4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
26 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
31 #include <sys/vdev_disk.h>
32 #include <sys/vdev_impl.h>
34 #include <sys/fs/zfs.h>
36 #include <linux/mod_compat.h>
37 #include <linux/msdos_fs.h>
39 char *zfs_vdev_scheduler
= VDEV_SCHEDULER
;
40 static void *zfs_vdev_holder
= VDEV_HOLDER
;
42 /* size of the "reserved" partition, in blocks */
43 #define EFI_MIN_RESV_SIZE (16 * 1024)
46 * Virtual device vector for disks.
48 typedef struct dio_request
{
49 zio_t
*dr_zio
; /* Parent ZIO */
50 atomic_t dr_ref
; /* References */
51 int dr_error
; /* Bio error */
52 int dr_bio_count
; /* Count of bio's */
53 struct bio
*dr_bio
[0]; /* Attached bio's */
57 #ifdef HAVE_OPEN_BDEV_EXCLUSIVE
59 vdev_bdev_mode(int smode
)
63 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
75 vdev_bdev_mode(int smode
)
79 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
81 if ((smode
& FREAD
) && !(smode
& FWRITE
))
86 #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
89 * Returns the usable capacity (in bytes) for the partition or disk.
92 bdev_capacity(struct block_device
*bdev
)
94 return (i_size_read(bdev
->bd_inode
));
98 * Returns the maximum expansion capacity of the block device (in bytes).
100 * It is possible to expand a vdev when it has been created as a wholedisk
101 * and the containing block device has increased in capacity. Or when the
102 * partition containing the pool has been manually increased in size.
104 * This function is only responsible for calculating the potential expansion
105 * size so it can be reported by 'zpool list'. The efi_use_whole_disk() is
106 * responsible for verifying the expected partition layout in the wholedisk
107 * case, and updating the partition table if appropriate. Once the partition
108 * size has been increased the additional capacity will be visible using
112 bdev_max_capacity(struct block_device
*bdev
, uint64_t wholedisk
)
117 if (wholedisk
&& bdev
->bd_part
!= NULL
&& bdev
!= bdev
->bd_contains
) {
119 * When reporting maximum expansion capacity for a wholedisk
120 * deduct any capacity which is expected to be lost due to
121 * alignment restrictions. Over reporting this value isn't
122 * harmful and would only result in slightly less capacity
123 * than expected post expansion.
125 available
= i_size_read(bdev
->bd_contains
->bd_inode
) -
126 ((EFI_MIN_RESV_SIZE
+ NEW_START_BLOCK
+
127 PARTITION_END_ALIGNMENT
) << SECTOR_BITS
);
131 psize
= bdev_capacity(bdev
);
133 psize
= bdev_capacity(bdev
);
140 vdev_disk_error(zio_t
*zio
)
142 zfs_dbgmsg("zio error=%d type=%d offset=%llu size=%llu flags=%x\n",
143 zio
->io_error
, zio
->io_type
, (u_longlong_t
)zio
->io_offset
,
144 (u_longlong_t
)zio
->io_size
, zio
->io_flags
);
148 * Use the Linux 'noop' elevator for zfs managed block devices. This
149 * strikes the ideal balance by allowing the zfs elevator to do all
150 * request ordering and prioritization. While allowing the Linux
151 * elevator to do the maximum front/back merging allowed by the
152 * physical device. This yields the largest possible requests for
153 * the device with the lowest total overhead.
156 vdev_elevator_switch(vdev_t
*v
, char *elevator
)
158 vdev_disk_t
*vd
= v
->vdev_tsd
;
159 struct request_queue
*q
;
163 for (int c
= 0; c
< v
->vdev_children
; c
++)
164 vdev_elevator_switch(v
->vdev_child
[c
], elevator
);
166 if (!v
->vdev_ops
->vdev_op_leaf
|| vd
->vd_bdev
== NULL
)
169 q
= bdev_get_queue(vd
->vd_bdev
);
170 device
= vd
->vd_bdev
->bd_disk
->disk_name
;
173 * Skip devices which are not whole disks (partitions).
