xref: /freebsd/sys/contrib/openzfs/module/zfs/vdev_label.c (revision d9497217456002b0ddad3cd319570d0b098daa29)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
26  * Copyright (c) 2017, Intel Corporation.
27  */
28 
29 /*
30  * Virtual Device Labels
31  * ---------------------
32  *
33  * The vdev label serves several distinct purposes:
34  *
35  *	1. Uniquely identify this device as part of a ZFS pool and confirm its
36  *	   identity within the pool.
37  *
38  *	2. Verify that all the devices given in a configuration are present
39  *         within the pool.
40  *
41  *	3. Determine the uberblock for the pool.
42  *
43  *	4. In case of an import operation, determine the configuration of the
44  *         toplevel vdev of which it is a part.
45  *
46  *	5. If an import operation cannot find all the devices in the pool,
47  *         provide enough information to the administrator to determine which
48  *         devices are missing.
49  *
50  * It is important to note that while the kernel is responsible for writing the
51  * label, it only consumes the information in the first three cases.  The
52  * latter information is only consumed in userland when determining the
53  * configuration to import a pool.
54  *
55  *
56  * Label Organization
57  * ------------------
58  *
59  * Before describing the contents of the label, it's important to understand how
60  * the labels are written and updated with respect to the uberblock.
61  *
62  * When the pool configuration is altered, either because it was newly created
63  * or a device was added, we want to update all the labels such that we can deal
64  * with fatal failure at any point.  To this end, each disk has two labels which
65  * are updated before and after the uberblock is synced.  Assuming we have
66  * labels and an uberblock with the following transaction groups:
67  *
68  *              L1          UB          L2
69  *           +------+    +------+    +------+
70  *           |      |    |      |    |      |
71  *           | t10  |    | t10  |    | t10  |
72  *           |      |    |      |    |      |
73  *           +------+    +------+    +------+
74  *
75  * In this stable state, the labels and the uberblock were all updated within
76  * the same transaction group (10).  Each label is mirrored and checksummed, so
77  * that we can detect when we fail partway through writing the label.
78  *
79  * In order to identify which labels are valid, the labels are written in the
80  * following manner:
81  *
82  *	1. For each vdev, update 'L1' to the new label
83  *	2. Update the uberblock
84  *	3. For each vdev, update 'L2' to the new label
85  *
86  * Given arbitrary failure, we can determine the correct label to use based on
87  * the transaction group.  If we fail after updating L1 but before updating the
88  * UB, we will notice that L1's transaction group is greater than the uberblock,
89  * so L2 must be valid.  If we fail after writing the uberblock but before
90  * writing L2, we will notice that L2's transaction group is less than L1, and
91  * therefore L1 is valid.
92  *
93  * Another added complexity is that not every label is updated when the config
94  * is synced.  If we add a single device, we do not want to have to re-write
95  * every label for every device in the pool.  This means that both L1 and L2 may
96  * be older than the pool uberblock, because the necessary information is stored
97  * on another vdev.
98  *
99  *
100  * On-disk Format
101  * --------------
102  *
103  * The vdev label consists of two distinct parts, and is wrapped within the
104  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
105  * VTOC disk labels, but is otherwise ignored.
106  *
107  * The first half of the label is a packed nvlist which contains pool wide
108  * properties, per-vdev properties, and configuration information.  It is
109  * described in more detail below.
110  *
111  * The latter half of the label consists of a redundant array of uberblocks.
112  * These uberblocks are updated whenever a transaction group is committed,
113  * or when the configuration is updated.  When a pool is loaded, we scan each
114  * vdev for the 'best' uberblock.
115  *
116  *
117  * Configuration Information
118  * -------------------------
119  *
120  * The nvlist describing the pool and vdev contains the following elements:
121  *
122  *	version		ZFS on-disk version
123  *	name		Pool name
124  *	state		Pool state
125  *	txg		Transaction group in which this label was written
126  *	pool_guid	Unique identifier for this pool
127  *	vdev_tree	An nvlist describing vdev tree.
128  *	features_for_read
129  *			An nvlist of the features necessary for reading the MOS.
130  *
131  * Each leaf device label also contains the following:
132  *
133  *	top_guid	Unique ID for top-level vdev in which this is contained
134  *	guid		Unique ID for the leaf vdev
135  *
136  * The 'vs' configuration follows the format described in 'spa_config.c'.
137  */
138 
139 #include <sys/zfs_context.h>
140 #include <sys/spa.h>
141 #include <sys/spa_impl.h>
142 #include <sys/dmu.h>
143 #include <sys/zap.h>
144 #include <sys/vdev.h>
145 #include <sys/vdev_impl.h>
146 #include <sys/vdev_raidz.h>
147 #include <sys/vdev_draid.h>
148 #include <sys/uberblock_impl.h>
149 #include <sys/metaslab.h>
150 #include <sys/metaslab_impl.h>
151 #include <sys/zio.h>
152 #include <sys/dsl_scan.h>
153 #include <sys/abd.h>
154 #include <sys/fs/zfs.h>
155 #include <sys/byteorder.h>
156 #include <sys/zfs_bootenv.h>
157 
158 /*
159  * Basic routines to read and write from a vdev label.
160  * Used throughout the rest of this file.
161  */
162 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)163 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
164 {
165 	ASSERT(offset < sizeof (vdev_label_t));
166 	ASSERT0(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t));
167 
168 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
169 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
170 }
171 
172 /*
173  * Returns back the vdev label associated with the passed in offset.
174  */
175 int
vdev_label_number(uint64_t psize,uint64_t offset)176 vdev_label_number(uint64_t psize, uint64_t offset)
177 {
178 	int l;
179 
180 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
181 		offset -= psize - VDEV_LABEL_END_SIZE;
182 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
183 	}
184 	l = offset / sizeof (vdev_label_t);
185 	return (l < VDEV_LABELS ? l : -1);
186 }
187 
188 static void
vdev_label_read(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)189 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
190     uint64_t size, zio_done_func_t *done, void *private, int flags)
191 {
192 	ASSERT(
193 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
194 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
195 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
196 
197 	zio_nowait(zio_read_phys(zio, vd,
198 	    vdev_label_offset(vd->vdev_psize, l, offset),
199 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
200 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
201 }
202 
203 void
vdev_label_write(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)204 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
205     uint64_t size, zio_done_func_t *done, void *private, int flags)
206 {
207 	ASSERT(
208 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
209 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
210 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
211 
212 	zio_nowait(zio_write_phys(zio, vd,
213 	    vdev_label_offset(vd->vdev_psize, l, offset),
214 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
215 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
216 }
217 
218 /*
219  * Generate the nvlist representing this vdev's stats
220  */
221 void
vdev_config_generate_stats(vdev_t * vd,nvlist_t * nv)222 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
223 {
224 	nvlist_t *nvx;
225 	vdev_stat_t *vs;
226 	vdev_stat_ex_t *vsx;
227 
228 	vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
229 	vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
230 
231 	vdev_get_stats_ex(vd, vs, vsx);
232 	fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
233 	    (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
234 
235 	/*
236 	 * Add extended stats into a special extended stats nvlist.  This keeps
237 	 * all the extended stats nicely grouped together.  The extended stats
238 	 * nvlist is then added to the main nvlist.
