xref: /freebsd/sys/contrib/openzfs/module/zfs/vdev_label.c (revision 5956d97f4b3204318ceb6aa9c77bd0bc6ea87a41)
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 	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_SPARE | VDEV_CONFIG_L2CACHE)) &&
490 	    vd == vd->vdev_top) {
491 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
492 		    vd->vdev_ms_array);
493 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
494 		    vd->vdev_ms_shift);
495 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
496 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
497 		    vd->vdev_asize);
498 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
499 		if (vd->vdev_noalloc) {
500 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
501 			    vd->vdev_noalloc);
502 		}
503 		if (vd->vdev_removing) {
504 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
505 			    vd->vdev_removing);
506 		}
507 
508 		/* zpool command expects alloc class data */
509 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
510 			const char *bias = NULL;
511 
512 			switch (vd->vdev_alloc_bias) {
513 			case VDEV_BIAS_LOG:
514 				bias = VDEV_ALLOC_BIAS_LOG;
515 				break;
516 			case VDEV_BIAS_SPECIAL:
517 				bias = VDEV_ALLOC_BIAS_SPECIAL;
518 				break;
519 			case VDEV_BIAS_DEDUP:
520 				bias = VDEV_ALLOC_BIAS_DEDUP;
521 				break;
522 			default:
523 				ASSERT3U(vd->vdev_alloc_bias, ==,
524 				    VDEV_BIAS_NONE);
525 			}
526 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
527 			    bias);
528 		}
529 	}
530 
531 	if (vd->vdev_dtl_sm != NULL) {
532 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
533 		    space_map_object(vd->vdev_dtl_sm));
534 	}
535 
536 	if (vic->vic_mapping_object != 0) {
537 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
538 		    vic->vic_mapping_object);
539 	}
540 
541 	if (vic->vic_births_object != 0) {
542 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
543 		    vic->vic_births_object);
544 	}
545 
546 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
547 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
548 		    vic->vic_prev_indirect_vdev);
549 	}
550 
551 	if (vd->vdev_crtxg)
552 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
553 
554 	if (vd->vdev_expansion_time)
555 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
556 		    vd->vdev_expansion_time);
557 
558 	if (flags & VDEV_CONFIG_MOS) {
559 		if (vd->vdev_leaf_zap != 0) {
560 			ASSERT(vd->vdev_ops->vdev_op_leaf);
561 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
562 			    vd->vdev_leaf_zap);
563 		}
564 
565 		if (vd->vdev_top_zap != 0) {
566 			ASSERT(vd == vd->vdev_top);
567 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
568 			    vd->vdev_top_zap);
569 		}
570 
571 		if (vd->vdev_resilver_deferred) {
572 			ASSERT(vd->vdev_ops->vdev_op_leaf);
573 			ASSERT(spa->spa_resilver_deferred);
574 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
575 		}
576 	}
577 
578 	if (getstats) {
579 		vdev_config_generate_stats(vd, nv);
580 
581 		root_vdev_actions_getprogress(vd, nv);
582 		top_vdev_actions_getprogress(vd, nv);
583 
584 		/*
585 		 * Note: this can be called from open context
586 		 * (spa_get_stats()), so we need the rwlock to prevent
587 		 * the mapping from being changed by condensing.
588 		 */
589 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
590 		if (vd->vdev_indirect_mapping != NULL) {
591 			ASSERT(vd->vdev_indirect_births != NULL);
592 			vdev_indirect_mapping_t *vim =
593 			    vd->vdev_indirect_mapping;
594 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
595 			    vdev_indirect_mapping_size(vim));
596 		}
597 		rw_exit(&vd->vdev_indirect_rwlock);
598 		if (vd->vdev_mg != NULL &&
599 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
600 			/*
601 			 * Compute approximately how much memory would be used
602 			 * for the indirect mapping if this device were to
603 			 * be removed.
604 			 *
605 			 * Note: If the frag metric is invalid, then not
606 			 * enough metaslabs have been converted to have
607 			 * histograms.
608 			 */
609 			uint64_t seg_count = 0;
610 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
611 
612 			/*
613 			 * There are the same number of allocated segments
614 			 * as free segments, so we will have at least one
615 			 * entry per free segment.  However, small free
616 			 * segments (smaller than vdev_removal_max_span)
617 			 * will be combined with adjacent allocated segments
618 			 * as a single mapping.
619 			 */
620 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
621 				if (i + 1 < highbit64(vdev_removal_max_span)
622 				    - 1) {
623 					to_alloc +=
624 					    vd->vdev_mg->mg_histogram[i] <<
625 					    (i + 1);
626 				} else {
627 					seg_count +=
628 					    vd->vdev_mg->mg_histogram[i];
629 				}
630 			}
631 
632 			/*
633 			 * The maximum length of a mapping is
634 			 * zfs_remove_max_segment, so we need at least one entry
635 			 * per zfs_remove_max_segment of allocated data.
636 			 */
637 			seg_count += to_alloc / spa_remove_max_segment(spa);
638 
639 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
640 			    seg_count *
641 			    sizeof (vdev_indirect_mapping_entry_phys_t));
642 		}
643 	}
644 
645 	if (!vd->vdev_ops->vdev_op_leaf) {
646 		nvlist_t **child;
647 		int c, idx;
648 
649 		ASSERT(!vd->vdev_ishole);
650 
651 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
652 		    KM_SLEEP);
653 
654 		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
655 			vdev_t *cvd = vd->vdev_child[c];
656 
657 			/*
658 			 * If we're generating an nvlist of removing
659 			 * vdevs then skip over any device which is
660 			 * not being removed.
