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 https://opensource.org/licenses/CDDL-1.0.
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, 2019 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>
33 #include <sys/vdev_trim.h>
35 #include <sys/fs/zfs.h>
37 #include <linux/blkpg.h>
38 #include <linux/msdos_fs.h>
39 #include <linux/vfs_compat.h>
40 #ifdef HAVE_LINUX_BLK_CGROUP_HEADER
41 #include <linux/blk-cgroup.h>
44 typedef struct vdev_disk
{
45 struct block_device
*vd_bdev
;
50 * Unique identifier for the exclusive vdev holder.
52 static void *zfs_vdev_holder
= VDEV_HOLDER
;
55 * Wait up to zfs_vdev_open_timeout_ms milliseconds before determining the
56 * device is missing. The missing path may be transient since the links
57 * can be briefly removed and recreated in response to udev events.
59 static uint_t zfs_vdev_open_timeout_ms
= 1000;
62 * Size of the "reserved" partition, in blocks.
64 #define EFI_MIN_RESV_SIZE (16 * 1024)
67 * Virtual device vector for disks.
69 typedef struct dio_request
{
70 zio_t
*dr_zio
; /* Parent ZIO */
71 atomic_t dr_ref
; /* References */
72 int dr_error
; /* Bio error */
73 int dr_bio_count
; /* Count of bio's */
74 struct bio
*dr_bio
[]; /* Attached bio's */
78 * BIO request failfast mask.
81 static unsigned int zfs_vdev_failfast_mask
= 1;
84 vdev_bdev_mode(spa_mode_t spa_mode
)
88 if (spa_mode
& SPA_MODE_READ
)
91 if (spa_mode
& SPA_MODE_WRITE
)
98 * Returns the usable capacity (in bytes) for the partition or disk.
101 bdev_capacity(struct block_device
*bdev
)
103 return (i_size_read(bdev
->bd_inode
));
106 #if !defined(HAVE_BDEV_WHOLE)
107 static inline struct block_device
*
108 bdev_whole(struct block_device
*bdev
)
110 return (bdev
->bd_contains
);
114 #if defined(HAVE_BDEVNAME)
115 #define vdev_bdevname(bdev, name) bdevname(bdev, name)
118 vdev_bdevname(struct block_device
*bdev
, char *name
)
120 snprintf(name
, BDEVNAME_SIZE
, "%pg", bdev
);
125 * Returns the maximum expansion capacity of the block device (in bytes).
127 * It is possible to expand a vdev when it has been created as a wholedisk
128 * and the containing block device has increased in capacity. Or when the
129 * partition containing the pool has been manually increased in size.
131 * This function is only responsible for calculating the potential expansion
132 * size so it can be reported by 'zpool list'. The efi_use_whole_disk() is
133 * responsible for verifying the expected partition layout in the wholedisk
134 * case, and updating the partition table if appropriate. Once the partition
135 * size has been increased the additional capacity will be visible using
138 * The returned maximum expansion capacity is always expected to be larger, or
139 * at the very least equal, to its usable capacity to prevent overestimating
140 * the pool expandsize.
143 bdev_max_capacity(struct block_device
*bdev
, uint64_t wholedisk
)
148 if (wholedisk
&& bdev
!= bdev_whole(bdev
)) {
150 * When reporting maximum expansion capacity for a wholedisk
151 * deduct any capacity which is expected to be lost due to
152 * alignment restrictions. Over reporting this value isn't
153 * harmful and would only result in slightly less capacity
154 * than expected post expansion.
155 * The estimated available space may be slightly smaller than
156 * bdev_capacity() for devices where the number of sectors is
157 * not a multiple of the alignment size and the partition layout
158 * is keeping less than PARTITION_END_ALIGNMENT bytes after the
159 * "reserved" EFI partition: in such cases return the device
162 available
= i_size_read(bdev_whole(bdev
)->bd_inode
) -
163 ((EFI_MIN_RESV_SIZE
+ NEW_START_BLOCK
+
164 PARTITION_END_ALIGNMENT
) << SECTOR_BITS
);
165 psize
= MAX(available
, bdev_capacity(bdev
));
167 psize
= bdev_capacity(bdev
);
174 vdev_disk_error(zio_t
*zio
)
177 * This function can be called in interrupt context, for instance while
178 * handling IRQs coming from a misbehaving disk device; use printk()
179 * which is safe from any context.
