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>
45 * Linux 6.8.x uses a bdev_handle as an instance/refcount for an underlying
46 * block_device. Since it carries the block_device inside, its convenient to
47 * just use the handle as a proxy. For pre-6.8, we just emulate this with
48 * a cast, since we don't need any of the other fields inside the handle.
50 #ifdef HAVE_BDEV_OPEN_BY_PATH
51 typedef struct bdev_handle zfs_bdev_handle_t
;
52 #define BDH_BDEV(bdh) ((bdh)->bdev)
53 #define BDH_IS_ERR(bdh) (IS_ERR(bdh))
54 #define BDH_PTR_ERR(bdh) (PTR_ERR(bdh))
55 #define BDH_ERR_PTR(err) (ERR_PTR(err))
57 typedef void zfs_bdev_handle_t
;
58 #define BDH_BDEV(bdh) ((struct block_device *)bdh)
59 #define BDH_IS_ERR(bdh) (IS_ERR(BDH_BDEV(bdh)))
60 #define BDH_PTR_ERR(bdh) (PTR_ERR(BDH_BDEV(bdh)))
61 #define BDH_ERR_PTR(err) (ERR_PTR(err))
64 typedef struct vdev_disk
{
65 zfs_bdev_handle_t
*vd_bdh
;
70 * Unique identifier for the exclusive vdev holder.
72 static void *zfs_vdev_holder
= VDEV_HOLDER
;
75 * Wait up to zfs_vdev_open_timeout_ms milliseconds before determining the
76 * device is missing. The missing path may be transient since the links
77 * can be briefly removed and recreated in response to udev events.
79 static uint_t zfs_vdev_open_timeout_ms
= 1000;
82 * Size of the "reserved" partition, in blocks.
84 #define EFI_MIN_RESV_SIZE (16 * 1024)
87 * Virtual device vector for disks.
89 typedef struct dio_request
{
90 zio_t
*dr_zio
; /* Parent ZIO */
91 atomic_t dr_ref
; /* References */
92 int dr_error
; /* Bio error */
93 int dr_bio_count
; /* Count of bio's */
94 struct bio
*dr_bio
[]; /* Attached bio's */
98 * BIO request failfast mask.
101 static unsigned int zfs_vdev_failfast_mask
= 1;
103 #ifdef HAVE_BLK_MODE_T
108 vdev_bdev_mode(spa_mode_t spa_mode
, boolean_t exclusive
)
110 #ifdef HAVE_BLK_MODE_T
113 if (spa_mode
& SPA_MODE_READ
)
114 mode
|= BLK_OPEN_READ
;
116 if (spa_mode
& SPA_MODE_WRITE
)
117 mode
|= BLK_OPEN_WRITE
;
120 mode
|= BLK_OPEN_EXCL
;
124 if (spa_mode
& SPA_MODE_READ
)
127 if (spa_mode
& SPA_MODE_WRITE
)
138 * Returns the usable capacity (in bytes) for the partition or disk.
141 bdev_capacity(struct block_device
*bdev
)
143 return (i_size_read(bdev
->bd_inode
));
146 #if !defined(HAVE_BDEV_WHOLE)
147 static inline struct block_device
*
148 bdev_whole(struct block_device
*bdev
)
150 return (bdev
->bd_contains
);
154 #if defined(HAVE_BDEVNAME)
155 #define vdev_bdevname(bdev, name) bdevname(bdev, name)
158 vdev_bdevname(struct block_device
*bdev
, char *name
)
160 snprintf(name
, BDEVNAME_SIZE
, "%pg", bdev
);
165 * Returns the maximum expansion capacity of the block device (in bytes).
167 * It is possible to expand a vdev when it has been created as a wholedisk
168 * and the containing block device has increased in capacity. Or when the
169 * partition containing the pool has been manually increased in size.
171 * This function is only responsible for calculating the potential expansion
172 * size so it can be reported by 'zpool list'. The efi_use_whole_disk() is
173 * responsible for verifying the expected partition layout in the wholedisk
174 * case, and updating the partition table if appropriate. Once the partition
175 * size has been increased the additional capacity will be visible using
178 * The returned maximum expansion capacity is always expected to be larger, or
179 * at the very least equal, to its usable capacity to prevent overestimating
180 * the pool expandsize.
183 bdev_max_capacity(struct block_device
*bdev
, uint64_t wholedisk
)
188 if (wholedisk
&& bdev
!= bdev_whole(bdev
)) {
190 * When reporting maximum expansion capacity for a wholedisk
191 * deduct any capacity which is expected to be lost due to
192 * alignment restrictions. Over reporting this value isn't
193 * harmful and would only result in slightly less capacity
194 * than expected post expansion.
