xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev.c (revision bb8b5132cbf0e1d3fd47148d13c12ace41e302bb)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa.h>
32 #include <sys/spa_impl.h>
33 #include <sys/dmu.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/uberblock_impl.h>
37 #include <sys/metaslab.h>
38 #include <sys/metaslab_impl.h>
39 #include <sys/space_map.h>
40 #include <sys/zio.h>
41 #include <sys/zap.h>
42 #include <sys/fs/zfs.h>
43 
44 /*
45  * Virtual device management.
46  */
47 
48 static vdev_ops_t *vdev_ops_table[] = {
49 	&vdev_root_ops,
50 	&vdev_raidz_ops,
51 	&vdev_mirror_ops,
52 	&vdev_replacing_ops,
53 	&vdev_spare_ops,
54 	&vdev_disk_ops,
55 	&vdev_file_ops,
56 	&vdev_missing_ops,
57 	NULL
58 };
59 
60 /* maximum scrub/resilver I/O queue */
61 int zfs_scrub_limit = 70;
62 
63 /*
64  * Given a vdev type, return the appropriate ops vector.
65  */
66 static vdev_ops_t *
67 vdev_getops(const char *type)
68 {
69 	vdev_ops_t *ops, **opspp;
70 
71 	for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
72 		if (strcmp(ops->vdev_op_type, type) == 0)
73 			break;
74 
75 	return (ops);
76 }
77 
78 /*
79  * Default asize function: return the MAX of psize with the asize of
80  * all children.  This is what's used by anything other than RAID-Z.
81  */
82 uint64_t
83 vdev_default_asize(vdev_t *vd, uint64_t psize)
84 {
85 	uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
86 	uint64_t csize;
87 	uint64_t c;
88 
89 	for (c = 0; c < vd->vdev_children; c++) {
90 		csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
91 		asize = MAX(asize, csize);
92 	}
93 
94 	return (asize);
95 }
96 
97 /*
98  * Get the replaceable or attachable device size.
99  * If the parent is a mirror or raidz, the replaceable size is the minimum
100  * psize of all its children. For the rest, just return our own psize.
101  *
102  * e.g.
103  *			psize	rsize
104  * root			-	-
105  *	mirror/raidz	-	-
106  *	    disk1	20g	20g
107  *	    disk2 	40g	20g
108  *	disk3 		80g	80g
109  */
110 uint64_t
111 vdev_get_rsize(vdev_t *vd)
112 {
113 	vdev_t *pvd, *cvd;
114 	uint64_t c, rsize;
115 
116 	pvd = vd->vdev_parent;
117 
118 	/*
119 	 * If our parent is NULL or the root, just return our own psize.
120 	 */
121 	if (pvd == NULL || pvd->vdev_parent == NULL)
122 		return (vd->vdev_psize);
123 
124 	rsize = 0;
125 
126 	for (c = 0; c < pvd->vdev_children; c++) {
127 		cvd = pvd->vdev_child[c];
128 		rsize = MIN(rsize - 1, cvd->vdev_psize - 1) + 1;
129 	}
130 
131 	return (rsize);
132 }
133 
134 vdev_t *
135 vdev_lookup_top(spa_t *spa, uint64_t vdev)
136 {
137 	vdev_t *rvd = spa->spa_root_vdev;
138 
139 	if (vdev < rvd->vdev_children)
140 		return (rvd->vdev_child[vdev]);
141 
142 	return (NULL);
143 }
144 
145 vdev_t *
146 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
147 {
148 	int c;
149 	vdev_t *mvd;
150 
151 	if (vd->vdev_guid == guid)
152 		return (vd);
153 
154 	for (c = 0; c < vd->vdev_children; c++)
155 		if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
156 		    NULL)
157 			return (mvd);
158 
159 	return (NULL);
160 }
161 
162 void
163 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
164 {
165 	size_t oldsize, newsize;
166 	uint64_t id = cvd->vdev_id;
167 	vdev_t **newchild;
168 
169 	ASSERT(spa_config_held(cvd->vdev_spa, RW_WRITER));
170 	ASSERT(cvd->vdev_parent == NULL);
171 
172 	cvd->vdev_parent = pvd;
173 
174 	if (pvd == NULL)
175 		return;
176 
177 	ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
178 
179 	oldsize = pvd->vdev_children * sizeof (vdev_t *);
180 	pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
181 	newsize = pvd->vdev_children * sizeof (vdev_t *);
182 
183 	newchild = kmem_zalloc(newsize, KM_SLEEP);
184 	if (pvd->vdev_child != NULL) {
185 		bcopy(pvd->vdev_child, newchild, oldsize);
186 		kmem_free(pvd->vdev_child, oldsize);
187 	}
188 
189 	pvd->vdev_child = newchild;
190 	pvd->vdev_child[id] = cvd;
191 
192 	cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
193 	ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
194 
195 	/*
196 	 * Walk up all ancestors to update guid sum.
197 	 */
198 	for (; pvd != NULL; pvd = pvd->vdev_parent)
199 		pvd->vdev_guid_sum += cvd->vdev_guid_sum;
200 
201 	if (cvd->vdev_ops->vdev_op_leaf)
202 		cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit;
203 }
204 
205 void
206 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
207 {
208 	int c;
209 	uint_t id = cvd->vdev_id;
210 
211 	ASSERT(cvd->vdev_parent == pvd);
212 
213 	if (pvd == NULL)
214 		return;
215 
216 	ASSERT(id < pvd->vdev_children);
217 	ASSERT(pvd->vdev_child[id] == cvd);
218 
219 	pvd->vdev_child[id] = NULL;
220 	cvd->vdev_parent = NULL;
221 
222 	for (c = 0; c < pvd->vdev_children; c++)
223 		if (pvd->vdev_child[c])
224 			break;
225 
226 	if (c == pvd->vdev_children) {
227 		kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
228 		pvd->vdev_child = NULL;
229 		pvd->vdev_children = 0;
230 	}
231 
232 	/*
233 	 * Walk up all ancestors to update guid sum.
234 	 */
235 	for (; pvd != NULL; pvd = pvd->vdev_parent)
236 		pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
237 
238 	if (cvd->vdev_ops->vdev_op_leaf)
239 		cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit;
240 }
241 
242 /*
243  * Remove any holes in the child array.
244  */
245 void
246 vdev_compact_children(vdev_t *pvd)
247 {
248 	vdev_t **newchild, *cvd;
249 	int oldc = pvd->vdev_children;
250 	int newc, c;
251 
252 	ASSERT(spa_config_held(pvd->vdev_spa, RW_WRITER));
253 
254 	for (c = newc = 0; c < oldc; c++)
255 		if (pvd->vdev_child[c])
256 			newc++;
257 
258 	newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
259 
260 	for (c = newc = 0; c < oldc; c++) {
261 		if ((cvd = pvd->vdev_child[c]) != NULL) {
262 			newchild[newc] = cvd;
263 			cvd->vdev_id = newc++;
264 		}
265 	}
266 
267 	kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
268 	pvd->vdev_child = newchild;
269 	pvd->vdev_children = newc;
270 }
271 
272 /*
273  * Allocate and minimally initialize a vdev_t.
274  */
275 static vdev_t *
276 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
277 {
278 	vdev_t *vd;
279 
280 	vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
281 
282 	if (spa->spa_root_vdev == NULL) {
283 		ASSERT(ops == &vdev_root_ops);
284 		spa->spa_root_vdev = vd;
285 	}
286 
287 	if (guid == 0) {
288 		if (spa->spa_root_vdev == vd) {
289 			/*
290 			 * The root vdev's guid will also be the pool guid,
291 			 * which must be unique among all pools.
292 			 */
293 			while (guid == 0 || spa_guid_exists(guid, 0))
294 				guid = spa_get_random(-1ULL);
295 		} else {
296 			/*
297 			 * Any other vdev's guid must be unique within the pool.
298 			 */
299 			while (guid == 0 ||
300 			    spa_guid_exists(spa_guid(spa), guid))
301 				guid = spa_get_random(-1ULL);
302 		}
303 		ASSERT(!spa_guid_exists(spa_guid(spa), guid));
304 	}
305 
306 	vd->vdev_spa = spa;
307 	vd->vdev_id = id;
308 	vd->vdev_guid = guid;
309 	vd->vdev_guid_sum = guid;
310 	vd->vdev_ops = ops;
311 	vd->vdev_state = VDEV_STATE_CLOSED;
312 
313 	mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
314 	mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
315 	space_map_create(&vd->vdev_dtl_map, 0, -1ULL, 0, &vd->vdev_dtl_lock);
316 	space_map_create(&vd->vdev_dtl_scrub, 0, -1ULL, 0, &vd->vdev_dtl_lock);
317 	txg_list_create(&vd->vdev_ms_list,
318 	    offsetof(struct metaslab, ms_txg_node));
319 	txg_list_create(&vd->vdev_dtl_list,
320 	    offsetof(struct vdev, vdev_dtl_node));
321 	vd->vdev_stat.vs_timestamp = gethrtime();
322 	vdev_queue_init(vd);
323 	vdev_cache_init(vd);
324 
325 	return (vd);
326 }
327 
328 /*
329  * Allocate a new vdev.  The 'alloctype' is used to control whether we are
330  * creating a new vdev or loading an existing one - the behavior is slightly
331  * different for each case.