181 printk(KERN_WARNING
"zio pool=%s vdev=%s error=%d type=%d "
182 "offset=%llu size=%llu flags=%llu\n", spa_name(zio
->io_spa
),
183 zio
->io_vd
->vdev_path
, zio
->io_error
, zio
->io_type
,
184 (u_longlong_t
)zio
->io_offset
, (u_longlong_t
)zio
->io_size
,
189 vdev_disk_kobj_evt_post(vdev_t
*v
)
191 vdev_disk_t
*vd
= v
->vdev_tsd
;
192 if (vd
&& vd
->vd_bdev
) {
193 spl_signal_kobj_evt(vd
->vd_bdev
);
195 vdev_dbgmsg(v
, "vdev_disk_t is NULL for VDEV:%s\n",
201 vdev_disk_open(vdev_t
*v
, uint64_t *psize
, uint64_t *max_psize
,
202 uint64_t *logical_ashift
, uint64_t *physical_ashift
)
204 struct block_device
*bdev
;
205 fmode_t mode
= vdev_bdev_mode(spa_mode(v
->vdev_spa
));
206 hrtime_t timeout
= MSEC2NSEC(zfs_vdev_open_timeout_ms
);
209 /* Must have a pathname and it must be absolute. */
210 if (v
->vdev_path
== NULL
|| v
->vdev_path
[0] != '/') {
211 v
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
212 vdev_dbgmsg(v
, "invalid vdev_path");
213 return (SET_ERROR(EINVAL
));
217 * Reopen the device if it is currently open. When expanding a
218 * partition force re-scanning the partition table if userland
219 * did not take care of this already. We need to do this while closed
220 * in order to get an accurate updated block device size. Then
221 * since udev may need to recreate the device links increase the
222 * open retry timeout before reporting the device as unavailable.
226 char disk_name
[BDEVNAME_SIZE
+ 6] = "/dev/";
227 boolean_t reread_part
= B_FALSE
;
229 rw_enter(&vd
->vd_lock
, RW_WRITER
);
234 if (v
->vdev_expanding
&& bdev
!= bdev_whole(bdev
)) {
235 vdev_bdevname(bdev_whole(bdev
), disk_name
+ 5);
237 * If userland has BLKPG_RESIZE_PARTITION,
238 * then it should have updated the partition
239 * table already. We can detect this by
240 * comparing our current physical size
241 * with that of the device. If they are
242 * the same, then we must not have
243 * BLKPG_RESIZE_PARTITION or it failed to
244 * update the partition table online. We
245 * fallback to rescanning the partition
246 * table from the kernel below. However,
247 * if the capacity already reflects the
248 * updated partition, then we skip
249 * rescanning the partition table here.
251 if (v
->vdev_psize
== bdev_capacity(bdev
))
252 reread_part
= B_TRUE
;
255 blkdev_put(bdev
, mode
| FMODE_EXCL
);
259 bdev
= blkdev_get_by_path(disk_name
, mode
| FMODE_EXCL
,
262 int error
= vdev_bdev_reread_part(bdev
);
263 blkdev_put(bdev
, mode
| FMODE_EXCL
);
266 zfs_vdev_open_timeout_ms
* 2);
271 vd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
273 rw_init(&vd
->vd_lock
, NULL
, RW_DEFAULT
, NULL
);
274 rw_enter(&vd
->vd_lock
, RW_WRITER
);
278 * Devices are always opened by the path provided at configuration
279 * time. This means that if the provided path is a udev by-id path
280 * then drives may be re-cabled without an issue. If the provided
281 * path is a udev by-path path, then the physical location information
282 * will be preserved. This can be critical for more complicated
283 * configurations where drives are located in specific physical
284 * locations to maximize the systems tolerance to component failure.
