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]
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
27 #include <sys/zfs_context.h>
28 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa_impl.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/uberblock_impl.h>
35 #include <sys/metaslab.h>
36 #include <sys/metaslab_impl.h>
37 #include <sys/space_map.h>
40 #include <sys/fs/zfs.h>
43 SYSCTL_DECL(_vfs_zfs);
44 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
47 * Virtual device management.
50 static vdev_ops_t *vdev_ops_table[] = {
66 /* maximum scrub/resilver I/O queue per leaf vdev */
67 int zfs_scrub_limit = 10;
69 TUNABLE_INT("vfs.zfs.scrub_limit", &zfs_scrub_limit);
70 SYSCTL_INT(_vfs_zfs, OID_AUTO, scrub_limit, CTLFLAG_RDTUN, &zfs_scrub_limit, 0,
71 "Maximum scrub/resilver I/O queue");
74 * Given a vdev type, return the appropriate ops vector.
77 vdev_getops(const char *type)
79 vdev_ops_t *ops, **opspp;
81 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
82 if (strcmp(ops->vdev_op_type, type) == 0)
89 * Default asize function: return the MAX of psize with the asize of
90 * all children. This is what's used by anything other than RAID-Z.
93 vdev_default_asize(vdev_t *vd, uint64_t psize)
95 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
99 for (c = 0; c < vd->vdev_children; c++) {
100 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
101 asize = MAX(asize, csize);
108 * Get the replaceable or attachable device size.
109 * If the parent is a mirror or raidz, the replaceable size is the minimum
110 * psize of all its children. For the rest, just return our own psize.
121 vdev_get_rsize(vdev_t *vd)
126 pvd = vd->vdev_parent;
129 * If our parent is NULL or the root, just return our own psize.
131 if (pvd == NULL || pvd->vdev_parent == NULL)
132 return (vd->vdev_psize);
136 for (c = 0; c < pvd->vdev_children; c++) {
137 cvd = pvd->vdev_child[c];
138 rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1;
145 vdev_lookup_top(spa_t *spa, uint64_t vdev)
147 vdev_t *rvd = spa->spa_root_vdev;
149 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
151 if (vdev < rvd->vdev_children) {
152 ASSERT(rvd->vdev_child[vdev] != NULL);
153 return (rvd->vdev_child[vdev]);
160 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
165 if (vd->vdev_guid == guid)
168 for (c = 0; c < vd->vdev_children; c++)
169 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
177 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
179 size_t oldsize, newsize;
180 uint64_t id = cvd->vdev_id;
183 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
184 ASSERT(cvd->vdev_parent == NULL);
186 cvd->vdev_parent = pvd;
191 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
193 oldsize = pvd->vdev_children * sizeof (vdev_t *);
194 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
195 newsize = pvd->vdev_children * sizeof (vdev_t *);
197 newchild = kmem_zalloc(newsize, KM_SLEEP);
198 if (pvd->vdev_child != NULL) {
199 bcopy(pvd->vdev_child, newchild, oldsize);
200 kmem_free(pvd->vdev_child, oldsize);
203 pvd->vdev_child = newchild;
204 pvd->vdev_child[id] = cvd;
206 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
207 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
210 * Walk up all ancestors to update guid sum.
212 for (; pvd != NULL; pvd = pvd->vdev_parent)
213 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
215 if (cvd->vdev_ops->vdev_op_leaf)
216 cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit;
220 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
223 uint_t id = cvd->vdev_id;
225 ASSERT(cvd->vdev_parent == pvd);
230 ASSERT(id < pvd->vdev_children);
231 ASSERT(pvd->vdev_child[id] == cvd);
233 pvd->vdev_child[id] = NULL;
234 cvd->vdev_parent = NULL;
236 for (c = 0; c < pvd->vdev_children; c++)
237 if (pvd->vdev_child[c])
240 if (c == pvd->vdev_children) {
241 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
242 pvd->vdev_child = NULL;
243 pvd->vdev_children = 0;
247 * Walk up all ancestors to update guid sum.
249 for (; pvd != NULL; pvd = pvd->vdev_parent)
250 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
252 if (cvd->vdev_ops->vdev_op_leaf)
253 cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit;
257 * Remove any holes in the child array.
260 vdev_compact_children(vdev_t *pvd)
262 vdev_t **newchild, *cvd;
263 int oldc = pvd->vdev_children;
266 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
268 for (c = newc = 0; c < oldc; c++)
269 if (pvd->vdev_child[c])
272 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
274 for (c = newc = 0; c < oldc; c++) {
275 if ((cvd = pvd->vdev_child[c]) != NULL) {
276 newchild[newc] = cvd;
277 cvd->vdev_id = newc++;
281 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
282 pvd->vdev_child = newchild;
283 pvd->vdev_children = newc;
287 * Allocate and minimally initialize a vdev_t.
290 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
294 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
296 if (spa->spa_root_vdev == NULL) {
297 ASSERT(ops == &vdev_root_ops);
298 spa->spa_root_vdev = vd;
302 if (spa->spa_root_vdev == vd) {
304 * The root vdev's guid will also be the pool guid,
305 * which must be unique among all pools.
307 while (guid == 0 || spa_guid_exists(guid, 0))
308 guid = spa_get_random(-1ULL);
311 * Any other vdev's guid must be unique within the pool.
314 spa_guid_exists(spa_guid(spa), guid))
315 guid = spa_get_random(-1ULL);
317 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
322 vd->vdev_guid = guid;
323 vd->vdev_guid_sum = guid;
325 vd->vdev_state = VDEV_STATE_CLOSED;
327 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
328 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
329 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
330 for (int t = 0; t < DTL_TYPES; t++) {
331 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
334 txg_list_create(&vd->vdev_ms_list,
335 offsetof(struct metaslab, ms_txg_node));
336 txg_list_create(&vd->vdev_dtl_list,
337 offsetof(struct vdev, vdev_dtl_node));
338 vd->vdev_stat.vs_timestamp = gethrtime();
346 * Allocate a new vdev. The 'alloctype' is used to control whether we are
347 * creating a new vdev or loading an existing one - the behavior is slightly
348 * different for each case.
351 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
356 uint64_t guid = 0, islog, nparity;
359 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
361 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
364 if ((ops = vdev_getops(type)) == NULL)
368 * If this is a load, get the vdev guid from the nvlist.
369 * Otherwise, vdev_alloc_common() will generate one for us.
371 if (alloctype == VDEV_ALLOC_LOAD) {
374 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
378 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
380 } else if (alloctype == VDEV_ALLOC_SPARE) {
381 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
383 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
384 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
389 * The first allocated vdev must be of type 'root'.
391 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
395 * Determine whether we're a log vdev.
398 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
399 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
403 * Set the nparity property for RAID-Z vdevs.
406 if (ops == &vdev_raidz_ops) {
407 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
410 * Currently, we can only support 2 parity devices.
412 if (nparity == 0 || nparity > 2)
415 * Older versions can only support 1 parity device.
418 spa_version(spa) < SPA_VERSION_RAID6)
422 * We require the parity to be specified for SPAs that
423 * support multiple parity levels.
425 if (spa_version(spa) >= SPA_VERSION_RAID6)
428 * Otherwise, we default to 1 parity device for RAID-Z.
435 ASSERT(nparity != -1ULL);
437 vd = vdev_alloc_common(spa, id, guid, ops);
439 vd->vdev_islog = islog;
440 vd->vdev_nparity = nparity;
442 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
443 vd->vdev_path = spa_strdup(vd->vdev_path);
444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
445 vd->vdev_devid = spa_strdup(vd->vdev_devid);
446 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
447 &vd->vdev_physpath) == 0)
448 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
449 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
450 vd->vdev_fru = spa_strdup(vd->vdev_fru);
453 * Set the whole_disk property. If it's not specified, leave the value
456 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
457 &vd->vdev_wholedisk) != 0)
458 vd->vdev_wholedisk = -1ULL;
461 * Look for the 'not present' flag. This will only be set if the device
462 * was not present at the time of import.