195 * The estimated available space may be slightly smaller than
196 * bdev_capacity() for devices where the number of sectors is
197 * not a multiple of the alignment size and the partition layout
198 * is keeping less than PARTITION_END_ALIGNMENT bytes after the
199 * "reserved" EFI partition: in such cases return the device
202 available
= i_size_read(bdev_whole(bdev
)->bd_inode
) -
203 ((EFI_MIN_RESV_SIZE
+ NEW_START_BLOCK
+
204 PARTITION_END_ALIGNMENT
) << SECTOR_BITS
);
205 psize
= MAX(available
, bdev_capacity(bdev
));
207 psize
= bdev_capacity(bdev
);
214 vdev_disk_error(zio_t
*zio
)
217 * This function can be called in interrupt context, for instance while
218 * handling IRQs coming from a misbehaving disk device; use printk()
219 * which is safe from any context.
221 printk(KERN_WARNING
"zio pool=%s vdev=%s error=%d type=%d "
222 "offset=%llu size=%llu flags=%llu\n", spa_name(zio
->io_spa
),
223 zio
->io_vd
->vdev_path
, zio
->io_error
, zio
->io_type
,
224 (u_longlong_t
)zio
->io_offset
, (u_longlong_t
)zio
->io_size
,
229 vdev_disk_kobj_evt_post(vdev_t
*v
)
231 vdev_disk_t
*vd
= v
->vdev_tsd
;
232 if (vd
&& vd
->vd_bdh
) {
233 spl_signal_kobj_evt(BDH_BDEV(vd
->vd_bdh
));
235 vdev_dbgmsg(v
, "vdev_disk_t is NULL for VDEV:%s\n",
240 static zfs_bdev_handle_t
*
241 vdev_blkdev_get_by_path(const char *path
, spa_mode_t mode
, void *holder
)
243 #if defined(HAVE_BDEV_OPEN_BY_PATH)
244 return (bdev_open_by_path(path
,
245 vdev_bdev_mode(mode
, B_TRUE
), holder
, NULL
));
246 #elif defined(HAVE_BLKDEV_GET_BY_PATH_4ARG)
247 return (blkdev_get_by_path(path
,
248 vdev_bdev_mode(mode
, B_TRUE
), holder
, NULL
));
250 return (blkdev_get_by_path(path
,
251 vdev_bdev_mode(mode
, B_TRUE
), holder
));
256 vdev_blkdev_put(zfs_bdev_handle_t
*bdh
, spa_mode_t mode
, void *holder
)
258 #if defined(HAVE_BDEV_RELEASE)
259 return (bdev_release(bdh
));
260 #elif defined(HAVE_BLKDEV_PUT_HOLDER)
261 return (blkdev_put(BDH_BDEV(bdh
), holder
));
263 return (blkdev_put(BDH_BDEV(bdh
),
264 vdev_bdev_mode(mode
, B_TRUE
)));
269 vdev_disk_open(vdev_t
*v
, uint64_t *psize
, uint64_t *max_psize
,
270 uint64_t *logical_ashift
, uint64_t *physical_ashift
)
272 zfs_bdev_handle_t
*bdh
;
273 #ifdef HAVE_BLK_MODE_T
274 blk_mode_t mode
= vdev_bdev_mode(spa_mode(v
->vdev_spa
), B_FALSE
);
276 fmode_t mode
= vdev_bdev_mode(spa_mode(v
->vdev_spa
), B_FALSE
);
278 hrtime_t timeout
= MSEC2NSEC(zfs_vdev_open_timeout_ms
);
281 /* Must have a pathname and it must be absolute. */
282 if (v
->vdev_path
== NULL
|| v
->vdev_path
[0] != '/') {
283 v
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
284 vdev_dbgmsg(v
, "invalid vdev_path");
285 return (SET_ERROR(EINVAL
));
289 * Reopen the device if it is currently open. When expanding a
290 * partition force re-scanning the partition table if userland
291 * did not take care of this already. We need to do this while closed
292 * in order to get an accurate updated block device size. Then
293 * since udev may need to recreate the device links increase the
294 * open retry timeout before reporting the device as unavailable.
298 char disk_name
[BDEVNAME_SIZE
+ 6] = "/dev/";
299 boolean_t reread_part
= B_FALSE
;
301 rw_enter(&vd
->vd_lock
, RW_WRITER
);
306 struct block_device
*bdev
= BDH_BDEV(bdh
);
307 if (v
->vdev_expanding
&& bdev
!= bdev_whole(bdev
)) {
308 vdev_bdevname(bdev_whole(bdev
), disk_name
+ 5);
310 * If userland has BLKPG_RESIZE_PARTITION,
311 * then it should have updated the partition
312 * table already. We can detect this by
313 * comparing our current physical size
314 * with that of the device. If they are
315 * the same, then we must not have
316 * BLKPG_RESIZE_PARTITION or it failed to
317 * update the partition table online. We
318 * fallback to rescanning the partition
319 * table from the kernel below. However,
320 * if the capacity already reflects the
321 * updated partition, then we skip
322 * rescanning the partition table here.