239 	 */
240 	nvx = fnvlist_alloc();
241 
242 	/* ZIOs in flight to disk */
243 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
244 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
245 
246 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
247 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
248 
249 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
250 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
251 
252 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
253 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
254 
255 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
256 	    vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
257 
258 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
259 	    vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
260 
261 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,
262 	    vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]);
263 
264 	/* ZIOs pending */
265 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
266 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
267 
268 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
269 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
270 
271 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
272 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
273 
274 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
275 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
276 
277 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
278 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
279 
280 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
281 	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
282 
283 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,
284 	    vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]);
285 
286 	/* Histograms */
287 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
288 	    vsx->vsx_total_histo[ZIO_TYPE_READ],
289 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
290 
291 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
292 	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
293 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
294 
295 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
296 	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
297 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
298 
299 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
300 	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
301 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
302 
303 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
304 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
305 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
306 
307 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
308 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
309 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
310 
311 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
312 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
313 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
314 
315 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
316 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
317 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
318 
319 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
320 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
321 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
322 
323 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
324 	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
325 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
326 
327 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,
328 	    vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD],
329 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD]));
330 
331 	/* Request sizes */
332 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
333 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
334 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
335 
336 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
337 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
338 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
339 
340 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
341 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
342 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
343 
344 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
345 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
346 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
347 
348 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
349 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
350 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
351 
352 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
353 	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
354 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
355 
356 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,
357 	    vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD],
358 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD]));
359 
360 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
361 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
362 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
363 
364 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
365 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
366 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
367 
368 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
369 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
370 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
371 
372 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
373 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
374 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
375 
376 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
377 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
378 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
379 
380 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
381 	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
382 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
383 
384 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,
385 	    vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD],
386 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD]));
387 
388 	/* IO delays */
389 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
390 
391 	/* Direct I/O write verify errors */
392 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_DIO_VERIFY_ERRORS,
393 	    vs->vs_dio_verify_errors);
394 
395 	/* Add extended stats nvlist to main nvlist */
396 	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
397 
398 	fnvlist_free(nvx);
399 	kmem_free(vs, sizeof (*vs));
400 	kmem_free(vsx, sizeof (*vsx));
401 }
402 
403 static void
root_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)404 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
405 {
406 	spa_t *spa = vd->vdev_spa;
407 
408 	if (vd != spa->spa_root_vdev)
409 		return;
410 
411 	/* provide either current or previous scan information */
412 	pool_scan_stat_t ps;
413 	if (spa_scan_get_stats(spa, &ps) == 0) {
414 		fnvlist_add_uint64_array(nvl,
415 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
416 		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
417 	}
418 
419 	pool_removal_stat_t prs;
420 	if (spa_removal_get_stats(spa, &prs) == 0) {
421 		fnvlist_add_uint64_array(nvl,
422 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
423 		    sizeof (prs) / sizeof (uint64_t));
424 	}
425 
426 	pool_checkpoint_stat_t pcs;
427 	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
428 		fnvlist_add_uint64_array(nvl,
429 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
430 		    sizeof (pcs) / sizeof (uint64_t));
431 	}
432 
433 	pool_raidz_expand_stat_t pres;
434 	if (spa_raidz_expand_get_stats(spa, &pres) == 0) {
435 		fnvlist_add_uint64_array(nvl,
436 		    ZPOOL_CONFIG_RAIDZ_EXPAND_STATS, (uint64_t *)&pres,
437 		    sizeof (pres) / sizeof (uint64_t));
438 	}
439 }
440 
441 static void
top_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)442 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
443 {
444 	if (vd == vd->vdev_top) {
445 		vdev_rebuild_stat_t vrs;
446 		if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
447 			fnvlist_add_uint64_array(nvl,
448 			    ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
449 			    sizeof (vrs) / sizeof (uint64_t));
450 		}
451 	}
452 }
453 
454 /*
455  * Generate the nvlist representing this vdev's config.
456  */
457 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)458 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
459     vdev_config_flag_t flags)
460 {
461 	nvlist_t *nv = NULL;
462 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
463 
464 	nv = fnvlist_alloc();
465 
466 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
467 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
468 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
469 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
470 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
471 	    vd->vdev_top != NULL) {
472 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_TOP_GUID,
473 		    vd->vdev_top->vdev_guid);
474 	}
475 
476 	if (vd->vdev_path != NULL)
477 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
478 
479 	if (vd->vdev_devid != NULL)
480 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
481 
482 	if (vd->vdev_physpath != NULL)
483 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
484 		    vd->vdev_physpath);
485 
486 	if (vd->vdev_enc_sysfs_path != NULL)
487 		fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
488 		    vd->vdev_enc_sysfs_path);
489 
490 	if (vd->vdev_fru != NULL)
491 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
492 
493 	if (vd->vdev_ops->vdev_op_config_generate != NULL)
494 		vd->vdev_ops->vdev_op_config_generate(vd, nv);
495 
496 	if (vd->vdev_wholedisk != -1ULL) {
497 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
498 		    vd->vdev_wholedisk);
499 	}
500 
501 	if (vd->vdev_ops->vdev_op_leaf) {
502 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROTATIONAL,
503 		    !vd->vdev_nonrot);
504 	}
505 
506 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
507 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
508 
509 	if (vd->vdev_isspare)
510 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
511 
512 	if (flags & VDEV_CONFIG_L2CACHE)
513 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
514 
515 	if ((flags & VDEV_CONFIG_SPARE) && vd->vdev_asize != 0)
516 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE, vd->vdev_asize);
517 
518 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
519 	    vd == vd->vdev_top) {
520 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
521 		    vd->vdev_ms_array);
522 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
523 		    vd->vdev_ms_shift);
524 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
525 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
526 		    vd->vdev_asize);
527 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_MIN_ALLOC,
528 		    vdev_get_min_alloc(vd));
529 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
530 		if (vd->vdev_noalloc) {
531 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
532 			    vd->vdev_noalloc);
533 		}
534 
535 		/*
536 		 * Slog devices are removed synchronously so don't
537 		 * persist the vdev_removing flag to the label.
