xref: /freebsd/sys/contrib/openzfs/module/zfs/vdev_label.c (revision ba3c1f5972d7b90feb6e6da47905ff2757e0fe57)
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 https://opensource.org/licenses/CDDL-1.0.
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 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,
260 	    vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]);
261 
262 	/* ZIOs pending */
263 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
264 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
265 
266 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
267 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
268 
269 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
270 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
271 
272 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
273 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
274 
275 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
276 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
277 
278 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
279 	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
280 
281 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,
282 	    vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]);
283 
284 	/* Histograms */
285 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
286 	    vsx->vsx_total_histo[ZIO_TYPE_READ],
287 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
288 
289 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
290 	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
291 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
292 
293 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
294 	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
295 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
296 
297 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
298 	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
299 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
300 
301 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
302 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
303 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
304 
305 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
306 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
307 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
308 
309 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
310 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
311 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
312 
313 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
314 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
315 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
316 
317 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
318 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
319 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
320 
321 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
322 	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
323 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
324 
325 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,
326 	    vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD],
327 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD]));
328 
329 	/* Request sizes */
330 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
331 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
332 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
333 
334 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
335 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
336 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
337 
338 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
339 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
340 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
341 
342 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
343 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
344 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
345 
346 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
347 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
348 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
349 
350 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
351 	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
352 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
353 
354 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,
355 	    vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD],
356 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD]));
357 
358 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
359 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
360 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
361 
362 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
363 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
364 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
365 
366 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
367 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
368 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
369 
370 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
371 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
372 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
373 
374 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
375 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
376 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
377 
378 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
379 	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
380 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
381 
382 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,
383 	    vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD],
384 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD]));
385 
386 	/* IO delays */
387 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
388 
389 	/* Add extended stats nvlist to main nvlist */
390 	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
391 
392 	fnvlist_free(nvx);
393 	kmem_free(vs, sizeof (*vs));
394 	kmem_free(vsx, sizeof (*vsx));
395 }
396 
397 static void
398 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
399 {
400 	spa_t *spa = vd->vdev_spa;
401 
402 	if (vd != spa->spa_root_vdev)
403 		return;
404 
405 	/* provide either current or previous scan information */
406 	pool_scan_stat_t ps;
407 	if (spa_scan_get_stats(spa, &ps) == 0) {
408 		fnvlist_add_uint64_array(nvl,
409 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
410 		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
411 	}
412 
413 	pool_removal_stat_t prs;
414 	if (spa_removal_get_stats(spa, &prs) == 0) {
415 		fnvlist_add_uint64_array(nvl,
416 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
417 		    sizeof (prs) / sizeof (uint64_t));
418 	}
419 
420 	pool_checkpoint_stat_t pcs;
421 	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
422 		fnvlist_add_uint64_array(nvl,
423 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
424 		    sizeof (pcs) / sizeof (uint64_t));
425 	}
426 }
427 
428 static void
429 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
430 {
431 	if (vd == vd->vdev_top) {
432 		vdev_rebuild_stat_t vrs;
433 		if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
434 			fnvlist_add_uint64_array(nvl,
435 			    ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
436 			    sizeof (vrs) / sizeof (uint64_t));
437 		}
438 	}
439 }
440 
441 /*
442  * Generate the nvlist representing this vdev's config.
443  */
444 nvlist_t *
445 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
446     vdev_config_flag_t flags)
447 {
448 	nvlist_t *nv = NULL;
449 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
450 
451 	nv = fnvlist_alloc();
452 
453 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
454 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
455 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
456 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
457 
458 	if (vd->vdev_path != NULL)
459 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
460 
461 	if (vd->vdev_devid != NULL)
462 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
463 
464 	if (vd->vdev_physpath != NULL)
465 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
466 		    vd->vdev_physpath);
467 
468 	if (vd->vdev_enc_sysfs_path != NULL)
469 		fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
470 		    vd->vdev_enc_sysfs_path);
471 
472 	if (vd->vdev_fru != NULL)
473 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
474 
475 	if (vd->vdev_ops->vdev_op_config_generate != NULL)
476 		vd->vdev_ops->vdev_op_config_generate(vd, nv);
477 
478 	if (vd->vdev_wholedisk != -1ULL) {
479 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
480 		    vd->vdev_wholedisk);
481 	}
482 
483 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
484 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
485 
486 	if (vd->vdev_isspare)
487 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
488 
489 	if (flags & VDEV_CONFIG_L2CACHE)
490 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
491 
492 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
493 	    vd == vd->vdev_top) {
494 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
495 		    vd->vdev_ms_array);
496 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
497 		    vd->vdev_ms_shift);
498 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
499 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
500 		    vd->vdev_asize);
501 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
502 		if (vd->vdev_noalloc) {
503 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
504 			    vd->vdev_noalloc);
505 		}
506 
507 		/*
508 		 * Slog devices are removed synchronously so don't
509 		 * persist the vdev_removing flag to the label.
510 		 */
511 		if (vd->vdev_removing && !vd->vdev_islog) {
512 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
513 			    vd->vdev_removing);
514 		}
515 
516 		/* zpool command expects alloc class data */
517 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
518 			const char *bias = NULL;
519 
520 			switch (vd->vdev_alloc_bias) {
521 			case VDEV_BIAS_LOG:
522 				bias = VDEV_ALLOC_BIAS_LOG;
523 				break;
524 			case VDEV_BIAS_SPECIAL:
525 				bias = VDEV_ALLOC_BIAS_SPECIAL;
526 				break;
527 			case VDEV_BIAS_DEDUP:
528 				bias = VDEV_ALLOC_BIAS_DEDUP;
529 				break;
530 			default:
531 				ASSERT3U(vd->vdev_alloc_bias, ==,
532 				    VDEV_BIAS_NONE);
533 			}
534 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
535 			    bias);
536 		}
537 	}
538 
539 	if (vd->vdev_dtl_sm != NULL) {
540 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
541 		    space_map_object(vd->vdev_dtl_sm));
542 	}
543 
544 	if (vic->vic_mapping_object != 0) {
545 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
546 		    vic->vic_mapping_object);
547 	}
548 
549 	if (vic->vic_births_object != 0) {
550 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
551 		    vic->vic_births_object);
552 	}
553 
554 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
555 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
556 		    vic->vic_prev_indirect_vdev);
557 	}
558 
559 	if (vd->vdev_crtxg)
560 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
561 
562 	if (vd->vdev_expansion_time)
563 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
564 		    vd->vdev_expansion_time);
565 
566 	if (flags & VDEV_CONFIG_MOS) {
567 		if (vd->vdev_leaf_zap != 0) {
568 			ASSERT(vd->vdev_ops->vdev_op_leaf);
569 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
570 			    vd->vdev_leaf_zap);
571 		}
572 
573 		if (vd->vdev_top_zap != 0) {
574 			ASSERT(vd == vd->vdev_top);
575 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
576 			    vd->vdev_top_zap);
577 		}
578 
579 		if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
580 		    spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
581 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
582 			    vd->vdev_root_zap);
583 		}
584 
585 		if (vd->vdev_resilver_deferred) {
586 			ASSERT(vd->vdev_ops->vdev_op_leaf);
587 			ASSERT(spa->spa_resilver_deferred);
588 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
589 		}
590 	}
591 
592 	if (getstats) {
593 		vdev_config_generate_stats(vd, nv);
594 
595 		root_vdev_actions_getprogress(vd, nv);
596 		top_vdev_actions_getprogress(vd, nv);
597 
598 		/*
599 		 * Note: this can be called from open context
600 		 * (spa_get_stats()), so we need the rwlock to prevent
601 		 * the mapping from being changed by condensing.