324 if (v
->vdev_psize
== bdev_capacity(bdev
))
325 reread_part
= B_TRUE
;
328 vdev_blkdev_put(bdh
, mode
, zfs_vdev_holder
);
332 bdh
= vdev_blkdev_get_by_path(disk_name
, mode
,
334 if (!BDH_IS_ERR(bdh
)) {
336 vdev_bdev_reread_part(BDH_BDEV(bdh
));
337 vdev_blkdev_put(bdh
, mode
, zfs_vdev_holder
);
340 zfs_vdev_open_timeout_ms
* 2);
345 vd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
347 rw_init(&vd
->vd_lock
, NULL
, RW_DEFAULT
, NULL
);
348 rw_enter(&vd
->vd_lock
, RW_WRITER
);
352 * Devices are always opened by the path provided at configuration
353 * time. This means that if the provided path is a udev by-id path
354 * then drives may be re-cabled without an issue. If the provided
355 * path is a udev by-path path, then the physical location information
356 * will be preserved. This can be critical for more complicated
357 * configurations where drives are located in specific physical
358 * locations to maximize the systems tolerance to component failure.
360 * Alternatively, you can provide your own udev rule to flexibly map
361 * the drives as you see fit. It is not advised that you use the
362 * /dev/[hd]d devices which may be reordered due to probing order.
363 * Devices in the wrong locations will be detected by the higher
364 * level vdev validation.
366 * The specified paths may be briefly removed and recreated in
367 * response to udev events. This should be exceptionally unlikely
368 * because the zpool command makes every effort to verify these paths
369 * have already settled prior to reaching this point. Therefore,
370 * a ENOENT failure at this point is highly likely to be transient
371 * and it is reasonable to sleep and retry before giving up. In
372 * practice delays have been observed to be on the order of 100ms.
374 * When ERESTARTSYS is returned it indicates the block device is
375 * a zvol which could not be opened due to the deadlock detection
376 * logic in zvol_open(). Extend the timeout and retry the open
377 * subsequent attempts are expected to eventually succeed.
379 hrtime_t start
= gethrtime();
380 bdh
= BDH_ERR_PTR(-ENXIO
);
381 while (BDH_IS_ERR(bdh
) && ((gethrtime() - start
) < timeout
)) {
382 bdh
= vdev_blkdev_get_by_path(v
->vdev_path
, mode
,
384 if (unlikely(BDH_PTR_ERR(bdh
) == -ENOENT
)) {
386 * There is no point of waiting since device is removed
392 schedule_timeout(MSEC_TO_TICK(10));
393 } else if (unlikely(BDH_PTR_ERR(bdh
) == -ERESTARTSYS
)) {
394 timeout
= MSEC2NSEC(zfs_vdev_open_timeout_ms
* 10);
396 } else if (BDH_IS_ERR(bdh
)) {
401 if (BDH_IS_ERR(bdh
)) {
402 int error
= -BDH_PTR_ERR(bdh
);
403 vdev_dbgmsg(v
, "open error=%d timeout=%llu/%llu", error
,
404 (u_longlong_t
)(gethrtime() - start
),
405 (u_longlong_t
)timeout
);
408 rw_exit(&vd
->vd_lock
);
409 return (SET_ERROR(error
));
413 rw_exit(&vd
->vd_lock
);
416 struct block_device
*bdev
= BDH_BDEV(vd
->vd_bdh
);
418 /* Determine the physical block size */
419 int physical_block_size
= bdev_physical_block_size(bdev
);
421 /* Determine the logical block size */
422 int logical_block_size
= bdev_logical_block_size(bdev
);
424 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
425 v
->vdev_nowritecache
= B_FALSE
;
427 /* Set when device reports it supports TRIM. */
428 v
->vdev_has_trim
= bdev_discard_supported(bdev
);
430 /* Set when device reports it supports secure TRIM. */
431 v
->vdev_has_securetrim
= bdev_secure_discard_supported(bdev
);
433 /* Inform the ZIO pipeline that we are non-rotational */
434 v
->vdev_nonrot