464 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
465 &vd->vdev_not_present);
468 * Get the alignment requirement.
470 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
473 * If we're a top-level vdev, try to load the allocation parameters.
475 if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) {
476 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
478 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
485 * If we're a leaf vdev, try to load the DTL object and other state.
487 if (vd->vdev_ops->vdev_op_leaf &&
488 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE)) {
489 if (alloctype == VDEV_ALLOC_LOAD) {
490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
491 &vd->vdev_dtl_smo.smo_object);
492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
495 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
499 * When importing a pool, we want to ignore the persistent fault
500 * state, as the diagnosis made on another system may not be
501 * valid in the current context.
503 if (spa->spa_load_state == SPA_LOAD_OPEN) {
504 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
506 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
508 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
514 * Add ourselves to the parent's list of children.
516 vdev_add_child(parent, vd);
524 vdev_free(vdev_t *vd)
527 spa_t *spa = vd->vdev_spa;
530 * vdev_free() implies closing the vdev first. This is simpler than
531 * trying to ensure complicated semantics for all callers.
535 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
540 for (c = 0; c < vd->vdev_children; c++)
541 vdev_free(vd->vdev_child[c]);
543 ASSERT(vd->vdev_child == NULL);
544 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
547 * Discard allocation state.
549 if (vd == vd->vdev_top)
550 vdev_metaslab_fini(vd);
552 ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
553 ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
554 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
557 * Remove this vdev from its parent's child list.
559 vdev_remove_child(vd->vdev_parent, vd);
561 ASSERT(vd->vdev_parent == NULL);
564 * Clean up vdev structure.
570 spa_strfree(vd->vdev_path);
572 spa_strfree(vd->vdev_devid);
573 if (vd->vdev_physpath)
574 spa_strfree(vd->vdev_physpath);
576 spa_strfree(vd->vdev_fru);
578 if (vd->vdev_isspare)
579 spa_spare_remove(vd);
580 if (vd->vdev_isl2cache)
581 spa_l2cache_remove(vd);
583 txg_list_destroy(&vd->vdev_ms_list);
584 txg_list_destroy(&vd->vdev_dtl_list);
586 mutex_enter(&vd->vdev_dtl_lock);
587 for (int t = 0; t < DTL_TYPES; t++) {
588 space_map_unload(&vd->vdev_dtl[t]);
589 space_map_destroy(&vd->vdev_dtl[t]);
591 mutex_exit(&vd->vdev_dtl_lock);
593 mutex_destroy(&vd->vdev_dtl_lock);
594 mutex_destroy(&vd->vdev_stat_lock);
595 mutex_destroy(&vd->vdev_probe_lock);
597 if (vd == spa->spa_root_vdev)
598 spa->spa_root_vdev = NULL;
600 kmem_free(vd, sizeof (vdev_t));
604 * Transfer top-level vdev state from svd to tvd.
607 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
609 spa_t *spa = svd->vdev_spa;
614 ASSERT(tvd == tvd->vdev_top);
616 tvd->vdev_ms_array = svd->vdev_ms_array;
617 tvd->vdev_ms_shift = svd->vdev_ms_shift;
618 tvd->vdev_ms_count = svd->vdev_ms_count;
620 svd->vdev_ms_array = 0;
621 svd->vdev_ms_shift = 0;
622 svd->vdev_ms_count = 0;
624 tvd->vdev_mg = svd->vdev_mg;
625 tvd->vdev_ms = svd->vdev_ms;
630 if (tvd->vdev_mg != NULL)
631 tvd->vdev_mg->mg_vd = tvd;
633 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
634 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
635 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
637 svd->vdev_stat.vs_alloc = 0;
638 svd->vdev_stat.vs_space = 0;
639 svd->vdev_stat.vs_dspace = 0;
641 for (t = 0; t < TXG_SIZE; t++) {
642 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
643 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
644 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
645 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
646 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
647 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
650 if (list_link_active(&svd->vdev_config_dirty_node)) {
651 vdev_config_clean(svd);
652 vdev_config_dirty(tvd);
655 if (list_link_active(&svd->vdev_state_dirty_node)) {
656 vdev_state_clean(svd);
657 vdev_state_dirty(tvd);
660 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
661 svd->vdev_deflate_ratio = 0;
663 tvd->vdev_islog = svd->vdev_islog;
668 vdev_top_update(vdev_t *tvd, vdev_t *vd)
677 for (c = 0; c < vd->vdev_children; c++)
678 vdev_top_update(tvd, vd->vdev_child[c]);
682 * Add a mirror/replacing vdev above an existing vdev.
685 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
687 spa_t *spa = cvd->vdev_spa;
688 vdev_t *pvd = cvd->vdev_parent;
691 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
693 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
695 mvd->vdev_asize = cvd->vdev_asize;
696 mvd->vdev_ashift = cvd->vdev_ashift;
697 mvd->vdev_state = cvd->vdev_state;
699 vdev_remove_child(pvd, cvd);
700 vdev_add_child(pvd, mvd);
701 cvd->vdev_id = mvd->vdev_children;
702 vdev_add_child(mvd, cvd);
703 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
705 if (mvd == mvd->vdev_top)
706 vdev_top_transfer(cvd, mvd);
712 * Remove a 1-way mirror/replacing vdev from the tree.
715 vdev_remove_parent(vdev_t *cvd)
717 vdev_t *mvd = cvd->vdev_parent;
718 vdev_t *pvd = mvd->vdev_parent;
720 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
722 ASSERT(mvd->vdev_children == 1);
723 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
724 mvd->vdev_ops == &vdev_replacing_ops ||
725 mvd->vdev_ops == &vdev_spare_ops);
726 cvd->vdev_ashift = mvd->vdev_ashift;
728 vdev_remove_child(mvd, cvd);
729 vdev_remove_child(pvd, mvd);
732 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
733 * Otherwise, we could have detached an offline device, and when we
734 * go to import the pool we'll think we have two top-level vdevs,
735 * instead of a different version of the same top-level vdev.
