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 * Virtual device vector for disks.
44 extern ldi_ident_t zfs_li;
46 static void vdev_disk_close(vdev_t *);
48 typedef struct vdev_disk_ldi_cb {
50 ldi_callback_id_t lcb_id;
54 * Bypass the devid when opening a disk vdev.
55 * There have been issues where the devids of several devices were shuffled,
56 * causing pool open failures. Note, that this flag is intended to be used
57 * for pool recovery only.
59 * Note that if a pool is imported with the devids bypassed, all its vdevs will
60 * cease storing devid information permanently. In practice, the devid is rarely
61 * useful as vdev paths do not tend to change unless the hardware is
62 * reconfigured. That said, if the paths do change and a pool fails to open
63 * automatically at boot, a simple zpool import should re-scan the paths and fix
66 boolean_t vdev_disk_bypass_devid = B_FALSE;
69 vdev_disk_alloc(vdev_t *vd)
73 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
75 * Create the LDI event callback list.
77 list_create(&dvd->vd_ldi_cbs, sizeof (vdev_disk_ldi_cb_t),
78 offsetof(vdev_disk_ldi_cb_t, lcb_next));
82 vdev_disk_free(vdev_t *vd)
84 vdev_disk_t *dvd = vd->vdev_tsd;
85 vdev_disk_ldi_cb_t *lcb;
91 * We have already closed the LDI handle. Clean up the LDI event
92 * callbacks and free vd->vdev_tsd.
94 while ((lcb = list_head(&dvd->vd_ldi_cbs)) != NULL) {
95 list_remove(&dvd->vd_ldi_cbs, lcb);
96 (void) ldi_ev_remove_callbacks(lcb->lcb_id);
97 kmem_free(lcb, sizeof (vdev_disk_ldi_cb_t));
99 list_destroy(&dvd->vd_ldi_cbs);
100 kmem_free(dvd, sizeof (vdev_disk_t));
106 vdev_disk_off_notify(ldi_handle_t lh, ldi_ev_cookie_t ecookie, void *arg,
109 vdev_t *vd = (vdev_t *)arg;
110 vdev_disk_t *dvd = vd->vdev_tsd;
113 * Ignore events other than offline.
115 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
116 return (LDI_EV_SUCCESS);
119 * All LDI handles must be closed for the state change to succeed, so
120 * call on vdev_disk_close() to do this.
122 * We inform vdev_disk_close that it is being called from offline
123 * notify context so it will defer cleanup of LDI event callbacks and
124 * freeing of vd->vdev_tsd to the offline finalize or a reopen.
126 dvd->vd_ldi_offline = B_TRUE;
130 * Now that the device is closed, request that the spa_async_thread
131 * mark the device as REMOVED and notify FMA of the removal.
133 zfs_post_remove(vd->vdev_spa, vd);
134 vd->vdev_remove_wanted = B_TRUE;
135 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
137 return (LDI_EV_SUCCESS);
142 vdev_disk_off_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
143 int ldi_result, void *arg, void *ev_data)
145 vdev_t *vd = (vdev_t *)arg;
148 * Ignore events other than offline.
150 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_OFFLINE) != 0)
154 * We have already closed the LDI handle in notify.
155 * Clean up the LDI event callbacks and free vd->vdev_tsd.
160 * Request that the vdev be reopened if the offline state change was
163 if (ldi_result != LDI_EV_SUCCESS) {
164 vd->vdev_probe_wanted = B_TRUE;
165 spa_async_request(vd->vdev_spa, SPA_ASYNC_PROBE);
169 static ldi_ev_callback_t vdev_disk_off_callb = {
170 .cb_vers = LDI_EV_CB_VERS,
171 .cb_notify = vdev_disk_off_notify,
172 .cb_finalize = vdev_disk_off_finalize
177 vdev_disk_dgrd_finalize(ldi_handle_t lh, ldi_ev_cookie_t ecookie,
178 int ldi_result, void *arg, void *ev_data)
180 vdev_t *vd = (vdev_t *)arg;
183 * Ignore events other than degrade.
185 if (strcmp(ldi_ev_get_type(ecookie), LDI_EV_DEGRADE) != 0)
189 * Degrade events always succeed. Mark the vdev as degraded.
