xref: /freebsd-src/sys/contrib/openzfs/module/zfs/vdev_label.c (revision e716630d4cf89e69ec3f675ebfceee09f1a85e05)
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_raidz.h>
146 #include <sys/vdev_draid.h>
147 #include <sys/uberblock_impl.h>
148 #include <sys/metaslab.h>
149 #include <sys/metaslab_impl.h>
150 #include <sys/zio.h>
151 #include <sys/dsl_scan.h>
152 #include <sys/abd.h>
153 #include <sys/fs/zfs.h>
154 #include <sys/byteorder.h>
155 #include <sys/zfs_bootenv.h>
156 
157 /*
158  * Basic routines to read and write from a vdev label.
159  * Used throughout the rest of this file.
160  */
161 uint64_t
162 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
163 {
164 	ASSERT(offset < sizeof (vdev_label_t));
165 	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
166 
167 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
168 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
169 }
170 
171 /*
172  * Returns back the vdev label associated with the passed in offset.
173  */
174 int
175 vdev_label_number(uint64_t psize, uint64_t offset)
176 {
177 	int l;
178 
179 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
180 		offset -= psize - VDEV_LABEL_END_SIZE;
181 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
182 	}
183 	l = offset / sizeof (vdev_label_t);
184 	return (l < VDEV_LABELS ? l : -1);
185 }
186 
187 static void
188 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
189     uint64_t size, zio_done_func_t *done, void *private, int flags)
190 {
191 	ASSERT(
192 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
193 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
194 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
195 
196 	zio_nowait(zio_read_phys(zio, vd,
197 	    vdev_label_offset(vd->vdev_psize, l, offset),
198 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
199 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
200 }
201 
202 void
203 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
204     uint64_t size, zio_done_func_t *done, void *private, int flags)
205 {
206 	ASSERT(
207 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
208 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
209 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
210 
211 	zio_nowait(zio_write_phys(zio, vd,
212 	    vdev_label_offset(vd->vdev_psize, l, offset),
213 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
214 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
215 }
216 
217 /*
218  * Generate the nvlist representing this vdev's stats
219  */
220 void
221 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
222 {
223 	nvlist_t *nvx;
224 	vdev_stat_t *vs;
225 	vdev_stat_ex_t *vsx;
226 
227 	vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
228 	vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
229 
230 	vdev_get_stats_ex(vd, vs, vsx);
231 	fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
232 	    (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
233 
234 	/*
235 	 * Add extended stats into a special extended stats nvlist.  This keeps
236 	 * all the extended stats nicely grouped together.  The extended stats
237 	 * nvlist is then added to the main nvlist.
238 	 */
239 	nvx = fnvlist_alloc();
240 
241 	/* ZIOs in flight to disk */
242 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
243 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
244 
245 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
246 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
247 
248 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
249 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
250 
251 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
252 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
253 
254 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
255 	    vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
256 
257 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
258 	    vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
259 
260 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,
261 	    vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]);
262 
263 	/* ZIOs pending */
264 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
265 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
266 
267 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
268 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
269 
270 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
271 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
272 
273 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
274 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
275 
276 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
277 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
278 
279 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
280 	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
281 
282 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,
283 	    vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]);
284 
285 	/* Histograms */
286 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
287 	    vsx->vsx_total_histo[ZIO_TYPE_READ],
288 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
289 
290 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
291 	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
292 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
293 
294 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
295 	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
296 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
297 
298 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
299 	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
300 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
301 
302 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
303 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
304 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
305 
306 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
307 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
308 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
309 
310 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
311 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
312 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
313 
314 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
315 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
316 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
317 
318 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
319 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
320 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
321 
322 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
323 	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
324 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
325 
326 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,
327 	    vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD],
328 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD]));
329 
330 	/* Request sizes */
331 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
332 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
333 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
334 
335 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
336 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
337 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
338 
339 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
340 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
341 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
342 
343 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
344 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
345 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
346 
347 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
348 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
349 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
350 
351 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
352 	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
353 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
354 
355 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,
356 	    vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD],
357 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD]));
358 
359 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
360 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
361 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
362 
363 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
364 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
365 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
366 
367 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
368 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
369 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
370 
371 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
372 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
373 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
374 
375 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
376 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
377 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
378 
379 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
380 	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
381 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
382 
383 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,
384 	    vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD],
385 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD]));
386 
387 	/* IO delays */
388 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
389 
390 	/* Add extended stats nvlist to main nvlist */
391 	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
392 
393 	fnvlist_free(nvx);
394 	kmem_free(vs, sizeof (*vs));
395 	kmem_free(vsx, sizeof (*vsx));
396 }
397 
398 static void
399 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
400 {
401 	spa_t *spa = vd->vdev_spa;
402 
403 	if (vd != spa->spa_root_vdev)
404 		return;
405 
406 	/* provide either current or previous scan information */
407 	pool_scan_stat_t ps;
408 	if (spa_scan_get_stats(spa, &ps) == 0) {
409 		fnvlist_add_uint64_array(nvl,
410 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
411 		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
412 	}
413 
414 	pool_removal_stat_t prs;
415 	if (spa_removal_get_stats(spa, &prs) == 0) {
416 		fnvlist_add_uint64_array(nvl,
417 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
418 		    sizeof (prs) / sizeof (uint64_t));
419 	}
420 
421 	pool_checkpoint_stat_t pcs;
422 	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
423 		fnvlist_add_uint64_array(nvl,
424 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
425 		    sizeof (pcs) / sizeof (uint64_t));
426 	}
427 
428 	pool_raidz_expand_stat_t pres;
429 	if (spa_raidz_expand_get_stats(spa, &pres) == 0) {
430 		fnvlist_add_uint64_array(nvl,
431 		    ZPOOL_CONFIG_RAIDZ_EXPAND_STATS, (uint64_t *)&pres,
432 		    sizeof (pres) / sizeof (uint64_t));
433 	}
434 }
435 
436 static void
437 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
438 {
439 	if (vd == vd->vdev_top) {
440 		vdev_rebuild_stat_t vrs;
441 		if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
442 			fnvlist_add_uint64_array(nvl,
443 			    ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
444 			    sizeof (vrs) / sizeof (uint64_t));
445 		}
446 	}
447 }
448 
449 /*
450  * Generate the nvlist representing this vdev's config.
451  */
452 nvlist_t *
453 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
454     vdev_config_flag_t flags)
455 {
456 	nvlist_t *nv = NULL;
457 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
458 
459 	nv = fnvlist_alloc();
460 
461 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
462 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
463 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
464 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
465 
466 	if (vd->vdev_path != NULL)
467 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
468 
469 	if (vd->vdev_devid != NULL)
470 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
471 
472 	if (vd->vdev_physpath != NULL)
473 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
474 		    vd->vdev_physpath);
475 
476 	if (vd->vdev_enc_sysfs_path != NULL)
477 		fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
478 		    vd->vdev_enc_sysfs_path);
479 
480 	if (vd->vdev_fru != NULL)
481 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
482 
483 	if (vd->vdev_ops->vdev_op_config_generate != NULL)
484 		vd->vdev_ops->vdev_op_config_generate(vd, nv);
485 
486 	if (vd->vdev_wholedisk != -1ULL) {
487 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
488 		    vd->vdev_wholedisk);
489 	}
490 
491 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
492 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
493 
494 	if (vd->vdev_isspare)
495 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
496 
497 	if (flags & VDEV_CONFIG_L2CACHE)
498 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
499 
500 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
501 	    vd == vd->vdev_top) {
502 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
503 		    vd->vdev_ms_array);
504 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
505 		    vd->vdev_ms_shift);
506 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
507 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
508 		    vd->vdev_asize);
509 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
510 		if (vd->vdev_noalloc) {
511 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
512 			    vd->vdev_noalloc);
513 		}
514 
515 		/*
516 		 * Slog devices are removed synchronously so don't
517 		 * persist the vdev_removing flag to the label.