538 		 */
539 		if (vd->vdev_removing && !vd->vdev_islog) {
540 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
541 			    vd->vdev_removing);
542 		}
543 
544 		/* zpool command expects alloc class data */
545 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
546 			const char *bias = NULL;
547 
548 			switch (vd->vdev_alloc_bias) {
549 			case VDEV_BIAS_LOG:
550 				bias = VDEV_ALLOC_BIAS_LOG;
551 				break;
552 			case VDEV_BIAS_SPECIAL:
553 				bias = VDEV_ALLOC_BIAS_SPECIAL;
554 				break;
555 			case VDEV_BIAS_DEDUP:
556 				bias = VDEV_ALLOC_BIAS_DEDUP;
557 				break;
558 			default:
559 				ASSERT3U(vd->vdev_alloc_bias, ==,
560 				    VDEV_BIAS_NONE);
561 			}
562 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
563 			    bias);
564 		}
565 	}
566 
567 	if (vd->vdev_dtl_sm != NULL) {
568 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
569 		    space_map_object(vd->vdev_dtl_sm));
570 	}
571 
572 	if (vic->vic_mapping_object != 0) {
573 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
574 		    vic->vic_mapping_object);
575 	}
576 
577 	if (vic->vic_births_object != 0) {
578 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
579 		    vic->vic_births_object);
580 	}
581 
582 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
583 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
584 		    vic->vic_prev_indirect_vdev);
585 	}
586 
587 	if (vd->vdev_crtxg)
588 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
589 
590 	if (vd->vdev_expansion_time)
591 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
592 		    vd->vdev_expansion_time);
593 
594 	if (flags & VDEV_CONFIG_MOS) {
595 		if (vd->vdev_leaf_zap != 0) {
596 			ASSERT(vd->vdev_ops->vdev_op_leaf);
597 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
598 			    vd->vdev_leaf_zap);
599 		}
600 
601 		if (vd->vdev_top_zap != 0) {
602 			ASSERT(vd == vd->vdev_top);
603 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
604 			    vd->vdev_top_zap);
605 		}
606 
607 		if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
608 		    spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
609 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
610 			    vd->vdev_root_zap);
611 		}
612 
613 		if (vd->vdev_resilver_deferred) {
614 			ASSERT(vd->vdev_ops->vdev_op_leaf);
615 			ASSERT(spa->spa_resilver_deferred);
616 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
617 		}
618 	}
619 
620 	if (getstats) {
621 		vdev_config_generate_stats(vd, nv);
622 
623 		root_vdev_actions_getprogress(vd, nv);
624 		top_vdev_actions_getprogress(vd, nv);
625 
626 		/*
627 		 * Note: this can be called from open context
628 		 * (spa_get_stats()), so we need the rwlock to prevent
629 		 * the mapping from being changed by condensing.
630 		 */
631 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
632 		if (vd->vdev_indirect_mapping != NULL) {
633 			ASSERT(vd->vdev_indirect_births != NULL);
634 			vdev_indirect_mapping_t *vim =
635 			    vd->vdev_indirect_mapping;
636 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
637 			    vdev_indirect_mapping_size(vim));
638 		}
639 		rw_exit(&vd->vdev_indirect_rwlock);
640 		if (vd->vdev_mg != NULL &&
641 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
642 			/*
643 			 * Compute approximately how much memory would be used
644 			 * for the indirect mapping if this device were to
645 			 * be removed.
646 			 *
647 			 * Note: If the frag metric is invalid, then not
648 			 * enough metaslabs have been converted to have
649 			 * histograms.
650 			 */
651 			uint64_t seg_count = 0;
652 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
653 
654 			/*
655 			 * There are the same number of allocated segments
656 			 * as free segments, so we will have at least one
657 			 * entry per free segment.  However, small free
658 			 * segments (smaller than vdev_removal_max_span)
659 			 * will be combined with adjacent allocated segments
660 			 * as a single mapping.
661 			 */
662 			for (int i = 0; i < ZFS_RANGE_TREE_HISTOGRAM_SIZE;
663 			    i++) {
664 				if (i + 1 < highbit64(vdev_removal_max_span)
665 				    - 1) {
666 					to_alloc +=
667 					    vd->vdev_mg->mg_histogram[i] <<
668 					    (i + 1);
669 				} else {
670 					seg_count +=
671 					    vd->vdev_mg->mg_histogram[i];
672 				}
673 			}
674 
675 			/*
676 			 * The maximum length of a mapping is
677 			 * zfs_remove_max_segment, so we need at least one entry
678 			 * per zfs_remove_max_segment of allocated data.
679 			 */
680 			seg_count += to_alloc / spa_remove_max_segment(spa);
681 
682 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
683 			    seg_count *
684 			    sizeof (vdev_indirect_mapping_entry_phys_t));
685 		}
686 	}
687 
688 	if (!vd->vdev_ops->vdev_op_leaf) {
689 		nvlist_t **child;
690 		uint64_t c;
691 
692 		ASSERT(!vd->vdev_ishole);
693 
694 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
695 		    KM_SLEEP);
696 
697 		for (c = 0; c < vd->vdev_children; c++) {
698 			child[c] = vdev_config_generate(spa, vd->vdev_child[c],
699 			    getstats, flags);
700 		}
701 
702 		fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
703 		    (const nvlist_t * const *)child, vd->vdev_children);
704 
705 		for (c = 0; c < vd->vdev_children; c++)
706 			nvlist_free(child[c]);
707 
708 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
709 
710 	} else {
711 		const char *aux = NULL;
712 
713 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
714 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
715 		if (vd->vdev_resilver_txg != 0)
716 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
717 			    vd->vdev_resilver_txg);
718 		if (vd->vdev_rebuild_txg != 0)
719 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
720 			    vd->vdev_rebuild_txg);
721 		if (vd->vdev_faulted)
722 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
723 		if (vd->vdev_degraded)
724 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
725 		if (vd->vdev_removed)
726 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
727 		if (vd->vdev_unspare)
728 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
729 		if (vd->vdev_ishole)
730 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
731 
732 		/* Set the reason why we're FAULTED/DEGRADED. */
733 		switch (vd->vdev_stat.vs_aux) {
734 		case VDEV_AUX_ERR_EXCEEDED:
735 			aux = "err_exceeded";
736 			break;
737 
738 		case VDEV_AUX_EXTERNAL:
739 			aux = "external";
740 			break;
741 		}
742 
743 		if (aux != NULL && !vd->vdev_tmpoffline) {
744 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
745 		} else {
746 			/*
747 			 * We're healthy - clear any previous AUX_STATE values.
748 			 */
749 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
750 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
751 		}
752 
753 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
754 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
755 			    vd->vdev_orig_guid);
756 		}
757 	}
758 
759 	return (nv);
760 }
761 
762 /*
763  * Generate a view of the top-level vdevs.  If we currently have holes
764  * in the namespace, then generate an array which contains a list of holey
765  * vdevs.  Additionally, add the number of top-level children that currently
766  * exist.