190 * This status is purely informative for the user.
192 (void) vdev_degrade(vd->vdev_spa, vd->vdev_guid, 0);
195 static ldi_ev_callback_t vdev_disk_dgrd_callb = {
196 .cb_vers = LDI_EV_CB_VERS,
198 .cb_finalize = vdev_disk_dgrd_finalize
202 vdev_disk_hold(vdev_t *vd)
207 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
210 * We must have a pathname, and it must be absolute.
212 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
216 * Only prefetch path and devid info if the device has
219 if (vd->vdev_tsd != NULL)
222 if (vd->vdev_wholedisk == -1ULL) {
223 size_t len = strlen(vd->vdev_path) + 3;
224 char *buf = kmem_alloc(len, KM_SLEEP);
226 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
228 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
232 if (vd->vdev_name_vp == NULL)
233 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
235 if (vd->vdev_devid != NULL &&
236 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
237 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
238 ddi_devid_str_free(minor);
239 ddi_devid_free(devid);
244 vdev_disk_rele(vdev_t *vd)
246 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
248 if (vd->vdev_name_vp) {
249 VN_RELE_ASYNC(vd->vdev_name_vp,
250 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
251 vd->vdev_name_vp = NULL;
253 if (vd->vdev_devid_vp) {
254 VN_RELE_ASYNC(vd->vdev_devid_vp,
255 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
256 vd->vdev_devid_vp = NULL;
261 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
262 * even a fallback to DKIOCGMEDIAINFO fails.
265 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
267 #define VDEV_DEBUG(...) /* Nothing... */
271 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
274 spa_t *spa = vd->vdev_spa;
275 vdev_disk_t *dvd = vd->vdev_tsd;
276 ldi_ev_cookie_t ecookie;
277 vdev_disk_ldi_cb_t *lcb;
279 struct dk_minfo_ext ude;
282 struct dk_minfo_ext *dkmext = &dks.ude;
283 struct dk_minfo *dkm = &dks.ud;
287 boolean_t validate_devid = B_FALSE;
289 uint64_t capacity = 0, blksz = 0, pbsize;
292 * We must have a pathname, and it must be absolute.
294 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
295 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
296 return (SET_ERROR(EINVAL));
300 * Reopen the device if it's not currently open. Otherwise,
301 * just update the physical size of the device.
304 if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
306 * If we are opening a device in its offline notify
307 * context, the LDI handle was just closed. Clean
308 * up the LDI event callbacks and free vd->vdev_tsd.
312 ASSERT(vd->vdev_reopening);
318 * Create vd->vdev_tsd.
324 * Allow bypassing the devid.
326 if (vd->vdev_devid != NULL && vdev_disk_bypass_devid) {
327 vdev_dbgmsg(vd, "vdev_disk_open, devid %s bypassed",
329 spa_strfree(vd->vdev_devid);
330 vd->vdev_devid = NULL;
334 * When opening a disk device, we want to preserve the user's original
335 * intent. We always want to open the device by the path the user gave
336 * us, even if it is one of multiple paths to the save device. But we
337 * also want to be able to survive disks being removed/recabled.
338 * Therefore the sequence of opening devices is:
340 * 1. Try opening the device by path. For legacy pools without the
341 * 'whole_disk' property, attempt to fix the path by appending 's0'.
343 * 2. If the devid of the device matches the stored value, return
346 * 3. Otherwise, the device may have moved. Try opening the device
347 * by the devid instead.
349 if (vd->vdev_devid != NULL) {
350 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
351 &dvd->vd_minor) != 0) {
352 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
353 vdev_dbgmsg(vd, "vdev_disk_open: invalid "
354 "vdev_devid '%s'", vd->vdev_devid);
355 return (SET_ERROR(EINVAL));
359 error = EINVAL; /* presume failure */
361 if (vd->vdev_path != NULL) {
363 if (vd->vdev_wholedisk == -1ULL) {
364 size_t len = strlen(vd->vdev_path) + 3;
365 char *buf = kmem_alloc(len, KM_SLEEP);
367 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
369 error = ldi_open_by_name(buf, spa_mode(spa), kcred,
370 &dvd->vd_lh, zfs_li);
372 spa_strfree(vd->vdev_path);
374 vd->vdev_wholedisk = 1ULL;
381 * If we have not yet opened the device, try to open it by the
385 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
386 kcred, &dvd->vd_lh, zfs_li);
390 * Compare the devid to the stored value.