518 		 */
519 		if (vd->vdev_removing && !vd->vdev_islog) {
520 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
521 			    vd->vdev_removing);
522 		}
523 
524 		/* zpool command expects alloc class data */
525 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
526 			const char *bias = NULL;
527 
528 			switch (vd->vdev_alloc_bias) {
529 			case VDEV_BIAS_LOG:
530 				bias = VDEV_ALLOC_BIAS_LOG;
531 				break;
532 			case VDEV_BIAS_SPECIAL:
533 				bias = VDEV_ALLOC_BIAS_SPECIAL;
534 				break;
535 			case VDEV_BIAS_DEDUP:
536 				bias = VDEV_ALLOC_BIAS_DEDUP;
537 				break;
538 			default:
539 				ASSERT3U(vd->vdev_alloc_bias, ==,
540 				    VDEV_BIAS_NONE);
541 			}
542 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
543 			    bias);
544 		}
545 	}
546 
547 	if (vd->vdev_dtl_sm != NULL) {
548 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
549 		    space_map_object(vd->vdev_dtl_sm));
550 	}
551 
552 	if (vic->vic_mapping_object != 0) {
553 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
554 		    vic->vic_mapping_object);
555 	}
556 
557 	if (vic->vic_births_object != 0) {
558 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
559 		    vic->vic_births_object);
560 	}
561 
562 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
563 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
564 		    vic->vic_prev_indirect_vdev);
565 	}
566 
567 	if (vd->vdev_crtxg)
568 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
569 
570 	if (vd->vdev_expansion_time)
571 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
572 		    vd->vdev_expansion_time);
573 
574 	if (flags & VDEV_CONFIG_MOS) {
575 		if (vd->vdev_leaf_zap != 0) {
576 			ASSERT(vd->vdev_ops->vdev_op_leaf);
577 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
578 			    vd->vdev_leaf_zap);
579 		}
580 
581 		if (vd->vdev_top_zap != 0) {
582 			ASSERT(vd == vd->vdev_top);
583 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
584 			    vd->vdev_top_zap);
585 		}
586 
587 		if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
588 		    spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
589 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
590 			    vd->vdev_root_zap);
591 		}
592 
593 		if (vd->vdev_resilver_deferred) {
594 			ASSERT(vd->vdev_ops->vdev_op_leaf);
595 			ASSERT(spa->spa_resilver_deferred);
596 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
597 		}
598 	}
599 
600 	if (getstats) {
601 		vdev_config_generate_stats(vd, nv);
602 
603 		root_vdev_actions_getprogress(vd, nv);
604 		top_vdev_actions_getprogress(vd, nv);
605 
606 		/*
607 		 * Note: this can be called from open context
608 		 * (spa_get_stats()), so we need the rwlock to prevent
609 		 * the mapping from being changed by condensing.
610 		 */
611 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
612 		if (vd->vdev_indirect_mapping != NULL) {
613 			ASSERT(vd->vdev_indirect_births != NULL);
614 			vdev_indirect_mapping_t *vim =
615 			    vd->vdev_indirect_mapping;
616 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
617 			    vdev_indirect_mapping_size(vim));
618 		}
619 		rw_exit(&vd->vdev_indirect_rwlock);
620 		if (vd->vdev_mg != NULL &&
621 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
622 			/*
623 			 * Compute approximately how much memory would be used
624 			 * for the indirect mapping if this device were to
625 			 * be removed.
626 			 *
627 			 * Note: If the frag metric is invalid, then not
628 			 * enough metaslabs have been converted to have
629 			 * histograms.
630 			 */
631 			uint64_t seg_count = 0;
632 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
633 
634 			/*
635 			 * There are the same number of allocated segments
636 			 * as free segments, so we will have at least one
637 			 * entry per free segment.  However, small free
638 			 * segments (smaller than vdev_removal_max_span)
639 			 * will be combined with adjacent allocated segments
640 			 * as a single mapping.
641 			 */
642 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
643 				if (i + 1 < highbit64(vdev_removal_max_span)
644 				    - 1) {
645 					to_alloc +=
646 					    vd->vdev_mg->mg_histogram[i] <<
647 					    (i + 1);
648 				} else {
649 					seg_count +=
650 					    vd->vdev_mg->mg_histogram[i];
651 				}
652 			}
653 
654 			/*
655 			 * The maximum length of a mapping is
656 			 * zfs_remove_max_segment, so we need at least one entry
657 			 * per zfs_remove_max_segment of allocated data.
658 			 */
659 			seg_count += to_alloc / spa_remove_max_segment(spa);
660 
661 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
662 			    seg_count *
663 			    sizeof (vdev_indirect_mapping_entry_phys_t));
664 		}
665 	}
666 
667 	if (!vd->vdev_ops->vdev_op_leaf) {
668 		nvlist_t **child;
669 		uint64_t c;
670 
671 		ASSERT(!vd->vdev_ishole);
672 
673 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
674 		    KM_SLEEP);
675 
676 		for (c = 0; c < vd->vdev_children; c++) {
677 			child[c] = vdev_config_generate(spa, vd->vdev_child[c],
678 			    getstats, flags);
679 		}
680 
681 		fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
682 		    (const nvlist_t * const *)child, vd->vdev_children);
683 
684 		for (c = 0; c < vd->vdev_children; c++)
685 			nvlist_free(child[c]);
686 
687 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
688 
689 	} else {
690 		const char *aux = NULL;
691 
692 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
693 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
694 		if (vd->vdev_resilver_txg != 0)
695 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
696 			    vd->vdev_resilver_txg);
697 		if (vd->vdev_rebuild_txg != 0)
698 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
699 			    vd->vdev_rebuild_txg);
700 		if (vd->vdev_faulted)
701 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
702 		if (vd->vdev_degraded)
703 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
704 		if (vd->vdev_removed)
705 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
706 		if (vd->vdev_unspare)
707 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
708 		if (vd->vdev_ishole)
709 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
710 
711 		/* Set the reason why we're FAULTED/DEGRADED. */
712 		switch (vd->vdev_stat.vs_aux) {
713 		case VDEV_AUX_ERR_EXCEEDED:
714 			aux = "err_exceeded";
715 			break;
716 
717 		case VDEV_AUX_EXTERNAL:
718 			aux = "external";
719 			break;
720 		}
721 
722 		if (aux != NULL && !vd->vdev_tmpoffline) {
723 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
724 		} else {
725 			/*
726 			 * We're healthy - clear any previous AUX_STATE values.