737 if (mvd->vdev_top == mvd) {
738 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
739 cvd->vdev_guid += guid_delta;
740 cvd->vdev_guid_sum += guid_delta;
742 cvd->vdev_id = mvd->vdev_id;
743 vdev_add_child(pvd, cvd);
744 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
746 if (cvd == cvd->vdev_top)
747 vdev_top_transfer(mvd, cvd);
749 ASSERT(mvd->vdev_children == 0);
754 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
756 spa_t *spa = vd->vdev_spa;
757 objset_t *mos = spa->spa_meta_objset;
758 metaslab_class_t *mc;
760 uint64_t oldc = vd->vdev_ms_count;
761 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
765 if (vd->vdev_ms_shift == 0) /* not being allocated from yet */
768 ASSERT(oldc <= newc);
771 mc = spa->spa_log_class;
773 mc = spa->spa_normal_class;
775 if (vd->vdev_mg == NULL)
776 vd->vdev_mg = metaslab_group_create(mc, vd);
778 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
781 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
782 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
786 vd->vdev_ms_count = newc;
788 for (m = oldc; m < newc; m++) {
789 space_map_obj_t smo = { 0, 0, 0 };
792 error = dmu_read(mos, vd->vdev_ms_array,
793 m * sizeof (uint64_t), sizeof (uint64_t), &object,
799 error = dmu_bonus_hold(mos, object, FTAG, &db);
802 ASSERT3U(db->db_size, >=, sizeof (smo));
803 bcopy(db->db_data, &smo, sizeof (smo));
804 ASSERT3U(smo.smo_object, ==, object);
805 dmu_buf_rele(db, FTAG);
808 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
809 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
816 vdev_metaslab_fini(vdev_t *vd)
819 uint64_t count = vd->vdev_ms_count;
821 if (vd->vdev_ms != NULL) {
822 for (m = 0; m < count; m++)
823 if (vd->vdev_ms[m] != NULL)
824 metaslab_fini(vd->vdev_ms[m]);
825 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
830 typedef struct vdev_probe_stats {
831 boolean_t vps_readable;
832 boolean_t vps_writeable;
834 } vdev_probe_stats_t;
837 vdev_probe_done(zio_t *zio)
839 spa_t *spa = zio->io_spa;
840 vdev_t *vd = zio->io_vd;
841 vdev_probe_stats_t *vps = zio->io_private;
843 ASSERT(vd->vdev_probe_zio != NULL);
845 if (zio->io_type == ZIO_TYPE_READ) {
846 if (zio->io_error == 0)
847 vps->vps_readable = 1;
848 if (zio->io_error == 0 && spa_writeable(spa)) {
849 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
850 zio->io_offset, zio->io_size, zio->io_data,
851 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
852 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
854 zio_buf_free(zio->io_data, zio->io_size);
856 } else if (zio->io_type == ZIO_TYPE_WRITE) {
857 if (zio->io_error == 0)
858 vps->vps_writeable = 1;
859 zio_buf_free(zio->io_data, zio->io_size);
860 } else if (zio->io_type == ZIO_TYPE_NULL) {
863 vd->vdev_cant_read |= !vps->vps_readable;
864 vd->vdev_cant_write |= !vps->vps_writeable;
866 if (vdev_readable(vd) &&
867 (vdev_writeable(vd) || !spa_writeable(spa))) {
870 ASSERT(zio->io_error != 0);
871 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
872 spa, vd, NULL, 0, 0);
873 zio->io_error = ENXIO;
876 mutex_enter(&vd->vdev_probe_lock);
877 ASSERT(vd->vdev_probe_zio == zio);
878 vd->vdev_probe_zio = NULL;
879 mutex_exit(&vd->vdev_probe_lock);
881 while ((pio = zio_walk_parents(zio)) != NULL)
882 if (!vdev_accessible(vd, pio))
883 pio->io_error = ENXIO;
885 kmem_free(vps, sizeof (*vps));
890 * Determine whether this device is accessible by reading and writing
891 * to several known locations: the pad regions of each vdev label
892 * but the first (which we leave alone in case it contains a VTOC).
895 vdev_probe(vdev_t *vd, zio_t *zio)
897 spa_t *spa = vd->vdev_spa;
898 vdev_probe_stats_t *vps = NULL;
901 ASSERT(vd->vdev_ops->vdev_op_leaf);
904 * Don't probe the probe.
906 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
910 * To prevent 'probe storms' when a device fails, we create
911 * just one probe i/o at a time. All zios that want to probe
912 * this vdev will become parents of the probe io.
914 mutex_enter(&vd->vdev_probe_lock);
916 if ((pio = vd->vdev_probe_zio) == NULL) {
917 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
919 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
920 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
923 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
925 * vdev_cant_read and vdev_cant_write can only
926 * transition from TRUE to FALSE when we have the
927 * SCL_ZIO lock as writer; otherwise they can only
928 * transition from FALSE to TRUE. This ensures that
929 * any zio looking at these values can assume that
930 * failures persist for the life of the I/O. That's
931 * important because when a device has intermittent
932 * connectivity problems, we want to ensure that
933 * they're ascribed to the device (ENXIO) and not
936 * Since we hold SCL_ZIO as writer here, clear both
937 * values so the probe can reevaluate from first
940 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
941 vd->vdev_cant_read = B_FALSE;
942 vd->vdev_cant_write = B_FALSE;
945 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
946 vdev_probe_done, vps,
947 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
950 vd->vdev_probe_wanted = B_TRUE;
951 spa_async_request(spa, SPA_ASYNC_PROBE);
956 zio_add_child(zio, pio);
958 mutex_exit(&vd->vdev_probe_lock);
965 for (int l = 1; l < VDEV_LABELS; l++) {
966 zio_nowait(zio_read_phys(pio, vd,
967 vdev_label_offset(vd->vdev_psize, l,
968 offsetof(vdev_label_t, vl_pad2)),
969 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
970 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
971 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
982 * Prepare a virtual device for access.
985 vdev_open(vdev_t *vd)
987 spa_t *spa = vd->vdev_spa;
991 uint64_t asize, psize;
994 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
996 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
997 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
998 vd->vdev_state == VDEV_STATE_OFFLINE);
1000 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1002 if (!vd->vdev_removed && vd->vdev_faulted) {
1003 ASSERT(vd->vdev_children == 0);
1004 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1005 VDEV_AUX_ERR_EXCEEDED);
1007 } else if (vd->vdev_offline) {
1008 ASSERT(vd->vdev_children == 0);
1009 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1013 error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
1015 if (zio_injection_enabled && error == 0)
1016 error = zio_handle_device_injection(vd, ENXIO);
1019 if (vd->vdev_removed &&
1020 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1021 vd->vdev_removed = B_FALSE;
1023 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1024 vd->vdev_stat.vs_aux);
1028 vd->vdev_removed = B_FALSE;
1030 if (vd->vdev_degraded) {
1031 ASSERT(vd->vdev_children == 0);
1032 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1033 VDEV_AUX_ERR_EXCEEDED);
1035 vd->vdev_state = VDEV_STATE_HEALTHY;
1038 for (c = 0; c < vd->vdev_children; c++)
1039 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1040 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1045 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1047 if (vd->vdev_children == 0) {
1048 if (osize < SPA_MINDEVSIZE) {
1049 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1050 VDEV_AUX_TOO_SMALL);
1054 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1056 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1057 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1058 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1059 VDEV_AUX_TOO_SMALL);
1066 vd->vdev_psize = psize;
1068 if (vd->vdev_asize == 0) {
1070 * This is the first-ever open, so use the computed values.
1071 * For testing purposes, a higher ashift can be requested.
1073 vd->vdev_asize = asize;
1074 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1077 * Make sure the alignment requirement hasn't increased.
1079 if (ashift > vd->vdev_top->vdev_ashift) {
1080 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1081 VDEV_AUX_BAD_LABEL);
1086 * Make sure the device hasn't shrunk.
1088 if (asize < vd->vdev_asize) {
1089 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1090 VDEV_AUX_BAD_LABEL);
1095 * If all children are healthy and the asize has increased,
1096 * then we've experienced dynamic LUN growth.
1098 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1099 asize > vd->vdev_asize) {
1100 vd->vdev_asize = asize;
1105 * Ensure we can issue some IO before declaring the
1106 * vdev open for business.
1108 if (vd->vdev_ops->vdev_op_leaf &&
1109 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1110 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1111 VDEV_AUX_IO_FAILURE);
1116 * If this is a top-level vdev, compute the raidz-deflation
1117 * ratio. Note, we hard-code in 128k (1<<17) because it is the
1118 * current "typical" blocksize. Even if SPA_MAXBLOCKSIZE
1119 * changes, this algorithm must never change, or we will
1120 * inconsistently account for existing bp's.
1122 if (vd->vdev_top == vd) {
1123 vd->vdev_deflate_ratio = (1<<17) /
1124 (vdev_psize_to_asize(vd, 1<<17) >> SPA_MINBLOCKSHIFT);
1128 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1129 * resilver. But don't do this if we are doing a reopen for a scrub,
1130 * since this would just restart the scrub we are already doing.
1132 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1133 vdev_resilver_needed(vd, NULL, NULL))
1134 spa_async_request(spa, SPA_ASYNC_RESILVER);
1140 * Called once the vdevs are all opened, this routine validates the label
1141 * contents. This needs to be done before vdev_load() so that we don't
1142 * inadvertently do repair I/Os to the wrong device.