602 		 */
603 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
604 		if (vd->vdev_indirect_mapping != NULL) {
605 			ASSERT(vd->vdev_indirect_births != NULL);
606 			vdev_indirect_mapping_t *vim =
607 			    vd->vdev_indirect_mapping;
608 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
609 			    vdev_indirect_mapping_size(vim));
610 		}
611 		rw_exit(&vd->vdev_indirect_rwlock);
612 		if (vd->vdev_mg != NULL &&
613 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
614 			/*
615 			 * Compute approximately how much memory would be used
616 			 * for the indirect mapping if this device were to
617 			 * be removed.
618 			 *
619 			 * Note: If the frag metric is invalid, then not
620 			 * enough metaslabs have been converted to have
621 			 * histograms.
622 			 */
623 			uint64_t seg_count = 0;
624 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
625 
626 			/*
627 			 * There are the same number of allocated segments
628 			 * as free segments, so we will have at least one
629 			 * entry per free segment.  However, small free
630 			 * segments (smaller than vdev_removal_max_span)
631 			 * will be combined with adjacent allocated segments
632 			 * as a single mapping.
633 			 */
634 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
635 				if (i + 1 < highbit64(vdev_removal_max_span)
636 				    - 1) {
637 					to_alloc +=
638 					    vd->vdev_mg->mg_histogram[i] <<
639 					    (i + 1);
640 				} else {
641 					seg_count +=
642 					    vd->vdev_mg->mg_histogram[i];
643 				}
644 			}
645 
646 			/*
647 			 * The maximum length of a mapping is
648 			 * zfs_remove_max_segment, so we need at least one entry
649 			 * per zfs_remove_max_segment of allocated data.
650 			 */
651 			seg_count += to_alloc / spa_remove_max_segment(spa);
652 
653 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
654 			    seg_count *
655 			    sizeof (vdev_indirect_mapping_entry_phys_t));
656 		}
657 	}
658 
659 	if (!vd->vdev_ops->vdev_op_leaf) {
660 		nvlist_t **child;
661 		uint64_t c;
662 
663 		ASSERT(!vd->vdev_ishole);
664 
665 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
666 		    KM_SLEEP);
667 
668 		for (c = 0; c < vd->vdev_children; c++) {
669 			child[c] = vdev_config_generate(spa, vd->vdev_child[c],
670 			    getstats, flags);
671 		}
672 
673 		fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
674 		    (const nvlist_t * const *)child, vd->vdev_children);
675 
676 		for (c = 0; c < vd->vdev_children; c++)
677 			nvlist_free(child[c]);
678 
679 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
680 
681 	} else {
682 		const char *aux = NULL;
683 
684 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
685 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
686 		if (vd->vdev_resilver_txg != 0)
687 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
688 			    vd->vdev_resilver_txg);
689 		if (vd->vdev_rebuild_txg != 0)
690 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
691 			    vd->vdev_rebuild_txg);
692 		if (vd->vdev_faulted)
693 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
694 		if (vd->vdev_degraded)
695 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
696 		if (vd->vdev_removed)
697 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
698 		if (vd->vdev_unspare)
699 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
700 		if (vd->vdev_ishole)
701 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
702 
703 		/* Set the reason why we're FAULTED/DEGRADED. */
704 		switch (vd->vdev_stat.vs_aux) {
705 		case VDEV_AUX_ERR_EXCEEDED:
706 			aux = "err_exceeded";
707 			break;
708 
709 		case VDEV_AUX_EXTERNAL:
710 			aux = "external";
711 			break;
712 		}
713 
714 		if (aux != NULL && !vd->vdev_tmpoffline) {
715 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
716 		} else {
717 			/*
718 			 * We're healthy - clear any previous AUX_STATE values.
719 			 */
720 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
721 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
722 		}
723 
724 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
725 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
726 			    vd->vdev_orig_guid);
727 		}
728 	}
729 
730 	return (nv);
731 }
732 
733 /*
734  * Generate a view of the top-level vdevs.  If we currently have holes
735  * in the namespace, then generate an array which contains a list of holey
736  * vdevs.  Additionally, add the number of top-level children that currently
737  * exist.
738  */
739 void
740 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
741 {
742 	vdev_t *rvd = spa->spa_root_vdev;
743 	uint64_t *array;
744 	uint_t c, idx;
745 
746 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
747 
748 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
749 		vdev_t *tvd = rvd->vdev_child[c];
750 
751 		if (tvd->vdev_ishole) {
752 			array[idx++] = c;
753 		}
754 	}
755 
756 	if (idx) {
757 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
758 		    array, idx) == 0);
759 	}
760 
761 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
762 	    rvd->vdev_children) == 0);
763 
764 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
765 }
766 
767 /*
768  * Returns the configuration from the label of the given vdev. For vdevs
769  * which don't have a txg value stored on their label (i.e. spares/cache)
770  * or have not been completely initialized (txg = 0) just return
771  * the configuration from the first valid label we find. Otherwise,
772  * find the most up-to-date label that does not exceed the specified
773  * 'txg' value.