174 * Device-mapper devices are excepted since they may be whole
175 * disks despite the vdev_wholedisk flag, in which case we can
176 * and should switch the elevator. If the device-mapper device
177 * does not have an elevator (i.e. dm-raid, dm-crypt, etc.) the
178 * "Skip devices without schedulers" check below will fail.
180 if (!v
->vdev_wholedisk
&& strncmp(device
, "dm-", 3) != 0)
183 /* Leave existing scheduler when set to "none" */
184 if ((strncmp(elevator
, "none", 4) == 0) && (strlen(elevator
) == 4))
188 * The elevator_change() function was available in kernels from
189 * 2.6.36 to 4.11. When not available fall back to using the user
190 * mode helper functionality to set the elevator via sysfs. This
191 * requires /bin/echo and sysfs to be mounted which may not be true
192 * early in the boot process.
194 #ifdef HAVE_ELEVATOR_CHANGE
195 error
= elevator_change(q
, elevator
);
197 #define SET_SCHEDULER_CMD \
198 "exec 0</dev/null " \
199 " 1>/sys/block/%s/queue/scheduler " \
203 char *argv
[] = { "/bin/sh", "-c", NULL
, NULL
};
204 char *envp
[] = { NULL
};
206 argv
[2] = kmem_asprintf(SET_SCHEDULER_CMD
, device
, elevator
);
207 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
209 #endif /* HAVE_ELEVATOR_CHANGE */
211 zfs_dbgmsg("Unable to set \"%s\" scheduler for %s (%s): %d\n",
212 elevator
, v
->vdev_path
, device
, error
);
217 vdev_disk_open(vdev_t
*v
, uint64_t *psize
, uint64_t *max_psize
,
220 struct block_device
*bdev
;
221 fmode_t mode
= vdev_bdev_mode(spa_mode(v
->vdev_spa
));
222 int count
= 0, block_size
;
223 int bdev_retry_count
= 50;
226 /* Must have a pathname and it must be absolute. */
227 if (v
->vdev_path
== NULL
|| v
->vdev_path
[0] != '/') {
228 v
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
229 vdev_dbgmsg(v
, "invalid vdev_path");
230 return (SET_ERROR(EINVAL
));
234 * Reopen the device if it is currently open. When expanding a
235 * partition force re-scanning the partition table while closed
236 * in order to get an accurate updated block device size. Then
237 * since udev may need to recreate the device links increase the
238 * open retry count before reporting the device as unavailable.
242 char disk_name
[BDEVNAME_SIZE
+ 6] = "/dev/";
243 boolean_t reread_part
= B_FALSE
;
245 rw_enter(&vd
->vd_lock
, RW_WRITER
);
250 if (v
->vdev_expanding
&& bdev
!= bdev
->bd_contains
) {
251 bdevname(bdev
->bd_contains
, disk_name
+ 5);
252 reread_part
= B_TRUE
;
255 vdev_bdev_close(bdev
, mode
);
259 bdev
= vdev_bdev_open(disk_name
, mode
, zfs_vdev_holder
);
261 int error
= vdev_bdev_reread_part(bdev
);
262 vdev_bdev_close(bdev
, mode
);
264 bdev_retry_count
= 100;
268 vd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
270 rw_init(&vd
->vd_lock
, NULL
, RW_DEFAULT
, NULL
);
271 rw_enter(&vd
->vd_lock
, RW_WRITER
);
275 * Devices are always opened by the path provided at configuration
276 * time. This means that if the provided path is a udev by-id path
277 * then drives may be re-cabled without an issue. If the provided
278 * path is a udev by-path path, then the physical location information
279 * will be preserved. This can be critical for more complicated
280 * configurations where drives are located in specific physical
281 * locations to maximize the systems tolerance to component failure.
283 * Alternatively, you can provide your own udev rule to flexibly map
284 * the drives as you see fit. It is not advised that you use the
285 * /dev/[hd]d devices which may be reordered due to probing order.
