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, 2018 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 * Bypass the devid when opening a disk vdev.
62 * There have been issues where the devids of several devices were shuffled,
63 * causing pool open failures. Note, that this flag is intended to be used
64 * for pool recovery only.
66 * Note that if a pool is imported with the devids bypassed, all its vdevs will
67 * cease storing devid information permanently. In practice, the devid is rarely
68 * useful as vdev paths do not tend to change unless the hardware is
69 * reconfigured. That said, if the paths do change and a pool fails to open
70 * automatically at boot, a simple zpool import should re-scan the paths and fix
73 boolean_t vdev_disk_bypass_devid
= B_FALSE
;
76 vdev_disk_alloc(vdev_t
*vd
)
80 dvd
= vd
->vdev_tsd
= kmem_zalloc(sizeof (vdev_disk_t
), KM_SLEEP
);
82 * Create the LDI event callback list.
84 list_create(&dvd
->vd_ldi_cbs
, sizeof (vdev_disk_ldi_cb_t
),
85 offsetof(vdev_disk_ldi_cb_t
, lcb_next
));
89 vdev_disk_free(vdev_t
*vd
)
91 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
92 vdev_disk_ldi_cb_t
*lcb
;
98 * We have already closed the LDI handle. Clean up the LDI event
99 * callbacks and free vd->vdev_tsd.
101 while ((lcb
= list_head(&dvd
->vd_ldi_cbs
)) != NULL
) {
102 list_remove(&dvd
->vd_ldi_cbs
, lcb
);
103 (void) ldi_ev_remove_callbacks(lcb
->lcb_id
);
104 kmem_free(lcb
, sizeof (vdev_disk_ldi_cb_t
));
106 list_destroy(&dvd
->vd_ldi_cbs
);
107 kmem_free(dvd
, sizeof (vdev_disk_t
));
113 vdev_disk_off_notify(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
, void *arg
,
116 vdev_t
*vd
= (vdev_t
*)arg
;
117 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
120 * Ignore events other than offline.
122 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
123 return (LDI_EV_SUCCESS
);
126 * All LDI handles must be closed for the state change to succeed, so
127 * call on vdev_disk_close() to do this.
129 * We inform vdev_disk_close that it is being called from offline
130 * notify context so it will defer cleanup of LDI event callbacks and
131 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
133 dvd
->vd_ldi_offline
= B_TRUE
;
137 * Now that the device is closed, request that the spa_async_thread
138 * mark the device as REMOVED and notify FMA of the removal.
140 zfs_post_remove(vd
->vdev_spa
, vd
);
141 vd
->vdev_remove_wanted
= B_TRUE
;
142 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_REMOVE
);
144 return (LDI_EV_SUCCESS
);
149 vdev_disk_off_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
150 int ldi_result
, void *arg
, void *ev_data
)
152 vdev_t
*vd
= (vdev_t
*)arg
;
155 * Ignore events other than offline.
157 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_OFFLINE
) != 0)
161 * We have already closed the LDI handle in notify.
162 * Clean up the LDI event callbacks and free vd->vdev_tsd.
167 * Request that the vdev be reopened if the offline state change was
170 if (ldi_result
!= LDI_EV_SUCCESS
) {
171 vd
->vdev_probe_wanted
= B_TRUE
;
172 spa_async_request(vd
->vdev_spa
, SPA_ASYNC_PROBE
);
176 static ldi_ev_callback_t vdev_disk_off_callb
= {
177 .cb_vers
= LDI_EV_CB_VERS
,
178 .cb_notify
= vdev_disk_off_notify
,
179 .cb_finalize
= vdev_disk_off_finalize
184 vdev_disk_dgrd_finalize(ldi_handle_t lh
, ldi_ev_cookie_t ecookie
,
185 int ldi_result
, void *arg
, void *ev_data
)
187 vdev_t
*vd
= (vdev_t
*)arg
;
190 * Ignore events other than degrade.
192 if (strcmp(ldi_ev_get_type(ecookie
), LDI_EV_DEGRADE
) != 0)
196 * Degrade events always succeed. Mark the vdev as degraded.