774  */
775 nvlist_t *
776 vdev_label_read_config(vdev_t *vd, uint64_t txg)
777 {
778 	spa_t *spa = vd->vdev_spa;
779 	nvlist_t *config = NULL;
780 	vdev_phys_t *vp[VDEV_LABELS];
781 	abd_t *vp_abd[VDEV_LABELS];
782 	zio_t *zio[VDEV_LABELS];
783 	uint64_t best_txg = 0;
784 	uint64_t label_txg = 0;
785 	int error = 0;
786 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
787 	    ZIO_FLAG_SPECULATIVE;
788 
789 	ASSERT(vd->vdev_validate_thread == curthread ||
790 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
791 
792 	if (!vdev_readable(vd))
793 		return (NULL);
794 
795 	/*
796 	 * The label for a dRAID distributed spare is not stored on disk.
797 	 * Instead it is generated when needed which allows us to bypass
798 	 * the pipeline when reading the config from the label.
799 	 */
800 	if (vd->vdev_ops == &vdev_draid_spare_ops)
801 		return (vdev_draid_read_config_spare(vd));
802 
803 	for (int l = 0; l < VDEV_LABELS; l++) {
804 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
805 		vp[l] = abd_to_buf(vp_abd[l]);
806 	}
807 
808 retry:
809 	for (int l = 0; l < VDEV_LABELS; l++) {
810 		zio[l] = zio_root(spa, NULL, NULL, flags);
811 
812 		vdev_label_read(zio[l], vd, l, vp_abd[l],
813 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
814 		    NULL, NULL, flags);
815 	}
816 	for (int l = 0; l < VDEV_LABELS; l++) {
817 		nvlist_t *label = NULL;
818 
819 		if (zio_wait(zio[l]) == 0 &&
820 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
821 		    &label, 0) == 0) {
822 			/*
823 			 * Auxiliary vdevs won't have txg values in their
824 			 * labels and newly added vdevs may not have been
825 			 * completely initialized so just return the
826 			 * configuration from the first valid label we
827 			 * encounter.
828 			 */
829 			error = nvlist_lookup_uint64(label,
830 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
831 			if ((error || label_txg == 0) && !config) {
832 				config = label;
833 				for (l++; l < VDEV_LABELS; l++)
834 					zio_wait(zio[l]);
835 				break;
836 			} else if (label_txg <= txg && label_txg > best_txg) {
837 				best_txg = label_txg;
838 				nvlist_free(config);
839 				config = fnvlist_dup(label);
840 			}
841 		}
842 
843 		if (label != NULL) {
844 			nvlist_free(label);
845 			label = NULL;
846 		}
847 	}
848 
849 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
850 		flags |= ZIO_FLAG_TRYHARD;
851 		goto retry;
852 	}
853 
854 	/*
855 	 * We found a valid label but it didn't pass txg restrictions.
856 	 */
857 	if (config == NULL && label_txg != 0) {
858 		vdev_dbgmsg(vd, "label discarded as txg is too large "
859 		    "(%llu > %llu)", (u_longlong_t)label_txg,
860 		    (u_longlong_t)txg);
861 	}
862 
863 	for (int l = 0; l < VDEV_LABELS; l++) {
864 		abd_free(vp_abd[l]);
865 	}
866 
867 	return (config);
868 }
869 
870 /*
871  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
872  * in with the device guid if this spare is active elsewhere on the system.
873  */
874 static boolean_t
875 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
876     uint64_t *spare_guid, uint64_t *l2cache_guid)
877 {
878 	spa_t *spa = vd->vdev_spa;
879 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
880 	uint64_t vdtxg = 0;
881 	nvlist_t *label;
882 
883 	if (spare_guid)
884 		*spare_guid = 0ULL;
885 	if (l2cache_guid)
886 		*l2cache_guid = 0ULL;
887 
888 	/*
889 	 * Read the label, if any, and perform some basic sanity checks.
890 	 */
891 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
892 		return (B_FALSE);
893 
894 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
895 	    &vdtxg);
896 
897 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
898 	    &state) != 0 ||
899 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
900 	    &device_guid) != 0) {
901 		nvlist_free(label);
902 		return (B_FALSE);
903 	}
904 
905 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
906 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
907 	    &pool_guid) != 0 ||
908 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
909 	    &txg) != 0)) {
910 		nvlist_free(label);
911 		return (B_FALSE);
912 	}
913 
914 	nvlist_free(label);
915 
916 	/*
917 	 * Check to see if this device indeed belongs to the pool it claims to
918 	 * be a part of.  The only way this is allowed is if the device is a hot
919 	 * spare (which we check for later on).
920 	 */
921 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
922 	    !spa_guid_exists(pool_guid, device_guid) &&
923 	    !spa_spare_exists(device_guid, NULL, NULL) &&
924 	    !spa_l2cache_exists(device_guid, NULL))
925 		return (B_FALSE);
926 
927 	/*
928 	 * If the transaction group is zero, then this an initialized (but
929 	 * unused) label.  This is only an error if the create transaction
930 	 * on-disk is the same as the one we're using now, in which case the
931 	 * user has attempted to add the same vdev multiple times in the same
932 	 * transaction.
933 	 */
934 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
935 	    txg == 0 && vdtxg == crtxg)
936 		return (B_TRUE);
937 
938 	/*
939 	 * Check to see if this is a spare device.  We do an explicit check for
940 	 * spa_has_spare() here because it may be on our pending list of spares
941 	 * to add.
942 	 */
943 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
944 	    spa_has_spare(spa, device_guid)) {
945 		if (spare_guid)
946 			*spare_guid = device_guid;
947 
948 		switch (reason) {
949 		case VDEV_LABEL_CREATE:
950 			return (B_TRUE);
951 
952 		case VDEV_LABEL_REPLACE:
953 			return (!spa_has_spare(spa, device_guid) ||
954 			    spare_pool != 0ULL);
955 
956 		case VDEV_LABEL_SPARE:
957 			return (spa_has_spare(spa, device_guid));
958 		default:
959 			break;
960 		}
961 	}
962 
963 	/*
964 	 * Check to see if this is an l2cache device.