661 			 */
662 			if ((flags & VDEV_CONFIG_REMOVING) &&
663 			    !cvd->vdev_removing)
664 				continue;
665 
666 			child[idx++] = vdev_config_generate(spa, cvd,
667 			    getstats, flags);
668 		}
669 
670 		if (idx) {
671 			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
672 			    (const nvlist_t * const *)child, idx);
673 		}
674 
675 		for (c = 0; c < idx; c++)
676 			nvlist_free(child[c]);
677 
678 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
679 
680 	} else {
681 		const char *aux = NULL;
682 
683 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
684 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
685 		if (vd->vdev_resilver_txg != 0)
686 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
687 			    vd->vdev_resilver_txg);
688 		if (vd->vdev_rebuild_txg != 0)
689 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
690 			    vd->vdev_rebuild_txg);
691 		if (vd->vdev_faulted)
692 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
693 		if (vd->vdev_degraded)
694 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
695 		if (vd->vdev_removed)
696 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
697 		if (vd->vdev_unspare)
698 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
699 		if (vd->vdev_ishole)
700 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
701 
702 		/* Set the reason why we're FAULTED/DEGRADED. */
703 		switch (vd->vdev_stat.vs_aux) {
704 		case VDEV_AUX_ERR_EXCEEDED:
705 			aux = "err_exceeded";
706 			break;
707 
708 		case VDEV_AUX_EXTERNAL:
709 			aux = "external";
710 			break;
711 		}
712 
713 		if (aux != NULL && !vd->vdev_tmpoffline) {
714 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
715 		} else {
716 			/*
717 			 * We're healthy - clear any previous AUX_STATE values.
718 			 */
719 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
720 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
721 		}
722 
723 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
724 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
725 			    vd->vdev_orig_guid);
726 		}
727 	}
728 
729 	return (nv);
730 }
731 
732 /*
733  * Generate a view of the top-level vdevs.  If we currently have holes
734  * in the namespace, then generate an array which contains a list of holey
735  * vdevs.  Additionally, add the number of top-level children that currently
736  * exist.
737  */
738 void
739 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
740 {
741 	vdev_t *rvd = spa->spa_root_vdev;
742 	uint64_t *array;
743 	uint_t c, idx;
744 
745 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
746 
747 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
748 		vdev_t *tvd = rvd->vdev_child[c];
749 
750 		if (tvd->vdev_ishole) {
751 			array[idx++] = c;
752 		}
753 	}
754 
755 	if (idx) {
756 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
757 		    array, idx) == 0);
758 	}
759 
760 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
761 	    rvd->vdev_children) == 0);
762 
763 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
764 }
765 
766 /*
767  * Returns the configuration from the label of the given vdev. For vdevs
768  * which don't have a txg value stored on their label (i.e. spares/cache)
769  * or have not been completely initialized (txg = 0) just return
770  * the configuration from the first valid label we find. Otherwise,
771  * find the most up-to-date label that does not exceed the specified
772  * 'txg' value.
773  */
774 nvlist_t *
775 vdev_label_read_config(vdev_t *vd, uint64_t txg)
776 {
777 	spa_t *spa = vd->vdev_spa;
778 	nvlist_t *config = NULL;
779 	vdev_phys_t *vp[VDEV_LABELS];
780 	abd_t *vp_abd[VDEV_LABELS];
781 	zio_t *zio[VDEV_LABELS];
782 	uint64_t best_txg = 0;
783 	uint64_t label_txg = 0;
784 	int error = 0;
785 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
786 	    ZIO_FLAG_SPECULATIVE;
787 
788 	ASSERT(vd->vdev_validate_thread == curthread ||
789 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
790 
791 	if (!vdev_readable(vd))
792 		return (NULL);
793 
794 	/*
795 	 * The label for a dRAID distributed spare is not stored on disk.
796 	 * Instead it is generated when needed which allows us to bypass
797 	 * the pipeline when reading the config from the label.
798 	 */
799 	if (vd->vdev_ops == &vdev_draid_spare_ops)
800 		return (vdev_draid_read_config_spare(vd));
801 
802 	for (int l = 0; l < VDEV_LABELS; l++) {
803 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
804 		vp[l] = abd_to_buf(vp_abd[l]);
805 	}
806 
807 retry:
808 	for (int l = 0; l < VDEV_LABELS; l++) {
809 		zio[l] = zio_root(spa, NULL, NULL, flags);
810 
811 		vdev_label_read(zio[l], vd, l, vp_abd[l],
812 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
813 		    NULL, NULL, flags);
814 	}
815 	for (int l = 0; l < VDEV_LABELS; l++) {
816 		nvlist_t *label = NULL;
817 
818 		if (zio_wait(zio[l]) == 0 &&
819 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
820 		    &label, 0) == 0) {
821 			/*
822 			 * Auxiliary vdevs won't have txg values in their
823 			 * labels and newly added vdevs may not have been
824 			 * completely initialized so just return the
825 			 * configuration from the first valid label we
826 			 * encounter.
827 			 */
828 			error = nvlist_lookup_uint64(label,
829 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
830 			if ((error || label_txg == 0) && !config) {
831 				config = label;
832 				for (l++; l < VDEV_LABELS; l++)
833 					zio_wait(zio[l]);
834 				break;
835 			} else if (label_txg <= txg && label_txg > best_txg) {
836 				best_txg = label_txg;
837 				nvlist_free(config);
838 				config = fnvlist_dup(label);
839 			}
840 		}
841 
842 		if (label != NULL) {
843 			nvlist_free(label);
844 			label = NULL;
845 		}
846 	}
847 
848 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
849 		flags |= ZIO_FLAG_TRYHARD;
850 		goto retry;
851 	}
852 
853 	/*
854 	 * We found a valid label but it didn't pass txg restrictions.
855 	 */
856 	if (config == NULL && label_txg != 0) {
857 		vdev_dbgmsg(vd, "label discarded as txg is too large "
858 		    "(%llu > %llu)", (u_longlong_t)label_txg,
859 		    (u_longlong_t)txg);
860 	}
861 
862 	for (int l = 0; l < VDEV_LABELS; l++) {
863 		abd_free(vp_abd[l]);
864 	}
865 
866 	return (config);
867 }
868 
869 /*
870  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
871  * in with the device guid if this spare is active elsewhere on the system.