332  */
333 int
334 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
335     int alloctype)
336 {
337 	vdev_ops_t *ops;
338 	char *type;
339 	uint64_t guid = 0, islog, nparity;
340 	vdev_t *vd;
341 
342 	ASSERT(spa_config_held(spa, RW_WRITER));
343 
344 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
345 		return (EINVAL);
346 
347 	if ((ops = vdev_getops(type)) == NULL)
348 		return (EINVAL);
349 
350 	/*
351 	 * If this is a load, get the vdev guid from the nvlist.
352 	 * Otherwise, vdev_alloc_common() will generate one for us.
353 	 */
354 	if (alloctype == VDEV_ALLOC_LOAD) {
355 		uint64_t label_id;
356 
357 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
358 		    label_id != id)
359 			return (EINVAL);
360 
361 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
362 			return (EINVAL);
363 	} else if (alloctype == VDEV_ALLOC_SPARE) {
364 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
365 			return (EINVAL);
366 	}
367 
368 	/*
369 	 * The first allocated vdev must be of type 'root'.
370 	 */
371 	if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
372 		return (EINVAL);
373 
374 	/*
375 	 * Determine whether we're a log vdev.
376 	 */
377 	islog = 0;
378 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
379 	if (islog && spa_version(spa) < ZFS_VERSION_SLOGS)
380 		return (ENOTSUP);
381 
382 	/*
383 	 * Set the nparity property for RAID-Z vdevs.
384 	 */
385 	nparity = -1ULL;
386 	if (ops == &vdev_raidz_ops) {
387 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
388 		    &nparity) == 0) {
389 			/*
390 			 * Currently, we can only support 2 parity devices.
391 			 */
392 			if (nparity == 0 || nparity > 2)
393 				return (EINVAL);
394 			/*
395 			 * Older versions can only support 1 parity device.
396 			 */
397 			if (nparity == 2 &&
398 			    spa_version(spa) < SPA_VERSION_RAID6)
399 				return (ENOTSUP);
400 		} else {
401 			/*
402 			 * We require the parity to be specified for SPAs that
403 			 * support multiple parity levels.
404 			 */
405 			if (spa_version(spa) >= SPA_VERSION_RAID6)
406 				return (EINVAL);
407 			/*
408 			 * Otherwise, we default to 1 parity device for RAID-Z.
409 			 */
410 			nparity = 1;
411 		}
412 	} else {
413 		nparity = 0;
414 	}
415 	ASSERT(nparity != -1ULL);
416 
417 	vd = vdev_alloc_common(spa, id, guid, ops);
418 
419 	vd->vdev_islog = islog;
420 	vd->vdev_nparity = nparity;
421 
422 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
423 		vd->vdev_path = spa_strdup(vd->vdev_path);
424 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
425 		vd->vdev_devid = spa_strdup(vd->vdev_devid);
426 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
427 	    &vd->vdev_physpath) == 0)
428 		vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
429 
430 	/*
431 	 * Set the whole_disk property.  If it's not specified, leave the value
432 	 * as -1.
433 	 */
434 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
435 	    &vd->vdev_wholedisk) != 0)
436 		vd->vdev_wholedisk = -1ULL;
437 
438 	/*
439 	 * Look for the 'not present' flag.  This will only be set if the device
440 	 * was not present at the time of import.
441 	 */
442 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
443 	    &vd->vdev_not_present);
444 
445 	/*
446 	 * Get the alignment requirement.
447 	 */
448 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
449 
450 	/*
451 	 * If we're a top-level vdev, try to load the allocation parameters.
452 	 */
453 	if (parent && !parent->vdev_parent && alloctype == VDEV_ALLOC_LOAD) {
454 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
455 		    &vd->vdev_ms_array);
456 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
457 		    &vd->vdev_ms_shift);
458 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
459 		    &vd->vdev_asize);
460 	}
461 
462 	/*
463 	 * If we're a leaf vdev, try to load the DTL object and other state.
464 	 */
465 	if (vd->vdev_ops->vdev_op_leaf && alloctype == VDEV_ALLOC_LOAD) {
466 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
467 		    &vd->vdev_dtl.smo_object);
468 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
469 		    &vd->vdev_offline);
470 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
471 		    &vd->vdev_unspare);
472 		/*
473 		 * When importing a pool, we want to ignore the persistent fault
474 		 * state, as the diagnosis made on another system may not be
475 		 * valid in the current context.
476 		 */
477 		if (spa->spa_load_state == SPA_LOAD_OPEN) {
478 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
479 			    &vd->vdev_faulted);
480 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
481 			    &vd->vdev_degraded);
482 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
483 			    &vd->vdev_removed);
484 		}
485 	}
486 
487 	/*
488 	 * Add ourselves to the parent's list of children.
489 	 */
490 	vdev_add_child(parent, vd);
491 
492 	*vdp = vd;
493 
494 	return (0);
495 }
496 
497 void
498 vdev_free(vdev_t *vd)
499 {
500 	int c;
501 	spa_t *spa = vd->vdev_spa;
502 
503 	/*
504 	 * vdev_free() implies closing the vdev first.  This is simpler than
505 	 * trying to ensure complicated semantics for all callers.
506 	 */
507 	vdev_close(vd);
508 
509 
510 	ASSERT(!list_link_active(&vd->vdev_dirty_node));
511 
512 	/*
513 	 * Free all children.
514 	 */
515 	for (c = 0; c < vd->vdev_children; c++)
516 		vdev_free(vd->vdev_child[c]);
517 
518 	ASSERT(vd->vdev_child == NULL);
519 	ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
520 
521 	/*
522 	 * Discard allocation state.
523 	 */
524 	if (vd == vd->vdev_top)
525 		vdev_metaslab_fini(vd);
526 
527 	ASSERT3U(vd->vdev_stat.vs_space, ==, 0);
528 	ASSERT3U(vd->vdev_stat.vs_dspace, ==, 0);
529 	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
530 
531 	/*
532 	 * Remove this vdev from its parent's child list.
533 	 */
534 	vdev_remove_child(vd->vdev_parent, vd);
535 
536 	ASSERT(vd->vdev_parent == NULL);
537 
538 	/*
539 	 * Clean up vdev structure.
540 	 */
541 	vdev_queue_fini(vd);
542 	vdev_cache_fini(vd);
543 
544 	if (vd->vdev_path)
545 		spa_strfree(vd->vdev_path);
546 	if (vd->vdev_devid)
547 		spa_strfree(vd->vdev_devid);
548 	if (vd->vdev_physpath)
549 		spa_strfree(vd->vdev_physpath);
550 
551 	if (vd->vdev_isspare)
552 		spa_spare_remove(vd);
553 
554 	txg_list_destroy(&vd->vdev_ms_list);
555 	txg_list_destroy(&vd->vdev_dtl_list);
556 	mutex_enter(&vd->vdev_dtl_lock);
557 	space_map_unload(&vd->vdev_dtl_map);
558 	space_map_destroy(&vd->vdev_dtl_map);
559 	space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL);
560 	space_map_destroy(&vd->vdev_dtl_scrub);
561 	mutex_exit(&vd->vdev_dtl_lock);
562 	mutex_destroy(&vd->vdev_dtl_lock);
563 	mutex_destroy(&vd->vdev_stat_lock);
564 
565 	if (vd == spa->spa_root_vdev)
566 		spa->spa_root_vdev = NULL;
567 
568 	kmem_free(vd, sizeof (vdev_t));
569 }
570 
571 /*
572  * Transfer top-level vdev state from svd to tvd.
573  */
574 static void
575 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
576 {
577 	spa_t *spa = svd->vdev_spa;
578 	metaslab_t *msp;
579 	vdev_t *vd;
580 	int t;
581 
582 	ASSERT(tvd == tvd->vdev_top);
583 
584 	tvd->vdev_ms_array = svd->vdev_ms_array;
585 	tvd->vdev_ms_shift = svd->vdev_ms_shift;
586 	tvd->vdev_ms_count = svd->vdev_ms_count;
587 
588 	svd->vdev_ms_array = 0;
589 	svd->vdev_ms_shift = 0;
590 	svd->vdev_ms_count = 0;
591 
592 	tvd->vdev_mg = svd->vdev_mg;
593 	tvd->vdev_ms = svd->vdev_ms;
594 
595 	svd->vdev_mg = NULL;
596 	svd->vdev_ms = NULL;
597 
598 	if (tvd->vdev_mg != NULL)
599 		tvd->vdev_mg->mg_vd = tvd;
600 
601 	tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
602 	tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
603 	tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
604 
605 	svd->vdev_stat.vs_alloc = 0;
606 	svd->vdev_stat.vs_space = 0;
607 	svd->vdev_stat.vs_dspace = 0;
608 
609 	for (t = 0; t < TXG_SIZE; t++) {
610 		while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
611 			(void) txg_list_add(&tvd->vdev_ms_list, msp, t);
612 		while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
613 			(void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
614 		if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
615 			(void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
616 	}
617 
618 	if (list_link_active(&svd->vdev_dirty_node)) {
619 		vdev_config_clean(svd);
620 		vdev_config_dirty(tvd);
621 	}
622 
623 	tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
624 	svd->vdev_deflate_ratio = 0;
625 
626 	tvd->vdev_islog = svd->vdev_islog;
627 	svd->vdev_islog = 0;
628 }
629 
630 static void
631 vdev_top_update(vdev_t *tvd, vdev_t *vd)
632 {
633 	int c;
634 
635 	if (vd == NULL)
636 		return;
637 
638 	vd->vdev_top = tvd;
639 
640 	for (c = 0; c < vd->vdev_children; c++)
641 		vdev_top_update(tvd, vd->vdev_child[c]);
642 }
643 
644 /*
645  * Add a mirror/replacing vdev above an existing vdev.
