xref: /freebsd/sys/contrib/openzfs/module/zfs/vdev_label.c (revision 0d66206fff44f864ea8a4b220c3a53b4caa959a0)
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_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 
504 		/*
505 		 * Slog devices are removed synchronously so don't
506 		 * persist the vdev_removing flag to the label.
507 		 */
508 		if (vd->vdev_removing && !vd->vdev_islog) {
509 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
510 			    vd->vdev_removing);
511 		}
512 
513 		/* zpool command expects alloc class data */
514 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
515 			const char *bias = NULL;
516 
517 			switch (vd->vdev_alloc_bias) {
518 			case VDEV_BIAS_LOG:
519 				bias = VDEV_ALLOC_BIAS_LOG;
520 				break;
521 			case VDEV_BIAS_SPECIAL:
522 				bias = VDEV_ALLOC_BIAS_SPECIAL;
523 				break;
524 			case VDEV_BIAS_DEDUP:
525 				bias = VDEV_ALLOC_BIAS_DEDUP;
526 				break;
527 			default:
528 				ASSERT3U(vd->vdev_alloc_bias, ==,
529 				    VDEV_BIAS_NONE);
530 			}
531 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
532 			    bias);
533 		}
534 	}
535 
536 	if (vd->vdev_dtl_sm != NULL) {
537 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
538 		    space_map_object(vd->vdev_dtl_sm));
539 	}
540 
541 	if (vic->vic_mapping_object != 0) {
542 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
543 		    vic->vic_mapping_object);
544 	}
545 
546 	if (vic->vic_births_object != 0) {
547 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
548 		    vic->vic_births_object);
549 	}
550 
551 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
552 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
553 		    vic->vic_prev_indirect_vdev);
554 	}
555 
556 	if (vd->vdev_crtxg)
557 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
558 
559 	if (vd->vdev_expansion_time)
560 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
561 		    vd->vdev_expansion_time);
562 
563 	if (flags & VDEV_CONFIG_MOS) {
564 		if (vd->vdev_leaf_zap != 0) {
565 			ASSERT(vd->vdev_ops->vdev_op_leaf);
566 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
567 			    vd->vdev_leaf_zap);
568 		}
569 
570 		if (vd->vdev_top_zap != 0) {
571 			ASSERT(vd == vd->vdev_top);
572 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
573 			    vd->vdev_top_zap);
574 		}
575 
576 		if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
577 		    spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
578 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
579 			    vd->vdev_root_zap);
580 		}
581 
582 		if (vd->vdev_resilver_deferred) {
583 			ASSERT(vd->vdev_ops->vdev_op_leaf);
584 			ASSERT(spa->spa_resilver_deferred);
585 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
586 		}
587 	}
588 
589 	if (getstats) {
590 		vdev_config_generate_stats(vd, nv);
591 
592 		root_vdev_actions_getprogress(vd, nv);
593 		top_vdev_actions_getprogress(vd, nv);
594 
595 		/*
596 		 * Note: this can be called from open context
597 		 * (spa_get_stats()), so we need the rwlock to prevent
598 		 * the mapping from being changed by condensing.
599 		 */
600 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
601 		if (vd->vdev_indirect_mapping != NULL) {
602 			ASSERT(vd->vdev_indirect_births != NULL);
603 			vdev_indirect_mapping_t *vim =
604 			    vd->vdev_indirect_mapping;
605 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
606 			    vdev_indirect_mapping_size(vim));
607 		}
608 		rw_exit(&vd->vdev_indirect_rwlock);
609 		if (vd->vdev_mg != NULL &&
610 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
611 			/*
612 			 * Compute approximately how much memory would be used
613 			 * for the indirect mapping if this device were to
614 			 * be removed.
615 			 *
616 			 * Note: If the frag metric is invalid, then not
617 			 * enough metaslabs have been converted to have
618 			 * histograms.
619 			 */
620 			uint64_t seg_count = 0;
621 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
622 
623 			/*
624 			 * There are the same number of allocated segments
625 			 * as free segments, so we will have at least one
626 			 * entry per free segment.  However, small free
627 			 * segments (smaller than vdev_removal_max_span)
628 			 * will be combined with adjacent allocated segments
629 			 * as a single mapping.
630 			 */
631 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
632 				if (i + 1 < highbit64(vdev_removal_max_span)
633 				    - 1) {
634 					to_alloc +=
635 					    vd->vdev_mg->mg_histogram[i] <<
636 					    (i + 1);
637 				} else {
638 					seg_count +=
639 					    vd->vdev_mg->mg_histogram[i];
640 				}
641 			}
642 
643 			/*
644 			 * The maximum length of a mapping is
645 			 * zfs_remove_max_segment, so we need at least one entry
646 			 * per zfs_remove_max_segment of allocated data.
647 			 */
648 			seg_count += to_alloc / spa_remove_max_segment(spa);
649 
650 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
651 			    seg_count *
652 			    sizeof (vdev_indirect_mapping_entry_phys_t));
653 		}
654 	}
655 
656 	if (!vd->vdev_ops->vdev_op_leaf) {
657 		nvlist_t **child;
658 		uint64_t c;
659 
660 		ASSERT(!vd->vdev_ishole);
661 
662 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
663 		    KM_SLEEP);
664 
665 		for (c = 0; c < vd->vdev_children; c++) {
666 			child[c] = vdev_config_generate(spa, vd->vdev_child[c],
667 			    getstats, flags);
668 		}
669 
670 		fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
671 		    (const nvlist_t * const *)child, vd->vdev_children);
672 
673 		for (c = 0; c < vd->vdev_children; c++)
674 			nvlist_free(child[c]);
675 
676 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
677 
678 	} else {
679 		const char *aux = NULL;
680 
681 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
682 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
683 		if (vd->vdev_resilver_txg != 0)
684 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
685 			    vd->vdev_resilver_txg);
686 		if (vd->vdev_rebuild_txg != 0)
687 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
688 			    vd->vdev_rebuild_txg);
689 		if (vd->vdev_faulted)
690 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
691 		if (vd->vdev_degraded)
692 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
693 		if (vd->vdev_removed)
694 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
695 		if (vd->vdev_unspare)
696 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
697 		if (vd->vdev_ishole)
698 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
699 
700 		/* Set the reason why we're FAULTED/DEGRADED. */
701 		switch (vd->vdev_stat.vs_aux) {
702 		case VDEV_AUX_ERR_EXCEEDED:
703 			aux = "err_exceeded";
704 			break;
705 
706 		case VDEV_AUX_EXTERNAL:
707 			aux = "external";
708 			break;
709 		}
710 
711 		if (aux != NULL && !vd->vdev_tmpoffline) {
712 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
713 		} else {
714 			/*
715 			 * We're healthy - clear any previous AUX_STATE values.