= blk_queue_nonrot(bdev_get_queue(bdev
));
436 /* Physical volume size in bytes for the partition */
437 *psize
= bdev_capacity(bdev
);
439 /* Physical volume size in bytes including possible expansion space */
440 *max_psize
= bdev_max_capacity(bdev
, v
->vdev_wholedisk
);
442 /* Based on the minimum sector size set the block size */
443 *physical_ashift
= highbit64(MAX(physical_block_size
,
444 SPA_MINBLOCKSIZE
)) - 1;
446 *logical_ashift
= highbit64(MAX(logical_block_size
,
447 SPA_MINBLOCKSIZE
)) - 1;
453 vdev_disk_close(vdev_t
*v
)
455 vdev_disk_t
*vd
= v
->vdev_tsd
;
457 if (v
->vdev_reopening
|| vd
== NULL
)
460 if (vd
->vd_bdh
!= NULL
) {
461 vdev_blkdev_put(vd
->vd_bdh
, spa_mode(v
->vdev_spa
),
465 rw_destroy(&vd
->vd_lock
);
466 kmem_free(vd
, sizeof (vdev_disk_t
));
470 static dio_request_t
*
471 vdev_disk_dio_alloc(int bio_count
)
473 dio_request_t
*dr
= kmem_zalloc(sizeof (dio_request_t
) +
474 sizeof (struct bio
*) * bio_count
, KM_SLEEP
);
475 atomic_set(&dr
->dr_ref
, 0);
476 dr
->dr_bio_count
= bio_count
;
479 for (int i
= 0; i
< dr
->dr_bio_count
; i
++)
480 dr
->dr_bio
[i
] = NULL
;
486 vdev_disk_dio_free(dio_request_t
*dr
)
490 for (i
= 0; i
< dr
->dr_bio_count
; i
++)
492 bio_put(dr
->dr_bio
[i
]);
494 kmem_free(dr
, sizeof (dio_request_t
) +
495 sizeof (struct bio
*) * dr
->dr_bio_count
);
499 vdev_disk_dio_get(dio_request_t
*dr
)
501 atomic_inc(&dr
->dr_ref
);
505 vdev_disk_dio_put(dio_request_t
*dr
)
507 int rc
= atomic_dec_return(&dr
->dr_ref
);
510 * Free the dio_request when the last reference is dropped and
511 * ensure zio_interpret is called only once with the correct zio
514 zio_t
*zio
= dr
->dr_zio
;
515 int error
= dr
->dr_error
;
517 vdev_disk_dio_free(dr
);
520 zio
->io_error
= error
;
521 ASSERT3S(zio
->io_error
, >=, 0);
523 vdev_disk_error(zio
);
525 zio_delay_interrupt(zio
);
530 BIO_END_IO_PROTO(vdev_disk_physio_completion
, bio
, error
)
532 dio_request_t
*dr
= bio
->bi_private
;
534 if (dr
->dr_error
== 0) {
535 #ifdef HAVE_1ARG_BIO_END_IO_T
536 dr
->dr_error
= BIO_END_IO_ERROR(bio
);
539 dr
->dr_error
= -(error
);
540 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
545 /* Drop reference acquired by __vdev_disk_physio */
546 vdev_disk_dio_put(dr
);
550 vdev_submit_bio_impl(struct bio
*bio
)
552 #ifdef HAVE_1ARG_SUBMIT_BIO
553 (void) submit_bio(bio
);
555 (void) submit_bio(bio_data_dir(bio
), bio
);
560 * preempt_schedule_notrace is GPL-only which breaks the ZFS build, so
561 * replace it with preempt_schedule under the following condition:
563 #if defined(CONFIG_ARM64) && \
564 defined(CONFIG_PREEMPTION) && \
565 defined(CONFIG_BLK_CGROUP)
566 #define preempt_schedule_notrace(x) preempt_schedule(x)
570 * As for the Linux 5.18 kernel bio_alloc() expects a block_device struct
571 * as an argument removing the need to set it with bio_set_dev(). This
572 * removes the need for all of the following compatibility code.
574 #if !defined(HAVE_BIO_ALLOC_4ARG)
576 #ifdef HAVE_BIO_SET_DEV
577 #if defined(CONFIG_BLK_CGROUP) && defined(HAVE_BIO_SET_DEV_GPL_ONLY)
579 * The Linux 5.5 kernel updated percpu_ref_tryget() which is inlined by
580 * blkg_tryget() to use rcu_read_lock() instead of rcu_read_lock_sched().
581 * As a side effect the function was converted to GPL-only. Define our
582 * own version when needed which uses rcu_read_lock_sched().
584 * The Linux 5.17 kernel split linux/blk-cgroup.h into a private and a public
585 * part, moving blkg_tryget into the private one. Define our own version.