872  */
873 static boolean_t
874 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
875     uint64_t *spare_guid, uint64_t *l2cache_guid)
876 {
877 	spa_t *spa = vd->vdev_spa;
878 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
879 	uint64_t vdtxg = 0;
880 	nvlist_t *label;
881 
882 	if (spare_guid)
883 		*spare_guid = 0ULL;
884 	if (l2cache_guid)
885 		*l2cache_guid = 0ULL;
886 
887 	/*
888 	 * Read the label, if any, and perform some basic sanity checks.
889 	 */
890 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
891 		return (B_FALSE);
892 
893 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
894 	    &vdtxg);
895 
896 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
897 	    &state) != 0 ||
898 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
899 	    &device_guid) != 0) {
900 		nvlist_free(label);
901 		return (B_FALSE);
902 	}
903 
904 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
905 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
906 	    &pool_guid) != 0 ||
907 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
908 	    &txg) != 0)) {
909 		nvlist_free(label);
910 		return (B_FALSE);
911 	}
912 
913 	nvlist_free(label);
914 
915 	/*
916 	 * Check to see if this device indeed belongs to the pool it claims to
917 	 * be a part of.  The only way this is allowed is if the device is a hot
918 	 * spare (which we check for later on).
919 	 */
920 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
921 	    !spa_guid_exists(pool_guid, device_guid) &&
922 	    !spa_spare_exists(device_guid, NULL, NULL) &&
923 	    !spa_l2cache_exists(device_guid, NULL))
924 		return (B_FALSE);
925 
926 	/*
927 	 * If the transaction group is zero, then this an initialized (but
928 	 * unused) label.  This is only an error if the create transaction
929 	 * on-disk is the same as the one we're using now, in which case the
930 	 * user has attempted to add the same vdev multiple times in the same
931 	 * transaction.
932 	 */
933 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
934 	    txg == 0 && vdtxg == crtxg)
935 		return (B_TRUE);
936 
937 	/*
938 	 * Check to see if this is a spare device.  We do an explicit check for
939 	 * spa_has_spare() here because it may be on our pending list of spares
940 	 * to add.
941 	 */
942 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
943 	    spa_has_spare(spa, device_guid)) {
944 		if (spare_guid)
945 			*spare_guid = device_guid;
946 
947 		switch (reason) {
948 		case VDEV_LABEL_CREATE:
949 			return (B_TRUE);
950 
951 		case VDEV_LABEL_REPLACE:
952 			return (!spa_has_spare(spa, device_guid) ||
953 			    spare_pool != 0ULL);
954 
955 		case VDEV_LABEL_SPARE:
956 			return (spa_has_spare(spa, device_guid));
957 		default:
958 			break;
959 		}
960 	}
961 
962 	/*
963 	 * Check to see if this is an l2cache device.
964 	 */
965 	if (spa_l2cache_exists(device_guid, NULL) ||
966 	    spa_has_l2cache(spa, device_guid)) {
967 		if (l2cache_guid)
968 			*l2cache_guid = device_guid;
969 
970 		switch (reason) {
971 		case VDEV_LABEL_CREATE:
972 			return (B_TRUE);
973 
974 		case VDEV_LABEL_REPLACE:
975 			return (!spa_has_l2cache(spa, device_guid));
976 
977 		case VDEV_LABEL_L2CACHE:
978 			return (spa_has_l2cache(spa, device_guid));
979 		default:
980 			break;
981 		}
982 	}
983 
984 	/*
985 	 * We can't rely on a pool's state if it's been imported
986 	 * read-only.  Instead we look to see if the pools is marked
987 	 * read-only in the namespace and set the state to active.
988 	 */
989 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
990 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
991 	    spa_mode(spa) == SPA_MODE_READ)
992 		state = POOL_STATE_ACTIVE;
993 
994 	/*
995 	 * If the device is marked ACTIVE, then this device is in use by another
996 	 * pool on the system.
997 	 */
998 	return (state == POOL_STATE_ACTIVE);
999 }
1000 
1001 /*
1002  * Initialize a vdev label.  We check to make sure each leaf device is not in
1003  * use, and writable.  We put down an initial label which we will later
1004  * overwrite with a complete label.  Note that it's important to do this
1005  * sequentially, not in parallel, so that we catch cases of multiple use of the
1006  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1007  * itself.
1008  */
1009 int
1010 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1011 {
1012 	spa_t *spa = vd->vdev_spa;
1013 	nvlist_t *label;
1014 	vdev_phys_t *vp;
1015 	abd_t *vp_abd;
1016 	abd_t *bootenv;
1017 	uberblock_t *ub;
1018 	abd_t *ub_abd;
1019 	zio_t *zio;
1020 	char *buf;
1021 	size_t buflen;
1022 	int error;
1023 	uint64_t spare_guid = 0, l2cache_guid = 0;
1024 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1025 
1026 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1027 
1028 	for (int c = 0; c < vd->vdev_children; c++)
1029 		if ((error = vdev_label_init(vd->vdev_child[c],
1030 		    crtxg, reason)) != 0)
1031 			return (error);
1032 
1033 	/* Track the creation time for this vdev */
1034 	vd->vdev_crtxg = crtxg;
1035 
1036 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1037 		return (0);
1038 
1039 	/*
1040 	 * Dead vdevs cannot be initialized.
1041 	 */
1042 	if (vdev_is_dead(vd))
1043 		return (SET_ERROR(EIO));
1044 
1045 	/*
1046 	 * Determine if the vdev is in use.
1047 	 */
1048 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1049 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1050 		return (SET_ERROR(EBUSY));
1051 
1052 	/*
1053 	 * If this is a request to add or replace a spare or l2cache device
1054 	 * that is in use elsewhere on the system, then we must update the
1055 	 * guid (which was initialized to a random value) to reflect the
1056 	 * actual GUID (which is shared between multiple pools).