716 			 */
717 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
718 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
719 		}
720 
721 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
722 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
723 			    vd->vdev_orig_guid);
724 		}
725 	}
726 
727 	return (nv);
728 }
729 
730 /*
731  * Generate a view of the top-level vdevs.  If we currently have holes
732  * in the namespace, then generate an array which contains a list of holey
733  * vdevs.  Additionally, add the number of top-level children that currently
734  * exist.
735  */
736 void
737 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
738 {
739 	vdev_t *rvd = spa->spa_root_vdev;
740 	uint64_t *array;
741 	uint_t c, idx;
742 
743 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
744 
745 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
746 		vdev_t *tvd = rvd->vdev_child[c];
747 
748 		if (tvd->vdev_ishole) {
749 			array[idx++] = c;
750 		}
751 	}
752 
753 	if (idx) {
754 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
755 		    array, idx) == 0);
756 	}
757 
758 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
759 	    rvd->vdev_children) == 0);
760 
761 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
762 }
763 
764 /*
765  * Returns the configuration from the label of the given vdev. For vdevs
766  * which don't have a txg value stored on their label (i.e. spares/cache)
767  * or have not been completely initialized (txg = 0) just return
768  * the configuration from the first valid label we find. Otherwise,
769  * find the most up-to-date label that does not exceed the specified
770  * 'txg' value.
771  */
772 nvlist_t *
773 vdev_label_read_config(vdev_t *vd, uint64_t txg)
774 {
775 	spa_t *spa = vd->vdev_spa;
776 	nvlist_t *config = NULL;
777 	vdev_phys_t *vp[VDEV_LABELS];
778 	abd_t *vp_abd[VDEV_LABELS];
779 	zio_t *zio[VDEV_LABELS];
780 	uint64_t best_txg = 0;
781 	uint64_t label_txg = 0;
782 	int error = 0;
783 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
784 	    ZIO_FLAG_SPECULATIVE;
785 
786 	ASSERT(vd->vdev_validate_thread == curthread ||
787 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
788 
789 	if (!vdev_readable(vd))
790 		return (NULL);
791 
792 	/*
793 	 * The label for a dRAID distributed spare is not stored on disk.
794 	 * Instead it is generated when needed which allows us to bypass
795 	 * the pipeline when reading the config from the label.
796 	 */
797 	if (vd->vdev_ops == &vdev_draid_spare_ops)
798 		return (vdev_draid_read_config_spare(vd));
799 
800 	for (int l = 0; l < VDEV_LABELS; l++) {
801 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
802 		vp[l] = abd_to_buf(vp_abd[l]);
803 	}
804 
805 retry:
806 	for (int l = 0; l < VDEV_LABELS; l++) {
807 		zio[l] = zio_root(spa, NULL, NULL, flags);
808 
809 		vdev_label_read(zio[l], vd, l, vp_abd[l],
810 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
811 		    NULL, NULL, flags);
812 	}
813 	for (int l = 0; l < VDEV_LABELS; l++) {
814 		nvlist_t *label = NULL;
815 
816 		if (zio_wait(zio[l]) == 0 &&
817 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
818 		    &label, 0) == 0) {
819 			/*
820 			 * Auxiliary vdevs won't have txg values in their
821 			 * labels and newly added vdevs may not have been
822 			 * completely initialized so just return the
823 			 * configuration from the first valid label we
824 			 * encounter.
825 			 */
826 			error = nvlist_lookup_uint64(label,
827 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
828 			if ((error || label_txg == 0) && !config) {
829 				config = label;
830 				for (l++; l < VDEV_LABELS; l++)
831 					zio_wait(zio[l]);
832 				break;
833 			} else if (label_txg <= txg && label_txg > best_txg) {
834 				best_txg = label_txg;
835 				nvlist_free(config);
836 				config = fnvlist_dup(label);
837 			}
838 		}
839 
840 		if (label != NULL) {
841 			nvlist_free(label);
842 			label = NULL;
843 		}
844 	}
845 
846 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
847 		flags |= ZIO_FLAG_TRYHARD;
848 		goto retry;
849 	}
850 
851 	/*
852 	 * We found a valid label but it didn't pass txg restrictions.
853 	 */
854 	if (config == NULL && label_txg != 0) {
855 		vdev_dbgmsg(vd, "label discarded as txg is too large "
856 		    "(%llu > %llu)", (u_longlong_t)label_txg,
857 		    (u_longlong_t)txg);
858 	}
859 
860 	for (int l = 0; l < VDEV_LABELS; l++) {
861 		abd_free(vp_abd[l]);
862 	}
863 
864 	return (config);
865 }
866 
867 /*
868  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
869  * in with the device guid if this spare is active elsewhere on the system.
