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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Joyent, Inc. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.h>
30 #include <sys/refcount.h>
31 #include <sys/vdev_disk.h>
32 #include <sys/vdev_impl.h>
34 #include <sys/fs/zfs.h>
36 #include <sys/sunldi.h>
37 #include <sys/efi_partition.h>
38 #include <sys/fm/fs/zfs.h>
41 * Tunable parameter for debugging or performance analysis. Setting this
42 * will cause pool corruption on power loss if a volatile out-of-order
43 * write cache is enabled.
45 boolean_t zfs_nocacheflush
= B_FALSE
;
48 * Virtual device vector for disks.
51 extern ldi_ident_t zfs_li
;
53 static void vdev_disk_close(vdev_t
*);
55 typedef struct vdev_disk_ldi_cb
{
57 ldi_callback_id_t lcb_id
;
61 vdev_disk_alloc(vdev_t
*vd
)
65 dvd
= vd
->vdev_tsd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
67 * Create the LDI event callback list.
69 list_create(&dvd
->vd_ldi_cbs
, sizeof (vdev_disk_ldi_cb_t
),
70 offsetof(vdev_disk_ldi_cb_t
, lcb_next
));
74 vdev_disk_free(vdev_t
*vd
)
76 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
77 vdev_disk_ldi_cb_t
*lcb
;
83 * We have already closed the LDI handle. Clean up the LDI event
84 * callbacks and free vd->vdev_tsd.
86 while ((lcb
= list_head(&dvd
->vd_ldi_cbs
)) != NULL
) {
87 list_remove(&dvd
->vd_ldi_cbs
, lcb
);
88 (void) ldi_ev_remove_callbacks(lcb
->lcb_id
);
89 kmem_free(lcb
, sizeof (vdev_disk_ldi_cb_t
));
91 list_destroy(&dvd
->vd_ldi_cbs
);
92 kmem_free(dvd
, sizeof (vdev_disk_t
));
98 vdev_disk_off_notify(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
, void *arg
,
101 vdev_t
*vd
= (vdev_t
*)arg
;
102 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
105 * Ignore events other than offline.
107 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
108 return (LDI_EV_SUCCESS
);
111 * All LDI handles must be closed for the state change to succeed, so
112 * call on vdev_disk_close() to do this.
114 * We inform vdev_disk_close that it is being called from offline
115 * notify context so it will defer cleanup of LDI event callbacks and
116 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
118 dvd
->vd_ldi_offline
= B_TRUE
;
122 * Now that the device is closed, request that the spa_async_thread
123 * mark the device as REMOVED and notify FMA of the removal.
125 zfs_post_remove(vd
->vdev_spa
, vd
);
126 vd
->vdev_remove_wanted
= B_TRUE
;
127 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_REMOVE
);
129 return (LDI_EV_SUCCESS
);
134 vdev_disk_off_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
135 int ldi_result
, void *arg
, void *ev_data
)
137 vdev_t
*vd
= (vdev_t
*)arg
;
140 * Ignore events other than offline.
142 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
146 * We have already closed the LDI handle in notify.
147 * Clean up the LDI event callbacks and free vd->vdev_tsd.
152 * Request that the vdev be reopened if the offline state change was
155 if (ldi_result
!= LDI_EV_SUCCESS
) {
156 vd
->vdev_probe_wanted
= B_TRUE
;
157 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_PROBE
);
161 static ldi_ev_callback_t vdev_disk_off_callb
= {
162 .cb_vers
= LDI_EV_CB_VERS
,
163 .cb_notify
= vdev_disk_off_notify
,
164 .cb_finalize
= vdev_disk_off_finalize
169 vdev_disk_dgrd_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
170 int ldi_result
, void *arg
, void *ev_data
)
172 vdev_t
*vd
= (vdev_t
*)arg
;
175 * Ignore events other than degrade.
177 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_DEGRADE
) != 0)
181 * Degrade events always succeed. Mark the vdev as degraded.
182 * This status is purely informative for the user.
