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>.
28 #include <sys/zfs_context.h>
30 #include <sys/vdev_disk.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/fs/zfs.h>
34 #include <sys/sunldi.h>
36 char *zfs_vdev_scheduler
= VDEV_SCHEDULER
;
37 static void *zfs_vdev_holder
= VDEV_HOLDER
;
40 * Virtual device vector for disks.
42 typedef struct dio_request
{
43 struct completion dr_comp
; /* Completion for sync IO */
44 atomic_t dr_ref
; /* References */
45 zio_t
*dr_zio
; /* Parent ZIO */
46 int dr_rw
; /* Read/Write */
47 int dr_error
; /* Bio error */
48 int dr_bio_count
; /* Count of bio's */
49 struct bio
*dr_bio
[0]; /* Attached bio's */
53 #ifdef HAVE_OPEN_BDEV_EXCLUSIVE
55 vdev_bdev_mode(int smode
)
59 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
71 vdev_bdev_mode(int smode
)
75 ASSERT3S(smode
& (FREAD
| FWRITE
), !=, 0);
77 if ((smode
& FREAD
) && !(smode
& FWRITE
))
82 #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
85 bdev_capacity(struct block_device
*bdev
)
87 struct hd_struct
*part
= bdev
->bd_part
;
89 /* The partition capacity referenced by the block device */
91 return (part
->nr_sects
<< 9);
93 /* Otherwise assume the full device capacity */
94 return (get_capacity(bdev
->bd_disk
) << 9);
98 vdev_disk_error(zio_t
*zio
)
101 printk("ZFS: zio error=%d type=%d offset=%llu size=%llu "
102 "flags=%x delay=%llu\n", zio
->io_error
, zio
->io_type
,
103 (u_longlong_t
)zio
->io_offset
, (u_longlong_t
)zio
->io_size
,
104 zio
->io_flags
, (u_longlong_t
)zio
->io_delay
);
109 * Use the Linux 'noop' elevator for zfs managed block devices. This
110 * strikes the ideal balance by allowing the zfs elevator to do all
111 * request ordering and prioritization. While allowing the Linux
112 * elevator to do the maximum front/back merging allowed by the
113 * physical device. This yields the largest possible requests for
114 * the device with the lowest total overhead.
117 vdev_elevator_switch(vdev_t
*v
, char *elevator
)
119 vdev_disk_t
*vd
= v
->vdev_tsd
;
120 struct block_device
*bdev
= vd
->vd_bdev
;
121 struct request_queue
*q
= bdev_get_queue(bdev
);
122 char *device
= bdev
->bd_disk
->disk_name
;
126 * Skip devices which are not whole disks (partitions).
127 * Device-mapper devices are excepted since they may be whole
128 * disks despite the vdev_wholedisk flag, in which case we can
129 * and should switch the elevator. If the device-mapper device
130 * does not have an elevator (i.e. dm-raid, dm-crypt, etc.) the
131 * "Skip devices without schedulers" check below will fail.
133 if (!v
->vdev_wholedisk
&& strncmp(device
, "dm-", 3) != 0)
136 /* Skip devices without schedulers (loop, ram, dm, etc) */
137 if (!q
->elevator
|| !blk_queue_stackable(q
))
140 /* Leave existing scheduler when set to "none" */
141 if (!strncmp(elevator
, "none", 4) && (strlen(elevator
) == 4))
144 #ifdef HAVE_ELEVATOR_CHANGE
145 error
= elevator_change(q
, elevator
);
147 /* For pre-2.6.36 kernels elevator_change() is not available.
148 * Therefore we fall back to using a usermodehelper to echo the
149 * elevator into sysfs; This requires /bin/echo and sysfs to be
150 * mounted which may not be true early in the boot process.