767  */
768 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)769 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
770 {
771 	vdev_t *rvd = spa->spa_root_vdev;
772 	uint64_t *array;
773 	uint_t c, idx;
774 
775 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
776 
777 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
778 		vdev_t *tvd = rvd->vdev_child[c];
779 
780 		if (tvd->vdev_ishole) {
781 			array[idx++] = c;
782 		}
783 	}
784 
785 	if (idx) {
786 		VERIFY0(nvlist_add_uint64_array(config,
787 		    ZPOOL_CONFIG_HOLE_ARRAY, array, idx));
788 	}
789 
790 	VERIFY0(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
791 	    rvd->vdev_children));
792 
793 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
794 }
795 
796 /*
797  * Returns the configuration from the label of the given vdev. For vdevs
798  * which don't have a txg value stored on their label (i.e. spares/cache)
799  * or have not been completely initialized (txg = 0) just return
800  * the configuration from the first valid label we find. Otherwise,
801  * find the most up-to-date label that does not exceed the specified
802  * 'txg' value.
803  */
804 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)805 vdev_label_read_config(vdev_t *vd, uint64_t txg)
806 {
807 	spa_t *spa = vd->vdev_spa;
808 	nvlist_t *config = NULL;
809 	vdev_phys_t *vp[VDEV_LABELS];
810 	abd_t *vp_abd[VDEV_LABELS];
811 	zio_t *zio[VDEV_LABELS];
812 	uint64_t best_txg = 0;
813 	uint64_t label_txg = 0;
814 	int error = 0;
815 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
816 	    ZIO_FLAG_SPECULATIVE;
817 
818 	ASSERT(vd->vdev_validate_thread == curthread ||
819 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
820 
821 	if (!vdev_readable(vd))
822 		return (NULL);
823 
824 	/*
825 	 * The label for a dRAID distributed spare is not stored on disk.
826 	 * Instead it is generated when needed which allows us to bypass
827 	 * the pipeline when reading the config from the label.
828 	 */
829 	if (vd->vdev_ops == &vdev_draid_spare_ops)
830 		return (vdev_draid_read_config_spare(vd));
831 
832 	for (int l = 0; l < VDEV_LABELS; l++) {
833 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
834 		vp[l] = abd_to_buf(vp_abd[l]);
835 	}
836 
837 retry:
838 	for (int l = 0; l < VDEV_LABELS; l++) {
839 		zio[l] = zio_root(spa, NULL, NULL, flags);
840 
841 		vdev_label_read(zio[l], vd, l, vp_abd[l],
842 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
843 		    NULL, NULL, flags);
844 	}
845 	for (int l = 0; l < VDEV_LABELS; l++) {
846 		nvlist_t *label = NULL;
847 
848 		if (zio_wait(zio[l]) == 0 &&
849 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
850 		    &label, 0) == 0) {
851 			/*
852 			 * Auxiliary vdevs won't have txg values in their
853 			 * labels and newly added vdevs may not have been
854 			 * completely initialized so just return the
855 			 * configuration from the first valid label we
856 			 * encounter.
857 			 */
858 			error = nvlist_lookup_uint64(label,
859 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
860 			if ((error || label_txg == 0) && !config) {
861 				config = label;
862 				for (l++; l < VDEV_LABELS; l++)
863 					zio_wait(zio[l]);
864 				break;
865 			} else if (label_txg <= txg && label_txg > best_txg) {
866 				best_txg = label_txg;
867 				nvlist_free(config);
868 				config = fnvlist_dup(label);
869 			}
870 		}
871 
872 		if (label != NULL) {
873 			nvlist_free(label);
874 			label = NULL;
875 		}
876 	}
877 
878 	if (config == NULL && !(flags & ZIO_FLAG_IO_RETRY)) {
879 		flags |= ZIO_FLAG_IO_RETRY;
880 		goto retry;
881 	}
882 
883 	/*
884 	 * We found a valid label but it didn't pass txg restrictions.
885 	 */
886 	if (config == NULL && label_txg != 0) {
887 		vdev_dbgmsg(vd, "label discarded as txg is too large "
888 		    "(%llu > %llu)", (u_longlong_t)label_txg,
889 		    (u_longlong_t)txg);
890 	}
891 
892 	for (int l = 0; l < VDEV_LABELS; l++) {
893 		abd_free(vp_abd[l]);
894 	}
895 
896 	return (config);
897 }
898 
899 /*
900  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
901  * in with the device guid if this spare is active elsewhere on the system.
902  */
903 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)904 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
905     uint64_t *spare_guid, uint64_t *l2cache_guid)
906 {
907 	spa_t *spa = vd->vdev_spa;
908 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
909 	uint64_t vdtxg = 0;
910 	nvlist_t *label;
911 
912 	if (spare_guid)
913 		*spare_guid = 0ULL;
914 	if (l2cache_guid)
915 		*l2cache_guid = 0ULL;
916 
917 	/*
918 	 * Read the label, if any, and perform some basic sanity checks.
919 	 */
920 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
921 		return (B_FALSE);
922 
923 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
924 	    &vdtxg);
925 
926 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
927 	    &state) != 0 ||
928 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
929 	    &device_guid) != 0) {
930 		nvlist_free(label);
931 		return (B_FALSE);
932 	}
933 
934 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
935 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
936 	    &pool_guid) != 0 ||
937 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
938 	    &txg) != 0)) {
939 		nvlist_free(label);
940 		return (B_FALSE);
941 	}
942 
943 	nvlist_free(label);
944 
945 	/*
946 	 * Check to see if this device indeed belongs to the pool it claims to
947 	 * be a part of.  The only way this is allowed is if the device is a hot
948 	 * spare (which we check for later on).
949 	 */
950 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
951 	    !spa_guid_exists(pool_guid, device_guid) &&
952 	    !spa_spare_exists(device_guid, NULL, NULL) &&
953 	    !spa_l2cache_exists(device_guid, NULL))
954 		return (B_FALSE);
955 
956 	/*
957 	 * If the transaction group is zero, then this an initialized (but
958 	 * unused) label.  This is only an error if the create transaction
959 	 * on-disk is the same as the one we're using now, in which case the
960 	 * user has attempted to add the same vdev multiple times in the same
961 	 * transaction.
962 	 */
963 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
964 	    txg == 0 && vdtxg == crtxg)
965 		return (B_TRUE);
966 
967 	/*
968 	 * Check to see if this is a spare device.  We do an explicit check for
969 	 * spa_has_spare() here because it may be on our pending list of spares
970 	 * to add.
971 	 */
972 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
973 	    spa_has_spare(spa, device_guid)) {
974 		if (spare_guid)
975 			*spare_guid = device_guid;
976 
977 		switch (reason) {
978 		case VDEV_LABEL_CREATE:
979 			return (B_TRUE);
980 
981 		case VDEV_LABEL_REPLACE:
982 			return (!spa_has_spare(spa, device_guid) ||
983 			    spare_pool != 0ULL);
984 
985 		case VDEV_LABEL_SPARE:
986 			return (spa_has_spare(spa, device_guid));
987 		default:
988 			break;
989 		}
990 	}
991 
992 	/*
993 	 * Check to see if this is an l2cache device.