197 * This status is purely informative for the user.
199 (void) vdev_degrade(vd
->vdev_spa
, vd
->vdev_guid
, 0);
202 static ldi_ev_callback_t vdev_disk_dgrd_callb
= {
203 .cb_vers
= LDI_EV_CB_VERS
,
205 .cb_finalize
= vdev_disk_dgrd_finalize
209 vdev_disk_hold(vdev_t
*vd
)
214 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
217 * We must have a pathname, and it must be absolute.
219 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/')
223 * Only prefetch path and devid info if the device has
226 if (vd
->vdev_tsd
!= NULL
)
229 if (vd
->vdev_wholedisk
== -1ULL) {
230 size_t len
= strlen(vd
->vdev_path
) + 3;
231 char *buf
= kmem_alloc(len
, KM_SLEEP
);
233 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
235 (void) ldi_vp_from_name(buf
, &vd
->vdev_name_vp
);
239 if (vd
->vdev_name_vp
== NULL
)
240 (void) ldi_vp_from_name(vd
->vdev_path
, &vd
->vdev_name_vp
);
242 if (vd
->vdev_devid
!= NULL
&&
243 ddi_devid_str_decode(vd
->vdev_devid
, &devid
, &minor
) == 0) {
244 (void) ldi_vp_from_devid(devid
, minor
, &vd
->vdev_devid_vp
);
245 ddi_devid_str_free(minor
);
246 ddi_devid_free(devid
);
251 vdev_disk_rele(vdev_t
*vd
)
253 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_WRITER
));
255 if (vd
->vdev_name_vp
) {
256 VN_RELE_ASYNC(vd
->vdev_name_vp
,
257 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
258 vd
->vdev_name_vp
= NULL
;
260 if (vd
->vdev_devid_vp
) {
261 VN_RELE_ASYNC(vd
->vdev_devid_vp
,
262 dsl_pool_vnrele_taskq(vd
->vdev_spa
->spa_dsl_pool
));
263 vd
->vdev_devid_vp
= NULL
;
268 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
269 * even a fallback to DKIOCGMEDIAINFO fails.
272 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
274 #define VDEV_DEBUG(...) /* Nothing... */
278 vdev_disk_open(vdev_t
*vd
, uint64_t *psize
, uint64_t *max_psize
,
281 spa_t
*spa
= vd
->vdev_spa
;
282 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
283 ldi_ev_cookie_t ecookie
;
284 vdev_disk_ldi_cb_t
*lcb
;
286 struct dk_minfo_ext ude
;
289 struct dk_minfo_ext
*dkmext
= &dks
.ude
;
290 struct dk_minfo
*dkm
= &dks
.ud
;
294 boolean_t validate_devid
= B_FALSE
;
296 uint64_t capacity
= 0, blksz
= 0, pbsize
;
299 * We must have a pathname, and it must be absolute.
301 if (vd
->vdev_path
== NULL
|| vd
->vdev_path
[0] != '/') {
302 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
303 return (SET_ERROR(EINVAL
));
307 * Reopen the device if it's not currently open. Otherwise,
308 * just update the physical size of the device.
311 if (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
) {
313 * If we are opening a device in its offline notify
314 * context, the LDI handle was just closed. Clean
315 * up the LDI event callbacks and free vd->vdev_tsd.
319 ASSERT(vd
->vdev_reopening
);
325 * Create vd->vdev_tsd.
331 * Allow bypassing the devid.
333 if (vd
->vdev_devid
!= NULL
&& vdev_disk_bypass_devid
) {
334 vdev_dbgmsg(vd
, "vdev_disk_open, devid %s bypassed",
336 spa_strfree(vd
->vdev_devid
);
337 vd
->vdev_devid
= NULL
;
341 * When opening a disk device, we want to preserve the user's original
342 * intent. We always want to open the device by the path the user gave
343 * us, even if it is one of multiple paths to the same device. But we
344 * also want to be able to survive disks being removed/recabled.