646  */
647 vdev_t *
648 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
649 {
650 	spa_t *spa = cvd->vdev_spa;
651 	vdev_t *pvd = cvd->vdev_parent;
652 	vdev_t *mvd;
653 
654 	ASSERT(spa_config_held(spa, RW_WRITER));
655 
656 	mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
657 
658 	mvd->vdev_asize = cvd->vdev_asize;
659 	mvd->vdev_ashift = cvd->vdev_ashift;
660 	mvd->vdev_state = cvd->vdev_state;
661 
662 	vdev_remove_child(pvd, cvd);
663 	vdev_add_child(pvd, mvd);
664 	cvd->vdev_id = mvd->vdev_children;
665 	vdev_add_child(mvd, cvd);
666 	vdev_top_update(cvd->vdev_top, cvd->vdev_top);
667 
668 	if (mvd == mvd->vdev_top)
669 		vdev_top_transfer(cvd, mvd);
670 
671 	return (mvd);
672 }
673 
674 /*
675  * Remove a 1-way mirror/replacing vdev from the tree.
676  */
677 void
678 vdev_remove_parent(vdev_t *cvd)
679 {
680 	vdev_t *mvd = cvd->vdev_parent;
681 	vdev_t *pvd = mvd->vdev_parent;
682 
683 	ASSERT(spa_config_held(cvd->vdev_spa, RW_WRITER));
684 
685 	ASSERT(mvd->vdev_children == 1);
686 	ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
687 	    mvd->vdev_ops == &vdev_replacing_ops ||
688 	    mvd->vdev_ops == &vdev_spare_ops);
689 	cvd->vdev_ashift = mvd->vdev_ashift;
690 
691 	vdev_remove_child(mvd, cvd);
692 	vdev_remove_child(pvd, mvd);
693 	cvd->vdev_id = mvd->vdev_id;
694 	vdev_add_child(pvd, cvd);
695 	/*
696 	 * If we created a new toplevel vdev, then we need to change the child's
697 	 * vdev GUID to match the old toplevel vdev.  Otherwise, we could have
698 	 * detached an offline device, and when we go to import the pool we'll
699 	 * think we have two toplevel vdevs, instead of a different version of
700 	 * the same toplevel vdev.
701 	 */
702 	if (cvd->vdev_top == cvd) {
703 		pvd->vdev_guid_sum -= cvd->vdev_guid;
704 		cvd->vdev_guid_sum -= cvd->vdev_guid;
705 		cvd->vdev_guid = mvd->vdev_guid;
706 		cvd->vdev_guid_sum += mvd->vdev_guid;
707 		pvd->vdev_guid_sum += cvd->vdev_guid;
708 	}
709 	vdev_top_update(cvd->vdev_top, cvd->vdev_top);
710 
711 	if (cvd == cvd->vdev_top)
712 		vdev_top_transfer(mvd, cvd);
713 
714 	ASSERT(mvd->vdev_children == 0);
715 	vdev_free(mvd);
716 }
717 
718 int
719 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
720 {
721 	spa_t *spa = vd->vdev_spa;
722 	objset_t *mos = spa->spa_meta_objset;
723 	metaslab_class_t *mc;
724 	uint64_t m;
725 	uint64_t oldc = vd->vdev_ms_count;
726 	uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
727 	metaslab_t **mspp;
728 	int error;
729 
730 	if (vd->vdev_ms_shift == 0)	/* not being allocated from yet */
731 		return (0);
732 
733 	dprintf("%s oldc %llu newc %llu\n", vdev_description(vd), oldc, newc);
734 
735 	ASSERT(oldc <= newc);
736 
737 	if (vd->vdev_islog)
738 		mc = spa->spa_log_class;
739 	else
740 		mc = spa->spa_normal_class;
741 
742 	if (vd->vdev_mg == NULL)
743 		vd->vdev_mg = metaslab_group_create(mc, vd);
744 
745 	mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
746 
747 	if (oldc != 0) {
748 		bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
749 		kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
750 	}
751 
752 	vd->vdev_ms = mspp;
753 	vd->vdev_ms_count = newc;
754 
755 	for (m = oldc; m < newc; m++) {
756 		space_map_obj_t smo = { 0, 0, 0 };
757 		if (txg == 0) {
758 			uint64_t object = 0;
759 			error = dmu_read(mos, vd->vdev_ms_array,
760 			    m * sizeof (uint64_t), sizeof (uint64_t), &object);
761 			if (error)
762 				return (error);
763 			if (object != 0) {
764 				dmu_buf_t *db;
765 				error = dmu_bonus_hold(mos, object, FTAG, &db);
766 				if (error)
767 					return (error);
768 				ASSERT3U(db->db_size, ==, sizeof (smo));
769 				bcopy(db->db_data, &smo, db->db_size);
770 				ASSERT3U(smo.smo_object, ==, object);
771 				dmu_buf_rele(db, FTAG);
772 			}
773 		}
774 		vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
775 		    m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
776 	}
777 
778 	return (0);
779 }
780 
781 void
782 vdev_metaslab_fini(vdev_t *vd)
783 {
784 	uint64_t m;
785 	uint64_t count = vd->vdev_ms_count;
786 
787 	if (vd->vdev_ms != NULL) {
788 		for (m = 0; m < count; m++)
789 			if (vd->vdev_ms[m] != NULL)
790 				metaslab_fini(vd->vdev_ms[m]);
791 		kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
792 		vd->vdev_ms = NULL;
793 	}
794 }
795 
796 /*
797  * Prepare a virtual device for access.
798  */
799 int
800 vdev_open(vdev_t *vd)
801 {
802 	int error;
803 	int c;
804 	uint64_t osize = 0;
805 	uint64_t asize, psize;
806 	uint64_t ashift = 0;
807 
808 	ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
809 	    vd->vdev_state == VDEV_STATE_CANT_OPEN ||
810 	    vd->vdev_state == VDEV_STATE_OFFLINE);
811 
812 	if (vd->vdev_fault_mode == VDEV_FAULT_COUNT)
813 		vd->vdev_fault_arg >>= 1;
814 	else
815 		vd->vdev_fault_mode = VDEV_FAULT_NONE;
816 
817 	vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
818 
819 	if (!vd->vdev_removed && vd->vdev_faulted) {
820 		ASSERT(vd->vdev_children == 0);
821 		vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
822 		    VDEV_AUX_ERR_EXCEEDED);
823 		return (ENXIO);
824 	} else if (vd->vdev_offline) {
825 		ASSERT(vd->vdev_children == 0);
826 		vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
827 		return (ENXIO);
828 	}
829 
830 	error = vd->vdev_ops->vdev_op_open(vd, &osize, &ashift);
831 
832 	if (zio_injection_enabled && error == 0)
833 		error = zio_handle_device_injection(vd, ENXIO);
834 
835 	if (error) {
836 		if (vd->vdev_removed &&
837 		    vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
838 			vd->vdev_removed = B_FALSE;
839 
840 		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
841 		    vd->vdev_stat.vs_aux);
842 		return (error);
843 	}
844 
845 	vd->vdev_removed = B_FALSE;
846 
847 	if (vd->vdev_degraded) {
848 		ASSERT(vd->vdev_children == 0);
849 		vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
850 		    VDEV_AUX_ERR_EXCEEDED);
851 	} else {
852 		vd->vdev_state = VDEV_STATE_HEALTHY;
853 	}
854 
855 	for (c = 0; c < vd->vdev_children; c++)
856 		if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
857 			vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
858 			    VDEV_AUX_NONE);
859 			break;
860 		}
861 
862 	osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
863 
864 	if (vd->vdev_children == 0) {
865 		if (osize < SPA_MINDEVSIZE) {
866 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
867 			    VDEV_AUX_TOO_SMALL);
868 			return (EOVERFLOW);
869 		}
870 		psize = osize;
871 		asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
872 	} else {
873 		if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
874 		    (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
875 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
876 			    VDEV_AUX_TOO_SMALL);
877 			return (EOVERFLOW);
878 		}
879 		psize = 0;
880 		asize = osize;
881 	}
882 
883 	vd->vdev_psize = psize;
884 
885 	if (vd->vdev_asize == 0) {
886 		/*
887 		 * This is the first-ever open, so use the computed values.