286 * Devices in the wrong locations will be detected by the higher
287 * level vdev validation.
289 * The specified paths may be briefly removed and recreated in
290 * response to udev events. This should be exceptionally unlikely
291 * because the zpool command makes every effort to verify these paths
292 * have already settled prior to reaching this point. Therefore,
293 * a ENOENT failure at this point is highly likely to be transient
294 * and it is reasonable to sleep and retry before giving up. In
295 * practice delays have been observed to be on the order of 100ms.
297 bdev
= ERR_PTR(-ENXIO
);
298 while (IS_ERR(bdev
) && count
< bdev_retry_count
) {
299 bdev
= vdev_bdev_open(v
->vdev_path
, mode
, zfs_vdev_holder
);
300 if (unlikely(PTR_ERR(bdev
) == -ENOENT
)) {
301 schedule_timeout(MSEC_TO_TICK(10));
303 } else if (IS_ERR(bdev
)) {
309 int error
= -PTR_ERR(bdev
);
310 vdev_dbgmsg(v
, "open error=%d count=%d\n", error
, count
);
313 rw_exit(&vd
->vd_lock
);
314 return (SET_ERROR(error
));
318 rw_exit(&vd
->vd_lock
);
321 /* Determine the physical block size */
322 block_size
= vdev_bdev_block_size(vd
->vd_bdev
);
324 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
325 v
->vdev_nowritecache
= B_FALSE
;
327 /* Inform the ZIO pipeline that we are non-rotational */
328 v
->vdev_nonrot
= blk_queue_nonrot(bdev_get_queue(vd
->vd_bdev
));
330 /* Physical volume size in bytes for the partition */
331 *psize
= bdev_capacity(vd
->vd_bdev
);
333 /* Physical volume size in bytes including possible expansion space */
334 *max_psize
= bdev_max_capacity(vd
->vd_bdev
, v
->vdev_wholedisk
);
336 /* Based on the minimum sector size set the block size */
337 *ashift
= highbit64(MAX(block_size
, SPA_MINBLOCKSIZE
)) - 1;
339 /* Try to set the io scheduler elevator algorithm */
340 (void) vdev_elevator_switch(v
, zfs_vdev_scheduler
);
346 vdev_disk_close(vdev_t
*v
)
348 vdev_disk_t
*vd
= v
->vdev_tsd
;
350 if (v
->vdev_reopening
|| vd
== NULL
)
353 if (vd
->vd_bdev
!= NULL
) {
354 vdev_bdev_close(vd
->vd_bdev
,
355 vdev_bdev_mode(spa_mode(v
->vdev_spa
)));
358 rw_destroy(&vd
->vd_lock
);
359 kmem_free(vd
, sizeof (vdev_disk_t
));
363 static dio_request_t
*
364 vdev_disk_dio_alloc(int bio_count
)
369 dr
= kmem_zalloc(sizeof (dio_request_t
) +
370 sizeof (struct bio
*) * bio_count
, KM_SLEEP
);
372 atomic_set(&dr
->dr_ref
, 0);
373 dr
->dr_bio_count
= bio_count
;
376 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
377 dr
->dr_bio
[i
] = NULL
;
384 vdev_disk_dio_free(dio_request_t
*dr
)
388 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
390 bio_put(dr
->dr_bio
[i
]);
392 kmem_free(dr
, sizeof (dio_request_t
) +
393 sizeof (struct bio
*) * dr
->dr_bio_count
);
397 vdev_disk_dio_get(dio_request_t
*dr
)
399 atomic_inc(&dr
->dr_ref
);
403 vdev_disk_dio_put(dio_request_t
*dr
)
405 int rc
= atomic_dec_return(&dr
->dr_ref
);
408 * Free the dio_request when the last reference is dropped and
409 * ensure zio_interpret is called only once with the correct zio
412 zio_t
*zio
= dr
->dr_zio
;
413 int error
= dr
->dr_error
;
415 vdev_disk_dio_free(dr
);
418 zio
->io_error
= error
;
419 ASSERT3S(zio
->io_error