345 * Therefore the sequence of opening devices is:
347 * 1. Try opening the device by path. For legacy pools without the
348 * 'whole_disk' property, attempt to fix the path by appending 's0'.
350 * 2. If the devid of the device matches the stored value, return
353 * 3. Otherwise, the device may have moved. Try opening the device
354 * by the devid instead.
356 if (vd
->vdev_devid
!= NULL
) {
357 if (ddi_devid_str_decode(vd
->vdev_devid
, &dvd
->vd_devid
,
358 &dvd
->vd_minor
) != 0) {
359 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
360 vdev_dbgmsg(vd
, "vdev_disk_open: invalid "
361 "vdev_devid '%s'", vd
->vdev_devid
);
362 return (SET_ERROR(EINVAL
));
366 error
= EINVAL
; /* presume failure */
368 if (vd
->vdev_path
!= NULL
) {
370 if (vd
->vdev_wholedisk
== -1ULL) {
371 size_t len
= strlen(vd
->vdev_path
) + 3;
372 char *buf
= kmem_alloc(len
, KM_SLEEP
);
374 (void) snprintf(buf
, len
, "%ss0", vd
->vdev_path
);
376 error
= ldi_open_by_name(buf
, spa_mode(spa
), kcred
,
377 &dvd
->vd_lh
, zfs_li
);
379 spa_strfree(vd
->vdev_path
);
381 vd
->vdev_wholedisk
= 1ULL;
388 * If we have not yet opened the device, try to open it by the
392 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
393 kcred
, &dvd
->vd_lh
, zfs_li
);
397 * Compare the devid to the stored value.
399 if (error
== 0 && vd
->vdev_devid
!= NULL
&&
400 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
401 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
403 * A mismatch here is unexpected, log it.
405 char *devid_str
= ddi_devid_str_encode(devid
,
407 vdev_dbgmsg(vd
, "vdev_disk_open: devid "
408 "mismatch: %s != %s", vd
->vdev_devid
,
410 cmn_err(CE_NOTE
, "vdev_disk_open %s: devid "
411 "mismatch: %s != %s", vd
->vdev_path
,
412 vd
->vdev_devid
, devid_str
);
413 ddi_devid_str_free(devid_str
);
415 error
= SET_ERROR(EINVAL
);
416 (void) ldi_close(dvd
->vd_lh
, spa_mode(spa
),
420 ddi_devid_free(devid
);
424 * If we succeeded in opening the device, but 'vdev_wholedisk'
425 * is not yet set, then this must be a slice.
427 if (error
== 0 && vd
->vdev_wholedisk
== -1ULL)
428 vd
->vdev_wholedisk
= 0;
432 * If we were unable to open by path, or the devid check fails, open by
435 if (error
!= 0 && vd
->vdev_devid
!= NULL
) {
436 error
= ldi_open_by_devid(dvd
->vd_devid
, dvd
->vd_minor
,
437 spa_mode(spa
), kcred
, &dvd
->vd_lh
, zfs_li
);
439 vdev_dbgmsg(vd
, "Failed to open by devid (%s)",
445 * If all else fails, then try opening by physical path (if available)
446 * or the logical path (if we failed due to the devid check). While not
447 * as reliable as the devid, this will give us something, and the higher
448 * level vdev validation will prevent us from opening the wrong device.
451 if (vd
->vdev_devid
!= NULL
)
452 validate_devid
= B_TRUE
;
454 if (vd
->vdev_physpath
!= NULL
&&
455 (dev
= ddi_pathname_to_dev_t(vd
->vdev_physpath
)) != NODEV
)
456 error
= ldi_open_by_dev(&dev
, OTYP_BLK
, spa_mode(spa
),
457 kcred
, &dvd
->vd_lh
, zfs_li
);
460 * Note that we don't support the legacy auto-wholedisk support
461 * as above. This hasn't been used in a very long time and we
462 * don't need to propagate its oddities to this edge condition.