965 	 */
966 	if (spa_l2cache_exists(device_guid, NULL) ||
967 	    spa_has_l2cache(spa, device_guid)) {
968 		if (l2cache_guid)
969 			*l2cache_guid = device_guid;
970 
971 		switch (reason) {
972 		case VDEV_LABEL_CREATE:
973 			return (B_TRUE);
974 
975 		case VDEV_LABEL_REPLACE:
976 			return (!spa_has_l2cache(spa, device_guid));
977 
978 		case VDEV_LABEL_L2CACHE:
979 			return (spa_has_l2cache(spa, device_guid));
980 		default:
981 			break;
982 		}
983 	}
984 
985 	/*
986 	 * We can't rely on a pool's state if it's been imported
987 	 * read-only.  Instead we look to see if the pools is marked
988 	 * read-only in the namespace and set the state to active.
989 	 */
990 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
991 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
992 	    spa_mode(spa) == SPA_MODE_READ)
993 		state = POOL_STATE_ACTIVE;
994 
995 	/*
996 	 * If the device is marked ACTIVE, then this device is in use by another
997 	 * pool on the system.
998 	 */
999 	return (state == POOL_STATE_ACTIVE);
1000 }
1001 
1002 /*
1003  * Initialize a vdev label.  We check to make sure each leaf device is not in
1004  * use, and writable.  We put down an initial label which we will later
1005  * overwrite with a complete label.  Note that it's important to do this
1006  * sequentially, not in parallel, so that we catch cases of multiple use of the
1007  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1008  * itself.
1009  */
1010 int
1011 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1012 {
1013 	spa_t *spa = vd->vdev_spa;
1014 	nvlist_t *label;
1015 	vdev_phys_t *vp;
1016 	abd_t *vp_abd;
1017 	abd_t *bootenv;
1018 	uberblock_t *ub;
1019 	abd_t *ub_abd;
1020 	zio_t *zio;
1021 	char *buf;
1022 	size_t buflen;
1023 	int error;
1024 	uint64_t spare_guid = 0, l2cache_guid = 0;
1025 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1026 
1027 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1028 
1029 	for (int c = 0; c < vd->vdev_children; c++)
1030 		if ((error = vdev_label_init(vd->vdev_child[c],
1031 		    crtxg, reason)) != 0)
1032 			return (error);
1033 
1034 	/* Track the creation time for this vdev */
1035 	vd->vdev_crtxg = crtxg;
1036 
1037 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1038 		return (0);
1039 
1040 	/*
1041 	 * Dead vdevs cannot be initialized.
1042 	 */
1043 	if (vdev_is_dead(vd))
1044 		return (SET_ERROR(EIO));
1045 
1046 	/*
1047 	 * Determine if the vdev is in use.
1048 	 */
1049 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1050 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1051 		return (SET_ERROR(EBUSY));
1052 
1053 	/*
1054 	 * If this is a request to add or replace a spare or l2cache device
1055 	 * that is in use elsewhere on the system, then we must update the
1056 	 * guid (which was initialized to a random value) to reflect the
1057 	 * actual GUID (which is shared between multiple pools).
1058 	 */
1059 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1060 	    spare_guid != 0ULL) {
1061 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1062 
1063 		vd->vdev_guid += guid_delta;
1064 
1065 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1066 			pvd->vdev_guid_sum += guid_delta;
1067 
1068 		/*
1069 		 * If this is a replacement, then we want to fallthrough to the
1070 		 * rest of the code.  If we're adding a spare, then it's already
1071 		 * labeled appropriately and we can just return.
1072 		 */
1073 		if (reason == VDEV_LABEL_SPARE)
1074 			return (0);
1075 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1076 		    reason == VDEV_LABEL_SPLIT);
1077 	}
1078 
1079 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1080 	    l2cache_guid != 0ULL) {
1081 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1082 
1083 		vd->vdev_guid += guid_delta;
1084 
1085 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1086 			pvd->vdev_guid_sum += guid_delta;
1087 
1088 		/*
1089 		 * If this is a replacement, then we want to fallthrough to the
1090 		 * rest of the code.  If we're adding an l2cache, then it's
1091 		 * already labeled appropriately and we can just return.
1092 		 */
1093 		if (reason == VDEV_LABEL_L2CACHE)
1094 			return (0);
1095 		ASSERT(reason == VDEV_LABEL_REPLACE);
1096 	}
1097 
1098 	/*
1099 	 * Initialize its label.
1100 	 */
1101 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1102 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1103 	vp = abd_to_buf(vp_abd);
1104 
1105 	/*
1106 	 * Generate a label describing the pool and our top-level vdev.
1107 	 * We mark it as being from txg 0 to indicate that it's not
1108 	 * really part of an active pool just yet.  The labels will
1109 	 * be written again with a meaningful txg by spa_sync().
1110 	 */
1111 	if (reason == VDEV_LABEL_SPARE ||
1112 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1113 		/*
1114 		 * For inactive hot spares, we generate a special label that
1115 		 * identifies as a mutually shared hot spare.  We write the
1116 		 * label if we are adding a hot spare, or if we are removing an
1117 		 * active hot spare (in which case we want to revert the
1118 		 * labels).
1119 		 */
1120 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1121 
1122 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1123 		    spa_version(spa)) == 0);
1124 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1125 		    POOL_STATE_SPARE) == 0);
1126 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1127 		    vd->vdev_guid) == 0);
1128 	} else if (reason == VDEV_LABEL_L2CACHE ||
1129 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1130 		/*
1131 		 * For level 2 ARC devices, add a special label.
1132 		 */
1133 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1134 
1135 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1136 		    spa_version(spa)) == 0);
1137 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1138 		    POOL_STATE_L2CACHE) == 0);
1139 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1140 		    vd->vdev_guid) == 0);
1141 	} else {
1142 		uint64_t txg = 0ULL;
1143 
1144 		if (reason == VDEV_LABEL_SPLIT)
1145 			txg = spa->spa_uberblock.ub_txg;
1146 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1147 
1148 		/*
1149 		 * Add our creation time.  This allows us to detect multiple
1150 		 * vdev uses as described above, and automatically expires if we
1151 		 * fail.
1152 		 */
1153 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1154 		    crtxg) == 0);
1155 	}
1156 
1157 	buf = vp->vp_nvlist;
1158 	buflen = sizeof (vp->vp_nvlist);
1159 
1160 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1161 	if (error != 0) {
1162 		nvlist_free(label);
1163 		abd_free(vp_abd);
1164 		/* EFAULT means nvlist_pack ran out of room */
1165 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1166 	}
1167 
1168 	/*
1169 	 * Initialize uberblock template.
1170 	 */
1171 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1172 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1173 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1174 	ub = abd_to_buf(ub_abd);
1175 	ub->ub_txg = 0;
1176 
1177 	/* Initialize the 2nd padding area. */
1178 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1179 	abd_zero(bootenv, VDEV_PAD_SIZE);
1180 
1181 	/*
1182 	 * Write everything in parallel.
