vdev.c revision 288569
1169695Skan/*
2169695Skan * CDDL HEADER START
3169695Skan *
4169695Skan * The contents of this file are subject to the terms of the
5169695Skan * Common Development and Distribution License (the "License").
6169695Skan * You may not use this file except in compliance with the License.
7169695Skan *
8169695Skan * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9169695Skan * or http://www.opensolaris.org/os/licensing.
10169695Skan * See the License for the specific language governing permissions
11169695Skan * and limitations under the License.
12169695Skan *
13169695Skan * When distributing Covered Code, include this CDDL HEADER in each
14169695Skan * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15169695Skan * If applicable, add the following below this CDDL HEADER, with the
16169695Skan * fields enclosed by brackets "[]" replaced with your own identifying
17169695Skan * information: Portions Copyright [yyyy] [name of copyright owner]
18169695Skan *
19169695Skan * CDDL HEADER END
20169695Skan */
21169695Skan
22169695Skan/*
23169695Skan * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24169695Skan * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25169695Skan * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
26169695Skan * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27169695Skan */
28169695Skan
29169695Skan#include <sys/zfs_context.h>
30169695Skan#include <sys/fm/fs/zfs.h>
31169695Skan#include <sys/spa.h>
32169695Skan#include <sys/spa_impl.h>
33169695Skan#include <sys/dmu.h>
34169695Skan#include <sys/dmu_tx.h>
35169695Skan#include <sys/vdev_impl.h>
36169695Skan#include <sys/uberblock_impl.h>
37169695Skan#include <sys/metaslab.h>
38169695Skan#include <sys/metaslab_impl.h>
39169695Skan#include <sys/space_map.h>
40169695Skan#include <sys/space_reftree.h>
41169695Skan#include <sys/zio.h>
42169695Skan#include <sys/zap.h>
43169695Skan#include <sys/fs/zfs.h>
44169695Skan#include <sys/arc.h>
45169695Skan#include <sys/zil.h>
46169695Skan#include <sys/dsl_scan.h>
47169695Skan#include <sys/trim_map.h>
48169695Skan
49169695SkanSYSCTL_DECL(_vfs_zfs);
50169695SkanSYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
51169695Skan
52169695Skan/*
53169695Skan * Virtual device management.
54169695Skan */
55169695Skan
56169695Skan/*
57169695Skan * The limit for ZFS to automatically increase a top-level vdev's ashift
58169695Skan * from logical ashift to physical ashift.
59169695Skan *
60169695Skan * Example: one or more 512B emulation child vdevs
61169695Skan *          child->vdev_ashift = 9 (512 bytes)
62169695Skan *          child->vdev_physical_ashift = 12 (4096 bytes)
63169695Skan *          zfs_max_auto_ashift = 11 (2048 bytes)
64169695Skan *          zfs_min_auto_ashift = 9 (512 bytes)
65169695Skan *
66169695Skan * On pool creation or the addition of a new top-level vdev, ZFS will
67169695Skan * increase the ashift of the top-level vdev to 2048 as limited by
68169695Skan * zfs_max_auto_ashift.
69169695Skan *
70169695Skan * Example: one or more 512B emulation child vdevs
71169695Skan *          child->vdev_ashift = 9 (512 bytes)
72169695Skan *          child->vdev_physical_ashift = 12 (4096 bytes)
73169695Skan *          zfs_max_auto_ashift = 13 (8192 bytes)
74169695Skan *          zfs_min_auto_ashift = 9 (512 bytes)
75169695Skan *
76169695Skan * On pool creation or the addition of a new top-level vdev, ZFS will
77169695Skan * increase the ashift of the top-level vdev to 4096 to match the
78169695Skan * max vdev_physical_ashift.
79169695Skan *
80169695Skan * Example: one or more 512B emulation child vdevs
81169695Skan *          child->vdev_ashift = 9 (512 bytes)
82169695Skan *          child->vdev_physical_ashift = 9 (512 bytes)
83169695Skan *          zfs_max_auto_ashift = 13 (8192 bytes)
84169695Skan *          zfs_min_auto_ashift = 12 (4096 bytes)
85169695Skan *
86169695Skan * On pool creation or the addition of a new top-level vdev, ZFS will
87169695Skan * increase the ashift of the top-level vdev to 4096 to match the
88169695Skan * zfs_min_auto_ashift.
89169695Skan */
90169695Skanstatic uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
91169695Skanstatic uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
92169695Skan
93169695Skanstatic int
94169695Skansysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
95169695Skan{
96169695Skan	uint64_t val;
97169695Skan	int err;
98169695Skan
99169695Skan	val = zfs_max_auto_ashift;
100169695Skan	err = sysctl_handle_64(oidp, &val, 0, req);
101169695Skan	if (err != 0 || req->newptr == NULL)
102169695Skan		return (err);
103169695Skan
104169695Skan	if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
105169695Skan		return (EINVAL);
106169695Skan
107169695Skan	zfs_max_auto_ashift = val;
108169695Skan
109169695Skan	return (0);
110169695Skan}
111169695SkanSYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
112169695Skan    CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
113169695Skan    sysctl_vfs_zfs_max_auto_ashift, "QU",
114169695Skan    "Max ashift used when optimising for logical -> physical sectors size on "
115169695Skan    "new top-level vdevs.");
116169695Skan
117169695Skanstatic int
118169695Skansysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
119169695Skan{
120169695Skan	uint64_t val;
121169695Skan	int err;
122169695Skan
123169695Skan	val = zfs_min_auto_ashift;
124169695Skan	err = sysctl_handle_64(oidp, &val, 0, req);
125169695Skan	if (err != 0 || req->newptr == NULL)
126169695Skan		return (err);
127169695Skan
128169695Skan	if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
129169695Skan		return (EINVAL);
130169695Skan
131169695Skan	zfs_min_auto_ashift = val;
132169695Skan
133169695Skan	return (0);
134169695Skan}
135169695SkanSYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
136169695Skan    CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
137169695Skan    sysctl_vfs_zfs_min_auto_ashift, "QU",
138169695Skan    "Min ashift used when creating new top-level vdevs.");
139169695Skan
140169695Skanstatic vdev_ops_t *vdev_ops_table[] = {
141169695Skan	&vdev_root_ops,
142169695Skan	&vdev_raidz_ops,
143169695Skan	&vdev_mirror_ops,
144169695Skan	&vdev_replacing_ops,
145169695Skan	&vdev_spare_ops,
146169695Skan#ifdef _KERNEL
147169695Skan	&vdev_geom_ops,
148169695Skan#else
149169695Skan	&vdev_disk_ops,
150169695Skan#endif
151169695Skan	&vdev_file_ops,
152169695Skan	&vdev_missing_ops,
153169695Skan	&vdev_hole_ops,
154169695Skan	NULL
155169695Skan};
156169695Skan
157169695Skan
158169695Skan/*
159169695Skan * When a vdev is added, it will be divided into approximately (but no
160169695Skan * more than) this number of metaslabs.
161169695Skan */
162169695Skanint metaslabs_per_vdev = 200;
163169695SkanSYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
164169695Skan    &metaslabs_per_vdev, 0,
165169695Skan    "When a vdev is added, how many metaslabs the vdev should be divided into");
166169695Skan
167169695Skan/*
168169695Skan * Given a vdev type, return the appropriate ops vector.
169169695Skan */
170169695Skanstatic vdev_ops_t *
171169695Skanvdev_getops(const char *type)
172169695Skan{
173169695Skan	vdev_ops_t *ops, **opspp;
174169695Skan
175169695Skan	for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
176169695Skan		if (strcmp(ops->vdev_op_type, type) == 0)
177169695Skan			break;
178169695Skan
179169695Skan	return (ops);
180169695Skan}
181169695Skan
182169695Skan/*
183169695Skan * Default asize function: return the MAX of psize with the asize of
184169695Skan * all children.  This is what's used by anything other than RAID-Z.
185169695Skan */
186169695Skanuint64_t
187169695Skanvdev_default_asize(vdev_t *vd, uint64_t psize)
188169695Skan{
189169695Skan	uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
190169695Skan	uint64_t csize;
191169695Skan
192169695Skan	for (int c = 0; c < vd->vdev_children; c++) {
193169695Skan		csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
194169695Skan		asize = MAX(asize, csize);
195169695Skan	}
196169695Skan
197169695Skan	return (asize);
198169695Skan}
199169695Skan
200169695Skan/*
201169695Skan * Get the minimum allocatable size. We define the allocatable size as
202169695Skan * the vdev's asize rounded to the nearest metaslab. This allows us to
203169695Skan * replace or attach devices which don't have the same physical size but
204169695Skan * can still satisfy the same number of allocations.
205169695Skan */
206169695Skanuint64_t
207169695Skanvdev_get_min_asize(vdev_t *vd)
208169695Skan{
209169695Skan	vdev_t *pvd = vd->vdev_parent;
210169695Skan
211169695Skan	/*
212169695Skan	 * If our parent is NULL (inactive spare or cache) or is the root,
213169695Skan	 * just return our own asize.
214169695Skan	 */
215169695Skan	if (pvd == NULL)
216169695Skan		return (vd->vdev_asize);
217169695Skan
218169695Skan	/*
219169695Skan	 * The top-level vdev just returns the allocatable size rounded
220169695Skan	 * to the nearest metaslab.
221169695Skan	 */
222169695Skan	if (vd == vd->vdev_top)
223169695Skan		return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
224169695Skan
225169695Skan	/*
226169695Skan	 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
227169695Skan	 * so each child must provide at least 1/Nth of its asize.
228169695Skan	 */
229169695Skan	if (pvd->vdev_ops == &vdev_raidz_ops)
230169695Skan		return (pvd->vdev_min_asize / pvd->vdev_children);
231169695Skan
232169695Skan	return (pvd->vdev_min_asize);
233169695Skan}
234169695Skan
235169695Skanvoid
236169695Skanvdev_set_min_asize(vdev_t *vd)
237169695Skan{
238169695Skan	vd->vdev_min_asize = vdev_get_min_asize(vd);
239169695Skan
240169695Skan	for (int c = 0; c < vd->vdev_children; c++)
241169695Skan		vdev_set_min_asize(vd->vdev_child[c]);
242169695Skan}
243169695Skan
244283010Spfgvdev_t *
245283010Spfgvdev_lookup_top(spa_t *spa, uint64_t vdev)
246283010Spfg{
247283010Spfg	vdev_t *rvd = spa->spa_root_vdev;
248283010Spfg
249283010Spfg	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
250283010Spfg
251283010Spfg	if (vdev < rvd->vdev_children) {
252283010Spfg		ASSERT(rvd->vdev_child[vdev] != NULL);
253283010Spfg		return (rvd->vdev_child[vdev]);
254283010Spfg	}
255283010Spfg
256283010Spfg	return (NULL);
257283010Spfg}
258283010Spfg
259283010Spfgvdev_t *
260283010Spfgvdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
261283010Spfg{
262283010Spfg	vdev_t *mvd;
263283010Spfg
264283010Spfg	if (vd->vdev_guid == guid)
265283010Spfg		return (vd);
266283010Spfg
267283010Spfg	for (int c = 0; c < vd->vdev_children; c++)
268283010Spfg		if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
269283010Spfg		    NULL)
270283010Spfg			return (mvd);
271283010Spfg
272283010Spfg	return (NULL);
273283010Spfg}
274283010Spfg
275283010Spfgstatic int
276283010Spfgvdev_count_leaves_impl(vdev_t *vd)
277283010Spfg{
278283010Spfg	int n = 0;
279283010Spfg
280283010Spfg	if (vd->vdev_ops->vdev_op_leaf)
281283010Spfg		return (1);
282283010Spfg
283283010Spfg	for (int c = 0; c < vd->vdev_children; c++)
284283010Spfg		n += vdev_count_leaves_impl(vd->vdev_child[c]);
285283010Spfg
286283010Spfg	return (n);
287283010Spfg}
288283010Spfg
289283010Spfgint
290283010Spfgvdev_count_leaves(spa_t *spa)
291283010Spfg{
292283010Spfg	return (vdev_count_leaves_impl(spa->spa_root_vdev));
293283010Spfg}
294283010Spfg
295283010Spfgvoid
296283010Spfgvdev_add_child(vdev_t *pvd, vdev_t *cvd)
297283010Spfg{
298283010Spfg	size_t oldsize, newsize;
299283010Spfg	uint64_t id = cvd->vdev_id;
300283010Spfg	vdev_t **newchild;
301283010Spfg	spa_t *spa = cvd->vdev_spa;
302283010Spfg
303283010Spfg	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
304283010Spfg	ASSERT(cvd->vdev_parent == NULL);
305283010Spfg
306283010Spfg	cvd->vdev_parent = pvd;
307283010Spfg
308283010Spfg	if (pvd == NULL)
309283010Spfg		return;
310283010Spfg
311283010Spfg	ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
312283010Spfg
313283010Spfg	oldsize = pvd->vdev_children * sizeof (vdev_t *);
314283010Spfg	pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
315283010Spfg	newsize = pvd->vdev_children * sizeof (vdev_t *);
316283010Spfg
317283010Spfg	newchild = kmem_zalloc(newsize, KM_SLEEP);
318283010Spfg	if (pvd->vdev_child != NULL) {
319283010Spfg		bcopy(pvd->vdev_child, newchild, oldsize);
320283010Spfg		kmem_free(pvd->vdev_child, oldsize);
321283010Spfg	}
322283010Spfg
323283010Spfg	pvd->vdev_child = newchild;
324283010Spfg	pvd->vdev_child[id] = cvd;
325283010Spfg
326283010Spfg	cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
327283010Spfg	ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
328283010Spfg
329283010Spfg	/*
330283010Spfg	 * Walk up all ancestors to update guid sum.
331283010Spfg	 */
332283010Spfg	for (; pvd != NULL; pvd = pvd->vdev_parent)
333283010Spfg		pvd->vdev_guid_sum += cvd->vdev_guid_sum;
334283010Spfg}
335283010Spfg
336283010Spfgvoid
337283010Spfgvdev_remove_child(vdev_t *pvd, vdev_t *cvd)
338283010Spfg{
339283010Spfg	int c;
340283010Spfg	uint_t id = cvd->vdev_id;
341283010Spfg
342283010Spfg	ASSERT(cvd->vdev_parent == pvd);
343283010Spfg
344283010Spfg	if (pvd == NULL)
345283010Spfg		return;
346283010Spfg
347283010Spfg	ASSERT(id < pvd->vdev_children);
348283010Spfg	ASSERT(pvd->vdev_child[id] == cvd);
349283010Spfg
350283010Spfg	pvd->vdev_child[id] = NULL;
351283010Spfg	cvd->vdev_parent = NULL;
352283010Spfg
353283010Spfg	for (c = 0; c < pvd->vdev_children; c++)
354283010Spfg		if (pvd->vdev_child[c])
355283010Spfg			break;
356283010Spfg
357283010Spfg	if (c == pvd->vdev_children) {
358283010Spfg		kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
359283010Spfg		pvd->vdev_child = NULL;
360283010Spfg		pvd->vdev_children = 0;
361283010Spfg	}
362283010Spfg
363283010Spfg	/*
364283010Spfg	 * Walk up all ancestors to update guid sum.
365283010Spfg	 */
366283010Spfg	for (; pvd != NULL; pvd = pvd->vdev_parent)
367283010Spfg		pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
368283010Spfg}
369283010Spfg
370283010Spfg/*
371283010Spfg * Remove any holes in the child array.
372283010Spfg */
373283010Spfgvoid
374283010Spfgvdev_compact_children(vdev_t *pvd)
375283010Spfg{
376283010Spfg	vdev_t **newchild, *cvd;
377283010Spfg	int oldc = pvd->vdev_children;
378283010Spfg	int newc;
379283010Spfg
380283010Spfg	ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
381283010Spfg
382283010Spfg	for (int c = newc = 0; c < oldc; c++)
383283010Spfg		if (pvd->vdev_child[c])
384283010Spfg			newc++;
385283010Spfg
386283010Spfg	newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
387283010Spfg
388283010Spfg	for (int c = newc = 0; c < oldc; c++) {
389283010Spfg		if ((cvd = pvd->vdev_child[c]) != NULL) {
390283010Spfg			newchild[newc] = cvd;
391283010Spfg			cvd->vdev_id = newc++;
392283010Spfg		}
393169695Skan	}
394169695Skan
395169695Skan	kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
396169695Skan	pvd->vdev_child = newchild;
397169695Skan	pvd->vdev_children = newc;
398169695Skan}
399169695Skan
400169695Skan/*
401169695Skan * Allocate and minimally initialize a vdev_t.
402169695Skan */
403169695Skanvdev_t *
404169695Skanvdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
405169695Skan{
406169695Skan	vdev_t *vd;
407169695Skan
408169695Skan	vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
409169695Skan
410169695Skan	if (spa->spa_root_vdev == NULL) {
411169695Skan		ASSERT(ops == &vdev_root_ops);
412169695Skan		spa->spa_root_vdev = vd;
413169695Skan		spa->spa_load_guid = spa_generate_guid(NULL);
414169695Skan	}
415169695Skan
416169695Skan	if (guid == 0 && ops != &vdev_hole_ops) {
417169695Skan		if (spa->spa_root_vdev == vd) {
418169695Skan			/*
419169695Skan			 * The root vdev's guid will also be the pool guid,
420169695Skan			 * which must be unique among all pools.
421169695Skan			 */
422169695Skan			guid = spa_generate_guid(NULL);
423169695Skan		} else {
424169695Skan			/*
425169695Skan			 * Any other vdev's guid must be unique within the pool.
426169695Skan			 */
427169695Skan			guid = spa_generate_guid(spa);
428169695Skan		}
429169695Skan		ASSERT(!spa_guid_exists(spa_guid(spa), guid));
430169695Skan	}
431169695Skan
432169695Skan	vd->vdev_spa = spa;
433169695Skan	vd->vdev_id = id;
434169695Skan	vd->vdev_guid = guid;
435169695Skan	vd->vdev_guid_sum = guid;
436169695Skan	vd->vdev_ops = ops;
437169695Skan	vd->vdev_state = VDEV_STATE_CLOSED;
438169695Skan	vd->vdev_ishole = (ops == &vdev_hole_ops);
439169695Skan
440169695Skan	mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
441169695Skan	mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
442169695Skan	mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
443169695Skan	for (int t = 0; t < DTL_TYPES; t++) {
444169695Skan		vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
445169695Skan		    &vd->vdev_dtl_lock);
446169695Skan	}
447169695Skan	txg_list_create(&vd->vdev_ms_list,
448169695Skan	    offsetof(struct metaslab, ms_txg_node));
449169695Skan	txg_list_create(&vd->vdev_dtl_list,
450169695Skan	    offsetof(struct vdev, vdev_dtl_node));
451169695Skan	vd->vdev_stat.vs_timestamp = gethrtime();
452169695Skan	vdev_queue_init(vd);
453169695Skan	vdev_cache_init(vd);
454169695Skan
455169695Skan	return (vd);
456169695Skan}
457169695Skan
458169695Skan/*
459169695Skan * Allocate a new vdev.  The 'alloctype' is used to control whether we are
460283010Spfg * creating a new vdev or loading an existing one - the behavior is slightly
461169695Skan * different for each case.