286 * Alternatively, you can provide your own udev rule to flexibly map
287 * the drives as you see fit. It is not advised that you use the
288 * /dev/[hd]d devices which may be reordered due to probing order.
289 * Devices in the wrong locations will be detected by the higher
290 * level vdev validation.
292 * The specified paths may be briefly removed and recreated in
293 * response to udev events. This should be exceptionally unlikely
294 * because the zpool command makes every effort to verify these paths
295 * have already settled prior to reaching this point. Therefore,
296 * a ENOENT failure at this point is highly likely to be transient
297 * and it is reasonable to sleep and retry before giving up. In
298 * practice delays have been observed to be on the order of 100ms.
300 * When ERESTARTSYS is returned it indicates the block device is
301 * a zvol which could not be opened due to the deadlock detection
302 * logic in zvol_open(). Extend the timeout and retry the open
303 * subsequent attempts are expected to eventually succeed.
305 hrtime_t start
= gethrtime();
306 bdev
= ERR_PTR(-ENXIO
);
307 while (IS_ERR(bdev
) && ((gethrtime() - start
) < timeout
)) {
308 bdev
= blkdev_get_by_path(v
->vdev_path
, mode
| FMODE_EXCL
,
310 if (unlikely(PTR_ERR(bdev
) == -ENOENT
)) {
312 * There is no point of waiting since device is removed
318 schedule_timeout(MSEC_TO_TICK(10));
319 } else if (unlikely(PTR_ERR(bdev
) == -ERESTARTSYS
)) {
320 timeout
= MSEC2NSEC(zfs_vdev_open_timeout_ms
* 10);
322 } else if (IS_ERR(bdev
)) {
328 int error
= -PTR_ERR(bdev
);
329 vdev_dbgmsg(v
, "open error=%d timeout=%llu/%llu", error
,
330 (u_longlong_t
)(gethrtime() - start
),
331 (u_longlong_t
)timeout
);
334 rw_exit(&vd
->vd_lock
);
335 return (SET_ERROR(error
));
339 rw_exit(&vd
->vd_lock
);
342 /* Determine the physical block size */
343 int physical_block_size
= bdev_physical_block_size(vd
->vd_bdev
);
345 /* Determine the logical block size */
346 int logical_block_size
= bdev_logical_block_size(vd
->vd_bdev
);
348 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
349 v
->vdev_nowritecache
= B_FALSE
;
351 /* Set when device reports it supports TRIM. */
352 v
->vdev_has_trim
= bdev_discard_supported(vd
->vd_bdev
);
354 /* Set when device reports it supports secure TRIM. */
355 v
->vdev_has_securetrim
= bdev_secure_discard_supported(vd
->vd_bdev
);
357 /* Inform the ZIO pipeline that we are non-rotational */
358 v
->vdev_nonrot
= blk_queue_nonrot(bdev_get_queue(vd
->vd_bdev
));
360 /* Physical volume size in bytes for the partition */
361 *psize
= bdev_capacity(vd
->vd_bdev
);
363 /* Physical volume size in bytes including possible expansion space */
364 *max_psize
= bdev_max_capacity(vd
->vd_bdev
, v
->vdev_wholedisk
);
366 /* Based on the minimum sector size set the block size */
367 *physical_ashift
= highbit64(MAX(physical_block_size
,
368 SPA_MINBLOCKSIZE
)) - 1;
370 *logical_ashift
= highbit64(MAX(logical_block_size
,
371 SPA_MINBLOCKSIZE
)) - 1;
377 vdev_disk_close(vdev_t
