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, 2015 by Delphix. All rights reserved.
24 * Copyright 2013 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>
33 #include <sys/fs/zfs.h>
35 #include <sys/sunldi.h>
36 #include <sys/efi_partition.h>
37 #include <sys/fm/fs/zfs.h>
40 * Virtual device vector for disks.
43 extern ldi_ident_t zfs_li
;
45 static void vdev_disk_close(vdev_t
*);
47 typedef struct vdev_disk_ldi_cb
{
49 ldi_callback_id_t lcb_id
;
53 vdev_disk_alloc(vdev_t
*vd
)
57 dvd
= vd
->vdev_tsd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
59 * Create the LDI event callback list.
61 list_create(&dvd
->vd_ldi_cbs
, sizeof (vdev_disk_ldi_cb_t
),
62 offsetof(vdev_disk_ldi_cb_t
, lcb_next
));
66 vdev_disk_free(vdev_t
*vd
)
68 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
69 vdev_disk_ldi_cb_t
*lcb
;
75 * We have already closed the LDI handle. Clean up the LDI event
76 * callbacks and free vd->vdev_tsd.
78 while ((lcb
= list_head(&dvd
->vd_ldi_cbs
)) != NULL
) {
79 list_remove(&dvd
->vd_ldi_cbs
, lcb
);
80 (void) ldi_ev_remove_callbacks(lcb
->lcb_id
);
81 kmem_free(lcb
, sizeof (vdev_disk_ldi_cb_t
));
83 list_destroy(&dvd
->vd_ldi_cbs
);
84 kmem_free(dvd
, sizeof (vdev_disk_t
));
90 vdev_disk_off_notify(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
, void *arg
,
93 vdev_t
*vd
= (vdev_t
*)arg
;
94 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
97 * Ignore events other than offline.
99 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
100 return (LDI_EV_SUCCESS
);
103 * All LDI handles must be closed for the state change to succeed, so
104 * call on vdev_disk_close() to do this.
106 * We inform vdev_disk_close that it is being called from offline
107 * notify context so it will defer cleanup of LDI event callbacks and
108 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
110 dvd
->vd_ldi_offline
= B_TRUE
;
114 * Now that the device is closed, request that the spa_async_thread
115 * mark the device as REMOVED and notify FMA of the removal.
117 zfs_post_remove(vd
->vdev_spa
, vd
);
118 vd
->vdev_remove_wanted
= B_TRUE
;
119 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_REMOVE
);
121 return (LDI_EV_SUCCESS
);
126 vdev_disk_off_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
127 int ldi_result
, void *arg
, void *ev_data
)
129 vdev_t
*vd
= (vdev_t
*)arg
;
132 * Ignore events other than offline.
134 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
138 * We have already closed the LDI handle in notify.
139 * Clean up the LDI event callbacks and free vd->vdev_tsd.
144 * Request that the vdev be reopened if the offline state change was
147 if (ldi_result
!= LDI_EV_SUCCESS
) {
148 vd
->vdev_probe_wanted
= B_TRUE
;
149 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_PROBE
);
153 static ldi_ev_callback_t vdev_disk_off_callb
= {
154 .cb_vers
= LDI_EV_CB_VERS
,
155 .cb_notify
= vdev_disk_off_notify
,
156 .cb_finalize
= vdev_disk_off_finalize
161 vdev_disk_dgrd_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
162 int ldi_result
, void *arg
, void *ev_data
)
164 vdev_t
*vd
= (vdev_t
*)arg
;
167 * Ignore events other than degrade.
169 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_DEGRADE
) != 0)
173 * Degrade events always succeed. Mark the vdev as degraded.
174 * This status is purely informative for the user.
176 (void) vdev_degrade(vd
->vdev_spa
, vd
->vdev_guid
, 0);
179 static ldi_ev_callback_t vdev_disk_dgrd_callb
= {
180 .cb_vers
= LDI_EV_CB_VERS
,
182 .cb_finalize
= vdev_disk_dgrd_finalize
186 vdev_disk_hold(vdev_t
*vd
)
191 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
194 * We must have a pathname, and it must be absolute.