1144 * This function will only return failure if one of the vdevs indicates that it
1145 * has since been destroyed or exported. This is only possible if
1146 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1147 * will be updated but the function will return 0.
1150 vdev_validate(vdev_t *vd)
1152 spa_t *spa = vd->vdev_spa;
1155 uint64_t guid, top_guid;
1158 for (c = 0; c < vd->vdev_children; c++)
1159 if (vdev_validate(vd->vdev_child[c]) != 0)
1163 * If the device has already failed, or was marked offline, don't do
1164 * any further validation. Otherwise, label I/O will fail and we will
1165 * overwrite the previous state.
1167 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1169 if ((label = vdev_label_read_config(vd)) == NULL) {
1170 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1171 VDEV_AUX_BAD_LABEL);
1175 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
1176 &guid) != 0 || guid != spa_guid(spa)) {
1177 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1178 VDEV_AUX_CORRUPT_DATA);
1184 * If this vdev just became a top-level vdev because its
1185 * sibling was detached, it will have adopted the parent's
1186 * vdev guid -- but the label may or may not be on disk yet.
1187 * Fortunately, either version of the label will have the
1188 * same top guid, so if we're a top-level vdev, we can
1189 * safely compare to that instead.
1191 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1193 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1195 (vd->vdev_guid != guid &&
1196 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1197 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1198 VDEV_AUX_CORRUPT_DATA);
1203 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1205 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1206 VDEV_AUX_CORRUPT_DATA);
1214 * If spa->spa_load_verbatim is true, no need to check the
1215 * state of the pool.
1217 if (!spa->spa_load_verbatim &&
1218 spa->spa_load_state == SPA_LOAD_OPEN &&
1219 state != POOL_STATE_ACTIVE)
1223 * If we were able to open and validate a vdev that was
1224 * previously marked permanently unavailable, clear that state
1227 if (vd->vdev_not_present)
1228 vd->vdev_not_present = 0;
1235 * Close a virtual device.
1238 vdev_close(vdev_t *vd)
1240 spa_t *spa = vd->vdev_spa;
1242 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1244 vd->vdev_ops->vdev_op_close(vd);
1246 vdev_cache_purge(vd);
1249 * We record the previous state before we close it, so that if we are
1250 * doing a reopen(), we don't generate FMA ereports if we notice that
1251 * it's still faulted.
1253 vd->vdev_prevstate = vd->vdev_state;
1255 if (vd->vdev_offline)
1256 vd->vdev_state = VDEV_STATE_OFFLINE;
1258 vd->vdev_state = VDEV_STATE_CLOSED;
1259 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1263 vdev_reopen(vdev_t *vd)
1265 spa_t *spa = vd->vdev_spa;
1267 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1270 (void) vdev_open(vd);
1273 * Call vdev_validate() here to make sure we have the same device.
1274 * Otherwise, a device with an invalid label could be successfully
1275 * opened in response to vdev_reopen().
1278 (void) vdev_validate_aux(vd);
1279 if (vdev_readable(vd) && vdev_writeable(vd) &&
1280 vd->vdev_aux == &spa->spa_l2cache &&
1281 !l2arc_vdev_present(vd)) {
1282 uint64_t size = vdev_get_rsize(vd);
1283 l2arc_add_vdev(spa, vd,
1284 VDEV_LABEL_START_SIZE,
1285 size - VDEV_LABEL_START_SIZE);
1288 (void) vdev_validate(vd);
1292 * Reassess parent vdev's health.
1294 vdev_propagate_state(vd);
1298 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1303 * Normally, partial opens (e.g. of a mirror) are allowed.
1304 * For a create, however, we want to fail the request if
1305 * there are any components we can't open.
1307 error = vdev_open(vd);
1309 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1311 return (error ? error : ENXIO);
1315 * Recursively initialize all labels.
1317 if ((error = vdev_label_init(vd, txg, isreplacing ?
1318 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1327 * The is the latter half of vdev_create(). It is distinct because it
1328 * involves initiating transactions in order to do metaslab creation.
1329 * For creation, we want to try to create all vdevs at once and then undo it
1330 * if anything fails; this is much harder if we have pending transactions.
1333 vdev_init(vdev_t *vd, uint64_t txg)
1336 * Aim for roughly 200 metaslabs per vdev.
1338 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1339 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1342 * Initialize the vdev's metaslabs. This can't fail because
1343 * there's nothing to read when creating all new metaslabs.
1345 VERIFY(vdev_metaslab_init(vd, txg) == 0);
1349 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1351 ASSERT(vd == vd->vdev_top);
1352 ASSERT(ISP2(flags));
1354 if (flags & VDD_METASLAB)
1355 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1357 if (flags & VDD_DTL)
1358 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1360 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1366 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1367 * the vdev has less than perfect replication. There are three kinds of DTL:
1369 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1371 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1373 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1374 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1375 * txgs that was scrubbed.
1377 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1378 * persistent errors or just some device being offline.
1379 * Unlike the other three, the DTL_OUTAGE map is not generally
1380 * maintained; it's only computed when needed, typically to
1381 * determine whether a device can be detached.
1383 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1384 * either has the data or it doesn't.
1386 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1387 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1388 * if any child is less than fully replicated, then so is its parent.
1389 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1390 * comprising only those txgs which appear in 'maxfaults' or more children;
1391 * those are the txgs we don't have enough replication to read. For example,
1392 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1393 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1394 * two child DTL_MISSING maps.
1396 * It should be clear from the above that to compute the DTLs and outage maps
1397 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1398 * Therefore, that is all we keep on disk. When loading the pool, or after
1399 * a configuration change, we generate all other DTLs from first principles.
1402 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1404 space_map_t *sm = &vd->vdev_dtl[t];
1406 ASSERT(t < DTL_TYPES);
1407 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1409 mutex_enter(sm->sm_lock);
1410 if (!space_map_contains(sm, txg, size))
1411 space_map_add(sm, txg, size);
1412 mutex_exit(sm->sm_lock);
1416 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1418 space_map_t *sm = &vd->vdev_dtl[t];
1419 boolean_t dirty = B_FALSE;
1421 ASSERT(t < DTL_TYPES);
1422 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1424 mutex_enter(sm->sm_lock);
1425 if (sm->sm_space != 0)
1426 dirty = space_map_contains(sm, txg, size);
1427 mutex_exit(sm->sm_lock);
1433 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1435 space_map_t *sm = &vd->vdev_dtl[t];
1438 mutex_enter(sm->sm_lock);
1439 empty = (sm->sm_space == 0);
1440 mutex_exit(sm->sm_lock);
1446 * Reassess DTLs after a config change or scrub completion.
1449 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1451 spa_t *spa = vd->vdev_spa;
1455 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1457 for (int c = 0; c < vd->vdev_children; c++)
1458 vdev_dtl_reassess(vd->vdev_child[c], txg,
1459 scrub_txg, scrub_done);
1461 if (vd == spa->spa_root_vdev)
1464 if (vd->vdev_ops->vdev_op_leaf) {
1465 mutex_enter(&vd->vdev_dtl_lock);
1466 if (scrub_txg != 0 &&
1467 (spa->spa_scrub_started || spa->spa_scrub_errors == 0)) {
1468 /* XXX should check scrub_done? */
1470 * We completed a scrub up to scrub_txg. If we
1471 * did it without rebooting, then the scrub dtl
1472 * will be valid, so excise the old region and
1473 * fold in the scrub dtl. Otherwise, leave the
1474 * dtl as-is if there was an error.
1476 * There's little trick here: to excise the beginning
1477 * of the DTL_MISSING map, we put it into a reference
1478 * tree and then add a segment with refcnt -1 that
1479 * covers the range [0, scrub_txg). This means
1480 * that each txg in that range has refcnt -1 or 0.