870  */
871 static boolean_t
872 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
873     uint64_t *spare_guid, uint64_t *l2cache_guid)
874 {
875 	spa_t *spa = vd->vdev_spa;
876 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
877 	uint64_t vdtxg = 0;
878 	nvlist_t *label;
879 
880 	if (spare_guid)
881 		*spare_guid = 0ULL;
882 	if (l2cache_guid)
883 		*l2cache_guid = 0ULL;
884 
885 	/*
886 	 * Read the label, if any, and perform some basic sanity checks.
887 	 */
888 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
889 		return (B_FALSE);
890 
891 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
892 	    &vdtxg);
893 
894 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
895 	    &state) != 0 ||
896 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
897 	    &device_guid) != 0) {
898 		nvlist_free(label);
899 		return (B_FALSE);
900 	}
901 
902 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
903 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
904 	    &pool_guid) != 0 ||
905 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
906 	    &txg) != 0)) {
907 		nvlist_free(label);
908 		return (B_FALSE);
909 	}
910 
911 	nvlist_free(label);
912 
913 	/*
914 	 * Check to see if this device indeed belongs to the pool it claims to
915 	 * be a part of.  The only way this is allowed is if the device is a hot
916 	 * spare (which we check for later on).
917 	 */
918 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
919 	    !spa_guid_exists(pool_guid, device_guid) &&
920 	    !spa_spare_exists(device_guid, NULL, NULL) &&
921 	    !spa_l2cache_exists(device_guid, NULL))
922 		return (B_FALSE);
923 
924 	/*
925 	 * If the transaction group is zero, then this an initialized (but
926 	 * unused) label.  This is only an error if the create transaction
927 	 * on-disk is the same as the one we're using now, in which case the
928 	 * user has attempted to add the same vdev multiple times in the same
929 	 * transaction.
930 	 */
931 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
932 	    txg == 0 && vdtxg == crtxg)
933 		return (B_TRUE);
934 
935 	/*
936 	 * Check to see if this is a spare device.  We do an explicit check for
937 	 * spa_has_spare() here because it may be on our pending list of spares
938 	 * to add.
939 	 */
940 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
941 	    spa_has_spare(spa, device_guid)) {
942 		if (spare_guid)
943 			*spare_guid = device_guid;
944 
945 		switch (reason) {
946 		case VDEV_LABEL_CREATE:
947 			return (B_TRUE);
948 
949 		case VDEV_LABEL_REPLACE:
950 			return (!spa_has_spare(spa, device_guid) ||
951 			    spare_pool != 0ULL);
952 
953 		case VDEV_LABEL_SPARE:
954 			return (spa_has_spare(spa, device_guid));
955 		default:
956 			break;
957 		}
958 	}
959 
960 	/*
961 	 * Check to see if this is an l2cache device.
962 	 */
963 	if (spa_l2cache_exists(device_guid, NULL) ||
964 	    spa_has_l2cache(spa, device_guid)) {
965 		if (l2cache_guid)
966 			*l2cache_guid = device_guid;
967 
968 		switch (reason) {
969 		case VDEV_LABEL_CREATE:
970 			return (B_TRUE);
971 
972 		case VDEV_LABEL_REPLACE:
973 			return (!spa_has_l2cache(spa, device_guid));
974 
975 		case VDEV_LABEL_L2CACHE:
976 			return (spa_has_l2cache(spa, device_guid));
977 		default:
978 			break;
979 		}
980 	}
981 
982 	/*
983 	 * We can't rely on a pool's state if it's been imported
984 	 * read-only.  Instead we look to see if the pools is marked
985 	 * read-only in the namespace and set the state to active.
986 	 */
987 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
988 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
989 	    spa_mode(spa) == SPA_MODE_READ)
990 		state = POOL_STATE_ACTIVE;
991 
992 	/*
993 	 * If the device is marked ACTIVE, then this device is in use by another
994 	 * pool on the system.
995 	 */
996 	return (state == POOL_STATE_ACTIVE);
997 }
998 
999 /*
1000  * Initialize a vdev label.  We check to make sure each leaf device is not in
1001  * use, and writable.  We put down an initial label which we will later
1002  * overwrite with a complete label.  Note that it's important to do this
1003  * sequentially, not in parallel, so that we catch cases of multiple use of the
1004  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1005  * itself.
1006  */
1007 int
1008 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1009 {
1010 	spa_t *spa = vd->vdev_spa;
1011 	nvlist_t *label;
1012 	vdev_phys_t *vp;
1013 	abd_t *vp_abd;
1014 	abd_t *bootenv;
1015 	uberblock_t *ub;
1016 	abd_t *ub_abd;
1017 	zio_t *zio;
1018 	char *buf;
1019 	size_t buflen;
1020 	int error;
1021 	uint64_t spare_guid = 0, l2cache_guid = 0;
1022 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1023 
1024 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1025 
1026 	for (int c = 0; c < vd->vdev_children; c++)
1027 		if ((error = vdev_label_init(vd->vdev_child[c],
1028 		    crtxg, reason)) != 0)
1029 			return (error);
1030 
1031 	/* Track the creation time for this vdev */
1032 	vd->vdev_crtxg = crtxg;
1033 
1034 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1035 		return (0);
1036 
1037 	/*
1038 	 * Dead vdevs cannot be initialized.
1039 	 */
1040 	if (vdev_is_dead(vd))
1041 		return (SET_ERROR(EIO));
1042 
1043 	/*
1044 	 * Determine if the vdev is in use.
1045 	 */
1046 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1047 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1048 		return (SET_ERROR(EBUSY));
1049 
1050 	/*
1051 	 * If this is a request to add or replace a spare or l2cache device
1052 	 * that is in use elsewhere on the system, then we must update the
1053 	 * guid (which was initialized to a random value) to reflect the
1054 	 * actual GUID (which is shared between multiple pools).