152 # define SET_SCHEDULER_CMD \
153 "exec 0</dev/null " \
154 " 1>/sys/block/%s/queue/scheduler " \
159 char *argv
[] = { "/bin/sh", "-c", NULL
, NULL
};
160 char *envp
[] = { NULL
};
162 argv
[2] = kmem_asprintf(SET_SCHEDULER_CMD
, device
, elevator
);
163 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
166 #endif /* HAVE_ELEVATOR_CHANGE */
168 printk("ZFS: Unable to set \"%s\" scheduler for %s (%s): %d\n",
169 elevator
, v
->vdev_path
, device
, error
);
175 * Expanding a whole disk vdev involves invoking BLKRRPART on the
176 * whole disk device. This poses a problem, because BLKRRPART will
177 * return EBUSY if one of the disk's partitions is open. That's why
178 * we have to do it here, just before opening the data partition.
179 * Unfortunately, BLKRRPART works by dropping all partitions and
180 * recreating them, which means that for a short time window, all
181 * /dev/sdxN device files disappear (until udev recreates them).
182 * This means two things:
183 * - When we open the data partition just after a BLKRRPART, we
184 * can't do it using the normal device file path because of the
185 * obvious race condition with udev. Instead, we use reliable
186 * kernel APIs to get a handle to the new partition device from
187 * the whole disk device.
188 * - Because vdev_disk_open() initially needs to find the device
189 * using its path, multiple vdev_disk_open() invocations in
190 * short succession on the same disk with BLKRRPARTs in the
191 * middle have a high probability of failure (because of the
192 * race condition with udev). A typical situation where this
193 * might happen is when the zpool userspace tool does a
194 * TRYIMPORT immediately followed by an IMPORT. For this
195 * reason, we only invoke BLKRRPART in the module when strictly
196 * necessary (zpool online -e case), and rely on userspace to
197 * do it when possible.
199 static struct block_device
*
200 vdev_disk_rrpart(const char *path
, int mode
, vdev_disk_t
*vd
)
202 #if defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK)
203 struct block_device
*bdev
, *result
= ERR_PTR(-ENXIO
);
204 struct gendisk
*disk
;
207 bdev
= vdev_bdev_open(path
, vdev_bdev_mode(mode
), zfs_vdev_holder
);
211 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
212 vdev_bdev_close(bdev
, vdev_bdev_mode(mode
));
215 bdev
= bdget(disk_devt(disk
));
217 error
= blkdev_get(bdev
, vdev_bdev_mode(mode
), vd
);
219 error
= ioctl_by_bdev(bdev
, BLKRRPART
, 0);
220 vdev_bdev_close(bdev
, vdev_bdev_mode(mode
));
223 bdev
= bdget_disk(disk
, partno
);
225 error
= blkdev_get(bdev
,
226 vdev_bdev_mode(mode
) | FMODE_EXCL
, vd
);
235 return ERR_PTR(-EOPNOTSUPP
);
236 #endif /* defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) */
240 vdev_disk_open(vdev_t
*v
, uint64_t *psize
, uint64_t *max_psize
,
243 struct block_device
*bdev
= ERR_PTR(-ENXIO
);
245 int mode
, block_size
;
247 /* Must have a pathname and it must be absolute. */
248 if (v
->vdev_path
== NULL
|| v
->vdev_path
[0] != '/') {
249 v
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
254 * Reopen the device if it's not currently open. Otherwise,
255 * just update the physical size of the device.
257 if (v
->vdev_tsd
!= NULL
) {
258 ASSERT(v
->vdev_reopening
);
263 vd
= kmem_zalloc(sizeof(vdev_disk_t
), KM_PUSHPAGE
);
268 * Devices are always opened by the path provided at configuration
269 * time. This means that if the provided path is a udev by-id path
270 * then drives may be recabled without an issue. If the provided
271 * path is a udev by-path path, then the physical location information
272 * will be preserved. This can be critical for more complicated
273 * configurations where drives are located in specific physical
274 * locations to maximize the systems tolerence to component failure.
275 * Alternatively, you can provide your own udev rule to flexibly map
276 * the drives as you see fit. It is not advised that you use the
277 * /dev/[hd]d devices which may be reordered due to probing order.