1183 	 */
1184 retry:
1185 	zio = zio_root(spa, NULL, NULL, flags);
1186 
1187 	for (int l = 0; l < VDEV_LABELS; l++) {
1188 
1189 		vdev_label_write(zio, vd, l, vp_abd,
1190 		    offsetof(vdev_label_t, vl_vdev_phys),
1191 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1192 
1193 		/*
1194 		 * Skip the 1st padding area.
1195 		 * Zero out the 2nd padding area where it might have
1196 		 * left over data from previous filesystem format.
1197 		 */
1198 		vdev_label_write(zio, vd, l, bootenv,
1199 		    offsetof(vdev_label_t, vl_be),
1200 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1201 
1202 		vdev_label_write(zio, vd, l, ub_abd,
1203 		    offsetof(vdev_label_t, vl_uberblock),
1204 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1205 	}
1206 
1207 	error = zio_wait(zio);
1208 
1209 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1210 		flags |= ZIO_FLAG_TRYHARD;
1211 		goto retry;
1212 	}
1213 
1214 	nvlist_free(label);
1215 	abd_free(bootenv);
1216 	abd_free(ub_abd);
1217 	abd_free(vp_abd);
1218 
1219 	/*
1220 	 * If this vdev hasn't been previously identified as a spare, then we
1221 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1222 	 * exists as a spare elsewhere in the system.  Do the same for
1223 	 * level 2 ARC devices.
1224 	 */
1225 	if (error == 0 && !vd->vdev_isspare &&
1226 	    (reason == VDEV_LABEL_SPARE ||
1227 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1228 		spa_spare_add(vd);
1229 
1230 	if (error == 0 && !vd->vdev_isl2cache &&
1231 	    (reason == VDEV_LABEL_L2CACHE ||
1232 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1233 		spa_l2cache_add(vd);
1234 
1235 	return (error);
1236 }
1237 
1238 /*
1239  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1240  * callback to finish, store our abd in the callback pointer. Otherwise, we
1241  * just free our abd and return.
1242  */
1243 static void
1244 vdev_label_read_bootenv_done(zio_t *zio)
1245 {
1246 	zio_t *rio = zio->io_private;
1247 	abd_t **cbp = rio->io_private;
1248 
1249 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1250 
1251 	if (zio->io_error == 0) {
1252 		mutex_enter(&rio->io_lock);
1253 		if (*cbp == NULL) {
1254 			/* Will free this buffer in vdev_label_read_bootenv. */
1255 			*cbp = zio->io_abd;
1256 		} else {
1257 			abd_free(zio->io_abd);
1258 		}
1259 		mutex_exit(&rio->io_lock);
1260 	} else {
1261 		abd_free(zio->io_abd);
1262 	}
1263 }
1264 
1265 static void
1266 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1267 {
1268 	for (int c = 0; c < vd->vdev_children; c++)
1269 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1270 
1271 	/*
1272 	 * We just use the first label that has a correct checksum; the
1273 	 * bootloader should have rewritten them all to be the same on boot,
1274 	 * and any changes we made since boot have been the same across all
1275 	 * labels.
1276 	 */
1277 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1278 		for (int l = 0; l < VDEV_LABELS; l++) {
1279 			vdev_label_read(zio, vd, l,
1280 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1281 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1282 			    vdev_label_read_bootenv_done, zio, flags);
1283 		}
1284 	}
1285 }
1286 
1287 int
1288 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1289 {
1290 	nvlist_t *config;
1291 	spa_t *spa = rvd->vdev_spa;
1292 	abd_t *abd = NULL;
1293 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1294 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1295 
1296 	ASSERT(bootenv);
1297 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1298 
1299 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1300 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1301 	int err = zio_wait(zio);
1302 
1303 	if (abd != NULL) {
1304 		char *buf;
1305 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1306 
1307 		vbe->vbe_version = ntohll(vbe->vbe_version);
1308 		switch (vbe->vbe_version) {
1309 		case VB_RAW:
1310 			/*
1311 			 * if we have textual data in vbe_bootenv, create nvlist
1312 			 * with key "envmap".
1313 			 */
1314 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1315 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1316 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1317 			    vbe->vbe_bootenv);
1318 			break;
1319 
1320 		case VB_NVLIST:
1321 			err = nvlist_unpack(vbe->vbe_bootenv,
1322 			    sizeof (vbe->vbe_bootenv), &config, 0);
1323 			if (err == 0) {
1324 				fnvlist_merge(bootenv, config);
1325 				nvlist_free(config);
1326 				break;
1327 			}
1328 			zfs_fallthrough;
1329 		default:
1330 			/* Check for FreeBSD zfs bootonce command string */
1331 			buf = abd_to_buf(abd);
1332 			if (*buf == '\0') {
1333 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1334 				    VB_NVLIST);
1335 				break;
1336 			}
1337 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1338 		}
1339 
1340 		/*
1341 		 * abd was allocated in vdev_label_read_bootenv_impl()
1342 		 */
1343 		abd_free(abd);
1344 		/*
1345 		 * If we managed to read any successfully,
1346 		 * return success.
1347 		 */
1348 		return (0);
1349 	}
1350 	return (err);
1351 }
1352 
1353 int
1354 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1355 {
1356 	zio_t *zio;
1357 	spa_t *spa = vd->vdev_spa;
1358 	vdev_boot_envblock_t *bootenv;
1359 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1360 	int error;
1361 	size_t nvsize;
1362 	char *nvbuf;
1363 	const char *tmp;
1364 
1365 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1366 	if (error != 0)
1367 		return (SET_ERROR(error));
1368 
1369 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1370 		return (SET_ERROR(E2BIG));
1371 	}
1372 
1373 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1374 
1375 	error = ENXIO;
1376 	for (int c = 0; c < vd->vdev_children; c++) {
1377 		int child_err;
1378 
1379 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1380 		/*
1381 		 * As long as any of the disks managed to write all of their
1382 		 * labels successfully, return success.
