xref: /freebsd/sys/contrib/openzfs/module/zfs/vdev_label.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25  * Copyright (c) 2017, Intel Corporation.
26  */
27 
28 /*
29  * Virtual Device Labels
30  * ---------------------
31  *
32  * The vdev label serves several distinct purposes:
33  *
34  *	1. Uniquely identify this device as part of a ZFS pool and confirm its
35  *	   identity within the pool.
36  *
37  *	2. Verify that all the devices given in a configuration are present
38  *         within the pool.
39  *
40  *	3. Determine the uberblock for the pool.
41  *
42  *	4. In case of an import operation, determine the configuration of the
43  *         toplevel vdev of which it is a part.
44  *
45  *	5. If an import operation cannot find all the devices in the pool,
46  *         provide enough information to the administrator to determine which
47  *         devices are missing.
48  *
49  * It is important to note that while the kernel is responsible for writing the
50  * label, it only consumes the information in the first three cases.  The
51  * latter information is only consumed in userland when determining the
52  * configuration to import a pool.
53  *
54  *
55  * Label Organization
56  * ------------------
57  *
58  * Before describing the contents of the label, it's important to understand how
59  * the labels are written and updated with respect to the uberblock.
60  *
61  * When the pool configuration is altered, either because it was newly created
62  * or a device was added, we want to update all the labels such that we can deal
63  * with fatal failure at any point.  To this end, each disk has two labels which
64  * are updated before and after the uberblock is synced.  Assuming we have
65  * labels and an uberblock with the following transaction groups:
66  *
67  *              L1          UB          L2
68  *           +------+    +------+    +------+
69  *           |      |    |      |    |      |
70  *           | t10  |    | t10  |    | t10  |
71  *           |      |    |      |    |      |
72  *           +------+    +------+    +------+
73  *
74  * In this stable state, the labels and the uberblock were all updated within
75  * the same transaction group (10).  Each label is mirrored and checksummed, so
76  * that we can detect when we fail partway through writing the label.
77  *
78  * In order to identify which labels are valid, the labels are written in the
79  * following manner:
80  *
81  *	1. For each vdev, update 'L1' to the new label
82  *	2. Update the uberblock
83  *	3. For each vdev, update 'L2' to the new label
84  *
85  * Given arbitrary failure, we can determine the correct label to use based on
86  * the transaction group.  If we fail after updating L1 but before updating the
87  * UB, we will notice that L1's transaction group is greater than the uberblock,
88  * so L2 must be valid.  If we fail after writing the uberblock but before
89  * writing L2, we will notice that L2's transaction group is less than L1, and
90  * therefore L1 is valid.
91  *
92  * Another added complexity is that not every label is updated when the config
93  * is synced.  If we add a single device, we do not want to have to re-write
94  * every label for every device in the pool.  This means that both L1 and L2 may
95  * be older than the pool uberblock, because the necessary information is stored
96  * on another vdev.
97  *
98  *
99  * On-disk Format
100  * --------------
101  *
102  * The vdev label consists of two distinct parts, and is wrapped within the
103  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
104  * VTOC disk labels, but is otherwise ignored.
105  *
106  * The first half of the label is a packed nvlist which contains pool wide
107  * properties, per-vdev properties, and configuration information.  It is
108  * described in more detail below.
109  *
110  * The latter half of the label consists of a redundant array of uberblocks.
111  * These uberblocks are updated whenever a transaction group is committed,
112  * or when the configuration is updated.  When a pool is loaded, we scan each
113  * vdev for the 'best' uberblock.
114  *
115  *
116  * Configuration Information
117  * -------------------------
118  *
119  * The nvlist describing the pool and vdev contains the following elements:
120  *
121  *	version		ZFS on-disk version
122  *	name		Pool name
123  *	state		Pool state
124  *	txg		Transaction group in which this label was written
125  *	pool_guid	Unique identifier for this pool
126  *	vdev_tree	An nvlist describing vdev tree.
127  *	features_for_read
128  *			An nvlist of the features necessary for reading the MOS.
129  *
130  * Each leaf device label also contains the following:
131  *
132  *	top_guid	Unique ID for top-level vdev in which this is contained
133  *	guid		Unique ID for the leaf vdev
134  *
135  * The 'vs' configuration follows the format described in 'spa_config.c'.
136  */
137 
138 #include <sys/zfs_context.h>
139 #include <sys/spa.h>
140 #include <sys/spa_impl.h>
141 #include <sys/dmu.h>
142 #include <sys/zap.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/vdev_draid.h>
146 #include <sys/uberblock_impl.h>
147 #include <sys/metaslab.h>
148 #include <sys/metaslab_impl.h>
149 #include <sys/zio.h>
150 #include <sys/dsl_scan.h>
151 #include <sys/abd.h>
152 #include <sys/fs/zfs.h>
153 #include <sys/byteorder.h>
154 #include <sys/zfs_bootenv.h>
155 
156 /*
157  * Basic routines to read and write from a vdev label.
158  * Used throughout the rest of this file.
159  */
160 uint64_t
161 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
162 {
163 	ASSERT(offset < sizeof (vdev_label_t));
164 	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
165 
166 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
167 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
168 }
169 
170 /*
171  * Returns back the vdev label associated with the passed in offset.
172  */
173 int
174 vdev_label_number(uint64_t psize, uint64_t offset)
175 {
176 	int l;
177 
178 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
179 		offset -= psize - VDEV_LABEL_END_SIZE;
180 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
181 	}
182 	l = offset / sizeof (vdev_label_t);
183 	return (l < VDEV_LABELS ? l : -1);
184 }
185 
186 static void
187 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
188     uint64_t size, zio_done_func_t *done, void *private, int flags)
189 {
190 	ASSERT(
191 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
192 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
193 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
194 
195 	zio_nowait(zio_read_phys(zio, vd,
196 	    vdev_label_offset(vd->vdev_psize, l, offset),
197 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
198 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
199 }
200 
201 void
202 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
203     uint64_t size, zio_done_func_t *done, void *private, int flags)
204 {
205 	ASSERT(
206 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
207 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
208 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
209 
210 	zio_nowait(zio_write_phys(zio, vd,
211 	    vdev_label_offset(vd->vdev_psize, l, offset),
212 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
213 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
214 }
215 
216 /*
217  * Generate the nvlist representing this vdev's stats
218  */
219 void
220 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
221 {
222 	nvlist_t *nvx;
223 	vdev_stat_t *vs;
224 	vdev_stat_ex_t *vsx;
225 
226 	vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
227 	vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
228 
229 	vdev_get_stats_ex(vd, vs, vsx);
230 	fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
231 	    (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
232 
233 	/*
234 	 * Add extended stats into a special extended stats nvlist.  This keeps
235 	 * all the extended stats nicely grouped together.  The extended stats
236 	 * nvlist is then added to the main nvlist.