196 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
200 * Only prefetch path and devid info if the device has
203 if (vd
->vdev_tsd
!= NULL
)
206 if (vd
->vdev_wholedisk
== -1ULL) {
207 size_t len
= strlen(vd
->vdev_path
) + 3;
208 char *buf
= kmem_alloc(len
, KM_SLEEP
);
210 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
212 (void) ldi_vp_from_name(buf
, &vd
->vdev_name_vp
);
216 if (vd
->vdev_name_vp
== NULL
)
217 (void) ldi_vp_from_name(vd
->vdev_path
, &vd
->vdev_name_vp
);
219 if (vd
->vdev_devid
!= NULL
&&
220 ddi_devid_str_decode(vd
->vdev_devid
, &devid
, &minor
) == 0) {
221 (void) ldi_vp_from_devid(devid
, minor
, &vd
->vdev_devid_vp
);
222 ddi_devid_str_free(minor
);
223 ddi_devid_free(devid
);
228 vdev_disk_rele(vdev_t
*vd
)
230 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
232 if (vd
->vdev_name_vp
) {
233 VN_RELE_ASYNC(vd
->vdev_name_vp
,
234 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
235 vd
->vdev_name_vp
= NULL
;
237 if (vd
->vdev_devid_vp
) {
238 VN_RELE_ASYNC(vd
->vdev_devid_vp
,
239 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
240 vd
->vdev_devid_vp
= NULL
;
245 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
246 * even a fallback to DKIOCGMEDIAINFO fails.
249 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
251 #define VDEV_DEBUG(...) /* Nothing... */
255 vdev_disk_open(vdev_t
*vd
, uint64_t *psize
, uint64_t *max_psize
,
258 spa_t
*spa
= vd
->vdev_spa
;
259 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
260 ldi_ev_cookie_t ecookie
;
261 vdev_disk_ldi_cb_t
*lcb
;
263 struct dk_minfo_ext ude
;
266 struct dk_minfo_ext
*dkmext
= &dks
.ude
;
267 struct dk_minfo
*dkm
= &dks
.ud
;
271 boolean_t validate_devid
= B_FALSE
;
273 uint64_t capacity
= 0, blksz
= 0, pbsize
;
276 * We must have a pathname, and it must be absolute.
278 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/') {
279 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
280 return (SET_ERROR(EINVAL
));
284 * Reopen the device if it's not currently open. Otherwise,
285 * just update the physical size of the device.
288 if (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
) {
290 * If we are opening a device in its offline notify
291 * context, the LDI handle was just closed. Clean
292 * up the LDI event callbacks and free vd->vdev_tsd.
296 ASSERT(vd
->vdev_reopening
);
302 * Create vd->vdev_tsd.
308 * When opening a disk device, we want to preserve the user's original
309 * intent. We always want to open the device by the path the user gave
310 * us, even if it is one of multiple paths to the same device. But we
311 * also want to be able to survive disks being removed/recabled.
312 * Therefore the sequence of opening devices is:
314 * 1. Try opening the device by path. For legacy pools without the
315 * 'whole_disk' property, attempt to fix the path by appending 's0'.
317 * 2. If the devid of the device matches the stored value, return
320 * 3. Otherwise, the device may have moved. Try opening the device
321 * by the devid instead.
323 if (vd
->vdev_devid
!= NULL
) {
324 if (ddi_devid_str_decode(vd
->vdev_devid
, &dvd
->vd_devid
,
325 &dvd
->vd_minor
) != 0) {
326 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
327 return (SET_ERROR(EINVAL
));
331 error
= EINVAL
; /* presume failure */
333 if (vd
->vdev_path
!= NULL
) {
335 if (vd
->vdev_wholedisk
== -1ULL) {
336 size_t len
= strlen(vd
->vdev_path
) + 3;
337 char *buf
= kmem_alloc(len
, KM_SLEEP
);
339 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
341 error
= ldi_open_by_name(buf
, spa_mode(spa
), kcred
,
342 &dvd
->vd_lh
, zfs_li
);
344 spa_strfree(vd
->vdev_path
);
346 vd
->vdev_wholedisk
= 1ULL;
353 * If we have not yet opened the device, try to open it by the
357 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
358 kcred
, &dvd
->vd_lh
, zfs_li
);
362 * Compare the devid to the stored value.