1481 * We then add DTL_SCRUB with a refcnt of 2, so that
1482 * entries in the range [0, scrub_txg) will have a
1483 * positive refcnt -- either 1 or 2. We then convert
1484 * the reference tree into the new DTL_MISSING map.
1486 space_map_ref_create(&reftree);
1487 space_map_ref_add_map(&reftree,
1488 &vd->vdev_dtl[DTL_MISSING], 1);
1489 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1490 space_map_ref_add_map(&reftree,
1491 &vd->vdev_dtl[DTL_SCRUB], 2);
1492 space_map_ref_generate_map(&reftree,
1493 &vd->vdev_dtl[DTL_MISSING], 1);
1494 space_map_ref_destroy(&reftree);
1496 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1497 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1498 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1500 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1501 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1502 if (!vdev_readable(vd))
1503 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1505 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1506 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1507 mutex_exit(&vd->vdev_dtl_lock);
1510 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1514 mutex_enter(&vd->vdev_dtl_lock);
1515 for (int t = 0; t < DTL_TYPES; t++) {
1516 /* account for child's outage in parent's missing map */
1517 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1519 continue; /* leaf vdevs only */
1520 if (t == DTL_PARTIAL)
1521 minref = 1; /* i.e. non-zero */
1522 else if (vd->vdev_nparity != 0)
1523 minref = vd->vdev_nparity + 1; /* RAID-Z */
1525 minref = vd->vdev_children; /* any kind of mirror */
1526 space_map_ref_create(&reftree);
1527 for (int c = 0; c < vd->vdev_children; c++) {
1528 vdev_t *cvd = vd->vdev_child[c];
1529 mutex_enter(&cvd->vdev_dtl_lock);
1530 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1531 mutex_exit(&cvd->vdev_dtl_lock);
1533 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1534 space_map_ref_destroy(&reftree);
1536 mutex_exit(&vd->vdev_dtl_lock);
1540 vdev_dtl_load(vdev_t *vd)
1542 spa_t *spa = vd->vdev_spa;
1543 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1544 objset_t *mos = spa->spa_meta_objset;
1548 ASSERT(vd->vdev_children == 0);
1550 if (smo->smo_object == 0)
1553 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1556 ASSERT3U(db->db_size, >=, sizeof (*smo));
1557 bcopy(db->db_data, smo, sizeof (*smo));
1558 dmu_buf_rele(db, FTAG);
1560 mutex_enter(&vd->vdev_dtl_lock);
1561 error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1562 NULL, SM_ALLOC, smo, mos);
1563 mutex_exit(&vd->vdev_dtl_lock);
1569 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1571 spa_t *spa = vd->vdev_spa;
1572 space_map_obj_t *smo = &vd->vdev_dtl_smo;
1573 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1574 objset_t *mos = spa->spa_meta_objset;
1580 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1582 if (vd->vdev_detached) {
1583 if (smo->smo_object != 0) {
1584 int err = dmu_object_free(mos, smo->smo_object, tx);
1585 ASSERT3U(err, ==, 0);
1586 smo->smo_object = 0;
1592 if (smo->smo_object == 0) {
1593 ASSERT(smo->smo_objsize == 0);
1594 ASSERT(smo->smo_alloc == 0);
1595 smo->smo_object = dmu_object_alloc(mos,
1596 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1597 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1598 ASSERT(smo->smo_object != 0);
1599 vdev_config_dirty(vd->vdev_top);
1602 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1604 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1607 mutex_enter(&smlock);
1609 mutex_enter(&vd->vdev_dtl_lock);
1610 space_map_walk(sm, space_map_add, &smsync);
1611 mutex_exit(&vd->vdev_dtl_lock);
1613 space_map_truncate(smo, mos, tx);
1614 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1616 space_map_destroy(&smsync);
1618 mutex_exit(&smlock);
1619 mutex_destroy(&smlock);
1621 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1622 dmu_buf_will_dirty(db, tx);
1623 ASSERT3U(db->db_size, >=, sizeof (*smo));
1624 bcopy(smo, db->db_data, sizeof (*smo));
1625 dmu_buf_rele(db, FTAG);
1631 * Determine whether the specified vdev can be offlined/detached/removed
1632 * without losing data.
1635 vdev_dtl_required(vdev_t *vd)
1637 spa_t *spa = vd->vdev_spa;
1638 vdev_t *tvd = vd->vdev_top;
1639 uint8_t cant_read = vd->vdev_cant_read;
1642 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1644 if (vd == spa->spa_root_vdev || vd == tvd)
1648 * Temporarily mark the device as unreadable, and then determine
1649 * whether this results in any DTL outages in the top-level vdev.
1650 * If not, we can safely offline/detach/remove the device.
1652 vd->vdev_cant_read = B_TRUE;
1653 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1654 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1655 vd->vdev_cant_read = cant_read;
1656 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1662 * Determine if resilver is needed, and if so the txg range.
1665 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1667 boolean_t needed = B_FALSE;
1668 uint64_t thismin = UINT64_MAX;
1669 uint64_t thismax = 0;
1671 if (vd->vdev_children == 0) {
1672 mutex_enter(&vd->vdev_dtl_lock);
1673 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1674 vdev_writeable(vd)) {
1677 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1678 thismin = ss->ss_start - 1;
1679 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1680 thismax = ss->ss_end;
1683 mutex_exit(&vd->vdev_dtl_lock);
1685 for (int c = 0; c < vd->vdev_children; c++) {
1686 vdev_t *cvd = vd->vdev_child[c];
1687 uint64_t cmin, cmax;
1689 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1690 thismin = MIN(thismin, cmin);
1691 thismax = MAX(thismax, cmax);
1697 if (needed && minp) {
1705 vdev_load(vdev_t *vd)
1708 * Recursively load all children.
1710 for (int c = 0; c < vd->vdev_children; c++)
1711 vdev_load(vd->vdev_child[c]);
1714 * If this is a top-level vdev, initialize its metaslabs.
1716 if (vd == vd->vdev_top &&
1717 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1718 vdev_metaslab_init(vd, 0) != 0))
1719 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1720 VDEV_AUX_CORRUPT_DATA);
1723 * If this is a leaf vdev, load its DTL.
1725 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1726 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1727 VDEV_AUX_CORRUPT_DATA);
1731 * The special vdev case is used for hot spares and l2cache devices. Its
1732 * sole purpose it to set the vdev state for the associated vdev. To do this,
1733 * we make sure that we can open the underlying device, then try to read the
1734 * label, and make sure that the label is sane and that it hasn't been
1735 * repurposed to another pool.
1738 vdev_validate_aux(vdev_t *vd)
1741 uint64_t guid, version;
1744 if (!vdev_readable(vd))
1747 if ((label = vdev_label_read_config(vd)) == NULL) {
1748 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1749 VDEV_AUX_CORRUPT_DATA);
1753 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1754 version > SPA_VERSION ||
1755 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1756 guid != vd->vdev_guid ||
1757 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1758 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1759 VDEV_AUX_CORRUPT_DATA);
1765 * We don't actually check the pool state here. If it's in fact in
1766 * use by another pool, we update this fact on the fly when requested.
1773 vdev_sync_done(vdev_t *vd, uint64_t txg)
1777 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
1778 metaslab_sync_done(msp, txg);
1782 vdev_sync(vdev_t *vd, uint64_t txg)
1784 spa_t *spa = vd->vdev_spa;
1789 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
1790 ASSERT(vd == vd->vdev_top);
1791 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1792 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
1793 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
1794 ASSERT(vd->vdev_ms_array != 0);
1795 vdev_config_dirty(vd);
1799 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
1800 metaslab_sync(msp, txg);
1801 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
1804 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
1805 vdev_dtl_sync(lvd, txg);
1807 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
1811 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
1813 return (vd->vdev_ops->vdev_op_asize(vd, psize));
1817 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
1818 * not be opened, and no I/O is attempted.