462169695Skan */
463169695Skanint
464169695Skanvdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
465169695Skan    int alloctype)
466169695Skan{
467169695Skan	vdev_ops_t *ops;
468169695Skan	char *type;
469169695Skan	uint64_t guid = 0, islog, nparity;
470169695Skan	vdev_t *vd;
471169695Skan
472169695Skan	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
473169695Skan
474169695Skan	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
475169695Skan		return (SET_ERROR(EINVAL));
476169695Skan
477169695Skan	if ((ops = vdev_getops(type)) == NULL)
478169695Skan		return (SET_ERROR(EINVAL));
479169695Skan
480169695Skan	/*
481169695Skan	 * If this is a load, get the vdev guid from the nvlist.
482169695Skan	 * Otherwise, vdev_alloc_common() will generate one for us.
483169695Skan	 */
484169695Skan	if (alloctype == VDEV_ALLOC_LOAD) {
485169695Skan		uint64_t label_id;
486169695Skan
487169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
488169695Skan		    label_id != id)
489169695Skan			return (SET_ERROR(EINVAL));
490169695Skan
491169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
492169695Skan			return (SET_ERROR(EINVAL));
493169695Skan	} else if (alloctype == VDEV_ALLOC_SPARE) {
494169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
495169695Skan			return (SET_ERROR(EINVAL));
496169695Skan	} else if (alloctype == VDEV_ALLOC_L2CACHE) {
497169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
498169695Skan			return (SET_ERROR(EINVAL));
499169695Skan	} else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
500169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
501169695Skan			return (SET_ERROR(EINVAL));
502169695Skan	}
503169695Skan
504169695Skan	/*
505169695Skan	 * The first allocated vdev must be of type 'root'.
506169695Skan	 */
507169695Skan	if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
508169695Skan		return (SET_ERROR(EINVAL));
509169695Skan
510169695Skan	/*
511169695Skan	 * Determine whether we're a log vdev.
512169695Skan	 */
513169695Skan	islog = 0;
514169695Skan	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
515169695Skan	if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
516169695Skan		return (SET_ERROR(ENOTSUP));
517169695Skan
518169695Skan	if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
519169695Skan		return (SET_ERROR(ENOTSUP));
520169695Skan
521169695Skan	/*
522169695Skan	 * Set the nparity property for RAID-Z vdevs.
523169695Skan	 */
524169695Skan	nparity = -1ULL;
525169695Skan	if (ops == &vdev_raidz_ops) {
526169695Skan		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
527169695Skan		    &nparity) == 0) {
528169695Skan			if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
529169695Skan				return (SET_ERROR(EINVAL));
530169695Skan			/*
531169695Skan			 * Previous versions could only support 1 or 2 parity
532169695Skan			 * device.
533169695Skan			 */
534169695Skan			if (nparity > 1 &&
535169695Skan			    spa_version(spa) < SPA_VERSION_RAIDZ2)
536169695Skan				return (SET_ERROR(ENOTSUP));
537169695Skan			if (nparity > 2 &&
538169695Skan			    spa_version(spa) < SPA_VERSION_RAIDZ3)
539169695Skan				return (SET_ERROR(ENOTSUP));
540169695Skan		} else {
541169695Skan			/*
542169695Skan			 * We require the parity to be specified for SPAs that
543169695Skan			 * support multiple parity levels.
544169695Skan			 */
545169695Skan			if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
546169695Skan				return (SET_ERROR(EINVAL));
547169695Skan			/*
548169695Skan			 * Otherwise, we default to 1 parity device for RAID-Z.
549169695Skan			 */
550169695Skan			nparity = 1;
551169695Skan		}
552169695Skan	} else {
553169695Skan		nparity = 0;
554169695Skan	}
555169695Skan	ASSERT(nparity != -1ULL);
556169695Skan
557169695Skan	vd = vdev_alloc_common(spa, id, guid, ops);
558169695Skan
559169695Skan	vd->vdev_islog = islog;
560169695Skan	vd->vdev_nparity = nparity;
561169695Skan
562169695Skan	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
563169695Skan		vd->vdev_path = spa_strdup(vd->vdev_path);
564169695Skan	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
565169695Skan		vd->vdev_devid = spa_strdup(vd->vdev_devid);
566169695Skan	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
567169695Skan	    &vd->vdev_physpath) == 0)
568169695Skan		vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
569169695Skan	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
570169695Skan		vd->vdev_fru = spa_strdup(vd->vdev_fru);
571169695Skan
572169695Skan	/*
573169695Skan	 * Set the whole_disk property.  If it's not specified, leave the value
574169695Skan	 * as -1.
575169695Skan	 */
576169695Skan	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
577169695Skan	    &vd->vdev_wholedisk) != 0)
578169695Skan		vd->vdev_wholedisk = -1ULL;
579169695Skan
580169695Skan	/*
581169695Skan	 * Look for the 'not present' flag.  This will only be set if the device
582169695Skan	 * was not present at the time of import.
583169695Skan	 */
584169695Skan	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
585169695Skan	    &vd->vdev_not_present);
586169695Skan
587169695Skan	/*
588169695Skan	 * Get the alignment requirement.
589169695Skan	 */
590169695Skan	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
591169695Skan
592169695Skan	/*
593169695Skan	 * Retrieve the vdev creation time.
594169695Skan	 */
595169695Skan	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
596169695Skan	    &vd->vdev_crtxg);
597169695Skan
598169695Skan	/*
599169695Skan	 * If we're a top-level vdev, try to load the allocation parameters.
600169695Skan	 */
601169695Skan	if (parent && !parent->vdev_parent &&
602169695Skan	    (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
603169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
604169695Skan		    &vd->vdev_ms_array);
605169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
606169695Skan		    &vd->vdev_ms_shift);
607169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
608169695Skan		    &vd->vdev_asize);
609169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
610169695Skan		    &vd->vdev_removing);
611169695Skan	}
612169695Skan
613169695Skan	if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
614169695Skan		ASSERT(alloctype == VDEV_ALLOC_LOAD ||
615169695Skan		    alloctype == VDEV_ALLOC_ADD ||
616169695Skan		    alloctype == VDEV_ALLOC_SPLIT ||
617169695Skan		    alloctype == VDEV_ALLOC_ROOTPOOL);
618169695Skan		vd->vdev_mg = metaslab_group_create(islog ?
619169695Skan		    spa_log_class(spa) : spa_normal_class(spa), vd);
620169695Skan	}
621169695Skan
622169695Skan	/*
623169695Skan	 * If we're a leaf vdev, try to load the DTL object and other state.
624169695Skan	 */
625169695Skan	if (vd->vdev_ops->vdev_op_leaf &&
626169695Skan	    (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
627169695Skan	    alloctype == VDEV_ALLOC_ROOTPOOL)) {
628169695Skan		if (alloctype == VDEV_ALLOC_LOAD) {
629169695Skan			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
630169695Skan			    &vd->vdev_dtl_object);
631169695Skan			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
632169695Skan			    &vd->vdev_unspare);
633169695Skan		}
634169695Skan
635169695Skan		if (alloctype == VDEV_ALLOC_ROOTPOOL) {
636169695Skan			uint64_t spare = 0;
637169695Skan
638169695Skan			if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
639169695Skan			    &spare) == 0 && spare)
640169695Skan				spa_spare_add(vd);
641169695Skan		}
642169695Skan
643169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
644169695Skan		    &vd->vdev_offline);
645169695Skan
646169695Skan		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
647169695Skan		    &vd->vdev_resilver_txg);
648169695Skan
649169695Skan		/*
650169695Skan		 * When importing a pool, we want to ignore the persistent fault
651169695Skan		 * state, as the diagnosis made on another system may not be
652169695Skan		 * valid in the current context.  Local vdevs will
653169695Skan		 * remain in the faulted state.
654169695Skan		 */
655169695Skan		if (spa_load_state(spa) == SPA_LOAD_OPEN) {
656169695Skan			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
657169695Skan			    &vd->vdev_faulted);
658169695Skan			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
659169695Skan			    &vd->vdev_degraded);
660169695Skan			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
661169695Skan			    &vd->vdev_removed);
662169695Skan
663169695Skan			if (vd->vdev_faulted || vd->vdev_degraded) {
664169695Skan				char *aux;
665169695Skan
666169695Skan				vd->vdev_label_aux =
667169695Skan				    VDEV_AUX_ERR_EXCEEDED;
668169695Skan				if (nvlist_lookup_string(nv,
669169695Skan				    ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
670169695Skan				    strcmp(aux, "external") == 0)
671169695Skan					vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
672169695Skan			}
673169695Skan		}
674169695Skan	}
675169695Skan
676169695Skan	/*
677169695Skan	 * Add ourselves to the parent's list of children.
678169695Skan	 */
679169695Skan	vdev_add_child(parent, vd);
680169695Skan
681169695Skan	*vdp = vd;
682169695Skan
683169695Skan	return (0);
684169695Skan}
685169695Skan
686169695Skanvoid
687169695Skanvdev_free(vdev_t *vd)
688169695Skan{
689169695Skan	spa_t *spa = vd->vdev_spa;
690169695Skan
691169695Skan	/*
692169695Skan	 * vdev_free() implies closing the vdev first.  This is simpler than
693169695Skan	 * trying to ensure complicated semantics for all callers.
694169695Skan	 */
695169695Skan	vdev_close(vd);
696169695Skan
697169695Skan	ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
698169695Skan	ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
699169695Skan
700169695Skan	/*
701169695Skan	 * Free all children.
702169695Skan	 */
703169695Skan	for (int c = 0; c < vd->vdev_children; c++)
704169695Skan		vdev_free(vd->vdev_child[c]);
705169695Skan
706169695Skan	ASSERT(vd->vdev_child == NULL);
707169695Skan	ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
708169695Skan
709169695Skan	/*
710169695Skan	 * Discard allocation state.
711169695Skan	 */
712169695Skan	if (vd->vdev_mg != NULL) {
713169695Skan		vdev_metaslab_fini(vd);
714169695Skan		metaslab_group_destroy(vd->vdev_mg);
715169695Skan	}
716169695Skan
717169695Skan	ASSERT0(vd->vdev_stat.vs_space);
718169695Skan	ASSERT0(vd->vdev_stat.vs_dspace);
719169695Skan	ASSERT0(vd->vdev_stat.vs_alloc);
720169695Skan
721169695Skan	/*
722169695Skan	 * Remove this vdev from its parent's child list.
723169695Skan	 */
724169695Skan	vdev_remove_child(vd->vdev_parent, vd);
725169695Skan
726169695Skan	ASSERT(vd->vdev_parent == NULL);
727169695Skan
728169695Skan	/*
729169695Skan	 * Clean up vdev structure.
730169695Skan	 */
731169695Skan	vdev_queue_fini(vd);
732169695Skan	vdev_cache_fini(vd);
733169695Skan
734169695Skan	if (vd->vdev_path)
735169695Skan		spa_strfree(vd->vdev_path);
736169695Skan	if (vd->vdev_devid)
737169695Skan		spa_strfree(vd->vdev_devid);
738169695Skan	if (vd->vdev_physpath)
739169695Skan		spa_strfree(vd->vdev_physpath);
740169695Skan	if (vd->vdev_fru)
741169695Skan		spa_strfree(vd->vdev_fru);
742169695Skan
743169695Skan	if (vd->vdev_isspare)
744169695Skan		spa_spare_remove(vd);
745169695Skan	if (vd->vdev_isl2cache)
746169695Skan		spa_l2cache_remove(vd);
747169695Skan
748169695Skan	txg_list_destroy(&vd->vdev_ms_list);
749169695Skan	txg_list_destroy(&vd->vdev_dtl_list);
750169695Skan
751169695Skan	mutex_enter(&vd->vdev_dtl_lock);
752169695Skan	space_map_close(vd->vdev_dtl_sm);
753169695Skan	for (int t = 0; t < DTL_TYPES; t++) {
754169695Skan		range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
755169695Skan		range_tree_destroy(vd->vdev_dtl[t]);
756169695Skan	}
757169695Skan	mutex_exit(&vd->vdev_dtl_lock);
758169695Skan
759169695Skan	mutex_destroy(&vd->vdev_dtl_lock);
760169695Skan	mutex_destroy(&vd->vdev_stat_lock);
761169695Skan	mutex_destroy(&vd->vdev_probe_lock);
762169695Skan
763169695Skan	if (vd == spa->spa_root_vdev)
764169695Skan		spa->spa_root_vdev = NULL;
765169695Skan
766169695Skan	kmem_free(vd, sizeof (vdev_t));
767169695Skan}
768169695Skan
769169695Skan/*
770169695Skan * Transfer top-level vdev state from svd to tvd.
771169695Skan */
772169695Skanstatic void
773169695Skanvdev_top_transfer(vdev_t *svd, vdev_t *tvd)
774169695Skan{
775169695Skan	spa_t *spa = svd->vdev_spa;
776169695Skan	metaslab_t *msp;
777169695Skan	vdev_t *vd;
778169695Skan	int t;
779169695Skan
780169695Skan	ASSERT(tvd == tvd->vdev_top);
781169695Skan
782169695Skan	tvd->vdev_ms_array = svd->vdev_ms_array;
783169695Skan	tvd->vdev_ms_shift = svd->vdev_ms_shift;
784169695Skan	tvd->vdev_ms_count = svd->vdev_ms_count;
785169695Skan
786169695Skan	svd->vdev_ms_array = 0;
787169695Skan	svd->vdev_ms_shift = 0;
788169695Skan	svd->vdev_ms_count = 0;
789169695Skan
790169695Skan	if (tvd->vdev_mg)
791169695Skan		ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
792169695Skan	tvd->vdev_mg = svd->vdev_mg;
793169695Skan	tvd->vdev_ms = svd->vdev_ms;
794169695Skan
795169695Skan	svd->vdev_mg = NULL;
796169695Skan	svd->vdev_ms = NULL;
797169695Skan
798169695Skan	if (tvd->vdev_mg != NULL)
799169695Skan		tvd->vdev_mg->mg_vd = tvd;
800169695Skan
801169695Skan	tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
802169695Skan	tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
803169695Skan	tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
804169695Skan
805169695Skan	svd->vdev_stat.vs_alloc = 0;
806169695Skan	svd->vdev_stat.vs_space = 0;
807169695Skan	svd->vdev_stat.vs_dspace = 0;
808169695Skan
809169695Skan	for (t = 0; t < TXG_SIZE; t++) {
810169695Skan		while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
811169695Skan			(void) txg_list_add(&tvd->vdev_ms_list, msp, t);
812169695Skan		while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
813169695Skan			(void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
814169695Skan		if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
815169695Skan			(void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
816169695Skan	}
817169695Skan
818169695Skan	if (list_link_active(&svd->vdev_config_dirty_node)) {
819169695Skan		vdev_config_clean(svd);
820169695Skan		vdev_config_dirty(tvd);
821169695Skan	}
822169695Skan
823169695Skan	if (list_link_active(&svd->vdev_state_dirty_node)) {
824169695Skan		vdev_state_clean(svd);
825169695Skan		vdev_state_dirty(tvd);
826169695Skan	}
827169695Skan
828169695Skan	tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
829169695Skan	svd->vdev_deflate_ratio = 0;
830169695Skan
831169695Skan	tvd->vdev_islog = svd->vdev_islog;
832169695Skan	svd->vdev_islog = 0;
833169695Skan}
834169695Skan
835169695Skanstatic void
836169695Skanvdev_top_update(vdev_t *tvd, vdev_t *vd)
837169695Skan{
838169695Skan	if (vd == NULL)
839169695Skan		return;
840169695Skan
841169695Skan	vd->vdev_top = tvd;
842169695Skan
843169695Skan	for (int c = 0; c < vd->vdev_children; c++)
844169695Skan		vdev_top_update(tvd, vd->vdev_child[c]);
845169695Skan}
846169695Skan
847169695Skan/*
848169695Skan * Add a mirror/replacing vdev above an existing vdev.
849169695Skan */
850169695Skanvdev_t *
851169695Skanvdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
852169695Skan{
853169695Skan	spa_t *spa = cvd->vdev_spa;
854169695Skan	vdev_t *pvd = cvd->vdev_parent;
855169695Skan	vdev_t *mvd;
856169695Skan
857169695Skan	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
858169695Skan
859169695Skan	mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
860169695Skan
861169695Skan	mvd->vdev_asize = cvd->vdev_asize;
862169695Skan	mvd->vdev_min_asize = cvd->vdev_min_asize;
863169695Skan	mvd->vdev_max_asize = cvd->vdev_max_asize;
864169695Skan	mvd->vdev_ashift = cvd->vdev_ashift;
865169695Skan	mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
866169695Skan	mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
867169695Skan	mvd->vdev_state = cvd->vdev_state;
868169695Skan	mvd->vdev_crtxg = cvd->vdev_crtxg;
869169695Skan
870283010Spfg	vdev_remove_child(pvd, cvd);
871169695Skan	vdev_add_child(pvd, mvd);
872169695Skan	cvd->vdev_id = mvd->vdev_children;
873169695Skan	vdev_add_child(mvd, cvd);
874169695Skan	vdev_top_update(cvd->vdev_top, cvd->vdev_top);
875169695Skan
876283010Spfg	if (mvd == mvd->vdev_top)
877169695Skan		vdev_top_transfer(cvd, mvd);
878169695Skan
879169695Skan	return (mvd);
880169695Skan}
881169695Skan
882169695Skan/*
883169695Skan * Remove a 1-way mirror/replacing vdev from the tree.
884169695Skan */
885283010Spfgvoid
886169695Skanvdev_remove_parent(vdev_t *cvd)
887169695Skan{
888169695Skan	vdev_t *mvd = cvd->vdev_parent;
889169695Skan	vdev_t *pvd = mvd->vdev_parent;
890169695Skan
891169695Skan	ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
892169695Skan
893169695Skan	ASSERT(mvd->vdev_children == 1);
894169695Skan	ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
895169695Skan	    mvd->vdev_ops == &vdev_replacing_ops ||
896169695Skan	    mvd->vdev_ops == &vdev_spare_ops);
897169695Skan	cvd->vdev_ashift = mvd->vdev_ashift;
898169695Skan	cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
899169695Skan	cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
900169695Skan
901283010Spfg	vdev_remove_child(mvd, cvd);
902283010Spfg	vdev_remove_child(pvd, mvd);
903283010Spfg
904283010Spfg	/*
905283010Spfg	 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
906283010Spfg	 * Otherwise, we could have detached an offline device, and when we
907283010Spfg	 * go to import the pool we'll think we have two top-level vdevs,
908283010Spfg	 * instead of a different version of the same top-level vdev.