237 	 */
238 	nvx = fnvlist_alloc();
239 
240 	/* ZIOs in flight to disk */
241 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
242 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
243 
244 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
245 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
246 
247 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
248 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
249 
250 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
251 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
252 
253 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
254 	    vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
255 
256 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
257 	    vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
258 
259 	/* ZIOs pending */
260 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
261 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
262 
263 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
264 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
265 
266 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
267 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
268 
269 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
270 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
271 
272 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
273 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
274 
275 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
276 	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
277 
278 	/* Histograms */
279 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
280 	    vsx->vsx_total_histo[ZIO_TYPE_READ],
281 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
282 
283 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
284 	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
285 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
286 
287 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
288 	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
289 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
290 
291 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
292 	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
293 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
294 
295 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
296 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
297 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
298 
299 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
300 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
301 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
302 
303 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
304 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
305 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
306 
307 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
308 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
309 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
310 
311 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
312 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
313 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
314 
315 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
316 	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
317 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
318 
319 	/* Request sizes */
320 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
321 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
322 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
323 
324 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
325 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
326 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
327 
328 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
329 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
330 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
331 
332 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
333 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
334 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
335 
336 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
337 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
338 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
339 
340 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
341 	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
342 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
343 
344 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
345 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
346 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
347 
348 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
349 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
350 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
351 
352 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
353 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
354 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
355 
356 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
357 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
358 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
359 
360 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
361 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
362 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
363 
364 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
365 	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
366 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
367 
368 	/* IO delays */
369 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
370 
371 	/* Add extended stats nvlist to main nvlist */
372 	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
373 
374 	fnvlist_free(nvx);
375 	kmem_free(vs, sizeof (*vs));
376 	kmem_free(vsx, sizeof (*vsx));
377 }
378 
379 static void
380 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
381 {
382 	spa_t *spa = vd->vdev_spa;
383 
384 	if (vd != spa->spa_root_vdev)
385 		return;
386 
387 	/* provide either current or previous scan information */
388 	pool_scan_stat_t ps;
389 	if (spa_scan_get_stats(spa, &ps) == 0) {
390 		fnvlist_add_uint64_array(nvl,
391 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
392 		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
393 	}
394 
395 	pool_removal_stat_t prs;
396 	if (spa_removal_get_stats(spa, &prs) == 0) {
397 		fnvlist_add_uint64_array(nvl,
398 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
399 		    sizeof (prs) / sizeof (uint64_t));
400 	}
401 
402 	pool_checkpoint_stat_t pcs;
403 	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
404 		fnvlist_add_uint64_array(nvl,
405 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
406 		    sizeof (pcs) / sizeof (uint64_t));
407 	}
408 }
409 
410 static void
411 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
412 {
413 	if (vd == vd->vdev_top) {
414 		vdev_rebuild_stat_t vrs;
415 		if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
416 			fnvlist_add_uint64_array(nvl,
417 			    ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
418 			    sizeof (vrs) / sizeof (uint64_t));
419 		}
420 	}
421 }
422 
423 /*
424  * Generate the nvlist representing this vdev's config.
425  */
426 nvlist_t *
427 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
428     vdev_config_flag_t flags)
429 {
430 	nvlist_t *nv = NULL;
431 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
432 
433 	nv = fnvlist_alloc();
434 
435 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
436 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
437 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
438 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
439 
440 	if (vd->vdev_path != NULL)
441 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
442 
443 	if (vd->vdev_devid != NULL)
444 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
445 
446 	if (vd->vdev_physpath != NULL)
447 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
448 		    vd->vdev_physpath);
449 
450 	if (vd->vdev_enc_sysfs_path != NULL)
451 		fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
452 		    vd->vdev_enc_sysfs_path);
453 
454 	if (vd->vdev_fru != NULL)
455 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
456 
457 	if (vd->vdev_ops->vdev_op_config_generate != NULL)
458 		vd->vdev_ops->vdev_op_config_generate(vd, nv);
459 
460 	if (vd->vdev_wholedisk != -1ULL) {
461 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
462 		    vd->vdev_wholedisk);
463 	}
464 
465 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
466 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
467 
468 	if (vd->vdev_isspare)
469 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
470 
471 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
472 	    vd == vd->vdev_top) {
473 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
474 		    vd->vdev_ms_array);
475 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
476 		    vd->vdev_ms_shift);
477 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
478 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
479 		    vd->vdev_asize);
480 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
481 		if (vd->vdev_removing) {
482 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
483 			    vd->vdev_removing);
484 		}
485 
486 		/* zpool command expects alloc class data */
487 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
488 			const char *bias = NULL;
489 
490 			switch (vd->vdev_alloc_bias) {
491 			case VDEV_BIAS_LOG:
492 				bias = VDEV_ALLOC_BIAS_LOG;
493 				break;
494 			case VDEV_BIAS_SPECIAL:
495 				bias = VDEV_ALLOC_BIAS_SPECIAL;
496 				break;
497 			case VDEV_BIAS_DEDUP:
498 				bias = VDEV_ALLOC_BIAS_DEDUP;
499 				break;
500 			default:
501 				ASSERT3U(vd->vdev_alloc_bias, ==,
502 				    VDEV_BIAS_NONE);
503 			}
504 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
505 			    bias);
506 		}
507 	}
508 
509 	if (vd->vdev_dtl_sm != NULL) {
510 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
511 		    space_map_object(vd->vdev_dtl_sm));
512 	}
513 
514 	if (vic->vic_mapping_object != 0) {
515 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
516 		    vic->vic_mapping_object);
517 	}
518 
519 	if (vic->vic_births_object != 0) {
520 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
521 		    vic->vic_births_object);
522 	}
523 
524 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
525 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
526 		    vic->vic_prev_indirect_vdev);
527 	}
528 
529 	if (vd->vdev_crtxg)
530 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
531 
532 	if (vd->vdev_expansion_time)
533 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
534 		    vd->vdev_expansion_time);
535 
536 	if (flags & VDEV_CONFIG_MOS) {
537 		if (vd->vdev_leaf_zap != 0) {
538 			ASSERT(vd->vdev_ops->vdev_op_leaf);
539 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
540 			    vd->vdev_leaf_zap);
541 		}
542 
543 		if (vd->vdev_top_zap != 0) {
544 			ASSERT(vd == vd->vdev_top);
545 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
546 			    vd->vdev_top_zap);
547 		}
548 
549 		if (vd->vdev_resilver_deferred) {
550 			ASSERT(vd->vdev_ops->vdev_op_leaf);
551 			ASSERT(spa->spa_resilver_deferred);
552 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
553 		}
554 	}
555 
556 	if (getstats) {
557 		vdev_config_generate_stats(vd, nv);
558 
559 		root_vdev_actions_getprogress(vd, nv);
560 		top_vdev_actions_getprogress(vd, nv);
561 
562 		/*
563 		 * Note: this can be called from open context
564 		 * (spa_get_stats()), so we need the rwlock to prevent
565 		 * the mapping from being changed by condensing.
566 		 */
567 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
568 		if (vd->vdev_indirect_mapping != NULL) {
569 			ASSERT(vd->vdev_indirect_births != NULL);
570 			vdev_indirect_mapping_t *vim =
571 			    vd->vdev_indirect_mapping;
572 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
573 			    vdev_indirect_mapping_size(vim));
574 		}
575 		rw_exit(&vd->vdev_indirect_rwlock);
576 		if (vd->vdev_mg != NULL &&
577 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
578 			/*
579 			 * Compute approximately how much memory would be used
580 			 * for the indirect mapping if this device were to
581 			 * be removed.