994 	 */
995 	if (spa_l2cache_exists(device_guid, NULL) ||
996 	    spa_has_l2cache(spa, device_guid)) {
997 		if (l2cache_guid)
998 			*l2cache_guid = device_guid;
999 
1000 		switch (reason) {
1001 		case VDEV_LABEL_CREATE:
1002 			return (B_TRUE);
1003 
1004 		case VDEV_LABEL_REPLACE:
1005 			return (!spa_has_l2cache(spa, device_guid));
1006 
1007 		case VDEV_LABEL_L2CACHE:
1008 			return (spa_has_l2cache(spa, device_guid));
1009 		default:
1010 			break;
1011 		}
1012 	}
1013 
1014 	/*
1015 	 * We can't rely on a pool's state if it's been imported
1016 	 * read-only.  Instead we look to see if the pools is marked
1017 	 * read-only in the namespace and set the state to active.
1018 	 */
1019 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1020 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
1021 	    spa_mode(spa) == SPA_MODE_READ)
1022 		state = POOL_STATE_ACTIVE;
1023 
1024 	/*
1025 	 * If the device is marked ACTIVE, then this device is in use by another
1026 	 * pool on the system.
1027 	 */
1028 	return (state == POOL_STATE_ACTIVE);
1029 }
1030 
1031 static nvlist_t *
vdev_aux_label_generate(vdev_t * vd,boolean_t reason_spare)1032 vdev_aux_label_generate(vdev_t *vd, boolean_t reason_spare)
1033 {
1034 	/*
1035 	 * For inactive hot spares and level 2 ARC devices, we generate
1036 	 * a special label that identifies as a mutually shared hot
1037 	 * spare or l2cache device. We write the label in case of
1038 	 * addition or removal of hot spare or l2cache vdev (in which
1039 	 * case we want to revert the labels).
1040 	 */
1041 	nvlist_t *label = fnvlist_alloc();
1042 	fnvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1043 	    spa_version(vd->vdev_spa));
1044 	fnvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, reason_spare ?
1045 	    POOL_STATE_SPARE : POOL_STATE_L2CACHE);
1046 	fnvlist_add_uint64(label, ZPOOL_CONFIG_GUID, vd->vdev_guid);
1047 
1048 	/*
1049 	 * This is merely to facilitate reporting the ashift of the
1050 	 * cache device through zdb. The actual retrieval of the
1051 	 * ashift (in vdev_alloc()) uses the nvlist
1052 	 * spa->spa_l2cache->sav_config (populated in
1053 	 * spa_ld_open_aux_vdevs()).
1054 	 */
1055 	if (!reason_spare)
1056 		fnvlist_add_uint64(label, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
1057 
1058 	/*
1059 	 * Add path information to help find it during pool import
1060 	 */
1061 	if (vd->vdev_path != NULL)
1062 		fnvlist_add_string(label, ZPOOL_CONFIG_PATH, vd->vdev_path);
1063 	if (vd->vdev_devid != NULL)
1064 		fnvlist_add_string(label, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
1065 	if (vd->vdev_physpath != NULL) {
1066 		fnvlist_add_string(label, ZPOOL_CONFIG_PHYS_PATH,
1067 		    vd->vdev_physpath);
1068 	}
1069 	return (label);
1070 }
1071 
1072 /*
1073  * Initialize a vdev label.  We check to make sure each leaf device is not in
1074  * use, and writable.  We put down an initial label which we will later
1075  * overwrite with a complete label.  Note that it's important to do this
1076  * sequentially, not in parallel, so that we catch cases of multiple use of the
1077  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1078  * itself.
1079  */
1080 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)1081 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1082 {
1083 	spa_t *spa = vd->vdev_spa;
1084 	nvlist_t *label;
1085 	vdev_phys_t *vp;
1086 	abd_t *vp_abd;
1087 	abd_t *bootenv;
1088 	uberblock_t *ub;
1089 	abd_t *ub_abd;
1090 	zio_t *zio;
1091 	char *buf;
1092 	size_t buflen;
1093 	int error;
1094 	uint64_t spare_guid = 0, l2cache_guid = 0;
1095 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1096 	    ZIO_FLAG_TRYHARD;
1097 	boolean_t reason_spare = (reason == VDEV_LABEL_SPARE || (reason ==
1098 	    VDEV_LABEL_REMOVE && vd->vdev_isspare));
1099 	boolean_t reason_l2cache = (reason == VDEV_LABEL_L2CACHE || (reason ==
1100 	    VDEV_LABEL_REMOVE && vd->vdev_isl2cache));
1101 
1102 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1103 
1104 	for (int c = 0; c < vd->vdev_children; c++)
1105 		if ((error = vdev_label_init(vd->vdev_child[c],
1106 		    crtxg, reason)) != 0)
1107 			return (error);
1108 
1109 	/* Track the creation time for this vdev */
1110 	vd->vdev_crtxg = crtxg;
1111 
1112 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1113 		return (0);
1114 
1115 	/*
1116 	 * Dead vdevs cannot be initialized.
1117 	 */
1118 	if (vdev_is_dead(vd))
1119 		return (SET_ERROR(EIO));
1120 
1121 	/*
1122 	 * Determine if the vdev is in use.
1123 	 */
1124 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1125 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid)) {
1126 		if (spa->spa_create_info == NULL) {
1127 			nvlist_t *nv = fnvlist_alloc();
1128 			nvlist_t *cfg;
1129 
1130 			if (vd->vdev_path != NULL)
1131 				fnvlist_add_string(nv,
1132 				    ZPOOL_CREATE_INFO_VDEV, vd->vdev_path);
1133 
1134 			cfg = vdev_label_read_config(vd, -1ULL);
1135 			if (cfg != NULL) {
1136 				const char *pname;
1137 				if (nvlist_lookup_string(cfg,
1138 				    ZPOOL_CONFIG_POOL_NAME, &pname) == 0)
1139 					fnvlist_add_string(nv,
1140 					    ZPOOL_CREATE_INFO_POOL, pname);
1141 				nvlist_free(cfg);
1142 			}
1143 
1144 			spa->spa_create_info = nv;
1145 		}
1146 		return (SET_ERROR(EBUSY));
1147 	}
1148 
1149 	/*
1150 	 * If this is a request to add or replace a spare or l2cache device
1151 	 * that is in use elsewhere on the system, then we must update the
1152 	 * guid (which was initialized to a random value) to reflect the
1153 	 * actual GUID (which is shared between multiple pools).
1154 	 */
1155 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1156 	    spare_guid != 0ULL) {
1157 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1158 
1159 		vd->vdev_guid += guid_delta;
1160 
1161 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1162 			pvd->vdev_guid_sum += guid_delta;
1163 
1164 		/*
1165 		 * If this is a replacement, then we want to fallthrough to the
1166 		 * rest of the code.  If we're adding a spare, then it's already
1167 		 * labeled appropriately and we can just return.