392 if (error == 0 && vd->vdev_devid != NULL &&
393 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
394 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
396 * A mismatch here is unexpected, log it.
398 char *devid_str = ddi_devid_str_encode(devid,
400 vdev_dbgmsg(vd, "vdev_disk_open: devid "
401 "mismatch: %s != %s", vd->vdev_devid,
403 cmn_err(CE_NOTE, "vdev_disk_open %s: devid "
404 "mismatch: %s != %s", vd->vdev_path,
405 vd->vdev_devid, devid_str);
406 ddi_devid_str_free(devid_str);
408 error = SET_ERROR(EINVAL);
409 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
413 ddi_devid_free(devid);
417 * If we succeeded in opening the device, but 'vdev_wholedisk'
418 * is not yet set, then this must be a slice.
420 if (error == 0 && vd->vdev_wholedisk == -1ULL)
421 vd->vdev_wholedisk = 0;
425 * If we were unable to open by path, or the devid check fails, open by
428 if (error != 0 && vd->vdev_devid != NULL) {
429 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
430 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
432 vdev_dbgmsg(vd, "Failed to open by devid (%s)",
438 * If all else fails, then try opening by physical path (if available)
439 * or the logical path (if we failed due to the devid check). While not
440 * as reliable as the devid, this will give us something, and the higher
441 * level vdev validation will prevent us from opening the wrong device.
444 if (vd->vdev_devid != NULL)
445 validate_devid = B_TRUE;
447 if (vd->vdev_physpath != NULL &&
448 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
449 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
450 kcred, &dvd->vd_lh, zfs_li);
453 * Note that we don't support the legacy auto-wholedisk support
454 * as above. This hasn't been used in a very long time and we
455 * don't need to propagate its oddities to this edge condition.
457 if (error && vd->vdev_path != NULL)
458 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
459 kcred, &dvd->vd_lh, zfs_li);
463 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
464 vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]",
470 * Now that the device has been successfully opened, update the devid
473 if (validate_devid && spa_writeable(spa) &&
474 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
475 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
478 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
479 vdev_dbgmsg(vd, "vdev_disk_open: update devid from "
480 "'%s' to '%s'", vd->vdev_devid, vd_devid);
481 cmn_err(CE_NOTE, "vdev_disk_open %s: update devid "
482 "from '%s' to '%s'", vd->vdev_path != NULL ?
483 vd->vdev_path : "?", vd->vdev_devid, vd_devid);
484 spa_strfree(vd->vdev_devid);
485 vd->vdev_devid = spa_strdup(vd_devid);
486 ddi_devid_str_free(vd_devid);
488 ddi_devid_free(devid);
492 * Once a device is opened, verify that the physical device path (if
493 * available) is up to date.
495 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
496 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
497 char *physpath, *minorname;
499 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
501 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
502 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
503 (vd->vdev_physpath == NULL ||
504 strcmp(vd->vdev_physpath, physpath) != 0)) {
505 if (vd->vdev_physpath)
506 spa_strfree(vd->vdev_physpath);
507 (void) strlcat(physpath, ":", MAXPATHLEN);
508 (void) strlcat(physpath, minorname, MAXPATHLEN);
509 vd->vdev_physpath = spa_strdup(physpath);
512 kmem_free(minorname, strlen(minorname) + 1);
513 kmem_free(physpath, MAXPATHLEN);
517 * Register callbacks for the LDI offline event.
519 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
521 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
522 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
523 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
524 &vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
528 * Register callbacks for the LDI degrade event.
530 if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
532 lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
533 list_insert_tail(&dvd->vd_ldi_cbs, lcb);
534 (void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
535 &vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
539 * Determine the actual size of the device.
541 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
542 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
543 vdev_dbgmsg(vd, "vdev_disk_open: failed to get size");
544 return (SET_ERROR(EINVAL));
550 * Determine the device's minimum transfer size.
551 * If the ioctl isn't supported, assume DEV_BSIZE.