1821 vdev_fault(spa_t *spa, uint64_t guid)
1825 spa_vdev_state_enter(spa);
1827 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1828 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1830 if (!vd->vdev_ops->vdev_op_leaf)
1831 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1834 * Faulted state takes precedence over degraded.
1836 vd->vdev_faulted = 1ULL;
1837 vd->vdev_degraded = 0ULL;
1838 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, VDEV_AUX_ERR_EXCEEDED);
1841 * If marking the vdev as faulted cause the top-level vdev to become
1842 * unavailable, then back off and simply mark the vdev as degraded
1845 if (vdev_is_dead(vd->vdev_top) && vd->vdev_aux == NULL) {
1846 vd->vdev_degraded = 1ULL;
1847 vd->vdev_faulted = 0ULL;
1850 * If we reopen the device and it's not dead, only then do we
1855 if (vdev_readable(vd)) {
1856 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1857 VDEV_AUX_ERR_EXCEEDED);
1861 return (spa_vdev_state_exit(spa, vd, 0));
1865 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
1866 * user that something is wrong. The vdev continues to operate as normal as far
1867 * as I/O is concerned.
1870 vdev_degrade(spa_t *spa, uint64_t guid)
1874 spa_vdev_state_enter(spa);
1876 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1877 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1879 if (!vd->vdev_ops->vdev_op_leaf)
1880 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1883 * If the vdev is already faulted, then don't do anything.
1885 if (vd->vdev_faulted || vd->vdev_degraded)
1886 return (spa_vdev_state_exit(spa, NULL, 0));
1888 vd->vdev_degraded = 1ULL;
1889 if (!vdev_is_dead(vd))
1890 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1891 VDEV_AUX_ERR_EXCEEDED);
1893 return (spa_vdev_state_exit(spa, vd, 0));
1897 * Online the given vdev. If 'unspare' is set, it implies two things. First,
1898 * any attached spare device should be detached when the device finishes
1899 * resilvering. Second, the online should be treated like a 'test' online case,
1900 * so no FMA events are generated if the device fails to open.
1903 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
1907 spa_vdev_state_enter(spa);
1909 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1910 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1912 if (!vd->vdev_ops->vdev_op_leaf)
1913 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1915 vd->vdev_offline = B_FALSE;
1916 vd->vdev_tmpoffline = B_FALSE;
1917 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
1918 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
1919 vdev_reopen(vd->vdev_top);
1920 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
1923 *newstate = vd->vdev_state;
1924 if ((flags & ZFS_ONLINE_UNSPARE) &&
1925 !vdev_is_dead(vd) && vd->vdev_parent &&
1926 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
1927 vd->vdev_parent->vdev_child[0] == vd)
1928 vd->vdev_unspare = B_TRUE;
1930 return (spa_vdev_state_exit(spa, vd, 0));
1934 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
1938 spa_vdev_state_enter(spa);
1940 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
1941 return (spa_vdev_state_exit(spa, NULL, ENODEV));
1943 if (!vd->vdev_ops->vdev_op_leaf)
1944 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
1947 * If the device isn't already offline, try to offline it.
1949 if (!vd->vdev_offline) {
1951 * If this device has the only valid copy of some data,
1952 * don't allow it to be offlined.
1954 if (vd->vdev_aux == NULL && vdev_dtl_required(vd))
1955 return (spa_vdev_state_exit(spa, NULL, EBUSY));
1958 * Offline this device and reopen its top-level vdev.
1959 * If this action results in the top-level vdev becoming
1960 * unusable, undo it and fail the request.
1962 vd->vdev_offline = B_TRUE;
1963 vdev_reopen(vd->vdev_top);
1964 if (vd->vdev_aux == NULL && vdev_is_dead(vd->vdev_top)) {
1965 vd->vdev_offline = B_FALSE;
1966 vdev_reopen(vd->vdev_top);
1967 return (spa_vdev_state_exit(spa, NULL, EBUSY));
1971 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
1973 return (spa_vdev_state_exit(spa, vd, 0));
1977 * Clear the error counts associated with this vdev. Unlike vdev_online() and
1978 * vdev_offline(), we assume the spa config is locked. We also clear all
1979 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
1982 vdev_clear(spa_t *spa, vdev_t *vd)
1984 vdev_t *rvd = spa->spa_root_vdev;
1986 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1991 vd->vdev_stat.vs_read_errors = 0;
1992 vd->vdev_stat.vs_write_errors = 0;
1993 vd->vdev_stat.vs_checksum_errors = 0;
1995 for (int c = 0; c < vd->vdev_children; c++)
1996 vdev_clear(spa, vd->vdev_child[c]);
1999 * If we're in the FAULTED state or have experienced failed I/O, then
2000 * clear the persistent state and attempt to reopen the device. We
2001 * also mark the vdev config dirty, so that the new faulted state is
2002 * written out to disk.
2004 if (vd->vdev_faulted || vd->vdev_degraded ||
2005 !vdev_readable(vd) || !vdev_writeable(vd)) {
2007 vd->vdev_faulted = vd->vdev_degraded = 0;
2008 vd->vdev_cant_read = B_FALSE;
2009 vd->vdev_cant_write = B_FALSE;
2014 vdev_state_dirty(vd->vdev_top);
2016 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2017 spa_async_request(spa, SPA_ASYNC_RESILVER);
2019 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2024 vdev_is_dead(vdev_t *vd)
2026 return (vd->vdev_state < VDEV_STATE_DEGRADED);
2030 vdev_readable(vdev_t *vd)
2032 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2036 vdev_writeable(vdev_t *vd)
2038 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2042 vdev_allocatable(vdev_t *vd)
2044 uint64_t state = vd->vdev_state;
2047 * We currently allow allocations from vdevs which may be in the
2048 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2049 * fails to reopen then we'll catch it later when we're holding
2050 * the proper locks. Note that we have to get the vdev state
2051 * in a local variable because although it changes atomically,
2052 * we're asking two separate questions about it.
2054 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2055 !vd->vdev_cant_write);
2059 vdev_accessible(vdev_t *vd, zio_t *zio)
2061 ASSERT(zio->io_vd == vd);
2063 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2066 if (zio->io_type == ZIO_TYPE_READ)
2067 return (!vd->vdev_cant_read);
2069 if (zio->io_type == ZIO_TYPE_WRITE)
2070 return (!vd->vdev_cant_write);
2076 * Get statistics for the given vdev.
2079 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2081 vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2083 mutex_enter(&vd->vdev_stat_lock);
2084 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2085 vs->vs_scrub_errors = vd->vdev_spa->spa_scrub_errors;
2086 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2087 vs->vs_state = vd->vdev_state;
2088 vs->vs_rsize = vdev_get_rsize(vd);
2089 mutex_exit(&vd->vdev_stat_lock);
2092 * If we're getting stats on the root vdev, aggregate the I/O counts
2093 * over all top-level vdevs (i.e. the direct children of the root).