582 			 *
583 			 * Note: If the frag metric is invalid, then not
584 			 * enough metaslabs have been converted to have
585 			 * histograms.
586 			 */
587 			uint64_t seg_count = 0;
588 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
589 
590 			/*
591 			 * There are the same number of allocated segments
592 			 * as free segments, so we will have at least one
593 			 * entry per free segment.  However, small free
594 			 * segments (smaller than vdev_removal_max_span)
595 			 * will be combined with adjacent allocated segments
596 			 * as a single mapping.
597 			 */
598 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
599 				if (i + 1 < highbit64(vdev_removal_max_span)
600 				    - 1) {
601 					to_alloc +=
602 					    vd->vdev_mg->mg_histogram[i] <<
603 					    (i + 1);
604 				} else {
605 					seg_count +=
606 					    vd->vdev_mg->mg_histogram[i];
607 				}
608 			}
609 
610 			/*
611 			 * The maximum length of a mapping is
612 			 * zfs_remove_max_segment, so we need at least one entry
613 			 * per zfs_remove_max_segment of allocated data.
614 			 */
615 			seg_count += to_alloc / spa_remove_max_segment(spa);
616 
617 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
618 			    seg_count *
619 			    sizeof (vdev_indirect_mapping_entry_phys_t));
620 		}
621 	}
622 
623 	if (!vd->vdev_ops->vdev_op_leaf) {
624 		nvlist_t **child;
625 		int c, idx;
626 
627 		ASSERT(!vd->vdev_ishole);
628 
629 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
630 		    KM_SLEEP);
631 
632 		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
633 			vdev_t *cvd = vd->vdev_child[c];
634 
635 			/*
636 			 * If we're generating an nvlist of removing
637 			 * vdevs then skip over any device which is
638 			 * not being removed.
639 			 */
640 			if ((flags & VDEV_CONFIG_REMOVING) &&
641 			    !cvd->vdev_removing)
642 				continue;
643 
644 			child[idx++] = vdev_config_generate(spa, cvd,
645 			    getstats, flags);
646 		}
647 
648 		if (idx) {
649 			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
650 			    child, idx);
651 		}
652 
653 		for (c = 0; c < idx; c++)
654 			nvlist_free(child[c]);
655 
656 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
657 
658 	} else {
659 		const char *aux = NULL;
660 
661 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
662 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
663 		if (vd->vdev_resilver_txg != 0)
664 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
665 			    vd->vdev_resilver_txg);
666 		if (vd->vdev_rebuild_txg != 0)
667 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
668 			    vd->vdev_rebuild_txg);
669 		if (vd->vdev_faulted)
670 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
671 		if (vd->vdev_degraded)
672 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
673 		if (vd->vdev_removed)
674 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
675 		if (vd->vdev_unspare)
676 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
677 		if (vd->vdev_ishole)
678 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
679 
680 		/* Set the reason why we're FAULTED/DEGRADED. */
681 		switch (vd->vdev_stat.vs_aux) {
682 		case VDEV_AUX_ERR_EXCEEDED:
683 			aux = "err_exceeded";
684 			break;
685 
686 		case VDEV_AUX_EXTERNAL:
687 			aux = "external";
688 			break;
689 		}
690 
691 		if (aux != NULL && !vd->vdev_tmpoffline) {
692 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
693 		} else {
694 			/*
695 			 * We're healthy - clear any previous AUX_STATE values.
696 			 */
697 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
698 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
699 		}
700 
701 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
702 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
703 			    vd->vdev_orig_guid);
704 		}
705 	}
706 
707 	return (nv);
708 }
709 
710 /*
711  * Generate a view of the top-level vdevs.  If we currently have holes
712  * in the namespace, then generate an array which contains a list of holey
713  * vdevs.  Additionally, add the number of top-level children that currently
714  * exist.
715  */
716 void
717 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
718 {
719 	vdev_t *rvd = spa->spa_root_vdev;
720 	uint64_t *array;
721 	uint_t c, idx;
722 
723 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
724 
725 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
726 		vdev_t *tvd = rvd->vdev_child[c];
727 
728 		if (tvd->vdev_ishole) {
729 			array[idx++] = c;
730 		}
731 	}
732 
733 	if (idx) {
734 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
735 		    array, idx) == 0);
736 	}
737 
738 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
739 	    rvd->vdev_children) == 0);
740 
741 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
742 }
743 
744 /*
745  * Returns the configuration from the label of the given vdev. For vdevs
746  * which don't have a txg value stored on their label (i.e. spares/cache)
747  * or have not been completely initialized (txg = 0) just return
748  * the configuration from the first valid label we find. Otherwise,
749  * find the most up-to-date label that does not exceed the specified
750  * 'txg' value.
751  */
752 nvlist_t *
753 vdev_label_read_config(vdev_t *vd, uint64_t txg)
754 {
755 	spa_t *spa = vd->vdev_spa;
756 	nvlist_t *config = NULL;
757 	vdev_phys_t *vp;
758 	abd_t *vp_abd;
759 	zio_t *zio;
760 	uint64_t best_txg = 0;
761 	uint64_t label_txg = 0;
762 	int error = 0;
763 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
764 	    ZIO_FLAG_SPECULATIVE;
765 
766 	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
767 
768 	if (!vdev_readable(vd))
769 		return (NULL);
770 
771 	/*
772 	 * The label for a dRAID distributed spare is not stored on disk.
773 	 * Instead it is generated when needed which allows us to bypass
774 	 * the pipeline when reading the config from the label.
775 	 */
776 	if (vd->vdev_ops == &vdev_draid_spare_ops)
777 		return (vdev_draid_read_config_spare(vd));
778 
779 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
780 	vp = abd_to_buf(vp_abd);
781 
782 retry:
783 	for (int l = 0; l < VDEV_LABELS; l++) {
784 		nvlist_t *label = NULL;
785 
786 		zio = zio_root(spa, NULL, NULL, flags);
787 
788 		vdev_label_read(zio, vd, l, vp_abd,
789 		    offsetof(vdev_label_t, vl_vdev_phys),
790 		    sizeof (vdev_phys_t), NULL, NULL, flags);
791 
792 		if (zio_wait(zio) == 0 &&
793 		    nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
794 		    &label, 0) == 0) {
795 			/*
796 			 * Auxiliary vdevs won't have txg values in their
797 			 * labels and newly added vdevs may not have been
798 			 * completely initialized so just return the
799 			 * configuration from the first valid label we
800 			 * encounter.
801 			 */
802 			error = nvlist_lookup_uint64(label,
803 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
804 			if ((error || label_txg == 0) && !config) {
805 				config = label;
806 				break;
807 			} else if (label_txg <= txg && label_txg > best_txg) {
808 				best_txg = label_txg;
809 				nvlist_free(config);
810 				config = fnvlist_dup(label);
811 			}
812 		}
813 
814 		if (label != NULL) {
815 			nvlist_free(label);
816 			label = NULL;
817 		}
818 	}
819 
820 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
821 		flags |= ZIO_FLAG_TRYHARD;
822 		goto retry;
823 	}
824 
825 	/*
826 	 * We found a valid label but it didn't pass txg restrictions.