909169695Skan	 */
910169695Skan	if (mvd->vdev_top == mvd) {
911169695Skan		uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
912169695Skan		cvd->vdev_orig_guid = cvd->vdev_guid;
913169695Skan		cvd->vdev_guid += guid_delta;
914169695Skan		cvd->vdev_guid_sum += guid_delta;
915169695Skan	}
916169695Skan	cvd->vdev_id = mvd->vdev_id;
917169695Skan	vdev_add_child(pvd, cvd);
918169695Skan	vdev_top_update(cvd->vdev_top, cvd->vdev_top);
919169695Skan
920169695Skan	if (cvd == cvd->vdev_top)
921169695Skan		vdev_top_transfer(mvd, cvd);
922169695Skan
923169695Skan	ASSERT(mvd->vdev_children == 0);
924169695Skan	vdev_free(mvd);
925169695Skan}
926169695Skan
927169695Skanint
928169695Skanvdev_metaslab_init(vdev_t *vd, uint64_t txg)
929169695Skan{
930169695Skan	spa_t *spa = vd->vdev_spa;
931169695Skan	objset_t *mos = spa->spa_meta_objset;
932169695Skan	uint64_t m;
933169695Skan	uint64_t oldc = vd->vdev_ms_count;
934169695Skan	uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
935169695Skan	metaslab_t **mspp;
936169695Skan	int error;
937169695Skan
938169695Skan	ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
939169695Skan
940169695Skan	/*
941169695Skan	 * This vdev is not being allocated from yet or is a hole.
942169695Skan	 */
943169695Skan	if (vd->vdev_ms_shift == 0)
944169695Skan		return (0);
945169695Skan
946169695Skan	ASSERT(!vd->vdev_ishole);
947169695Skan
948169695Skan	/*
949169695Skan	 * Compute the raidz-deflation ratio.  Note, we hard-code
950169695Skan	 * in 128k (1 << 17) because it is the "typical" blocksize.
951169695Skan	 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
952169695Skan	 * otherwise it would inconsistently account for existing bp's.
953169695Skan	 */
954169695Skan	vd->vdev_deflate_ratio = (1 << 17) /
955169695Skan	    (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
956169695Skan
957169695Skan	ASSERT(oldc <= newc);
958169695Skan
959169695Skan	mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
960169695Skan
961169695Skan	if (oldc != 0) {
962169695Skan		bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
963169695Skan		kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
964169695Skan	}
965169695Skan
966169695Skan	vd->vdev_ms = mspp;
967169695Skan	vd->vdev_ms_count = newc;
968169695Skan
969169695Skan	for (m = oldc; m < newc; m++) {
970169695Skan		uint64_t object = 0;
971169695Skan
972169695Skan		if (txg == 0) {
973169695Skan			error = dmu_read(mos, vd->vdev_ms_array,
974169695Skan			    m * sizeof (uint64_t), sizeof (uint64_t), &object,
975169695Skan			    DMU_READ_PREFETCH);
976169695Skan			if (error)
977169695Skan				return (error);
978169695Skan		}
979169695Skan
980169695Skan		error = metaslab_init(vd->vdev_mg, m, object, txg,
981169695Skan		    &(vd->vdev_ms[m]));
982169695Skan		if (error)
983169695Skan			return (error);
984169695Skan	}
985169695Skan
986169695Skan	if (txg == 0)
987169695Skan		spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
988169695Skan
989169695Skan	/*
990169695Skan	 * If the vdev is being removed we don't activate
991169695Skan	 * the metaslabs since we want to ensure that no new
992169695Skan	 * allocations are performed on this device.
993169695Skan	 */
994169695Skan	if (oldc == 0 && !vd->vdev_removing)
995169695Skan		metaslab_group_activate(vd->vdev_mg);
996169695Skan
997169695Skan	if (txg == 0)
998169695Skan		spa_config_exit(spa, SCL_ALLOC, FTAG);
999169695Skan
1000169695Skan	return (0);
1001169695Skan}
1002169695Skan
1003169695Skanvoid
1004169695Skanvdev_metaslab_fini(vdev_t *vd)
1005169695Skan{
1006169695Skan	uint64_t m;
1007169695Skan	uint64_t count = vd->vdev_ms_count;
1008169695Skan
1009169695Skan	if (vd->vdev_ms != NULL) {
1010169695Skan		metaslab_group_passivate(vd->vdev_mg);
1011169695Skan		for (m = 0; m < count; m++) {
1012169695Skan			metaslab_t *msp = vd->vdev_ms[m];
1013283010Spfg
1014283010Spfg			if (msp != NULL)
1015283010Spfg				metaslab_fini(msp);
1016283010Spfg		}
1017283010Spfg		kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1018283010Spfg		vd->vdev_ms = NULL;
1019169695Skan	}
1020169695Skan}
1021169695Skan
1022283010Spfgtypedef struct vdev_probe_stats {
1023283010Spfg	boolean_t	vps_readable;
1024169695Skan	boolean_t	vps_writeable;
1025169695Skan	int		vps_flags;
1026169695Skan} vdev_probe_stats_t;
1027169695Skan
1028169695Skanstatic void
1029169695Skanvdev_probe_done(zio_t *zio)
1030169695Skan{
1031283010Spfg	spa_t *spa = zio->io_spa;
1032283010Spfg	vdev_t *vd = zio->io_vd;
1033283010Spfg	vdev_probe_stats_t *vps = zio->io_private;
1034169695Skan
1035169695Skan	ASSERT(vd->vdev_probe_zio != NULL);
1036169695Skan
1037169695Skan	if (zio->io_type == ZIO_TYPE_READ) {
1038169695Skan		if (zio->io_error == 0)
1039169695Skan			vps->vps_readable = 1;
1040169695Skan		if (zio->io_error == 0 && spa_writeable(spa)) {
1041169695Skan			zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1042283010Spfg			    zio->io_offset, zio->io_size, zio->io_data,
1043169695Skan			    ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1044169695Skan			    ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1045169695Skan		} else {
1046169695Skan			zio_buf_free(zio->io_data, zio->io_size);
1047169695Skan		}
1048169695Skan	} else if (zio->io_type == ZIO_TYPE_WRITE) {
1049169695Skan		if (zio->io_error == 0)
1050169695Skan			vps->vps_writeable = 1;
1051169695Skan		zio_buf_free(zio->io_data, zio->io_size);
1052169695Skan	} else if (zio->io_type == ZIO_TYPE_NULL) {
1053169695Skan		zio_t *pio;
1054169695Skan
1055169695Skan		vd->vdev_cant_read |= !vps->vps_readable;
1056169695Skan		vd->vdev_cant_write |= !vps->vps_writeable;
1057169695Skan
1058169695Skan		if (vdev_readable(vd) &&
1059169695Skan		    (vdev_writeable(vd) || !spa_writeable(spa))) {
1060169695Skan			zio->io_error = 0;
1061169695Skan		} else {
1062169695Skan			ASSERT(zio->io_error != 0);
1063169695Skan			zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1064169695Skan			    spa, vd, NULL, 0, 0);
1065169695Skan			zio->io_error = SET_ERROR(ENXIO);
1066169695Skan		}
1067169695Skan
1068169695Skan		mutex_enter(&vd->vdev_probe_lock);
1069169695Skan		ASSERT(vd->vdev_probe_zio == zio);
1070169695Skan		vd->vdev_probe_zio = NULL;
1071169695Skan		mutex_exit(&vd->vdev_probe_lock);
1072169695Skan
1073169695Skan		while ((pio = zio_walk_parents(zio)) != NULL)
1074169695Skan			if (!vdev_accessible(vd, pio))
1075169695Skan				pio->io_error = SET_ERROR(ENXIO);
1076169695Skan
1077169695Skan		kmem_free(vps, sizeof (*vps));
1078169695Skan	}
1079169695Skan}
1080169695Skan
1081169695Skan/*
1082169695Skan * Determine whether this device is accessible.
1083169695Skan *
1084169695Skan * Read and write to several known locations: the pad regions of each
1085169695Skan * vdev label but the first, which we leave alone in case it contains
1086169695Skan * a VTOC.
1087169695Skan */
1088169695Skanzio_t *
1089169695Skanvdev_probe(vdev_t *vd, zio_t *zio)
1090169695Skan{
1091169695Skan	spa_t *spa = vd->vdev_spa;
1092169695Skan	vdev_probe_stats_t *vps = NULL;
1093169695Skan	zio_t *pio;
1094169695Skan
1095169695Skan	ASSERT(vd->vdev_ops->vdev_op_leaf);
1096169695Skan
1097169695Skan	/*
1098169695Skan	 * Don't probe the probe.
1099169695Skan	 */
1100169695Skan	if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1101169695Skan		return (NULL);
1102169695Skan
1103169695Skan	/*
1104169695Skan	 * To prevent 'probe storms' when a device fails, we create
1105169695Skan	 * just one probe i/o at a time.  All zios that want to probe
1106169695Skan	 * this vdev will become parents of the probe io.
1107169695Skan	 */
1108169695Skan	mutex_enter(&vd->vdev_probe_lock);
1109169695Skan
1110169695Skan	if ((pio = vd->vdev_probe_zio) == NULL) {
1111169695Skan		vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1112169695Skan
1113169695Skan		vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1114169695Skan		    ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1115169695Skan		    ZIO_FLAG_TRYHARD;
1116169695Skan
1117169695Skan		if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1118169695Skan			/*
1119169695Skan			 * vdev_cant_read and vdev_cant_write can only
1120169695Skan			 * transition from TRUE to FALSE when we have the
1121169695Skan			 * SCL_ZIO lock as writer; otherwise they can only
1122169695Skan			 * transition from FALSE to TRUE.  This ensures that
1123169695Skan			 * any zio looking at these values can assume that
1124169695Skan			 * failures persist for the life of the I/O.  That's
1125169695Skan			 * important because when a device has intermittent
1126169695Skan			 * connectivity problems, we want to ensure that
1127169695Skan			 * they're ascribed to the device (ENXIO) and not
1128169695Skan			 * the zio (EIO).
1129169695Skan			 *
1130169695Skan			 * Since we hold SCL_ZIO as writer here, clear both
1131169695Skan			 * values so the probe can reevaluate from first
1132169695Skan			 * principles.
1133169695Skan			 */
1134169695Skan			vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1135169695Skan			vd->vdev_cant_read = B_FALSE;
1136169695Skan			vd->vdev_cant_write = B_FALSE;
1137169695Skan		}
1138169695Skan
1139169695Skan		vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1140169695Skan		    vdev_probe_done, vps,
1141169695Skan		    vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1142169695Skan
1143169695Skan		/*
1144169695Skan		 * We can't change the vdev state in this context, so we
1145169695Skan		 * kick off an async task to do it on our behalf.
1146169695Skan		 */
1147169695Skan		if (zio != NULL) {
1148169695Skan			vd->vdev_probe_wanted = B_TRUE;
1149169695Skan			spa_async_request(spa, SPA_ASYNC_PROBE);
1150169695Skan		}
1151169695Skan	}
1152169695Skan
1153169695Skan	if (zio != NULL)
1154169695Skan		zio_add_child(zio, pio);
1155169695Skan
1156169695Skan	mutex_exit(&vd->vdev_probe_lock);
1157169695Skan
1158169695Skan	if (vps == NULL) {
1159169695Skan		ASSERT(zio != NULL);
1160169695Skan		return (NULL);
1161169695Skan	}
1162169695Skan
1163169695Skan	for (int l = 1; l < VDEV_LABELS; l++) {
1164169695Skan		zio_nowait(zio_read_phys(pio, vd,
1165169695Skan		    vdev_label_offset(vd->vdev_psize, l,
1166169695Skan		    offsetof(vdev_label_t, vl_pad2)),
1167169695Skan		    VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1168169695Skan		    ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1169169695Skan		    ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1170169695Skan	}
1171169695Skan
1172169695Skan	if (zio == NULL)
1173169695Skan		return (pio);
1174169695Skan
1175169695Skan	zio_nowait(pio);
1176169695Skan	return (NULL);
1177169695Skan}
1178169695Skan
1179169695Skanstatic void
1180169695Skanvdev_open_child(void *arg)
1181169695Skan{
1182169695Skan	vdev_t *vd = arg;
1183169695Skan
1184169695Skan	vd->vdev_open_thread = curthread;
1185169695Skan	vd->vdev_open_error = vdev_open(vd);
1186169695Skan	vd->vdev_open_thread = NULL;
1187169695Skan}
1188169695Skan
1189169695Skanboolean_t
1190169695Skanvdev_uses_zvols(vdev_t *vd)
1191169695Skan{
1192169695Skan	if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1193169695Skan	    strlen(ZVOL_DIR)) == 0)
1194169695Skan		return (B_TRUE);
1195169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1196169695Skan		if (vdev_uses_zvols(vd->vdev_child[c]))
1197169695Skan			return (B_TRUE);
1198169695Skan	return (B_FALSE);
1199169695Skan}
1200169695Skan
1201169695Skanvoid
1202169695Skanvdev_open_children(vdev_t *vd)
1203169695Skan{
1204169695Skan	taskq_t *tq;
1205169695Skan	int children = vd->vdev_children;
1206169695Skan
1207169695Skan	/*
1208169695Skan	 * in order to handle pools on top of zvols, do the opens
1209169695Skan	 * in a single thread so that the same thread holds the
1210169695Skan	 * spa_namespace_lock
1211169695Skan	 */
1212169695Skan	if (B_TRUE || vdev_uses_zvols(vd)) {
1213169695Skan		for (int c = 0; c < children; c++)
1214169695Skan			vd->vdev_child[c]->vdev_open_error =
1215169695Skan			    vdev_open(vd->vdev_child[c]);
1216169695Skan		return;
1217169695Skan	}
1218169695Skan	tq = taskq_create("vdev_open", children, minclsyspri,
1219169695Skan	    children, children, TASKQ_PREPOPULATE);
1220169695Skan
1221169695Skan	for (int c = 0; c < children; c++)
1222169695Skan		VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1223169695Skan		    TQ_SLEEP) != 0);
1224169695Skan
1225169695Skan	taskq_destroy(tq);
1226169695Skan}
1227169695Skan
1228169695Skan/*
1229169695Skan * Prepare a virtual device for access.
1230169695Skan */
1231169695Skanint
1232169695Skanvdev_open(vdev_t *vd)
1233169695Skan{
1234169695Skan	spa_t *spa = vd->vdev_spa;
1235169695Skan	int error;
1236169695Skan	uint64_t osize = 0;
1237169695Skan	uint64_t max_osize = 0;
1238169695Skan	uint64_t asize, max_asize, psize;
1239169695Skan	uint64_t logical_ashift = 0;
1240169695Skan	uint64_t physical_ashift = 0;
1241169695Skan
1242169695Skan	ASSERT(vd->vdev_open_thread == curthread ||
1243169695Skan	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1244169695Skan	ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1245169695Skan	    vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1246169695Skan	    vd->vdev_state == VDEV_STATE_OFFLINE);
1247169695Skan
1248169695Skan	vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1249169695Skan	vd->vdev_cant_read = B_FALSE;
1250169695Skan	vd->vdev_cant_write = B_FALSE;
1251169695Skan	vd->vdev_notrim = B_FALSE;
1252169695Skan	vd->vdev_min_asize = vdev_get_min_asize(vd);
1253169695Skan
1254169695Skan	/*
1255169695Skan	 * If this vdev is not removed, check its fault status.  If it's
1256169695Skan	 * faulted, bail out of the open.
1257169695Skan	 */
1258169695Skan	if (!vd->vdev_removed && vd->vdev_faulted) {
1259169695Skan		ASSERT(vd->vdev_children == 0);
1260169695Skan		ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1261169695Skan		    vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1262169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1263169695Skan		    vd->vdev_label_aux);
1264169695Skan		return (SET_ERROR(ENXIO));
1265169695Skan	} else if (vd->vdev_offline) {
1266169695Skan		ASSERT(vd->vdev_children == 0);
1267169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1268169695Skan		return (SET_ERROR(ENXIO));
1269169695Skan	}
1270169695Skan
1271169695Skan	error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1272169695Skan	    &logical_ashift, &physical_ashift);
1273169695Skan
1274169695Skan	/*
1275169695Skan	 * Reset the vdev_reopening flag so that we actually close
1276169695Skan	 * the vdev on error.
1277169695Skan	 */
1278169695Skan	vd->vdev_reopening = B_FALSE;
1279169695Skan	if (zio_injection_enabled && error == 0)
1280169695Skan		error = zio_handle_device_injection(vd, NULL, ENXIO);
1281169695Skan
1282169695Skan	if (error) {
1283169695Skan		if (vd->vdev_removed &&
1284169695Skan		    vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1285169695Skan			vd->vdev_removed = B_FALSE;
1286169695Skan
1287169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1288169695Skan		    vd->vdev_stat.vs_aux);
1289169695Skan		return (error);
1290169695Skan	}
1291169695Skan
1292169695Skan	vd->vdev_removed = B_FALSE;
1293169695Skan
1294169695Skan	/*
1295169695Skan	 * Recheck the faulted flag now that we have confirmed that
1296169695Skan	 * the vdev is accessible.  If we're faulted, bail.
1297169695Skan	 */
1298169695Skan	if (vd->vdev_faulted) {
1299169695Skan		ASSERT(vd->vdev_children == 0);
1300169695Skan		ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1301169695Skan		    vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1302169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1303169695Skan		    vd->vdev_label_aux);
1304169695Skan		return (SET_ERROR(ENXIO));
1305169695Skan	}
1306169695Skan
1307169695Skan	if (vd->vdev_degraded) {
1308169695Skan		ASSERT(vd->vdev_children == 0);
1309169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1310169695Skan		    VDEV_AUX_ERR_EXCEEDED);
1311169695Skan	} else {
1312169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1313169695Skan	}
1314169695Skan
1315169695Skan	/*
1316169695Skan	 * For hole or missing vdevs we just return success.
1317169695Skan	 */
1318169695Skan	if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1319169695Skan		return (0);
1320169695Skan
1321169695Skan	if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1322169695Skan		trim_map_create(vd);
1323169695Skan
1324169695Skan	for (int c = 0; c < vd->vdev_children; c++) {
1325169695Skan		if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1326169695Skan			vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1327169695Skan			    VDEV_AUX_NONE);
1328169695Skan			break;
1329169695Skan		}
1330169695Skan	}
1331169695Skan
1332169695Skan	osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1333169695Skan	max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1334169695Skan
1335169695Skan	if (vd->vdev_children == 0) {
1336169695Skan		if (osize < SPA_MINDEVSIZE) {
1337169695Skan			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1338169695Skan			    VDEV_AUX_TOO_SMALL);
1339169695Skan			return (SET_ERROR(EOVERFLOW));
1340169695Skan		}
1341169695Skan		psize = osize;
1342169695Skan		asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1343169695Skan		max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1344169695Skan		    VDEV_LABEL_END_SIZE);
1345169695Skan	} else {
1346169695Skan		if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1347169695Skan		    (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1348169695Skan			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1349169695Skan			    VDEV_AUX_TOO_SMALL);
1350169695Skan			return (SET_ERROR(EOVERFLOW));
1351169695Skan		}
1352169695Skan		psize = 0;
1353169695Skan		asize = osize;
1354169695Skan		max_asize = max_osize;
1355169695Skan	}
1356169695Skan
1357169695Skan	vd->vdev_psize = psize;
1358169695Skan
1359169695Skan	/*
1360169695Skan	 * Make sure the allocatable size hasn't shrunk.