, >=, 0);
421 vdev_disk_error(zio
);
423 zio_delay_interrupt(zio
);
430 BIO_END_IO_PROTO(vdev_disk_physio_completion
, bio
, error
)
432 dio_request_t
*dr
= bio
->bi_private
;
435 if (dr
->dr_error
== 0) {
436 #ifdef HAVE_1ARG_BIO_END_IO_T
437 dr
->dr_error
= BIO_END_IO_ERROR(bio
);
440 dr
->dr_error
= -(error
);
441 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
446 /* Drop reference acquired by __vdev_disk_physio */
447 rc
= vdev_disk_dio_put(dr
);
451 bio_map(struct bio
*bio
, void *bio_ptr
, unsigned int bio_size
)
453 unsigned int offset
, size
, i
;
456 offset
= offset_in_page(bio_ptr
);
457 for (i
= 0; i
< bio
->bi_max_vecs
; i
++) {
458 size
= PAGE_SIZE
- offset
;
466 if (is_vmalloc_addr(bio_ptr
))
467 page
= vmalloc_to_page(bio_ptr
);
469 page
= virt_to_page(bio_ptr
);
472 * Some network related block device uses tcp_sendpage, which
473 * doesn't behave well when using 0-count page, this is a
474 * safety net to catch them.
476 ASSERT3S(page_count(page
), >, 0);
478 if (bio_add_page(bio
, page
, size
, offset
) != size
)
490 bio_map_abd_off(struct bio
*bio
, abd_t
*abd
, unsigned int size
, size_t off
)
492 if (abd_is_linear(abd
))
493 return (bio_map(bio
, ((char *)abd_to_buf(abd
)) + off
, size
));
495 return (abd_scatter_bio_map_off(bio
, abd
, size
, off
));
499 vdev_submit_bio_impl(struct bio
*bio
)
501 #ifdef HAVE_1ARG_SUBMIT_BIO
508 #ifndef HAVE_BIO_SET_DEV
510 bio_set_dev(struct bio
*bio
, struct block_device
*bdev
)
514 #endif /* !HAVE_BIO_SET_DEV */
517 vdev_submit_bio(struct bio
*bio
)
519 #ifdef HAVE_CURRENT_BIO_TAIL
520 struct bio
**bio_tail
= current
->bio_tail
;
521 current
->bio_tail
= NULL
;
522 vdev_submit_bio_impl(bio
);
523 current
->bio_tail
= bio_tail
;
525 struct bio_list
*bio_list
= current
->bio_list
;
526 current
->bio_list
= NULL
;
527 vdev_submit_bio_impl(bio
);
528 current
->bio_list
= bio_list
;
533 __vdev_disk_physio(struct block_device
*bdev
, zio_t
*zio
,
534 size_t io_size
, uint64_t io_offset
, int rw
, int flags
)
539 int bio_size
, bio_count
= 16;
540 int i
= 0, error
= 0;
541 #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
542 struct blk_plug plug
;
545 * Accessing outside the block device is never allowed.
547 if (io_offset
+ io_size
> bdev
->bd_inode
->i_size
) {
548 vdev_dbgmsg(zio
->io_vd
,
549 "Illegal access %llu size %llu, device size %llu",
550 io_offset
, io_size
, i_size_read(bdev
->bd_inode
));
551 return (SET_ERROR(EIO
));
555 dr
= vdev_disk_dio_alloc(bio_count
);
557 return (SET_ERROR(ENOMEM
));
559 if (zio
&& !(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)))
560 bio_set_flags_failfast(bdev
, &flags
);
565 * When the IO size exceeds the maximum bio size for the request
566 * queue we are forced to break the IO in multiple bio's and wait
567 * for them all to complete. Ideally, all pool users will set
568 * their volume block size to match the maximum request size and
569 * the common case will be one bio per vdev IO request.
573 bio_offset
= io_offset
;
575 for (i
= 0; i
<= dr
->dr_bio_count
; i
++) {
577 /* Finished constructing bio's for given buffer */
582 * By default only 'bio_count' bio's per dio are allowed.