587 #if defined(HAVE_BLKG_TRYGET_GPL_ONLY) || !defined(HAVE_BLKG_TRYGET)
589 vdev_blkg_tryget(struct blkcg_gq
*blkg
)
591 struct percpu_ref
*ref
= &blkg
->refcnt
;
592 unsigned long __percpu
*count
;
595 rcu_read_lock_sched();
597 if (__ref_is_percpu(ref
, &count
)) {
598 this_cpu_inc(*count
);
601 #ifdef ZFS_PERCPU_REF_COUNT_IN_DATA
602 rc
= atomic_long_inc_not_zero(&ref
->data
->count
);
604 rc
= atomic_long_inc_not_zero(&ref
->count
);
608 rcu_read_unlock_sched();
613 #define vdev_blkg_tryget(bg) blkg_tryget(bg)
615 #ifdef HAVE_BIO_SET_DEV_MACRO
617 * The Linux 5.0 kernel updated the bio_set_dev() macro so it calls the
618 * GPL-only bio_associate_blkg() symbol thus inadvertently converting
619 * the entire macro. Provide a minimal version which always assigns the
620 * request queue's root_blkg to the bio.
623 vdev_bio_associate_blkg(struct bio
*bio
)
625 #if defined(HAVE_BIO_BDEV_DISK)
626 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
628 struct request_queue
*q
= bio
->bi_disk
->queue
;
631 ASSERT3P(q
, !=, NULL
);
632 ASSERT3P(bio
->bi_blkg
, ==, NULL
);
634 if (q
->root_blkg
&& vdev_blkg_tryget(q
->root_blkg
))
635 bio
->bi_blkg
= q
->root_blkg
;
638 #define bio_associate_blkg vdev_bio_associate_blkg
641 vdev_bio_set_dev(struct bio
*bio
, struct block_device
*bdev
)
643 #if defined(HAVE_BIO_BDEV_DISK)
644 struct request_queue
*q
= bdev
->bd_disk
->queue
;
646 struct request_queue
*q
= bio
->bi_disk
->queue
;
648 bio_clear_flag(bio
, BIO_REMAPPED
);
649 if (bio
->bi_bdev
!= bdev
)
650 bio_clear_flag(bio
, BIO_THROTTLED
);
653 ASSERT3P(q
, !=, NULL
);
654 ASSERT3P(bio
->bi_blkg
, ==, NULL
);
656 if (q
->root_blkg
&& vdev_blkg_tryget(q
->root_blkg
))
657 bio
->bi_blkg
= q
->root_blkg
;
659 #define bio_set_dev vdev_bio_set_dev
664 * Provide a bio_set_dev() helper macro for pre-Linux 4.14 kernels.
667 bio_set_dev(struct bio
*bio
, struct block_device
*bdev
)
671 #endif /* HAVE_BIO_SET_DEV */
672 #endif /* !HAVE_BIO_ALLOC_4ARG */
675 vdev_submit_bio(struct bio
*bio
)
677 struct bio_list
*bio_list
= current
->bio_list
;
678 current
->bio_list
= NULL
;
679 vdev_submit_bio_impl(bio
);
680 current
->bio_list
= bio_list
;
683 static inline struct bio
*
684 vdev_bio_alloc(struct block_device
*bdev
, gfp_t gfp_mask
,
685 unsigned short nr_vecs
)
689 #ifdef HAVE_BIO_ALLOC_4ARG
690 bio
= bio_alloc(bdev
, nr_vecs
, 0, gfp_mask
);
692 bio
= bio_alloc(gfp_mask
, nr_vecs
);
693 if (likely(bio
!= NULL
))
694 bio_set_dev(bio
, bdev
);
700 static inline unsigned int
701 vdev_bio_max_segs(zio_t
*zio
, int bio_size
, uint64_t abd_offset
)
703 unsigned long nr_segs
= abd_nr_pages_off(zio
->io_abd
,
704 bio_size
, abd_offset
);
706 #ifdef HAVE_BIO_MAX_SEGS
707 return (bio_max_segs(nr_segs
));
709 return (MIN(nr_segs
, BIO_MAX_PAGES
));
714 __vdev_disk_physio(struct block_device
*bdev
, zio_t
*zio
,
715 size_t io_size
, uint64_t io_offset
, int rw
, int flags
)
723 struct blk_plug plug
;
724 unsigned short nr_vecs
;
727 * Accessing outside the block device is never allowed.