1168 		 */
1169 		if (reason == VDEV_LABEL_SPARE)
1170 			return (0);
1171 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1172 		    reason == VDEV_LABEL_SPLIT);
1173 	}
1174 
1175 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1176 	    l2cache_guid != 0ULL) {
1177 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1178 
1179 		vd->vdev_guid += guid_delta;
1180 
1181 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1182 			pvd->vdev_guid_sum += guid_delta;
1183 
1184 		/*
1185 		 * If this is a replacement, then we want to fallthrough to the
1186 		 * rest of the code.  If we're adding an l2cache, then it's
1187 		 * already labeled appropriately and we can just return.
1188 		 */
1189 		if (reason == VDEV_LABEL_L2CACHE)
1190 			return (0);
1191 		ASSERT(reason == VDEV_LABEL_REPLACE);
1192 	}
1193 
1194 	/*
1195 	 * Initialize its label.
1196 	 */
1197 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1198 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1199 	vp = abd_to_buf(vp_abd);
1200 
1201 	/*
1202 	 * Generate a label describing the pool and our top-level vdev.
1203 	 * We mark it as being from txg 0 to indicate that it's not
1204 	 * really part of an active pool just yet.  The labels will
1205 	 * be written again with a meaningful txg by spa_sync().
1206 	 */
1207 	if (reason_spare || reason_l2cache) {
1208 		label = vdev_aux_label_generate(vd, reason_spare);
1209 
1210 		/*
1211 		 * When spare or l2cache (aux) vdev is added during pool
1212 		 * creation, spa->spa_uberblock is not written until this
1213 		 * point. Write it on next config sync.
1214 		 */
1215 		if (uberblock_verify(&spa->spa_uberblock))
1216 			spa->spa_aux_sync_uber = B_TRUE;
1217 	} else {
1218 		uint64_t txg = 0ULL;
1219 
1220 		if (reason == VDEV_LABEL_SPLIT)
1221 			txg = spa->spa_uberblock.ub_txg;
1222 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1223 
1224 		/*
1225 		 * Add our creation time.  This allows us to detect multiple
1226 		 * vdev uses as described above, and automatically expires if we
1227 		 * fail.
1228 		 */
1229 		VERIFY0(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1230 		    crtxg));
1231 	}
1232 
1233 	buf = vp->vp_nvlist;
1234 	buflen = sizeof (vp->vp_nvlist);
1235 
1236 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1237 	if (error != 0) {
1238 		nvlist_free(label);
1239 		abd_free(vp_abd);
1240 		/* EFAULT means nvlist_pack ran out of room */
1241 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1242 	}
1243 
1244 	/*
1245 	 * Initialize uberblock template.
1246 	 */
1247 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1248 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1249 	abd_zero_off(ub_abd, sizeof (uberblock_t),
1250 	    VDEV_UBERBLOCK_RING - sizeof (uberblock_t));
1251 	ub = abd_to_buf(ub_abd);
1252 	ub->ub_txg = 0;
1253 
1254 	/* Initialize the 2nd padding area. */
1255 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1256 	abd_zero(bootenv, VDEV_PAD_SIZE);
1257 
1258 	/*
1259 	 * Write everything in parallel.
1260 	 */
1261 	zio = zio_root(spa, NULL, NULL, flags);
1262 
1263 	for (int l = 0; l < VDEV_LABELS; l++) {
1264 
1265 		vdev_label_write(zio, vd, l, vp_abd,
1266 		    offsetof(vdev_label_t, vl_vdev_phys),
1267 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1268 
1269 		/*
1270 		 * Skip the 1st padding area.
1271 		 * Zero out the 2nd padding area where it might have
1272 		 * left over data from previous filesystem format.
1273 		 */
1274 		vdev_label_write(zio, vd, l, bootenv,
1275 		    offsetof(vdev_label_t, vl_be),
1276 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1277 
1278 		vdev_label_write(zio, vd, l, ub_abd,
1279 		    offsetof(vdev_label_t, vl_uberblock),
1280 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1281 	}
1282 
1283 	error = zio_wait(zio);
1284 
1285 	nvlist_free(label);
1286 	abd_free(bootenv);
1287 	abd_free(ub_abd);
1288 	abd_free(vp_abd);
1289 
1290 	/*
1291 	 * If this vdev hasn't been previously identified as a spare, then we
1292 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1293 	 * exists as a spare elsewhere in the system.  Do the same for
1294 	 * level 2 ARC devices.
1295 	 */
1296 	if (error == 0 && !vd->vdev_isspare &&
1297 	    (reason == VDEV_LABEL_SPARE ||
1298 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1299 		spa_spare_add(vd);
1300 
1301 	if (error == 0 && !vd->vdev_isl2cache &&
1302 	    (reason == VDEV_LABEL_L2CACHE ||
1303 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1304 		spa_l2cache_add(vd);
1305 
1306 	return (error);
1307 }
1308 
1309 /*
1310  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1311  * callback to finish, store our abd in the callback pointer. Otherwise, we
1312  * just free our abd and return.
1313  */
1314 static void
vdev_label_read_bootenv_done(zio_t * zio)1315 vdev_label_read_bootenv_done(zio_t *zio)
1316 {
1317 	zio_t *rio = zio->io_private;
1318 	abd_t **cbp = rio->io_private;
1319 
1320 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1321 
1322 	if (zio->io_error == 0) {
1323 		mutex_enter(&rio->io_lock);
1324 		if (*cbp == NULL) {
1325 			/* Will free this buffer in vdev_label_read_bootenv. */
1326 			*cbp = zio->io_abd;
1327 		} else {
1328 			abd_free(zio->io_abd);
1329 		}
1330 		mutex_exit(&rio->io_lock);
1331 	} else {
1332 		abd_free(zio->io_abd);
1333 	}
1334 }
1335 
1336 static void
vdev_label_read_bootenv_impl(zio_t * zio,vdev_t * vd,int flags)1337 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1338 {
1339 	for (int c = 0; c < vd->vdev_children; c++)
1340 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1341 
1342 	/*
1343 	 * We just use the first label that has a correct checksum; the
1344 	 * bootloader should have rewritten them all to be the same on boot,
1345 	 * and any changes we made since boot have been the same across all
1346 	 * labels.