727 			 */
728 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
729 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
730 		}
731 
732 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
733 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
734 			    vd->vdev_orig_guid);
735 		}
736 	}
737 
738 	return (nv);
739 }
740 
741 /*
742  * Generate a view of the top-level vdevs.  If we currently have holes
743  * in the namespace, then generate an array which contains a list of holey
744  * vdevs.  Additionally, add the number of top-level children that currently
745  * exist.
746  */
747 void
748 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
749 {
750 	vdev_t *rvd = spa->spa_root_vdev;
751 	uint64_t *array;
752 	uint_t c, idx;
753 
754 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
755 
756 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
757 		vdev_t *tvd = rvd->vdev_child[c];
758 
759 		if (tvd->vdev_ishole) {
760 			array[idx++] = c;
761 		}
762 	}
763 
764 	if (idx) {
765 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
766 		    array, idx) == 0);
767 	}
768 
769 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
770 	    rvd->vdev_children) == 0);
771 
772 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
773 }
774 
775 /*
776  * Returns the configuration from the label of the given vdev. For vdevs
777  * which don't have a txg value stored on their label (i.e. spares/cache)
778  * or have not been completely initialized (txg = 0) just return
779  * the configuration from the first valid label we find. Otherwise,
780  * find the most up-to-date label that does not exceed the specified
781  * 'txg' value.
782  */
783 nvlist_t *
784 vdev_label_read_config(vdev_t *vd, uint64_t txg)
785 {
786 	spa_t *spa = vd->vdev_spa;
787 	nvlist_t *config = NULL;
788 	vdev_phys_t *vp[VDEV_LABELS];
789 	abd_t *vp_abd[VDEV_LABELS];
790 	zio_t *zio[VDEV_LABELS];
791 	uint64_t best_txg = 0;
792 	uint64_t label_txg = 0;
793 	int error = 0;
794 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
795 	    ZIO_FLAG_SPECULATIVE;
796 
797 	ASSERT(vd->vdev_validate_thread == curthread ||
798 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
799 
800 	if (!vdev_readable(vd))
801 		return (NULL);
802 
803 	/*
804 	 * The label for a dRAID distributed spare is not stored on disk.
805 	 * Instead it is generated when needed which allows us to bypass
806 	 * the pipeline when reading the config from the label.
807 	 */
808 	if (vd->vdev_ops == &vdev_draid_spare_ops)
809 		return (vdev_draid_read_config_spare(vd));
810 
811 	for (int l = 0; l < VDEV_LABELS; l++) {
812 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
813 		vp[l] = abd_to_buf(vp_abd[l]);
814 	}
815 
816 retry:
817 	for (int l = 0; l < VDEV_LABELS; l++) {
818 		zio[l] = zio_root(spa, NULL, NULL, flags);
819 
820 		vdev_label_read(zio[l], vd, l, vp_abd[l],
821 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
822 		    NULL, NULL, flags);
823 	}
824 	for (int l = 0; l < VDEV_LABELS; l++) {
825 		nvlist_t *label = NULL;
826 
827 		if (zio_wait(zio[l]) == 0 &&
828 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
829 		    &label, 0) == 0) {
830 			/*
831 			 * Auxiliary vdevs won't have txg values in their
832 			 * labels and newly added vdevs may not have been
833 			 * completely initialized so just return the
834 			 * configuration from the first valid label we
835 			 * encounter.
836 			 */
837 			error = nvlist_lookup_uint64(label,
838 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
839 			if ((error || label_txg == 0) && !config) {
840 				config = label;
841 				for (l++; l < VDEV_LABELS; l++)
842 					zio_wait(zio[l]);
843 				break;
844 			} else if (label_txg <= txg && label_txg > best_txg) {
845 				best_txg = label_txg;
846 				nvlist_free(config);
847 				config = fnvlist_dup(label);
848 			}
849 		}
850 
851 		if (label != NULL) {
852 			nvlist_free(label);
853 			label = NULL;
854 		}
855 	}
856 
857 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
858 		flags |= ZIO_FLAG_TRYHARD;
859 		goto retry;
860 	}
861 
862 	/*
863 	 * We found a valid label but it didn't pass txg restrictions.
864 	 */
865 	if (config == NULL && label_txg != 0) {
866 		vdev_dbgmsg(vd, "label discarded as txg is too large "
867 		    "(%llu > %llu)", (u_longlong_t)label_txg,
868 		    (u_longlong_t)txg);
869 	}
870 
871 	for (int l = 0; l < VDEV_LABELS; l++) {
872 		abd_free(vp_abd[l]);
873 	}
874 
875 	return (config);
876 }
877 
878 /*
879  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
880  * in with the device guid if this spare is active elsewhere on the system.
881  */
882 static boolean_t
883 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
884     uint64_t *spare_guid, uint64_t *l2cache_guid)
885 {
886 	spa_t *spa = vd->vdev_spa;
887 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
888 	uint64_t vdtxg = 0;
889 	nvlist_t *label;
890 
891 	if (spare_guid)
892 		*spare_guid = 0ULL;
893 	if (l2cache_guid)
894 		*l2cache_guid = 0ULL;
895 
896 	/*
897 	 * Read the label, if any, and perform some basic sanity checks.
898 	 */
899 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
900 		return (B_FALSE);
901 
902 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
903 	    &vdtxg);
904 
905 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
906 	    &state) != 0 ||
907 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
908 	    &device_guid) != 0) {
909 		nvlist_free(label);
910 		return (B_FALSE);
911 	}
912 
913 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
914 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
915 	    &pool_guid) != 0 ||
916 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
917 	    &txg) != 0)) {
918 		nvlist_free(label);
919 		return (B_FALSE);
920 	}
921 
922 	nvlist_free(label);
923 
924 	/*
925 	 * Check to see if this device indeed belongs to the pool it claims to
926 	 * be a part of.  The only way this is allowed is if the device is a hot
927 	 * spare (which we check for later on).
928 	 */
929 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
930 	    !spa_guid_exists(pool_guid, device_guid) &&
931 	    !spa_spare_exists(device_guid, NULL, NULL) &&
932 	    !spa_l2cache_exists(device_guid, NULL))
933 		return (B_FALSE);
934 
935 	/*
936 	 * If the transaction group is zero, then this an initialized (but
937 	 * unused) label.  This is only an error if the create transaction
938 	 * on-disk is the same as the one we're using now, in which case the
939 	 * user has attempted to add the same vdev multiple times in the same
940 	 * transaction.
941 	 */
942 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
943 	    txg == 0 && vdtxg == crtxg)
944 		return (B_TRUE);
945 
946 	/*
947 	 * Check to see if this is a spare device.  We do an explicit check for
948 	 * spa_has_spare() here because it may be on our pending list of spares
949 	 * to add.