184 (void) vdev_degrade(vd
->vdev_spa
, vd
->vdev_guid
, 0);
187 static ldi_ev_callback_t vdev_disk_dgrd_callb
= {
188 .cb_vers
= LDI_EV_CB_VERS
,
190 .cb_finalize
= vdev_disk_dgrd_finalize
194 vdev_disk_hold(vdev_t
*vd
)
199 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
202 * We must have a pathname, and it must be absolute.
204 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
208 * Only prefetch path and devid info if the device has
211 if (vd
->vdev_tsd
!= NULL
)
214 if (vd
->vdev_wholedisk
== -1ULL) {
215 size_t len
= strlen(vd
->vdev_path
) + 3;
216 char *buf
= kmem_alloc(len
, KM_SLEEP
);
218 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
220 (void) ldi_vp_from_name(buf
, &vd
->vdev_name_vp
);
224 if (vd
->vdev_name_vp
== NULL
)
225 (void) ldi_vp_from_name(vd
->vdev_path
, &vd
->vdev_name_vp
);
227 if (vd
->vdev_devid
!= NULL
&&
228 ddi_devid_str_decode(vd
->vdev_devid
, &devid
, &minor
) == 0) {
229 (void) ldi_vp_from_devid(devid
, minor
, &vd
->vdev_devid_vp
);
230 ddi_devid_str_free(minor
);
231 ddi_devid_free(devid
);
236 vdev_disk_rele(vdev_t
*vd
)
238 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
240 if (vd
->vdev_name_vp
) {
241 VN_RELE_ASYNC(vd
->vdev_name_vp
,
242 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
243 vd
->vdev_name_vp
= NULL
;
245 if (vd
->vdev_devid_vp
) {
246 VN_RELE_ASYNC(vd
->vdev_devid_vp
,
247 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
248 vd
->vdev_devid_vp
= NULL
;
253 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
254 * even a fallback to DKIOCGMEDIAINFO fails.
257 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
259 #define VDEV_DEBUG(...) /* Nothing... */
263 vdev_disk_open(vdev_t
*vd
, uint64_t *psize
, uint64_t *max_psize
,
266 spa_t
*spa
= vd
->vdev_spa
;
267 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
268 ldi_ev_cookie_t ecookie
;
269 vdev_disk_ldi_cb_t
*lcb
;
271 struct dk_minfo_ext ude
;
274 struct dk_minfo_ext
*dkmext
= &dks
.ude
;
275 struct dk_minfo
*dkm
= &dks
.ud
;
279 boolean_t validate_devid
= B_FALSE
;
281 uint64_t capacity
= 0, blksz
= 0, pbsize
;
284 * We must have a pathname, and it must be absolute.
286 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/') {
287 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
288 return (SET_ERROR(EINVAL
));
292 * Reopen the device if it's not currently open. Otherwise,
293 * just update the physical size of the device.
296 if (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
) {
298 * If we are opening a device in its offline notify
299 * context, the LDI handle was just closed. Clean
300 * up the LDI event callbacks and free vd->vdev_tsd.
304 ASSERT(vd
->vdev_reopening
);
310 * Create vd->vdev_tsd.
316 * When opening a disk device, we want to preserve the user's original
317 * intent. We always want to open the device by the path the user gave
318 * us, even if it is one of multiple paths to the same device. But we
319 * also want to be able to survive disks being removed/recabled.
320 * Therefore the sequence of opening devices is:
322 * 1. Try opening the device by path. For legacy pools without the
323 * 'whole_disk' property, attempt to fix the path by appending 's0'.
325 * 2. If the devid of the device matches the stored value, return
328 * 3. Otherwise, the device may have moved. Try opening the device
329 * by the devid instead.
331 if (vd
->vdev_devid
!= NULL
) {
332 if (ddi_devid_str_decode(vd
->vdev_devid
, &dvd
->vd_devid
,
333 &dvd
->vd_minor
) != 0) {
334 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
335 vdev_dbgmsg(vd
, "vdev_disk_open: invalid "
336 "vdev_devid '%s'", vd
->vdev_devid
);
337 return (SET_ERROR(EINVAL
));
341 error
= EINVAL
; /* presume failure */
343 if (vd
->vdev_path
!= NULL
) {
345 if (vd
->vdev_wholedisk
== -1ULL) {
346 size_t len
= strlen(vd
->vdev_path
) + 3;
347 char *buf
= kmem_alloc(len
, KM_SLEEP
);
349 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
351 error
= ldi_open_by_name(buf
, spa_mode(spa
), kcred
,
352 &dvd
->vd_lh
, zfs_li
);
354 spa_strfree(vd
->vdev_path
);
356 vd
->vdev_wholedisk
= 1ULL;
363 * If we have not yet opened the device, try to open it by the
367 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
368 kcred
, &dvd
->vd_lh
, zfs_li
);
372 * Compare the devid to the stored value.