1347 	 */
1348 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1349 		for (int l = 0; l < VDEV_LABELS; l++) {
1350 			vdev_label_read(zio, vd, l,
1351 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1352 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1353 			    vdev_label_read_bootenv_done, zio, flags);
1354 		}
1355 	}
1356 }
1357 
1358 int
vdev_label_read_bootenv(vdev_t * rvd,nvlist_t * bootenv)1359 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1360 {
1361 	nvlist_t *config;
1362 	spa_t *spa = rvd->vdev_spa;
1363 	abd_t *abd = NULL;
1364 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1365 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1366 
1367 	ASSERT(bootenv);
1368 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1369 
1370 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1371 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1372 	int err = zio_wait(zio);
1373 
1374 	if (abd != NULL) {
1375 		char *buf;
1376 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1377 
1378 		vbe->vbe_version = ntohll(vbe->vbe_version);
1379 		switch (vbe->vbe_version) {
1380 		case VB_RAW:
1381 			/*
1382 			 * if we have textual data in vbe_bootenv, create nvlist
1383 			 * with key "envmap".
1384 			 */
1385 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1386 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1387 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1388 			    vbe->vbe_bootenv);
1389 			break;
1390 
1391 		case VB_NVLIST:
1392 			err = nvlist_unpack(vbe->vbe_bootenv,
1393 			    sizeof (vbe->vbe_bootenv), &config, 0);
1394 			if (err == 0) {
1395 				fnvlist_merge(bootenv, config);
1396 				nvlist_free(config);
1397 				break;
1398 			}
1399 			zfs_fallthrough;
1400 		default:
1401 			/* Check for FreeBSD zfs bootonce command string */
1402 			buf = abd_to_buf(abd);
1403 			if (*buf == '\0') {
1404 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1405 				    VB_NVLIST);
1406 				break;
1407 			}
1408 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1409 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1410 		}
1411 
1412 		/*
1413 		 * abd was allocated in vdev_label_read_bootenv_impl()
1414 		 */
1415 		abd_free(abd);
1416 		/*
1417 		 * If we managed to read any successfully,
1418 		 * return success.
1419 		 */
1420 		return (0);
1421 	}
1422 	return (err);
1423 }
1424 
1425 int
vdev_label_write_bootenv(vdev_t * vd,nvlist_t * env)1426 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1427 {
1428 	zio_t *zio;
1429 	spa_t *spa = vd->vdev_spa;
1430 	vdev_boot_envblock_t *bootenv;
1431 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1432 	    ZIO_FLAG_TRYHARD;
1433 	int error;
1434 	size_t nvsize;
1435 	char *nvbuf;
1436 	const char *tmp;
1437 
1438 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1439 	if (error != 0)
1440 		return (SET_ERROR(error));
1441 
1442 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1443 		return (SET_ERROR(E2BIG));
1444 	}
1445 
1446 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1447 
1448 	error = ENXIO;
1449 	for (int c = 0; c < vd->vdev_children; c++) {
1450 		int child_err;
1451 
1452 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1453 		/*
1454 		 * As long as any of the disks managed to write all of their
1455 		 * labels successfully, return success.
1456 		 */
1457 		if (child_err == 0)
1458 			error = child_err;
1459 	}
1460 
1461 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1462 	    !vdev_writeable(vd)) {
1463 		return (error);
1464 	}
1465 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1466 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1467 	abd_zero(abd, VDEV_PAD_SIZE);
1468 
1469 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1470 	nvbuf = bootenv->vbe_bootenv;
1471 	nvsize = sizeof (bootenv->vbe_bootenv);
1472 
1473 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1474 	switch (bootenv->vbe_version) {
1475 	case VB_RAW:
1476 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1477 			(void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1478 		}
1479 		error = 0;
1480 		break;
1481 
1482 	case VB_NVLIST:
1483 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1484 		    KM_SLEEP);
1485 		break;
1486 
1487 	default:
1488 		error = EINVAL;
1489 		break;
1490 	}
1491 
1492 	if (error == 0) {
1493 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1494 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1495 	} else {
1496 		abd_free(abd);
1497 		return (SET_ERROR(error));
1498 	}
1499 
1500 	zio = zio_root(spa, NULL, NULL, flags);
1501 	for (int l = 0; l < VDEV_LABELS; l++) {
1502 		vdev_label_write(zio, vd, l, abd,
1503 		    offsetof(vdev_label_t, vl_be),
1504 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1505 	}
1506 
1507 	error = zio_wait(zio);
1508 
1509 	abd_free(abd);
1510 	return (error);
1511 }
1512 
1513 /*
1514  * ==========================================================================
1515  * uberblock load/sync
1516  * ==========================================================================
1517  */
1518 
1519 /*
1520  * Consider the following situation: txg is safely synced to disk.  We've
1521  * written the first uberblock for txg + 1, and then we lose power.  When we
1522  * come back up, we fail to see the uberblock for txg + 1 because, say,
1523  * it was on a mirrored device and the replica to which we wrote txg + 1
1524  * is now offline.  If we then make some changes and sync txg + 1, and then
1525  * the missing replica comes back, then for a few seconds we'll have two
1526  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1527  * among uberblocks with equal txg, choose the one with the latest timestamp.
1528  */
1529 int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1530 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1531 {
1532 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1533 
1534 	if (likely(cmp))
1535 		return (cmp);
1536 
1537 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1538 	if (likely(cmp))
1539 		return (cmp);
1540 
1541 	/*
1542 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1543 	 * ZFS, e.g. OpenZFS >= 0.7.
1544 	 *
1545 	 * If one ub has MMP and the other does not, they were written by
1546 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1547 	 * a 0 value.
1548 	 *
1549 	 * Since timestamp and txg are the same if we get this far, either is
1550 	 * acceptable for importing the pool.
1551 	 */
1552 	unsigned int seq1 = 0;
1553 	unsigned int seq2 = 0;
1554 
1555 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1556 		seq1 = MMP_SEQ(ub1);
1557 
1558 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1559 		seq2 = MMP_SEQ(ub2);
1560 
1561 	return (TREE_CMP(seq1, seq2));
1562 }
1563 
1564 struct ubl_cbdata {
1565 	uberblock_t	ubl_latest;	/* Most recent uberblock */
1566 	uberblock_t	*ubl_ubbest;	/* Best uberblock (w/r/t max_txg) */
1567 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1568 };
1569 
1570 static void
vdev_uberblock_load_done(zio_t * zio)1571 vdev_uberblock_load_done(zio_t *zio)
1572 {
1573 	vdev_t *vd = zio->io_vd;
1574 	spa_t *spa = zio->io_spa;
1575 	zio_t *rio = zio->io_private;
1576 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1577 	struct ubl_cbdata *cbp = rio->io_private;
1578 
1579 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1580 
1581 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1582 		mutex_enter(&rio->io_lock);
1583 		if (vdev_uberblock_compare(ub, &cbp->ubl_latest) > 0) {
1584 			cbp->ubl_latest = *ub;
1585 		}
1586 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1587 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1588 			/*
1589 			 * Keep track of the vdev in which this uberblock
1590 			 * was found. We will use this information later
1591 			 * to obtain the config nvlist associated with
1592 			 * this uberblock.