464 if (error
&& vd
->vdev_path
!= NULL
)
465 error
= ldi_open_by_name(vd
->vdev_path
, spa_mode(spa
),
466 kcred
, &dvd
->vd_lh
, zfs_li
);
470 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
471 vdev_dbgmsg(vd
, "vdev_disk_open: failed to open [error=%d]",
477 * Now that the device has been successfully opened, update the devid
480 if (validate_devid
&& spa_writeable(spa
) &&
481 ldi_get_devid(dvd
->vd_lh
, &devid
) == 0) {
482 if (ddi_devid_compare(devid
, dvd
->vd_devid
) != 0) {
485 vd_devid
= ddi_devid_str_encode(devid
, dvd
->vd_minor
);
486 vdev_dbgmsg(vd
, "vdev_disk_open: update devid from "
487 "'%s' to '%s'", vd
->vdev_devid
, vd_devid
);
488 cmn_err(CE_NOTE
, "vdev_disk_open %s: update devid "
489 "from '%s' to '%s'", vd
->vdev_path
!= NULL
?
490 vd
->vdev_path
: "?", vd
->vdev_devid
, vd_devid
);
491 spa_strfree(vd
->vdev_devid
);
492 vd
->vdev_devid
= spa_strdup(vd_devid
);
493 ddi_devid_str_free(vd_devid
);
495 ddi_devid_free(devid
);
499 * Once a device is opened, verify that the physical device path (if
500 * available) is up to date.
502 if (ldi_get_dev(dvd
->vd_lh
, &dev
) == 0 &&
503 ldi_get_otyp(dvd
->vd_lh
, &otyp
) == 0) {
504 char *physpath
, *minorname
;
506 physpath
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
508 if (ddi_dev_pathname(dev
, otyp
, physpath
) == 0 &&
509 ldi_get_minor_name(dvd
->vd_lh
, &minorname
) == 0 &&
510 (vd
->vdev_physpath
== NULL
||
511 strcmp(vd
->vdev_physpath
, physpath
) != 0)) {
512 if (vd
->vdev_physpath
)
513 spa_strfree(vd
->vdev_physpath
);
514 (void) strlcat(physpath
, ":", MAXPATHLEN
);
515 (void) strlcat(physpath
, minorname
, MAXPATHLEN
);
516 vd
->vdev_physpath
= spa_strdup(physpath
);
519 kmem_free(minorname
, strlen(minorname
) + 1);
520 kmem_free(physpath
, MAXPATHLEN
);
524 * Register callbacks for the LDI offline event.
526 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_OFFLINE
, &ecookie
) ==
528 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
529 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
530 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
531 &vdev_disk_off_callb
, (void *) vd
, &lcb
->lcb_id
);
535 * Register callbacks for the LDI degrade event.
537 if (ldi_ev_get_cookie(dvd
->vd_lh
, LDI_EV_DEGRADE
, &ecookie
) ==
539 lcb
= kmem_zalloc(sizeof (vdev_disk_ldi_cb_t
), KM_SLEEP
);
540 list_insert_tail(&dvd
->vd_ldi_cbs
, lcb
);
541 (void) ldi_ev_register_callbacks(dvd
->vd_lh
, ecookie
,
542 &vdev_disk_dgrd_callb
, (void *) vd
, &lcb
->lcb_id
);
546 * Determine the actual size of the device.
548 if (ldi_get_size(dvd
->vd_lh
, psize
) != 0) {
549 vd
->vdev_stat
.vs_aux
= VDEV_AUX_OPEN_FAILED
;
550 vdev_dbgmsg(vd
, "vdev_disk_open: failed to get size");
551 return (SET_ERROR(EINVAL
));
557 * Determine the device's minimum transfer size.
558 * If the ioctl isn't supported, assume DEV_BSIZE.