950 	 */
951 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
952 	    spa_has_spare(spa, device_guid)) {
953 		if (spare_guid)
954 			*spare_guid = device_guid;
955 
956 		switch (reason) {
957 		case VDEV_LABEL_CREATE:
958 			return (B_TRUE);
959 
960 		case VDEV_LABEL_REPLACE:
961 			return (!spa_has_spare(spa, device_guid) ||
962 			    spare_pool != 0ULL);
963 
964 		case VDEV_LABEL_SPARE:
965 			return (spa_has_spare(spa, device_guid));
966 		default:
967 			break;
968 		}
969 	}
970 
971 	/*
972 	 * Check to see if this is an l2cache device.
973 	 */
974 	if (spa_l2cache_exists(device_guid, NULL) ||
975 	    spa_has_l2cache(spa, device_guid)) {
976 		if (l2cache_guid)
977 			*l2cache_guid = device_guid;
978 
979 		switch (reason) {
980 		case VDEV_LABEL_CREATE:
981 			return (B_TRUE);
982 
983 		case VDEV_LABEL_REPLACE:
984 			return (!spa_has_l2cache(spa, device_guid));
985 
986 		case VDEV_LABEL_L2CACHE:
987 			return (spa_has_l2cache(spa, device_guid));
988 		default:
989 			break;
990 		}
991 	}
992 
993 	/*
994 	 * We can't rely on a pool's state if it's been imported
995 	 * read-only.  Instead we look to see if the pools is marked
996 	 * read-only in the namespace and set the state to active.
997 	 */
998 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
999 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
1000 	    spa_mode(spa) == SPA_MODE_READ)
1001 		state = POOL_STATE_ACTIVE;
1002 
1003 	/*
1004 	 * If the device is marked ACTIVE, then this device is in use by another
1005 	 * pool on the system.
1006 	 */
1007 	return (state == POOL_STATE_ACTIVE);
1008 }
1009 
1010 /*
1011  * Initialize a vdev label.  We check to make sure each leaf device is not in
1012  * use, and writable.  We put down an initial label which we will later
1013  * overwrite with a complete label.  Note that it's important to do this
1014  * sequentially, not in parallel, so that we catch cases of multiple use of the
1015  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1016  * itself.
1017  */
1018 int
1019 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1020 {
1021 	spa_t *spa = vd->vdev_spa;
1022 	nvlist_t *label;
1023 	vdev_phys_t *vp;
1024 	abd_t *vp_abd;
1025 	abd_t *bootenv;
1026 	uberblock_t *ub;
1027 	abd_t *ub_abd;
1028 	zio_t *zio;
1029 	char *buf;
1030 	size_t buflen;
1031 	int error;
1032 	uint64_t spare_guid = 0, l2cache_guid = 0;
1033 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1034 
1035 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1036 
1037 	for (int c = 0; c < vd->vdev_children; c++)
1038 		if ((error = vdev_label_init(vd->vdev_child[c],
1039 		    crtxg, reason)) != 0)
1040 			return (error);
1041 
1042 	/* Track the creation time for this vdev */
1043 	vd->vdev_crtxg = crtxg;
1044 
1045 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1046 		return (0);
1047 
1048 	/*
1049 	 * Dead vdevs cannot be initialized.
1050 	 */
1051 	if (vdev_is_dead(vd))
1052 		return (SET_ERROR(EIO));
1053 
1054 	/*
1055 	 * Determine if the vdev is in use.
1056 	 */
1057 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1058 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1059 		return (SET_ERROR(EBUSY));
1060 
1061 	/*
1062 	 * If this is a request to add or replace a spare or l2cache device
1063 	 * that is in use elsewhere on the system, then we must update the
1064 	 * guid (which was initialized to a random value) to reflect the
1065 	 * actual GUID (which is shared between multiple pools).
1066 	 */
1067 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1068 	    spare_guid != 0ULL) {
1069 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1070 
1071 		vd->vdev_guid += guid_delta;
1072 
1073 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1074 			pvd->vdev_guid_sum += guid_delta;
1075 
1076 		/*
1077 		 * If this is a replacement, then we want to fallthrough to the
1078 		 * rest of the code.  If we're adding a spare, then it's already
1079 		 * labeled appropriately and we can just return.
1080 		 */
1081 		if (reason == VDEV_LABEL_SPARE)
1082 			return (0);
1083 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1084 		    reason == VDEV_LABEL_SPLIT);
1085 	}
1086 
1087 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1088 	    l2cache_guid != 0ULL) {
1089 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1090 
1091 		vd->vdev_guid += guid_delta;
1092 
1093 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1094 			pvd->vdev_guid_sum += guid_delta;
1095 
1096 		/*
1097 		 * If this is a replacement, then we want to fallthrough to the
1098 		 * rest of the code.  If we're adding an l2cache, then it's
1099 		 * already labeled appropriately and we can just return.
1100 		 */
1101 		if (reason == VDEV_LABEL_L2CACHE)
1102 			return (0);
1103 		ASSERT(reason == VDEV_LABEL_REPLACE);
1104 	}
1105 
1106 	/*
1107 	 * Initialize its label.
1108 	 */
1109 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1110 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1111 	vp = abd_to_buf(vp_abd);
1112 
1113 	/*
1114 	 * Generate a label describing the pool and our top-level vdev.
1115 	 * We mark it as being from txg 0 to indicate that it's not
1116 	 * really part of an active pool just yet.  The labels will
1117 	 * be written again with a meaningful txg by spa_sync().
1118 	 */
1119 	if (reason == VDEV_LABEL_SPARE ||
1120 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1121 		/*
1122 		 * For inactive hot spares, we generate a special label that
1123 		 * identifies as a mutually shared hot spare.  We write the
1124 		 * label if we are adding a hot spare, or if we are removing an
1125 		 * active hot spare (in which case we want to revert the
1126 		 * labels).
1127 		 */
1128 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1129 
1130 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1131 		    spa_version(spa)) == 0);
1132 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1133 		    POOL_STATE_SPARE) == 0);
1134 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1135 		    vd->vdev_guid) == 0);
1136 	} else if (reason == VDEV_LABEL_L2CACHE ||
1137 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1138 		/*
1139 		 * For level 2 ARC devices, add a special label.
1140 		 */
1141 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1142 
1143 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1144 		    spa_version(spa)) == 0);
1145 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1146 		    POOL_STATE_L2CACHE) == 0);
1147 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1148 		    vd->vdev_guid) == 0);
1149 
1150 		/*
1151 		 * This is merely to facilitate reporting the ashift of the
1152 		 * cache device through zdb. The actual retrieval of the
1153 		 * ashift (in vdev_alloc()) uses the nvlist
1154 		 * spa->spa_l2cache->sav_config (populated in
1155 		 * spa_ld_open_aux_vdevs()).
1156 		 */
1157 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_ASHIFT,
1158 		    vd->vdev_ashift) == 0);
1159 	} else {
1160 		uint64_t txg = 0ULL;
1161 
1162 		if (reason == VDEV_LABEL_SPLIT)
1163 			txg = spa->spa_uberblock.ub_txg;
1164 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1165 
1166 		/*
1167 		 * Add our creation time.  This allows us to detect multiple
1168 		 * vdev uses as described above, and automatically expires if we
1169 		 * fail.