827 	 */
828 	if (config == NULL && label_txg != 0) {
829 		vdev_dbgmsg(vd, "label discarded as txg is too large "
830 		    "(%llu > %llu)", (u_longlong_t)label_txg,
831 		    (u_longlong_t)txg);
832 	}
833 
834 	abd_free(vp_abd);
835 
836 	return (config);
837 }
838 
839 /*
840  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
841  * in with the device guid if this spare is active elsewhere on the system.
842  */
843 static boolean_t
844 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
845     uint64_t *spare_guid, uint64_t *l2cache_guid)
846 {
847 	spa_t *spa = vd->vdev_spa;
848 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
849 	uint64_t vdtxg = 0;
850 	nvlist_t *label;
851 
852 	if (spare_guid)
853 		*spare_guid = 0ULL;
854 	if (l2cache_guid)
855 		*l2cache_guid = 0ULL;
856 
857 	/*
858 	 * Read the label, if any, and perform some basic sanity checks.
859 	 */
860 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
861 		return (B_FALSE);
862 
863 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
864 	    &vdtxg);
865 
866 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
867 	    &state) != 0 ||
868 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
869 	    &device_guid) != 0) {
870 		nvlist_free(label);
871 		return (B_FALSE);
872 	}
873 
874 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
875 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
876 	    &pool_guid) != 0 ||
877 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
878 	    &txg) != 0)) {
879 		nvlist_free(label);
880 		return (B_FALSE);
881 	}
882 
883 	nvlist_free(label);
884 
885 	/*
886 	 * Check to see if this device indeed belongs to the pool it claims to
887 	 * be a part of.  The only way this is allowed is if the device is a hot
888 	 * spare (which we check for later on).
889 	 */
890 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
891 	    !spa_guid_exists(pool_guid, device_guid) &&
892 	    !spa_spare_exists(device_guid, NULL, NULL) &&
893 	    !spa_l2cache_exists(device_guid, NULL))
894 		return (B_FALSE);
895 
896 	/*
897 	 * If the transaction group is zero, then this an initialized (but
898 	 * unused) label.  This is only an error if the create transaction
899 	 * on-disk is the same as the one we're using now, in which case the
900 	 * user has attempted to add the same vdev multiple times in the same
901 	 * transaction.
902 	 */
903 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
904 	    txg == 0 && vdtxg == crtxg)
905 		return (B_TRUE);
906 
907 	/*
908 	 * Check to see if this is a spare device.  We do an explicit check for
909 	 * spa_has_spare() here because it may be on our pending list of spares
910 	 * to add.  We also check if it is an l2cache device.
911 	 */
912 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
913 	    spa_has_spare(spa, device_guid)) {
914 		if (spare_guid)
915 			*spare_guid = device_guid;
916 
917 		switch (reason) {
918 		case VDEV_LABEL_CREATE:
919 		case VDEV_LABEL_L2CACHE:
920 			return (B_TRUE);
921 
922 		case VDEV_LABEL_REPLACE:
923 			return (!spa_has_spare(spa, device_guid) ||
924 			    spare_pool != 0ULL);
925 
926 		case VDEV_LABEL_SPARE:
927 			return (spa_has_spare(spa, device_guid));
928 		default:
929 			break;
930 		}
931 	}
932 
933 	/*
934 	 * Check to see if this is an l2cache device.
935 	 */
936 	if (spa_l2cache_exists(device_guid, NULL))
937 		return (B_TRUE);
938 
939 	/*
940 	 * We can't rely on a pool's state if it's been imported
941 	 * read-only.  Instead we look to see if the pools is marked
942 	 * read-only in the namespace and set the state to active.
943 	 */
944 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
945 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
946 	    spa_mode(spa) == SPA_MODE_READ)
947 		state = POOL_STATE_ACTIVE;
948 
949 	/*
950 	 * If the device is marked ACTIVE, then this device is in use by another
951 	 * pool on the system.
952 	 */
953 	return (state == POOL_STATE_ACTIVE);
954 }
955 
956 /*
957  * Initialize a vdev label.  We check to make sure each leaf device is not in
958  * use, and writable.  We put down an initial label which we will later
959  * overwrite with a complete label.  Note that it's important to do this
960  * sequentially, not in parallel, so that we catch cases of multiple use of the
961  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
962  * itself.
963  */
964 int
965 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
966 {
967 	spa_t *spa = vd->vdev_spa;
968 	nvlist_t *label;
969 	vdev_phys_t *vp;
970 	abd_t *vp_abd;
971 	abd_t *bootenv;
972 	uberblock_t *ub;
973 	abd_t *ub_abd;
974 	zio_t *zio;
975 	char *buf;
976 	size_t buflen;
977 	int error;
978 	uint64_t spare_guid = 0, l2cache_guid = 0;
979 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
980 
981 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
982 
983 	for (int c = 0; c < vd->vdev_children; c++)
984 		if ((error = vdev_label_init(vd->vdev_child[c],
985 		    crtxg, reason)) != 0)
986 			return (error);
987 
988 	/* Track the creation time for this vdev */
989 	vd->vdev_crtxg = crtxg;
990 
991 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
992 		return (0);
993 
994 	/*
995 	 * Dead vdevs cannot be initialized.
996 	 */
997 	if (vdev_is_dead(vd))
998 		return (SET_ERROR(EIO));
999 
1000 	/*
1001 	 * Determine if the vdev is in use.
1002 	 */
1003 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1004 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1005 		return (SET_ERROR(EBUSY));
1006 
1007 	/*
1008 	 * If this is a request to add or replace a spare or l2cache device
1009 	 * that is in use elsewhere on the system, then we must update the
1010 	 * guid (which was initialized to a random value) to reflect the
1011 	 * actual GUID (which is shared between multiple pools).
1012 	 */
1013 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1014 	    spare_guid != 0ULL) {
1015 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1016 
1017 		vd->vdev_guid += guid_delta;
1018 
1019 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1020 			pvd->vdev_guid_sum += guid_delta;
1021 
1022 		/*
1023 		 * If this is a replacement, then we want to fallthrough to the
1024 		 * rest of the code.  If we're adding a spare, then it's already
1025 		 * labeled appropriately and we can just return.
1026 		 */
1027 		if (reason == VDEV_LABEL_SPARE)
1028 			return (0);
1029 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1030 		    reason == VDEV_LABEL_SPLIT);
1031 	}
1032 
1033 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1034 	    l2cache_guid != 0ULL) {
1035 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1036 
1037 		vd->vdev_guid += guid_delta;
1038 
1039 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1040 			pvd->vdev_guid_sum += guid_delta;
1041 
1042 		/*
1043 		 * If this is a replacement, then we want to fallthrough to the
1044 		 * rest of the code.  If we're adding an l2cache, then it's
1045 		 * already labeled appropriately and we can just return.