888 		 * For testing purposes, a higher ashift can be requested.
889 		 */
890 		vd->vdev_asize = asize;
891 		vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
892 	} else {
893 		/*
894 		 * Make sure the alignment requirement hasn't increased.
895 		 */
896 		if (ashift > vd->vdev_top->vdev_ashift) {
897 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
898 			    VDEV_AUX_BAD_LABEL);
899 			return (EINVAL);
900 		}
901 
902 		/*
903 		 * Make sure the device hasn't shrunk.
904 		 */
905 		if (asize < vd->vdev_asize) {
906 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
907 			    VDEV_AUX_BAD_LABEL);
908 			return (EINVAL);
909 		}
910 
911 		/*
912 		 * If all children are healthy and the asize has increased,
913 		 * then we've experienced dynamic LUN growth.
914 		 */
915 		if (vd->vdev_state == VDEV_STATE_HEALTHY &&
916 		    asize > vd->vdev_asize) {
917 			vd->vdev_asize = asize;
918 		}
919 	}
920 
921 	/*
922 	 * If this is a top-level vdev, compute the raidz-deflation
923 	 * ratio.  Note, we hard-code in 128k (1<<17) because it is the
924 	 * current "typical" blocksize.  Even if SPA_MAXBLOCKSIZE
925 	 * changes, this algorithm must never change, or we will
926 	 * inconsistently account for existing bp's.
927 	 */
928 	if (vd->vdev_top == vd) {
929 		vd->vdev_deflate_ratio = (1<<17) /
930 		    (vdev_psize_to_asize(vd, 1<<17) >> SPA_MINBLOCKSHIFT);
931 	}
932 
933 	/*
934 	 * This allows the ZFS DE to close cases appropriately.  If a device
935 	 * goes away and later returns, we want to close the associated case.
936 	 * But it's not enough to simply post this only when a device goes from
937 	 * CANT_OPEN -> HEALTHY.  If we reboot the system and the device is
938 	 * back, we also need to close the case (otherwise we will try to replay
939 	 * it).  So we have to post this notifier every time.  Since this only
940 	 * occurs during pool open or error recovery, this should not be an
941 	 * issue.
942 	 */
943 	zfs_post_ok(vd->vdev_spa, vd);
944 
945 	return (0);
946 }
947 
948 /*
949  * Called once the vdevs are all opened, this routine validates the label
950  * contents.  This needs to be done before vdev_load() so that we don't
951  * inadvertently do repair I/Os to the wrong device.
952  *
953  * This function will only return failure if one of the vdevs indicates that it
954  * has since been destroyed or exported.  This is only possible if
955  * /etc/zfs/zpool.cache was readonly at the time.  Otherwise, the vdev state
956  * will be updated but the function will return 0.
957  */
958 int
959 vdev_validate(vdev_t *vd)
960 {
961 	spa_t *spa = vd->vdev_spa;
962 	int c;
963 	nvlist_t *label;
964 	uint64_t guid;
965 	uint64_t state;
966 
967 	for (c = 0; c < vd->vdev_children; c++)
968 		if (vdev_validate(vd->vdev_child[c]) != 0)
969 			return (EBADF);
970 
971 	/*
972 	 * If the device has already failed, or was marked offline, don't do
973 	 * any further validation.  Otherwise, label I/O will fail and we will
974 	 * overwrite the previous state.
975 	 */
976 	if (vd->vdev_ops->vdev_op_leaf && !vdev_is_dead(vd)) {
977 
978 		if ((label = vdev_label_read_config(vd)) == NULL) {
979 			vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
980 			    VDEV_AUX_BAD_LABEL);
981 			return (0);
982 		}
983 
984 		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
985 		    &guid) != 0 || guid != spa_guid(spa)) {
986 			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
987 			    VDEV_AUX_CORRUPT_DATA);
988 			nvlist_free(label);
989 			return (0);
990 		}
991 
992 		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
993 		    &guid) != 0 || guid != vd->vdev_guid) {
994 			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
995 			    VDEV_AUX_CORRUPT_DATA);
996 			nvlist_free(label);
997 			return (0);
998 		}
999 
1000 		if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1001 		    &state) != 0) {
1002 			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1003 			    VDEV_AUX_CORRUPT_DATA);
1004 			nvlist_free(label);
1005 			return (0);
1006 		}
1007 
1008 		nvlist_free(label);
1009 
1010 		if (spa->spa_load_state == SPA_LOAD_OPEN &&
1011 		    state != POOL_STATE_ACTIVE)
1012 			return (EBADF);
1013 	}
1014 
1015 	/*
1016 	 * If we were able to open and validate a vdev that was previously
1017 	 * marked permanently unavailable, clear that state now.
1018 	 */
1019 	if (vd->vdev_not_present)
1020 		vd->vdev_not_present = 0;
1021 
1022 	return (0);
1023 }
1024 
1025 /*
1026  * Close a virtual device.
1027  */
1028 void
1029 vdev_close(vdev_t *vd)
1030 {
1031 	vd->vdev_ops->vdev_op_close(vd);
1032 
1033 	vdev_cache_purge(vd);
1034 
1035 	/*
1036 	 * We record the previous state before we close it, so  that if we are
1037 	 * doing a reopen(), we don't generate FMA ereports if we notice that
1038 	 * it's still faulted.
1039 	 */
1040 	vd->vdev_prevstate = vd->vdev_state;
1041 
1042 	if (vd->vdev_offline)
1043 		vd->vdev_state = VDEV_STATE_OFFLINE;
1044 	else
1045 		vd->vdev_state = VDEV_STATE_CLOSED;
1046 	vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1047 }
1048 
1049 void
1050 vdev_reopen(vdev_t *vd)
1051 {
1052 	spa_t *spa = vd->vdev_spa;
1053 
1054 	ASSERT(spa_config_held(spa, RW_WRITER));
1055 
1056 	vdev_close(vd);
1057 	(void) vdev_open(vd);
1058 
1059 	/*
1060 	 * Call vdev_validate() here to make sure we have the same device.
1061 	 * Otherwise, a device with an invalid label could be successfully
1062 	 * opened in response to vdev_reopen().
1063 	 */
1064 	(void) vdev_validate(vd);
1065 
1066 	/*
1067 	 * Reassess parent vdev's health.
1068 	 */
1069 	vdev_propagate_state(vd);
1070 }
1071 
1072 int
1073 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1074 {
1075 	int error;
1076 
1077 	/*
1078 	 * Normally, partial opens (e.g. of a mirror) are allowed.
1079 	 * For a create, however, we want to fail the request if
1080 	 * there are any components we can't open.
1081 	 */
1082 	error = vdev_open(vd);
1083 
1084 	if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1085 		vdev_close(vd);
1086 		return (error ? error : ENXIO);
1087 	}
1088 
1089 	/*
1090 	 * Recursively initialize all labels.
1091 	 */
1092 	if ((error = vdev_label_init(vd, txg, isreplacing ?
1093 	    VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1094 		vdev_close(vd);
1095 		return (error);
1096 	}
1097 
1098 	return (0);
1099 }
1100 
1101 /*
1102  * The is the latter half of vdev_create().  It is distinct because it
1103  * involves initiating transactions in order to do metaslab creation.
1104  * For creation, we want to try to create all vdevs at once and then undo it
1105  * if anything fails; this is much harder if we have pending transactions.
1106  */
1107 void
1108 vdev_init(vdev_t *vd, uint64_t txg)
1109 {
1110 	/*
1111 	 * Aim for roughly 200 metaslabs per vdev.
1112 	 */
1113 	vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1114 	vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1115 
1116 	/*
1117 	 * Initialize the vdev's metaslabs.  This can't fail because
1118 	 * there's nothing to read when creating all new metaslabs.
1119 	 */
1120 	VERIFY(vdev_metaslab_init(vd, txg) == 0);
1121 }
1122 
1123 void
1124 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1125 {
1126 	ASSERT(vd == vd->vdev_top);
1127 	ASSERT(ISP2(flags));
1128 
1129 	if (flags & VDD_METASLAB)
1130 		(void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1131 
1132 	if (flags & VDD_DTL)
1133 		(void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1134 
1135 	(void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1136 }
1137 
1138 void
1139 vdev_dtl_dirty(space_map_t *sm, uint64_t txg, uint64_t size)
1140 {
1141 	mutex_enter(sm->sm_lock);
1142 	if (!space_map_contains(sm, txg, size))
1143 		space_map_add(sm, txg, size);
1144 	mutex_exit(sm->sm_lock);
1145 }
1146 
1147 int
1148 vdev_dtl_contains(space_map_t *sm, uint64_t txg, uint64_t size)
1149 {
1150 	int dirty;
1151 
1152 	/*
1153 	 * Quick test without the lock -- covers the common case that
1154 	 * there are no dirty time segments.