553 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
554 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
555 capacity = dkmext->dki_capacity - 1;
556 blksz = dkmext->dki_lbsize;
557 pbsize = dkmext->dki_pbsize;
558 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
559 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
561 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
563 capacity = dkm->dki_capacity - 1;
564 blksz = dkm->dki_lbsize;
567 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
568 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
569 vd->vdev_path, error);
573 *ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
575 if (vd->vdev_wholedisk == 1) {
580 * If we have the capability to expand, we'd have
581 * found out via success from DKIOCGMEDIAINFO{,EXT}.
582 * Adjust max_psize upward accordingly since we know
583 * we own the whole disk now.
585 *max_psize = capacity * blksz;
589 * Since we own the whole disk, try to enable disk write
590 * caching. We ignore errors because it's OK if we can't do it.
592 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
593 FKIOCTL, kcred, NULL);
597 * Clear the nowritecache bit, so that on a vdev_reopen() we will
600 vd->vdev_nowritecache = B_FALSE;
606 vdev_disk_close(vdev_t *vd)
608 vdev_disk_t *dvd = vd->vdev_tsd;
610 if (vd->vdev_reopening || dvd == NULL)
613 if (dvd->vd_minor != NULL) {
614 ddi_devid_str_free(dvd->vd_minor);
615 dvd->vd_minor = NULL;
618 if (dvd->vd_devid != NULL) {
619 ddi_devid_free(dvd->vd_devid);
620 dvd->vd_devid = NULL;
623 if (dvd->vd_lh != NULL) {
624 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
628 vd->vdev_delayed_close = B_FALSE;
630 * If we closed the LDI handle due to an offline notify from LDI,
631 * don't free vd->vdev_tsd or unregister the callbacks here;
632 * the offline finalize callback or a reopen will take care of it.
634 if (dvd->vd_ldi_offline)
641 vdev_disk_physio(vdev_t *vd, caddr_t data,
642 size_t size, uint64_t offset, int flags, boolean_t isdump)
644 vdev_disk_t *dvd = vd->vdev_tsd;
647 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
648 * Nothing to be done here but return failure.
650 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL))
653 ASSERT(vd->vdev_ops == &vdev_disk_ops);
656 * If in the context of an active crash dump, use the ldi_dump(9F)
657 * call instead of ldi_strategy(9F) as usual.
660 ASSERT3P(dvd, !=, NULL);
661 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
665 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
669 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
670 size_t size, uint64_t offset, int flags)
676 return (SET_ERROR(EINVAL));
678 ASSERT(flags & B_READ || flags & B_WRITE);
680 bp = getrbuf(KM_SLEEP);
681 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
683 bp->b_un.b_addr = (void *)data;
684 bp->b_lblkno = lbtodb(offset);
685 bp->b_bufsize = size;
687 error = ldi_strategy(vd_lh, bp);
689 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
690 error = SET_ERROR(EIO);
697 vdev_disk_io_intr(buf_t *bp)
699 vdev_buf_t *vb = (vdev_buf_t *)bp;
700 zio_t *zio = vb->vb_io;
703 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
704 * Rather than teach the rest of the stack about other error
705 * possibilities (EFAULT, etc), we normalize the error value here.
707 zio->io_error = (geterror(bp) != 0 ? SET_ERROR(EIO) : 0);
709 if (zio->io_error == 0 && bp->b_resid != 0)
710 zio->io_error = SET_ERROR(EIO);
712 if (zio->io_type == ZIO_TYPE_READ) {
713 abd_return_buf_copy(zio->io_abd, bp->b_un.b_addr, zio->io_size);
715 abd_return_buf(zio->io_abd, bp->b_un.b_addr, zio->io_size);
718 kmem_free(vb, sizeof (vdev_buf_t));
720 zio_delay_interrupt(zio);
724 vdev_disk_ioctl_free(zio_t *zio)
726 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
729 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
730 vdev_disk_ioctl_free,
731 zio_vsd_default_cksum_report
735 vdev_disk_ioctl_done(void *zio_arg, int error)
737 zio_t *zio = zio_arg;
739 zio->io_error = error;
745 vdev_disk_io_start(zio_t *zio)
747 vdev_t *vd = zio->io_vd;
748 vdev_disk_t *dvd = vd->vdev_tsd;
750 struct dk_callback *dkc;
755 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
756 * Nothing to be done here but return failure.