1055 	 */
1056 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1057 	    spare_guid != 0ULL) {
1058 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1059 
1060 		vd->vdev_guid += guid_delta;
1061 
1062 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1063 			pvd->vdev_guid_sum += guid_delta;
1064 
1065 		/*
1066 		 * If this is a replacement, then we want to fallthrough to the
1067 		 * rest of the code.  If we're adding a spare, then it's already
1068 		 * labeled appropriately and we can just return.
1069 		 */
1070 		if (reason == VDEV_LABEL_SPARE)
1071 			return (0);
1072 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1073 		    reason == VDEV_LABEL_SPLIT);
1074 	}
1075 
1076 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1077 	    l2cache_guid != 0ULL) {
1078 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1079 
1080 		vd->vdev_guid += guid_delta;
1081 
1082 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1083 			pvd->vdev_guid_sum += guid_delta;
1084 
1085 		/*
1086 		 * If this is a replacement, then we want to fallthrough to the
1087 		 * rest of the code.  If we're adding an l2cache, then it's
1088 		 * already labeled appropriately and we can just return.
1089 		 */
1090 		if (reason == VDEV_LABEL_L2CACHE)
1091 			return (0);
1092 		ASSERT(reason == VDEV_LABEL_REPLACE);
1093 	}
1094 
1095 	/*
1096 	 * Initialize its label.
1097 	 */
1098 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1099 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1100 	vp = abd_to_buf(vp_abd);
1101 
1102 	/*
1103 	 * Generate a label describing the pool and our top-level vdev.
1104 	 * We mark it as being from txg 0 to indicate that it's not
1105 	 * really part of an active pool just yet.  The labels will
1106 	 * be written again with a meaningful txg by spa_sync().
1107 	 */
1108 	if (reason == VDEV_LABEL_SPARE ||
1109 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1110 		/*
1111 		 * For inactive hot spares, we generate a special label that
1112 		 * identifies as a mutually shared hot spare.  We write the
1113 		 * label if we are adding a hot spare, or if we are removing an
1114 		 * active hot spare (in which case we want to revert the
1115 		 * labels).
1116 		 */
1117 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1118 
1119 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1120 		    spa_version(spa)) == 0);
1121 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1122 		    POOL_STATE_SPARE) == 0);
1123 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1124 		    vd->vdev_guid) == 0);
1125 	} else if (reason == VDEV_LABEL_L2CACHE ||
1126 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1127 		/*
1128 		 * For level 2 ARC devices, add a special label.
1129 		 */
1130 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1131 
1132 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1133 		    spa_version(spa)) == 0);
1134 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1135 		    POOL_STATE_L2CACHE) == 0);
1136 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1137 		    vd->vdev_guid) == 0);
1138 	} else {
1139 		uint64_t txg = 0ULL;
1140 
1141 		if (reason == VDEV_LABEL_SPLIT)
1142 			txg = spa->spa_uberblock.ub_txg;
1143 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1144 
1145 		/*
1146 		 * Add our creation time.  This allows us to detect multiple
1147 		 * vdev uses as described above, and automatically expires if we
1148 		 * fail.
1149 		 */
1150 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1151 		    crtxg) == 0);
1152 	}
1153 
1154 	buf = vp->vp_nvlist;
1155 	buflen = sizeof (vp->vp_nvlist);
1156 
1157 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1158 	if (error != 0) {
1159 		nvlist_free(label);
1160 		abd_free(vp_abd);
1161 		/* EFAULT means nvlist_pack ran out of room */
1162 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1163 	}
1164 
1165 	/*
1166 	 * Initialize uberblock template.
1167 	 */
1168 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1169 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1170 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1171 	ub = abd_to_buf(ub_abd);
1172 	ub->ub_txg = 0;
1173 
1174 	/* Initialize the 2nd padding area. */
1175 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1176 	abd_zero(bootenv, VDEV_PAD_SIZE);
1177 
1178 	/*
1179 	 * Write everything in parallel.
1180 	 */
1181 retry:
1182 	zio = zio_root(spa, NULL, NULL, flags);
1183 
1184 	for (int l = 0; l < VDEV_LABELS; l++) {
1185 
1186 		vdev_label_write(zio, vd, l, vp_abd,
1187 		    offsetof(vdev_label_t, vl_vdev_phys),
1188 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1189 
1190 		/*
1191 		 * Skip the 1st padding area.
1192 		 * Zero out the 2nd padding area where it might have
1193 		 * left over data from previous filesystem format.
1194 		 */
1195 		vdev_label_write(zio, vd, l, bootenv,
1196 		    offsetof(vdev_label_t, vl_be),
1197 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1198 
1199 		vdev_label_write(zio, vd, l, ub_abd,
1200 		    offsetof(vdev_label_t, vl_uberblock),
1201 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1202 	}
1203 
1204 	error = zio_wait(zio);
1205 
1206 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1207 		flags |= ZIO_FLAG_TRYHARD;
1208 		goto retry;
1209 	}
1210 
1211 	nvlist_free(label);
1212 	abd_free(bootenv);
1213 	abd_free(ub_abd);
1214 	abd_free(vp_abd);
1215 
1216 	/*
1217 	 * If this vdev hasn't been previously identified as a spare, then we
1218 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1219 	 * exists as a spare elsewhere in the system.  Do the same for
1220 	 * level 2 ARC devices.
1221 	 */
1222 	if (error == 0 && !vd->vdev_isspare &&
1223 	    (reason == VDEV_LABEL_SPARE ||
1224 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1225 		spa_spare_add(vd);
1226 
1227 	if (error == 0 && !vd->vdev_isl2cache &&
1228 	    (reason == VDEV_LABEL_L2CACHE ||
1229 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1230 		spa_l2cache_add(vd);
1231 
1232 	return (error);
1233 }
1234 
1235 /*
1236  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1237  * callback to finish, store our abd in the callback pointer. Otherwise, we
1238  * just free our abd and return.