364 if (error
== 0 && vd
->vdev_devid
!= NULL
&&
365 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
366 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
367 error
= SET_ERROR(EINVAL
);
368 (void) ldi_close(dvd
->vd_lh
, spa_mode(spa
),
372 ddi_devid_free(devid
);
376 * If we succeeded in opening the device, but 'vdev_wholedisk'
377 * is not yet set, then this must be a slice.
379 if (error
== 0 && vd
->vdev_wholedisk
== -1ULL)
380 vd
->vdev_wholedisk
= 0;
384 * If we were unable to open by path, or the devid check fails, open by
387 if (error
!= 0 && vd
->vdev_devid
!= NULL
) {
388 error
= ldi_open_by_devid(dvd
->vd_devid
, dvd
->vd_minor
,
389 spa_mode(spa
), kcred
, &dvd
->vd_lh
, zfs_li
);
393 * If all else fails, then try opening by physical path (if available)
394 * or the logical path (if we failed due to the devid check). While not
395 * as reliable as the devid, this will give us something, and the higher
396 * level vdev validation will prevent us from opening the wrong device.
399 if (vd
->vdev_devid
!= NULL
)
400 validate_devid
= B_TRUE
;
402 if (vd
->vdev_physpath
!= NULL
&&
403 (dev
= ddi_pathname_to_dev_t(vd
->vdev_physpath
)) != NODEV
)
404 error
= ldi_open_by_dev(&dev
, OTYP_BLK
, spa_mode(spa
),
405 kcred
, &dvd
->vd_lh
, zfs_li
);
408 * Note that we don't support the legacy auto-wholedisk support
409 * as above. This hasn't been used in a very long time and we
410 * don't need to propagate its oddities to this edge condition.
412 if (error
&& vd
->vdev_path
!= NULL
)
413 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
414 kcred
, &dvd
->vd_lh
, zfs_li
);
418 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
423 * Now that the device has been successfully opened, update the devid
426 if (validate_devid
&& spa_writeable(spa
) &&
427 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
428 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
431 vd_devid
= ddi_devid_str_encode(devid
, dvd
->vd_minor
);
432 zfs_dbgmsg("vdev %s: update devid from %s, "
433 "to %s", vd
->vdev_path
, vd
->vdev_devid
, vd_devid
);
434 spa_strfree(vd
->vdev_devid
);
435 vd
->vdev_devid
= spa_strdup(vd_devid
);
436 ddi_devid_str_free(vd_devid
);
438 ddi_devid_free(devid
);
442 * Once a device is opened, verify that the physical device path (if
443 * available) is up to date.
445 if (ldi_get_dev(dvd
->vd_lh
, &dev
) == 0 &&
446 ldi_get_otyp(dvd
->vd_lh
, &otyp
) == 0) {
447 char *physpath
, *minorname
;
449 physpath
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
451 if (ddi_dev_pathname(dev
, otyp
, physpath
) == 0 &&
452 ldi_get_minor_name(dvd
->vd_lh
, &minorname
) == 0 &&
453 (vd
->vdev_physpath
== NULL
||
454 strcmp(vd
->vdev_physpath
, physpath
) != 0)) {
455 if (vd
->vdev_physpath
)
456 spa_strfree(vd
->vdev_physpath
);
457 (void) strlcat(physpath
, ":", MAXPATHLEN
);
458 (void) strlcat(physpath
, minorname
, MAXPATHLEN
);
459 vd
->vdev_physpath
= spa_strdup(physpath
);
462 kmem_free(minorname
, strlen(minorname
) + 1);
463 kmem_free(physpath
, MAXPATHLEN
);
467 * Register callbacks for the LDI offline event.