*v
)
379 vdev_disk_t
*vd
= v
->vdev_tsd
;
381 if (v
->vdev_reopening
|| vd
== NULL
)
384 if (vd
->vd_bdev
!= NULL
) {
385 blkdev_put(vd
->vd_bdev
,
386 vdev_bdev_mode(spa_mode(v
->vdev_spa
)) | FMODE_EXCL
);
389 rw_destroy(&vd
->vd_lock
);
390 kmem_free(vd
, sizeof (vdev_disk_t
));
394 static dio_request_t
*
395 vdev_disk_dio_alloc(int bio_count
)
397 dio_request_t
*dr
= kmem_zalloc(sizeof (dio_request_t
) +
398 sizeof (struct bio
*) * bio_count
, KM_SLEEP
);
399 atomic_set(&dr
->dr_ref
, 0);
400 dr
->dr_bio_count
= bio_count
;
403 for (int i
= 0; i
< dr
->dr_bio_count
; i
++)
404 dr
->dr_bio
[i
] = NULL
;
410 vdev_disk_dio_free(dio_request_t
*dr
)
414 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
416 bio_put(dr
->dr_bio
[i
]);
418 kmem_free(dr
, sizeof (dio_request_t
) +
419 sizeof (struct bio
*) * dr
->dr_bio_count
);
423 vdev_disk_dio_get(dio_request_t
*dr
)
425 atomic_inc(&dr
->dr_ref
);
429 vdev_disk_dio_put(dio_request_t
*dr
)
431 int rc
= atomic_dec_return(&dr
->dr_ref
);
434 * Free the dio_request when the last reference is dropped and
435 * ensure zio_interpret is called only once with the correct zio
438 zio_t
*zio
= dr
->dr_zio
;
439 int error
= dr
->dr_error
;
441 vdev_disk_dio_free(dr
);
444 zio
->io_error
= error
;
445 ASSERT3S(zio
->io_error
, >=, 0);
447 vdev_disk_error(zio
);
449 zio_delay_interrupt(zio
);
454 BIO_END_IO_PROTO(vdev_disk_physio_completion
, bio
, error
)
456 dio_request_t
*dr
= bio
->bi_private
;
458 if (dr
->dr_error
== 0) {
459 #ifdef HAVE_1ARG_BIO_END_IO_T
460 dr
->dr_error
= BIO_END_IO_ERROR(bio
);
463 dr
->dr_error
= -(error
);
464 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
469 /* Drop reference acquired by __vdev_disk_physio */
470 vdev_disk_dio_put(dr
);
474 vdev_submit_bio_impl(struct bio
*bio
)
476 #ifdef HAVE_1ARG_SUBMIT_BIO
477 (void) submit_bio(bio
);
479 (void) submit_bio(bio_data_dir(bio
), bio
);
484 * preempt_schedule_notrace is GPL-only which breaks the ZFS build, so
485 * replace it with preempt_schedule under the following condition:
487 #if defined(CONFIG_ARM64) && \
488 defined(CONFIG_PREEMPTION) && \
489 defined(CONFIG_BLK_CGROUP)
490 #define preempt_schedule_notrace(x) preempt_schedule(x)
494 * As for the Linux 5.18 kernel bio_alloc() expects a block_device struct
495 * as an argument removing the need to set it with bio_set_dev(). This
496 * removes the need for all of the following compatibility code.
498 #if !defined(HAVE_BIO_ALLOC_4ARG)
500 #ifdef HAVE_BIO_SET_DEV
501 #if defined(CONFIG_BLK_CGROUP) && defined(HAVE_BIO_SET_DEV_GPL_ONLY)
503 * The Linux 5.5 kernel updated percpu_ref_tryget() which is inlined by
504 * blkg_tryget() to use rcu_read_lock() instead of rcu_read_lock_sched().
505 * As a side effect the function was converted to GPL-only. Define our
506 * own version when needed which uses rcu_read_lock_sched().
508 * The Linux 5.17 kernel split linux/blk-cgroup.h into a private and a public
509 * part, moving blkg_tryget into the private one. Define our own version.
