the generic_make_request()
function. This eliminates the layer of code that broke IO requests on
zvols into a top half and a bottom half. This has several benefits:
1. No per zvol spinlocks.
2. No redundant IO elevator processing.
3. Interrupts are disabled only when actually necessary.
4. No redispatching of IOs when all taskq threads are busy.
5. Linux's page out routines will properly block.
6. Many autotools checks become obsolete.
An unfortunate consequence of eliminating the layer that
generic_make_request() is that we no longer calls the instrumentation
hooks for block IO accounting. Those hooks are GPL-exported, so we
cannot call them ourselves and consequently, we lose the ability to do
IO monitoring via iostat. Since zvols are internally files mapped as
block devices, this should be okay. Anyone who is willing to accept the
performance penalty for the block IO layer's accounting could use the
loop device in between the zvol and its consumer. Alternatively, perf
and ftrace likely could be used. Also, tools like latencytop will still
work. Tools such as latencytop sometimes provide a better view of
performance bottlenecks than the traditional block IO accounting tools
do.
Lastly, if direct reclaim occurs during spacemap loading and swap is on
a zvol, this code will deadlock. That deadlock could already occur with
sync=always on zvols. Given that swap on zvols is not yet production
ready, this is not a blocker.
Signed-off-by: Richard Yao <ryao@gentoo.org>
2014-07-05 02:43:47 +04:00
|
|
|
#else
|
|
|
|
#error "Allowing the build will cause discard requests to become writes "
|
|
|
|
"potentially triggering the DMU_MAX_ACCESS assertion. Please file a "
|
|
|
|
"an issue report at: https://github.com/zfsonlinux/zfs/issues/new"
|
|
|
|
#endif
|
2011-09-02 17:23:12 +04:00
|
|
|
|
2012-08-01 12:29:59 +04:00
|
|
|
/*
|
|
|
|
* 2.6.33 API change
|
|
|
|
* Discard granularity and alignment restrictions may now be set. For
|
|
|
|
* older kernels which do not support this it is safe to skip it.
|
|
|
|
*/
|
|
|
|
#ifdef HAVE_DISCARD_GRANULARITY
|
|
|
|
static inline void
|
|
|
|
blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
|
|
|
|
{
|
|
|
|
q->limits.discard_granularity = dg;
|
|
|
|
}
|
|
|
|
#else
|
2013-11-01 23:26:11 +04:00
|
|
|
#define blk_queue_discard_granularity(x, dg) ((void)0)
|
2012-08-01 12:29:59 +04:00
|
|
|
#endif /* HAVE_DISCARD_GRANULARITY */
|
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
/*
|
|
|
|
* 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"
|
2010-08-26 22:45:02 +04:00
|
|
|
|
2013-02-27 05:02:27 +04:00
|
|
|
/*
|
|
|
|
* 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
|
|
|
|
* allow them to open the device multiple times. Other kernel callers and
|
|
|
|
* user space processes which don't pass this value will get EBUSY. This is
|
|
|
|
* currently required for the correct operation of hot spares.
|
|
|
|
*/
|
2013-11-01 23:26:11 +04:00
|
|
|
#define VDEV_HOLDER ((void *)0x2401de7)
|
2013-02-27 05:02:27 +04:00
|
|
|
|
2011-02-22 23:15:13 +03:00
|
|
|
#endif /* _ZFS_BLKDEV_H */
|