1170 		 */
1171 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1172 		    crtxg) == 0);
1173 	}
1174 
1175 	buf = vp->vp_nvlist;
1176 	buflen = sizeof (vp->vp_nvlist);
1177 
1178 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1179 	if (error != 0) {
1180 		nvlist_free(label);
1181 		abd_free(vp_abd);
1182 		/* EFAULT means nvlist_pack ran out of room */
1183 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1184 	}
1185 
1186 	/*
1187 	 * Initialize uberblock template.
1188 	 */
1189 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1190 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1191 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1192 	ub = abd_to_buf(ub_abd);
1193 	ub->ub_txg = 0;
1194 
1195 	/* Initialize the 2nd padding area. */
1196 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1197 	abd_zero(bootenv, VDEV_PAD_SIZE);
1198 
1199 	/*
1200 	 * Write everything in parallel.
1201 	 */
1202 retry:
1203 	zio = zio_root(spa, NULL, NULL, flags);
1204 
1205 	for (int l = 0; l < VDEV_LABELS; l++) {
1206 
1207 		vdev_label_write(zio, vd, l, vp_abd,
1208 		    offsetof(vdev_label_t, vl_vdev_phys),
1209 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1210 
1211 		/*
1212 		 * Skip the 1st padding area.
1213 		 * Zero out the 2nd padding area where it might have
1214 		 * left over data from previous filesystem format.
1215 		 */
1216 		vdev_label_write(zio, vd, l, bootenv,
1217 		    offsetof(vdev_label_t, vl_be),
1218 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1219 
1220 		vdev_label_write(zio, vd, l, ub_abd,
1221 		    offsetof(vdev_label_t, vl_uberblock),
1222 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1223 	}
1224 
1225 	error = zio_wait(zio);
1226 
1227 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1228 		flags |= ZIO_FLAG_TRYHARD;
1229 		goto retry;
1230 	}
1231 
1232 	nvlist_free(label);
1233 	abd_free(bootenv);
1234 	abd_free(ub_abd);
1235 	abd_free(vp_abd);
1236 
1237 	/*
1238 	 * If this vdev hasn't been previously identified as a spare, then we
1239 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1240 	 * exists as a spare elsewhere in the system.  Do the same for
1241 	 * level 2 ARC devices.
1242 	 */
1243 	if (error == 0 && !vd->vdev_isspare &&
1244 	    (reason == VDEV_LABEL_SPARE ||
1245 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1246 		spa_spare_add(vd);
1247 
1248 	if (error == 0 && !vd->vdev_isl2cache &&
1249 	    (reason == VDEV_LABEL_L2CACHE ||
1250 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1251 		spa_l2cache_add(vd);
1252 
1253 	return (error);
1254 }
1255 
1256 /*
1257  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1258  * callback to finish, store our abd in the callback pointer. Otherwise, we
1259  * just free our abd and return.
1260  */
1261 static void
1262 vdev_label_read_bootenv_done(zio_t *zio)
1263 {
1264 	zio_t *rio = zio->io_private;
1265 	abd_t **cbp = rio->io_private;
1266 
1267 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1268 
1269 	if (zio->io_error == 0) {
1270 		mutex_enter(&rio->io_lock);
1271 		if (*cbp == NULL) {
1272 			/* Will free this buffer in vdev_label_read_bootenv. */
1273 			*cbp = zio->io_abd;
1274 		} else {
1275 			abd_free(zio->io_abd);
1276 		}
1277 		mutex_exit(&rio->io_lock);
1278 	} else {
1279 		abd_free(zio->io_abd);
1280 	}
1281 }
1282 
1283 static void
1284 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1285 {
1286 	for (int c = 0; c < vd->vdev_children; c++)
1287 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1288 
1289 	/*
1290 	 * We just use the first label that has a correct checksum; the
1291 	 * bootloader should have rewritten them all to be the same on boot,
1292 	 * and any changes we made since boot have been the same across all
1293 	 * labels.
1294 	 */
1295 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1296 		for (int l = 0; l < VDEV_LABELS; l++) {
1297 			vdev_label_read(zio, vd, l,
1298 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1299 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1300 			    vdev_label_read_bootenv_done, zio, flags);
1301 		}
1302 	}
1303 }
1304 
1305 int
1306 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1307 {
1308 	nvlist_t *config;
1309 	spa_t *spa = rvd->vdev_spa;
1310 	abd_t *abd = NULL;
1311 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1312 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1313 
1314 	ASSERT(bootenv);
1315 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1316 
1317 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1318 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1319 	int err = zio_wait(zio);
1320 
1321 	if (abd != NULL) {
1322 		char *buf;
1323 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1324 
1325 		vbe->vbe_version = ntohll(vbe->vbe_version);
1326 		switch (vbe->vbe_version) {
1327 		case VB_RAW:
1328 			/*
1329 			 * if we have textual data in vbe_bootenv, create nvlist
1330 			 * with key "envmap".
1331 			 */
1332 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1333 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1334 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1335 			    vbe->vbe_bootenv);
1336 			break;
1337 
1338 		case VB_NVLIST:
1339 			err = nvlist_unpack(vbe->vbe_bootenv,
1340 			    sizeof (vbe->vbe_bootenv), &config, 0);
1341 			if (err == 0) {
1342 				fnvlist_merge(bootenv, config);
1343 				nvlist_free(config);
1344 				break;
1345 			}
1346 			zfs_fallthrough;
1347 		default:
1348 			/* Check for FreeBSD zfs bootonce command string */
1349 			buf = abd_to_buf(abd);
1350 			if (*buf == '\0') {
1351 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1352 				    VB_NVLIST);
1353 				break;
1354 			}
1355 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1356 		}
1357 
1358 		/*
1359 		 * abd was allocated in vdev_label_read_bootenv_impl()
1360 		 */
1361 		abd_free(abd);
1362 		/*
1363 		 * If we managed to read any successfully,
1364 		 * return success.
1365 		 */
1366 		return (0);
1367 	}
1368 	return (err);
1369 }
1370 
1371 int
1372 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1373 {
1374 	zio_t *zio;
1375 	spa_t *spa = vd->vdev_spa;
1376 	vdev_boot_envblock_t *bootenv;
1377 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1378 	int error;
1379 	size_t nvsize;
1380 	char *nvbuf;
1381 	const char *tmp;
1382 
1383 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1384 	if (error != 0)
1385 		return (SET_ERROR(error));
1386 
1387 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1388 		return (SET_ERROR(E2BIG));
1389 	}
1390 
1391 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1392 
1393 	error = ENXIO;
1394 	for (int c = 0; c < vd->vdev_children; c++) {
1395 		int child_err;
1396 
1397 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1398 		/*
1399 		 * As long as any of the disks managed to write all of their
1400 		 * labels successfully, return success.