1239  */
1240 static void
1241 vdev_label_read_bootenv_done(zio_t *zio)
1242 {
1243 	zio_t *rio = zio->io_private;
1244 	abd_t **cbp = rio->io_private;
1245 
1246 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1247 
1248 	if (zio->io_error == 0) {
1249 		mutex_enter(&rio->io_lock);
1250 		if (*cbp == NULL) {
1251 			/* Will free this buffer in vdev_label_read_bootenv. */
1252 			*cbp = zio->io_abd;
1253 		} else {
1254 			abd_free(zio->io_abd);
1255 		}
1256 		mutex_exit(&rio->io_lock);
1257 	} else {
1258 		abd_free(zio->io_abd);
1259 	}
1260 }
1261 
1262 static void
1263 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1264 {
1265 	for (int c = 0; c < vd->vdev_children; c++)
1266 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1267 
1268 	/*
1269 	 * We just use the first label that has a correct checksum; the
1270 	 * bootloader should have rewritten them all to be the same on boot,
1271 	 * and any changes we made since boot have been the same across all
1272 	 * labels.
1273 	 */
1274 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1275 		for (int l = 0; l < VDEV_LABELS; l++) {
1276 			vdev_label_read(zio, vd, l,
1277 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1278 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1279 			    vdev_label_read_bootenv_done, zio, flags);
1280 		}
1281 	}
1282 }
1283 
1284 int
1285 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1286 {
1287 	nvlist_t *config;
1288 	spa_t *spa = rvd->vdev_spa;
1289 	abd_t *abd = NULL;
1290 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1291 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1292 
1293 	ASSERT(bootenv);
1294 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1295 
1296 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1297 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1298 	int err = zio_wait(zio);
1299 
1300 	if (abd != NULL) {
1301 		char *buf;
1302 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1303 
1304 		vbe->vbe_version = ntohll(vbe->vbe_version);
1305 		switch (vbe->vbe_version) {
1306 		case VB_RAW:
1307 			/*
1308 			 * if we have textual data in vbe_bootenv, create nvlist
1309 			 * with key "envmap".
1310 			 */
1311 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1312 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1313 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1314 			    vbe->vbe_bootenv);
1315 			break;
1316 
1317 		case VB_NVLIST:
1318 			err = nvlist_unpack(vbe->vbe_bootenv,
1319 			    sizeof (vbe->vbe_bootenv), &config, 0);
1320 			if (err == 0) {
1321 				fnvlist_merge(bootenv, config);
1322 				nvlist_free(config);
1323 				break;
1324 			}
1325 			zfs_fallthrough;
1326 		default:
1327 			/* Check for FreeBSD zfs bootonce command string */
1328 			buf = abd_to_buf(abd);
1329 			if (*buf == '\0') {
1330 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1331 				    VB_NVLIST);
1332 				break;
1333 			}
1334 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1335 		}
1336 
1337 		/*
1338 		 * abd was allocated in vdev_label_read_bootenv_impl()
1339 		 */
1340 		abd_free(abd);
1341 		/*
1342 		 * If we managed to read any successfully,
1343 		 * return success.
1344 		 */
1345 		return (0);
1346 	}
1347 	return (err);
1348 }
1349 
1350 int
1351 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1352 {
1353 	zio_t *zio;
1354 	spa_t *spa = vd->vdev_spa;
1355 	vdev_boot_envblock_t *bootenv;
1356 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1357 	int error;
1358 	size_t nvsize;
1359 	char *nvbuf;
1360 	const char *tmp;
1361 
1362 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1363 	if (error != 0)
1364 		return (SET_ERROR(error));
1365 
1366 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1367 		return (SET_ERROR(E2BIG));
1368 	}
1369 
1370 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1371 
1372 	error = ENXIO;
1373 	for (int c = 0; c < vd->vdev_children; c++) {
1374 		int child_err;
1375 
1376 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1377 		/*
1378 		 * As long as any of the disks managed to write all of their
1379 		 * labels successfully, return success.
1380 		 */
1381 		if (child_err == 0)
1382 			error = child_err;
1383 	}
1384 
1385 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1386 	    !vdev_writeable(vd)) {
1387 		return (error);
1388 	}
1389 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1390 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1391 	abd_zero(abd, VDEV_PAD_SIZE);
1392 
1393 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1394 	nvbuf = bootenv->vbe_bootenv;
1395 	nvsize = sizeof (bootenv->vbe_bootenv);
1396 
1397 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1398 	switch (bootenv->vbe_version) {
1399 	case VB_RAW:
1400 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1401 			(void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1402 		}
1403 		error = 0;
1404 		break;
1405 
1406 	case VB_NVLIST:
1407 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1408 		    KM_SLEEP);
1409 		break;
1410 
1411 	default:
1412 		error = EINVAL;
1413 		break;
1414 	}
1415 
1416 	if (error == 0) {
1417 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1418 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1419 	} else {
1420 		abd_free(abd);
1421 		return (SET_ERROR(error));
1422 	}
1423 
1424 retry:
1425 	zio = zio_root(spa, NULL, NULL, flags);
1426 	for (int l = 0; l < VDEV_LABELS; l++) {
1427 		vdev_label_write(zio, vd, l, abd,
1428 		    offsetof(vdev_label_t, vl_be),
1429 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1430 	}
1431 
1432 	error = zio_wait(zio);
1433 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1434 		flags |= ZIO_FLAG_TRYHARD;
1435 		goto retry;
1436 	}
1437 
1438 	abd_free(abd);
1439 	return (error);
1440 }
1441 
1442 /*
1443  * ==========================================================================
1444  * uberblock load/sync
1445  * ==========================================================================
1446  */
1447 
1448 /*
1449  * Consider the following situation: txg is safely synced to disk.  We've
1450  * written the first uberblock for txg + 1, and then we lose power.  When we
1451  * come back up, we fail to see the uberblock for txg + 1 because, say,
1452  * it was on a mirrored device and the replica to which we wrote txg + 1
1453  * is now offline.  If we then make some changes and sync txg + 1, and then
1454  * the missing replica comes back, then for a few seconds we'll have two
1455  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1456  * among uberblocks with equal txg, choose the one with the latest timestamp.