560 if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFOEXT
,
561 (intptr_t)dkmext
, FKIOCTL
, kcred
, NULL
)) == 0) {
562 capacity
= dkmext
->dki_capacity
- 1;
563 blksz
= dkmext
->dki_lbsize
;
564 pbsize
= dkmext
->dki_pbsize
;
565 } else if ((error
= ldi_ioctl(dvd
->vd_lh
, DKIOCGMEDIAINFO
,
566 (intptr_t)dkm
, FKIOCTL
, kcred
, NULL
)) == 0) {
568 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
570 capacity
= dkm
->dki_capacity
- 1;
571 blksz
= dkm
->dki_lbsize
;
574 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
575 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
576 vd
->vdev_path
, error
);
580 *ashift
= highbit64(MAX(pbsize
, SPA_MINBLOCKSIZE
)) - 1;
582 if (vd
->vdev_wholedisk
== 1) {
587 * If we have the capability to expand, we'd have
588 * found out via success from DKIOCGMEDIAINFO{,EXT}.
589 * Adjust max_psize upward accordingly since we know
590 * we own the whole disk now.
592 *max_psize
= capacity
* blksz
;
596 * Since we own the whole disk, try to enable disk write
597 * caching. We ignore errors because it's OK if we can't do it.
599 (void) ldi_ioctl(dvd
->vd_lh
, DKIOCSETWCE
, (intptr_t)&wce
,
600 FKIOCTL
, kcred
, NULL
);
604 * Clear the nowritecache bit, so that on a vdev_reopen() we will
607 vd
->vdev_nowritecache
= B_FALSE
;
613 vdev_disk_close(vdev_t
*vd
)
615 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
617 if (vd
->vdev_reopening
|| dvd
== NULL
)
620 if (dvd
->vd_minor
!= NULL
) {
621 ddi_devid_str_free(dvd
->vd_minor
);
622 dvd
->vd_minor
= NULL
;
625 if (dvd
->vd_devid
!= NULL
) {
626 ddi_devid_free(dvd
->vd_devid
);
627 dvd
->vd_devid
= NULL
;
630 if (dvd
->vd_lh
!= NULL
) {
631 (void) ldi_close(dvd
->vd_lh
, spa_mode(vd
->vdev_spa
), kcred
);
635 vd
->vdev_delayed_close
= B_FALSE
;
637 * If we closed the LDI handle due to an offline notify from LDI,
638 * don't free vd->vdev_tsd or unregister the callbacks here;
639 * the offline finalize callback or a reopen will take care of it.
641 if (dvd
->vd_ldi_offline
)
648 vdev_disk_physio(vdev_t
*vd
, caddr_t data
,
649 size_t size
, uint64_t offset
, int flags
, boolean_t isdump
)
651 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
654 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
655 * Nothing to be done here but return failure.
657 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
))
660 ASSERT(vd
->vdev_ops
== &vdev_disk_ops
);
663 * If in the context of an active crash dump, use the ldi_dump(9F)
664 * call instead of ldi_strategy(9F) as usual.
667 ASSERT3P(dvd
, !=, NULL
);
668 return (ldi_dump(dvd
->vd_lh
, data
, lbtodb(offset
),
672 return (vdev_disk_ldi_physio(dvd
->vd_lh
, data
, size
, offset
, flags
));
676 vdev_disk_ldi_physio(ldi_handle_t vd_lh
, caddr_t data
,
677 size_t size
, uint64_t offset
, int flags
)
683 return (SET_ERROR(EINVAL
));
685 ASSERT(flags
& B_READ
|| flags
& B_WRITE
);
687 bp
= getrbuf(KM_SLEEP
);
688 bp
->b_flags
= flags
| B_BUSY
| B_NOCACHE
| B_FAILFAST
;
690 bp
->b_un
.b_addr
= (void *)data
;
691 bp
->b_lblkno
= lbtodb(offset
);
692 bp
->b_bufsize
= size
;
694 error
= ldi_strategy(vd_lh
, bp
);
696 if ((error
= biowait(bp
)) == 0 && bp
->b_resid
!= 0)
697 error
= SET_ERROR(EIO
);
704 vdev_disk_io_intr(buf_t
*bp
)
706 vdev_buf_t
*vb
= (vdev_buf_t
*)bp
;
707 zio_t
*zio
= vb
->vb_io
;
710 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
711 * Rather than teach the rest of the stack about other error
712 * possibilities (EFAULT, etc), we normalize the error value here.