1361169695Skan	 */
1362169695Skan	if (asize < vd->vdev_min_asize) {
1363169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1364169695Skan		    VDEV_AUX_BAD_LABEL);
1365169695Skan		return (SET_ERROR(EINVAL));
1366169695Skan	}
1367169695Skan
1368169695Skan	vd->vdev_physical_ashift =
1369169695Skan	    MAX(physical_ashift, vd->vdev_physical_ashift);
1370169695Skan	vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1371169695Skan	vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1372169695Skan
1373169695Skan	if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1374169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1375169695Skan		    VDEV_AUX_ASHIFT_TOO_BIG);
1376169695Skan		return (EINVAL);
1377169695Skan	}
1378169695Skan
1379169695Skan	if (vd->vdev_asize == 0) {
1380169695Skan		/*
1381169695Skan		 * This is the first-ever open, so use the computed values.
1382169695Skan		 * For testing purposes, a higher ashift can be requested.
1383169695Skan		 */
1384169695Skan		vd->vdev_asize = asize;
1385169695Skan		vd->vdev_max_asize = max_asize;
1386169695Skan	} else {
1387169695Skan		/*
1388169695Skan		 * Make sure the alignment requirement hasn't increased.
1389169695Skan		 */
1390169695Skan		if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1391169695Skan		    vd->vdev_ops->vdev_op_leaf) {
1392169695Skan			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1393169695Skan			    VDEV_AUX_BAD_LABEL);
1394169695Skan			return (EINVAL);
1395169695Skan		}
1396169695Skan		vd->vdev_max_asize = max_asize;
1397169695Skan	}
1398169695Skan
1399169695Skan	/*
1400169695Skan	 * If all children are healthy and the asize has increased,
1401169695Skan	 * then we've experienced dynamic LUN growth.  If automatic
1402169695Skan	 * expansion is enabled then use the additional space.
1403169695Skan	 */
1404169695Skan	if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1405169695Skan	    (vd->vdev_expanding || spa->spa_autoexpand))
1406169695Skan		vd->vdev_asize = asize;
1407169695Skan
1408169695Skan	vdev_set_min_asize(vd);
1409169695Skan
1410169695Skan	/*
1411169695Skan	 * Ensure we can issue some IO before declaring the
1412169695Skan	 * vdev open for business.
1413169695Skan	 */
1414169695Skan	if (vd->vdev_ops->vdev_op_leaf &&
1415169695Skan	    (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1416169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1417169695Skan		    VDEV_AUX_ERR_EXCEEDED);
1418169695Skan		return (error);
1419169695Skan	}
1420169695Skan
1421169695Skan	/*
1422169695Skan	 * Track the min and max ashift values for normal data devices.
1423169695Skan	 */
1424169695Skan	if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1425169695Skan	    !vd->vdev_islog && vd->vdev_aux == NULL) {
1426169695Skan		if (vd->vdev_ashift > spa->spa_max_ashift)
1427169695Skan			spa->spa_max_ashift = vd->vdev_ashift;
1428169695Skan		if (vd->vdev_ashift < spa->spa_min_ashift)
1429169695Skan			spa->spa_min_ashift = vd->vdev_ashift;
1430169695Skan	}
1431169695Skan
1432169695Skan	/*
1433169695Skan	 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1434169695Skan	 * resilver.  But don't do this if we are doing a reopen for a scrub,
1435169695Skan	 * since this would just restart the scrub we are already doing.
1436169695Skan	 */
1437169695Skan	if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1438169695Skan	    vdev_resilver_needed(vd, NULL, NULL))
1439169695Skan		spa_async_request(spa, SPA_ASYNC_RESILVER);
1440169695Skan
1441169695Skan	return (0);
1442169695Skan}
1443169695Skan
1444169695Skan/*
1445169695Skan * Called once the vdevs are all opened, this routine validates the label
1446169695Skan * contents.  This needs to be done before vdev_load() so that we don't
1447169695Skan * inadvertently do repair I/Os to the wrong device.
1448169695Skan *
1449169695Skan * If 'strict' is false ignore the spa guid check. This is necessary because
1450169695Skan * if the machine crashed during a re-guid the new guid might have been written
1451169695Skan * to all of the vdev labels, but not the cached config. The strict check
1452169695Skan * will be performed when the pool is opened again using the mos config.
1453169695Skan *
1454169695Skan * This function will only return failure if one of the vdevs indicates that it
1455169695Skan * has since been destroyed or exported.  This is only possible if
1456169695Skan * /etc/zfs/zpool.cache was readonly at the time.  Otherwise, the vdev state
1457169695Skan * will be updated but the function will return 0.
1458169695Skan */
1459169695Skanint
1460169695Skanvdev_validate(vdev_t *vd, boolean_t strict)
1461169695Skan{
1462169695Skan	spa_t *spa = vd->vdev_spa;
1463169695Skan	nvlist_t *label;
1464169695Skan	uint64_t guid = 0, top_guid;
1465169695Skan	uint64_t state;
1466169695Skan
1467169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1468169695Skan		if (vdev_validate(vd->vdev_child[c], strict) != 0)
1469169695Skan			return (SET_ERROR(EBADF));
1470169695Skan
1471169695Skan	/*
1472169695Skan	 * If the device has already failed, or was marked offline, don't do
1473169695Skan	 * any further validation.  Otherwise, label I/O will fail and we will
1474169695Skan	 * overwrite the previous state.
1475169695Skan	 */
1476169695Skan	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1477169695Skan		uint64_t aux_guid = 0;
1478169695Skan		nvlist_t *nvl;
1479169695Skan		uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1480169695Skan		    spa_last_synced_txg(spa) : -1ULL;
1481169695Skan
1482169695Skan		if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1483169695Skan			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1484169695Skan			    VDEV_AUX_BAD_LABEL);
1485169695Skan			return (0);
1486169695Skan		}
1487169695Skan
1488169695Skan		/*
1489169695Skan		 * Determine if this vdev has been split off into another
1490169695Skan		 * pool.  If so, then refuse to open it.
1491169695Skan		 */
1492169695Skan		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1493169695Skan		    &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1494169695Skan			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1495169695Skan			    VDEV_AUX_SPLIT_POOL);
1496169695Skan			nvlist_free(label);
1497169695Skan			return (0);
1498169695Skan		}
1499169695Skan
1500169695Skan		if (strict && (nvlist_lookup_uint64(label,
1501169695Skan		    ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1502169695Skan		    guid != spa_guid(spa))) {
1503169695Skan			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1504169695Skan			    VDEV_AUX_CORRUPT_DATA);
1505169695Skan			nvlist_free(label);
1506169695Skan			return (0);
1507169695Skan		}
1508169695Skan
1509169695Skan		if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1510169695Skan		    != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1511169695Skan		    &aux_guid) != 0)
1512169695Skan			aux_guid = 0;
1513169695Skan
1514169695Skan		/*
1515169695Skan		 * If this vdev just became a top-level vdev because its
1516169695Skan		 * sibling was detached, it will have adopted the parent's
1517169695Skan		 * vdev guid -- but the label may or may not be on disk yet.
1518169695Skan		 * Fortunately, either version of the label will have the
1519169695Skan		 * same top guid, so if we're a top-level vdev, we can
1520169695Skan		 * safely compare to that instead.
1521169695Skan		 *
1522169695Skan		 * If we split this vdev off instead, then we also check the
1523169695Skan		 * original pool's guid.  We don't want to consider the vdev
1524169695Skan		 * corrupt if it is partway through a split operation.
1525169695Skan		 */
1526169695Skan		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1527169695Skan		    &guid) != 0 ||
1528169695Skan		    nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1529169695Skan		    &top_guid) != 0 ||
1530169695Skan		    ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1531169695Skan		    (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1532169695Skan			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1533169695Skan			    VDEV_AUX_CORRUPT_DATA);
1534169695Skan			nvlist_free(label);
1535169695Skan			return (0);
1536169695Skan		}
1537169695Skan
1538169695Skan		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1539169695Skan		    &state) != 0) {
1540169695Skan			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1541169695Skan			    VDEV_AUX_CORRUPT_DATA);
1542169695Skan			nvlist_free(label);
1543169695Skan			return (0);
1544169695Skan		}
1545169695Skan
1546169695Skan		nvlist_free(label);
1547169695Skan
1548169695Skan		/*
1549169695Skan		 * If this is a verbatim import, no need to check the
1550169695Skan		 * state of the pool.
1551169695Skan		 */
1552169695Skan		if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1553169695Skan		    spa_load_state(spa) == SPA_LOAD_OPEN &&
1554169695Skan		    state != POOL_STATE_ACTIVE)
1555169695Skan			return (SET_ERROR(EBADF));
1556169695Skan
1557169695Skan		/*
1558169695Skan		 * If we were able to open and validate a vdev that was
1559169695Skan		 * previously marked permanently unavailable, clear that state
1560169695Skan		 * now.
1561169695Skan		 */
1562169695Skan		if (vd->vdev_not_present)
1563169695Skan			vd->vdev_not_present = 0;
1564169695Skan	}
1565169695Skan
1566169695Skan	return (0);
1567169695Skan}
1568169695Skan
1569169695Skan/*
1570169695Skan * Close a virtual device.
1571169695Skan */
1572169695Skanvoid
1573169695Skanvdev_close(vdev_t *vd)
1574169695Skan{
1575169695Skan	spa_t *spa = vd->vdev_spa;
1576169695Skan	vdev_t *pvd = vd->vdev_parent;
1577169695Skan
1578169695Skan	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1579169695Skan
1580169695Skan	/*
1581169695Skan	 * If our parent is reopening, then we are as well, unless we are
1582169695Skan	 * going offline.
1583169695Skan	 */
1584169695Skan	if (pvd != NULL && pvd->vdev_reopening)
1585169695Skan		vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1586169695Skan
1587169695Skan	vd->vdev_ops->vdev_op_close(vd);
1588169695Skan
1589169695Skan	vdev_cache_purge(vd);
1590169695Skan
1591169695Skan	if (vd->vdev_ops->vdev_op_leaf)
1592169695Skan		trim_map_destroy(vd);
1593169695Skan
1594169695Skan	/*
1595169695Skan	 * We record the previous state before we close it, so that if we are
1596169695Skan	 * doing a reopen(), we don't generate FMA ereports if we notice that
1597169695Skan	 * it's still faulted.
1598169695Skan	 */
1599169695Skan	vd->vdev_prevstate = vd->vdev_state;
1600169695Skan
1601169695Skan	if (vd->vdev_offline)
1602169695Skan		vd->vdev_state = VDEV_STATE_OFFLINE;
1603169695Skan	else
1604169695Skan		vd->vdev_state = VDEV_STATE_CLOSED;
1605169695Skan	vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1606169695Skan}
1607169695Skan
1608169695Skanvoid
1609169695Skanvdev_hold(vdev_t *vd)
1610169695Skan{
1611169695Skan	spa_t *spa = vd->vdev_spa;
1612169695Skan
1613169695Skan	ASSERT(spa_is_root(spa));
1614169695Skan	if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1615169695Skan		return;
1616169695Skan
1617169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1618169695Skan		vdev_hold(vd->vdev_child[c]);
1619169695Skan
1620169695Skan	if (vd->vdev_ops->vdev_op_leaf)
1621169695Skan		vd->vdev_ops->vdev_op_hold(vd);
1622169695Skan}
1623169695Skan
1624169695Skanvoid
1625169695Skanvdev_rele(vdev_t *vd)
1626169695Skan{
1627169695Skan	spa_t *spa = vd->vdev_spa;
1628169695Skan
1629169695Skan	ASSERT(spa_is_root(spa));
1630169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1631169695Skan		vdev_rele(vd->vdev_child[c]);
1632169695Skan
1633169695Skan	if (vd->vdev_ops->vdev_op_leaf)
1634169695Skan		vd->vdev_ops->vdev_op_rele(vd);
1635169695Skan}
1636169695Skan
1637169695Skan/*
1638169695Skan * Reopen all interior vdevs and any unopened leaves.  We don't actually
1639169695Skan * reopen leaf vdevs which had previously been opened as they might deadlock
1640169695Skan * on the spa_config_lock.  Instead we only obtain the leaf's physical size.
1641169695Skan * If the leaf has never been opened then open it, as usual.
1642169695Skan */
1643169695Skanvoid
1644169695Skanvdev_reopen(vdev_t *vd)
1645169695Skan{
1646169695Skan	spa_t *spa = vd->vdev_spa;
1647169695Skan
1648169695Skan	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1649169695Skan
1650169695Skan	/* set the reopening flag unless we're taking the vdev offline */
1651169695Skan	vd->vdev_reopening = !vd->vdev_offline;
1652169695Skan	vdev_close(vd);
1653169695Skan	(void) vdev_open(vd);
1654169695Skan
1655169695Skan	/*
1656169695Skan	 * Call vdev_validate() here to make sure we have the same device.
1657169695Skan	 * Otherwise, a device with an invalid label could be successfully
1658169695Skan	 * opened in response to vdev_reopen().
1659169695Skan	 */
1660169695Skan	if (vd->vdev_aux) {
1661169695Skan		(void) vdev_validate_aux(vd);
1662169695Skan		if (vdev_readable(vd) && vdev_writeable(vd) &&
1663169695Skan		    vd->vdev_aux == &spa->spa_l2cache &&
1664169695Skan		    !l2arc_vdev_present(vd))
1665169695Skan			l2arc_add_vdev(spa, vd);
1666169695Skan	} else {
1667169695Skan		(void) vdev_validate(vd, B_TRUE);
1668169695Skan	}
1669169695Skan
1670169695Skan	/*
1671169695Skan	 * Reassess parent vdev's health.
1672169695Skan	 */
1673169695Skan	vdev_propagate_state(vd);
1674169695Skan}
1675169695Skan
1676169695Skanint
1677169695Skanvdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1678169695Skan{
1679169695Skan	int error;
1680169695Skan
1681169695Skan	/*
1682169695Skan	 * Normally, partial opens (e.g. of a mirror) are allowed.
1683169695Skan	 * For a create, however, we want to fail the request if
1684169695Skan	 * there are any components we can't open.
1685169695Skan	 */
1686169695Skan	error = vdev_open(vd);
1687169695Skan
1688169695Skan	if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1689169695Skan		vdev_close(vd);
1690169695Skan		return (error ? error : ENXIO);
1691169695Skan	}
1692283010Spfg
1693169695Skan	/*
1694169695Skan	 * Recursively load DTLs and initialize all labels.
1695169695Skan	 */
1696169695Skan	if ((error = vdev_dtl_load(vd)) != 0 ||
1697169695Skan	    (error = vdev_label_init(vd, txg, isreplacing ?
1698169695Skan	    VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1699169695Skan		vdev_close(vd);
1700169695Skan		return (error);
1701169695Skan	}
1702169695Skan
1703169695Skan	return (0);
1704169695Skan}
1705169695Skan
1706169695Skanvoid
1707169695Skanvdev_metaslab_set_size(vdev_t *vd)
1708169695Skan{
1709169695Skan	/*
1710169695Skan	 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1711169695Skan	 */
1712169695Skan	vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1713169695Skan	vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1714169695Skan}
1715169695Skan
1716169695Skan/*
1717169695Skan * Maximize performance by inflating the configured ashift for top level
1718169695Skan * vdevs to be as close to the physical ashift as possible while maintaining
1719169695Skan * administrator defined limits and ensuring it doesn't go below the
1720169695Skan * logical ashift.
1721169695Skan */
1722169695Skanvoid
1723169695Skanvdev_ashift_optimize(vdev_t *vd)
1724169695Skan{
1725169695Skan	if (vd == vd->vdev_top) {
1726169695Skan		if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1727169695Skan			vd->vdev_ashift = MIN(
1728169695Skan			    MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1729169695Skan			    MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1730169695Skan		} else {
1731169695Skan			/*
1732169695Skan			 * Unusual case where logical ashift > physical ashift
1733169695Skan			 * so we can't cap the calculated ashift based on max
1734169695Skan			 * ashift as that would cause failures.
1735169695Skan			 * We still check if we need to increase it to match
1736169695Skan			 * the min ashift.
1737169695Skan			 */
1738169695Skan			vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1739169695Skan			    vd->vdev_ashift);
1740169695Skan		}
1741169695Skan	}
1742169695Skan}
1743169695Skan
1744169695Skanvoid
1745169695Skanvdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1746169695Skan{
1747169695Skan	ASSERT(vd == vd->vdev_top);
1748169695Skan	ASSERT(!vd->vdev_ishole);
1749169695Skan	ASSERT(ISP2(flags));
1750169695Skan	ASSERT(spa_writeable(vd->vdev_spa));
1751169695Skan
1752169695Skan	if (flags & VDD_METASLAB)
1753169695Skan		(void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1754169695Skan
1755169695Skan	if (flags & VDD_DTL)
1756169695Skan		(void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1757169695Skan
1758169695Skan	(void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1759169695Skan}
1760169695Skan
1761169695Skanvoid
1762169695Skanvdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1763169695Skan{
1764169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1765169695Skan		vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1766169695Skan
1767169695Skan	if (vd->vdev_ops->vdev_op_leaf)
1768169695Skan		vdev_dirty(vd->vdev_top, flags, vd, txg);
1769169695Skan}
1770169695Skan
1771169695Skan/*
1772169695Skan * DTLs.
1773169695Skan *
1774169695Skan * A vdev's DTL (dirty time log) is the set of transaction groups for which
1775169695Skan * the vdev has less than perfect replication.  There are four kinds of DTL:
1776169695Skan *
1777169695Skan * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1778169695Skan *
1779169695Skan * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1780169695Skan *
1781169695Skan * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1782169695Skan *	scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1783169695Skan *	txgs that was scrubbed.
1784169695Skan *
1785169695Skan * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1786169695Skan *	persistent errors or just some device being offline.
1787169695Skan *	Unlike the other three, the DTL_OUTAGE map is not generally
1788169695Skan *	maintained; it's only computed when needed, typically to
1789169695Skan *	determine whether a device can be detached.
1790169695Skan *
1791169695Skan * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1792169695Skan * either has the data or it doesn't.
1793169695Skan *
1794169695Skan * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1795169695Skan * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1796169695Skan * if any child is less than fully replicated, then so is its parent.
1797169695Skan * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1798169695Skan * comprising only those txgs which appear in 'maxfaults' or more children;
1799169695Skan * those are the txgs we don't have enough replication to read.  For example,
1800169695Skan * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1801169695Skan * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1802169695Skan * two child DTL_MISSING maps.
1803169695Skan *
1804169695Skan * It should be clear from the above that to compute the DTLs and outage maps
1805169695Skan * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1806169695Skan * Therefore, that is all we keep on disk.  When loading the pool, or after
1807169695Skan * a configuration change, we generate all other DTLs from first principles.