2096 for (int c = 0; c < rvd->vdev_children; c++) {
2097 vdev_t *cvd = rvd->vdev_child[c];
2098 vdev_stat_t *cvs = &cvd->vdev_stat;
2100 mutex_enter(&vd->vdev_stat_lock);
2101 for (int t = 0; t < ZIO_TYPES; t++) {
2102 vs->vs_ops[t] += cvs->vs_ops[t];
2103 vs->vs_bytes[t] += cvs->vs_bytes[t];
2105 vs->vs_scrub_examined += cvs->vs_scrub_examined;
2106 mutex_exit(&vd->vdev_stat_lock);
2112 vdev_clear_stats(vdev_t *vd)
2114 mutex_enter(&vd->vdev_stat_lock);
2115 vd->vdev_stat.vs_space = 0;
2116 vd->vdev_stat.vs_dspace = 0;
2117 vd->vdev_stat.vs_alloc = 0;
2118 mutex_exit(&vd->vdev_stat_lock);
2122 vdev_stat_update(zio_t *zio, uint64_t psize)
2124 spa_t *spa = zio->io_spa;
2125 vdev_t *rvd = spa->spa_root_vdev;
2126 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2128 uint64_t txg = zio->io_txg;
2129 vdev_stat_t *vs = &vd->vdev_stat;
2130 zio_type_t type = zio->io_type;
2131 int flags = zio->io_flags;
2134 * If this i/o is a gang leader, it didn't do any actual work.
2136 if (zio->io_gang_tree)
2139 if (zio->io_error == 0) {
2141 * If this is a root i/o, don't count it -- we've already
2142 * counted the top-level vdevs, and vdev_get_stats() will
2143 * aggregate them when asked. This reduces contention on
2144 * the root vdev_stat_lock and implicitly handles blocks
2145 * that compress away to holes, for which there is no i/o.
2146 * (Holes never create vdev children, so all the counters
2147 * remain zero, which is what we want.)
2149 * Note: this only applies to successful i/o (io_error == 0)
2150 * because unlike i/o counts, errors are not additive.
2151 * When reading a ditto block, for example, failure of
2152 * one top-level vdev does not imply a root-level error.
2157 ASSERT(vd == zio->io_vd);
2159 if (flags & ZIO_FLAG_IO_BYPASS)
2162 mutex_enter(&vd->vdev_stat_lock);
2164 if (flags & ZIO_FLAG_IO_REPAIR) {
2165 if (flags & ZIO_FLAG_SCRUB_THREAD)
2166 vs->vs_scrub_repaired += psize;
2167 if (flags & ZIO_FLAG_SELF_HEAL)
2168 vs->vs_self_healed += psize;
2172 vs->vs_bytes[type] += psize;
2174 mutex_exit(&vd->vdev_stat_lock);
2178 if (flags & ZIO_FLAG_SPECULATIVE)
2181 mutex_enter(&vd->vdev_stat_lock);
2182 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2183 if (zio->io_error == ECKSUM)
2184 vs->vs_checksum_errors++;
2186 vs->vs_read_errors++;
2188 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2189 vs->vs_write_errors++;
2190 mutex_exit(&vd->vdev_stat_lock);
2192 if (type == ZIO_TYPE_WRITE && txg != 0 &&
2193 (!(flags & ZIO_FLAG_IO_REPAIR) ||
2194 (flags & ZIO_FLAG_SCRUB_THREAD))) {
2196 * This is either a normal write (not a repair), or it's a
2197 * repair induced by the scrub thread. In the normal case,
2198 * we commit the DTL change in the same txg as the block
2199 * was born. In the scrub-induced repair case, we know that
2200 * scrubs run in first-pass syncing context, so we commit
2201 * the DTL change in spa->spa_syncing_txg.
2203 * We currently do not make DTL entries for failed spontaneous
2204 * self-healing writes triggered by normal (non-scrubbing)
2205 * reads, because we have no transactional context in which to
2206 * do so -- and it's not clear that it'd be desirable anyway.
2208 if (vd->vdev_ops->vdev_op_leaf) {
2209 uint64_t commit_txg = txg;
2210 if (flags & ZIO_FLAG_SCRUB_THREAD) {
2211 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2212 ASSERT(spa_sync_pass(spa) == 1);
2213 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2214 commit_txg = spa->spa_syncing_txg;
2216 ASSERT(commit_txg >= spa->spa_syncing_txg);
2217 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2219 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2220 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2221 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2224 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2229 vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete)
2232 vdev_stat_t *vs = &vd->vdev_stat;
2234 for (c = 0; c < vd->vdev_children; c++)
2235 vdev_scrub_stat_update(vd->vdev_child[c], type, complete);
2237 mutex_enter(&vd->vdev_stat_lock);
2239 if (type == POOL_SCRUB_NONE) {
2241 * Update completion and end time. Leave everything else alone
2242 * so we can report what happened during the previous scrub.
2244 vs->vs_scrub_complete = complete;
2245 vs->vs_scrub_end = gethrestime_sec();
2247 vs->vs_scrub_type = type;
2248 vs->vs_scrub_complete = 0;
2249 vs->vs_scrub_examined = 0;
2250 vs->vs_scrub_repaired = 0;
2251 vs->vs_scrub_start = gethrestime_sec();
2252 vs->vs_scrub_end = 0;
2255 mutex_exit(&vd->vdev_stat_lock);
2259 * Update the in-core space usage stats for this vdev and the root vdev.
2262 vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta,
2263 boolean_t update_root)
2265 int64_t dspace_delta = space_delta;
2266 spa_t *spa = vd->vdev_spa;
2267 vdev_t *rvd = spa->spa_root_vdev;
2269 ASSERT(vd == vd->vdev_top);
2272 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2273 * factor. We must calculate this here and not at the root vdev
2274 * because the root vdev's psize-to-asize is simply the max of its
2275 * childrens', thus not accurate enough for us.
2277 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2278 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2279 vd->vdev_deflate_ratio;
2281 mutex_enter(&vd->vdev_stat_lock);
2282 vd->vdev_stat.vs_space += space_delta;
2283 vd->vdev_stat.vs_alloc += alloc_delta;
2284 vd->vdev_stat.vs_dspace += dspace_delta;
2285 mutex_exit(&vd->vdev_stat_lock);
2288 ASSERT(rvd == vd->vdev_parent);
2289 ASSERT(vd->vdev_ms_count != 0);
2292 * Don't count non-normal (e.g. intent log) space as part of
2293 * the pool's capacity.
2295 if (vd->vdev_mg->mg_class != spa->spa_normal_class)
2298 mutex_enter(&rvd->vdev_stat_lock);
2299 rvd->vdev_stat.vs_space += space_delta;
2300 rvd->vdev_stat.vs_alloc += alloc_delta;
2301 rvd->vdev_stat.vs_dspace += dspace_delta;
2302 mutex_exit(&rvd->vdev_stat_lock);
2307 * Mark a top-level vdev's config as dirty, placing it on the dirty list
2308 * so that it will be written out next time the vdev configuration is synced.
2309 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2312 vdev_config_dirty(vdev_t *vd)
2314 spa_t *spa = vd->vdev_spa;
2315 vdev_t *rvd = spa->spa_root_vdev;
2319 * If this is an aux vdev (as with l2cache and spare devices), then we
2320 * update the vdev config manually and set the sync flag.
2322 if (vd->vdev_aux != NULL) {
2323 spa_aux_vdev_t *sav = vd->vdev_aux;
2327 for (c = 0; c < sav->sav_count; c++) {
2328 if (sav->sav_vdevs[c] == vd)
2332 if (c == sav->sav_count) {
2334 * We're being removed. There's nothing more to do.
2336 ASSERT(sav->sav_sync == B_TRUE);
2340 sav->sav_sync = B_TRUE;
2342 if (nvlist_lookup_nvlist_array(sav->sav_config,
2343 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2344 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2345 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2351 * Setting the nvlist in the middle if the array is a little
2352 * sketchy, but it will work.
2354 nvlist_free(aux[c]);
2355 aux[c] = vdev_config_generate(spa, vd, B_TRUE, B_FALSE, B_TRUE);
2361 * The dirty list is protected by the SCL_CONFIG lock. The caller
2362 * must either hold SCL_CONFIG as writer, or must be the sync thread
2363 * (which holds SCL_CONFIG as reader). There's only one sync thread,
2364 * so this is sufficient to ensure mutual exclusion.