374 if (error
== 0 && vd
->vdev_devid
!= NULL
&&
375 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
376 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
377 error
= SET_ERROR(EINVAL
);
378 (void) ldi_close(dvd
->vd_lh
, spa_mode(spa
),
382 ddi_devid_free(devid
);
386 * If we succeeded in opening the device, but 'vdev_wholedisk'
387 * is not yet set, then this must be a slice.
389 if (error
== 0 && vd
->vdev_wholedisk
== -1ULL)
390 vd
->vdev_wholedisk
= 0;
394 * If we were unable to open by path, or the devid check fails, open by
397 if (error
!= 0 && vd
->vdev_devid
!= NULL
) {
398 error
= ldi_open_by_devid(dvd
->vd_devid
, dvd
->vd_minor
,
399 spa_mode(spa
), kcred
, &dvd
->vd_lh
, zfs_li
);
403 * If all else fails, then try opening by physical path (if available)
404 * or the logical path (if we failed due to the devid check). While not
405 * as reliable as the devid, this will give us something, and the higher
406 * level vdev validation will prevent us from opening the wrong device.
409 if (vd
->vdev_devid
!= NULL
)
410 validate_devid
= B_TRUE
;
412 if (vd
->vdev_physpath
!= NULL
&&
413 (dev
= ddi_pathname_to_dev_t(vd
->vdev_physpath
)) != NODEV
)
414 error
= ldi_open_by_dev(&dev
, OTYP_BLK
, spa_mode(spa
),
415 kcred
, &dvd
->vd_lh
, zfs_li
);
418 * Note that we don't support the legacy auto-wholedisk support
419 * as above. This hasn't been used in a very long time and we
420 * don't need to propagate its oddities to this edge condition.
422 if (error
&& vd
->vdev_path
!= NULL
)
423 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
424 kcred
, &dvd
->vd_lh
, zfs_li
);
428 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
429 vdev_dbgmsg(vd
, "vdev_disk_open: failed to open [error=%d]",
435 * Now that the device has been successfully opened, update the devid
438 if (validate_devid
&& spa_writeable(spa
) &&
439 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
440 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
443 vd_devid
= ddi_devid_str_encode(devid
, dvd
->vd_minor
);
444 vdev_dbgmsg(vd
, "vdev_disk_open: update devid from "
445 "'%s' to '%s'", vd
->vdev_devid
, vd_devid
);
446 spa_strfree(vd
->vdev_devid
);
447 vd
->vdev_devid
= spa_strdup(vd_devid
);
448 ddi_devid_str_free(vd_devid
);
450 ddi_devid_free(devid
);
454 * Once a device is opened, verify that the physical device path (if
455 * available) is up to date.
457 if (ldi_get_dev(dvd
->vd_lh
, &dev
) == 0 &&
458 ldi_get_otyp(dvd
->vd_lh
, &otyp
) == 0) {
459 char *physpath
, *minorname
;
461 physpath
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
463 if (ddi_dev_pathname(dev
, otyp
, physpath
) == 0 &&
464 ldi_get_minor_name(dvd
->vd_lh
, &minorname
) == 0 &&
465 (vd
->vdev_physpath
== NULL
||
466 strcmp(vd
->vdev_physpath
, physpath
) != 0)) {
467 if (vd
->vdev_physpath
)
468 spa_strfree(vd
->vdev_physpath
);
469 (void) strlcat(physpath
, ":", MAXPATHLEN
);
470 (void) strlcat(physpath
, minorname
, MAXPATHLEN
);
471 vd
->vdev_physpath
= spa_strdup(physpath
);
474 kmem_free(minorname
, strlen(minorname
) + 1);
475 kmem_free(physpath
, MAXPATHLEN
);
479 * Register callbacks for the LDI offline event.