511 #if defined(HAVE_BLKG_TRYGET_GPL_ONLY) || !defined(HAVE_BLKG_TRYGET)
513 vdev_blkg_tryget(struct blkcg_gq
*blkg
)
515 struct percpu_ref
*ref
= &blkg
->refcnt
;
516 unsigned long __percpu
*count
;
519 rcu_read_lock_sched();
521 if (__ref_is_percpu(ref
, &count
)) {
522 this_cpu_inc(*count
);
525 #ifdef ZFS_PERCPU_REF_COUNT_IN_DATA
526 rc
= atomic_long_inc_not_zero(&ref
->data
->count
);
528 rc
= atomic_long_inc_not_zero(&ref
->count
);
532 rcu_read_unlock_sched();
537 #define vdev_blkg_tryget(bg) blkg_tryget(bg)
539 #ifdef HAVE_BIO_SET_DEV_MACRO
541 * The Linux 5.0 kernel updated the bio_set_dev() macro so it calls the
542 * GPL-only bio_associate_blkg() symbol thus inadvertently converting
543 * the entire macro. Provide a minimal version which always assigns the
544 * request queue's root_blkg to the bio.
547 vdev_bio_associate_blkg(struct bio
*bio
)
549 #if defined(HAVE_BIO_BDEV_DISK)
550 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
552 struct request_queue
*q
= bio
->bi_disk
->queue
;
555 ASSERT3P(q
, !=, NULL
);
556 ASSERT3P(bio
->bi_blkg
, ==, NULL
);
558 if (q
->root_blkg
&& vdev_blkg_tryget(q
->root_blkg
))
559 bio
->bi_blkg
= q
->root_blkg
;
562 #define bio_associate_blkg vdev_bio_associate_blkg
565 vdev_bio_set_dev(struct bio
*bio
, struct block_device
*bdev
)
567 #if defined(HAVE_BIO_BDEV_DISK)
568 struct request_queue
*q
= bdev
->bd_disk
->queue
;
570 struct request_queue
*q
= bio
->bi_disk
->queue
;
572 bio_clear_flag(bio
, BIO_REMAPPED
);
573 if (bio
->bi_bdev
!= bdev
)
574 bio_clear_flag(bio
, BIO_THROTTLED
);
577 ASSERT3P(q
, !=, NULL
);
578 ASSERT3P(bio
->bi_blkg
, ==, NULL
);
580 if (q
->root_blkg
&& vdev_blkg_tryget(q
->root_blkg
))
581 bio
->bi_blkg
= q
->root_blkg
;
583 #define bio_set_dev vdev_bio_set_dev
588 * Provide a bio_set_dev() helper macro for pre-Linux 4.14 kernels.
591 bio_set_dev(struct bio
*bio
, struct block_device
*bdev
)
595 #endif /* HAVE_BIO_SET_DEV */
596 #endif /* !HAVE_BIO_ALLOC_4ARG */
599 vdev_submit_bio(struct bio
*bio
)
601 struct bio_list
*bio_list
= current
->bio_list
;
602 current
->bio_list
= NULL
;
603 vdev_submit_bio_impl(bio
);
604 current
->bio_list
= bio_list
;
607 static inline struct bio
*
608 vdev_bio_alloc(struct block_device
*bdev
, gfp_t gfp_mask
,
609 unsigned short nr_vecs
)
613 #ifdef HAVE_BIO_ALLOC_4ARG
614 bio
= bio_alloc(bdev
, nr_vecs
, 0, gfp_mask
);
616 bio
= bio_alloc(gfp_mask
, nr_vecs
);
617 if (likely(bio
!= NULL
))
618 bio_set_dev(bio
, bdev
);
624 static inline unsigned int
625 vdev_bio_max_segs(zio_t
*zio
, int bio_size
, uint64_t abd_offset
)
627 unsigned long nr_segs
= abd_nr_pages_off(zio
->io_abd
,
628 bio_size
, abd_offset
);
630 #ifdef HAVE_BIO_MAX_SEGS
631 return (bio_max_segs(nr_segs
));
633 return (MIN(nr_segs
, BIO_MAX_PAGES
));
638 __vdev_disk_physio(struct block_device
*bdev
, zio_t
*zio
,
639 size_t io_size
, uint64_t io_offset
, int rw
, int flags
)
647 struct blk_plug plug
;
648 unsigned short nr_vecs
;
651 * Accessing outside the block device is never allowed.