1155 	 */
1156 	if (sm->sm_space == 0)
1157 		return (0);
1158 
1159 	mutex_enter(sm->sm_lock);
1160 	dirty = space_map_contains(sm, txg, size);
1161 	mutex_exit(sm->sm_lock);
1162 
1163 	return (dirty);
1164 }
1165 
1166 /*
1167  * Reassess DTLs after a config change or scrub completion.
1168  */
1169 void
1170 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1171 {
1172 	spa_t *spa = vd->vdev_spa;
1173 	int c;
1174 
1175 	ASSERT(spa_config_held(spa, RW_WRITER));
1176 
1177 	if (vd->vdev_children == 0) {
1178 		mutex_enter(&vd->vdev_dtl_lock);
1179 		/*
1180 		 * We're successfully scrubbed everything up to scrub_txg.
1181 		 * Therefore, excise all old DTLs up to that point, then
1182 		 * fold in the DTLs for everything we couldn't scrub.
1183 		 */
1184 		if (scrub_txg != 0) {
1185 			space_map_excise(&vd->vdev_dtl_map, 0, scrub_txg);
1186 			space_map_union(&vd->vdev_dtl_map, &vd->vdev_dtl_scrub);
1187 		}
1188 		if (scrub_done)
1189 			space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL);
1190 		mutex_exit(&vd->vdev_dtl_lock);
1191 		if (txg != 0)
1192 			vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1193 		return;
1194 	}
1195 
1196 	/*
1197 	 * Make sure the DTLs are always correct under the scrub lock.
1198 	 */
1199 	if (vd == spa->spa_root_vdev)
1200 		mutex_enter(&spa->spa_scrub_lock);
1201 
1202 	mutex_enter(&vd->vdev_dtl_lock);
1203 	space_map_vacate(&vd->vdev_dtl_map, NULL, NULL);
1204 	space_map_vacate(&vd->vdev_dtl_scrub, NULL, NULL);
1205 	mutex_exit(&vd->vdev_dtl_lock);
1206 
1207 	for (c = 0; c < vd->vdev_children; c++) {
1208 		vdev_t *cvd = vd->vdev_child[c];
1209 		vdev_dtl_reassess(cvd, txg, scrub_txg, scrub_done);
1210 		mutex_enter(&vd->vdev_dtl_lock);
1211 		space_map_union(&vd->vdev_dtl_map, &cvd->vdev_dtl_map);
1212 		space_map_union(&vd->vdev_dtl_scrub, &cvd->vdev_dtl_scrub);
1213 		mutex_exit(&vd->vdev_dtl_lock);
1214 	}
1215 
1216 	if (vd == spa->spa_root_vdev)
1217 		mutex_exit(&spa->spa_scrub_lock);
1218 }
1219 
1220 static int
1221 vdev_dtl_load(vdev_t *vd)
1222 {
1223 	spa_t *spa = vd->vdev_spa;
1224 	space_map_obj_t *smo = &vd->vdev_dtl;
1225 	objset_t *mos = spa->spa_meta_objset;
1226 	dmu_buf_t *db;
1227 	int error;
1228 
1229 	ASSERT(vd->vdev_children == 0);
1230 
1231 	if (smo->smo_object == 0)
1232 		return (0);
1233 
1234 	if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1235 		return (error);
1236 
1237 	ASSERT3U(db->db_size, ==, sizeof (*smo));
1238 	bcopy(db->db_data, smo, db->db_size);
1239 	dmu_buf_rele(db, FTAG);
1240 
1241 	mutex_enter(&vd->vdev_dtl_lock);
1242 	error = space_map_load(&vd->vdev_dtl_map, NULL, SM_ALLOC, smo, mos);
1243 	mutex_exit(&vd->vdev_dtl_lock);
1244 
1245 	return (error);
1246 }
1247 
1248 void
1249 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1250 {
1251 	spa_t *spa = vd->vdev_spa;
1252 	space_map_obj_t *smo = &vd->vdev_dtl;
1253 	space_map_t *sm = &vd->vdev_dtl_map;
1254 	objset_t *mos = spa->spa_meta_objset;
1255 	space_map_t smsync;
1256 	kmutex_t smlock;
1257 	dmu_buf_t *db;
1258 	dmu_tx_t *tx;
1259 
1260 	dprintf("%s in txg %llu pass %d\n",
1261 	    vdev_description(vd), (u_longlong_t)txg, spa_sync_pass(spa));
1262 
1263 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1264 
1265 	if (vd->vdev_detached) {
1266 		if (smo->smo_object != 0) {
1267 			int err = dmu_object_free(mos, smo->smo_object, tx);
1268 			ASSERT3U(err, ==, 0);
1269 			smo->smo_object = 0;
1270 		}
1271 		dmu_tx_commit(tx);
1272 		dprintf("detach %s committed in txg %llu\n",
1273 		    vdev_description(vd), txg);
1274 		return;
1275 	}
1276 
1277 	if (smo->smo_object == 0) {
1278 		ASSERT(smo->smo_objsize == 0);
1279 		ASSERT(smo->smo_alloc == 0);
1280 		smo->smo_object = dmu_object_alloc(mos,
1281 		    DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1282 		    DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1283 		ASSERT(smo->smo_object != 0);
1284 		vdev_config_dirty(vd->vdev_top);
1285 	}
1286 
1287 	mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1288 
1289 	space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1290 	    &smlock);
1291 
1292 	mutex_enter(&smlock);
1293 
1294 	mutex_enter(&vd->vdev_dtl_lock);
1295 	space_map_walk(sm, space_map_add, &smsync);
1296 	mutex_exit(&vd->vdev_dtl_lock);
1297 
1298 	space_map_truncate(smo, mos, tx);
1299 	space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1300 
1301 	space_map_destroy(&smsync);
1302 
1303 	mutex_exit(&smlock);
1304 	mutex_destroy(&smlock);
1305 
1306 	VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1307 	dmu_buf_will_dirty(db, tx);
1308 	ASSERT3U(db->db_size, ==, sizeof (*smo));
1309 	bcopy(smo, db->db_data, db->db_size);
1310 	dmu_buf_rele(db, FTAG);
1311 
1312 	dmu_tx_commit(tx);
1313 }
1314 
1315 void
1316 vdev_load(vdev_t *vd)
1317 {
1318 	int c;
1319 
1320 	/*
1321 	 * Recursively load all children.
1322 	 */
1323 	for (c = 0; c < vd->vdev_children; c++)
1324 		vdev_load(vd->vdev_child[c]);
1325 
1326 	/*
1327 	 * If this is a top-level vdev, initialize its metaslabs.
1328 	 */
1329 	if (vd == vd->vdev_top &&
1330 	    (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1331 	    vdev_metaslab_init(vd, 0) != 0))
1332 		vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1333 		    VDEV_AUX_CORRUPT_DATA);
1334 
1335 	/*
1336 	 * If this is a leaf vdev, load its DTL.
1337 	 */
1338 	if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1339 		vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1340 		    VDEV_AUX_CORRUPT_DATA);
1341 }
1342 
1343 /*
1344  * This special case of vdev_spare() is used for hot spares.  It's sole purpose
1345  * it to set the vdev state for the associated vdev.  To do this, we make sure
1346  * that we can open the underlying device, then try to read the label, and make
1347  * sure that the label is sane and that it hasn't been repurposed to another
1348  * pool.
1349  */
1350 int
1351 vdev_validate_spare(vdev_t *vd)
1352 {
1353 	nvlist_t *label;
1354 	uint64_t guid, version;
1355 	uint64_t state;
1356 
1357 	if ((label = vdev_label_read_config(vd)) == NULL) {
1358 		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1359 		    VDEV_AUX_CORRUPT_DATA);
1360 		return (-1);
1361 	}
1362 
1363 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1364 	    version > SPA_VERSION ||
1365 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1366 	    guid != vd->vdev_guid ||
1367 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1368 		vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1369 		    VDEV_AUX_CORRUPT_DATA);
1370 		nvlist_free(label);
1371 		return (-1);
1372 	}
1373 
1374 	spa_spare_add(vd);
1375 
1376 	/*
1377 	 * We don't actually check the pool state here.  If it's in fact in
1378 	 * use by another pool, we update this fact on the fly when requested.