729 if (io_offset
+ io_size
> bdev
->bd_inode
->i_size
) {
730 vdev_dbgmsg(zio
->io_vd
,
731 "Illegal access %llu size %llu, device size %llu",
732 (u_longlong_t
)io_offset
,
733 (u_longlong_t
)io_size
,
734 (u_longlong_t
)i_size_read(bdev
->bd_inode
));
735 return (SET_ERROR(EIO
));
739 dr
= vdev_disk_dio_alloc(bio_count
);
741 if (!(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)) &&
742 zio
->io_vd
->vdev_failfast
== B_TRUE
) {
743 bio_set_flags_failfast(bdev
, &flags
, zfs_vdev_failfast_mask
& 1,
744 zfs_vdev_failfast_mask
& 2, zfs_vdev_failfast_mask
& 4);
750 * Since bio's can have up to BIO_MAX_PAGES=256 iovec's, each of which
751 * is at least 512 bytes and at most PAGESIZE (typically 4K), one bio
752 * can cover at least 128KB and at most 1MB. When the required number
753 * of iovec's exceeds this, we are forced to break the IO in multiple
754 * bio's and wait for them all to complete. This is likely if the
755 * recordsize property is increased beyond 1MB. The default
756 * bio_count=16 should typically accommodate the maximum-size zio of
761 bio_offset
= io_offset
;
763 for (int i
= 0; i
<= dr
->dr_bio_count
; i
++) {
765 /* Finished constructing bio's for given buffer */
770 * If additional bio's are required, we have to retry, but
771 * this should be rare - see the comment above.
773 if (dr
->dr_bio_count
== i
) {
774 vdev_disk_dio_free(dr
);
779 nr_vecs
= vdev_bio_max_segs(zio
, bio_size
, abd_offset
);
780 dr
->dr_bio
[i
] = vdev_bio_alloc(bdev
, GFP_NOIO
, nr_vecs
);
781 if (unlikely(dr
->dr_bio
[i
] == NULL
)) {
782 vdev_disk_dio_free(dr
);
783 return (SET_ERROR(ENOMEM
));
786 /* Matching put called by vdev_disk_physio_completion */
787 vdev_disk_dio_get(dr
);
789 BIO_BI_SECTOR(dr
->dr_bio
[i
]) = bio_offset
>> 9;
790 dr
->dr_bio
[i
]->bi_end_io
= vdev_disk_physio_completion
;
791 dr
->dr_bio
[i
]->bi_private
= dr
;
792 bio_set_op_attrs(dr
->dr_bio
[i
], rw
, flags
);
794 /* Remaining size is returned to become the new size */
795 bio_size
= abd_bio_map_off(dr
->dr_bio
[i
], zio
->io_abd
,
796 bio_size
, abd_offset
);
798 /* Advance in buffer and construct another bio if needed */
799 abd_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
800 bio_offset
+= BIO_BI_SIZE(dr
->dr_bio
[i
]);
803 /* Extra reference to protect dio_request during vdev_submit_bio */
804 vdev_disk_dio_get(dr
);
806 if (dr
->dr_bio_count
> 1)
807 blk_start_plug(&plug
);
809 /* Submit all bio's associated with this dio */
810 for (int i
= 0; i
< dr
->dr_bio_count
; i
++) {
812 vdev_submit_bio(dr
->dr_bio
[i
]);
815 if (dr
->dr_bio_count
> 1)
816 blk_finish_plug(&plug
);
818 vdev_disk_dio_put(dr
);
823 BIO_END_IO_PROTO(vdev_disk_io_flush_completion
, bio
, error
)
825 zio_t
*zio
= bio
->bi_private
;
826 #ifdef HAVE_1ARG_BIO_END_IO_T
827 zio
->io_error
= BIO_END_IO_ERROR(bio
);
829 zio
->io_error
= -error
;
832 if (zio
->io_error
&& (zio
->io_error
== EOPNOTSUPP
))
833 zio
->io_vd
->vdev_nowritecache
= B_TRUE
;
836 ASSERT3S(zio
->io_error
, >=, 0);
838 vdev_disk_error(zio
);
843 vdev_disk_io_flush(struct block_device
*bdev
, zio_t
*zio
)
845 struct request_queue
*q
;
848 q
= bdev_get_queue(bdev
);
850 return (SET_ERROR(ENXIO
));
852 bio
= vdev_bio_alloc(bdev
, GFP_NOIO
, 0);
853 if (unlikely(bio
== NULL
))
854 return (SET_ERROR(ENOMEM
));
856 bio
->bi_end_io
= vdev_disk_io_flush_completion
;
857 bio
->bi_private
= zio
;
859 vdev_submit_bio(bio
);
860 invalidate_bdev(bdev
);