583 * However, if we find ourselves in a situation where more
584 * are needed we allocate a larger dio and warn the user.
586 if (dr
->dr_bio_count
== i
) {
587 vdev_disk_dio_free(dr
);
592 /* bio_alloc() with __GFP_WAIT never returns NULL */
593 dr
->dr_bio
[i
] = bio_alloc(GFP_NOIO
,
594 MIN(abd_nr_pages_off(zio
->io_abd
, bio_size
, abd_offset
),
596 if (unlikely(dr
->dr_bio
[i
] == NULL
)) {
597 vdev_disk_dio_free(dr
);
598 return (SET_ERROR(ENOMEM
));
601 /* Matching put called by vdev_disk_physio_completion */
602 vdev_disk_dio_get(dr
);
604 bio_set_dev(dr
->dr_bio
[i
], bdev
);
605 BIO_BI_SECTOR(dr
->dr_bio
[i
]) = bio_offset
>> 9;
606 dr
->dr_bio
[i
]->bi_end_io
= vdev_disk_physio_completion
;
607 dr
->dr_bio
[i
]->bi_private
= dr
;
608 bio_set_op_attrs(dr
->dr_bio
[i
], rw
, flags
);
610 /* Remaining size is returned to become the new size */
611 bio_size
= bio_map_abd_off(dr
->dr_bio
[i
], zio
->io_abd
,
612 bio_size
, abd_offset
);
614 /* Advance in buffer and construct another bio if needed */
615 abd_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
616 bio_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
619 /* Extra reference to protect dio_request during vdev_submit_bio */
620 vdev_disk_dio_get(dr
);
622 #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
623 if (dr
->dr_bio_count
> 1)
624 blk_start_plug(&plug
);
627 /* Submit all bio's associated with this dio */
628 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
630 vdev_submit_bio(dr
->dr_bio
[i
]);
632 #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
633 if (dr
->dr_bio_count
> 1)
634 blk_finish_plug(&plug
);
637 (void) vdev_disk_dio_put(dr
);
642 BIO_END_IO_PROTO(vdev_disk_io_flush_completion
, bio
, error
)
644 zio_t
*zio
= bio
->bi_private
;
645 #ifdef HAVE_1ARG_BIO_END_IO_T
646 zio
->io_error
= BIO_END_IO_ERROR(bio
);
648 zio
->io_error
= -error
;
651 if (zio
->io_error
&& (zio
->io_error
== EOPNOTSUPP
))
652 zio
->io_vd
->vdev_nowritecache
= B_TRUE
;
655 ASSERT3S(zio
->io_error
, >=, 0);
657 vdev_disk_error(zio
);
662 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
664 struct request_queue
*q
;
667 q
= bdev_get_queue(bdev
);
669 return (SET_ERROR(ENXIO
));
671 bio
= bio_alloc(GFP_NOIO
, 0);
672 /* bio_alloc() with __GFP_WAIT never returns NULL */
673 if (unlikely(bio
== NULL
))
674 return (SET_ERROR(ENOMEM
));
676 bio
->bi_end_io
= vdev_disk_io_flush_completion
;
677 bio
->bi_private
= zio
;
678 bio_set_dev(bio
, bdev
);
680 vdev_submit_bio(bio
);
681 invalidate_bdev(bdev
);
687 vdev_disk_io_start(zio_t
*zio
)
689 vdev_t
*v
= zio
->io_vd
;
690 vdev_disk_t
*vd
= v
->vdev_tsd
;
691 int rw
, flags
, error
;
694 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
695 * Nothing to be done here but return failure.
698 zio
->io_error
= ENXIO
;
703 rw_enter(&vd
->vd_lock
, RW_READER
);
706 * If the vdev is closed, it's likely due to a failed reopen and is
707 * in the UNAVAIL state. Nothing to be done here but return failure.