1379 	 */
1380 	nvlist_free(label);
1381 	return (0);
1382 }
1383 
1384 void
1385 vdev_sync_done(vdev_t *vd, uint64_t txg)
1386 {
1387 	metaslab_t *msp;
1388 
1389 	dprintf("%s txg %llu\n", vdev_description(vd), txg);
1390 
1391 	while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
1392 		metaslab_sync_done(msp, txg);
1393 }
1394 
1395 void
1396 vdev_sync(vdev_t *vd, uint64_t txg)
1397 {
1398 	spa_t *spa = vd->vdev_spa;
1399 	vdev_t *lvd;
1400 	metaslab_t *msp;
1401 	dmu_tx_t *tx;
1402 
1403 	dprintf("%s txg %llu pass %d\n",
1404 	    vdev_description(vd), (u_longlong_t)txg, spa_sync_pass(spa));
1405 
1406 	if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
1407 		ASSERT(vd == vd->vdev_top);
1408 		tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1409 		vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
1410 		    DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
1411 		ASSERT(vd->vdev_ms_array != 0);
1412 		vdev_config_dirty(vd);
1413 		dmu_tx_commit(tx);
1414 	}
1415 
1416 	while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
1417 		metaslab_sync(msp, txg);
1418 		(void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
1419 	}
1420 
1421 	while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
1422 		vdev_dtl_sync(lvd, txg);
1423 
1424 	(void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
1425 }
1426 
1427 uint64_t
1428 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
1429 {
1430 	return (vd->vdev_ops->vdev_op_asize(vd, psize));
1431 }
1432 
1433 void
1434 vdev_io_start(zio_t *zio)
1435 {
1436 	zio->io_vd->vdev_ops->vdev_op_io_start(zio);
1437 }
1438 
1439 void
1440 vdev_io_done(zio_t *zio)
1441 {
1442 	zio->io_vd->vdev_ops->vdev_op_io_done(zio);
1443 }
1444 
1445 const char *
1446 vdev_description(vdev_t *vd)
1447 {
1448 	if (vd == NULL || vd->vdev_ops == NULL)
1449 		return ("<unknown>");
1450 
1451 	if (vd->vdev_path != NULL)
1452 		return (vd->vdev_path);
1453 
1454 	if (vd->vdev_parent == NULL)
1455 		return (spa_name(vd->vdev_spa));
1456 
1457 	return (vd->vdev_ops->vdev_op_type);
1458 }
1459 
1460 /*
1461  * Mark the given vdev faulted.  A faulted vdev behaves as if the device could
1462  * not be opened, and no I/O is attempted.
1463  */
1464 int
1465 vdev_fault(spa_t *spa, uint64_t guid)
1466 {
1467 	vdev_t *rvd, *vd;
1468 	uint64_t txg;
1469 
1470 	txg = spa_vdev_enter(spa);
1471 
1472 	rvd = spa->spa_root_vdev;
1473 
1474 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1475 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1476 	if (!vd->vdev_ops->vdev_op_leaf)
1477 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1478 
1479 	/*
1480 	 * Faulted state takes precedence over degraded.
1481 	 */
1482 	vd->vdev_faulted = 1ULL;
1483 	vd->vdev_degraded = 0ULL;
1484 	vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED,
1485 	    VDEV_AUX_ERR_EXCEEDED);
1486 
1487 	/*
1488 	 * If marking the vdev as faulted cause the toplevel vdev to become
1489 	 * unavailable, then back off and simply mark the vdev as degraded
1490 	 * instead.
1491 	 */
1492 	if (vdev_is_dead(vd->vdev_top)) {
1493 		vd->vdev_degraded = 1ULL;
1494 		vd->vdev_faulted = 0ULL;
1495 
1496 		/*
1497 		 * If we reopen the device and it's not dead, only then do we
1498 		 * mark it degraded.
1499 		 */
1500 		vdev_reopen(vd);
1501 
1502 		if (!vdev_is_dead(vd)) {
1503 			vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1504 			    VDEV_AUX_ERR_EXCEEDED);
1505 		}
1506 	}
1507 
1508 	vdev_config_dirty(vd->vdev_top);
1509 
1510 	(void) spa_vdev_exit(spa, NULL, txg, 0);
1511 
1512 	return (0);
1513 }
1514 
1515 /*
1516  * Mark the given vdev degraded.  A degraded vdev is purely an indication to the
1517  * user that something is wrong.  The vdev continues to operate as normal as far
1518  * as I/O is concerned.
1519  */
1520 int
1521 vdev_degrade(spa_t *spa, uint64_t guid)
1522 {
1523 	vdev_t *rvd, *vd;
1524 	uint64_t txg;
1525 
1526 	txg = spa_vdev_enter(spa);
1527 
1528 	rvd = spa->spa_root_vdev;
1529 
1530 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1531 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1532 	if (!vd->vdev_ops->vdev_op_leaf)
1533 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1534 
1535 	/*
1536 	 * If the vdev is already faulted, then don't do anything.
1537 	 */
1538 	if (vd->vdev_faulted || vd->vdev_degraded) {
1539 		(void) spa_vdev_exit(spa, NULL, txg, 0);
1540 		return (0);
1541 	}
1542 
1543 	vd->vdev_degraded = 1ULL;
1544 	if (!vdev_is_dead(vd))
1545 		vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
1546 		    VDEV_AUX_ERR_EXCEEDED);
1547 	vdev_config_dirty(vd->vdev_top);
1548 
1549 	(void) spa_vdev_exit(spa, NULL, txg, 0);
1550 
1551 	return (0);
1552 }
1553 
1554 /*
1555  * Online the given vdev.  If 'unspare' is set, it implies two things.  First,
1556  * any attached spare device should be detached when the device finishes
1557  * resilvering.  Second, the online should be treated like a 'test' online case,
1558  * so no FMA events are generated if the device fails to open.
1559  */
1560 int
1561 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags,
1562     vdev_state_t *newstate)
1563 {
1564 	vdev_t *rvd, *vd;
1565 	uint64_t txg;
1566 
1567 	txg = spa_vdev_enter(spa);
1568 
1569 	rvd = spa->spa_root_vdev;
1570 
1571 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1572 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1573 
1574 	if (!vd->vdev_ops->vdev_op_leaf)
1575 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1576 
1577 	vd->vdev_offline = B_FALSE;
1578 	vd->vdev_tmpoffline = B_FALSE;
1579 	vd->vdev_checkremove = (flags & ZFS_ONLINE_CHECKREMOVE) ?
1580 	    B_TRUE : B_FALSE;
1581 	vd->vdev_forcefault = (flags & ZFS_ONLINE_FORCEFAULT) ?
1582 	    B_TRUE : B_FALSE;
1583 	vdev_reopen(vd->vdev_top);
1584 	vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
1585 
1586 	if (newstate)
1587 		*newstate = vd->vdev_state;
1588 	if ((flags & ZFS_ONLINE_UNSPARE) &&
1589 	    !vdev_is_dead(vd) && vd->vdev_parent &&
1590 	    vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
1591 	    vd->vdev_parent->vdev_child[0] == vd)
1592 		vd->vdev_unspare = B_TRUE;
1593 
1594 	vdev_config_dirty(vd->vdev_top);
1595 
1596 	(void) spa_vdev_exit(spa, NULL, txg, 0);
1597 
1598 	/*
1599 	 * Must hold spa_namespace_lock in order to post resilver sysevent
1600 	 * w/pool name.
1601 	 */
1602 	mutex_enter(&spa_namespace_lock);
1603 	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1604 	mutex_exit(&spa_namespace_lock);
1605 
1606 	return (0);
1607 }
1608 
1609 int
1610 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
1611 {
1612 	vdev_t *rvd, *vd;
1613 	uint64_t txg;
1614 
1615 	txg = spa_vdev_enter(spa);
1616 
1617 	rvd = spa->spa_root_vdev;
1618 
1619 	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1620 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1621 
1622 	if (!vd->vdev_ops->vdev_op_leaf)
1623 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1624 
1625 	/*
1626 	 * If the device isn't already offline, try to offline it.
1627 	 */
1628 	if (!vd->vdev_offline) {
1629 		/*
1630 		 * If this device's top-level vdev has a non-empty DTL,
1631 		 * don't allow the device to be offlined.
1632 		 *
1633 		 * XXX -- make this more precise by allowing the offline
1634 		 * as long as the remaining devices don't have any DTL holes.
1635 		 */
1636 		if (vd->vdev_top->vdev_dtl_map.sm_space != 0)
1637 			return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1638 
1639 		/*
1640 		 * Offline this device and reopen its top-level vdev.
1641 		 * If this action results in the top-level vdev becoming
1642 		 * unusable, undo it and fail the request.
1643 		 */
1644 		vd->vdev_offline = B_TRUE;
1645 		vdev_reopen(vd->vdev_top);
1646 		if (vdev_is_dead(vd->vdev_top)) {
1647 			vd->vdev_offline = B_FALSE;
1648 			vdev_reopen(vd->vdev_top);
1649 			return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1650 		}
1651 	}
1652 
1653 	vd->vdev_tmpoffline = (flags & ZFS_OFFLINE_TEMPORARY) ?
1654 	    B_TRUE : B_FALSE;
1655 
1656 	vdev_config_dirty(vd->vdev_top);
1657 
1658 	return (spa_vdev_exit(spa, NULL, txg, 0));
1659 }
1660 
1661 /*
1662  * Clear the error counts associated with this vdev.  Unlike vdev_online() and
1663  * vdev_offline(), we assume the spa config is locked.  We also clear all
1664  * children.  If 'vd' is NULL, then the user wants to clear all vdevs.