1046 		 */
1047 		if (reason == VDEV_LABEL_L2CACHE)
1048 			return (0);
1049 		ASSERT(reason == VDEV_LABEL_REPLACE);
1050 	}
1051 
1052 	/*
1053 	 * Initialize its label.
1054 	 */
1055 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1056 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1057 	vp = abd_to_buf(vp_abd);
1058 
1059 	/*
1060 	 * Generate a label describing the pool and our top-level vdev.
1061 	 * We mark it as being from txg 0 to indicate that it's not
1062 	 * really part of an active pool just yet.  The labels will
1063 	 * be written again with a meaningful txg by spa_sync().
1064 	 */
1065 	if (reason == VDEV_LABEL_SPARE ||
1066 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1067 		/*
1068 		 * For inactive hot spares, we generate a special label that
1069 		 * identifies as a mutually shared hot spare.  We write the
1070 		 * label if we are adding a hot spare, or if we are removing an
1071 		 * active hot spare (in which case we want to revert the
1072 		 * labels).
1073 		 */
1074 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1075 
1076 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1077 		    spa_version(spa)) == 0);
1078 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1079 		    POOL_STATE_SPARE) == 0);
1080 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1081 		    vd->vdev_guid) == 0);
1082 	} else if (reason == VDEV_LABEL_L2CACHE ||
1083 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1084 		/*
1085 		 * For level 2 ARC devices, add a special label.
1086 		 */
1087 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1088 
1089 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1090 		    spa_version(spa)) == 0);
1091 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1092 		    POOL_STATE_L2CACHE) == 0);
1093 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1094 		    vd->vdev_guid) == 0);
1095 	} else {
1096 		uint64_t txg = 0ULL;
1097 
1098 		if (reason == VDEV_LABEL_SPLIT)
1099 			txg = spa->spa_uberblock.ub_txg;
1100 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1101 
1102 		/*
1103 		 * Add our creation time.  This allows us to detect multiple
1104 		 * vdev uses as described above, and automatically expires if we
1105 		 * fail.
1106 		 */
1107 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1108 		    crtxg) == 0);
1109 	}
1110 
1111 	buf = vp->vp_nvlist;
1112 	buflen = sizeof (vp->vp_nvlist);
1113 
1114 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1115 	if (error != 0) {
1116 		nvlist_free(label);
1117 		abd_free(vp_abd);
1118 		/* EFAULT means nvlist_pack ran out of room */
1119 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1120 	}
1121 
1122 	/*
1123 	 * Initialize uberblock template.
1124 	 */
1125 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1126 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1127 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1128 	ub = abd_to_buf(ub_abd);
1129 	ub->ub_txg = 0;
1130 
1131 	/* Initialize the 2nd padding area. */
1132 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1133 	abd_zero(bootenv, VDEV_PAD_SIZE);
1134 
1135 	/*
1136 	 * Write everything in parallel.
1137 	 */
1138 retry:
1139 	zio = zio_root(spa, NULL, NULL, flags);
1140 
1141 	for (int l = 0; l < VDEV_LABELS; l++) {
1142 
1143 		vdev_label_write(zio, vd, l, vp_abd,
1144 		    offsetof(vdev_label_t, vl_vdev_phys),
1145 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1146 
1147 		/*
1148 		 * Skip the 1st padding area.
1149 		 * Zero out the 2nd padding area where it might have
1150 		 * left over data from previous filesystem format.
1151 		 */
1152 		vdev_label_write(zio, vd, l, bootenv,
1153 		    offsetof(vdev_label_t, vl_be),
1154 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1155 
1156 		vdev_label_write(zio, vd, l, ub_abd,
1157 		    offsetof(vdev_label_t, vl_uberblock),
1158 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1159 	}
1160 
1161 	error = zio_wait(zio);
1162 
1163 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1164 		flags |= ZIO_FLAG_TRYHARD;
1165 		goto retry;
1166 	}
1167 
1168 	nvlist_free(label);
1169 	abd_free(bootenv);
1170 	abd_free(ub_abd);
1171 	abd_free(vp_abd);
1172 
1173 	/*
1174 	 * If this vdev hasn't been previously identified as a spare, then we
1175 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1176 	 * exists as a spare elsewhere in the system.  Do the same for
1177 	 * level 2 ARC devices.
1178 	 */
1179 	if (error == 0 && !vd->vdev_isspare &&
1180 	    (reason == VDEV_LABEL_SPARE ||
1181 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1182 		spa_spare_add(vd);
1183 
1184 	if (error == 0 && !vd->vdev_isl2cache &&
1185 	    (reason == VDEV_LABEL_L2CACHE ||
1186 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1187 		spa_l2cache_add(vd);
1188 
1189 	return (error);
1190 }
1191 
1192 /*
1193  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1194  * callback to finish, store our abd in the callback pointer. Otherwise, we
1195  * just free our abd and return.
1196  */
1197 static void
1198 vdev_label_read_bootenv_done(zio_t *zio)
1199 {
1200 	zio_t *rio = zio->io_private;
1201 	abd_t **cbp = rio->io_private;
1202 
1203 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1204 
1205 	if (zio->io_error == 0) {
1206 		mutex_enter(&rio->io_lock);
1207 		if (*cbp == NULL) {
1208 			/* Will free this buffer in vdev_label_read_bootenv. */
1209 			*cbp = zio->io_abd;
1210 		} else {
1211 			abd_free(zio->io_abd);
1212 		}
1213 		mutex_exit(&rio->io_lock);
1214 	} else {
1215 		abd_free(zio->io_abd);
1216 	}
1217 }
1218 
1219 static void
1220 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1221 {
1222 	for (int c = 0; c < vd->vdev_children; c++)
1223 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1224 
1225 	/*
1226 	 * We just use the first label that has a correct checksum; the
1227 	 * bootloader should have rewritten them all to be the same on boot,
1228 	 * and any changes we made since boot have been the same across all
1229 	 * labels.
1230 	 */
1231 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1232 		for (int l = 0; l < VDEV_LABELS; l++) {
1233 			vdev_label_read(zio, vd, l,
1234 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1235 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1236 			    vdev_label_read_bootenv_done, zio, flags);
1237 		}
1238 	}
1239 }
1240 
1241 int
1242 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1243 {
1244 	nvlist_t *config;
1245 	spa_t *spa = rvd->vdev_spa;
1246 	abd_t *abd = NULL;
1247 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1248 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1249 
1250 	ASSERT(bootenv);
1251 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1252 
1253 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1254 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1255 	int err = zio_wait(zio);
1256 
1257 	if (abd != NULL) {
1258 		char *buf;
1259 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1260 
1261 		vbe->vbe_version = ntohll(vbe->vbe_version);
1262 		switch (vbe->vbe_version) {
1263 		case VB_RAW:
1264 			/*
1265 			 * if we have textual data in vbe_bootenv, create nvlist
1266 			 * with key "envmap".