1057 	 */
1058 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1059 	    spare_guid != 0ULL) {
1060 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1061 
1062 		vd->vdev_guid += guid_delta;
1063 
1064 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1065 			pvd->vdev_guid_sum += guid_delta;
1066 
1067 		/*
1068 		 * If this is a replacement, then we want to fallthrough to the
1069 		 * rest of the code.  If we're adding a spare, then it's already
1070 		 * labeled appropriately and we can just return.
1071 		 */
1072 		if (reason == VDEV_LABEL_SPARE)
1073 			return (0);
1074 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1075 		    reason == VDEV_LABEL_SPLIT);
1076 	}
1077 
1078 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1079 	    l2cache_guid != 0ULL) {
1080 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1081 
1082 		vd->vdev_guid += guid_delta;
1083 
1084 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1085 			pvd->vdev_guid_sum += guid_delta;
1086 
1087 		/*
1088 		 * If this is a replacement, then we want to fallthrough to the
1089 		 * rest of the code.  If we're adding an l2cache, then it's
1090 		 * already labeled appropriately and we can just return.
1091 		 */
1092 		if (reason == VDEV_LABEL_L2CACHE)
1093 			return (0);
1094 		ASSERT(reason == VDEV_LABEL_REPLACE);
1095 	}
1096 
1097 	/*
1098 	 * Initialize its label.
1099 	 */
1100 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1101 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1102 	vp = abd_to_buf(vp_abd);
1103 
1104 	/*
1105 	 * Generate a label describing the pool and our top-level vdev.
1106 	 * We mark it as being from txg 0 to indicate that it's not
1107 	 * really part of an active pool just yet.  The labels will
1108 	 * be written again with a meaningful txg by spa_sync().
1109 	 */
1110 	if (reason == VDEV_LABEL_SPARE ||
1111 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1112 		/*
1113 		 * For inactive hot spares, we generate a special label that
1114 		 * identifies as a mutually shared hot spare.  We write the
1115 		 * label if we are adding a hot spare, or if we are removing an
1116 		 * active hot spare (in which case we want to revert the
1117 		 * labels).
1118 		 */
1119 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1120 
1121 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1122 		    spa_version(spa)) == 0);
1123 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1124 		    POOL_STATE_SPARE) == 0);
1125 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1126 		    vd->vdev_guid) == 0);
1127 	} else if (reason == VDEV_LABEL_L2CACHE ||
1128 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1129 		/*
1130 		 * For level 2 ARC devices, add a special label.
1131 		 */
1132 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1133 
1134 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1135 		    spa_version(spa)) == 0);
1136 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1137 		    POOL_STATE_L2CACHE) == 0);
1138 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1139 		    vd->vdev_guid) == 0);
1140 	} else {
1141 		uint64_t txg = 0ULL;
1142 
1143 		if (reason == VDEV_LABEL_SPLIT)
1144 			txg = spa->spa_uberblock.ub_txg;
1145 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1146 
1147 		/*
1148 		 * Add our creation time.  This allows us to detect multiple
1149 		 * vdev uses as described above, and automatically expires if we
1150 		 * fail.
1151 		 */
1152 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1153 		    crtxg) == 0);
1154 	}
1155 
1156 	buf = vp->vp_nvlist;
1157 	buflen = sizeof (vp->vp_nvlist);
1158 
1159 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1160 	if (error != 0) {
1161 		nvlist_free(label);
1162 		abd_free(vp_abd);
1163 		/* EFAULT means nvlist_pack ran out of room */
1164 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1165 	}
1166 
1167 	/*
1168 	 * Initialize uberblock template.
1169 	 */
1170 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1171 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1172 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1173 	ub = abd_to_buf(ub_abd);
1174 	ub->ub_txg = 0;
1175 
1176 	/* Initialize the 2nd padding area. */
1177 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1178 	abd_zero(bootenv, VDEV_PAD_SIZE);
1179 
1180 	/*
1181 	 * Write everything in parallel.
1182 	 */
1183 retry:
1184 	zio = zio_root(spa, NULL, NULL, flags);
1185 
1186 	for (int l = 0; l < VDEV_LABELS; l++) {
1187 
1188 		vdev_label_write(zio, vd, l, vp_abd,
1189 		    offsetof(vdev_label_t, vl_vdev_phys),
1190 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1191 
1192 		/*
1193 		 * Skip the 1st padding area.
1194 		 * Zero out the 2nd padding area where it might have
1195 		 * left over data from previous filesystem format.
1196 		 */
1197 		vdev_label_write(zio, vd, l, bootenv,
1198 		    offsetof(vdev_label_t, vl_be),
1199 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1200 
1201 		vdev_label_write(zio, vd, l, ub_abd,
1202 		    offsetof(vdev_label_t, vl_uberblock),
1203 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1204 	}
1205 
1206 	error = zio_wait(zio);
1207 
1208 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1209 		flags |= ZIO_FLAG_TRYHARD;
1210 		goto retry;
1211 	}
1212 
1213 	nvlist_free(label);
1214 	abd_free(bootenv);
1215 	abd_free(ub_abd);
1216 	abd_free(vp_abd);
1217 
1218 	/*
1219 	 * If this vdev hasn't been previously identified as a spare, then we
1220 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1221 	 * exists as a spare elsewhere in the system.  Do the same for
1222 	 * level 2 ARC devices.
1223 	 */
1224 	if (error == 0 && !vd->vdev_isspare &&
1225 	    (reason == VDEV_LABEL_SPARE ||
1226 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1227 		spa_spare_add(vd);
1228 
1229 	if (error == 0 && !vd->vdev_isl2cache &&
1230 	    (reason == VDEV_LABEL_L2CACHE ||
1231 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1232 		spa_l2cache_add(vd);
1233 
1234 	return (error);
1235 }
1236 
1237 /*
1238  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1239  * callback to finish, store our abd in the callback pointer. Otherwise, we
1240  * just free our abd and return.