1383 		 */
1384 		if (child_err == 0)
1385 			error = child_err;
1386 	}
1387 
1388 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1389 	    !vdev_writeable(vd)) {
1390 		return (error);
1391 	}
1392 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1393 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1394 	abd_zero(abd, VDEV_PAD_SIZE);
1395 
1396 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1397 	nvbuf = bootenv->vbe_bootenv;
1398 	nvsize = sizeof (bootenv->vbe_bootenv);
1399 
1400 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1401 	switch (bootenv->vbe_version) {
1402 	case VB_RAW:
1403 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1404 			(void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1405 		}
1406 		error = 0;
1407 		break;
1408 
1409 	case VB_NVLIST:
1410 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1411 		    KM_SLEEP);
1412 		break;
1413 
1414 	default:
1415 		error = EINVAL;
1416 		break;
1417 	}
1418 
1419 	if (error == 0) {
1420 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1421 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1422 	} else {
1423 		abd_free(abd);
1424 		return (SET_ERROR(error));
1425 	}
1426 
1427 retry:
1428 	zio = zio_root(spa, NULL, NULL, flags);
1429 	for (int l = 0; l < VDEV_LABELS; l++) {
1430 		vdev_label_write(zio, vd, l, abd,
1431 		    offsetof(vdev_label_t, vl_be),
1432 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1433 	}
1434 
1435 	error = zio_wait(zio);
1436 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1437 		flags |= ZIO_FLAG_TRYHARD;
1438 		goto retry;
1439 	}
1440 
1441 	abd_free(abd);
1442 	return (error);
1443 }
1444 
1445 /*
1446  * ==========================================================================
1447  * uberblock load/sync
1448  * ==========================================================================
1449  */
1450 
1451 /*
1452  * Consider the following situation: txg is safely synced to disk.  We've
1453  * written the first uberblock for txg + 1, and then we lose power.  When we
1454  * come back up, we fail to see the uberblock for txg + 1 because, say,
1455  * it was on a mirrored device and the replica to which we wrote txg + 1
1456  * is now offline.  If we then make some changes and sync txg + 1, and then
1457  * the missing replica comes back, then for a few seconds we'll have two
1458  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1459  * among uberblocks with equal txg, choose the one with the latest timestamp.
1460  */
1461 static int
1462 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1463 {
1464 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1465 
1466 	if (likely(cmp))
1467 		return (cmp);
1468 
1469 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1470 	if (likely(cmp))
1471 		return (cmp);
1472 
1473 	/*
1474 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1475 	 * ZFS, e.g. OpenZFS >= 0.7.
1476 	 *
1477 	 * If one ub has MMP and the other does not, they were written by
1478 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1479 	 * a 0 value.
1480 	 *
1481 	 * Since timestamp and txg are the same if we get this far, either is
1482 	 * acceptable for importing the pool.
1483 	 */
1484 	unsigned int seq1 = 0;
1485 	unsigned int seq2 = 0;
1486 
1487 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1488 		seq1 = MMP_SEQ(ub1);
1489 
1490 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1491 		seq2 = MMP_SEQ(ub2);
1492 
1493 	return (TREE_CMP(seq1, seq2));
1494 }
1495 
1496 struct ubl_cbdata {
1497 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1498 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1499 };
1500 
1501 static void
1502 vdev_uberblock_load_done(zio_t *zio)
1503 {
1504 	vdev_t *vd = zio->io_vd;
1505 	spa_t *spa = zio->io_spa;
1506 	zio_t *rio = zio->io_private;
1507 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1508 	struct ubl_cbdata *cbp = rio->io_private;
1509 
1510 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1511 
1512 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1513 		mutex_enter(&rio->io_lock);
1514 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1515 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1516 			/*
1517 			 * Keep track of the vdev in which this uberblock
1518 			 * was found. We will use this information later
1519 			 * to obtain the config nvlist associated with
1520 			 * this uberblock.
1521 			 */
1522 			*cbp->ubl_ubbest = *ub;
1523 			cbp->ubl_vd = vd;
1524 		}
1525 		mutex_exit(&rio->io_lock);
1526 	}
1527 
1528 	abd_free(zio->io_abd);
1529 }
1530 
1531 static void
1532 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1533     struct ubl_cbdata *cbp)
1534 {
1535 	for (int c = 0; c < vd->vdev_children; c++)
1536 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1537 
1538 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1539 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1540 		for (int l = 0; l < VDEV_LABELS; l++) {
1541 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1542 				vdev_label_read(zio, vd, l,
1543 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1544 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1545 				    VDEV_UBERBLOCK_SIZE(vd),
1546 				    vdev_uberblock_load_done, zio, flags);
1547 			}
1548 		}
1549 	}
1550 }
1551 
1552 /*
1553  * Reads the 'best' uberblock from disk along with its associated
1554  * configuration. First, we read the uberblock array of each label of each
1555  * vdev, keeping track of the uberblock with the highest txg in each array.
1556  * Then, we read the configuration from the same vdev as the best uberblock.
1557  */
1558 void
1559 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1560 {
1561 	zio_t *zio;
1562 	spa_t *spa = rvd->vdev_spa;
1563 	struct ubl_cbdata cb;
1564 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1565 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1566 
1567 	ASSERT(ub);
1568 	ASSERT(config);
1569 
1570 	memset(ub, 0, sizeof (uberblock_t));
1571 	*config = NULL;
1572 
1573 	cb.ubl_ubbest = ub;
1574 	cb.ubl_vd = NULL;
1575 
1576 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1577 	zio = zio_root(spa, NULL, &cb, flags);
1578 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1579 	(void) zio_wait(zio);
1580 
1581 	/*
1582 	 * It's possible that the best uberblock was discovered on a label
1583 	 * that has a configuration which was written in a future txg.
1584 	 * Search all labels on this vdev to find the configuration that
1585 	 * matches the txg for our uberblock.
1586 	 */
1587 	if (cb.ubl_vd != NULL) {
1588 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1589 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1590 
1591 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1592 		if (*config == NULL && spa->spa_extreme_rewind) {
1593 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1594 			    "Trying again without txg restrictions.");
1595 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1596 		}
1597 		if (*config == NULL) {
1598 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1599 		}
1600 	}
1601 	spa_config_exit(spa, SCL_ALL, FTAG);
1602 }
1603 
1604 /*
1605  * For use when a leaf vdev is expanded.
1606  * The location of labels 2 and 3 changed, and at the new location the
1607  * uberblock rings are either empty or contain garbage.  The sync will write
1608  * new configs there because the vdev is dirty, but expansion also needs the
1609  * uberblock rings copied.  Read them from label 0 which did not move.