865 #if defined(HAVE_BLKDEV_ISSUE_SECURE_ERASE) || \
866 defined(HAVE_BLKDEV_ISSUE_DISCARD_ASYNC)
867 BIO_END_IO_PROTO(vdev_disk_discard_end_io
, bio
, error
)
869 zio_t
*zio
= bio
->bi_private
;
870 #ifdef HAVE_1ARG_BIO_END_IO_T
871 zio
->io_error
= BIO_END_IO_ERROR(bio
);
873 zio
->io_error
= -error
;
877 vdev_disk_error(zio
);
882 vdev_issue_discard_trim(zio_t
*zio
, unsigned long flags
)
885 struct bio
*bio
= NULL
;
887 #if defined(BLKDEV_DISCARD_SECURE)
888 ret
= - __blkdev_issue_discard(
889 BDH_BDEV(((vdev_disk_t
*)zio
->io_vd
->vdev_tsd
)->vd_bdh
),
890 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
, flags
, &bio
);
893 ret
= - __blkdev_issue_discard(
894 BDH_BDEV(((vdev_disk_t
*)zio
->io_vd
->vdev_tsd
)->vd_bdh
),
895 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
, &bio
);
898 bio
->bi_private
= zio
;
899 bio
->bi_end_io
= vdev_disk_discard_end_io
;
900 vdev_submit_bio(bio
);
907 vdev_disk_io_trim(zio_t
*zio
)
909 unsigned long trim_flags
= 0;
910 if (zio
->io_trim_flags
& ZIO_TRIM_SECURE
) {
911 #if defined(HAVE_BLKDEV_ISSUE_SECURE_ERASE)
912 return (-blkdev_issue_secure_erase(
913 BDH_BDEV(((vdev_disk_t
*)zio
->io_vd
->vdev_tsd
)->vd_bdh
),
914 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
));
915 #elif defined(BLKDEV_DISCARD_SECURE)
916 trim_flags
|= BLKDEV_DISCARD_SECURE
;
919 #if defined(HAVE_BLKDEV_ISSUE_SECURE_ERASE) || \
920 defined(HAVE_BLKDEV_ISSUE_DISCARD_ASYNC)
921 return (vdev_issue_discard_trim(zio
, trim_flags
));
922 #elif defined(HAVE_BLKDEV_ISSUE_DISCARD)
923 return (-blkdev_issue_discard(
924 BDH_BDEV(((vdev_disk_t
*)zio
->io_vd
->vdev_tsd
)->vd_bdh
),
925 zio
->io_offset
>> 9, zio
->io_size
>> 9, GFP_NOFS
, trim_flags
));
927 #error "Unsupported kernel"
932 vdev_disk_io_start(zio_t
*zio
)
934 vdev_t
*v
= zio
->io_vd
;
935 vdev_disk_t
*vd
= v
->vdev_tsd
;
939 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
940 * Nothing to be done here but return failure.
943 zio
->io_error
= ENXIO
;
948 rw_enter(&vd
->vd_lock
, RW_READER
);
951 * If the vdev is closed, it's likely due to a failed reopen and is
952 * in the UNAVAIL state. Nothing to be done here but return failure.
954 if (vd
->vd_bdh
== NULL
) {
955 rw_exit(&vd
->vd_lock
);
956 zio
->io_error
= ENXIO
;
961 switch (zio
->io_type
) {
964 if (!vdev_readable(v
)) {
965 rw_exit(&vd
->vd_lock
);
966 zio
->io_error
= SET_ERROR(ENXIO
);
971 switch (zio
->io_cmd
) {
972 case DKIOCFLUSHWRITECACHE
:
974 if (zfs_nocacheflush
)
977 if (v
->vdev_nowritecache
) {
978 zio
->io_error
= SET_ERROR(ENOTSUP
);
982 error
= vdev_disk_io_flush(BDH_BDEV(vd
->vd_bdh
), zio
);
984 rw_exit(&vd
->vd_lock
);
988 zio
->io_error
= error
;
993 zio
->io_error
= SET_ERROR(ENOTSUP
);
996 rw_exit(&vd
->vd_lock
);
1008 zio
->io_error
= vdev_disk_io_trim(zio
);
1009 rw_exit(&vd
->vd_lock
);
1010 #if defined(HAVE_BLKDEV_ISSUE_SECURE_ERASE)
1011 if (zio
->io_trim_flags
& ZIO_TRIM_SECURE
)
1013 #elif defined(HAVE_BLKDEV_ISSUE_DISCARD)
1019 rw_exit(&vd
->vd_lock
);
1020 zio
->io_error
= SET_ERROR(ENOTSUP
);
1025 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
1026 error
= __vdev_disk_physio(BDH_BDEV(vd
->vd_bdh
), zio
,
1027 zio
->io_size
, zio
->io_offset
, rw
, 0);
1028 rw_exit(&vd
->vd_lock
);
1031 zio
->io_error
= error
;
1038 vdev_disk_io_done(zio_t
*zio
)
1041 * If the device returned EIO, we revalidate the media. If it is
1042 * determined the media has changed this triggers the asynchronous
1043 * removal of the device from the configuration.