1808169695Skan */
1809169695Skanvoid
1810169695Skanvdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1811169695Skan{
1812169695Skan	range_tree_t *rt = vd->vdev_dtl[t];
1813169695Skan
1814169695Skan	ASSERT(t < DTL_TYPES);
1815169695Skan	ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1816169695Skan	ASSERT(spa_writeable(vd->vdev_spa));
1817169695Skan
1818169695Skan	mutex_enter(rt->rt_lock);
1819169695Skan	if (!range_tree_contains(rt, txg, size))
1820169695Skan		range_tree_add(rt, txg, size);
1821169695Skan	mutex_exit(rt->rt_lock);
1822169695Skan}
1823169695Skan
1824169695Skanboolean_t
1825169695Skanvdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1826169695Skan{
1827169695Skan	range_tree_t *rt = vd->vdev_dtl[t];
1828169695Skan	boolean_t dirty = B_FALSE;
1829169695Skan
1830169695Skan	ASSERT(t < DTL_TYPES);
1831169695Skan	ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1832169695Skan
1833169695Skan	mutex_enter(rt->rt_lock);
1834169695Skan	if (range_tree_space(rt) != 0)
1835169695Skan		dirty = range_tree_contains(rt, txg, size);
1836169695Skan	mutex_exit(rt->rt_lock);
1837169695Skan
1838169695Skan	return (dirty);
1839169695Skan}
1840169695Skan
1841169695Skanboolean_t
1842169695Skanvdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1843169695Skan{
1844169695Skan	range_tree_t *rt = vd->vdev_dtl[t];
1845169695Skan	boolean_t empty;
1846169695Skan
1847169695Skan	mutex_enter(rt->rt_lock);
1848169695Skan	empty = (range_tree_space(rt) == 0);
1849169695Skan	mutex_exit(rt->rt_lock);
1850169695Skan
1851169695Skan	return (empty);
1852169695Skan}
1853169695Skan
1854169695Skan/*
1855169695Skan * Returns the lowest txg in the DTL range.
1856169695Skan */
1857169695Skanstatic uint64_t
1858169695Skanvdev_dtl_min(vdev_t *vd)
1859169695Skan{
1860169695Skan	range_seg_t *rs;
1861169695Skan
1862169695Skan	ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1863169695Skan	ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1864169695Skan	ASSERT0(vd->vdev_children);
1865169695Skan
1866169695Skan	rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1867169695Skan	return (rs->rs_start - 1);
1868169695Skan}
1869169695Skan
1870169695Skan/*
1871169695Skan * Returns the highest txg in the DTL.
1872169695Skan */
1873169695Skanstatic uint64_t
1874169695Skanvdev_dtl_max(vdev_t *vd)
1875169695Skan{
1876169695Skan	range_seg_t *rs;
1877169695Skan
1878169695Skan	ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1879169695Skan	ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1880169695Skan	ASSERT0(vd->vdev_children);
1881169695Skan
1882169695Skan	rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1883169695Skan	return (rs->rs_end);
1884169695Skan}
1885169695Skan
1886169695Skan/*
1887169695Skan * Determine if a resilvering vdev should remove any DTL entries from
1888169695Skan * its range. If the vdev was resilvering for the entire duration of the
1889169695Skan * scan then it should excise that range from its DTLs. Otherwise, this
1890169695Skan * vdev is considered partially resilvered and should leave its DTL
1891169695Skan * entries intact. The comment in vdev_dtl_reassess() describes how we
1892169695Skan * excise the DTLs.
1893169695Skan */
1894169695Skanstatic boolean_t
1895169695Skanvdev_dtl_should_excise(vdev_t *vd)
1896169695Skan{
1897169695Skan	spa_t *spa = vd->vdev_spa;
1898169695Skan	dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1899169695Skan
1900169695Skan	ASSERT0(scn->scn_phys.scn_errors);
1901169695Skan	ASSERT0(vd->vdev_children);
1902169695Skan
1903169695Skan	if (vd->vdev_resilver_txg == 0 ||
1904169695Skan	    range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1905169695Skan		return (B_TRUE);
1906169695Skan
1907169695Skan	/*
1908169695Skan	 * When a resilver is initiated the scan will assign the scn_max_txg
1909169695Skan	 * value to the highest txg value that exists in all DTLs. If this
1910169695Skan	 * device's max DTL is not part of this scan (i.e. it is not in
1911169695Skan	 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1912169695Skan	 * for excision.
1913169695Skan	 */
1914169695Skan	if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1915169695Skan		ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1916169695Skan		ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1917169695Skan		ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1918169695Skan		return (B_TRUE);
1919169695Skan	}
1920169695Skan	return (B_FALSE);
1921169695Skan}
1922169695Skan
1923169695Skan/*
1924169695Skan * Reassess DTLs after a config change or scrub completion.
1925169695Skan */
1926169695Skanvoid
1927169695Skanvdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1928169695Skan{
1929169695Skan	spa_t *spa = vd->vdev_spa;
1930169695Skan	avl_tree_t reftree;
1931169695Skan	int minref;
1932169695Skan
1933169695Skan	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1934169695Skan
1935169695Skan	for (int c = 0; c < vd->vdev_children; c++)
1936169695Skan		vdev_dtl_reassess(vd->vdev_child[c], txg,
1937169695Skan		    scrub_txg, scrub_done);
1938169695Skan
1939169695Skan	if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1940169695Skan		return;
1941169695Skan
1942169695Skan	if (vd->vdev_ops->vdev_op_leaf) {
1943169695Skan		dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1944169695Skan
1945169695Skan		mutex_enter(&vd->vdev_dtl_lock);
1946169695Skan
1947169695Skan		/*
1948169695Skan		 * If we've completed a scan cleanly then determine
1949169695Skan		 * if this vdev should remove any DTLs. We only want to
1950169695Skan		 * excise regions on vdevs that were available during
1951169695Skan		 * the entire duration of this scan.
1952169695Skan		 */
1953169695Skan		if (scrub_txg != 0 &&
1954169695Skan		    (spa->spa_scrub_started ||
1955169695Skan		    (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1956169695Skan		    vdev_dtl_should_excise(vd)) {
1957169695Skan			/*
1958169695Skan			 * We completed a scrub up to scrub_txg.  If we
1959169695Skan			 * did it without rebooting, then the scrub dtl
1960169695Skan			 * will be valid, so excise the old region and
1961169695Skan			 * fold in the scrub dtl.  Otherwise, leave the
1962169695Skan			 * dtl as-is if there was an error.
1963169695Skan			 *
1964169695Skan			 * There's little trick here: to excise the beginning
1965169695Skan			 * of the DTL_MISSING map, we put it into a reference
1966169695Skan			 * tree and then add a segment with refcnt -1 that
1967169695Skan			 * covers the range [0, scrub_txg).  This means
1968169695Skan			 * that each txg in that range has refcnt -1 or 0.
1969169695Skan			 * We then add DTL_SCRUB with a refcnt of 2, so that
1970169695Skan			 * entries in the range [0, scrub_txg) will have a
1971169695Skan			 * positive refcnt -- either 1 or 2.  We then convert
1972169695Skan			 * the reference tree into the new DTL_MISSING map.
1973169695Skan			 */
1974169695Skan			space_reftree_create(&reftree);
1975169695Skan			space_reftree_add_map(&reftree,
1976169695Skan			    vd->vdev_dtl[DTL_MISSING], 1);
1977169695Skan			space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
1978169695Skan			space_reftree_add_map(&reftree,
1979169695Skan			    vd->vdev_dtl[DTL_SCRUB], 2);
1980169695Skan			space_reftree_generate_map(&reftree,
1981169695Skan			    vd->vdev_dtl[DTL_MISSING], 1);
1982169695Skan			space_reftree_destroy(&reftree);
1983169695Skan		}
1984169695Skan		range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1985169695Skan		range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1986169695Skan		    range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
1987169695Skan		if (scrub_done)
1988169695Skan			range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1989169695Skan		range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1990169695Skan		if (!vdev_readable(vd))
1991169695Skan			range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1992169695Skan		else
1993169695Skan			range_tree_walk(vd->vdev_dtl[DTL_MISSING],
1994169695Skan			    range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
1995169695Skan
1996169695Skan		/*
1997169695Skan		 * If the vdev was resilvering and no longer has any
1998169695Skan		 * DTLs then reset its resilvering flag and dirty
1999169695Skan		 * the top level so that we persist the change.
2000169695Skan		 */
2001169695Skan		if (vd->vdev_resilver_txg != 0 &&
2002169695Skan		    range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2003169695Skan		    range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2004169695Skan			vd->vdev_resilver_txg = 0;
2005169695Skan			vdev_config_dirty(vd->vdev_top);
2006169695Skan		}
2007169695Skan
2008169695Skan		mutex_exit(&vd->vdev_dtl_lock);
2009169695Skan
2010169695Skan		if (txg != 0)
2011169695Skan			vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2012169695Skan		return;
2013169695Skan	}
2014169695Skan
2015169695Skan	mutex_enter(&vd->vdev_dtl_lock);
2016169695Skan	for (int t = 0; t < DTL_TYPES; t++) {
2017169695Skan		/* account for child's outage in parent's missing map */
2018169695Skan		int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2019169695Skan		if (t == DTL_SCRUB)
2020169695Skan			continue;			/* leaf vdevs only */
2021169695Skan		if (t == DTL_PARTIAL)
2022169695Skan			minref = 1;			/* i.e. non-zero */
2023169695Skan		else if (vd->vdev_nparity != 0)
2024169695Skan			minref = vd->vdev_nparity + 1;	/* RAID-Z */
2025169695Skan		else
2026169695Skan			minref = vd->vdev_children;	/* any kind of mirror */
2027169695Skan		space_reftree_create(&reftree);
2028169695Skan		for (int c = 0; c < vd->vdev_children; c++) {
2029169695Skan			vdev_t *cvd = vd->vdev_child[c];
2030169695Skan			mutex_enter(&cvd->vdev_dtl_lock);
2031169695Skan			space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2032169695Skan			mutex_exit(&cvd->vdev_dtl_lock);
2033169695Skan		}
2034169695Skan		space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2035169695Skan		space_reftree_destroy(&reftree);
2036169695Skan	}
2037169695Skan	mutex_exit(&vd->vdev_dtl_lock);
2038169695Skan}
2039169695Skan
2040169695Skanint
2041169695Skanvdev_dtl_load(vdev_t *vd)
2042169695Skan{
2043169695Skan	spa_t *spa = vd->vdev_spa;
2044169695Skan	objset_t *mos = spa->spa_meta_objset;
2045169695Skan	int error = 0;
2046169695Skan
2047169695Skan	if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2048169695Skan		ASSERT(!vd->vdev_ishole);
2049169695Skan
2050169695Skan		error = space_map_open(&vd->vdev_dtl_sm, mos,
2051169695Skan		    vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2052169695Skan		if (error)
2053169695Skan			return (error);
2054169695Skan		ASSERT(vd->vdev_dtl_sm != NULL);
2055169695Skan
2056169695Skan		mutex_enter(&vd->vdev_dtl_lock);
2057169695Skan
2058169695Skan		/*
2059169695Skan		 * Now that we've opened the space_map we need to update
2060169695Skan		 * the in-core DTL.
2061169695Skan		 */
2062169695Skan		space_map_update(vd->vdev_dtl_sm);
2063169695Skan
2064169695Skan		error = space_map_load(vd->vdev_dtl_sm,
2065169695Skan		    vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2066169695Skan		mutex_exit(&vd->vdev_dtl_lock);
2067169695Skan
2068169695Skan		return (error);
2069169695Skan	}
2070169695Skan
2071169695Skan	for (int c = 0; c < vd->vdev_children; c++) {
2072169695Skan		error = vdev_dtl_load(vd->vdev_child[c]);
2073169695Skan		if (error != 0)
2074169695Skan			break;
2075169695Skan	}
2076169695Skan
2077169695Skan	return (error);
2078169695Skan}
2079169695Skan
2080169695Skanvoid
2081169695Skanvdev_dtl_sync(vdev_t *vd, uint64_t txg)
2082169695Skan{
2083169695Skan	spa_t *spa = vd->vdev_spa;
2084169695Skan	range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2085169695Skan	objset_t *mos = spa->spa_meta_objset;
2086169695Skan	range_tree_t *rtsync;
2087169695Skan	kmutex_t rtlock;
2088169695Skan	dmu_tx_t *tx;
2089169695Skan	uint64_t object = space_map_object(vd->vdev_dtl_sm);
2090169695Skan
2091169695Skan	ASSERT(!vd->vdev_ishole);
2092169695Skan	ASSERT(vd->vdev_ops->vdev_op_leaf);
2093169695Skan
2094169695Skan	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2095169695Skan
2096169695Skan	if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2097169695Skan		mutex_enter(&vd->vdev_dtl_lock);
2098169695Skan		space_map_free(vd->vdev_dtl_sm, tx);
2099169695Skan		space_map_close(vd->vdev_dtl_sm);
2100169695Skan		vd->vdev_dtl_sm = NULL;
2101169695Skan		mutex_exit(&vd->vdev_dtl_lock);
2102169695Skan		dmu_tx_commit(tx);
2103169695Skan		return;
2104169695Skan	}
2105169695Skan
2106169695Skan	if (vd->vdev_dtl_sm == NULL) {
2107169695Skan		uint64_t new_object;
2108169695Skan
2109169695Skan		new_object = space_map_alloc(mos, tx);
2110169695Skan		VERIFY3U(new_object, !=, 0);
2111169695Skan
2112169695Skan		VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2113169695Skan		    0, -1ULL, 0, &vd->vdev_dtl_lock));
2114169695Skan		ASSERT(vd->vdev_dtl_sm != NULL);
2115169695Skan	}
2116169695Skan
2117169695Skan	bzero(&rtlock, sizeof(rtlock));
2118169695Skan	mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2119169695Skan
2120169695Skan	rtsync = range_tree_create(NULL, NULL, &rtlock);
2121169695Skan
2122169695Skan	mutex_enter(&rtlock);
2123169695Skan
2124169695Skan	mutex_enter(&vd->vdev_dtl_lock);
2125169695Skan	range_tree_walk(rt, range_tree_add, rtsync);
2126169695Skan	mutex_exit(&vd->vdev_dtl_lock);
2127169695Skan
2128169695Skan	space_map_truncate(vd->vdev_dtl_sm, tx);
2129169695Skan	space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2130169695Skan	range_tree_vacate(rtsync, NULL, NULL);
2131169695Skan
2132169695Skan	range_tree_destroy(rtsync);
2133169695Skan
2134169695Skan	mutex_exit(&rtlock);
2135169695Skan	mutex_destroy(&rtlock);
2136169695Skan
2137169695Skan	/*
2138169695Skan	 * If the object for the space map has changed then dirty
2139169695Skan	 * the top level so that we update the config.
2140169695Skan	 */
2141169695Skan	if (object != space_map_object(vd->vdev_dtl_sm)) {
2142169695Skan		zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2143169695Skan		    "new object %llu", txg, spa_name(spa), object,
2144169695Skan		    space_map_object(vd->vdev_dtl_sm));
2145169695Skan		vdev_config_dirty(vd->vdev_top);
2146169695Skan	}
2147169695Skan
2148169695Skan	dmu_tx_commit(tx);
2149169695Skan
2150169695Skan	mutex_enter(&vd->vdev_dtl_lock);
2151169695Skan	space_map_update(vd->vdev_dtl_sm);
2152169695Skan	mutex_exit(&vd->vdev_dtl_lock);
2153169695Skan}
2154169695Skan
2155169695Skan/*
2156169695Skan * Determine whether the specified vdev can be offlined/detached/removed
2157169695Skan * without losing data.
2158169695Skan */
2159169695Skanboolean_t
2160169695Skanvdev_dtl_required(vdev_t *vd)
2161169695Skan{
2162169695Skan	spa_t *spa = vd->vdev_spa;
2163169695Skan	vdev_t *tvd = vd->vdev_top;
2164169695Skan	uint8_t cant_read = vd->vdev_cant_read;
2165169695Skan	boolean_t required;
2166169695Skan
2167169695Skan	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2168169695Skan
2169169695Skan	if (vd == spa->spa_root_vdev || vd == tvd)
2170169695Skan		return (B_TRUE);
2171169695Skan
2172169695Skan	/*
2173169695Skan	 * Temporarily mark the device as unreadable, and then determine
2174169695Skan	 * whether this results in any DTL outages in the top-level vdev.
2175169695Skan	 * If not, we can safely offline/detach/remove the device.
2176169695Skan	 */
2177169695Skan	vd->vdev_cant_read = B_TRUE;
2178169695Skan	vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2179169695Skan	required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2180169695Skan	vd->vdev_cant_read = cant_read;
2181169695Skan	vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2182169695Skan
2183169695Skan	if (!required && zio_injection_enabled)
2184169695Skan		required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2185169695Skan
2186169695Skan	return (required);
2187169695Skan}
2188169695Skan
2189169695Skan/*
2190169695Skan * Determine if resilver is needed, and if so the txg range.
2191169695Skan */
2192169695Skanboolean_t
2193169695Skanvdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2194169695Skan{
2195169695Skan	boolean_t needed = B_FALSE;
2196169695Skan	uint64_t thismin = UINT64_MAX;
2197169695Skan	uint64_t thismax = 0;
2198169695Skan
2199169695Skan	if (vd->vdev_children == 0) {
2200169695Skan		mutex_enter(&vd->vdev_dtl_lock);
2201169695Skan		if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2202169695Skan		    vdev_writeable(vd)) {
2203169695Skan
2204169695Skan			thismin = vdev_dtl_min(vd);
2205169695Skan			thismax = vdev_dtl_max(vd);
2206169695Skan			needed = B_TRUE;
2207169695Skan		}
2208169695Skan		mutex_exit(&vd->vdev_dtl_lock);
2209169695Skan	} else {
2210169695Skan		for (int c = 0; c < vd->vdev_children; c++) {
2211169695Skan			vdev_t *cvd = vd->vdev_child[c];
2212169695Skan			uint64_t cmin, cmax;
2213169695Skan
2214169695Skan			if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2215169695Skan				thismin = MIN(thismin, cmin);
2216169695Skan				thismax = MAX(thismax, cmax);
2217169695Skan				needed = B_TRUE;
2218169695Skan			}
2219169695Skan		}
2220169695Skan	}
2221169695Skan
2222169695Skan	if (needed && minp) {
2223169695Skan		*minp = thismin;
2224169695Skan		*maxp = thismax;
2225169695Skan	}
2226169695Skan	return (needed);
2227169695Skan}
2228169695Skan
2229169695Skanvoid
2230169695Skanvdev_load(vdev_t *vd)
2231169695Skan{
2232169695Skan	/*
2233169695Skan	 * Recursively load all children.
2234169695Skan	 */
2235169695Skan	for (int c = 0; c < vd->vdev_children; c++)
2236169695Skan		vdev_load(vd->vdev_child[c]);
2237169695Skan
2238169695Skan	/*
2239169695Skan	 * If this is a top-level vdev, initialize its metaslabs.
2240169695Skan	 */
2241169695Skan	if (vd == vd->vdev_top && !vd->vdev_ishole &&
2242169695Skan	    (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2243169695Skan	    vdev_metaslab_init(vd, 0) != 0))
2244169695Skan		vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2245169695Skan		    VDEV_AUX_CORRUPT_DATA);
2246169695Skan
2247169695Skan	/*
2248169695Skan	 * If this is a leaf vdev, load its DTL.