1457  */
1458 static int
1459 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1460 {
1461 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1462 
1463 	if (likely(cmp))
1464 		return (cmp);
1465 
1466 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1467 	if (likely(cmp))
1468 		return (cmp);
1469 
1470 	/*
1471 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1472 	 * ZFS, e.g. OpenZFS >= 0.7.
1473 	 *
1474 	 * If one ub has MMP and the other does not, they were written by
1475 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1476 	 * a 0 value.
1477 	 *
1478 	 * Since timestamp and txg are the same if we get this far, either is
1479 	 * acceptable for importing the pool.
1480 	 */
1481 	unsigned int seq1 = 0;
1482 	unsigned int seq2 = 0;
1483 
1484 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1485 		seq1 = MMP_SEQ(ub1);
1486 
1487 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1488 		seq2 = MMP_SEQ(ub2);
1489 
1490 	return (TREE_CMP(seq1, seq2));
1491 }
1492 
1493 struct ubl_cbdata {
1494 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1495 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1496 };
1497 
1498 static void
1499 vdev_uberblock_load_done(zio_t *zio)
1500 {
1501 	vdev_t *vd = zio->io_vd;
1502 	spa_t *spa = zio->io_spa;
1503 	zio_t *rio = zio->io_private;
1504 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1505 	struct ubl_cbdata *cbp = rio->io_private;
1506 
1507 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1508 
1509 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1510 		mutex_enter(&rio->io_lock);
1511 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1512 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1513 			/*
1514 			 * Keep track of the vdev in which this uberblock
1515 			 * was found. We will use this information later
1516 			 * to obtain the config nvlist associated with
1517 			 * this uberblock.
1518 			 */
1519 			*cbp->ubl_ubbest = *ub;
1520 			cbp->ubl_vd = vd;
1521 		}
1522 		mutex_exit(&rio->io_lock);
1523 	}
1524 
1525 	abd_free(zio->io_abd);
1526 }
1527 
1528 static void
1529 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1530     struct ubl_cbdata *cbp)
1531 {
1532 	for (int c = 0; c < vd->vdev_children; c++)
1533 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1534 
1535 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1536 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1537 		for (int l = 0; l < VDEV_LABELS; l++) {
1538 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1539 				vdev_label_read(zio, vd, l,
1540 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1541 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1542 				    VDEV_UBERBLOCK_SIZE(vd),
1543 				    vdev_uberblock_load_done, zio, flags);
1544 			}
1545 		}
1546 	}
1547 }
1548 
1549 /*
1550  * Reads the 'best' uberblock from disk along with its associated
1551  * configuration. First, we read the uberblock array of each label of each
1552  * vdev, keeping track of the uberblock with the highest txg in each array.
1553  * Then, we read the configuration from the same vdev as the best uberblock.
1554  */
1555 void
1556 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1557 {
1558 	zio_t *zio;
1559 	spa_t *spa = rvd->vdev_spa;
1560 	struct ubl_cbdata cb;
1561 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1562 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1563 
1564 	ASSERT(ub);
1565 	ASSERT(config);
1566 
1567 	memset(ub, 0, sizeof (uberblock_t));
1568 	*config = NULL;
1569 
1570 	cb.ubl_ubbest = ub;
1571 	cb.ubl_vd = NULL;
1572 
1573 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1574 	zio = zio_root(spa, NULL, &cb, flags);
1575 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1576 	(void) zio_wait(zio);
1577 
1578 	/*
1579 	 * It's possible that the best uberblock was discovered on a label
1580 	 * that has a configuration which was written in a future txg.
1581 	 * Search all labels on this vdev to find the configuration that
1582 	 * matches the txg for our uberblock.
1583 	 */
1584 	if (cb.ubl_vd != NULL) {
1585 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1586 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1587 
1588 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1589 		if (*config == NULL && spa->spa_extreme_rewind) {
1590 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1591 			    "Trying again without txg restrictions.");
1592 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1593 		}
1594 		if (*config == NULL) {
1595 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1596 		}
1597 	}
1598 	spa_config_exit(spa, SCL_ALL, FTAG);
1599 }
1600 
1601 /*
1602  * For use when a leaf vdev is expanded.
1603  * The location of labels 2 and 3 changed, and at the new location the
1604  * uberblock rings are either empty or contain garbage.  The sync will write
1605  * new configs there because the vdev is dirty, but expansion also needs the
1606  * uberblock rings copied.  Read them from label 0 which did not move.
1607  *
1608  * Since the point is to populate labels {2,3} with valid uberblocks,
1609  * we zero uberblocks we fail to read or which are not valid.
1610  */
1611 
1612 static void
1613 vdev_copy_uberblocks(vdev_t *vd)
1614 {
1615 	abd_t *ub_abd;
1616 	zio_t *write_zio;
1617 	int locks = (SCL_L2ARC | SCL_ZIO);
1618 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1619 	    ZIO_FLAG_SPECULATIVE;
1620 
1621 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1622 	    SCL_STATE);
1623 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1624 
1625 	/*
1626 	 * No uberblocks are stored on distributed spares, they may be
1627 	 * safely skipped when expanding a leaf vdev.
1628 	 */
1629 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1630 		return;
1631 
1632 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1633 
1634 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1635 
1636 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1637 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1638 		const int src_label = 0;
1639 		zio_t *zio;
1640 
1641 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1642 		vdev_label_read(zio, vd, src_label, ub_abd,
1643 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1644 		    NULL, NULL, flags);
1645 
1646 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1647 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1648 
1649 		for (int l = 2; l < VDEV_LABELS; l++)
1650 			vdev_label_write(write_zio, vd, l, ub_abd,
1651 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1652 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1653 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1654 	}
1655 	(void) zio_wait(write_zio);
1656 
1657 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1658 
1659 	abd_free(ub_abd);
1660 }
1661 
1662 /*
1663  * On success, increment root zio's count of good writes.