714 zio
->io_error
= (geterror(bp
) != 0 ? EIO
: 0);
716 if (zio
->io_error
== 0 && bp
->b_resid
!= 0)
717 zio
->io_error
= SET_ERROR(EIO
);
719 if (zio
->io_type
== ZIO_TYPE_READ
) {
720 abd_return_buf_copy(zio
->io_abd
, bp
->b_un
.b_addr
, zio
->io_size
);
722 abd_return_buf(zio
->io_abd
, bp
->b_un
.b_addr
, zio
->io_size
);
725 kmem_free(vb
, sizeof (vdev_buf_t
));
727 zio_delay_interrupt(zio
);
732 vdev_disk_ioctl_free(zio_t
*zio
)
734 kmem_free(zio
->io_vsd
, sizeof (struct dk_callback
));
737 static const zio_vsd_ops_t vdev_disk_vsd_ops
= {
738 vdev_disk_ioctl_free
,
739 zio_vsd_default_cksum_report
743 vdev_disk_ioctl_done(void *zio_arg
, int error
)
745 zio_t
*zio
= zio_arg
;
747 zio
->io_error
= error
;
753 vdev_disk_io_start(zio_t
*zio
)
755 vdev_t
*vd
= zio
->io_vd
;
756 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
758 struct dk_callback
*dkc
;
763 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
764 * Nothing to be done here but return failure.
766 if (dvd
== NULL
|| (dvd
->vd_ldi_offline
&& dvd
->vd_lh
== NULL
)) {
767 zio
->io_error
= ENXIO
;
772 if (zio
->io_type
== ZIO_TYPE_IOCTL
) {
774 if (!vdev_readable(vd
)) {
775 zio
->io_error
= SET_ERROR(ENXIO
);
780 switch (zio
->io_cmd
) {
782 case DKIOCFLUSHWRITECACHE
:
784 if (zfs_nocacheflush
)
787 if (vd
->vdev_nowritecache
) {
788 zio
->io_error
= SET_ERROR(ENOTSUP
);
792 zio
->io_vsd
= dkc
= kmem_alloc(sizeof (*dkc
), KM_SLEEP
);
793 zio
->io_vsd_ops
= &vdev_disk_vsd_ops
;
795 dkc
->dkc_callback
= vdev_disk_ioctl_done
;
796 dkc
->dkc_flag
= FLUSH_VOLATILE
;
797 dkc
->dkc_cookie
= zio
;
799 error
= ldi_ioctl(dvd
->vd_lh
, zio
->io_cmd
,
800 (uintptr_t)dkc
, FKIOCTL
, kcred
, NULL
);
804 * The ioctl will be done asychronously,
805 * and will call vdev_disk_ioctl_done()
811 zio
->io_error
= error
;
816 zio
->io_error
= SET_ERROR(ENOTSUP
);
823 ASSERT(zio
->io_type
== ZIO_TYPE_READ
|| zio
->io_type
== ZIO_TYPE_WRITE
);
824 zio
->io_target_timestamp
= zio_handle_io_delay(zio
);
826 vb
= kmem_alloc(sizeof (vdev_buf_t
), KM_SLEEP
);
832 bp
->b_flags
= B_BUSY
| B_NOCACHE
|
833 (zio
->io_type
== ZIO_TYPE_READ
? B_READ
: B_WRITE
);
834 if (!(zio
->io_flags
& (ZIO_FLAG_IO_RETRY
| ZIO_FLAG_TRYHARD
)))
835 bp
->b_flags
|= B_FAILFAST
;
836 bp
->b_bcount
= zio
->io_size
;
838 if (zio
->io_type
== ZIO_TYPE_READ
) {
840 abd_borrow_buf(zio
->io_abd
, zio
->io_size
);
843 abd_borrow_buf_copy(zio
->io_abd
, zio
->io_size
);
846 bp
->b_lblkno
= lbtodb(zio
->io_offset
);
847 bp
->b_bufsize
= zio
->io_size
;
848 bp
->b_iodone
= vdev_disk_io_intr
;
850 /* ldi_strategy() will return non-zero only on programming errors */
851 VERIFY(ldi_strategy(dvd
->vd_lh
, bp
) == 0);
855 vdev_disk_io_done(zio_t
*zio
)
857 vdev_t
*vd
= zio
->io_vd
;
860 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
861 * the device has been removed. If this is the case, then we trigger an
862 * asynchronous removal of the device. Otherwise, probe the device and
863 * make sure it's still accessible.