2249169695Skan	 */
2250169695Skan	if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2251169695Skan		vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2252169695Skan		    VDEV_AUX_CORRUPT_DATA);
2253169695Skan}
2254169695Skan
2255169695Skan/*
2256169695Skan * The special vdev case is used for hot spares and l2cache devices.  Its
2257169695Skan * sole purpose it to set the vdev state for the associated vdev.  To do this,
2258169695Skan * we make sure that we can open the underlying device, then try to read the
2259169695Skan * label, and make sure that the label is sane and that it hasn't been
2260169695Skan * repurposed to another pool.
2261169695Skan */
2262169695Skanint
2263169695Skanvdev_validate_aux(vdev_t *vd)
2264169695Skan{
2265169695Skan	nvlist_t *label;
2266169695Skan	uint64_t guid, version;
2267169695Skan	uint64_t state;
2268169695Skan
2269169695Skan	if (!vdev_readable(vd))
2270169695Skan		return (0);
2271169695Skan
2272169695Skan	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2273169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2274169695Skan		    VDEV_AUX_CORRUPT_DATA);
2275169695Skan		return (-1);
2276169695Skan	}
2277169695Skan
2278169695Skan	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2279169695Skan	    !SPA_VERSION_IS_SUPPORTED(version) ||
2280169695Skan	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2281169695Skan	    guid != vd->vdev_guid ||
2282169695Skan	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2283169695Skan		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2284169695Skan		    VDEV_AUX_CORRUPT_DATA);
2285169695Skan		nvlist_free(label);
2286169695Skan		return (-1);
2287169695Skan	}
2288169695Skan
2289169695Skan	/*
2290169695Skan	 * We don't actually check the pool state here.  If it's in fact in
2291169695Skan	 * use by another pool, we update this fact on the fly when requested.
2292169695Skan	 */
2293169695Skan	nvlist_free(label);
2294169695Skan	return (0);
2295169695Skan}
2296169695Skan
2297169695Skanvoid
2298169695Skanvdev_remove(vdev_t *vd, uint64_t txg)
2299169695Skan{
2300169695Skan	spa_t *spa = vd->vdev_spa;
2301169695Skan	objset_t *mos = spa->spa_meta_objset;
2302169695Skan	dmu_tx_t *tx;
2303169695Skan
2304169695Skan	tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2305169695Skan
2306169695Skan	if (vd->vdev_ms != NULL) {
2307169695Skan		metaslab_group_t *mg = vd->vdev_mg;
2308169695Skan
2309169695Skan		metaslab_group_histogram_verify(mg);
2310169695Skan		metaslab_class_histogram_verify(mg->mg_class);
2311169695Skan
2312169695Skan		for (int m = 0; m < vd->vdev_ms_count; m++) {
2313169695Skan			metaslab_t *msp = vd->vdev_ms[m];
2314169695Skan
2315169695Skan			if (msp == NULL || msp->ms_sm == NULL)
2316169695Skan				continue;
2317169695Skan
2318169695Skan			mutex_enter(&msp->ms_lock);
2319169695Skan			/*
2320169695Skan			 * If the metaslab was not loaded when the vdev
2321169695Skan			 * was removed then the histogram accounting may
2322169695Skan			 * not be accurate. Update the histogram information
2323169695Skan			 * here so that we ensure that the metaslab group
2324169695Skan			 * and metaslab class are up-to-date.
2325169695Skan			 */
2326169695Skan			metaslab_group_histogram_remove(mg, msp);
2327169695Skan
2328169695Skan			VERIFY0(space_map_allocated(msp->ms_sm));
2329169695Skan			space_map_free(msp->ms_sm, tx);
2330169695Skan			space_map_close(msp->ms_sm);
2331169695Skan			msp->ms_sm = NULL;
2332169695Skan			mutex_exit(&msp->ms_lock);
2333169695Skan		}
2334169695Skan
2335169695Skan		metaslab_group_histogram_verify(mg);
2336169695Skan		metaslab_class_histogram_verify(mg->mg_class);
2337169695Skan		for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2338169695Skan			ASSERT0(mg->mg_histogram[i]);
2339169695Skan
2340169695Skan	}
2341169695Skan
2342169695Skan	if (vd->vdev_ms_array) {
2343169695Skan		(void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2344169695Skan		vd->vdev_ms_array = 0;
2345169695Skan	}
2346169695Skan	dmu_tx_commit(tx);
2347169695Skan}
2348169695Skan
2349169695Skanvoid
2350169695Skanvdev_sync_done(vdev_t *vd, uint64_t txg)
2351169695Skan{
2352169695Skan	metaslab_t *msp;
2353169695Skan	boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2354169695Skan
2355169695Skan	ASSERT(!vd->vdev_ishole);
2356169695Skan
2357169695Skan	while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2358169695Skan		metaslab_sync_done(msp, txg);
2359169695Skan
2360169695Skan	if (reassess)
2361169695Skan		metaslab_sync_reassess(vd->vdev_mg);
2362169695Skan}
2363169695Skan
2364169695Skanvoid
2365169695Skanvdev_sync(vdev_t *vd, uint64_t txg)
2366169695Skan{
2367169695Skan	spa_t *spa = vd->vdev_spa;
2368169695Skan	vdev_t *lvd;
2369169695Skan	metaslab_t *msp;
2370169695Skan	dmu_tx_t *tx;
2371169695Skan
2372169695Skan	ASSERT(!vd->vdev_ishole);
2373169695Skan
2374169695Skan	if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2375169695Skan		ASSERT(vd == vd->vdev_top);
2376169695Skan		tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2377169695Skan		vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2378169695Skan		    DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2379169695Skan		ASSERT(vd->vdev_ms_array != 0);
2380169695Skan		vdev_config_dirty(vd);
2381169695Skan		dmu_tx_commit(tx);
2382169695Skan	}
2383169695Skan
2384169695Skan	/*
2385169695Skan	 * Remove the metadata associated with this vdev once it's empty.
2386169695Skan	 */
2387169695Skan	if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2388169695Skan		vdev_remove(vd, txg);
2389169695Skan
2390169695Skan	while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2391169695Skan		metaslab_sync(msp, txg);
2392169695Skan		(void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2393169695Skan	}
2394169695Skan
2395169695Skan	while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2396169695Skan		vdev_dtl_sync(lvd, txg);
2397169695Skan
2398169695Skan	(void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2399169695Skan}
2400169695Skan
2401169695Skanuint64_t
2402169695Skanvdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2403169695Skan{
2404169695Skan	return (vd->vdev_ops->vdev_op_asize(vd, psize));
2405169695Skan}
2406169695Skan
2407169695Skan/*
2408169695Skan * Mark the given vdev faulted.  A faulted vdev behaves as if the device could
2409169695Skan * not be opened, and no I/O is attempted.
2410169695Skan */
2411169695Skanint
2412169695Skanvdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2413169695Skan{
2414169695Skan	vdev_t *vd, *tvd;
2415169695Skan
2416169695Skan	spa_vdev_state_enter(spa, SCL_NONE);
2417169695Skan
2418169695Skan	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2419169695Skan		return (spa_vdev_state_exit(spa, NULL, ENODEV));
2420169695Skan
2421169695Skan	if (!vd->vdev_ops->vdev_op_leaf)
2422169695Skan		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2423169695Skan
2424169695Skan	tvd = vd->vdev_top;
2425169695Skan
2426169695Skan	/*
2427169695Skan	 * We don't directly use the aux state here, but if we do a
2428169695Skan	 * vdev_reopen(), we need this value to be present to remember why we
2429169695Skan	 * were faulted.
2430169695Skan	 */
2431169695Skan	vd->vdev_label_aux = aux;
2432169695Skan
2433169695Skan	/*
2434169695Skan	 * Faulted state takes precedence over degraded.
2435169695Skan	 */
2436169695Skan	vd->vdev_delayed_close = B_FALSE;
2437169695Skan	vd->vdev_faulted = 1ULL;
2438169695Skan	vd->vdev_degraded = 0ULL;
2439169695Skan	vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2440169695Skan
2441169695Skan	/*
2442169695Skan	 * If this device has the only valid copy of the data, then
2443169695Skan	 * back off and simply mark the vdev as degraded instead.
2444169695Skan	 */
2445169695Skan	if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2446169695Skan		vd->vdev_degraded = 1ULL;
2447169695Skan		vd->vdev_faulted = 0ULL;
2448169695Skan
2449169695Skan		/*
2450169695Skan		 * If we reopen the device and it's not dead, only then do we
2451169695Skan		 * mark it degraded.
2452169695Skan		 */
2453169695Skan		vdev_reopen(tvd);
2454169695Skan
2455169695Skan		if (vdev_readable(vd))
2456169695Skan			vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2457169695Skan	}
2458169695Skan
2459169695Skan	return (spa_vdev_state_exit(spa, vd, 0));
2460169695Skan}
2461169695Skan
2462169695Skan/*
2463169695Skan * Mark the given vdev degraded.  A degraded vdev is purely an indication to the
2464169695Skan * user that something is wrong.  The vdev continues to operate as normal as far
2465169695Skan * as I/O is concerned.
2466169695Skan */
2467169695Skanint
2468169695Skanvdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2469169695Skan{
2470169695Skan	vdev_t *vd;
2471169695Skan
2472169695Skan	spa_vdev_state_enter(spa, SCL_NONE);
2473169695Skan
2474169695Skan	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2475169695Skan		return (spa_vdev_state_exit(spa, NULL, ENODEV));
2476169695Skan
2477169695Skan	if (!vd->vdev_ops->vdev_op_leaf)
2478169695Skan		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2479169695Skan
2480169695Skan	/*
2481169695Skan	 * If the vdev is already faulted, then don't do anything.
2482169695Skan	 */
2483169695Skan	if (vd->vdev_faulted || vd->vdev_degraded)
2484169695Skan		return (spa_vdev_state_exit(spa, NULL, 0));
2485169695Skan
2486169695Skan	vd->vdev_degraded = 1ULL;
2487169695Skan	if (!vdev_is_dead(vd))
2488169695Skan		vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2489169695Skan		    aux);
2490169695Skan
2491169695Skan	return (spa_vdev_state_exit(spa, vd, 0));
2492169695Skan}
2493169695Skan
2494169695Skan/*
2495169695Skan * Online the given vdev.
2496169695Skan *
2497169695Skan * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things.  First, any attached
2498169695Skan * spare device should be detached when the device finishes resilvering.
2499169695Skan * Second, the online should be treated like a 'test' online case, so no FMA
2500169695Skan * events are generated if the device fails to open.
2501169695Skan */
2502169695Skanint
2503169695Skanvdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2504169695Skan{
2505169695Skan	vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2506169695Skan
2507169695Skan	spa_vdev_state_enter(spa, SCL_NONE);
2508169695Skan
2509169695Skan	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2510169695Skan		return (spa_vdev_state_exit(spa, NULL, ENODEV));
2511169695Skan
2512169695Skan	if (!vd->vdev_ops->vdev_op_leaf)
2513169695Skan		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2514169695Skan
2515169695Skan	tvd = vd->vdev_top;
2516169695Skan	vd->vdev_offline = B_FALSE;
2517169695Skan	vd->vdev_tmpoffline = B_FALSE;
2518169695Skan	vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2519169695Skan	vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2520169695Skan
2521169695Skan	/* XXX - L2ARC 1.0 does not support expansion */
2522169695Skan	if (!vd->vdev_aux) {
2523169695Skan		for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2524169695Skan			pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2525169695Skan	}
2526169695Skan
2527169695Skan	vdev_reopen(tvd);
2528169695Skan	vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2529169695Skan
2530169695Skan	if (!vd->vdev_aux) {
2531169695Skan		for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2532169695Skan			pvd->vdev_expanding = B_FALSE;
2533169695Skan	}
2534169695Skan
2535169695Skan	if (newstate)
2536169695Skan		*newstate = vd->vdev_state;
2537169695Skan	if ((flags & ZFS_ONLINE_UNSPARE) &&
2538169695Skan	    !vdev_is_dead(vd) && vd->vdev_parent &&
2539169695Skan	    vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2540169695Skan	    vd->vdev_parent->vdev_child[0] == vd)
2541169695Skan		vd->vdev_unspare = B_TRUE;
2542169695Skan
2543169695Skan	if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2544169695Skan
2545169695Skan		/* XXX - L2ARC 1.0 does not support expansion */
2546169695Skan		if (vd->vdev_aux)
2547169695Skan			return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2548169695Skan		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2549169695Skan	}
2550169695Skan	return (spa_vdev_state_exit(spa, vd, 0));
2551169695Skan}
2552169695Skan
2553169695Skanstatic int
2554169695Skanvdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2555169695Skan{
2556169695Skan	vdev_t *vd, *tvd;
2557169695Skan	int error = 0;
2558169695Skan	uint64_t generation;
2559169695Skan	metaslab_group_t *mg;
2560169695Skan
2561169695Skantop:
2562169695Skan	spa_vdev_state_enter(spa, SCL_ALLOC);
2563169695Skan
2564169695Skan	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2565169695Skan		return (spa_vdev_state_exit(spa, NULL, ENODEV));
2566169695Skan
2567169695Skan	if (!vd->vdev_ops->vdev_op_leaf)
2568169695Skan		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2569169695Skan
2570169695Skan	tvd = vd->vdev_top;
2571169695Skan	mg = tvd->vdev_mg;
2572169695Skan	generation = spa->spa_config_generation + 1;
2573169695Skan
2574169695Skan	/*
2575169695Skan	 * If the device isn't already offline, try to offline it.
2576169695Skan	 */
2577169695Skan	if (!vd->vdev_offline) {
2578169695Skan		/*
2579169695Skan		 * If this device has the only valid copy of some data,
2580169695Skan		 * don't allow it to be offlined. Log devices are always
2581169695Skan		 * expendable.
2582169695Skan		 */
2583169695Skan		if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2584169695Skan		    vdev_dtl_required(vd))
2585169695Skan			return (spa_vdev_state_exit(spa, NULL, EBUSY));
2586169695Skan
2587169695Skan		/*
2588169695Skan		 * If the top-level is a slog and it has had allocations
2589169695Skan		 * then proceed.  We check that the vdev's metaslab group
2590169695Skan		 * is not NULL since it's possible that we may have just
2591169695Skan		 * added this vdev but not yet initialized its metaslabs.
2592169695Skan		 */
2593169695Skan		if (tvd->vdev_islog && mg != NULL) {
2594169695Skan			/*
2595169695Skan			 * Prevent any future allocations.
2596169695Skan			 */
2597169695Skan			metaslab_group_passivate(mg);
2598169695Skan			(void) spa_vdev_state_exit(spa, vd, 0);
2599169695Skan
2600169695Skan			error = spa_offline_log(spa);
2601169695Skan
2602169695Skan			spa_vdev_state_enter(spa, SCL_ALLOC);
2603169695Skan
2604169695Skan			/*
2605169695Skan			 * Check to see if the config has changed.
2606169695Skan			 */
2607169695Skan			if (error || generation != spa->spa_config_generation) {
2608169695Skan				metaslab_group_activate(mg);
2609169695Skan				if (error)
2610169695Skan					return (spa_vdev_state_exit(spa,
2611169695Skan					    vd, error));
2612169695Skan				(void) spa_vdev_state_exit(spa, vd, 0);
2613169695Skan				goto top;
2614169695Skan			}
2615169695Skan			ASSERT0(tvd->vdev_stat.vs_alloc);
2616169695Skan		}
2617169695Skan
2618169695Skan		/*
2619169695Skan		 * Offline this device and reopen its top-level vdev.
2620169695Skan		 * If the top-level vdev is a log device then just offline
2621169695Skan		 * it. Otherwise, if this action results in the top-level
2622169695Skan		 * vdev becoming unusable, undo it and fail the request.
2623169695Skan		 */
2624169695Skan		vd->vdev_offline = B_TRUE;
2625169695Skan		vdev_reopen(tvd);
2626169695Skan
2627169695Skan		if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2628169695Skan		    vdev_is_dead(tvd)) {
2629169695Skan			vd->vdev_offline = B_FALSE;
2630169695Skan			vdev_reopen(tvd);
2631169695Skan			return (spa_vdev_state_exit(spa, NULL, EBUSY));
2632169695Skan		}
2633169695Skan
2634169695Skan		/*
2635169695Skan		 * Add the device back into the metaslab rotor so that
2636169695Skan		 * once we online the device it's open for business.
2637169695Skan		 */
2638169695Skan		if (tvd->vdev_islog && mg != NULL)
2639169695Skan			metaslab_group_activate(mg);
2640169695Skan	}
2641169695Skan
2642169695Skan	vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2643169695Skan
2644169695Skan	return (spa_vdev_state_exit(spa, vd, 0));
2645169695Skan}
2646169695Skan
2647169695Skanint
2648169695Skanvdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2649169695Skan{
2650169695Skan	int error;
2651169695Skan
2652169695Skan	mutex_enter(&spa->spa_vdev_top_lock);
2653169695Skan	error = vdev_offline_locked(spa, guid, flags);
2654169695Skan	mutex_exit(&spa->spa_vdev_top_lock);
2655169695Skan
2656169695Skan	return (error);
2657169695Skan}
2658169695Skan
2659169695Skan/*
2660169695Skan * Clear the error counts associated with this vdev.  Unlike vdev_online() and
2661169695Skan * vdev_offline(), we assume the spa config is locked.  We also clear all
2662169695Skan * children.  If 'vd' is NULL, then the user wants to clear all vdevs.
2663169695Skan */
2664169695Skanvoid
2665169695Skanvdev_clear(spa_t *spa, vdev_t *vd)
2666169695Skan{
2667169695Skan	vdev_t *rvd = spa->spa_root_vdev;
2668169695Skan
2669169695Skan	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2670169695Skan
2671169695Skan	if (vd == NULL)
2672169695Skan		vd = rvd;
2673169695Skan
2674169695Skan	vd->vdev_stat.vs_read_errors = 0;
2675169695Skan	vd->vdev_stat.vs_write_errors = 0;
2676169695Skan	vd->vdev_stat.vs_checksum_errors = 0;
2677169695Skan
2678169695Skan	for (int c = 0; c < vd->vdev_children; c++)
2679169695Skan		vdev_clear(spa, vd->vdev_child[c]);
2680169695Skan
2681169695Skan	if (vd == rvd) {
2682169695Skan		for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2683169695Skan			vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2684169695Skan
2685169695Skan		for (int c = 0; c < spa->spa_spares.sav_count; c++)
2686169695Skan			vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2687169695Skan	}
2688169695Skan
2689169695Skan	/*
2690169695Skan	 * If we're in the FAULTED state or have experienced failed I/O, then
2691169695Skan	 * clear the persistent state and attempt to reopen the device.  We
2692169695Skan	 * also mark the vdev config dirty, so that the new faulted state is
2693169695Skan	 * written out to disk.
2694169695Skan	 */
2695169695Skan	if (vd->vdev_faulted || vd->vdev_degraded ||
2696169695Skan	    !vdev_readable(vd) || !vdev_writeable(vd)) {
2697169695Skan
2698169695Skan		/*
2699169695Skan		 * When reopening in reponse to a clear event, it may be due to
2700169695Skan		 * a fmadm repair request.  In this case, if the device is
2701169695Skan		 * still broken, we want to still post the ereport again.