1610  *
1611  * Since the point is to populate labels {2,3} with valid uberblocks,
1612  * we zero uberblocks we fail to read or which are not valid.
1613  */
1614 
1615 static void
1616 vdev_copy_uberblocks(vdev_t *vd)
1617 {
1618 	abd_t *ub_abd;
1619 	zio_t *write_zio;
1620 	int locks = (SCL_L2ARC | SCL_ZIO);
1621 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1622 	    ZIO_FLAG_SPECULATIVE;
1623 
1624 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1625 	    SCL_STATE);
1626 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1627 
1628 	/*
1629 	 * No uberblocks are stored on distributed spares, they may be
1630 	 * safely skipped when expanding a leaf vdev.
1631 	 */
1632 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1633 		return;
1634 
1635 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1636 
1637 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1638 
1639 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1640 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1641 		const int src_label = 0;
1642 		zio_t *zio;
1643 
1644 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1645 		vdev_label_read(zio, vd, src_label, ub_abd,
1646 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1647 		    NULL, NULL, flags);
1648 
1649 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1650 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1651 
1652 		for (int l = 2; l < VDEV_LABELS; l++)
1653 			vdev_label_write(write_zio, vd, l, ub_abd,
1654 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1655 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1656 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1657 	}
1658 	(void) zio_wait(write_zio);
1659 
1660 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1661 
1662 	abd_free(ub_abd);
1663 }
1664 
1665 /*
1666  * On success, increment root zio's count of good writes.
1667  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1668  */
1669 static void
1670 vdev_uberblock_sync_done(zio_t *zio)
1671 {
1672 	uint64_t *good_writes = zio->io_private;
1673 
1674 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1675 		atomic_inc_64(good_writes);
1676 }
1677 
1678 /*
1679  * Write the uberblock to all labels of all leaves of the specified vdev.
1680  */
1681 static void
1682 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1683     uberblock_t *ub, vdev_t *vd, int flags)
1684 {
1685 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1686 		vdev_uberblock_sync(zio, good_writes,
1687 		    ub, vd->vdev_child[c], flags);
1688 	}
1689 
1690 	if (!vd->vdev_ops->vdev_op_leaf)
1691 		return;
1692 
1693 	if (!vdev_writeable(vd))
1694 		return;
1695 
1696 	/*
1697 	 * There's no need to write uberblocks to a distributed spare, they
1698 	 * are already stored on all the leaves of the parent dRAID.  For
1699 	 * this same reason vdev_uberblock_load_impl() skips distributed
1700 	 * spares when reading uberblocks.
1701 	 */
1702 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1703 		return;
1704 
1705 	/* If the vdev was expanded, need to copy uberblock rings. */
1706 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1707 	    vd->vdev_copy_uberblocks == B_TRUE) {
1708 		vdev_copy_uberblocks(vd);
1709 		vd->vdev_copy_uberblocks = B_FALSE;
1710 	}
1711 
1712 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1713 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1714 
1715 	/* Copy the uberblock_t into the ABD */
1716 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1717 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1718 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1719 
1720 	for (int l = 0; l < VDEV_LABELS; l++)
1721 		vdev_label_write(zio, vd, l, ub_abd,
1722 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1723 		    vdev_uberblock_sync_done, good_writes,
1724 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1725 
1726 	abd_free(ub_abd);
1727 }
1728 
1729 /* Sync the uberblocks to all vdevs in svd[] */
1730 static int
1731 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1732 {
1733 	spa_t *spa = svd[0]->vdev_spa;
1734 	zio_t *zio;
1735 	uint64_t good_writes = 0;
1736 
1737 	zio = zio_root(spa, NULL, NULL, flags);
1738 
1739 	for (int v = 0; v < svdcount; v++)
1740 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1741 
1742 	(void) zio_wait(zio);
1743 
1744 	/*
1745 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1746 	 * are no longer needed (because the new uberblocks and the even
1747 	 * labels are safely on disk), so it is safe to overwrite them.
1748 	 */
1749 	zio = zio_root(spa, NULL, NULL, flags);
1750 
1751 	for (int v = 0; v < svdcount; v++) {
1752 		if (vdev_writeable(svd[v])) {
1753 			zio_flush(zio, svd[v]);
1754 		}
1755 	}
1756 
1757 	(void) zio_wait(zio);
1758 
1759 	return (good_writes >= 1 ? 0 : EIO);
1760 }
1761 
1762 /*
1763  * On success, increment the count of good writes for our top-level vdev.
1764  */
1765 static void
1766 vdev_label_sync_done(zio_t *zio)
1767 {
1768 	uint64_t *good_writes = zio->io_private;
1769 
1770 	if (zio->io_error == 0)
1771 		atomic_inc_64(good_writes);
1772 }
1773 
1774 /*
1775  * If there weren't enough good writes, indicate failure to the parent.
1776  */
1777 static void
1778 vdev_label_sync_top_done(zio_t *zio)
1779 {
1780 	uint64_t *good_writes = zio->io_private;
1781 
1782 	if (*good_writes == 0)
1783 		zio->io_error = SET_ERROR(EIO);
1784 
1785 	kmem_free(good_writes, sizeof (uint64_t));
1786 }
1787 
1788 /*
1789  * We ignore errors for log and cache devices, simply free the private data.
1790  */
1791 static void
1792 vdev_label_sync_ignore_done(zio_t *zio)
1793 {
1794 	kmem_free(zio->io_private, sizeof (uint64_t));
1795 }
1796 
1797 /*
1798  * Write all even or odd labels to all leaves of the specified vdev.
1799  */
1800 static void
1801 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1802     vdev_t *vd, int l, uint64_t txg, int flags)
1803 {
1804 	nvlist_t *label;
1805 	vdev_phys_t *vp;
1806 	abd_t *vp_abd;
1807 	char *buf;
1808 	size_t buflen;
1809 
1810 	for (int c = 0; c < vd->vdev_children; c++) {
1811 		vdev_label_sync(zio, good_writes,
1812 		    vd->vdev_child[c], l, txg, flags);
1813 	}
1814 
1815 	if (!vd->vdev_ops->vdev_op_leaf)
1816 		return;
1817 
1818 	if (!vdev_writeable(vd))
1819 		return;
1820 
1821 	/*
1822 	 * The top-level config never needs to be written to a distributed
1823 	 * spare.  When read vdev_dspare_label_read_config() will generate
1824 	 * the config for the vdev_label_read_config().