481 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_OFFLINE
, &ecookie
) ==
483 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
484 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
485 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
486 &vdev_disk_off_callb
, (void *) vd
, &lcb
->lcb_id
);
490 * Register callbacks for the LDI degrade event.
492 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_DEGRADE
, &ecookie
) ==
494 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
495 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
496 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
497 &vdev_disk_dgrd_callb
, (void *) vd
, &lcb
->lcb_id
);
501 * Determine the actual size of the device.
503 if (ldi_get_size(dvd
->vd_lh
, psize
) != 0) {
504 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
505 vdev_dbgmsg(vd
, "vdev_disk_open: failed to get size");
506 return (SET_ERROR(EINVAL
));
512 * Determine the device's minimum transfer size.
513 * If the ioctl isn't supported, assume DEV_BSIZE.
515 if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFOEXT
,
516 (intptr_t)dkmext
, FKIOCTL
, kcred
, NULL
)) == 0) {
517 capacity
= dkmext
->dki_capacity
- 1;
518 blksz
= dkmext
->dki_lbsize
;
519 pbsize
= dkmext
->dki_pbsize
;
520 } else if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFO
,
521 (intptr_t)dkm
, FKIOCTL
, kcred
, NULL
)) == 0) {
523 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
525 capacity
= dkm
->dki_capacity
- 1;
526 blksz
= dkm
->dki_lbsize
;
529 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
530 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
531 vd
->vdev_path
, error
);
535 *ashift
= highbit64(MAX(pbsize
, SPA_MINBLOCKSIZE
)) - 1;
537 if (vd
->vdev_wholedisk
== 1) {
542 * If we have the capability to expand, we'd have
543 * found out via success from DKIOCGMEDIAINFO{,EXT}.
544 * Adjust max_psize upward accordingly since we know
545 * we own the whole disk now.
547 *max_psize
= capacity
* blksz
;
551 * Since we own the whole disk, try to enable disk write
552 * caching. We ignore errors because it's OK if we can't do it.
554 (void) ldi_ioctl(dvd
->vd_lh
, DKIOCSETWCE
, (intptr_t)&wce
,
555 FKIOCTL
, kcred
, NULL
);
559 * Clear the nowritecache bit, so that on a vdev_reopen() we will
562 vd
->vdev_nowritecache
= B_FALSE
;
568 vdev_disk_close(vdev_t
*vd
)
570 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
572 if (vd
->vdev_reopening
|| dvd
== NULL
)
575 if (dvd
->vd_minor
!= NULL
) {
576 ddi_devid_str_free(dvd
->vd_minor
);
577 dvd
->vd_minor
= NULL
;
580 if (dvd
->vd_devid
!= NULL
) {
581 ddi_devid_free(dvd
->vd_devid
);
582 dvd
->vd_devid
= NULL
;
585 if (dvd
->vd_lh
!= NULL
) {
586 (void) ldi_close(dvd
->vd_lh
, spa_mode(vd
->vdev_spa
), kcred
);
590 vd
->vdev_delayed_close
= B_FALSE
;
592 * If we closed the LDI handle due to an offline notify from LDI,
593 * don't free vd->vdev_tsd or unregister the callbacks here;
594 * the offline finalize callback or a reopen will take care of it.
596 if (dvd
->vd_ldi_offline
)
603 vdev_disk_physio(vdev_t
*vd
, caddr_t data
,
604 size_t size
, uint64_t offset
, int flags
, boolean_t isdump
)
606 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
609 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
610 * Nothing to be done here but return failure.
612 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
))
615 ASSERT(vd
->vdev_ops
== &vdev_disk_ops
);
618 * If in the context of an active crash dump, use the ldi_dump(9F)
619 * call instead of ldi_strategy(9F) as usual.