2702169695Skan		 */
2703169695Skan		vd->vdev_forcefault = B_TRUE;
2704169695Skan
2705169695Skan		vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2706169695Skan		vd->vdev_cant_read = B_FALSE;
2707169695Skan		vd->vdev_cant_write = B_FALSE;
2708169695Skan
2709169695Skan		vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2710169695Skan
2711169695Skan		vd->vdev_forcefault = B_FALSE;
2712169695Skan
2713169695Skan		if (vd != rvd && vdev_writeable(vd->vdev_top))
2714169695Skan			vdev_state_dirty(vd->vdev_top);
2715169695Skan
2716169695Skan		if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2717169695Skan			spa_async_request(spa, SPA_ASYNC_RESILVER);
2718169695Skan
2719169695Skan		spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2720169695Skan	}
2721169695Skan
2722169695Skan	/*
2723169695Skan	 * When clearing a FMA-diagnosed fault, we always want to
2724169695Skan	 * unspare the device, as we assume that the original spare was
2725169695Skan	 * done in response to the FMA fault.
2726169695Skan	 */
2727169695Skan	if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2728169695Skan	    vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2729169695Skan	    vd->vdev_parent->vdev_child[0] == vd)
2730169695Skan		vd->vdev_unspare = B_TRUE;
2731169695Skan}
2732169695Skan
2733169695Skanboolean_t
2734169695Skanvdev_is_dead(vdev_t *vd)
2735169695Skan{
2736169695Skan	/*
2737169695Skan	 * Holes and missing devices are always considered "dead".
2738169695Skan	 * This simplifies the code since we don't have to check for
2739169695Skan	 * these types of devices in the various code paths.
2740169695Skan	 * Instead we rely on the fact that we skip over dead devices
2741169695Skan	 * before issuing I/O to them.
2742169695Skan	 */
2743169695Skan	return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2744169695Skan	    vd->vdev_ops == &vdev_missing_ops);
2745169695Skan}
2746169695Skan
2747169695Skanboolean_t
2748169695Skanvdev_readable(vdev_t *vd)
2749169695Skan{
2750169695Skan	return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2751169695Skan}
2752169695Skan
2753169695Skanboolean_t
2754169695Skanvdev_writeable(vdev_t *vd)
2755169695Skan{
2756169695Skan	return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2757169695Skan}
2758169695Skan
2759169695Skanboolean_t
2760169695Skanvdev_allocatable(vdev_t *vd)
2761169695Skan{
2762169695Skan	uint64_t state = vd->vdev_state;
2763169695Skan
2764169695Skan	/*
2765169695Skan	 * We currently allow allocations from vdevs which may be in the
2766169695Skan	 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2767169695Skan	 * fails to reopen then we'll catch it later when we're holding
2768169695Skan	 * the proper locks.  Note that we have to get the vdev state
2769169695Skan	 * in a local variable because although it changes atomically,
2770169695Skan	 * we're asking two separate questions about it.
2771169695Skan	 */
2772169695Skan	return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2773169695Skan	    !vd->vdev_cant_write && !vd->vdev_ishole);
2774169695Skan}
2775169695Skan
2776169695Skanboolean_t
2777169695Skanvdev_accessible(vdev_t *vd, zio_t *zio)
2778169695Skan{
2779169695Skan	ASSERT(zio->io_vd == vd);
2780169695Skan
2781169695Skan	if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2782169695Skan		return (B_FALSE);
2783169695Skan
2784169695Skan	if (zio->io_type == ZIO_TYPE_READ)
2785169695Skan		return (!vd->vdev_cant_read);
2786169695Skan
2787169695Skan	if (zio->io_type == ZIO_TYPE_WRITE)
2788169695Skan		return (!vd->vdev_cant_write);
2789169695Skan
2790169695Skan	return (B_TRUE);
2791169695Skan}
2792169695Skan
2793169695Skan/*
2794169695Skan * Get statistics for the given vdev.
2795169695Skan */
2796169695Skanvoid
2797169695Skanvdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2798169695Skan{
2799169695Skan	spa_t *spa = vd->vdev_spa;
2800169695Skan	vdev_t *rvd = spa->spa_root_vdev;
2801169695Skan
2802169695Skan	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2803169695Skan
2804169695Skan	mutex_enter(&vd->vdev_stat_lock);
2805169695Skan	bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2806169695Skan	vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2807169695Skan	vs->vs_state = vd->vdev_state;
2808169695Skan	vs->vs_rsize = vdev_get_min_asize(vd);
2809169695Skan	if (vd->vdev_ops->vdev_op_leaf)
2810169695Skan		vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2811169695Skan	vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2812169695Skan	vs->vs_configured_ashift = vd->vdev_top != NULL
2813169695Skan	    ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2814169695Skan	vs->vs_logical_ashift = vd->vdev_logical_ashift;
2815169695Skan	vs->vs_physical_ashift = vd->vdev_physical_ashift;
2816169695Skan	if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2817169695Skan		vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2818169695Skan	}
2819169695Skan
2820169695Skan	/*
2821169695Skan	 * If we're getting stats on the root vdev, aggregate the I/O counts
2822169695Skan	 * over all top-level vdevs (i.e. the direct children of the root).
2823169695Skan	 */
2824169695Skan	if (vd == rvd) {
2825169695Skan		for (int c = 0; c < rvd->vdev_children; c++) {
2826169695Skan			vdev_t *cvd = rvd->vdev_child[c];
2827169695Skan			vdev_stat_t *cvs = &cvd->vdev_stat;
2828169695Skan
2829169695Skan			for (int t = 0; t < ZIO_TYPES; t++) {
2830169695Skan				vs->vs_ops[t] += cvs->vs_ops[t];
2831169695Skan				vs->vs_bytes[t] += cvs->vs_bytes[t];
2832169695Skan			}
2833169695Skan			cvs->vs_scan_removing = cvd->vdev_removing;
2834169695Skan		}
2835169695Skan	}
2836169695Skan	mutex_exit(&vd->vdev_stat_lock);
2837169695Skan}
2838169695Skan
2839169695Skanvoid
2840169695Skanvdev_clear_stats(vdev_t *vd)
2841169695Skan{
2842169695Skan	mutex_enter(&vd->vdev_stat_lock);
2843169695Skan	vd->vdev_stat.vs_space = 0;
2844169695Skan	vd->vdev_stat.vs_dspace = 0;
2845169695Skan	vd->vdev_stat.vs_alloc = 0;
2846169695Skan	mutex_exit(&vd->vdev_stat_lock);
2847169695Skan}
2848169695Skan
2849169695Skanvoid
2850169695Skanvdev_scan_stat_init(vdev_t *vd)
2851169695Skan{
2852169695Skan	vdev_stat_t *vs = &vd->vdev_stat;
2853169695Skan
2854169695Skan	for (int c = 0; c < vd->vdev_children; c++)
2855169695Skan		vdev_scan_stat_init(vd->vdev_child[c]);
2856169695Skan
2857169695Skan	mutex_enter(&vd->vdev_stat_lock);
2858169695Skan	vs->vs_scan_processed = 0;
2859169695Skan	mutex_exit(&vd->vdev_stat_lock);
2860169695Skan}
2861169695Skan
2862169695Skanvoid
2863169695Skanvdev_stat_update(zio_t *zio, uint64_t psize)
2864169695Skan{
2865169695Skan	spa_t *spa = zio->io_spa;
2866169695Skan	vdev_t *rvd = spa->spa_root_vdev;
2867169695Skan	vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2868169695Skan	vdev_t *pvd;
2869169695Skan	uint64_t txg = zio->io_txg;
2870169695Skan	vdev_stat_t *vs = &vd->vdev_stat;
2871169695Skan	zio_type_t type = zio->io_type;
2872169695Skan	int flags = zio->io_flags;
2873169695Skan
2874169695Skan	/*
2875169695Skan	 * If this i/o is a gang leader, it didn't do any actual work.
2876169695Skan	 */
2877169695Skan	if (zio->io_gang_tree)
2878169695Skan		return;
2879169695Skan
2880169695Skan	if (zio->io_error == 0) {
2881169695Skan		/*
2882169695Skan		 * If this is a root i/o, don't count it -- we've already
2883169695Skan		 * counted the top-level vdevs, and vdev_get_stats() will
2884169695Skan		 * aggregate them when asked.  This reduces contention on
2885169695Skan		 * the root vdev_stat_lock and implicitly handles blocks
2886169695Skan		 * that compress away to holes, for which there is no i/o.
2887169695Skan		 * (Holes never create vdev children, so all the counters
2888169695Skan		 * remain zero, which is what we want.)
2889169695Skan		 *
2890169695Skan		 * Note: this only applies to successful i/o (io_error == 0)
2891169695Skan		 * because unlike i/o counts, errors are not additive.
2892169695Skan		 * When reading a ditto block, for example, failure of
2893169695Skan		 * one top-level vdev does not imply a root-level error.
2894169695Skan		 */
2895169695Skan		if (vd == rvd)
2896169695Skan			return;
2897169695Skan
2898169695Skan		ASSERT(vd == zio->io_vd);
2899169695Skan
2900169695Skan		if (flags & ZIO_FLAG_IO_BYPASS)
2901169695Skan			return;
2902169695Skan
2903169695Skan		mutex_enter(&vd->vdev_stat_lock);
2904169695Skan
2905169695Skan		if (flags & ZIO_FLAG_IO_REPAIR) {
2906169695Skan			if (flags & ZIO_FLAG_SCAN_THREAD) {
2907169695Skan				dsl_scan_phys_t *scn_phys =
2908169695Skan				    &spa->spa_dsl_pool->dp_scan->scn_phys;
2909169695Skan				uint64_t *processed = &scn_phys->scn_processed;
2910169695Skan
2911169695Skan				/* XXX cleanup? */
2912169695Skan				if (vd->vdev_ops->vdev_op_leaf)
2913169695Skan					atomic_add_64(processed, psize);
2914169695Skan				vs->vs_scan_processed += psize;
2915169695Skan			}
2916169695Skan
2917169695Skan			if (flags & ZIO_FLAG_SELF_HEAL)
2918169695Skan				vs->vs_self_healed += psize;
2919169695Skan		}
2920169695Skan
2921169695Skan		vs->vs_ops[type]++;
2922169695Skan		vs->vs_bytes[type] += psize;
2923169695Skan
2924169695Skan		mutex_exit(&vd->vdev_stat_lock);
2925169695Skan		return;
2926169695Skan	}
2927169695Skan
2928169695Skan	if (flags & ZIO_FLAG_SPECULATIVE)
2929169695Skan		return;
2930169695Skan
2931169695Skan	/*
2932169695Skan	 * If this is an I/O error that is going to be retried, then ignore the
2933169695Skan	 * error.  Otherwise, the user may interpret B_FAILFAST I/O errors as
2934169695Skan	 * hard errors, when in reality they can happen for any number of
2935169695Skan	 * innocuous reasons (bus resets, MPxIO link failure, etc).
2936169695Skan	 */
2937169695Skan	if (zio->io_error == EIO &&
2938169695Skan	    !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2939169695Skan		return;
2940169695Skan
2941169695Skan	/*
2942169695Skan	 * Intent logs writes won't propagate their error to the root
2943169695Skan	 * I/O so don't mark these types of failures as pool-level
2944169695Skan	 * errors.
2945169695Skan	 */
2946169695Skan	if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2947169695Skan		return;
2948169695Skan
2949169695Skan	mutex_enter(&vd->vdev_stat_lock);
2950169695Skan	if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2951169695Skan		if (zio->io_error == ECKSUM)
2952169695Skan			vs->vs_checksum_errors++;
2953169695Skan		else
2954169695Skan			vs->vs_read_errors++;
2955169695Skan	}
2956169695Skan	if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2957169695Skan		vs->vs_write_errors++;
2958169695Skan	mutex_exit(&vd->vdev_stat_lock);
2959169695Skan
2960169695Skan	if (type == ZIO_TYPE_WRITE && txg != 0 &&
2961169695Skan	    (!(flags & ZIO_FLAG_IO_REPAIR) ||
2962169695Skan	    (flags & ZIO_FLAG_SCAN_THREAD) ||
2963169695Skan	    spa->spa_claiming)) {
2964169695Skan		/*
2965169695Skan		 * This is either a normal write (not a repair), or it's
2966169695Skan		 * a repair induced by the scrub thread, or it's a repair
2967169695Skan		 * made by zil_claim() during spa_load() in the first txg.
2968169695Skan		 * In the normal case, we commit the DTL change in the same
2969169695Skan		 * txg as the block was born.  In the scrub-induced repair
2970169695Skan		 * case, we know that scrubs run in first-pass syncing context,
2971169695Skan		 * so we commit the DTL change in spa_syncing_txg(spa).
2972169695Skan		 * In the zil_claim() case, we commit in spa_first_txg(spa).
2973169695Skan		 *
2974169695Skan		 * We currently do not make DTL entries for failed spontaneous
2975169695Skan		 * self-healing writes triggered by normal (non-scrubbing)
2976169695Skan		 * reads, because we have no transactional context in which to
2977169695Skan		 * do so -- and it's not clear that it'd be desirable anyway.
2978169695Skan		 */
2979169695Skan		if (vd->vdev_ops->vdev_op_leaf) {
2980169695Skan			uint64_t commit_txg = txg;
2981169695Skan			if (flags & ZIO_FLAG_SCAN_THREAD) {
2982169695Skan				ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2983169695Skan				ASSERT(spa_sync_pass(spa) == 1);
2984169695Skan				vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2985169695Skan				commit_txg = spa_syncing_txg(spa);
2986169695Skan			} else if (spa->spa_claiming) {
2987169695Skan				ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2988169695Skan				commit_txg = spa_first_txg(spa);
2989169695Skan			}
2990169695Skan			ASSERT(commit_txg >= spa_syncing_txg(spa));
2991169695Skan			if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2992169695Skan				return;
2993169695Skan			for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2994169695Skan				vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2995169695Skan			vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2996169695Skan		}
2997169695Skan		if (vd != rvd)
2998169695Skan			vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2999169695Skan	}
3000169695Skan}
3001169695Skan
3002169695Skan/*
3003169695Skan * Update the in-core space usage stats for this vdev, its metaslab class,
3004169695Skan * and the root vdev.
3005169695Skan */
3006169695Skanvoid
3007169695Skanvdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3008169695Skan    int64_t space_delta)
3009169695Skan{
3010169695Skan	int64_t dspace_delta = space_delta;
3011169695Skan	spa_t *spa = vd->vdev_spa;
3012169695Skan	vdev_t *rvd = spa->spa_root_vdev;
3013169695Skan	metaslab_group_t *mg = vd->vdev_mg;
3014169695Skan	metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3015169695Skan
3016169695Skan	ASSERT(vd == vd->vdev_top);
3017169695Skan
3018169695Skan	/*
3019169695Skan	 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3020169695Skan	 * factor.  We must calculate this here and not at the root vdev
3021169695Skan	 * because the root vdev's psize-to-asize is simply the max of its
3022169695Skan	 * childrens', thus not accurate enough for us.
3023169695Skan	 */
3024169695Skan	ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3025169695Skan	ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3026169695Skan	dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3027169695Skan	    vd->vdev_deflate_ratio;
3028169695Skan
3029169695Skan	mutex_enter(&vd->vdev_stat_lock);
3030169695Skan	vd->vdev_stat.vs_alloc += alloc_delta;
3031169695Skan	vd->vdev_stat.vs_space += space_delta;
3032169695Skan	vd->vdev_stat.vs_dspace += dspace_delta;
3033169695Skan	mutex_exit(&vd->vdev_stat_lock);
3034169695Skan
3035169695Skan	if (mc == spa_normal_class(spa)) {
3036169695Skan		mutex_enter(&rvd->vdev_stat_lock);
3037169695Skan		rvd->vdev_stat.vs_alloc += alloc_delta;
3038169695Skan		rvd->vdev_stat.vs_space += space_delta;
3039169695Skan		rvd->vdev_stat.vs_dspace += dspace_delta;
3040169695Skan		mutex_exit(&rvd->vdev_stat_lock);
3041169695Skan	}
3042169695Skan
3043169695Skan	if (mc != NULL) {
3044169695Skan		ASSERT(rvd == vd->vdev_parent);
3045169695Skan		ASSERT(vd->vdev_ms_count != 0);
3046169695Skan
3047169695Skan		metaslab_class_space_update(mc,
3048169695Skan		    alloc_delta, defer_delta, space_delta, dspace_delta);
3049169695Skan	}
3050169695Skan}
3051169695Skan
3052169695Skan/*
3053169695Skan * Mark a top-level vdev's config as dirty, placing it on the dirty list
3054169695Skan * so that it will be written out next time the vdev configuration is synced.
3055169695Skan * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3056169695Skan */
3057169695Skanvoid
3058169695Skanvdev_config_dirty(vdev_t *vd)
3059169695Skan{
3060169695Skan	spa_t *spa = vd->vdev_spa;
3061169695Skan	vdev_t *rvd = spa->spa_root_vdev;
3062169695Skan	int c;
3063169695Skan
3064169695Skan	ASSERT(spa_writeable(spa));
3065169695Skan
3066169695Skan	/*
3067169695Skan	 * If this is an aux vdev (as with l2cache and spare devices), then we
3068169695Skan	 * update the vdev config manually and set the sync flag.
3069169695Skan	 */
3070169695Skan	if (vd->vdev_aux != NULL) {
3071169695Skan		spa_aux_vdev_t *sav = vd->vdev_aux;
3072169695Skan		nvlist_t **aux;
3073169695Skan		uint_t naux;
3074169695Skan
3075169695Skan		for (c = 0; c < sav->sav_count; c++) {
3076169695Skan			if (sav->sav_vdevs[c] == vd)
3077169695Skan				break;
3078169695Skan		}
3079169695Skan
3080169695Skan		if (c == sav->sav_count) {
3081169695Skan			/*
3082169695Skan			 * We're being removed.  There's nothing more to do.
3083169695Skan			 */
3084169695Skan			ASSERT(sav->sav_sync == B_TRUE);
3085169695Skan			return;
3086169695Skan		}
3087169695Skan
3088169695Skan		sav->sav_sync = B_TRUE;
3089169695Skan
3090169695Skan		if (nvlist_lookup_nvlist_array(sav->sav_config,
3091169695Skan		    ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3092169695Skan			VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3093169695Skan			    ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3094169695Skan		}
3095169695Skan
3096169695Skan		ASSERT(c < naux);
3097169695Skan
3098169695Skan		/*
3099169695Skan		 * Setting the nvlist in the middle if the array is a little
3100169695Skan		 * sketchy, but it will work.
3101169695Skan		 */
3102169695Skan		nvlist_free(aux[c]);
3103169695Skan		aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3104169695Skan
3105169695Skan		return;
3106169695Skan	}
3107169695Skan
3108169695Skan	/*
3109169695Skan	 * The dirty list is protected by the SCL_CONFIG lock.  The caller
3110169695Skan	 * must either hold SCL_CONFIG as writer, or must be the sync thread
3111169695Skan	 * (which holds SCL_CONFIG as reader).  There's only one sync thread,
3112169695Skan	 * so this is sufficient to ensure mutual exclusion.