469 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_OFFLINE
, &ecookie
) ==
471 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
472 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
473 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
474 &vdev_disk_off_callb
, (void *) vd
, &lcb
->lcb_id
);
478 * Register callbacks for the LDI degrade event.
480 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_DEGRADE
, &ecookie
) ==
482 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
483 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
484 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
485 &vdev_disk_dgrd_callb
, (void *) vd
, &lcb
->lcb_id
);
489 * Determine the actual size of the device.
491 if (ldi_get_size(dvd
->vd_lh
, psize
) != 0) {
492 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
493 return (SET_ERROR(EINVAL
));
499 * Determine the device's minimum transfer size.
500 * If the ioctl isn't supported, assume DEV_BSIZE.
502 if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFOEXT
,
503 (intptr_t)dkmext
, FKIOCTL
, kcred
, NULL
)) == 0) {
504 capacity
= dkmext
->dki_capacity
- 1;
505 blksz
= dkmext
->dki_lbsize
;
506 pbsize
= dkmext
->dki_pbsize
;
507 } else if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFO
,
508 (intptr_t)dkm
, FKIOCTL
, kcred
, NULL
)) == 0) {
510 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
512 capacity
= dkm
->dki_capacity
- 1;
513 blksz
= dkm
->dki_lbsize
;
516 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
517 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
518 vd
->vdev_path
, error
);
522 *ashift
= highbit64(MAX(pbsize
, SPA_MINBLOCKSIZE
)) - 1;
524 if (vd
->vdev_wholedisk
== 1) {
529 * If we have the capability to expand, we'd have
530 * found out via success from DKIOCGMEDIAINFO{,EXT}.
531 * Adjust max_psize upward accordingly since we know
532 * we own the whole disk now.
534 *max_psize
= capacity
* blksz
;
538 * Since we own the whole disk, try to enable disk write
539 * caching. We ignore errors because it's OK if we can't do it.
541 (void) ldi_ioctl(dvd
->vd_lh
, DKIOCSETWCE
, (intptr_t)&wce
,
542 FKIOCTL
, kcred
, NULL
);
546 * Clear the nowritecache bit, so that on a vdev_reopen() we will
549 vd
->vdev_nowritecache
= B_FALSE
;
555 vdev_disk_close(vdev_t
*vd
)
557 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
559 if (vd
->vdev_reopening
|| dvd
== NULL
)
562 if (dvd
->vd_minor
!= NULL
) {
563 ddi_devid_str_free(dvd
->vd_minor
);
564 dvd
->vd_minor
= NULL
;
567 if (dvd
->vd_devid
!= NULL
) {
568 ddi_devid_free(dvd
->vd_devid
);
569 dvd
->vd_devid
= NULL
;
572 if (dvd
->vd_lh
!= NULL
) {
573 (void) ldi_close(dvd
->vd_lh
, spa_mode(vd
->vdev_spa
), kcred
);
577 vd
->vdev_delayed_close
= B_FALSE
;
579 * If we closed the LDI handle due to an offline notify from LDI,
580 * don't free vd->vdev_tsd or unregister the callbacks here;
581 * the offline finalize callback or a reopen will take care of it.
583 if (dvd
->vd_ldi_offline
)
590 vdev_disk_physio(vdev_t
*vd
, caddr_t data
,
591 size_t size
, uint64_t offset
, int flags
, boolean_t isdump
)
593 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
596 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
597 * Nothing to be done here but return failure.
599 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
))
602 ASSERT(vd
->vdev_ops
== &vdev_disk_ops
);
605 * If in the context of an active crash dump, use the ldi_dump(9F)
606 * call instead of ldi_strategy(9F) as usual.