1241  */
1242 static void
1243 vdev_label_read_bootenv_done(zio_t *zio)
1244 {
1245 	zio_t *rio = zio->io_private;
1246 	abd_t **cbp = rio->io_private;
1247 
1248 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1249 
1250 	if (zio->io_error == 0) {
1251 		mutex_enter(&rio->io_lock);
1252 		if (*cbp == NULL) {
1253 			/* Will free this buffer in vdev_label_read_bootenv. */
1254 			*cbp = zio->io_abd;
1255 		} else {
1256 			abd_free(zio->io_abd);
1257 		}
1258 		mutex_exit(&rio->io_lock);
1259 	} else {
1260 		abd_free(zio->io_abd);
1261 	}
1262 }
1263 
1264 static void
1265 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1266 {
1267 	for (int c = 0; c < vd->vdev_children; c++)
1268 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1269 
1270 	/*
1271 	 * We just use the first label that has a correct checksum; the
1272 	 * bootloader should have rewritten them all to be the same on boot,
1273 	 * and any changes we made since boot have been the same across all
1274 	 * labels.
1275 	 */
1276 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1277 		for (int l = 0; l < VDEV_LABELS; l++) {
1278 			vdev_label_read(zio, vd, l,
1279 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1280 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1281 			    vdev_label_read_bootenv_done, zio, flags);
1282 		}
1283 	}
1284 }
1285 
1286 int
1287 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1288 {
1289 	nvlist_t *config;
1290 	spa_t *spa = rvd->vdev_spa;
1291 	abd_t *abd = NULL;
1292 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1293 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1294 
1295 	ASSERT(bootenv);
1296 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1297 
1298 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1299 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1300 	int err = zio_wait(zio);
1301 
1302 	if (abd != NULL) {
1303 		char *buf;
1304 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1305 
1306 		vbe->vbe_version = ntohll(vbe->vbe_version);
1307 		switch (vbe->vbe_version) {
1308 		case VB_RAW:
1309 			/*
1310 			 * if we have textual data in vbe_bootenv, create nvlist
1311 			 * with key "envmap".
1312 			 */
1313 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1314 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1315 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1316 			    vbe->vbe_bootenv);
1317 			break;
1318 
1319 		case VB_NVLIST:
1320 			err = nvlist_unpack(vbe->vbe_bootenv,
1321 			    sizeof (vbe->vbe_bootenv), &config, 0);
1322 			if (err == 0) {
1323 				fnvlist_merge(bootenv, config);
1324 				nvlist_free(config);
1325 				break;
1326 			}
1327 			zfs_fallthrough;
1328 		default:
1329 			/* Check for FreeBSD zfs bootonce command string */
1330 			buf = abd_to_buf(abd);
1331 			if (*buf == '\0') {
1332 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1333 				    VB_NVLIST);
1334 				break;
1335 			}
1336 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1337 		}
1338 
1339 		/*
1340 		 * abd was allocated in vdev_label_read_bootenv_impl()
1341 		 */
1342 		abd_free(abd);
1343 		/*
1344 		 * If we managed to read any successfully,
1345 		 * return success.
1346 		 */
1347 		return (0);
1348 	}
1349 	return (err);
1350 }
1351 
1352 int
1353 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1354 {
1355 	zio_t *zio;
1356 	spa_t *spa = vd->vdev_spa;
1357 	vdev_boot_envblock_t *bootenv;
1358 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1359 	int error;
1360 	size_t nvsize;
1361 	char *nvbuf;
1362 
1363 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1364 	if (error != 0)
1365 		return (SET_ERROR(error));
1366 
1367 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1368 		return (SET_ERROR(E2BIG));
1369 	}
1370 
1371 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1372 
1373 	error = ENXIO;
1374 	for (int c = 0; c < vd->vdev_children; c++) {
1375 		int child_err;
1376 
1377 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1378 		/*
1379 		 * As long as any of the disks managed to write all of their
1380 		 * labels successfully, return success.
1381 		 */
1382 		if (child_err == 0)
1383 			error = child_err;
1384 	}
1385 
1386 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1387 	    !vdev_writeable(vd)) {
1388 		return (error);
1389 	}
1390 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1391 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1392 	abd_zero(abd, VDEV_PAD_SIZE);
1393 
1394 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1395 	nvbuf = bootenv->vbe_bootenv;
1396 	nvsize = sizeof (bootenv->vbe_bootenv);
1397 
1398 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1399 	switch (bootenv->vbe_version) {
1400 	case VB_RAW:
1401 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) {
1402 			(void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize);
1403 		}
1404 		error = 0;
1405 		break;
1406 
1407 	case VB_NVLIST:
1408 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1409 		    KM_SLEEP);
1410 		break;
1411 
1412 	default:
1413 		error = EINVAL;
1414 		break;
1415 	}
1416 
1417 	if (error == 0) {
1418 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1419 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1420 	} else {
1421 		abd_free(abd);
1422 		return (SET_ERROR(error));
1423 	}
1424 
1425 retry:
1426 	zio = zio_root(spa, NULL, NULL, flags);
1427 	for (int l = 0; l < VDEV_LABELS; l++) {
1428 		vdev_label_write(zio, vd, l, abd,
1429 		    offsetof(vdev_label_t, vl_be),
1430 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1431 	}
1432 
1433 	error = zio_wait(zio);
1434 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1435 		flags |= ZIO_FLAG_TRYHARD;
1436 		goto retry;
1437 	}
1438 
1439 	abd_free(abd);
1440 	return (error);
1441 }
1442 
1443 /*
1444  * ==========================================================================
1445  * uberblock load/sync
1446  * ==========================================================================
1447  */
1448 
1449 /*
1450  * Consider the following situation: txg is safely synced to disk.  We've
1451  * written the first uberblock for txg + 1, and then we lose power.  When we
1452  * come back up, we fail to see the uberblock for txg + 1 because, say,
1453  * it was on a mirrored device and the replica to which we wrote txg + 1
1454  * is now offline.  If we then make some changes and sync txg + 1, and then
1455  * the missing replica comes back, then for a few seconds we'll have two
1456  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1457  * among uberblocks with equal txg, choose the one with the latest timestamp.