709 if (vd
->vd_bdev
== NULL
) {
710 rw_exit(&vd
->vd_lock
);
711 zio
->io_error
= ENXIO
;
716 switch (zio
->io_type
) {
719 if (!vdev_readable(v
)) {
720 rw_exit(&vd
->vd_lock
);
721 zio
->io_error
= SET_ERROR(ENXIO
);
726 switch (zio
->io_cmd
) {
727 case DKIOCFLUSHWRITECACHE
:
729 if (zfs_nocacheflush
)
732 if (v
->vdev_nowritecache
) {
733 zio
->io_error
= SET_ERROR(ENOTSUP
);
737 error
= vdev_disk_io_flush(vd
->vd_bdev
, zio
);
739 rw_exit(&vd
->vd_lock
);
743 zio
->io_error
= error
;
748 zio
->io_error
= SET_ERROR(ENOTSUP
);
751 rw_exit(&vd
->vd_lock
);
756 #if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG)
757 flags
= (1 << BIO_RW_UNPLUG
);
758 #elif defined(REQ_UNPLUG)
767 #if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG)
768 flags
= (1 << BIO_RW_UNPLUG
);
769 #elif defined(REQ_UNPLUG)
777 rw_exit(&vd
->vd_lock
);
778 zio
->io_error
= SET_ERROR(ENOTSUP
);
783 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
784 error
= __vdev_disk_physio(vd
->vd_bdev
, zio
,
785 zio
->io_size
, zio
->io_offset
, rw
, flags
);
786 rw_exit(&vd
->vd_lock
);
789 zio
->io_error
= error
;
796 vdev_disk_io_done(zio_t
*zio
)
799 * If the device returned EIO, we revalidate the media. If it is
800 * determined the media has changed this triggers the asynchronous
801 * removal of the device from the configuration.
803 if (zio
->io_error
== EIO
) {
804 vdev_t
*v
= zio
->io_vd
;
805 vdev_disk_t
*vd
= v
->vdev_tsd
;
807 if (check_disk_change(vd
->vd_bdev
)) {
808 vdev_bdev_invalidate(vd
->vd_bdev
);
809 v
->vdev_remove_wanted
= B_TRUE
;
810 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
816 vdev_disk_hold(vdev_t
*vd
)
818 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
820 /* We must have a pathname, and it must be absolute. */
821 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
825 * Only prefetch path and devid info if the device has
828 if (vd
->vdev_tsd
!= NULL
)
831 /* XXX: Implement me as a vnode lookup for the device */
832 vd
->vdev_name_vp
= NULL
;
833 vd
->vdev_devid_vp
= NULL
;
837 vdev_disk_rele(vdev_t
*vd
)
839 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
841 /* XXX: Implement me as a vnode rele for the device */
845 param_set_vdev_scheduler(const char *val
, zfs_kernel_param_t
*kp
)
851 return (SET_ERROR(-EINVAL
));
853 if ((p
= strchr(val
, '\n')) != NULL
)
856 if (spa_mode_global
!= 0) {
857 mutex_enter(&spa_namespace_lock
);
858 while ((spa
= spa_next(spa
)) != NULL
) {
859 if (spa_state(spa
) != POOL_STATE_ACTIVE
||
860 !spa_writeable(spa
) || spa_suspended(spa
))
863 spa_open_ref(spa
, FTAG
);
864 mutex_exit(&spa_namespace_lock
);
865 vdev_elevator_switch(spa
->spa_root_vdev
, (char *)val
);
866 mutex_enter(&spa_namespace_lock
);
867 spa_close(spa
, FTAG
);
869 mutex_exit(&spa_namespace_lock
);
872 return (param_set_charp(val
, kp
));
875 vdev_ops_t vdev_disk_ops
= {
886 VDEV_TYPE_DISK
, /* name of this vdev type */
887 B_TRUE
/* leaf vdev */
890 module_param_call(zfs_vdev_scheduler
, param_set_vdev_scheduler
,
891 param_get_charp
, &zfs_vdev_scheduler
, 0644);
892 MODULE_PARM_DESC(zfs_vdev_scheduler
, "I/O scheduler");