758 if (dvd == NULL || (dvd->vd_ldi_offline && dvd->vd_lh == NULL)) {
759 zio->io_error = SET_ERROR(ENXIO);
764 if (zio->io_type == ZIO_TYPE_IOCTL) {
766 if (!vdev_readable(vd)) {
767 zio->io_error = SET_ERROR(ENXIO);
772 switch (zio->io_cmd) {
774 case DKIOCFLUSHWRITECACHE:
776 if (zfs_nocacheflush)
779 if (vd->vdev_nowritecache) {
780 zio->io_error = SET_ERROR(ENOTSUP);
784 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
785 zio->io_vsd_ops = &vdev_disk_vsd_ops;
787 dkc->dkc_callback = vdev_disk_ioctl_done;
788 dkc->dkc_flag = FLUSH_VOLATILE;
789 dkc->dkc_cookie = zio;
791 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
792 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
796 * The ioctl will be done asychronously,
797 * and will call vdev_disk_ioctl_done()
803 zio->io_error = error;
808 zio->io_error = SET_ERROR(ENOTSUP);
815 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
816 zio->io_target_timestamp = zio_handle_io_delay(zio);
818 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
824 bp->b_flags = B_BUSY | B_NOCACHE |
825 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
826 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
827 bp->b_flags |= B_FAILFAST;
828 bp->b_bcount = zio->io_size;
830 if (zio->io_type == ZIO_TYPE_READ) {
832 abd_borrow_buf(zio->io_abd, zio->io_size);
835 abd_borrow_buf_copy(zio->io_abd, zio->io_size);
838 bp->b_lblkno = lbtodb(zio->io_offset);
839 bp->b_bufsize = zio->io_size;
840 bp->b_iodone = (int (*)())vdev_disk_io_intr;
842 /* ldi_strategy() will return non-zero only on programming errors */
843 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
847 vdev_disk_io_done(zio_t *zio)
849 vdev_t *vd = zio->io_vd;
852 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
853 * the device has been removed. If this is the case, then we trigger an
854 * asynchronous removal of the device. Otherwise, probe the device and
855 * make sure it's still accessible.
857 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
858 vdev_disk_t *dvd = vd->vdev_tsd;
859 int state = DKIO_NONE;
861 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
862 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
864 * We post the resource as soon as possible, instead of
865 * when the async removal actually happens, because the
866 * DE is using this information to discard previous I/O
869 zfs_post_remove(zio->io_spa, vd);
870 vd->vdev_remove_wanted = B_TRUE;
871 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
872 } else if (!vd->vdev_delayed_close) {
873 vd->vdev_delayed_close = B_TRUE;
878 vdev_ops_t vdev_disk_ops = {
890 VDEV_TYPE_DISK, /* name of this vdev type */
891 B_TRUE /* leaf vdev */
895 * Given the root disk device devid or pathname, read the label from
896 * the device, and construct a configuration nvlist.
899 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
905 ddi_devid_t tmpdevid;
910 * Read the device label and build the nvlist.
912 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
914 error = ldi_open_by_devid(tmpdevid, minor_name,
915 FREAD, kcred, &vd_lh, zfs_li);
916 ddi_devid_free(tmpdevid);
917 ddi_devid_str_free(minor_name);
920 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
924 if (ldi_get_size(vd_lh, &s)) {
925 (void) ldi_close(vd_lh, FREAD, kcred);
926 return (SET_ERROR(EIO));
929 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
930 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
933 for (l = 0; l < VDEV_LABELS; l++) {
934 uint64_t offset, state, txg = 0;
936 /* read vdev label */
937 offset = vdev_label_offset(size, l, 0);
938 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
939 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
942 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
943 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
948 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
949 &state) != 0 || state >= POOL_STATE_DESTROYED) {
950 nvlist_free(*config);
955 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
956 &txg) != 0 || txg == 0) {
957 nvlist_free(*config);
965 kmem_free(label, sizeof (vdev_label_t));
966 (void) ldi_close(vd_lh, FREAD, kcred);
968 error = SET_ERROR(EIDRM);