1458  */
1459 static int
1460 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1461 {
1462 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1463 
1464 	if (likely(cmp))
1465 		return (cmp);
1466 
1467 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1468 	if (likely(cmp))
1469 		return (cmp);
1470 
1471 	/*
1472 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1473 	 * ZFS, e.g. OpenZFS >= 0.7.
1474 	 *
1475 	 * If one ub has MMP and the other does not, they were written by
1476 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1477 	 * a 0 value.
1478 	 *
1479 	 * Since timestamp and txg are the same if we get this far, either is
1480 	 * acceptable for importing the pool.
1481 	 */
1482 	unsigned int seq1 = 0;
1483 	unsigned int seq2 = 0;
1484 
1485 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1486 		seq1 = MMP_SEQ(ub1);
1487 
1488 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1489 		seq2 = MMP_SEQ(ub2);
1490 
1491 	return (TREE_CMP(seq1, seq2));
1492 }
1493 
1494 struct ubl_cbdata {
1495 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1496 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1497 };
1498 
1499 static void
1500 vdev_uberblock_load_done(zio_t *zio)
1501 {
1502 	vdev_t *vd = zio->io_vd;
1503 	spa_t *spa = zio->io_spa;
1504 	zio_t *rio = zio->io_private;
1505 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1506 	struct ubl_cbdata *cbp = rio->io_private;
1507 
1508 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1509 
1510 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1511 		mutex_enter(&rio->io_lock);
1512 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1513 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1514 			/*
1515 			 * Keep track of the vdev in which this uberblock
1516 			 * was found. We will use this information later
1517 			 * to obtain the config nvlist associated with
1518 			 * this uberblock.
1519 			 */
1520 			*cbp->ubl_ubbest = *ub;
1521 			cbp->ubl_vd = vd;
1522 		}
1523 		mutex_exit(&rio->io_lock);
1524 	}
1525 
1526 	abd_free(zio->io_abd);
1527 }
1528 
1529 static void
1530 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1531     struct ubl_cbdata *cbp)
1532 {
1533 	for (int c = 0; c < vd->vdev_children; c++)
1534 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1535 
1536 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1537 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1538 		for (int l = 0; l < VDEV_LABELS; l++) {
1539 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1540 				vdev_label_read(zio, vd, l,
1541 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1542 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1543 				    VDEV_UBERBLOCK_SIZE(vd),
1544 				    vdev_uberblock_load_done, zio, flags);
1545 			}
1546 		}
1547 	}
1548 }
1549 
1550 /*
1551  * Reads the 'best' uberblock from disk along with its associated
1552  * configuration. First, we read the uberblock array of each label of each
1553  * vdev, keeping track of the uberblock with the highest txg in each array.
1554  * Then, we read the configuration from the same vdev as the best uberblock.
1555  */
1556 void
1557 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1558 {
1559 	zio_t *zio;
1560 	spa_t *spa = rvd->vdev_spa;
1561 	struct ubl_cbdata cb;
1562 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1563 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1564 
1565 	ASSERT(ub);
1566 	ASSERT(config);
1567 
1568 	memset(ub, 0, sizeof (uberblock_t));
1569 	*config = NULL;
1570 
1571 	cb.ubl_ubbest = ub;
1572 	cb.ubl_vd = NULL;
1573 
1574 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1575 	zio = zio_root(spa, NULL, &cb, flags);
1576 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1577 	(void) zio_wait(zio);
1578 
1579 	/*
1580 	 * It's possible that the best uberblock was discovered on a label
1581 	 * that has a configuration which was written in a future txg.
1582 	 * Search all labels on this vdev to find the configuration that
1583 	 * matches the txg for our uberblock.
1584 	 */
1585 	if (cb.ubl_vd != NULL) {
1586 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1587 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1588 
1589 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1590 		if (*config == NULL && spa->spa_extreme_rewind) {
1591 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1592 			    "Trying again without txg restrictions.");
1593 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1594 		}
1595 		if (*config == NULL) {
1596 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1597 		}
1598 	}
1599 	spa_config_exit(spa, SCL_ALL, FTAG);
1600 }
1601 
1602 /*
1603  * For use when a leaf vdev is expanded.
1604  * The location of labels 2 and 3 changed, and at the new location the
1605  * uberblock rings are either empty or contain garbage.  The sync will write
1606  * new configs there because the vdev is dirty, but expansion also needs the
1607  * uberblock rings copied.  Read them from label 0 which did not move.
1608  *
1609  * Since the point is to populate labels {2,3} with valid uberblocks,
1610  * we zero uberblocks we fail to read or which are not valid.
1611  */
1612 
1613 static void
1614 vdev_copy_uberblocks(vdev_t *vd)
1615 {
1616 	abd_t *ub_abd;
1617 	zio_t *write_zio;
1618 	int locks = (SCL_L2ARC | SCL_ZIO);
1619 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1620 	    ZIO_FLAG_SPECULATIVE;
1621 
1622 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1623 	    SCL_STATE);
1624 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1625 
1626 	/*
1627 	 * No uberblocks are stored on distributed spares, they may be
1628 	 * safely skipped when expanding a leaf vdev.
1629 	 */
1630 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1631 		return;
1632 
1633 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1634 
1635 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1636 
1637 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1638 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1639 		const int src_label = 0;
1640 		zio_t *zio;
1641 
1642 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1643 		vdev_label_read(zio, vd, src_label, ub_abd,
1644 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1645 		    NULL, NULL, flags);
1646 
1647 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1648 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1649 
1650 		for (int l = 2; l < VDEV_LABELS; l++)
1651 			vdev_label_write(write_zio, vd, l, ub_abd,
1652 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1653 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1654 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1655 	}
1656 	(void) zio_wait(write_zio);
1657 
1658 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1659 
1660 	abd_free(ub_abd);
1661 }
1662 
1663 /*
1664  * On success, increment root zio's count of good writes.