1267 			 */
1268 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1269 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1270 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1271 			    vbe->vbe_bootenv);
1272 			break;
1273 
1274 		case VB_NVLIST:
1275 			err = nvlist_unpack(vbe->vbe_bootenv,
1276 			    sizeof (vbe->vbe_bootenv), &config, 0);
1277 			if (err == 0) {
1278 				fnvlist_merge(bootenv, config);
1279 				nvlist_free(config);
1280 				break;
1281 			}
1282 			/* FALLTHROUGH */
1283 		default:
1284 			/* Check for FreeBSD zfs bootonce command string */
1285 			buf = abd_to_buf(abd);
1286 			if (*buf == '\0') {
1287 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1288 				    VB_NVLIST);
1289 				break;
1290 			}
1291 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1292 		}
1293 
1294 		/*
1295 		 * abd was allocated in vdev_label_read_bootenv_impl()
1296 		 */
1297 		abd_free(abd);
1298 		/*
1299 		 * If we managed to read any successfully,
1300 		 * return success.
1301 		 */
1302 		return (0);
1303 	}
1304 	return (err);
1305 }
1306 
1307 int
1308 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1309 {
1310 	zio_t *zio;
1311 	spa_t *spa = vd->vdev_spa;
1312 	vdev_boot_envblock_t *bootenv;
1313 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1314 	int error;
1315 	size_t nvsize;
1316 	char *nvbuf;
1317 
1318 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1319 	if (error != 0)
1320 		return (SET_ERROR(error));
1321 
1322 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1323 		return (SET_ERROR(E2BIG));
1324 	}
1325 
1326 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1327 
1328 	error = ENXIO;
1329 	for (int c = 0; c < vd->vdev_children; c++) {
1330 		int child_err;
1331 
1332 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1333 		/*
1334 		 * As long as any of the disks managed to write all of their
1335 		 * labels successfully, return success.
1336 		 */
1337 		if (child_err == 0)
1338 			error = child_err;
1339 	}
1340 
1341 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1342 	    !vdev_writeable(vd)) {
1343 		return (error);
1344 	}
1345 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1346 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1347 	abd_zero(abd, VDEV_PAD_SIZE);
1348 
1349 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1350 	nvbuf = bootenv->vbe_bootenv;
1351 	nvsize = sizeof (bootenv->vbe_bootenv);
1352 
1353 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1354 	switch (bootenv->vbe_version) {
1355 	case VB_RAW:
1356 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) {
1357 			(void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize);
1358 		}
1359 		error = 0;
1360 		break;
1361 
1362 	case VB_NVLIST:
1363 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1364 		    KM_SLEEP);
1365 		break;
1366 
1367 	default:
1368 		error = EINVAL;
1369 		break;
1370 	}
1371 
1372 	if (error == 0) {
1373 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1374 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1375 	} else {
1376 		abd_free(abd);
1377 		return (SET_ERROR(error));
1378 	}
1379 
1380 retry:
1381 	zio = zio_root(spa, NULL, NULL, flags);
1382 	for (int l = 0; l < VDEV_LABELS; l++) {
1383 		vdev_label_write(zio, vd, l, abd,
1384 		    offsetof(vdev_label_t, vl_be),
1385 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1386 	}
1387 
1388 	error = zio_wait(zio);
1389 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1390 		flags |= ZIO_FLAG_TRYHARD;
1391 		goto retry;
1392 	}
1393 
1394 	abd_free(abd);
1395 	return (error);
1396 }
1397 
1398 /*
1399  * ==========================================================================
1400  * uberblock load/sync
1401  * ==========================================================================
1402  */
1403 
1404 /*
1405  * Consider the following situation: txg is safely synced to disk.  We've
1406  * written the first uberblock for txg + 1, and then we lose power.  When we
1407  * come back up, we fail to see the uberblock for txg + 1 because, say,
1408  * it was on a mirrored device and the replica to which we wrote txg + 1
1409  * is now offline.  If we then make some changes and sync txg + 1, and then
1410  * the missing replica comes back, then for a few seconds we'll have two
1411  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1412  * among uberblocks with equal txg, choose the one with the latest timestamp.
1413  */
1414 static int
1415 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1416 {
1417 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1418 
1419 	if (likely(cmp))
1420 		return (cmp);
1421 
1422 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1423 	if (likely(cmp))
1424 		return (cmp);
1425 
1426 	/*
1427 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1428 	 * ZFS, e.g. OpenZFS >= 0.7.
1429 	 *
1430 	 * If one ub has MMP and the other does not, they were written by
1431 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1432 	 * a 0 value.
1433 	 *
1434 	 * Since timestamp and txg are the same if we get this far, either is
1435 	 * acceptable for importing the pool.
1436 	 */
1437 	unsigned int seq1 = 0;
1438 	unsigned int seq2 = 0;
1439 
1440 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1441 		seq1 = MMP_SEQ(ub1);
1442 
1443 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1444 		seq2 = MMP_SEQ(ub2);
1445 
1446 	return (TREE_CMP(seq1, seq2));
1447 }
1448 
1449 struct ubl_cbdata {
1450 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1451 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1452 };
1453 
1454 static void
1455 vdev_uberblock_load_done(zio_t *zio)
1456 {
1457 	vdev_t *vd = zio->io_vd;
1458 	spa_t *spa = zio->io_spa;
1459 	zio_t *rio = zio->io_private;
1460 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1461 	struct ubl_cbdata *cbp = rio->io_private;
1462 
1463 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1464 
1465 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1466 		mutex_enter(&rio->io_lock);
1467 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1468 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1469 			/*
1470 			 * Keep track of the vdev in which this uberblock
1471 			 * was found. We will use this information later
1472 			 * to obtain the config nvlist associated with
1473 			 * this uberblock.
1474 			 */
1475 			*cbp->ubl_ubbest = *ub;
1476 			cbp->ubl_vd = vd;
1477 		}
1478 		mutex_exit(&rio->io_lock);
1479 	}
1480 
1481 	abd_free(zio->io_abd);
1482 }
1483 
1484 static void
1485 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1486     struct ubl_cbdata *cbp)
1487 {
1488 	for (int c = 0; c < vd->vdev_children; c++)
1489 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1490 
1491 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1492 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1493 		for (int l = 0; l < VDEV_LABELS; l++) {
1494 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1495 				vdev_label_read(zio, vd, l,
1496 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1497 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1498 				    VDEV_UBERBLOCK_SIZE(vd),
1499 				    vdev_uberblock_load_done, zio, flags);
1500 			}
1501 		}
1502 	}
1503 }
1504 
1505 /*
1506  * Reads the 'best' uberblock from disk along with its associated
1507  * configuration. First, we read the uberblock array of each label of each
1508  * vdev, keeping track of the uberblock with the highest txg in each array.
1509  * Then, we read the configuration from the same vdev as the best uberblock.