1665  */
1666 void
1667 vdev_clear(spa_t *spa, vdev_t *vd)
1668 {
1669 	int c;
1670 
1671 	if (vd == NULL)
1672 		vd = spa->spa_root_vdev;
1673 
1674 	vd->vdev_stat.vs_read_errors = 0;
1675 	vd->vdev_stat.vs_write_errors = 0;
1676 	vd->vdev_stat.vs_checksum_errors = 0;
1677 
1678 	for (c = 0; c < vd->vdev_children; c++)
1679 		vdev_clear(spa, vd->vdev_child[c]);
1680 
1681 	/*
1682 	 * If we're in the FAULTED state, then clear the persistent state and
1683 	 * attempt to reopen the device.  We also mark the vdev config dirty, so
1684 	 * that the new faulted state is written out to disk.
1685 	 */
1686 	if (vd->vdev_faulted || vd->vdev_degraded) {
1687 		vd->vdev_faulted = vd->vdev_degraded = 0;
1688 		vdev_reopen(vd);
1689 		vdev_config_dirty(vd->vdev_top);
1690 
1691 		if (vd->vdev_faulted)
1692 			spa_async_request(spa, SPA_ASYNC_RESILVER);
1693 
1694 		spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
1695 	}
1696 }
1697 
1698 int
1699 vdev_is_dead(vdev_t *vd)
1700 {
1701 	return (vd->vdev_state < VDEV_STATE_DEGRADED);
1702 }
1703 
1704 int
1705 vdev_error_inject(vdev_t *vd, zio_t *zio)
1706 {
1707 	int error = 0;
1708 
1709 	if (vd->vdev_fault_mode == VDEV_FAULT_NONE)
1710 		return (0);
1711 
1712 	if (((1ULL << zio->io_type) & vd->vdev_fault_mask) == 0)
1713 		return (0);
1714 
1715 	switch (vd->vdev_fault_mode) {
1716 	case VDEV_FAULT_RANDOM:
1717 		if (spa_get_random(vd->vdev_fault_arg) == 0)
1718 			error = EIO;
1719 		break;
1720 
1721 	case VDEV_FAULT_COUNT:
1722 		if ((int64_t)--vd->vdev_fault_arg <= 0)
1723 			vd->vdev_fault_mode = VDEV_FAULT_NONE;
1724 		error = EIO;
1725 		break;
1726 	}
1727 
1728 	return (error);
1729 }
1730 
1731 /*
1732  * Get statistics for the given vdev.
1733  */
1734 void
1735 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
1736 {
1737 	vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
1738 	int c, t;
1739 
1740 	mutex_enter(&vd->vdev_stat_lock);
1741 	bcopy(&vd->vdev_stat, vs, sizeof (*vs));
1742 	vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
1743 	vs->vs_state = vd->vdev_state;
1744 	vs->vs_rsize = vdev_get_rsize(vd);
1745 	mutex_exit(&vd->vdev_stat_lock);
1746 
1747 	/*
1748 	 * If we're getting stats on the root vdev, aggregate the I/O counts
1749 	 * over all top-level vdevs (i.e. the direct children of the root).
1750 	 */
1751 	if (vd == rvd) {
1752 		for (c = 0; c < rvd->vdev_children; c++) {
1753 			vdev_t *cvd = rvd->vdev_child[c];
1754 			vdev_stat_t *cvs = &cvd->vdev_stat;
1755 
1756 			mutex_enter(&vd->vdev_stat_lock);
1757 			for (t = 0; t < ZIO_TYPES; t++) {
1758 				vs->vs_ops[t] += cvs->vs_ops[t];
1759 				vs->vs_bytes[t] += cvs->vs_bytes[t];
1760 			}
1761 			vs->vs_read_errors += cvs->vs_read_errors;
1762 			vs->vs_write_errors += cvs->vs_write_errors;
1763 			vs->vs_checksum_errors += cvs->vs_checksum_errors;
1764 			vs->vs_scrub_examined += cvs->vs_scrub_examined;
1765 			vs->vs_scrub_errors += cvs->vs_scrub_errors;
1766 			mutex_exit(&vd->vdev_stat_lock);
1767 		}
1768 	}
1769 }
1770 
1771 void
1772 vdev_stat_update(zio_t *zio)
1773 {
1774 	vdev_t *vd = zio->io_vd;
1775 	vdev_t *pvd;
1776 	uint64_t txg = zio->io_txg;
1777 	vdev_stat_t *vs = &vd->vdev_stat;
1778 	zio_type_t type = zio->io_type;
1779 	int flags = zio->io_flags;
1780 
1781 	if (zio->io_error == 0) {
1782 		if (!(flags & ZIO_FLAG_IO_BYPASS)) {
1783 			mutex_enter(&vd->vdev_stat_lock);
1784 			vs->vs_ops[type]++;
1785 			vs->vs_bytes[type] += zio->io_size;
1786 			mutex_exit(&vd->vdev_stat_lock);
1787 		}
1788 		if ((flags & ZIO_FLAG_IO_REPAIR) &&
1789 		    zio->io_delegate_list == NULL) {
1790 			mutex_enter(&vd->vdev_stat_lock);
1791 			if (flags & ZIO_FLAG_SCRUB_THREAD)
1792 				vs->vs_scrub_repaired += zio->io_size;
1793 			else
1794 				vs->vs_self_healed += zio->io_size;
1795 			mutex_exit(&vd->vdev_stat_lock);
1796 		}
1797 		return;
1798 	}
1799 
1800 	if (flags & ZIO_FLAG_SPECULATIVE)
1801 		return;
1802 
1803 	if (!vdev_is_dead(vd)) {
1804 		mutex_enter(&vd->vdev_stat_lock);
1805 		if (type == ZIO_TYPE_READ) {
1806 			if (zio->io_error == ECKSUM)
1807 				vs->vs_checksum_errors++;
1808 			else
1809 				vs->vs_read_errors++;
1810 		}
1811 		if (type == ZIO_TYPE_WRITE)
1812 			vs->vs_write_errors++;
1813 		mutex_exit(&vd->vdev_stat_lock);
1814 	}
1815 
1816 	if (type == ZIO_TYPE_WRITE) {
1817 		if (txg == 0 || vd->vdev_children != 0)
1818 			return;
1819 		if (flags & ZIO_FLAG_SCRUB_THREAD) {
1820 			ASSERT(flags & ZIO_FLAG_IO_REPAIR);
1821 			for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1822 				vdev_dtl_dirty(&pvd->vdev_dtl_scrub, txg, 1);
1823 		}
1824 		if (!(flags & ZIO_FLAG_IO_REPAIR)) {
1825 			if (vdev_dtl_contains(&vd->vdev_dtl_map, txg, 1))
1826 				return;
1827 			vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1828 			for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1829 				vdev_dtl_dirty(&pvd->vdev_dtl_map, txg, 1);
1830 		}
1831 	}
1832 }
1833 
1834 void
1835 vdev_scrub_stat_update(vdev_t *vd, pool_scrub_type_t type, boolean_t complete)
1836 {
1837 	int c;
1838 	vdev_stat_t *vs = &vd->vdev_stat;
1839 
1840 	for (c = 0; c < vd->vdev_children; c++)
1841 		vdev_scrub_stat_update(vd->vdev_child[c], type, complete);
1842 
1843 	mutex_enter(&vd->vdev_stat_lock);
1844 
1845 	if (type == POOL_SCRUB_NONE) {
1846 		/*
1847 		 * Update completion and end time.  Leave everything else alone
1848 		 * so we can report what happened during the previous scrub.
1849 		 */
1850 		vs->vs_scrub_complete = complete;
1851 		vs->vs_scrub_end = gethrestime_sec();
1852 	} else {
1853 		vs->vs_scrub_type = type;
1854 		vs->vs_scrub_complete = 0;
1855 		vs->vs_scrub_examined = 0;
1856 		vs->vs_scrub_repaired = 0;
1857 		vs->vs_scrub_errors = 0;
1858 		vs->vs_scrub_start = gethrestime_sec();
1859 		vs->vs_scrub_end = 0;
1860 	}
1861 
1862 	mutex_exit(&vd->vdev_stat_lock);
1863 }
1864 
1865 /*
1866  * Update the in-core space usage stats for this vdev and the root vdev.
1867  */
1868 void
1869 vdev_space_update(vdev_t *vd, int64_t space_delta, int64_t alloc_delta)
1870 {
1871 	int64_t dspace_delta = space_delta;
1872 	spa_t *spa = vd->vdev_spa;
1873 	vdev_t *rvd = spa->spa_root_vdev;
1874 
1875 	ASSERT(vd == vd->vdev_top);
1876 	ASSERT(rvd == vd->vdev_parent);
1877 	ASSERT(vd->vdev_ms_count != 0);
1878 
1879 	/*
1880 	 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
1881 	 * factor.  We must calculate this here and not at the root vdev
1882 	 * because the root vdev's psize-to-asize is simply the max of its
1883 	 * childrens', thus not accurate enough for us.