1664  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1665  */
1666 static void
1667 vdev_uberblock_sync_done(zio_t *zio)
1668 {
1669 	uint64_t *good_writes = zio->io_private;
1670 
1671 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1672 		atomic_inc_64(good_writes);
1673 }
1674 
1675 /*
1676  * Write the uberblock to all labels of all leaves of the specified vdev.
1677  */
1678 static void
1679 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1680     uberblock_t *ub, vdev_t *vd, int flags)
1681 {
1682 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1683 		vdev_uberblock_sync(zio, good_writes,
1684 		    ub, vd->vdev_child[c], flags);
1685 	}
1686 
1687 	if (!vd->vdev_ops->vdev_op_leaf)
1688 		return;
1689 
1690 	if (!vdev_writeable(vd))
1691 		return;
1692 
1693 	/*
1694 	 * There's no need to write uberblocks to a distributed spare, they
1695 	 * are already stored on all the leaves of the parent dRAID.  For
1696 	 * this same reason vdev_uberblock_load_impl() skips distributed
1697 	 * spares when reading uberblocks.
1698 	 */
1699 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1700 		return;
1701 
1702 	/* If the vdev was expanded, need to copy uberblock rings. */
1703 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1704 	    vd->vdev_copy_uberblocks == B_TRUE) {
1705 		vdev_copy_uberblocks(vd);
1706 		vd->vdev_copy_uberblocks = B_FALSE;
1707 	}
1708 
1709 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1710 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1711 
1712 	/* Copy the uberblock_t into the ABD */
1713 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1714 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1715 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1716 
1717 	for (int l = 0; l < VDEV_LABELS; l++)
1718 		vdev_label_write(zio, vd, l, ub_abd,
1719 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1720 		    vdev_uberblock_sync_done, good_writes,
1721 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1722 
1723 	abd_free(ub_abd);
1724 }
1725 
1726 /* Sync the uberblocks to all vdevs in svd[] */
1727 static int
1728 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1729 {
1730 	spa_t *spa = svd[0]->vdev_spa;
1731 	zio_t *zio;
1732 	uint64_t good_writes = 0;
1733 
1734 	zio = zio_root(spa, NULL, NULL, flags);
1735 
1736 	for (int v = 0; v < svdcount; v++)
1737 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1738 
1739 	(void) zio_wait(zio);
1740 
1741 	/*
1742 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1743 	 * are no longer needed (because the new uberblocks and the even
1744 	 * labels are safely on disk), so it is safe to overwrite them.
1745 	 */
1746 	zio = zio_root(spa, NULL, NULL, flags);
1747 
1748 	for (int v = 0; v < svdcount; v++) {
1749 		if (vdev_writeable(svd[v])) {
1750 			zio_flush(zio, svd[v]);
1751 		}
1752 	}
1753 
1754 	(void) zio_wait(zio);
1755 
1756 	return (good_writes >= 1 ? 0 : EIO);
1757 }
1758 
1759 /*
1760  * On success, increment the count of good writes for our top-level vdev.
1761  */
1762 static void
1763 vdev_label_sync_done(zio_t *zio)
1764 {
1765 	uint64_t *good_writes = zio->io_private;
1766 
1767 	if (zio->io_error == 0)
1768 		atomic_inc_64(good_writes);
1769 }
1770 
1771 /*
1772  * If there weren't enough good writes, indicate failure to the parent.
1773  */
1774 static void
1775 vdev_label_sync_top_done(zio_t *zio)
1776 {
1777 	uint64_t *good_writes = zio->io_private;
1778 
1779 	if (*good_writes == 0)
1780 		zio->io_error = SET_ERROR(EIO);
1781 
1782 	kmem_free(good_writes, sizeof (uint64_t));
1783 }
1784 
1785 /*
1786  * We ignore errors for log and cache devices, simply free the private data.
1787  */
1788 static void
1789 vdev_label_sync_ignore_done(zio_t *zio)
1790 {
1791 	kmem_free(zio->io_private, sizeof (uint64_t));
1792 }
1793 
1794 /*
1795  * Write all even or odd labels to all leaves of the specified vdev.
1796  */
1797 static void
1798 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1799     vdev_t *vd, int l, uint64_t txg, int flags)
1800 {
1801 	nvlist_t *label;
1802 	vdev_phys_t *vp;
1803 	abd_t *vp_abd;
1804 	char *buf;
1805 	size_t buflen;
1806 
1807 	for (int c = 0; c < vd->vdev_children; c++) {
1808 		vdev_label_sync(zio, good_writes,
1809 		    vd->vdev_child[c], l, txg, flags);
1810 	}
1811 
1812 	if (!vd->vdev_ops->vdev_op_leaf)
1813 		return;
1814 
1815 	if (!vdev_writeable(vd))
1816 		return;
1817 
1818 	/*
1819 	 * The top-level config never needs to be written to a distributed
1820 	 * spare.  When read vdev_dspare_label_read_config() will generate
1821 	 * the config for the vdev_label_read_config().
1822 	 */
1823 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1824 		return;
1825 
1826 	/*
1827 	 * Generate a label describing the top-level config to which we belong.
1828 	 */
1829 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1830 
1831 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1832 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1833 	vp = abd_to_buf(vp_abd);
1834 
1835 	buf = vp->vp_nvlist;
1836 	buflen = sizeof (vp->vp_nvlist);
1837 
1838 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1839 		for (; l < VDEV_LABELS; l += 2) {
1840 			vdev_label_write(zio, vd, l, vp_abd,
1841 			    offsetof(vdev_label_t, vl_vdev_phys),
1842 			    sizeof (vdev_phys_t),
1843 			    vdev_label_sync_done, good_writes,
1844 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1845 		}
1846 	}
1847 
1848 	abd_free(vp_abd);
1849 	nvlist_free(label);
1850 }
1851 
1852 static int
1853 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1854 {
1855 	list_t *dl = &spa->spa_config_dirty_list;
1856 	vdev_t *vd;
1857 	zio_t *zio;
1858 	int error;
1859 
1860 	/*
1861 	 * Write the new labels to disk.