1510  */
1511 void
1512 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1513 {
1514 	zio_t *zio;
1515 	spa_t *spa = rvd->vdev_spa;
1516 	struct ubl_cbdata cb;
1517 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1518 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1519 
1520 	ASSERT(ub);
1521 	ASSERT(config);
1522 
1523 	bzero(ub, sizeof (uberblock_t));
1524 	*config = NULL;
1525 
1526 	cb.ubl_ubbest = ub;
1527 	cb.ubl_vd = NULL;
1528 
1529 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1530 	zio = zio_root(spa, NULL, &cb, flags);
1531 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1532 	(void) zio_wait(zio);
1533 
1534 	/*
1535 	 * It's possible that the best uberblock was discovered on a label
1536 	 * that has a configuration which was written in a future txg.
1537 	 * Search all labels on this vdev to find the configuration that
1538 	 * matches the txg for our uberblock.
1539 	 */
1540 	if (cb.ubl_vd != NULL) {
1541 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1542 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1543 
1544 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1545 		if (*config == NULL && spa->spa_extreme_rewind) {
1546 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1547 			    "Trying again without txg restrictions.");
1548 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1549 		}
1550 		if (*config == NULL) {
1551 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1552 		}
1553 	}
1554 	spa_config_exit(spa, SCL_ALL, FTAG);
1555 }
1556 
1557 /*
1558  * For use when a leaf vdev is expanded.
1559  * The location of labels 2 and 3 changed, and at the new location the
1560  * uberblock rings are either empty or contain garbage.  The sync will write
1561  * new configs there because the vdev is dirty, but expansion also needs the
1562  * uberblock rings copied.  Read them from label 0 which did not move.
1563  *
1564  * Since the point is to populate labels {2,3} with valid uberblocks,
1565  * we zero uberblocks we fail to read or which are not valid.
1566  */
1567 
1568 static void
1569 vdev_copy_uberblocks(vdev_t *vd)
1570 {
1571 	abd_t *ub_abd;
1572 	zio_t *write_zio;
1573 	int locks = (SCL_L2ARC | SCL_ZIO);
1574 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1575 	    ZIO_FLAG_SPECULATIVE;
1576 
1577 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1578 	    SCL_STATE);
1579 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1580 
1581 	/*
1582 	 * No uberblocks are stored on distributed spares, they may be
1583 	 * safely skipped when expanding a leaf vdev.
1584 	 */
1585 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1586 		return;
1587 
1588 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1589 
1590 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1591 
1592 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1593 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1594 		const int src_label = 0;
1595 		zio_t *zio;
1596 
1597 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1598 		vdev_label_read(zio, vd, src_label, ub_abd,
1599 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1600 		    NULL, NULL, flags);
1601 
1602 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1603 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1604 
1605 		for (int l = 2; l < VDEV_LABELS; l++)
1606 			vdev_label_write(write_zio, vd, l, ub_abd,
1607 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1608 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1609 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1610 	}
1611 	(void) zio_wait(write_zio);
1612 
1613 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1614 
1615 	abd_free(ub_abd);
1616 }
1617 
1618 /*
1619  * On success, increment root zio's count of good writes.
1620  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1621  */
1622 static void
1623 vdev_uberblock_sync_done(zio_t *zio)
1624 {
1625 	uint64_t *good_writes = zio->io_private;
1626 
1627 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1628 		atomic_inc_64(good_writes);
1629 }
1630 
1631 /*
1632  * Write the uberblock to all labels of all leaves of the specified vdev.
1633  */
1634 static void
1635 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1636     uberblock_t *ub, vdev_t *vd, int flags)
1637 {
1638 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1639 		vdev_uberblock_sync(zio, good_writes,
1640 		    ub, vd->vdev_child[c], flags);
1641 	}
1642 
1643 	if (!vd->vdev_ops->vdev_op_leaf)
1644 		return;
1645 
1646 	if (!vdev_writeable(vd))
1647 		return;
1648 
1649 	/*
1650 	 * There's no need to write uberblocks to a distributed spare, they
1651 	 * are already stored on all the leaves of the parent dRAID.  For
1652 	 * this same reason vdev_uberblock_load_impl() skips distributed
1653 	 * spares when reading uberblocks.
1654 	 */
1655 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1656 		return;
1657 
1658 	/* If the vdev was expanded, need to copy uberblock rings. */
1659 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1660 	    vd->vdev_copy_uberblocks == B_TRUE) {
1661 		vdev_copy_uberblocks(vd);
1662 		vd->vdev_copy_uberblocks = B_FALSE;
1663 	}
1664 
1665 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1666 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1667 
1668 	/* Copy the uberblock_t into the ABD */
1669 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1670 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1671 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1672 
1673 	for (int l = 0; l < VDEV_LABELS; l++)
1674 		vdev_label_write(zio, vd, l, ub_abd,
1675 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1676 		    vdev_uberblock_sync_done, good_writes,
1677 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1678 
1679 	abd_free(ub_abd);
1680 }
1681 
1682 /* Sync the uberblocks to all vdevs in svd[] */
1683 static int
1684 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1685 {
1686 	spa_t *spa = svd[0]->vdev_spa;
1687 	zio_t *zio;
1688 	uint64_t good_writes = 0;
1689 
1690 	zio = zio_root(spa, NULL, NULL, flags);
1691 
1692 	for (int v = 0; v < svdcount; v++)
1693 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1694 
1695 	(void) zio_wait(zio);
1696 
1697 	/*
1698 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1699 	 * are no longer needed (because the new uberblocks and the even
1700 	 * labels are safely on disk), so it is safe to overwrite them.
1701 	 */
1702 	zio = zio_root(spa, NULL, NULL, flags);
1703 
1704 	for (int v = 0; v < svdcount; v++) {
1705 		if (vdev_writeable(svd[v])) {
1706 			zio_flush(zio, svd[v]);
1707 		}
1708 	}
1709 
1710 	(void) zio_wait(zio);
1711 
1712 	return (good_writes >= 1 ? 0 : EIO);
1713 }
1714 
1715 /*
1716  * On success, increment the count of good writes for our top-level vdev.
1717  */
1718 static void
1719 vdev_label_sync_done(zio_t *zio)
1720 {
1721 	uint64_t *good_writes = zio->io_private;
1722 
1723 	if (zio->io_error == 0)
1724 		atomic_inc_64(good_writes);
1725 }
1726 
1727 /*
1728  * If there weren't enough good writes, indicate failure to the parent.
1729  */
1730 static void
1731 vdev_label_sync_top_done(zio_t *zio)
1732 {
1733 	uint64_t *good_writes = zio->io_private;
1734 
1735 	if (*good_writes == 0)
1736 		zio->io_error = SET_ERROR(EIO);
1737 
1738 	kmem_free(good_writes, sizeof (uint64_t));
1739 }
1740 
1741 /*
1742  * We ignore errors for log and cache devices, simply free the private data.
1743  */
1744 static void
1745 vdev_label_sync_ignore_done(zio_t *zio)
1746 {
1747 	kmem_free(zio->io_private, sizeof (uint64_t));
1748 }
1749 
1750 /*
1751  * Write all even or odd labels to all leaves of the specified vdev.