622 ASSERT3P(dvd
, !=, NULL
);
623 return (ldi_dump(dvd
->vd_lh
, data
, lbtodb(offset
),
627 return (vdev_disk_ldi_physio(dvd
->vd_lh
, data
, size
, offset
, flags
));
631 vdev_disk_ldi_physio(ldi_handle_t vd_lh
, caddr_t data
,
632 size_t size
, uint64_t offset
, int flags
)
638 return (SET_ERROR(EINVAL
));
640 ASSERT(flags
& B_READ
|| flags
& B_WRITE
);
642 bp
= getrbuf(KM_SLEEP
);
643 bp
->b_flags
= flags
| B_BUSY
| B_NOCACHE
| B_FAILFAST
;
645 bp
->b_un
.b_addr
= (void *)data
;
646 bp
->b_lblkno
= lbtodb(offset
);
647 bp
->b_bufsize
= size
;
649 error
= ldi_strategy(vd_lh
, bp
);
651 if ((error
= biowait(bp
)) == 0 && bp
->b_resid
!= 0)
652 error
= SET_ERROR(EIO
);
659 vdev_disk_io_intr(buf_t
*bp
)
661 vdev_buf_t
*vb
= (vdev_buf_t
*)bp
;
662 zio_t
*zio
= vb
->vb_io
;
665 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
666 * Rather than teach the rest of the stack about other error
667 * possibilities (EFAULT, etc), we normalize the error value here.
669 zio
->io_error
= (geterror(bp
) != 0 ? EIO
: 0);
671 if (zio
->io_error
== 0 && bp
->b_resid
!= 0)
672 zio
->io_error
= SET_ERROR(EIO
);
674 if (zio
->io_type
== ZIO_TYPE_READ
) {
675 abd_return_buf_copy(zio
->io_abd
, bp
->b_un
.b_addr
, zio
->io_size
);
677 abd_return_buf(zio
->io_abd
, bp
->b_un
.b_addr
, zio
->io_size
);
680 kmem_free(vb
, sizeof (vdev_buf_t
));
682 zio_delay_interrupt(zio
);
686 vdev_disk_ioctl_free(zio_t
*zio
)
688 kmem_free(zio
->io_vsd
, sizeof (struct dk_callback
));
691 static const zio_vsd_ops_t vdev_disk_vsd_ops
= {
692 vdev_disk_ioctl_free
,
693 zio_vsd_default_cksum_report
697 vdev_disk_ioctl_done(void *zio_arg
, int error
)
699 zio_t
*zio
= zio_arg
;
701 zio
->io_error
= error
;
707 vdev_disk_io_start(zio_t
*zio
)
709 vdev_t
*vd
= zio
->io_vd
;
710 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
712 struct dk_callback
*dkc
;
717 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
718 * Nothing to be done here but return failure.
720 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
)) {
721 zio
->io_error
= ENXIO
;
726 if (zio
->io_type
== ZIO_TYPE_IOCTL
) {
728 if (!vdev_readable(vd
)) {
729 zio
->io_error
= SET_ERROR(ENXIO
);
734 switch (zio
->io_cmd
) {
736 case DKIOCFLUSHWRITECACHE
:
738 if (zfs_nocacheflush
)
741 if (vd
->vdev_nowritecache
) {
742 zio
->io_error
= SET_ERROR(ENOTSUP
);
746 zio
->io_vsd
= dkc
= kmem_alloc(sizeof (*dkc
), KM_SLEEP
);
747 zio
->io_vsd_ops
= &vdev_disk_vsd_ops
;
749 dkc
->dkc_callback
= vdev_disk_ioctl_done
;
750 dkc
->dkc_flag
= FLUSH_VOLATILE
;
751 dkc
->dkc_cookie
= zio
;
753 error
= ldi_ioctl(dvd
->vd_lh
, zio
->io_cmd
,
754 (uintptr_t)dkc
, FKIOCTL
, kcred
, NULL
);
758 * The ioctl will be done asychronously,
759 * and will call vdev_disk_ioctl_done()
765 zio
->io_error
= error
;
770 zio
->io_error
= SET_ERROR(ENOTSUP
);
777 ASSERT(zio
->io_type
== ZIO_TYPE_READ
|| zio
->io_type
== ZIO_TYPE_WRITE
);
778 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
780 vb
= kmem_alloc(sizeof (vdev_buf_t
), KM_SLEEP
);
786 bp
->b_flags
= B_BUSY
| B_NOCACHE
|
787 (zio
->io_type
== ZIO_TYPE_READ
? B_READ
: B_WRITE
);
788 if (!(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)))
789 bp
->b_flags
|= B_FAILFAST
;
790 bp
->b_bcount
= zio
->io_size
;
792 if (zio
->io_type
== ZIO_TYPE_READ
) {
794 abd_borrow_buf(zio
->io_abd
, zio
->io_size
);
797 abd_borrow_buf_copy(zio
->io_abd
, zio
->io_size
);
800 bp
->b_lblkno
= lbtodb(zio
->io_offset
);
801 bp
->b_bufsize
= zio
->io_size
;
802 bp
->b_iodone
= (int (*)())vdev_disk_io_intr
;
804 /* ldi_strategy() will return non-zero only on programming errors */
805 VERIFY(ldi_strategy(dvd
->vd_lh
, bp
) == 0);
809 vdev_disk_io_done(zio_t
*zio
)
811 vdev_t
*vd
= zio
->io_vd
;
814 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
815 * the device has been removed. If this is the case, then we trigger an
816 * asynchronous removal of the device. Otherwise, probe the device and
817 * make sure it's still accessible.