278 * Devices in the wrong locations will be detected by the higher
279 * level vdev validation.
281 mode
= spa_mode(v
->vdev_spa
);
282 if (v
->vdev_wholedisk
&& v
->vdev_expanding
)
283 bdev
= vdev_disk_rrpart(v
->vdev_path
, mode
, vd
);
285 bdev
= vdev_bdev_open(v
->vdev_path
,
286 vdev_bdev_mode(mode
), zfs_vdev_holder
);
288 kmem_free(vd
, sizeof(vdev_disk_t
));
289 return -PTR_ERR(bdev
);
296 /* Determine the physical block size */
297 block_size
= vdev_bdev_block_size(vd
->vd_bdev
);
299 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
300 v
->vdev_nowritecache
= B_FALSE
;
302 /* Physical volume size in bytes */
303 *psize
= bdev_capacity(vd
->vd_bdev
);
305 /* TODO: report possible expansion size */
308 /* Based on the minimum sector size set the block size */
309 *ashift
= highbit(MAX(block_size
, SPA_MINBLOCKSIZE
)) - 1;
311 /* Try to set the io scheduler elevator algorithm */
312 (void) vdev_elevator_switch(v
, zfs_vdev_scheduler
);
318 vdev_disk_close(vdev_t
*v
)
320 vdev_disk_t
*vd
= v
->vdev_tsd
;
322 if (v
->vdev_reopening
|| vd
== NULL
)
325 if (vd
->vd_bdev
!= NULL
)
326 vdev_bdev_close(vd
->vd_bdev
,
327 vdev_bdev_mode(spa_mode(v
->vdev_spa
)));
329 kmem_free(vd
, sizeof(vdev_disk_t
));
333 static dio_request_t
*
334 vdev_disk_dio_alloc(int bio_count
)
339 dr
= kmem_zalloc(sizeof(dio_request_t
) +
340 sizeof(struct bio
*) * bio_count
, KM_PUSHPAGE
);
342 init_completion(&dr
->dr_comp
);
343 atomic_set(&dr
->dr_ref
, 0);
344 dr
->dr_bio_count
= bio_count
;
347 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
348 dr
->dr_bio
[i
] = NULL
;
355 vdev_disk_dio_free(dio_request_t
*dr
)
359 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
361 bio_put(dr
->dr_bio
[i
]);
363 kmem_free(dr
, sizeof(dio_request_t
) +
364 sizeof(struct bio
*) * dr
->dr_bio_count
);
368 vdev_disk_dio_is_sync(dio_request_t
*dr
)
370 #ifdef HAVE_BIO_RW_SYNC
371 /* BIO_RW_SYNC preferred interface from 2.6.12-2.6.29 */
372 return (dr
->dr_rw
& (1 << BIO_RW_SYNC
));
374 # ifdef HAVE_BIO_RW_SYNCIO
375 /* BIO_RW_SYNCIO preferred interface from 2.6.30-2.6.35 */
376 return (dr
->dr_rw
& (1 << BIO_RW_SYNCIO
));
378 # ifdef HAVE_REQ_SYNC
379 /* REQ_SYNC preferred interface from 2.6.36-2.6.xx */
380 return (dr
->dr_rw
& REQ_SYNC
);
382 # error "Unable to determine bio sync flag"
383 # endif /* HAVE_REQ_SYNC */
384 # endif /* HAVE_BIO_RW_SYNC */
385 #endif /* HAVE_BIO_RW_SYNCIO */
389 vdev_disk_dio_get(dio_request_t
*dr
)
391 atomic_inc(&dr
->dr_ref
);
395 vdev_disk_dio_put(dio_request_t
*dr
)
397 int rc
= atomic_dec_return(&dr
->dr_ref
);
400 * Free the dio_request when the last reference is dropped and
401 * ensure zio_interpret is called only once with the correct zio
404 zio_t
*zio
= dr
->dr_zio
;
405 int error
= dr
->dr_error
;
407 vdev_disk_dio_free(dr
);
410 zio
->io_delay