865 if (zio
->io_error
== EIO
&& !vd
->vdev_remove_wanted
) {
866 vdev_disk_t
*dvd
= vd
->vdev_tsd
;
867 int state
= DKIO_NONE
;
869 if (ldi_ioctl(dvd
->vd_lh
, DKIOCSTATE
, (intptr_t)&state
,
870 FKIOCTL
, kcred
, NULL
) == 0 && state
!= DKIO_INSERTED
) {
872 * We post the resource as soon as possible, instead of
873 * when the async removal actually happens, because the
874 * DE is using this information to discard previous I/O
877 zfs_post_remove(zio
->io_spa
, vd
);
878 vd
->vdev_remove_wanted
= B_TRUE
;
879 spa_async_request(zio
->io_spa
, SPA_ASYNC_REMOVE
);
880 } else if (!vd
->vdev_delayed_close
) {
881 vd
->vdev_delayed_close
= B_TRUE
;
886 vdev_ops_t vdev_disk_ops
= {
897 VDEV_TYPE_DISK
, /* name of this vdev type */
898 B_TRUE
/* leaf vdev */
902 * Given the root disk device devid or pathname, read the label from
903 * the device, and construct a configuration nvlist.
906 vdev_disk_read_rootlabel(char *devpath
, char *devid
, nvlist_t
**config
)
912 ddi_devid_t tmpdevid
;
917 * Read the device label and build the nvlist.
919 if (devid
!= NULL
&& ddi_devid_str_decode(devid
, &tmpdevid
,
921 error
= ldi_open_by_devid(tmpdevid
, minor_name
,
922 FREAD
, kcred
, &vd_lh
, zfs_li
);
923 ddi_devid_free(tmpdevid
);
924 ddi_devid_str_free(minor_name
);
927 if (error
&& (error
= ldi_open_by_name(devpath
, FREAD
, kcred
, &vd_lh
,
931 if (ldi_get_size(vd_lh
, &s
)) {
932 (void) ldi_close(vd_lh
, FREAD
, kcred
);
933 return (SET_ERROR(EIO
));
936 size
= P2ALIGN_TYPED(s
, sizeof (vdev_label_t
), uint64_t);
937 label
= kmem_alloc(sizeof (vdev_label_t
), KM_SLEEP
);
940 for (l
= 0; l
< VDEV_LABELS
; l
++) {
941 uint64_t offset
, state
, txg
= 0;
943 /* read vdev label */
944 offset
= vdev_label_offset(size
, l
, 0);
945 if (vdev_disk_ldi_physio(vd_lh
, (caddr_t
)label
,
946 VDEV_SKIP_SIZE
+ VDEV_PHYS_SIZE
, offset
, B_READ
) != 0)
949 if (nvlist_unpack(label
->vl_vdev_phys
.vp_nvlist
,
950 sizeof (label
->vl_vdev_phys
.vp_nvlist
), config
, 0) != 0) {
955 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_STATE
,
956 &state
) != 0 || state
>= POOL_STATE_DESTROYED
) {
957 nvlist_free(*config
);
962 if (nvlist_lookup_uint64(*config
, ZPOOL_CONFIG_POOL_TXG
,
963 &txg
) != 0 || txg
== 0) {
964 nvlist_free(*config
);
972 kmem_free(label
, sizeof (vdev_label_t
));
973 (void) ldi_close(vd_lh
, FREAD
, kcred
);
975 error
= SET_ERROR(EIDRM
);