2366 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2367 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2368 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2371 for (c = 0; c < rvd->vdev_children; c++)
2372 vdev_config_dirty(rvd->vdev_child[c]);
2374 ASSERT(vd == vd->vdev_top);
2376 if (!list_link_active(&vd->vdev_config_dirty_node))
2377 list_insert_head(&spa->spa_config_dirty_list, vd);
2382 vdev_config_clean(vdev_t *vd)
2384 spa_t *spa = vd->vdev_spa;
2386 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2387 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2388 spa_config_held(spa, SCL_CONFIG, RW_READER)));
2390 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2391 list_remove(&spa->spa_config_dirty_list, vd);
2395 * Mark a top-level vdev's state as dirty, so that the next pass of
2396 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
2397 * the state changes from larger config changes because they require
2398 * much less locking, and are often needed for administrative actions.
2401 vdev_state_dirty(vdev_t *vd)
2403 spa_t *spa = vd->vdev_spa;
2405 ASSERT(vd == vd->vdev_top);
2408 * The state list is protected by the SCL_STATE lock. The caller
2409 * must either hold SCL_STATE as writer, or must be the sync thread
2410 * (which holds SCL_STATE as reader). There's only one sync thread,
2411 * so this is sufficient to ensure mutual exclusion.
2413 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2414 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2415 spa_config_held(spa, SCL_STATE, RW_READER)));
2417 if (!list_link_active(&vd->vdev_state_dirty_node))
2418 list_insert_head(&spa->spa_state_dirty_list, vd);
2422 vdev_state_clean(vdev_t *vd)
2424 spa_t *spa = vd->vdev_spa;
2426 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2427 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2428 spa_config_held(spa, SCL_STATE, RW_READER)));
2430 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2431 list_remove(&spa->spa_state_dirty_list, vd);
2435 * Propagate vdev state up from children to parent.
2438 vdev_propagate_state(vdev_t *vd)
2440 spa_t *spa = vd->vdev_spa;
2441 vdev_t *rvd = spa->spa_root_vdev;
2442 int degraded = 0, faulted = 0;
2447 if (vd->vdev_children > 0) {
2448 for (c = 0; c < vd->vdev_children; c++) {
2449 child = vd->vdev_child[c];
2451 if (!vdev_readable(child) ||
2452 (!vdev_writeable(child) && spa_writeable(spa))) {
2454 * Root special: if there is a top-level log
2455 * device, treat the root vdev as if it were
2458 if (child->vdev_islog && vd == rvd)
2462 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2466 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2470 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2473 * Root special: if there is a top-level vdev that cannot be
2474 * opened due to corrupted metadata, then propagate the root
2475 * vdev's aux state as 'corrupt' rather than 'insufficient
2478 if (corrupted && vd == rvd &&
2479 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2480 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2481 VDEV_AUX_CORRUPT_DATA);
2484 if (vd->vdev_parent)
2485 vdev_propagate_state(vd->vdev_parent);
2489 * Set a vdev's state. If this is during an open, we don't update the parent
2490 * state, because we're in the process of opening children depth-first.
2491 * Otherwise, we propagate the change to the parent.
2493 * If this routine places a device in a faulted state, an appropriate ereport is
2497 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2499 uint64_t save_state;
2500 spa_t *spa = vd->vdev_spa;
2502 if (state == vd->vdev_state) {
2503 vd->vdev_stat.vs_aux = aux;
2507 save_state = vd->vdev_state;
2509 vd->vdev_state = state;
2510 vd->vdev_stat.vs_aux = aux;
2513 * If we are setting the vdev state to anything but an open state, then
2514 * always close the underlying device. Otherwise, we keep accessible
2515 * but invalid devices open forever. We don't call vdev_close() itself,
2516 * because that implies some extra checks (offline, etc) that we don't
2517 * want here. This is limited to leaf devices, because otherwise
2518 * closing the device will affect other children.
2520 if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf)
2521 vd->vdev_ops->vdev_op_close(vd);
2523 if (vd->vdev_removed &&
2524 state == VDEV_STATE_CANT_OPEN &&
2525 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2527 * If the previous state is set to VDEV_STATE_REMOVED, then this
2528 * device was previously marked removed and someone attempted to
2529 * reopen it. If this failed due to a nonexistent device, then
2530 * keep the device in the REMOVED state. We also let this be if
2531 * it is one of our special test online cases, which is only
2532 * attempting to online the device and shouldn't generate an FMA
2535 vd->vdev_state = VDEV_STATE_REMOVED;
2536 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2537 } else if (state == VDEV_STATE_REMOVED) {
2539 * Indicate to the ZFS DE that this device has been removed, and
2540 * any recent errors should be ignored.
2542 zfs_post_remove(spa, vd);
2543 vd->vdev_removed = B_TRUE;
2544 } else if (state == VDEV_STATE_CANT_OPEN) {
2546 * If we fail to open a vdev during an import, we mark it as
2547 * "not available", which signifies that it was never there to
2548 * begin with. Failure to open such a device is not considered
2551 if (spa->spa_load_state == SPA_LOAD_IMPORT &&
2552 vd->vdev_ops->vdev_op_leaf)
2553 vd->vdev_not_present = 1;
2556 * Post the appropriate ereport. If the 'prevstate' field is
2557 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2558 * that this is part of a vdev_reopen(). In this case, we don't
2559 * want to post the ereport if the device was already in the
2560 * CANT_OPEN state beforehand.
2562 * If the 'checkremove' flag is set, then this is an attempt to
2563 * online the device in response to an insertion event. If we
2564 * hit this case, then we have detected an insertion event for a
2565 * faulted or offline device that wasn't in the removed state.
2566 * In this scenario, we don't post an ereport because we are
2567 * about to replace the device, or attempt an online with
2568 * vdev_forcefault, which will generate the fault for us.
2570 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2571 !vd->vdev_not_present && !vd->vdev_checkremove &&
2572 vd != spa->spa_root_vdev) {
2576 case VDEV_AUX_OPEN_FAILED:
2577 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
2579 case VDEV_AUX_CORRUPT_DATA:
2580 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
2582 case VDEV_AUX_NO_REPLICAS:
2583 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
2585 case VDEV_AUX_BAD_GUID_SUM:
2586 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
2588 case VDEV_AUX_TOO_SMALL:
2589 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
2591 case VDEV_AUX_BAD_LABEL:
2592 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
2594 case VDEV_AUX_IO_FAILURE:
2595 class = FM_EREPORT_ZFS_IO_FAILURE;
2598 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
2601 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
2604 /* Erase any notion of persistent removed state */
2605 vd->vdev_removed = B_FALSE;
2607 vd->vdev_removed = B_FALSE;
2610 if (!isopen && vd->vdev_parent)
2611 vdev_propagate_state(vd->vdev_parent);
2615 * Check the vdev configuration to ensure that it's capable of supporting
2618 * On Solaris, we do not support RAID-Z or partial configuration. In
2619 * addition, only a single top-level vdev is allowed and none of the
2620 * leaves can be wholedisks.
2622 * For FreeBSD, we can boot from any configuration. There is a
2623 * limitation that the boot filesystem must be either uncompressed or
2624 * compresses with lzjb compression but I'm not sure how to enforce
2628 vdev_is_bootable(vdev_t *vd)
2630 #ifdef __FreeBSD_version
2635 if (!vd->vdev_ops->vdev_op_leaf) {
2636 char *vdev_type = vd->vdev_ops->vdev_op_type;
2638 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
2639 vd->vdev_children > 1) {
2641 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
2642 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
2645 } else if (vd->vdev_wholedisk == 1) {
2649 for (c = 0; c < vd->vdev_children; c++) {
2650 if (!vdev_is_bootable(vd->vdev_child[c]))