= jiffies_64
- zio
->io_delay
;
411 zio
->io_error
= error
;
412 ASSERT3S(zio
->io_error
, >=, 0);
414 vdev_disk_error(zio
);
422 BIO_END_IO_PROTO(vdev_disk_physio_completion
, bio
, size
, error
)
424 dio_request_t
*dr
= bio
->bi_private
;
427 /* Fatal error but print some useful debugging before asserting */
429 PANIC("dr == NULL, bio->bi_private == NULL\n"
430 "bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d\n"
431 "bi_idx: %d, bi_size: %d, bi_end_io: %p, bi_cnt: %d\n",
432 bio
->bi_next
, bio
->bi_flags
, bio
->bi_rw
, bio
->bi_vcnt
,
433 bio
->bi_idx
, bio
->bi_size
, bio
->bi_end_io
,
434 atomic_read(&bio
->bi_cnt
));
436 #ifndef HAVE_2ARGS_BIO_END_IO_T
439 #endif /* HAVE_2ARGS_BIO_END_IO_T */
441 if (error
== 0 && !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
444 if (dr
->dr_error
== 0)
445 dr
->dr_error
= -error
;
447 /* Drop reference aquired by __vdev_disk_physio */
448 rc
= vdev_disk_dio_put(dr
);
450 /* Wake up synchronous waiter this is the last outstanding bio */
451 if ((rc
== 1) && vdev_disk_dio_is_sync(dr
))
452 complete(&dr
->dr_comp
);
454 BIO_END_IO_RETURN(0);
457 static inline unsigned long
458 bio_nr_pages(void *bio_ptr
, unsigned int bio_size
)
460 return ((((unsigned long)bio_ptr
+ bio_size
+ PAGE_SIZE
- 1) >>
461 PAGE_SHIFT
) - ((unsigned long)bio_ptr
>> PAGE_SHIFT
));
465 bio_map(struct bio
*bio
, void *bio_ptr
, unsigned int bio_size
)
467 unsigned int offset
, size
, i
;
470 offset
= offset_in_page(bio_ptr
);
471 for (i
= 0; i
< bio
->bi_max_vecs
; i
++) {
472 size
= PAGE_SIZE
- offset
;
480 if (kmem_virt(bio_ptr
))
481 page
= vmalloc_to_page(bio_ptr
);
483 page
= virt_to_page(bio_ptr
);
485 if (bio_add_page(bio
, page
, size
, offset
) != size
)
497 __vdev_disk_physio(struct block_device
*bdev
, zio_t
*zio
, caddr_t kbuf_ptr
,
498 size_t kbuf_size
, uint64_t kbuf_offset
, int flags
)
503 int bio_size
, bio_count
= 16;
504 int i
= 0, error
= 0;
506 ASSERT3U(kbuf_offset
+ kbuf_size
, <=, bdev
->bd_inode
->i_size
);
509 dr
= vdev_disk_dio_alloc(bio_count
);
513 if (zio
&& !(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)))
514 bio_set_flags_failfast(bdev
, &flags
);
520 * When the IO size exceeds the maximum bio size for the request
521 * queue we are forced to break the IO in multiple bio's and wait
522 * for them all to complete. Ideally, all pool users will set
523 * their volume block size to match the maximum request size and
524 * the common case will be one bio per vdev IO request.
527 bio_offset
= kbuf_offset
;
528 bio_size
= kbuf_size
;
529 for (i
= 0; i
<= dr
->dr_bio_count
; i
++) {
531 /* Finished constructing bio's for given buffer */
536 * By default only 'bio_count' bio's per dio are allowed.
537 * However, if we find ourselves in a situation where more
538 * are needed we allocate a larger dio and warn the user.