1665  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1666  */
1667 static void
1668 vdev_uberblock_sync_done(zio_t *zio)
1669 {
1670 	uint64_t *good_writes = zio->io_private;
1671 
1672 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1673 		atomic_inc_64(good_writes);
1674 }
1675 
1676 /*
1677  * Write the uberblock to all labels of all leaves of the specified vdev.
1678  */
1679 static void
1680 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1681     uberblock_t *ub, vdev_t *vd, int flags)
1682 {
1683 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1684 		vdev_uberblock_sync(zio, good_writes,
1685 		    ub, vd->vdev_child[c], flags);
1686 	}
1687 
1688 	if (!vd->vdev_ops->vdev_op_leaf)
1689 		return;
1690 
1691 	if (!vdev_writeable(vd))
1692 		return;
1693 
1694 	/*
1695 	 * There's no need to write uberblocks to a distributed spare, they
1696 	 * are already stored on all the leaves of the parent dRAID.  For
1697 	 * this same reason vdev_uberblock_load_impl() skips distributed
1698 	 * spares when reading uberblocks.
1699 	 */
1700 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1701 		return;
1702 
1703 	/* If the vdev was expanded, need to copy uberblock rings. */
1704 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1705 	    vd->vdev_copy_uberblocks == B_TRUE) {
1706 		vdev_copy_uberblocks(vd);
1707 		vd->vdev_copy_uberblocks = B_FALSE;
1708 	}
1709 
1710 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1711 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1712 
1713 	/* Copy the uberblock_t into the ABD */
1714 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1715 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1716 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1717 
1718 	for (int l = 0; l < VDEV_LABELS; l++)
1719 		vdev_label_write(zio, vd, l, ub_abd,
1720 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1721 		    vdev_uberblock_sync_done, good_writes,
1722 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1723 
1724 	abd_free(ub_abd);
1725 }
1726 
1727 /* Sync the uberblocks to all vdevs in svd[] */
1728 static int
1729 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1730 {
1731 	spa_t *spa = svd[0]->vdev_spa;
1732 	zio_t *zio;
1733 	uint64_t good_writes = 0;
1734 
1735 	zio = zio_root(spa, NULL, NULL, flags);
1736 
1737 	for (int v = 0; v < svdcount; v++)
1738 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1739 
1740 	(void) zio_wait(zio);
1741 
1742 	/*
1743 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1744 	 * are no longer needed (because the new uberblocks and the even
1745 	 * labels are safely on disk), so it is safe to overwrite them.
1746 	 */
1747 	zio = zio_root(spa, NULL, NULL, flags);
1748 
1749 	for (int v = 0; v < svdcount; v++) {
1750 		if (vdev_writeable(svd[v])) {
1751 			zio_flush(zio, svd[v]);
1752 		}
1753 	}
1754 
1755 	(void) zio_wait(zio);
1756 
1757 	return (good_writes >= 1 ? 0 : EIO);
1758 }
1759 
1760 /*
1761  * On success, increment the count of good writes for our top-level vdev.
1762  */
1763 static void
1764 vdev_label_sync_done(zio_t *zio)
1765 {
1766 	uint64_t *good_writes = zio->io_private;
1767 
1768 	if (zio->io_error == 0)
1769 		atomic_inc_64(good_writes);
1770 }
1771 
1772 /*
1773  * If there weren't enough good writes, indicate failure to the parent.
1774  */
1775 static void
1776 vdev_label_sync_top_done(zio_t *zio)
1777 {
1778 	uint64_t *good_writes = zio->io_private;
1779 
1780 	if (*good_writes == 0)
1781 		zio->io_error = SET_ERROR(EIO);
1782 
1783 	kmem_free(good_writes, sizeof (uint64_t));
1784 }
1785 
1786 /*
1787  * We ignore errors for log and cache devices, simply free the private data.
1788  */
1789 static void
1790 vdev_label_sync_ignore_done(zio_t *zio)
1791 {
1792 	kmem_free(zio->io_private, sizeof (uint64_t));
1793 }
1794 
1795 /*
1796  * Write all even or odd labels to all leaves of the specified vdev.
1797  */
1798 static void
1799 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1800     vdev_t *vd, int l, uint64_t txg, int flags)
1801 {
1802 	nvlist_t *label;
1803 	vdev_phys_t *vp;
1804 	abd_t *vp_abd;
1805 	char *buf;
1806 	size_t buflen;
1807 
1808 	for (int c = 0; c < vd->vdev_children; c++) {
1809 		vdev_label_sync(zio, good_writes,
1810 		    vd->vdev_child[c], l, txg, flags);
1811 	}
1812 
1813 	if (!vd->vdev_ops->vdev_op_leaf)
1814 		return;
1815 
1816 	if (!vdev_writeable(vd))
1817 		return;
1818 
1819 	/*
1820 	 * The top-level config never needs to be written to a distributed
1821 	 * spare.  When read vdev_dspare_label_read_config() will generate
1822 	 * the config for the vdev_label_read_config().
1823 	 */
1824 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1825 		return;
1826 
1827 	/*
1828 	 * Generate a label describing the top-level config to which we belong.