653 if (io_offset
+ io_size
> bdev
->bd_inode
->i_size
) {
654 vdev_dbgmsg(zio
->io_vd
,
655 "Illegal access %llu size %llu, device size %llu",
656 (u_longlong_t
)io_offset
,
657 (u_longlong_t
)io_size
,
658 (u_longlong_t
)i_size_read(bdev
->bd_inode
));
659 return (SET_ERROR(EIO
));
663 dr
= vdev_disk_dio_alloc(bio_count
);
665 if (!(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)) &&
666 zio
->io_vd
->vdev_failfast
== B_TRUE
) {
667 bio_set_flags_failfast(bdev
, &flags
, zfs_vdev_failfast_mask
& 1,
668 zfs_vdev_failfast_mask
& 2, zfs_vdev_failfast_mask
& 4);
674 * Since bio's can have up to BIO_MAX_PAGES=256 iovec's, each of which
675 * is at least 512 bytes and at most PAGESIZE (typically 4K), one bio
676 * can cover at least 128KB and at most 1MB. When the required number
677 * of iovec's exceeds this, we are forced to break the IO in multiple
678 * bio's and wait for them all to complete. This is likely if the
679 * recordsize property is increased beyond 1MB. The default
680 * bio_count=16 should typically accommodate the maximum-size zio of
685 bio_offset
= io_offset
;
687 for (int i
= 0; i
<= dr
->dr_bio_count
; i
++) {
689 /* Finished constructing bio's for given buffer */
694 * If additional bio's are required, we have to retry, but
695 * this should be rare - see the comment above.
697 if (dr
->dr_bio_count
== i
) {
698 vdev_disk_dio_free(dr
);
703 nr_vecs
= vdev_bio_max_segs(zio
, bio_size
, abd_offset
);
704 dr
->dr_bio
[i
] = vdev_bio_alloc(bdev
, GFP_NOIO
, nr_vecs
);
705 if (unlikely(dr
->dr_bio
[i
] == NULL
)) {
706 vdev_disk_dio_free(dr
);
707 return (SET_ERROR(ENOMEM
));
710 /* Matching put called by vdev_disk_physio_completion */
711 vdev_disk_dio_get(dr
);
713 BIO_BI_SECTOR(dr
->dr_bio
[i
]) = bio_offset
>> 9;
714 dr
->dr_bio
[i
]->bi_end_io
= vdev_disk_physio_completion
;
715 dr
->dr_bio
[i
]->bi_private
= dr
;
716 bio_set_op_attrs(dr
->dr_bio
[i
], rw
, flags
);
718 /* Remaining size is returned to become the new size */
719 bio_size
= abd_bio_map_off(dr
->dr_bio
[i
], zio
->io_abd
,
720 bio_size
, abd_offset
);
722 /* Advance in buffer and construct another bio if needed */
723 abd_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
724 bio_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
727 /* Extra reference to protect dio_request during vdev_submit_bio */
728 vdev_disk_dio_get(dr
);
730 if (dr
->dr_bio_count
> 1)
731 blk_start_plug(&plug
);
733 /* Submit all bio's associated with this dio */
734 for (int i
= 0; i
< dr
->dr_bio_count
; i
++) {
736 vdev_submit_bio(dr
->dr_bio
[i
]);
739 if (dr
->dr_bio_count
> 1)
740 blk_finish_plug(&plug
);
742 vdev_disk_dio_put(dr
);
747 BIO_END_IO_PROTO(vdev_disk_io_flush_completion
, bio
, error
)
749 zio_t
*zio
= bio
->bi_private
;
750 #ifdef HAVE_1ARG_BIO_END_IO_T
751 zio
->io_error
= BIO_END_IO_ERROR(bio
);
753 zio
->io_error
= -error
;
756 if (zio
->io_error
&& (zio