1752  */
1753 static void
1754 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1755     vdev_t *vd, int l, uint64_t txg, int flags)
1756 {
1757 	nvlist_t *label;
1758 	vdev_phys_t *vp;
1759 	abd_t *vp_abd;
1760 	char *buf;
1761 	size_t buflen;
1762 
1763 	for (int c = 0; c < vd->vdev_children; c++) {
1764 		vdev_label_sync(zio, good_writes,
1765 		    vd->vdev_child[c], l, txg, flags);
1766 	}
1767 
1768 	if (!vd->vdev_ops->vdev_op_leaf)
1769 		return;
1770 
1771 	if (!vdev_writeable(vd))
1772 		return;
1773 
1774 	/*
1775 	 * The top-level config never needs to be written to a distributed
1776 	 * spare.  When read vdev_dspare_label_read_config() will generate
1777 	 * the config for the vdev_label_read_config().
1778 	 */
1779 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1780 		return;
1781 
1782 	/*
1783 	 * Generate a label describing the top-level config to which we belong.
1784 	 */
1785 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1786 
1787 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1788 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1789 	vp = abd_to_buf(vp_abd);
1790 
1791 	buf = vp->vp_nvlist;
1792 	buflen = sizeof (vp->vp_nvlist);
1793 
1794 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1795 		for (; l < VDEV_LABELS; l += 2) {
1796 			vdev_label_write(zio, vd, l, vp_abd,
1797 			    offsetof(vdev_label_t, vl_vdev_phys),
1798 			    sizeof (vdev_phys_t),
1799 			    vdev_label_sync_done, good_writes,
1800 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1801 		}
1802 	}
1803 
1804 	abd_free(vp_abd);
1805 	nvlist_free(label);
1806 }
1807 
1808 static int
1809 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1810 {
1811 	list_t *dl = &spa->spa_config_dirty_list;
1812 	vdev_t *vd;
1813 	zio_t *zio;
1814 	int error;
1815 
1816 	/*
1817 	 * Write the new labels to disk.
1818 	 */
1819 	zio = zio_root(spa, NULL, NULL, flags);
1820 
1821 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1822 		uint64_t *good_writes;
1823 
1824 		ASSERT(!vd->vdev_ishole);
1825 
1826 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1827 		zio_t *vio = zio_null(zio, spa, NULL,
1828 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1829 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1830 		    good_writes, flags);
1831 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1832 		zio_nowait(vio);
1833 	}
1834 
1835 	error = zio_wait(zio);
1836 
1837 	/*
1838 	 * Flush the new labels to disk.
1839 	 */
1840 	zio = zio_root(spa, NULL, NULL, flags);
1841 
1842 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1843 		zio_flush(zio, vd);
1844 
1845 	(void) zio_wait(zio);
1846 
1847 	return (error);
1848 }
1849 
1850 /*
1851  * Sync the uberblock and any changes to the vdev configuration.
1852  *
1853  * The order of operations is carefully crafted to ensure that
1854  * if the system panics or loses power at any time, the state on disk
1855  * is still transactionally consistent.  The in-line comments below
1856  * describe the failure semantics at each stage.
1857  *
1858  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1859  * at any time, you can just call it again, and it will resume its work.
1860  */
1861 int
1862 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1863 {
1864 	spa_t *spa = svd[0]->vdev_spa;
1865 	uberblock_t *ub = &spa->spa_uberblock;
1866 	int error = 0;
1867 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1868 
1869 	ASSERT(svdcount != 0);
1870 retry:
1871 	/*
1872 	 * Normally, we don't want to try too hard to write every label and
1873 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1874 	 * sync process to block while we retry.  But if we can't write a
1875 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1876 	 * bailing out and declaring the pool faulted.
1877 	 */
1878 	if (error != 0) {
1879 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1880 			return (error);
1881 		flags |= ZIO_FLAG_TRYHARD;
1882 	}
1883 
1884 	ASSERT(ub->ub_txg <= txg);
1885 
1886 	/*
1887 	 * If this isn't a resync due to I/O errors,
1888 	 * and nothing changed in this transaction group,
1889 	 * and the vdev configuration hasn't changed,
1890 	 * then there's nothing to do.
1891 	 */
1892 	if (ub->ub_txg < txg) {
1893 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1894 		    txg, spa->spa_mmp.mmp_delay);
1895 
1896 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1897 			return (0);
1898 	}
1899 
1900 	if (txg > spa_freeze_txg(spa))
1901 		return (0);
1902 
1903 	ASSERT(txg <= spa->spa_final_txg);
1904 
1905 	/*
1906 	 * Flush the write cache of every disk that's been written to
1907 	 * in this transaction group.  This ensures that all blocks
1908 	 * written in this txg will be committed to stable storage
1909 	 * before any uberblock that references them.
1910 	 */
1911 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1912 
1913 	for (vdev_t *vd =
1914 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1915 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1916 		zio_flush(zio, vd);
1917 
1918 	(void) zio_wait(zio);
1919 
1920 	/*
1921 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1922 	 * system dies in the middle of this process, that's OK: all of the
1923 	 * even labels that made it to disk will be newer than any uberblock,
1924 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1925 	 * which have not yet been touched, will still be valid.  We flush
1926 	 * the new labels to disk to ensure that all even-label updates
1927 	 * are committed to stable storage before the uberblock update.
1928 	 */
1929 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1930 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1931 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1932 			    "for pool '%s' when syncing out the even labels "
1933 			    "of dirty vdevs", error, spa_name(spa));
1934 		}
1935 		goto retry;
1936 	}
1937 
1938 	/*
1939 	 * Sync the uberblocks to all vdevs in svd[].
1940 	 * If the system dies in the middle of this step, there are two cases
1941 	 * to consider, and the on-disk state is consistent either way:
1942 	 *
1943 	 * (1)	If none of the new uberblocks made it to disk, then the
1944 	 *	previous uberblock will be the newest, and the odd labels
1945 	 *	(which had not yet been touched) will be valid with respect
1946 	 *	to that uberblock.
1947 	 *
1948 	 * (2)	If one or more new uberblocks made it to disk, then they
1949 	 *	will be the newest, and the even labels (which had all
1950 	 *	been successfully committed) will be valid with respect
1951 	 *	to the new uberblocks.
1952 	 */
1953 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1954 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1955 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1956 			    "%d for pool '%s'", error, spa_name(spa));
1957 		}
1958 		goto retry;
1959 	}
1960 
1961 	if (spa_multihost(spa))
1962 		mmp_update_uberblock(spa, ub);
1963 
1964 	/*
1965 	 * Sync out odd labels for every dirty vdev.  If the system dies
1966 	 * in the middle of this process, the even labels and the new
1967 	 * uberblocks will suffice to open the pool.  The next time
1968 	 * the pool is opened, the first thing we'll do -- before any
1969 	 * user data is modified -- is mark every vdev dirty so that
1970 	 * all labels will be brought up to date.  We flush the new labels
1971 	 * to disk to ensure that all odd-label updates are committed to
1972 	 * stable storage before the next transaction group begins.
1973 	 */
1974 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1975 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1976 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1977 			    "for pool '%s' when syncing out the odd labels of "
1978 			    "dirty vdevs", error, spa_name(spa));
1979 		}
1980 		goto retry;
1981 	}
1982 
1983 	return (0);
1984 }
1985