1829 	 */
1830 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1831 
1832 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1833 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1834 	vp = abd_to_buf(vp_abd);
1835 
1836 	buf = vp->vp_nvlist;
1837 	buflen = sizeof (vp->vp_nvlist);
1838 
1839 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1840 		for (; l < VDEV_LABELS; l += 2) {
1841 			vdev_label_write(zio, vd, l, vp_abd,
1842 			    offsetof(vdev_label_t, vl_vdev_phys),
1843 			    sizeof (vdev_phys_t),
1844 			    vdev_label_sync_done, good_writes,
1845 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1846 		}
1847 	}
1848 
1849 	abd_free(vp_abd);
1850 	nvlist_free(label);
1851 }
1852 
1853 static int
1854 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1855 {
1856 	list_t *dl = &spa->spa_config_dirty_list;
1857 	vdev_t *vd;
1858 	zio_t *zio;
1859 	int error;
1860 
1861 	/*
1862 	 * Write the new labels to disk.
1863 	 */
1864 	zio = zio_root(spa, NULL, NULL, flags);
1865 
1866 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1867 		uint64_t *good_writes;
1868 
1869 		ASSERT(!vd->vdev_ishole);
1870 
1871 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1872 		zio_t *vio = zio_null(zio, spa, NULL,
1873 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1874 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1875 		    good_writes, flags);
1876 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1877 		zio_nowait(vio);
1878 	}
1879 
1880 	error = zio_wait(zio);
1881 
1882 	/*
1883 	 * Flush the new labels to disk.
1884 	 */
1885 	zio = zio_root(spa, NULL, NULL, flags);
1886 
1887 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1888 		zio_flush(zio, vd);
1889 
1890 	(void) zio_wait(zio);
1891 
1892 	return (error);
1893 }
1894 
1895 /*
1896  * Sync the uberblock and any changes to the vdev configuration.
1897  *
1898  * The order of operations is carefully crafted to ensure that
1899  * if the system panics or loses power at any time, the state on disk
1900  * is still transactionally consistent.  The in-line comments below
1901  * describe the failure semantics at each stage.
1902  *
1903  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1904  * at any time, you can just call it again, and it will resume its work.
1905  */
1906 int
1907 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1908 {
1909 	spa_t *spa = svd[0]->vdev_spa;
1910 	uberblock_t *ub = &spa->spa_uberblock;
1911 	int error = 0;
1912 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1913 
1914 	ASSERT(svdcount != 0);
1915 retry:
1916 	/*
1917 	 * Normally, we don't want to try too hard to write every label and
1918 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1919 	 * sync process to block while we retry.  But if we can't write a
1920 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1921 	 * bailing out and declaring the pool faulted.
1922 	 */
1923 	if (error != 0) {
1924 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1925 			return (error);
1926 		flags |= ZIO_FLAG_TRYHARD;
1927 	}
1928 
1929 	ASSERT(ub->ub_txg <= txg);
1930 
1931 	/*
1932 	 * If this isn't a resync due to I/O errors,
1933 	 * and nothing changed in this transaction group,
1934 	 * and the vdev configuration hasn't changed,
1935 	 * then there's nothing to do.
1936 	 */
1937 	if (ub->ub_txg < txg) {
1938 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1939 		    txg, spa->spa_mmp.mmp_delay);
1940 
1941 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1942 			return (0);
1943 	}
1944 
1945 	if (txg > spa_freeze_txg(spa))
1946 		return (0);
1947 
1948 	ASSERT(txg <= spa->spa_final_txg);
1949 
1950 	/*
1951 	 * Flush the write cache of every disk that's been written to
1952 	 * in this transaction group.  This ensures that all blocks
1953 	 * written in this txg will be committed to stable storage
1954 	 * before any uberblock that references them.
1955 	 */
1956 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1957 
1958 	for (vdev_t *vd =
1959 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1960 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1961 		zio_flush(zio, vd);
1962 
1963 	(void) zio_wait(zio);
1964 
1965 	/*
1966 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1967 	 * system dies in the middle of this process, that's OK: all of the
1968 	 * even labels that made it to disk will be newer than any uberblock,
1969 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1970 	 * which have not yet been touched, will still be valid.  We flush
1971 	 * the new labels to disk to ensure that all even-label updates
1972 	 * are committed to stable storage before the uberblock update.
1973 	 */
1974 	if ((error = vdev_label_sync_list(spa, 0, 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 even labels "
1978 			    "of dirty vdevs", error, spa_name(spa));
1979 		}
1980 		goto retry;
1981 	}
1982 
1983 	/*
1984 	 * Sync the uberblocks to all vdevs in svd[].
1985 	 * If the system dies in the middle of this step, there are two cases
1986 	 * to consider, and the on-disk state is consistent either way:
1987 	 *
1988 	 * (1)	If none of the new uberblocks made it to disk, then the
1989 	 *	previous uberblock will be the newest, and the odd labels
1990 	 *	(which had not yet been touched) will be valid with respect
1991 	 *	to that uberblock.
1992 	 *
1993 	 * (2)	If one or more new uberblocks made it to disk, then they
1994 	 *	will be the newest, and the even labels (which had all
1995 	 *	been successfully committed) will be valid with respect
1996 	 *	to the new uberblocks.
1997 	 */
1998 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1999 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2000 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2001 			    "%d for pool '%s'", error, spa_name(spa));
2002 		}
2003 		goto retry;
2004 	}
2005 
2006 	if (spa_multihost(spa))
2007 		mmp_update_uberblock(spa, ub);
2008 
2009 	/*
2010 	 * Sync out odd labels for every dirty vdev.  If the system dies
2011 	 * in the middle of this process, the even labels and the new
2012 	 * uberblocks will suffice to open the pool.  The next time
2013 	 * the pool is opened, the first thing we'll do -- before any
2014 	 * user data is modified -- is mark every vdev dirty so that
2015 	 * all labels will be brought up to date.  We flush the new labels
2016 	 * to disk to ensure that all odd-label updates are committed to
2017 	 * stable storage before the next transaction group begins.
2018 	 */
2019 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2020 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2021 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2022 			    "for pool '%s' when syncing out the odd labels of "
2023 			    "dirty vdevs", error, spa_name(spa));
2024 		}
2025 		goto retry;
2026 	}
2027 
2028 	return (0);
2029 }
2030