1401 		 */
1402 		if (child_err == 0)
1403 			error = child_err;
1404 	}
1405 
1406 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1407 	    !vdev_writeable(vd)) {
1408 		return (error);
1409 	}
1410 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1411 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1412 	abd_zero(abd, VDEV_PAD_SIZE);
1413 
1414 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1415 	nvbuf = bootenv->vbe_bootenv;
1416 	nvsize = sizeof (bootenv->vbe_bootenv);
1417 
1418 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1419 	switch (bootenv->vbe_version) {
1420 	case VB_RAW:
1421 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1422 			(void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1423 		}
1424 		error = 0;
1425 		break;
1426 
1427 	case VB_NVLIST:
1428 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1429 		    KM_SLEEP);
1430 		break;
1431 
1432 	default:
1433 		error = EINVAL;
1434 		break;
1435 	}
1436 
1437 	if (error == 0) {
1438 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1439 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1440 	} else {
1441 		abd_free(abd);
1442 		return (SET_ERROR(error));
1443 	}
1444 
1445 retry:
1446 	zio = zio_root(spa, NULL, NULL, flags);
1447 	for (int l = 0; l < VDEV_LABELS; l++) {
1448 		vdev_label_write(zio, vd, l, abd,
1449 		    offsetof(vdev_label_t, vl_be),
1450 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1451 	}
1452 
1453 	error = zio_wait(zio);
1454 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1455 		flags |= ZIO_FLAG_TRYHARD;
1456 		goto retry;
1457 	}
1458 
1459 	abd_free(abd);
1460 	return (error);
1461 }
1462 
1463 /*
1464  * ==========================================================================
1465  * uberblock load/sync
1466  * ==========================================================================
1467  */
1468 
1469 /*
1470  * Consider the following situation: txg is safely synced to disk.  We've
1471  * written the first uberblock for txg + 1, and then we lose power.  When we
1472  * come back up, we fail to see the uberblock for txg + 1 because, say,
1473  * it was on a mirrored device and the replica to which we wrote txg + 1
1474  * is now offline.  If we then make some changes and sync txg + 1, and then
1475  * the missing replica comes back, then for a few seconds we'll have two
1476  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1477  * among uberblocks with equal txg, choose the one with the latest timestamp.
1478  */
1479 static int
1480 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1481 {
1482 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1483 
1484 	if (likely(cmp))
1485 		return (cmp);
1486 
1487 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1488 	if (likely(cmp))
1489 		return (cmp);
1490 
1491 	/*
1492 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1493 	 * ZFS, e.g. OpenZFS >= 0.7.
1494 	 *
1495 	 * If one ub has MMP and the other does not, they were written by
1496 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1497 	 * a 0 value.
1498 	 *
1499 	 * Since timestamp and txg are the same if we get this far, either is
1500 	 * acceptable for importing the pool.
1501 	 */
1502 	unsigned int seq1 = 0;
1503 	unsigned int seq2 = 0;
1504 
1505 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1506 		seq1 = MMP_SEQ(ub1);
1507 
1508 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1509 		seq2 = MMP_SEQ(ub2);
1510 
1511 	return (TREE_CMP(seq1, seq2));
1512 }
1513 
1514 struct ubl_cbdata {
1515 	uberblock_t	ubl_latest;	/* Most recent uberblock */
1516 	uberblock_t	*ubl_ubbest;	/* Best uberblock (w/r/t max_txg) */
1517 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1518 };
1519 
1520 static void
1521 vdev_uberblock_load_done(zio_t *zio)
1522 {
1523 	vdev_t *vd = zio->io_vd;
1524 	spa_t *spa = zio->io_spa;
1525 	zio_t *rio = zio->io_private;
1526 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1527 	struct ubl_cbdata *cbp = rio->io_private;
1528 
1529 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1530 
1531 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1532 		mutex_enter(&rio->io_lock);
1533 		if (vdev_uberblock_compare(ub, &cbp->ubl_latest) > 0) {
1534 			cbp->ubl_latest = *ub;
1535 		}
1536 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1537 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1538 			/*
1539 			 * Keep track of the vdev in which this uberblock
1540 			 * was found. We will use this information later
1541 			 * to obtain the config nvlist associated with
1542 			 * this uberblock.
1543 			 */
1544 			*cbp->ubl_ubbest = *ub;
1545 			cbp->ubl_vd = vd;
1546 		}
1547 		mutex_exit(&rio->io_lock);
1548 	}
1549 
1550 	abd_free(zio->io_abd);
1551 }
1552 
1553 static void
1554 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1555     struct ubl_cbdata *cbp)
1556 {
1557 	for (int c = 0; c < vd->vdev_children; c++)
1558 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1559 
1560 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1561 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1562 		for (int l = 0; l < VDEV_LABELS; l++) {
1563 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1564 				vdev_label_read(zio, vd, l,
1565 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1566 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1567 				    VDEV_UBERBLOCK_SIZE(vd),
1568 				    vdev_uberblock_load_done, zio, flags);
1569 			}
1570 		}
1571 	}
1572 }
1573 
1574 /*
1575  * Reads the 'best' uberblock from disk along with its associated
1576  * configuration. First, we read the uberblock array of each label of each
1577  * vdev, keeping track of the uberblock with the highest txg in each array.
1578  * Then, we read the configuration from the same vdev as the best uberblock.
1579  */
1580 void
1581 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1582 {
1583 	zio_t *zio;
1584 	spa_t *spa = rvd->vdev_spa;
1585 	struct ubl_cbdata cb;
1586 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1587 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1588 
1589 	ASSERT(ub);
1590 	ASSERT(config);
1591 
1592 	memset(ub, 0, sizeof (uberblock_t));
1593 	memset(&cb, 0, sizeof (cb));
1594 	*config = NULL;
1595 
1596 	cb.ubl_ubbest = ub;
1597 
1598 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1599 	zio = zio_root(spa, NULL, &cb, flags);
1600 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1601 	(void) zio_wait(zio);
1602 
1603 	/*
1604 	 * It's possible that the best uberblock was discovered on a label
1605 	 * that has a configuration which was written in a future txg.
1606 	 * Search all labels on this vdev to find the configuration that
1607 	 * matches the txg for our uberblock.
1608 	 */
1609 	if (cb.ubl_vd != NULL) {
1610 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1611 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1612 
1613 		if (ub->ub_raidz_reflow_info !=
1614 		    cb.ubl_latest.ub_raidz_reflow_info) {
1615 			vdev_dbgmsg(cb.ubl_vd,
1616 			    "spa=%s best uberblock (txg=%llu info=0x%llx) "
1617 			    "has different raidz_reflow_info than latest "
1618 			    "uberblock (txg=%llu info=0x%llx)",
1619 			    spa->spa_name,
1620 			    (u_longlong_t)ub->ub_txg,
1621 			    (u_longlong_t)ub->ub_raidz_reflow_info,
1622 			    (u_longlong_t)cb.ubl_latest.ub_txg,
1623 			    (u_longlong_t)cb.ubl_latest.ub_raidz_reflow_info);
1624 			memset(ub, 0, sizeof (uberblock_t));
1625 			spa_config_exit(spa, SCL_ALL, FTAG);
1626 			return;
1627 		}
1628 
1629 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1630 		if (*config == NULL && spa->spa_extreme_rewind) {
1631 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1632 			    "Trying again without txg restrictions.");
1633 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1634 		}
1635 		if (*config == NULL) {
1636 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1637 		}
1638 	}
1639 	spa_config_exit(spa, SCL_ALL, FTAG);
1640 }
1641 
1642 /*
1643  * For use when a leaf vdev is expanded.