->io_error
== EOPNOTSUPP
))
757 zio
->io_vd
->vdev_nowritecache
= B_TRUE
;
760 ASSERT3S(zio
->io_error
, >=, 0);
762 vdev_disk_error(zio
);
767 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
769 struct request_queue
*q
;
772 q
= bdev_get_queue(bdev
);
774 return (SET_ERROR(ENXIO
));
776 bio
= vdev_bio_alloc(bdev
, GFP_NOIO
, 0);
777 if (unlikely(bio
== NULL
))
778 return (SET_ERROR(ENOMEM
));
780 bio
->bi_end_io
= vdev_disk_io_flush_completion
;
781 bio
->bi_private
= zio
;
783 vdev_submit_bio(bio
);
784 invalidate_bdev(bdev
);
790 vdev_disk_io_trim(zio_t
*zio
)
792 vdev_t
*v
= zio
->io_vd
;
793 vdev_disk_t
*vd
= v
->vdev_tsd
;
795 #if defined(HAVE_BLKDEV_ISSUE_SECURE_ERASE)
796 if (zio
->io_trim_flags
& ZIO_TRIM_SECURE
) {
797 return (-blkdev_issue_secure_erase(vd
->vd_bdev
,
798 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
));
800 return (-blkdev_issue_discard(vd
->vd_bdev
,
801 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
));
803 #elif defined(HAVE_BLKDEV_ISSUE_DISCARD)
804 unsigned long trim_flags
= 0;
805 #if defined(BLKDEV_DISCARD_SECURE)
806 if (zio
->io_trim_flags
& ZIO_TRIM_SECURE
)
807 trim_flags
|= BLKDEV_DISCARD_SECURE
;
809 return (-blkdev_issue_discard(vd
->vd_bdev
,
810 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
, trim_flags
));
812 #error "Unsupported kernel"
817 vdev_disk_io_start(zio_t
*zio
)
819 vdev_t
*v
= zio
->io_vd
;
820 vdev_disk_t
*vd
= v
->vdev_tsd
;
824 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
825 * Nothing to be done here but return failure.
828 zio
->io_error
= ENXIO
;
833 rw_enter(&vd
->vd_lock
, RW_READER
);
836 * If the vdev is closed, it's likely due to a failed reopen and is
837 * in the UNAVAIL state. Nothing to be done here but return failure.
839 if (vd
->vd_bdev
== NULL
) {
840 rw_exit(&vd
->vd_lock
);
841 zio
->io_error
= ENXIO
;
846 switch (zio
->io_type
) {
849 if (!vdev_readable(v
)) {
850 rw_exit(&vd
->vd_lock
);
851 zio
->io_error
= SET_ERROR(ENXIO
);
856 switch (zio
->io_cmd
) {
857 case DKIOCFLUSHWRITECACHE
:
859 if (zfs_nocacheflush
)
862 if (v
->vdev_nowritecache
) {
863 zio
->io_error
= SET_ERROR(ENOTSUP
);
867 error
= vdev_disk_io_flush(vd
->vd_bdev
, zio
);
869 rw_exit(&vd
->vd_lock
);
873 zio
->io_error
= error
;
878 zio
->io_error
= SET_ERROR(ENOTSUP
);
881 rw_exit(&vd
->vd_lock
);
893 zio
->io_error
= vdev_disk_io_trim(zio
);
894 rw_exit(&vd
->vd_lock
);
899 rw_exit(&vd
->vd_lock
);
900 zio
->io_error
= SET_ERROR(ENOTSUP
);
905 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
906 error
= __vdev_disk_physio(vd
->vd_bdev
, zio
,
907 zio
->io_size
, zio
->io_offset
, rw
, 0);
908 rw_exit(&vd
->vd_lock
);
911 zio
->io_error
= error
;
918 vdev_disk_io_done(zio_t
*zio
)
921 * If the device returned EIO, we revalidate the media. If it is
922 * determined the media has changed this triggers the asynchronous
923 * removal of the device from the configuration.