1825 	 */
1826 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1827 		return;
1828 
1829 	/*
1830 	 * Generate a label describing the top-level config to which we belong.
1831 	 */
1832 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1833 
1834 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1835 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1836 	vp = abd_to_buf(vp_abd);
1837 
1838 	buf = vp->vp_nvlist;
1839 	buflen = sizeof (vp->vp_nvlist);
1840 
1841 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1842 		for (; l < VDEV_LABELS; l += 2) {
1843 			vdev_label_write(zio, vd, l, vp_abd,
1844 			    offsetof(vdev_label_t, vl_vdev_phys),
1845 			    sizeof (vdev_phys_t),
1846 			    vdev_label_sync_done, good_writes,
1847 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1848 		}
1849 	}
1850 
1851 	abd_free(vp_abd);
1852 	nvlist_free(label);
1853 }
1854 
1855 static int
1856 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1857 {
1858 	list_t *dl = &spa->spa_config_dirty_list;
1859 	vdev_t *vd;
1860 	zio_t *zio;
1861 	int error;
1862 
1863 	/*
1864 	 * Write the new labels to disk.
1865 	 */
1866 	zio = zio_root(spa, NULL, NULL, flags);
1867 
1868 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1869 		uint64_t *good_writes;
1870 
1871 		ASSERT(!vd->vdev_ishole);
1872 
1873 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1874 		zio_t *vio = zio_null(zio, spa, NULL,
1875 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1876 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1877 		    good_writes, flags);
1878 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1879 		zio_nowait(vio);
1880 	}
1881 
1882 	error = zio_wait(zio);
1883 
1884 	/*
1885 	 * Flush the new labels to disk.
1886 	 */
1887 	zio = zio_root(spa, NULL, NULL, flags);
1888 
1889 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1890 		zio_flush(zio, vd);
1891 
1892 	(void) zio_wait(zio);
1893 
1894 	return (error);
1895 }
1896 
1897 /*
1898  * Sync the uberblock and any changes to the vdev configuration.
1899  *
1900  * The order of operations is carefully crafted to ensure that
1901  * if the system panics or loses power at any time, the state on disk
1902  * is still transactionally consistent.  The in-line comments below
1903  * describe the failure semantics at each stage.
1904  *
1905  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1906  * at any time, you can just call it again, and it will resume its work.
1907  */
1908 int
1909 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1910 {
1911 	spa_t *spa = svd[0]->vdev_spa;
1912 	uberblock_t *ub = &spa->spa_uberblock;
1913 	int error = 0;
1914 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1915 
1916 	ASSERT(svdcount != 0);
1917 retry:
1918 	/*
1919 	 * Normally, we don't want to try too hard to write every label and
1920 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1921 	 * sync process to block while we retry.  But if we can't write a
1922 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1923 	 * bailing out and declaring the pool faulted.
1924 	 */
1925 	if (error != 0) {
1926 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1927 			return (error);
1928 		flags |= ZIO_FLAG_TRYHARD;
1929 	}
1930 
1931 	ASSERT(ub->ub_txg <= txg);
1932 
1933 	/*
1934 	 * If this isn't a resync due to I/O errors,
1935 	 * and nothing changed in this transaction group,
1936 	 * and the vdev configuration hasn't changed,
1937 	 * then there's nothing to do.
1938 	 */
1939 	if (ub->ub_txg < txg) {
1940 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1941 		    txg, spa->spa_mmp.mmp_delay);
1942 
1943 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1944 			return (0);
1945 	}
1946 
1947 	if (txg > spa_freeze_txg(spa))
1948 		return (0);
1949 
1950 	ASSERT(txg <= spa->spa_final_txg);
1951 
1952 	/*
1953 	 * Flush the write cache of every disk that's been written to
1954 	 * in this transaction group.  This ensures that all blocks
1955 	 * written in this txg will be committed to stable storage
1956 	 * before any uberblock that references them.
1957 	 */
1958 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1959 
1960 	for (vdev_t *vd =
1961 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1962 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1963 		zio_flush(zio, vd);
1964 
1965 	(void) zio_wait(zio);
1966 
1967 	/*
1968 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1969 	 * system dies in the middle of this process, that's OK: all of the
1970 	 * even labels that made it to disk will be newer than any uberblock,
1971 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1972 	 * which have not yet been touched, will still be valid.  We flush
1973 	 * the new labels to disk to ensure that all even-label updates
1974 	 * are committed to stable storage before the uberblock update.
1975 	 */
1976 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1977 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1978 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1979 			    "for pool '%s' when syncing out the even labels "
1980 			    "of dirty vdevs", error, spa_name(spa));
1981 		}
1982 		goto retry;
1983 	}
1984 
1985 	/*
1986 	 * Sync the uberblocks to all vdevs in svd[].
1987 	 * If the system dies in the middle of this step, there are two cases
1988 	 * to consider, and the on-disk state is consistent either way:
1989 	 *
1990 	 * (1)	If none of the new uberblocks made it to disk, then the
1991 	 *	previous uberblock will be the newest, and the odd labels
1992 	 *	(which had not yet been touched) will be valid with respect
1993 	 *	to that uberblock.
1994 	 *
1995 	 * (2)	If one or more new uberblocks made it to disk, then they
1996 	 *	will be the newest, and the even labels (which had all
1997 	 *	been successfully committed) will be valid with respect
1998 	 *	to the new uberblocks.
1999 	 */
2000 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
2001 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2002 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2003 			    "%d for pool '%s'", error, spa_name(spa));
2004 		}
2005 		goto retry;
2006 	}
2007 
2008 	if (spa_multihost(spa))
2009 		mmp_update_uberblock(spa, ub);
2010 
2011 	/*
2012 	 * Sync out odd labels for every dirty vdev.  If the system dies
2013 	 * in the middle of this process, the even labels and the new
2014 	 * uberblocks will suffice to open the pool.  The next time
2015 	 * the pool is opened, the first thing we'll do -- before any
2016 	 * user data is modified -- is mark every vdev dirty so that
2017 	 * all labels will be brought up to date.  We flush the new labels
2018 	 * to disk to ensure that all odd-label updates are committed to
2019 	 * stable storage before the next transaction group begins.
2020 	 */
2021 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2022 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2023 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2024 			    "for pool '%s' when syncing out the odd labels of "
2025 			    "dirty vdevs", error, spa_name(spa));
2026 		}
2027 		goto retry;
2028 	}
2029 
2030 	return (0);
2031 }
2032