540 if (dr
->dr_bio_count
== i
) {
541 vdev_disk_dio_free(dr
);
546 dr
->dr_bio
[i
] = bio_alloc(GFP_NOIO
,
547 bio_nr_pages(bio_ptr
, bio_size
));
548 if (dr
->dr_bio
[i
] == NULL
) {
549 vdev_disk_dio_free(dr
);
553 /* Matching put called by vdev_disk_physio_completion */
554 vdev_disk_dio_get(dr
);
556 dr
->dr_bio
[i
]->bi_bdev
= bdev
;
557 dr
->dr_bio
[i
]->bi_sector
= bio_offset
>> 9;
558 dr
->dr_bio
[i
]->bi_rw
= dr
->dr_rw
;
559 dr
->dr_bio
[i
]->bi_end_io
= vdev_disk_physio_completion
;
560 dr
->dr_bio
[i
]->bi_private
= dr
;
562 /* Remaining size is returned to become the new size */
563 bio_size
= bio_map(dr
->dr_bio
[i
], bio_ptr
, bio_size
);
565 /* Advance in buffer and construct another bio if needed */
566 bio_ptr
+= dr
->dr_bio
[i
]->bi_size
;
567 bio_offset
+= dr
->dr_bio
[i
]->bi_size
;
570 /* Extra reference to protect dio_request during submit_bio */
571 vdev_disk_dio_get(dr
);
573 zio
->io_delay
= jiffies_64
;
575 /* Submit all bio's associated with this dio */
576 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
578 submit_bio(dr
->dr_rw
, dr
->dr_bio
[i
]);
581 * On synchronous blocking requests we wait for all bio the completion
582 * callbacks to run. We will be woken when the last callback runs
583 * for this dio. We are responsible for putting the last dio_request
584 * reference will in turn put back the last bio references. The
585 * only synchronous consumer is vdev_disk_read_rootlabel() all other
586 * IO originating from vdev_disk_io_start() is asynchronous.
588 if (vdev_disk_dio_is_sync(dr
)) {
589 wait_for_completion(&dr
->dr_comp
);
590 error
= dr
->dr_error
;
591 ASSERT3S(atomic_read(&dr
->dr_ref
), ==, 1);
594 (void)vdev_disk_dio_put(dr
);
600 vdev_disk_physio(struct block_device
*bdev
, caddr_t kbuf
,
601 size_t size
, uint64_t offset
, int flags
)
603 bio_set_flags_failfast(bdev
, &flags
);
604 return __vdev_disk_physio(bdev
, NULL
, kbuf
, size
, offset
, flags
);
607 BIO_END_IO_PROTO(vdev_disk_io_flush_completion
, bio
, size
, rc
)
609 zio_t
*zio
= bio
->bi_private
;
611 zio
->io_delay
= jiffies_64
- zio
->io_delay
;
613 if (rc
&& (rc
== -EOPNOTSUPP
))
614 zio
->io_vd
->vdev_nowritecache
= B_TRUE
;
617 ASSERT3S(zio
->io_error
, >=, 0);
619 vdev_disk_error(zio
);
622 BIO_END_IO_RETURN(0);
626 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
628 struct request_queue
*q
;
631 q
= bdev_get_queue(bdev
);
635 bio
= bio_alloc(GFP_NOIO
, 0);
639 bio
->bi_end_io
= vdev_disk_io_flush_completion
;
640 bio
->bi_private
= zio
;
642 zio
->io_delay
= jiffies_64
;
643 submit_bio(VDEV_WRITE_FLUSH_FUA
, bio
);
649 vdev_disk_io_start(zio_t
*zio
)
651 vdev_t
*v
= zio
->io_vd
;
652 vdev_disk_t
*vd
= v
->vdev_tsd
;
655 switch (zio
->io_type
) {
658 if (!vdev_readable(v
)) {
659 zio
->io_error
= ENXIO
;
660 return ZIO_PIPELINE_CONTINUE
;
663 switch (zio
->io_cmd
) {
664 case DKIOCFLUSHWRITECACHE
:
666 if (zfs_nocacheflush
)
669 if (v
->vdev_nowritecache
) {
670 zio
->io_error
= ENOTSUP
;
674 error
= vdev_disk_io_flush(vd
->vd_bdev
, zio
);
676 return ZIO_PIPELINE_STOP
;
678 zio
->io_error
= error
;
679 if (error
== ENOTSUP
)
680 v
->vdev_nowritecache
= B_TRUE
;
685 zio
->io_error
= ENOTSUP
;
688 return ZIO_PIPELINE_CONTINUE
;
699 zio
->io_error
= ENOTSUP
;
700 return ZIO_PIPELINE_CONTINUE
;
703 error
= __vdev_disk_physio(vd
->vd_bdev
, zio
, zio
->io_data
,
704 zio
->io_size
, zio
->io_offset
, flags
);
706 zio
->io_error
= error
;
707 return ZIO_PIPELINE_CONTINUE
;
710 return ZIO_PIPELINE_STOP
;
714 vdev_disk_io_done(zio_t
*zio
)
717 * If the device returned EIO, we revalidate the media. If it is
718 * determined the media has changed this triggers the asynchronous
719 * removal of the device from the configuration.