925 if (zio
->io_error
== EIO
) {
926 vdev_t
*v
= zio
->io_vd
;
927 vdev_disk_t
*vd
= v
->vdev_tsd
;
929 if (!zfs_check_disk_status(vd
->vd_bdev
)) {
930 invalidate_bdev(vd
->vd_bdev
);
931 v
->vdev_remove_wanted
= B_TRUE
;
932 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
938 vdev_disk_hold(vdev_t
*vd
)
940 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
942 /* We must have a pathname, and it must be absolute. */
943 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
947 * Only prefetch path and devid info if the device has
950 if (vd
->vdev_tsd
!= NULL
)
956 vdev_disk_rele(vdev_t
*vd
)
958 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
960 /* XXX: Implement me as a vnode rele for the device */
963 vdev_ops_t vdev_disk_ops
= {
964 .vdev_op_init
= NULL
,
965 .vdev_op_fini
= NULL
,
966 .vdev_op_open
= vdev_disk_open
,
967 .vdev_op_close
= vdev_disk_close
,
968 .vdev_op_asize
= vdev_default_asize
,
969 .vdev_op_min_asize
= vdev_default_min_asize
,
970 .vdev_op_min_alloc
= NULL
,
971 .vdev_op_io_start
= vdev_disk_io_start
,
972 .vdev_op_io_done
= vdev_disk_io_done
,
973 .vdev_op_state_change
= NULL
,
974 .vdev_op_need_resilver
= NULL
,
975 .vdev_op_hold
= vdev_disk_hold
,
976 .vdev_op_rele
= vdev_disk_rele
,
977 .vdev_op_remap
= NULL
,
978 .vdev_op_xlate
= vdev_default_xlate
,
979 .vdev_op_rebuild_asize
= NULL
,
980 .vdev_op_metaslab_init
= NULL
,
981 .vdev_op_config_generate
= NULL
,
982 .vdev_op_nparity
= NULL
,
983 .vdev_op_ndisks
= NULL
,
984 .vdev_op_type
= VDEV_TYPE_DISK
, /* name of this vdev type */
985 .vdev_op_leaf
= B_TRUE
, /* leaf vdev */
986 .vdev_op_kobj_evt_post
= vdev_disk_kobj_evt_post
990 * The zfs_vdev_scheduler module option has been deprecated. Setting this
991 * value no longer has any effect. It has not yet been entirely removed
992 * to allow the module to be loaded if this option is specified in the
993 * /etc/modprobe.d/zfs.conf file. The following warning will be logged.
996 param_set_vdev_scheduler(const char *val
, zfs_kernel_param_t
*kp
)
998 int error
= param_set_charp(val
, kp
);
1000 printk(KERN_INFO
"The 'zfs_vdev_scheduler' module option "
1001 "is not supported.\n");
1007 static const char *zfs_vdev_scheduler
= "unused";
1008 module_param_call(zfs_vdev_scheduler
, param_set_vdev_scheduler
,
1009 param_get_charp
, &zfs_vdev_scheduler
, 0644);
1010 MODULE_PARM_DESC(zfs_vdev_scheduler
, "I/O scheduler");
1013 param_set_min_auto_ashift(const char *buf
, zfs_kernel_param_t
*kp
)
1018 error
= kstrtouint(buf
, 0, &val
);
1020 return (SET_ERROR(error
));
1022 if (val
< ASHIFT_MIN
|| val
> zfs_vdev_max_auto_ashift
)
1023 return (SET_ERROR(-EINVAL
));
1025 error
= param_set_uint(buf
, kp
);
1027 return (SET_ERROR(error
));
1033 param_set_max_auto_ashift(const char *buf
, zfs_kernel_param_t
*kp
)
1038 error
= kstrtouint(buf
, 0, &val
);
1040 return (SET_ERROR(error
));
1042 if (val
> ASHIFT_MAX
|| val
< zfs_vdev_min_auto_ashift
)
1043 return (SET_ERROR(-EINVAL
));
1045 error
= param_set_uint(buf
, kp
);
1047 return (SET_ERROR(error
));
1052 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, open_timeout_ms
, UINT
, ZMOD_RW
,
1053 "Timeout before determining that a device is missing");
1055 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, failfast_mask
, UINT
, ZMOD_RW
,
1056 "Defines failfast mask: 1 - device, 2 - transport, 4 - driver");