609 ASSERT3P(dvd
, !=, NULL
);
610 return (ldi_dump(dvd
->vd_lh
, data
, lbtodb(offset
),
614 return (vdev_disk_ldi_physio(dvd
->vd_lh
, data
, size
, offset
, flags
));
618 vdev_disk_ldi_physio(ldi_handle_t vd_lh
, caddr_t data
,
619 size_t size
, uint64_t offset
, int flags
)
625 return (SET_ERROR(EINVAL
));
627 ASSERT(flags
& B_READ
|| flags
& B_WRITE
);
629 bp
= getrbuf(KM_SLEEP
);
630 bp
->b_flags
= flags
| B_BUSY
| B_NOCACHE
| B_FAILFAST
;
632 bp
->b_un
.b_addr
= (void *)data
;
633 bp
->b_lblkno
= lbtodb(offset
);
634 bp
->b_bufsize
= size
;
636 error
= ldi_strategy(vd_lh
, bp
);
638 if ((error
= biowait(bp
)) == 0 && bp
->b_resid
!= 0)
639 error
= SET_ERROR(EIO
);
646 vdev_disk_io_intr(buf_t
*bp
)
648 vdev_buf_t
*vb
= (vdev_buf_t
*)bp
;
649 zio_t
*zio
= vb
->vb_io
;
652 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
653 * Rather than teach the rest of the stack about other error
654 * possibilities (EFAULT, etc), we normalize the error value here.
656 zio
->io_error
= (geterror(bp
) != 0 ? EIO
: 0);
658 if (zio
->io_error
== 0 && bp
->b_resid
!= 0)
659 zio
->io_error
= SET_ERROR(EIO
);
661 kmem_free(vb
, sizeof (vdev_buf_t
));
663 zio_delay_interrupt(zio
);
667 vdev_disk_ioctl_free(zio_t
*zio
)
669 kmem_free(zio
->io_vsd
, sizeof (struct dk_callback
));
672 static const zio_vsd_ops_t vdev_disk_vsd_ops
= {
673 vdev_disk_ioctl_free
,
674 zio_vsd_default_cksum_report
678 vdev_disk_ioctl_done(void *zio_arg
, int error
)
680 zio_t
*zio
= zio_arg
;
682 zio
->io_error
= error
;
688 vdev_disk_io_start(zio_t
*zio
)
690 vdev_t
*vd
= zio
->io_vd
;
691 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
693 struct dk_callback
*dkc
;
698 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
699 * Nothing to be done here but return failure.
701 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
)) {
702 zio
->io_error
= ENXIO
;
707 if (zio
->io_type
== ZIO_TYPE_IOCTL
) {
709 if (!vdev_readable(vd
)) {
710 zio
->io_error
= SET_ERROR(ENXIO
);
715 switch (zio
->io_cmd
) {
717 case DKIOCFLUSHWRITECACHE
:
719 if (zfs_nocacheflush
)
722 if (vd
->vdev_nowritecache
) {
723 zio
->io_error
= SET_ERROR(ENOTSUP
);
727 zio
->io_vsd
= dkc
= kmem_alloc(sizeof (*dkc
), KM_SLEEP
);
728 zio
->io_vsd_ops
= &vdev_disk_vsd_ops
;
730 dkc
->dkc_callback
= vdev_disk_ioctl_done
;
731 dkc
->dkc_flag
= FLUSH_VOLATILE
;
732 dkc
->dkc_cookie
= zio
;
734 error
= ldi_ioctl(dvd
->vd_lh
, zio
->io_cmd
,
735 (uintptr_t)dkc
, FKIOCTL
, kcred
, NULL
);
739 * The ioctl will be done asychronously,
740 * and will call vdev_disk_ioctl_done()
746 if (error
== ENOTSUP
|| error
== ENOTTY
) {
748 * If we get ENOTSUP or ENOTTY, we know that
749 * no future attempts will ever succeed.