1644  * The location of labels 2 and 3 changed, and at the new location the
1645  * uberblock rings are either empty or contain garbage.  The sync will write
1646  * new configs there because the vdev is dirty, but expansion also needs the
1647  * uberblock rings copied.  Read them from label 0 which did not move.
1648  *
1649  * Since the point is to populate labels {2,3} with valid uberblocks,
1650  * we zero uberblocks we fail to read or which are not valid.
1651  */
1652 
1653 static void
1654 vdev_copy_uberblocks(vdev_t *vd)
1655 {
1656 	abd_t *ub_abd;
1657 	zio_t *write_zio;
1658 	int locks = (SCL_L2ARC | SCL_ZIO);
1659 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1660 	    ZIO_FLAG_SPECULATIVE;
1661 
1662 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1663 	    SCL_STATE);
1664 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1665 
1666 	/*
1667 	 * No uberblocks are stored on distributed spares, they may be
1668 	 * safely skipped when expanding a leaf vdev.
1669 	 */
1670 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1671 		return;
1672 
1673 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1674 
1675 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1676 
1677 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1678 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1679 		const int src_label = 0;
1680 		zio_t *zio;
1681 
1682 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1683 		vdev_label_read(zio, vd, src_label, ub_abd,
1684 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1685 		    NULL, NULL, flags);
1686 
1687 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1688 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1689 
1690 		for (int l = 2; l < VDEV_LABELS; l++)
1691 			vdev_label_write(write_zio, vd, l, ub_abd,
1692 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1693 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1694 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1695 	}
1696 	(void) zio_wait(write_zio);
1697 
1698 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1699 
1700 	abd_free(ub_abd);
1701 }
1702 
1703 /*
1704  * On success, increment root zio's count of good writes.
1705  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1706  */
1707 static void
1708 vdev_uberblock_sync_done(zio_t *zio)
1709 {
1710 	uint64_t *good_writes = zio->io_private;
1711 
1712 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1713 		atomic_inc_64(good_writes);
1714 }
1715 
1716 /*
1717  * Write the uberblock to all labels of all leaves of the specified vdev.
1718  */
1719 static void
1720 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1721     uberblock_t *ub, vdev_t *vd, int flags)
1722 {
1723 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1724 		vdev_uberblock_sync(zio, good_writes,
1725 		    ub, vd->vdev_child[c], flags);
1726 	}
1727 
1728 	if (!vd->vdev_ops->vdev_op_leaf)
1729 		return;
1730 
1731 	if (!vdev_writeable(vd))
1732 		return;
1733 
1734 	/*
1735 	 * There's no need to write uberblocks to a distributed spare, they
1736 	 * are already stored on all the leaves of the parent dRAID.  For
1737 	 * this same reason vdev_uberblock_load_impl() skips distributed
1738 	 * spares when reading uberblocks.
1739 	 */
1740 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1741 		return;
1742 
1743 	/* If the vdev was expanded, need to copy uberblock rings. */
1744 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1745 	    vd->vdev_copy_uberblocks == B_TRUE) {
1746 		vdev_copy_uberblocks(vd);
1747 		vd->vdev_copy_uberblocks = B_FALSE;
1748 	}
1749 
1750 	/*
1751 	 * We chose a slot based on the txg.  If this uberblock has a special
1752 	 * RAIDZ expansion state, then it is essentially an update of the
1753 	 * current uberblock (it has the same txg).  However, the current
1754 	 * state is committed, so we want to write it to a different slot. If
1755 	 * we overwrote the same slot, and we lose power during the uberblock
1756 	 * write, and the disk does not do single-sector overwrites
1757 	 * atomically (even though it is required to - i.e. we should see
1758 	 * either the old or the new uberblock), then we could lose this
1759 	 * txg's uberblock. Rewinding to the previous txg's uberblock may not
1760 	 * be possible because RAIDZ expansion may have already overwritten
1761 	 * some of the data, so we need the progress indicator in the
1762 	 * uberblock.
1763 	 */
1764 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1765 	int n = (ub->ub_txg - (RRSS_GET_STATE(ub) == RRSS_SCRATCH_VALID)) %
1766 	    (VDEV_UBERBLOCK_COUNT(vd) - m);
1767 
1768 	/* Copy the uberblock_t into the ABD */
1769 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1770 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1771 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1772 
1773 	for (int l = 0; l < VDEV_LABELS; l++)
1774 		vdev_label_write(zio, vd, l, ub_abd,
1775 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1776 		    vdev_uberblock_sync_done, good_writes,
1777 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1778 
1779 	abd_free(ub_abd);
1780 }
1781 
1782 /* Sync the uberblocks to all vdevs in svd[] */
1783 int
1784 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1785 {
1786 	spa_t *spa = svd[0]->vdev_spa;
1787 	zio_t *zio;
1788 	uint64_t good_writes = 0;
1789 
1790 	zio = zio_root(spa, NULL, NULL, flags);
1791 
1792 	for (int v = 0; v < svdcount; v++)
1793 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1794 
1795 	(void) zio_wait(zio);
1796 
1797 	/*
1798 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1799 	 * are no longer needed (because the new uberblocks and the even
1800 	 * labels are safely on disk), so it is safe to overwrite them.
1801 	 */
1802 	zio = zio_root(spa, NULL, NULL, flags);
1803 
1804 	for (int v = 0; v < svdcount; v++) {
1805 		if (vdev_writeable(svd[v])) {
1806 			zio_flush(zio, svd[v]);
1807 		}
1808 	}
1809 
1810 	(void) zio_wait(zio);
1811 
1812 	return (good_writes >= 1 ? 0 : EIO);
1813 }
1814 
1815 /*
1816  * On success, increment the count of good writes for our top-level vdev.
1817  */
1818 static void
1819 vdev_label_sync_done(zio_t *zio)
1820 {
1821 	uint64_t *good_writes = zio->io_private;
1822 
1823 	if (zio->io_error == 0)
1824 		atomic_inc_64(good_writes);
1825 }
1826 
1827 /*
1828  * If there weren't enough good writes, indicate failure to the parent.
1829  */
1830 static void
1831 vdev_label_sync_top_done(zio_t *zio)
1832 {
1833 	uint64_t *good_writes = zio->io_private;
1834 
1835 	if (*good_writes == 0)
1836 		zio->io_error = SET_ERROR(EIO);
1837 
1838 	kmem_free(good_writes, sizeof (uint64_t));
1839 }
1840 
1841 /*
1842  * We ignore errors for log and cache devices, simply free the private data.
1843  */
1844 static void
1845 vdev_label_sync_ignore_done(zio_t *zio)
1846 {
1847 	kmem_free(zio->io_private, sizeof (uint64_t));
1848 }
1849 
1850 /*
1851  * Write all even or odd labels to all leaves of the specified vdev.