1045 if (zio
->io_error
== EIO
) {
1046 vdev_t
*v
= zio
->io_vd
;
1047 vdev_disk_t
*vd
= v
->vdev_tsd
;
1049 if (!zfs_check_disk_status(BDH_BDEV(vd
->vd_bdh
))) {
1050 invalidate_bdev(BDH_BDEV(vd
->vd_bdh
));
1051 v
->vdev_remove_wanted
= B_TRUE
;
1052 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
1058 vdev_disk_hold(vdev_t
*vd
)
1060 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
1062 /* We must have a pathname, and it must be absolute. */
1063 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
1067 * Only prefetch path and devid info if the device has
1068 * never been opened.
1070 if (vd
->vdev_tsd
!= NULL
)
1076 vdev_disk_rele(vdev_t
*vd
)
1078 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
1080 /* XXX: Implement me as a vnode rele for the device */
1083 vdev_ops_t vdev_disk_ops
= {
1084 .vdev_op_init
= NULL
,
1085 .vdev_op_fini
= NULL
,
1086 .vdev_op_open
= vdev_disk_open
,
1087 .vdev_op_close
= vdev_disk_close
,
1088 .vdev_op_asize
= vdev_default_asize
,
1089 .vdev_op_min_asize
= vdev_default_min_asize
,
1090 .vdev_op_min_alloc
= NULL
,
1091 .vdev_op_io_start
= vdev_disk_io_start
,
1092 .vdev_op_io_done
= vdev_disk_io_done
,
1093 .vdev_op_state_change
= NULL
,
1094 .vdev_op_need_resilver
= NULL
,
1095 .vdev_op_hold
= vdev_disk_hold
,
1096 .vdev_op_rele
= vdev_disk_rele
,
1097 .vdev_op_remap
= NULL
,
1098 .vdev_op_xlate
= vdev_default_xlate
,
1099 .vdev_op_rebuild_asize
= NULL
,
1100 .vdev_op_metaslab_init
= NULL
,
1101 .vdev_op_config_generate
= NULL
,
1102 .vdev_op_nparity
= NULL
,
1103 .vdev_op_ndisks
= NULL
,
1104 .vdev_op_type
= VDEV_TYPE_DISK
, /* name of this vdev type */
1105 .vdev_op_leaf
= B_TRUE
, /* leaf vdev */
1106 .vdev_op_kobj_evt_post
= vdev_disk_kobj_evt_post
1110 * The zfs_vdev_scheduler module option has been deprecated. Setting this
1111 * value no longer has any effect. It has not yet been entirely removed
1112 * to allow the module to be loaded if this option is specified in the
1113 * /etc/modprobe.d/zfs.conf file. The following warning will be logged.
1116 param_set_vdev_scheduler(const char *val
, zfs_kernel_param_t
*kp
)
1118 int error
= param_set_charp(val
, kp
);
1120 printk(KERN_INFO
"The 'zfs_vdev_scheduler' module option "
1121 "is not supported.\n");
1127 static const char *zfs_vdev_scheduler
= "unused";
1128 module_param_call(zfs_vdev_scheduler
, param_set_vdev_scheduler
,
1129 param_get_charp
, &zfs_vdev_scheduler
, 0644);
1130 MODULE_PARM_DESC(zfs_vdev_scheduler
, "I/O scheduler");
1133 param_set_min_auto_ashift(const char *buf
, zfs_kernel_param_t
*kp
)
1138 error
= kstrtouint(buf
, 0, &val
);
1140 return (SET_ERROR(error
));
1142 if (val
< ASHIFT_MIN
|| val
> zfs_vdev_max_auto_ashift
)
1143 return (SET_ERROR(-EINVAL
));
1145 error
= param_set_uint(buf
, kp
);
1147 return (SET_ERROR(error
));
1153 param_set_max_auto_ashift(const char *buf
, zfs_kernel_param_t
*kp
)
1158 error
= kstrtouint(buf
, 0, &val
);
1160 return (SET_ERROR(error
));
1162 if (val
> ASHIFT_MAX
|| val
< zfs_vdev_min_auto_ashift
)
1163 return (SET_ERROR(-EINVAL
));
1165 error
= param_set_uint(buf
, kp
);
1167 return (SET_ERROR(error
));
1172 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, open_timeout_ms
, UINT
, ZMOD_RW
,
1173 "Timeout before determining that a device is missing");
1175 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, failfast_mask
, UINT
, ZMOD_RW
,
1176 "Defines failfast mask: 1 - device, 2 - transport, 4 - driver");