819 if (zio
->io_error
== EIO
&& !vd
->vdev_remove_wanted
) {
820 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
821 int state
= DKIO_NONE
;
823 if (ldi_ioctl(dvd
->vd_lh
, DKIOCSTATE
, (intptr_t)&state
,
824 FKIOCTL
, kcred
, NULL
) == 0 && state
!= DKIO_INSERTED
) {
826 * We post the resource as soon as possible, instead of
827 * when the async removal actually happens, because the
828 * DE is using this information to discard previous I/O
831 zfs_post_remove(zio
->io_spa
, vd
);
832 vd
->vdev_remove_wanted
= B_TRUE
;
833 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
834 } else if (!vd
->vdev_delayed_close
) {
835 vd
->vdev_delayed_close
= B_TRUE
;
840 vdev_ops_t vdev_disk_ops
= {
851 VDEV_TYPE_DISK
, /* name of this vdev type */
852 B_TRUE
/* leaf vdev */
856 * Given the root disk device devid or pathname, read the label from
857 * the device, and construct a configuration nvlist.
860 vdev_disk_read_rootlabel(char *devpath
, char *devid
, nvlist_t
**config
)
866 ddi_devid_t tmpdevid
;
871 * Read the device label and build the nvlist.
873 if (devid
!= NULL
&& ddi_devid_str_decode(devid
, &tmpdevid
,
875 error
= ldi_open_by_devid(tmpdevid
, minor_name
,
876 FREAD
, kcred
, &vd_lh
, zfs_li
);
877 ddi_devid_free(tmpdevid
);
878 ddi_devid_str_free(minor_name
);
881 if (error
&& (error
= ldi_open_by_name(devpath
, FREAD
, kcred
, &vd_lh
,
885 if (ldi_get_size(vd_lh
, &s
)) {
886 (void) ldi_close(vd_lh
, FREAD
, kcred
);
887 return (SET_ERROR(EIO
));
890 size
= P2ALIGN_TYPED(s
, sizeof (vdev_label_t
), uint64_t);
891 label
= kmem_alloc(sizeof (vdev_label_t
), KM_SLEEP
);
894 for (l
= 0; l
< VDEV_LABELS
; l
++) {
895 uint64_t offset
, state
, txg
= 0;
897 /* read vdev label */
898 offset
= vdev_label_offset(size
, l
, 0);
899 if (vdev_disk_ldi_physio(vd_lh
, (caddr_t
)label
,
900 VDEV_SKIP_SIZE
+ VDEV_PHYS_SIZE
, offset
, B_READ
) != 0)
903 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
904 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0) {
909 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
910 &state
) != 0 || state
>= POOL_STATE_DESTROYED
) {
911 nvlist_free(*config
);
916 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
917 &txg
) != 0 || txg
== 0) {
918 nvlist_free(*config
);
926 kmem_free(label
, sizeof (vdev_label_t
));
927 (void) ldi_close(vd_lh
, FREAD
, kcred
);
929 error
= SET_ERROR(EIDRM
);