3113169695Skan	 */
3114169695Skan	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3115169695Skan	    (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3116169695Skan	    spa_config_held(spa, SCL_CONFIG, RW_READER)));
3117169695Skan
3118169695Skan	if (vd == rvd) {
3119169695Skan		for (c = 0; c < rvd->vdev_children; c++)
3120169695Skan			vdev_config_dirty(rvd->vdev_child[c]);
3121169695Skan	} else {
3122169695Skan		ASSERT(vd == vd->vdev_top);
3123169695Skan
3124169695Skan		if (!list_link_active(&vd->vdev_config_dirty_node) &&
3125169695Skan		    !vd->vdev_ishole)
3126169695Skan			list_insert_head(&spa->spa_config_dirty_list, vd);
3127169695Skan	}
3128169695Skan}
3129169695Skan
3130169695Skanvoid
3131169695Skanvdev_config_clean(vdev_t *vd)
3132169695Skan{
3133169695Skan	spa_t *spa = vd->vdev_spa;
3134169695Skan
3135169695Skan	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3136169695Skan	    (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3137169695Skan	    spa_config_held(spa, SCL_CONFIG, RW_READER)));
3138169695Skan
3139169695Skan	ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3140169695Skan	list_remove(&spa->spa_config_dirty_list, vd);
3141169695Skan}
3142169695Skan
3143169695Skan/*
3144169695Skan * Mark a top-level vdev's state as dirty, so that the next pass of
3145169695Skan * spa_sync() can convert this into vdev_config_dirty().  We distinguish
3146169695Skan * the state changes from larger config changes because they require
3147169695Skan * much less locking, and are often needed for administrative actions.
3148169695Skan */
3149169695Skanvoid
3150169695Skanvdev_state_dirty(vdev_t *vd)
3151169695Skan{
3152169695Skan	spa_t *spa = vd->vdev_spa;
3153169695Skan
3154169695Skan	ASSERT(spa_writeable(spa));
3155169695Skan	ASSERT(vd == vd->vdev_top);
3156169695Skan
3157169695Skan	/*
3158169695Skan	 * The state list is protected by the SCL_STATE lock.  The caller
3159169695Skan	 * must either hold SCL_STATE as writer, or must be the sync thread
3160169695Skan	 * (which holds SCL_STATE as reader).  There's only one sync thread,
3161169695Skan	 * so this is sufficient to ensure mutual exclusion.
3162169695Skan	 */
3163169695Skan	ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3164169695Skan	    (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3165169695Skan	    spa_config_held(spa, SCL_STATE, RW_READER)));
3166169695Skan
3167169695Skan	if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3168169695Skan		list_insert_head(&spa->spa_state_dirty_list, vd);
3169169695Skan}
3170169695Skan
3171169695Skanvoid
3172169695Skanvdev_state_clean(vdev_t *vd)
3173169695Skan{
3174169695Skan	spa_t *spa = vd->vdev_spa;
3175169695Skan
3176169695Skan	ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3177169695Skan	    (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3178169695Skan	    spa_config_held(spa, SCL_STATE, RW_READER)));
3179169695Skan
3180169695Skan	ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3181169695Skan	list_remove(&spa->spa_state_dirty_list, vd);
3182169695Skan}
3183169695Skan
3184169695Skan/*
3185169695Skan * Propagate vdev state up from children to parent.
3186169695Skan */
3187169695Skanvoid
3188169695Skanvdev_propagate_state(vdev_t *vd)
3189169695Skan{
3190169695Skan	spa_t *spa = vd->vdev_spa;
3191169695Skan	vdev_t *rvd = spa->spa_root_vdev;
3192169695Skan	int degraded = 0, faulted = 0;
3193169695Skan	int corrupted = 0;
3194169695Skan	vdev_t *child;
3195169695Skan
3196169695Skan	if (vd->vdev_children > 0) {
3197169695Skan		for (int c = 0; c < vd->vdev_children; c++) {
3198169695Skan			child = vd->vdev_child[c];
3199169695Skan
3200169695Skan			/*
3201169695Skan			 * Don't factor holes into the decision.
3202169695Skan			 */
3203169695Skan			if (child->vdev_ishole)
3204169695Skan				continue;
3205169695Skan
3206169695Skan			if (!vdev_readable(child) ||
3207169695Skan			    (!vdev_writeable(child) && spa_writeable(spa))) {
3208169695Skan				/*
3209169695Skan				 * Root special: if there is a top-level log
3210169695Skan				 * device, treat the root vdev as if it were
3211169695Skan				 * degraded.
3212169695Skan				 */
3213169695Skan				if (child->vdev_islog && vd == rvd)
3214169695Skan					degraded++;
3215169695Skan				else
3216169695Skan					faulted++;
3217169695Skan			} else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3218169695Skan				degraded++;
3219169695Skan			}
3220169695Skan
3221169695Skan			if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3222169695Skan				corrupted++;
3223169695Skan		}
3224169695Skan
3225169695Skan		vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3226169695Skan
3227169695Skan		/*
3228169695Skan		 * Root special: if there is a top-level vdev that cannot be
3229169695Skan		 * opened due to corrupted metadata, then propagate the root
3230169695Skan		 * vdev's aux state as 'corrupt' rather than 'insufficient
3231169695Skan		 * replicas'.
3232169695Skan		 */
3233169695Skan		if (corrupted && vd == rvd &&
3234169695Skan		    rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3235169695Skan			vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3236169695Skan			    VDEV_AUX_CORRUPT_DATA);
3237169695Skan	}
3238169695Skan
3239169695Skan	if (vd->vdev_parent)
3240169695Skan		vdev_propagate_state(vd->vdev_parent);
3241169695Skan}
3242169695Skan
3243169695Skan/*
3244169695Skan * Set a vdev's state.  If this is during an open, we don't update the parent
3245169695Skan * state, because we're in the process of opening children depth-first.
3246169695Skan * Otherwise, we propagate the change to the parent.
3247169695Skan *
3248169695Skan * If this routine places a device in a faulted state, an appropriate ereport is
3249169695Skan * generated.
3250169695Skan */
3251169695Skanvoid
3252169695Skanvdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3253169695Skan{
3254169695Skan	uint64_t save_state;
3255169695Skan	spa_t *spa = vd->vdev_spa;
3256169695Skan
3257169695Skan	if (state == vd->vdev_state) {
3258169695Skan		vd->vdev_stat.vs_aux = aux;
3259169695Skan		return;
3260169695Skan	}
3261169695Skan
3262169695Skan	save_state = vd->vdev_state;
3263169695Skan
3264169695Skan	vd->vdev_state = state;
3265169695Skan	vd->vdev_stat.vs_aux = aux;
3266169695Skan
3267169695Skan	/*
3268169695Skan	 * If we are setting the vdev state to anything but an open state, then
3269169695Skan	 * always close the underlying device unless the device has requested
3270169695Skan	 * a delayed close (i.e. we're about to remove or fault the device).
3271169695Skan	 * Otherwise, we keep accessible but invalid devices open forever.
3272169695Skan	 * We don't call vdev_close() itself, because that implies some extra
3273169695Skan	 * checks (offline, etc) that we don't want here.  This is limited to
3274169695Skan	 * leaf devices, because otherwise closing the device will affect other
3275169695Skan	 * children.
3276169695Skan	 */
3277169695Skan	if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3278169695Skan	    vd->vdev_ops->vdev_op_leaf)
3279169695Skan		vd->vdev_ops->vdev_op_close(vd);
3280169695Skan
3281169695Skan	/*
3282169695Skan	 * If we have brought this vdev back into service, we need
3283169695Skan	 * to notify fmd so that it can gracefully repair any outstanding
3284169695Skan	 * cases due to a missing device.  We do this in all cases, even those
3285169695Skan	 * that probably don't correlate to a repaired fault.  This is sure to
3286169695Skan	 * catch all cases, and we let the zfs-retire agent sort it out.  If
3287169695Skan	 * this is a transient state it's OK, as the retire agent will
3288169695Skan	 * double-check the state of the vdev before repairing it.
3289169695Skan	 */
3290169695Skan	if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
3291169695Skan	    vd->vdev_prevstate != state)
3292169695Skan		zfs_post_state_change(spa, vd);
3293169695Skan
3294169695Skan	if (vd->vdev_removed &&
3295169695Skan	    state == VDEV_STATE_CANT_OPEN &&
3296169695Skan	    (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3297169695Skan		/*
3298169695Skan		 * If the previous state is set to VDEV_STATE_REMOVED, then this
3299169695Skan		 * device was previously marked removed and someone attempted to
3300169695Skan		 * reopen it.  If this failed due to a nonexistent device, then
3301169695Skan		 * keep the device in the REMOVED state.  We also let this be if
3302169695Skan		 * it is one of our special test online cases, which is only
3303169695Skan		 * attempting to online the device and shouldn't generate an FMA
3304169695Skan		 * fault.
3305169695Skan		 */
3306169695Skan		vd->vdev_state = VDEV_STATE_REMOVED;
3307169695Skan		vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3308169695Skan	} else if (state == VDEV_STATE_REMOVED) {
3309169695Skan		vd->vdev_removed = B_TRUE;
3310169695Skan	} else if (state == VDEV_STATE_CANT_OPEN) {
3311169695Skan		/*
3312169695Skan		 * If we fail to open a vdev during an import or recovery, we
3313169695Skan		 * mark it as "not available", which signifies that it was
3314169695Skan		 * never there to begin with.  Failure to open such a device
3315169695Skan		 * is not considered an error.
3316169695Skan		 */
3317169695Skan		if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3318169695Skan		    spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3319169695Skan		    vd->vdev_ops->vdev_op_leaf)
3320169695Skan			vd->vdev_not_present = 1;
3321169695Skan
3322169695Skan		/*
3323169695Skan		 * Post the appropriate ereport.  If the 'prevstate' field is
3324169695Skan		 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3325169695Skan		 * that this is part of a vdev_reopen().  In this case, we don't
3326169695Skan		 * want to post the ereport if the device was already in the
3327169695Skan		 * CANT_OPEN state beforehand.
3328169695Skan		 *
3329169695Skan		 * If the 'checkremove' flag is set, then this is an attempt to
3330169695Skan		 * online the device in response to an insertion event.  If we
3331169695Skan		 * hit this case, then we have detected an insertion event for a
3332169695Skan		 * faulted or offline device that wasn't in the removed state.
3333169695Skan		 * In this scenario, we don't post an ereport because we are
3334169695Skan		 * about to replace the device, or attempt an online with
3335169695Skan		 * vdev_forcefault, which will generate the fault for us.
3336169695Skan		 */
3337169695Skan		if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3338169695Skan		    !vd->vdev_not_present && !vd->vdev_checkremove &&
3339169695Skan		    vd != spa->spa_root_vdev) {
3340169695Skan			const char *class;
3341169695Skan
3342169695Skan			switch (aux) {
3343169695Skan			case VDEV_AUX_OPEN_FAILED:
3344169695Skan				class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3345169695Skan				break;
3346169695Skan			case VDEV_AUX_CORRUPT_DATA:
3347169695Skan				class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3348169695Skan				break;
3349169695Skan			case VDEV_AUX_NO_REPLICAS:
3350169695Skan				class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3351169695Skan				break;
3352169695Skan			case VDEV_AUX_BAD_GUID_SUM:
3353169695Skan				class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3354169695Skan				break;
3355169695Skan			case VDEV_AUX_TOO_SMALL:
3356169695Skan				class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3357169695Skan				break;
3358169695Skan			case VDEV_AUX_BAD_LABEL:
3359169695Skan				class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3360169695Skan				break;
3361169695Skan			default:
3362169695Skan				class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3363169695Skan			}
3364169695Skan
3365169695Skan			zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3366169695Skan		}
3367169695Skan
3368169695Skan		/* Erase any notion of persistent removed state */
3369169695Skan		vd->vdev_removed = B_FALSE;
3370169695Skan	} else {
3371169695Skan		vd->vdev_removed = B_FALSE;
3372169695Skan	}
3373169695Skan
3374169695Skan	if (!isopen && vd->vdev_parent)
3375169695Skan		vdev_propagate_state(vd->vdev_parent);
3376169695Skan}
3377169695Skan
3378169695Skan/*
3379169695Skan * Check the vdev configuration to ensure that it's capable of supporting
3380169695Skan * a root pool.
3381169695Skan *
3382169695Skan * On Solaris, we do not support RAID-Z or partial configuration.  In
3383169695Skan * addition, only a single top-level vdev is allowed and none of the
3384169695Skan * leaves can be wholedisks.
3385169695Skan *
3386169695Skan * For FreeBSD, we can boot from any configuration. There is a
3387169695Skan * limitation that the boot filesystem must be either uncompressed or
3388169695Skan * compresses with lzjb compression but I'm not sure how to enforce
3389169695Skan * that here.
3390169695Skan */
3391169695Skanboolean_t
3392169695Skanvdev_is_bootable(vdev_t *vd)
3393169695Skan{
3394169695Skan#ifdef sun
3395169695Skan	if (!vd->vdev_ops->vdev_op_leaf) {
3396169695Skan		char *vdev_type = vd->vdev_ops->vdev_op_type;
3397169695Skan
3398169695Skan		if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3399169695Skan		    vd->vdev_children > 1) {
3400169695Skan			return (B_FALSE);
3401169695Skan		} else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3402169695Skan		    strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3403169695Skan			return (B_FALSE);
3404169695Skan		}
3405169695Skan	} else if (vd->vdev_wholedisk == 1) {
3406169695Skan		return (B_FALSE);
3407169695Skan	}
3408169695Skan
3409169695Skan	for (int c = 0; c < vd->vdev_children; c++) {
3410169695Skan		if (!vdev_is_bootable(vd->vdev_child[c]))
3411169695Skan			return (B_FALSE);
3412169695Skan	}
3413169695Skan#endif	/* sun */
3414169695Skan	return (B_TRUE);
3415169695Skan}
3416169695Skan
3417169695Skan/*
3418169695Skan * Load the state from the original vdev tree (ovd) which
3419169695Skan * we've retrieved from the MOS config object. If the original
3420169695Skan * vdev was offline or faulted then we transfer that state to the
3421169695Skan * device in the current vdev tree (nvd).
3422169695Skan */
3423169695Skanvoid
3424169695Skanvdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3425169695Skan{
3426169695Skan	spa_t *spa = nvd->vdev_spa;
3427169695Skan
3428169695Skan	ASSERT(nvd->vdev_top->vdev_islog);
3429169695Skan	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3430169695Skan	ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3431169695Skan
3432169695Skan	for (int c = 0; c < nvd->vdev_children; c++)
3433169695Skan		vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3434169695Skan
3435169695Skan	if (nvd->vdev_ops->vdev_op_leaf) {
3436169695Skan		/*
3437169695Skan		 * Restore the persistent vdev state
3438169695Skan		 */
3439169695Skan		nvd->vdev_offline = ovd->vdev_offline;
3440169695Skan		nvd->vdev_faulted = ovd->vdev_faulted;
3441169695Skan		nvd->vdev_degraded = ovd->vdev_degraded;
3442169695Skan		nvd->vdev_removed = ovd->vdev_removed;
3443169695Skan	}
3444169695Skan}
3445169695Skan
3446169695Skan/*
3447169695Skan * Determine if a log device has valid content.  If the vdev was
3448169695Skan * removed or faulted in the MOS config then we know that
3449169695Skan * the content on the log device has already been written to the pool.
3450169695Skan */
3451169695Skanboolean_t
3452169695Skanvdev_log_state_valid(vdev_t *vd)
3453169695Skan{
3454169695Skan	if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3455169695Skan	    !vd->vdev_removed)
3456169695Skan		return (B_TRUE);
3457169695Skan
3458169695Skan	for (int c = 0; c < vd->vdev_children; c++)
3459169695Skan		if (vdev_log_state_valid(vd->vdev_child[c]))
3460169695Skan			return (B_TRUE);
3461169695Skan
3462169695Skan	return (B_FALSE);
3463169695Skan}
3464169695Skan
3465169695Skan/*
3466169695Skan * Expand a vdev if possible.
3467169695Skan */
3468169695Skanvoid
3469169695Skanvdev_expand(vdev_t *vd, uint64_t txg)
3470169695Skan{
3471169695Skan	ASSERT(vd->vdev_top == vd);
3472169695Skan	ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3473169695Skan
3474169695Skan	if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3475169695Skan		VERIFY(vdev_metaslab_init(vd, txg) == 0);
3476169695Skan		vdev_config_dirty(vd);
3477169695Skan	}
3478169695Skan}
3479169695Skan
3480169695Skan/*
3481169695Skan * Split a vdev.
3482169695Skan */
3483169695Skanvoid
3484169695Skanvdev_split(vdev_t *vd)
3485169695Skan{
3486169695Skan	vdev_t *cvd, *pvd = vd->vdev_parent;
3487169695Skan
3488169695Skan	vdev_remove_child(pvd, vd);
3489169695Skan	vdev_compact_children(pvd);
3490169695Skan
3491169695Skan	cvd = pvd->vdev_child[0];
3492169695Skan	if (pvd->vdev_children == 1) {
3493169695Skan		vdev_remove_parent(cvd);
3494169695Skan		cvd->vdev_splitting = B_TRUE;
3495169695Skan	}
3496169695Skan	vdev_propagate_state(cvd);
3497169695Skan}
3498169695Skan
3499169695Skanvoid
3500169695Skanvdev_deadman(vdev_t *vd)
3501169695Skan{
3502169695Skan	for (int c = 0; c < vd->vdev_children; c++) {
3503169695Skan		vdev_t *cvd = vd->vdev_child[c];
3504169695Skan
3505169695Skan		vdev_deadman(cvd);
3506169695Skan	}
3507169695Skan
3508169695Skan	if (vd->vdev_ops->vdev_op_leaf) {
3509169695Skan		vdev_queue_t *vq = &vd->vdev_queue;
3510169695Skan
3511169695Skan		mutex_enter(&vq->vq_lock);
3512169695Skan		if (avl_numnodes(&vq->vq_active_tree) > 0) {
3513169695Skan			spa_t *spa = vd->vdev_spa;
3514169695Skan			zio_t *fio;
3515169695Skan			uint64_t delta;
3516169695Skan
3517169695Skan			/*
3518169695Skan			 * Look at the head of all the pending queues,
3519169695Skan			 * if any I/O has been outstanding for longer than
3520169695Skan			 * the spa_deadman_synctime we panic the system.
3521169695Skan			 */
3522169695Skan			fio = avl_first(&vq->vq_active_tree);
3523169695Skan			delta = gethrtime() - fio->io_timestamp;
3524169695Skan			if (delta > spa_deadman_synctime(spa)) {
3525169695Skan				zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3526169695Skan				    "delta %lluns, last io %lluns",
3527169695Skan				    fio->io_timestamp, delta,
3528169695Skan				    vq->vq_io_complete_ts);
3529169695Skan				fm_panic("I/O to pool '%s' appears to be "
3530169695Skan				    "hung on vdev guid %llu at '%s'.",
3531169695Skan				    spa_name(spa),
3532169695Skan				    (long long unsigned int) vd->vdev_guid,
3533169695Skan				    vd->vdev_path);
3534169695Skan			}
3535169695Skan		}
3536169695Skan		mutex_exit(&vq->vq_lock);
3537169695Skan	}
3538169695Skan}
3539169695Skan