1862 	 */
1863 	zio = zio_root(spa, NULL, NULL, flags);
1864 
1865 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1866 		uint64_t *good_writes;
1867 
1868 		ASSERT(!vd->vdev_ishole);
1869 
1870 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1871 		zio_t *vio = zio_null(zio, spa, NULL,
1872 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1873 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1874 		    good_writes, flags);
1875 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1876 		zio_nowait(vio);
1877 	}
1878 
1879 	error = zio_wait(zio);
1880 
1881 	/*
1882 	 * Flush the new labels to disk.
1883 	 */
1884 	zio = zio_root(spa, NULL, NULL, flags);
1885 
1886 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1887 		zio_flush(zio, vd);
1888 
1889 	(void) zio_wait(zio);
1890 
1891 	return (error);
1892 }
1893 
1894 /*
1895  * Sync the uberblock and any changes to the vdev configuration.
1896  *
1897  * The order of operations is carefully crafted to ensure that
1898  * if the system panics or loses power at any time, the state on disk
1899  * is still transactionally consistent.  The in-line comments below
1900  * describe the failure semantics at each stage.
1901  *
1902  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1903  * at any time, you can just call it again, and it will resume its work.
1904  */
1905 int
1906 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1907 {
1908 	spa_t *spa = svd[0]->vdev_spa;
1909 	uberblock_t *ub = &spa->spa_uberblock;
1910 	int error = 0;
1911 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1912 
1913 	ASSERT(svdcount != 0);
1914 retry:
1915 	/*
1916 	 * Normally, we don't want to try too hard to write every label and
1917 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1918 	 * sync process to block while we retry.  But if we can't write a
1919 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1920 	 * bailing out and declaring the pool faulted.
1921 	 */
1922 	if (error != 0) {
1923 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1924 			return (error);
1925 		flags |= ZIO_FLAG_TRYHARD;
1926 	}
1927 
1928 	ASSERT(ub->ub_txg <= txg);
1929 
1930 	/*
1931 	 * If this isn't a resync due to I/O errors,
1932 	 * and nothing changed in this transaction group,
1933 	 * and the vdev configuration hasn't changed,
1934 	 * then there's nothing to do.
1935 	 */
1936 	if (ub->ub_txg < txg) {
1937 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1938 		    txg, spa->spa_mmp.mmp_delay);
1939 
1940 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1941 			return (0);
1942 	}
1943 
1944 	if (txg > spa_freeze_txg(spa))
1945 		return (0);
1946 
1947 	ASSERT(txg <= spa->spa_final_txg);
1948 
1949 	/*
1950 	 * Flush the write cache of every disk that's been written to
1951 	 * in this transaction group.  This ensures that all blocks
1952 	 * written in this txg will be committed to stable storage
1953 	 * before any uberblock that references them.
1954 	 */
1955 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1956 
1957 	for (vdev_t *vd =
1958 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1959 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1960 		zio_flush(zio, vd);
1961 
1962 	(void) zio_wait(zio);
1963 
1964 	/*
1965 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1966 	 * system dies in the middle of this process, that's OK: all of the
1967 	 * even labels that made it to disk will be newer than any uberblock,
1968 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1969 	 * which have not yet been touched, will still be valid.  We flush
1970 	 * the new labels to disk to ensure that all even-label updates
1971 	 * are committed to stable storage before the uberblock update.
1972 	 */
1973 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1974 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1975 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1976 			    "for pool '%s' when syncing out the even labels "
1977 			    "of dirty vdevs", error, spa_name(spa));
1978 		}
1979 		goto retry;
1980 	}
1981 
1982 	/*
1983 	 * Sync the uberblocks to all vdevs in svd[].
1984 	 * If the system dies in the middle of this step, there are two cases
1985 	 * to consider, and the on-disk state is consistent either way:
1986 	 *
1987 	 * (1)	If none of the new uberblocks made it to disk, then the
1988 	 *	previous uberblock will be the newest, and the odd labels
1989 	 *	(which had not yet been touched) will be valid with respect
1990 	 *	to that uberblock.
1991 	 *
1992 	 * (2)	If one or more new uberblocks made it to disk, then they
1993 	 *	will be the newest, and the even labels (which had all
1994 	 *	been successfully committed) will be valid with respect
1995 	 *	to the new uberblocks.
1996 	 */
1997 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1998 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1999 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2000 			    "%d for pool '%s'", error, spa_name(spa));
2001 		}
2002 		goto retry;
2003 	}
2004 
2005 	if (spa_multihost(spa))
2006 		mmp_update_uberblock(spa, ub);
2007 
2008 	/*
2009 	 * Sync out odd labels for every dirty vdev.  If the system dies
2010 	 * in the middle of this process, the even labels and the new
2011 	 * uberblocks will suffice to open the pool.  The next time
2012 	 * the pool is opened, the first thing we'll do -- before any
2013 	 * user data is modified -- is mark every vdev dirty so that
2014 	 * all labels will be brought up to date.  We flush the new labels
2015 	 * to disk to ensure that all odd-label updates are committed to
2016 	 * stable storage before the next transaction group begins.
2017 	 */
2018 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2019 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2020 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2021 			    "for pool '%s' when syncing out the odd labels of "
2022 			    "dirty vdevs", error, spa_name(spa));
2023 		}
2024 		goto retry;
2025 	}
2026 
2027 	return (0);
2028 }
2029