1884 	 */
1885 	ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
1886 	dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
1887 	    vd->vdev_deflate_ratio;
1888 
1889 	mutex_enter(&vd->vdev_stat_lock);
1890 	vd->vdev_stat.vs_space += space_delta;
1891 	vd->vdev_stat.vs_alloc += alloc_delta;
1892 	vd->vdev_stat.vs_dspace += dspace_delta;
1893 	mutex_exit(&vd->vdev_stat_lock);
1894 
1895 	/*
1896 	 * Don't count non-normal (e.g. intent log) space as part of
1897 	 * the pool's capacity.
1898 	 */
1899 	if (vd->vdev_mg->mg_class != spa->spa_normal_class)
1900 		return;
1901 
1902 	mutex_enter(&rvd->vdev_stat_lock);
1903 	rvd->vdev_stat.vs_space += space_delta;
1904 	rvd->vdev_stat.vs_alloc += alloc_delta;
1905 	rvd->vdev_stat.vs_dspace += dspace_delta;
1906 	mutex_exit(&rvd->vdev_stat_lock);
1907 }
1908 
1909 /*
1910  * Mark a top-level vdev's config as dirty, placing it on the dirty list
1911  * so that it will be written out next time the vdev configuration is synced.
1912  * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
1913  */
1914 void
1915 vdev_config_dirty(vdev_t *vd)
1916 {
1917 	spa_t *spa = vd->vdev_spa;
1918 	vdev_t *rvd = spa->spa_root_vdev;
1919 	int c;
1920 
1921 	/*
1922 	 * The dirty list is protected by the config lock.  The caller must
1923 	 * either hold the config lock as writer, or must be the sync thread
1924 	 * (which holds the lock as reader).  There's only one sync thread,
1925 	 * so this is sufficient to ensure mutual exclusion.
1926 	 */
1927 	ASSERT(spa_config_held(spa, RW_WRITER) ||
1928 	    dsl_pool_sync_context(spa_get_dsl(spa)));
1929 
1930 	if (vd == rvd) {
1931 		for (c = 0; c < rvd->vdev_children; c++)
1932 			vdev_config_dirty(rvd->vdev_child[c]);
1933 	} else {
1934 		ASSERT(vd == vd->vdev_top);
1935 
1936 		if (!list_link_active(&vd->vdev_dirty_node))
1937 			list_insert_head(&spa->spa_dirty_list, vd);
1938 	}
1939 }
1940 
1941 void
1942 vdev_config_clean(vdev_t *vd)
1943 {
1944 	spa_t *spa = vd->vdev_spa;
1945 
1946 	ASSERT(spa_config_held(spa, RW_WRITER) ||
1947 	    dsl_pool_sync_context(spa_get_dsl(spa)));
1948 
1949 	ASSERT(list_link_active(&vd->vdev_dirty_node));
1950 	list_remove(&spa->spa_dirty_list, vd);
1951 }
1952 
1953 void
1954 vdev_propagate_state(vdev_t *vd)
1955 {
1956 	vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
1957 	int degraded = 0, faulted = 0;
1958 	int corrupted = 0;
1959 	int c;
1960 	vdev_t *child;
1961 
1962 	if (vd->vdev_children > 0) {
1963 		for (c = 0; c < vd->vdev_children; c++) {
1964 			child = vd->vdev_child[c];
1965 			if (vdev_is_dead(child))
1966 				faulted++;
1967 			else if (child->vdev_state == VDEV_STATE_DEGRADED)
1968 				degraded++;
1969 
1970 			if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
1971 				corrupted++;
1972 		}
1973 
1974 		vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
1975 
1976 		/*
1977 		 * Root special: if there is a toplevel vdev that cannot be
1978 		 * opened due to corrupted metadata, then propagate the root
1979 		 * vdev's aux state as 'corrupt' rather than 'insufficient
1980 		 * replicas'.
1981 		 */
1982 		if (corrupted && vd == rvd &&
1983 		    rvd->vdev_state == VDEV_STATE_CANT_OPEN)
1984 			vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
1985 			    VDEV_AUX_CORRUPT_DATA);
1986 	}
1987 
1988 	if (vd->vdev_parent && !vd->vdev_islog)
1989 		vdev_propagate_state(vd->vdev_parent);
1990 }
1991 
1992 /*
1993  * Set a vdev's state.  If this is during an open, we don't update the parent
1994  * state, because we're in the process of opening children depth-first.
1995  * Otherwise, we propagate the change to the parent.
1996  *
1997  * If this routine places a device in a faulted state, an appropriate ereport is
1998  * generated.
1999  */
2000 void
2001 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2002 {
2003 	uint64_t save_state;
2004 
2005 	if (state == vd->vdev_state) {
2006 		vd->vdev_stat.vs_aux = aux;
2007 		return;
2008 	}
2009 
2010 	save_state = vd->vdev_state;
2011 
2012 	vd->vdev_state = state;
2013 	vd->vdev_stat.vs_aux = aux;
2014 
2015 	/*
2016 	 * If we are setting the vdev state to anything but an open state, then
2017 	 * always close the underlying device.  Otherwise, we keep accessible
2018 	 * but invalid devices open forever.  We don't call vdev_close() itself,
2019 	 * because that implies some extra checks (offline, etc) that we don't
2020 	 * want here.  This is limited to leaf devices, because otherwise
2021 	 * closing the device will affect other children.
2022 	 */
2023 	if (vdev_is_dead(vd) && vd->vdev_ops->vdev_op_leaf)
2024 		vd->vdev_ops->vdev_op_close(vd);
2025 
2026 	if (vd->vdev_removed &&
2027 	    state == VDEV_STATE_CANT_OPEN &&
2028 	    (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2029 		/*
2030 		 * If the previous state is set to VDEV_STATE_REMOVED, then this
2031 		 * device was previously marked removed and someone attempted to
2032 		 * reopen it.  If this failed due to a nonexistent device, then
2033 		 * keep the device in the REMOVED state.  We also let this be if
2034 		 * it is one of our special test online cases, which is only
2035 		 * attempting to online the device and shouldn't generate an FMA
2036 		 * fault.
2037 		 */
2038 		vd->vdev_state = VDEV_STATE_REMOVED;
2039 		vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2040 	} else if (state == VDEV_STATE_REMOVED) {
2041 		/*
2042 		 * Indicate to the ZFS DE that this device has been removed, and
2043 		 * any recent errors should be ignored.
2044 		 */
2045 		zfs_post_remove(vd->vdev_spa, vd);
2046 		vd->vdev_removed = B_TRUE;
2047 	} else if (state == VDEV_STATE_CANT_OPEN) {
2048 		/*
2049 		 * If we fail to open a vdev during an import, we mark it as
2050 		 * "not available", which signifies that it was never there to
2051 		 * begin with.  Failure to open such a device is not considered
2052 		 * an error.
2053 		 */
2054 		if (vd->vdev_spa->spa_load_state == SPA_LOAD_IMPORT &&
2055 		    vd->vdev_ops->vdev_op_leaf)
2056 			vd->vdev_not_present = 1;
2057 
2058 		/*
2059 		 * Post the appropriate ereport.  If the 'prevstate' field is
2060 		 * set to something other than VDEV_STATE_UNKNOWN, it indicates
2061 		 * that this is part of a vdev_reopen().  In this case, we don't
2062 		 * want to post the ereport if the device was already in the
2063 		 * CANT_OPEN state beforehand.
2064 		 *
2065 		 * If the 'checkremove' flag is set, then this is an attempt to
2066 		 * online the device in response to an insertion event.  If we
2067 		 * hit this case, then we have detected an insertion event for a
2068 		 * faulted or offline device that wasn't in the removed state.
2069 		 * In this scenario, we don't post an ereport because we are
2070 		 * about to replace the device, or attempt an online with
2071 		 * vdev_forcefault, which will generate the fault for us.
2072 		 */
2073 		if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
2074 		    !vd->vdev_not_present && !vd->vdev_checkremove &&
2075 		    vd != vd->vdev_spa->spa_root_vdev) {
2076 			const char *class;
2077 
2078 			switch (aux) {
2079 			case VDEV_AUX_OPEN_FAILED:
2080 				class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
2081 				break;
2082 			case VDEV_AUX_CORRUPT_DATA:
2083 				class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
2084 				break;
2085 			case VDEV_AUX_NO_REPLICAS:
2086 				class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
2087 				break;
2088 			case VDEV_AUX_BAD_GUID_SUM:
2089 				class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
2090 				break;
2091 			case VDEV_AUX_TOO_SMALL:
2092 				class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
2093 				break;
2094 			case VDEV_AUX_BAD_LABEL:
2095 				class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
2096 				break;
2097 			default:
2098 				class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
2099 			}
2100 
2101 			zfs_ereport_post(class, vd->vdev_spa,
2102 			    vd, NULL, save_state, 0);
2103 		}
2104 
2105 		/* Erase any notion of persistent removed state */
2106 		vd->vdev_removed = B_FALSE;
2107 	} else {
2108 		vd->vdev_removed = B_FALSE;
2109 	}
2110 
2111 	if (!isopen)
2112 		vdev_propagate_state(vd);
2113 }
2114