1852  */
1853 static void
1854 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1855     vdev_t *vd, int l, uint64_t txg, int flags)
1856 {
1857 	nvlist_t *label;
1858 	vdev_phys_t *vp;
1859 	abd_t *vp_abd;
1860 	char *buf;
1861 	size_t buflen;
1862 
1863 	for (int c = 0; c < vd->vdev_children; c++) {
1864 		vdev_label_sync(zio, good_writes,
1865 		    vd->vdev_child[c], l, txg, flags);
1866 	}
1867 
1868 	if (!vd->vdev_ops->vdev_op_leaf)
1869 		return;
1870 
1871 	if (!vdev_writeable(vd))
1872 		return;
1873 
1874 	/*
1875 	 * The top-level config never needs to be written to a distributed
1876 	 * spare.  When read vdev_dspare_label_read_config() will generate
1877 	 * the config for the vdev_label_read_config().
1878 	 */
1879 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1880 		return;
1881 
1882 	/*
1883 	 * Generate a label describing the top-level config to which we belong.
1884 	 */
1885 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1886 
1887 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1888 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1889 	vp = abd_to_buf(vp_abd);
1890 
1891 	buf = vp->vp_nvlist;
1892 	buflen = sizeof (vp->vp_nvlist);
1893 
1894 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1895 		for (; l < VDEV_LABELS; l += 2) {
1896 			vdev_label_write(zio, vd, l, vp_abd,
1897 			    offsetof(vdev_label_t, vl_vdev_phys),
1898 			    sizeof (vdev_phys_t),
1899 			    vdev_label_sync_done, good_writes,
1900 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1901 		}
1902 	}
1903 
1904 	abd_free(vp_abd);
1905 	nvlist_free(label);
1906 }
1907 
1908 static int
1909 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1910 {
1911 	list_t *dl = &spa->spa_config_dirty_list;
1912 	vdev_t *vd;
1913 	zio_t *zio;
1914 	int error;
1915 
1916 	/*
1917 	 * Write the new labels to disk.
1918 	 */
1919 	zio = zio_root(spa, NULL, NULL, flags);
1920 
1921 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1922 		uint64_t *good_writes;
1923 
1924 		ASSERT(!vd->vdev_ishole);
1925 
1926 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1927 		zio_t *vio = zio_null(zio, spa, NULL,
1928 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1929 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1930 		    good_writes, flags);
1931 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1932 		zio_nowait(vio);
1933 	}
1934 
1935 	error = zio_wait(zio);
1936 
1937 	/*
1938 	 * Flush the new labels to disk.
1939 	 */
1940 	zio = zio_root(spa, NULL, NULL, flags);
1941 
1942 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1943 		zio_flush(zio, vd);
1944 
1945 	(void) zio_wait(zio);
1946 
1947 	return (error);
1948 }
1949 
1950 /*
1951  * Sync the uberblock and any changes to the vdev configuration.
1952  *
1953  * The order of operations is carefully crafted to ensure that
1954  * if the system panics or loses power at any time, the state on disk
1955  * is still transactionally consistent.  The in-line comments below
1956  * describe the failure semantics at each stage.
1957  *
1958  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1959  * at any time, you can just call it again, and it will resume its work.
1960  */
1961 int
1962 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1963 {
1964 	spa_t *spa = svd[0]->vdev_spa;
1965 	uberblock_t *ub = &spa->spa_uberblock;
1966 	int error = 0;
1967 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1968 
1969 	ASSERT(svdcount != 0);
1970 retry:
1971 	/*
1972 	 * Normally, we don't want to try too hard to write every label and
1973 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1974 	 * sync process to block while we retry.  But if we can't write a
1975 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1976 	 * bailing out and declaring the pool faulted.
1977 	 */
1978 	if (error != 0) {
1979 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1980 			return (error);
1981 		flags |= ZIO_FLAG_TRYHARD;
1982 	}
1983 
1984 	ASSERT(ub->ub_txg <= txg);
1985 
1986 	/*
1987 	 * If this isn't a resync due to I/O errors,
1988 	 * and nothing changed in this transaction group,
1989 	 * and the vdev configuration hasn't changed,
1990 	 * then there's nothing to do.
1991 	 */
1992 	if (ub->ub_txg < txg) {
1993 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1994 		    txg, spa->spa_mmp.mmp_delay);
1995 
1996 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1997 			return (0);
1998 	}
1999 
2000 	if (txg > spa_freeze_txg(spa))
2001 		return (0);
2002 
2003 	ASSERT(txg <= spa->spa_final_txg);
2004 
2005 	/*
2006 	 * Flush the write cache of every disk that's been written to
2007 	 * in this transaction group.  This ensures that all blocks
2008 	 * written in this txg will be committed to stable storage
2009 	 * before any uberblock that references them.
2010 	 */
2011 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
2012 
2013 	for (vdev_t *vd =
2014 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
2015 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
2016 		zio_flush(zio, vd);
2017 
2018 	(void) zio_wait(zio);
2019 
2020 	/*
2021 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
2022 	 * system dies in the middle of this process, that's OK: all of the
2023 	 * even labels that made it to disk will be newer than any uberblock,
2024 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
2025 	 * which have not yet been touched, will still be valid.  We flush
2026 	 * the new labels to disk to ensure that all even-label updates
2027 	 * are committed to stable storage before the uberblock update.
2028 	 */
2029 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
2030 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2031 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2032 			    "for pool '%s' when syncing out the even labels "
2033 			    "of dirty vdevs", error, spa_name(spa));
2034 		}
2035 		goto retry;
2036 	}
2037 
2038 	/*
2039 	 * Sync the uberblocks to all vdevs in svd[].
2040 	 * If the system dies in the middle of this step, there are two cases
2041 	 * to consider, and the on-disk state is consistent either way:
2042 	 *
2043 	 * (1)	If none of the new uberblocks made it to disk, then the
2044 	 *	previous uberblock will be the newest, and the odd labels
2045 	 *	(which had not yet been touched) will be valid with respect
2046 	 *	to that uberblock.
2047 	 *
2048 	 * (2)	If one or more new uberblocks made it to disk, then they
2049 	 *	will be the newest, and the even labels (which had all
2050 	 *	been successfully committed) will be valid with respect
2051 	 *	to the new uberblocks.
2052 	 */
2053 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
2054 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2055 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2056 			    "%d for pool '%s'", error, spa_name(spa));
2057 		}
2058 		goto retry;
2059 	}
2060 
2061 	if (spa_multihost(spa))
2062 		mmp_update_uberblock(spa, ub);
2063 
2064 	/*
2065 	 * Sync out odd labels for every dirty vdev.  If the system dies
2066 	 * in the middle of this process, the even labels and the new
2067 	 * uberblocks will suffice to open the pool.  The next time
2068 	 * the pool is opened, the first thing we'll do -- before any
2069 	 * user data is modified -- is mark every vdev dirty so that
2070 	 * all labels will be brought up to date.  We flush the new labels
2071 	 * to disk to ensure that all odd-label updates are committed to
2072 	 * stable storage before the next transaction group begins.
2073 	 */
2074 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2075 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2076 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2077 			    "for pool '%s' when syncing out the odd labels of "
2078 			    "dirty vdevs", error, spa_name(spa));
2079 		}
2080 		goto retry;
2081 	}
2082 
2083 	return (0);
2084 }
2085