721 if (zio
->io_error
== EIO
) {
722 vdev_t
*v
= zio
->io_vd
;
723 vdev_disk_t
*vd
= v
->vdev_tsd
;
725 if (check_disk_change(vd
->vd_bdev
)) {
726 vdev_bdev_invalidate(vd
->vd_bdev
);
727 v
->vdev_remove_wanted
= B_TRUE
;
728 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
734 vdev_disk_hold(vdev_t
*vd
)
736 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
738 /* We must have a pathname, and it must be absolute. */
739 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
743 * Only prefetch path and devid info if the device has
746 if (vd
->vdev_tsd
!= NULL
)
749 /* XXX: Implement me as a vnode lookup for the device */
750 vd
->vdev_name_vp
= NULL
;
751 vd
->vdev_devid_vp
= NULL
;
755 vdev_disk_rele(vdev_t
*vd
)
757 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
759 /* XXX: Implement me as a vnode rele for the device */
762 vdev_ops_t vdev_disk_ops
= {
771 VDEV_TYPE_DISK
, /* name of this vdev type */
772 B_TRUE
/* leaf vdev */
776 * Given the root disk device devid or pathname, read the label from
777 * the device, and construct a configuration nvlist.
780 vdev_disk_read_rootlabel(char *devpath
, char *devid
, nvlist_t
**config
)
782 struct block_device
*bdev
;
787 bdev
= vdev_bdev_open(devpath
, vdev_bdev_mode(FREAD
), zfs_vdev_holder
);
789 return -PTR_ERR(bdev
);
791 s
= bdev_capacity(bdev
);
793 vdev_bdev_close(bdev
, vdev_bdev_mode(FREAD
));
797 size
= P2ALIGN_TYPED(s
, sizeof(vdev_label_t
), uint64_t);
798 label
= vmem_alloc(sizeof(vdev_label_t
), KM_PUSHPAGE
);
800 for (i
= 0; i
< VDEV_LABELS
; i
++) {
801 uint64_t offset
, state
, txg
= 0;
803 /* read vdev label */
804 offset
= vdev_label_offset(size
, i
, 0);
805 if (vdev_disk_physio(bdev
, (caddr_t
)label
,
806 VDEV_SKIP_SIZE
+ VDEV_PHYS_SIZE
, offset
, READ_SYNC
) != 0)
809 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
810 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0) {
815 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
816 &state
) != 0 || state
>= POOL_STATE_DESTROYED
) {
817 nvlist_free(*config
);
822 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
823 &txg
) != 0 || txg
== 0) {
824 nvlist_free(*config
);
832 vmem_free(label
, sizeof(vdev_label_t
));
833 vdev_bdev_close(bdev
, vdev_bdev_mode(FREAD
));
838 module_param(zfs_vdev_scheduler
, charp
, 0644);
839 MODULE_PARM_DESC(zfs_vdev_scheduler
, "I/O scheduler");