750 * In this case we set a persistent bit so
751 * that we don't bother with the ioctl in the
754 vd
->vdev_nowritecache
= B_TRUE
;
756 zio
->io_error
= error
;
761 zio
->io_error
= SET_ERROR(ENOTSUP
);
768 ASSERT(zio
->io_type
== ZIO_TYPE_READ
|| zio
->io_type
== ZIO_TYPE_WRITE
);
769 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
771 vb
= kmem_alloc(sizeof (vdev_buf_t
), KM_SLEEP
);
777 bp
->b_flags
= B_BUSY
| B_NOCACHE
|
778 (zio
->io_type
== ZIO_TYPE_READ
? B_READ
: B_WRITE
);
779 if (!(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)))
780 bp
->b_flags
|= B_FAILFAST
;
781 bp
->b_bcount
= zio
->io_size
;
782 bp
->b_un
.b_addr
= zio
->io_data
;
783 bp
->b_lblkno
= lbtodb(zio
->io_offset
);
784 bp
->b_bufsize
= zio
->io_size
;
785 bp
->b_iodone
= (int (*)())vdev_disk_io_intr
;
787 /* ldi_strategy() will return non-zero only on programming errors */
788 VERIFY(ldi_strategy(dvd
->vd_lh
, bp
) == 0);
792 vdev_disk_io_done(zio_t
*zio
)
794 vdev_t
*vd
= zio
->io_vd
;
797 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
798 * the device has been removed. If this is the case, then we trigger an
799 * asynchronous removal of the device. Otherwise, probe the device and
800 * make sure it's still accessible.
802 if (zio
->io_error
== EIO
&& !vd
->vdev_remove_wanted
) {
803 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
804 int state
= DKIO_NONE
;
806 if (ldi_ioctl(dvd
->vd_lh
, DKIOCSTATE
, (intptr_t)&state
,
807 FKIOCTL
, kcred
, NULL
) == 0 && state
!= DKIO_INSERTED
) {
809 * We post the resource as soon as possible, instead of
810 * when the async removal actually happens, because the
811 * DE is using this information to discard previous I/O
814 zfs_post_remove(zio
->io_spa
, vd
);
815 vd
->vdev_remove_wanted
= B_TRUE
;
816 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
817 } else if (!vd
->vdev_delayed_close
) {
818 vd
->vdev_delayed_close
= B_TRUE
;
823 vdev_ops_t vdev_disk_ops
= {
832 VDEV_TYPE_DISK
, /* name of this vdev type */
833 B_TRUE
/* leaf vdev */
837 * Given the root disk device devid or pathname, read the label from
838 * the device, and construct a configuration nvlist.
841 vdev_disk_read_rootlabel(char *devpath
, char *devid
, nvlist_t
**config
)
847 ddi_devid_t tmpdevid
;
852 * Read the device label and build the nvlist.
854 if (devid
!= NULL
&& ddi_devid_str_decode(devid
, &tmpdevid
,
856 error
= ldi_open_by_devid(tmpdevid
, minor_name
,
857 FREAD
, kcred
, &vd_lh
, zfs_li
);
858 ddi_devid_free(tmpdevid
);
859 ddi_devid_str_free(minor_name
);
862 if (error
&& (error
= ldi_open_by_name(devpath
, FREAD
, kcred
, &vd_lh
,
866 if (ldi_get_size(vd_lh
, &s
)) {
867 (void) ldi_close(vd_lh
, FREAD
, kcred
);
868 return (SET_ERROR(EIO
));
871 size
= P2ALIGN_TYPED(s
, sizeof (vdev_label_t
), uint64_t);
872 label
= kmem_alloc(sizeof (vdev_label_t
), KM_SLEEP
);
875 for (l
= 0; l
< VDEV_LABELS
; l
++) {
876 uint64_t offset
, state
, txg
= 0;
878 /* read vdev label */
879 offset
= vdev_label_offset(size
, l
, 0);
880 if (vdev_disk_ldi_physio(vd_lh
, (caddr_t
)label
,
881 VDEV_SKIP_SIZE
+ VDEV_PHYS_SIZE
, offset
, B_READ
) != 0)
884 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
885 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0) {
890 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
891 &state
) != 0 || state
>= POOL_STATE_DESTROYED
) {
892 nvlist_free(*config
);
897 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
898 &txg
) != 0 || txg
== 0) {
899 nvlist_free(*config
);
907 kmem_free(label
, sizeof (vdev_label_t
));
908 (void) ldi_close(vd_lh
, FREAD
, kcred
);
910 error
= SET_ERROR(EIDRM
);