1593 			 */
1594 			*cbp->ubl_ubbest = *ub;
1595 			cbp->ubl_vd = vd;
1596 		}
1597 		mutex_exit(&rio->io_lock);
1598 	}
1599 
1600 	abd_free(zio->io_abd);
1601 }
1602 
1603 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1604 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1605     struct ubl_cbdata *cbp)
1606 {
1607 	for (int c = 0; c < vd->vdev_children; c++)
1608 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1609 
1610 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1611 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1612 		for (int l = 0; l < VDEV_LABELS; l++) {
1613 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1614 				vdev_label_read(zio, vd, l,
1615 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1616 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1617 				    VDEV_UBERBLOCK_SIZE(vd),
1618 				    vdev_uberblock_load_done, zio, flags);
1619 			}
1620 		}
1621 	}
1622 }
1623 
1624 /*
1625  * Reads the 'best' uberblock from disk along with its associated
1626  * configuration. First, we read the uberblock array of each label of each
1627  * vdev, keeping track of the uberblock with the highest txg in each array.
1628  * Then, we read the configuration from the same vdev as the best uberblock.
1629  */
1630 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1631 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1632 {
1633 	zio_t *zio;
1634 	spa_t *spa = rvd->vdev_spa;
1635 	struct ubl_cbdata cb;
1636 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1637 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1638 
1639 	ASSERT(ub);
1640 	ASSERT(config);
1641 
1642 	memset(ub, 0, sizeof (uberblock_t));
1643 	memset(&cb, 0, sizeof (cb));
1644 	*config = NULL;
1645 
1646 	cb.ubl_ubbest = ub;
1647 
1648 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1649 	zio = zio_root(spa, NULL, &cb, flags);
1650 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1651 	(void) zio_wait(zio);
1652 
1653 	/*
1654 	 * It's possible that the best uberblock was discovered on a label
1655 	 * that has a configuration which was written in a future txg.
1656 	 * Search all labels on this vdev to find the configuration that
1657 	 * matches the txg for our uberblock.
1658 	 */
1659 	if (cb.ubl_vd != NULL) {
1660 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s, "
1661 		    "txg=%llu seq=%llu", spa_load_name(spa),
1662 		    (u_longlong_t)ub->ub_txg,
1663 		    (u_longlong_t)(MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
1664 
1665 		if (ub->ub_raidz_reflow_info !=
1666 		    cb.ubl_latest.ub_raidz_reflow_info) {
1667 			vdev_dbgmsg(cb.ubl_vd,
1668 			    "spa=%s best uberblock (txg=%llu info=0x%llx) "
1669 			    "has different raidz_reflow_info than latest "
1670 			    "uberblock (txg=%llu info=0x%llx)",
1671 			    spa_load_name(spa),
1672 			    (u_longlong_t)ub->ub_txg,
1673 			    (u_longlong_t)ub->ub_raidz_reflow_info,
1674 			    (u_longlong_t)cb.ubl_latest.ub_txg,
1675 			    (u_longlong_t)cb.ubl_latest.ub_raidz_reflow_info);
1676 			memset(ub, 0, sizeof (uberblock_t));
1677 			spa_config_exit(spa, SCL_ALL, FTAG);
1678 			return;
1679 		}
1680 
1681 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1682 		if (*config == NULL && spa->spa_extreme_rewind) {
1683 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1684 			    "Trying again without txg restrictions.");
1685 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1686 		}
1687 		if (*config == NULL) {
1688 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1689 		}
1690 	}
1691 	spa_config_exit(spa, SCL_ALL, FTAG);
1692 }
1693 
1694 /*
1695  * For use when a leaf vdev is expanded.
1696  * The location of labels 2 and 3 changed, and at the new location the
1697  * uberblock rings are either empty or contain garbage.  The sync will write
1698  * new configs there because the vdev is dirty, but expansion also needs the
1699  * uberblock rings copied.  Read them from label 0 which did not move.
1700  *
1701  * Since the point is to populate labels {2,3} with valid uberblocks,
1702  * we zero uberblocks we fail to read or which are not valid.
1703  */
1704 
1705 static void
vdev_copy_uberblocks(vdev_t * vd)1706 vdev_copy_uberblocks(vdev_t *vd)
1707 {
1708 	abd_t *ub_abd;
1709 	zio_t *write_zio;
1710 	int locks = (SCL_L2ARC | SCL_ZIO);
1711 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1712 	    ZIO_FLAG_SPECULATIVE;
1713 
1714 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1715 	    SCL_STATE);
1716 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1717 
1718 	/*
1719 	 * No uberblocks are stored on distributed spares, they may be
1720 	 * safely skipped when expanding a leaf vdev.
1721 	 */
1722 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1723 		return;
1724 
1725 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1726 
1727 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1728 
1729 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1730 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1731 		const int src_label = 0;
1732 		zio_t *zio;
1733 
1734 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1735 		vdev_label_read(zio, vd, src_label, ub_abd,
1736 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1737 		    NULL, NULL, flags);
1738 
1739 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1740 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1741 
1742 		for (int l = 2; l < VDEV_LABELS; l++)
1743 			vdev_label_write(write_zio, vd, l, ub_abd,
1744 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1745 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1746 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1747 	}
1748 	(void) zio_wait(write_zio);
1749 
1750 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1751 
1752 	abd_free(ub_abd);
1753 }
1754 
1755 /*
1756  * On success, increment root zio's count of good writes.
1757  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1758  */
1759 static void
vdev_uberblock_sync_done(zio_t * zio)1760 vdev_uberblock_sync_done(zio_t *zio)
1761 {
1762 	uint64_t *good_writes = zio->io_private;
1763 
1764 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1765 		atomic_inc_64(good_writes);
1766 }
1767 
1768 /*
1769  * Write the uberblock to all labels of all leaves of the specified vdev.
1770  */
1771 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1772 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1773     uberblock_t *ub, vdev_t *vd, int flags)
1774 {
1775 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1776 		vdev_uberblock_sync(zio, good_writes,
1777 		    ub, vd->vdev_child[c], flags);
1778 	}
1779 
1780 	if (!vd->vdev_ops->vdev_op_leaf)
1781 		return;
1782 
1783 	if (!vdev_writeable(vd))
1784 		return;
1785 
1786 	/*
1787 	 * There's no need to write uberblocks to a distributed spare, they
1788 	 * are already stored on all the leaves of the parent dRAID.  For
1789 	 * this same reason vdev_uberblock_load_impl() skips distributed
1790 	 * spares when reading uberblocks.
1791 	 */
1792 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1793 		return;
1794 
1795 	/* If the vdev was expanded, need to copy uberblock rings. */
1796 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1797 	    vd->vdev_copy_uberblocks == B_TRUE) {
1798 		vdev_copy_uberblocks(vd);
1799 		vd->vdev_copy_uberblocks = B_FALSE;
1800 	}
1801 
1802 	/*
1803 	 * We chose a slot based on the txg.  If this uberblock has a special
1804 	 * RAIDZ expansion state, then it is essentially an update of the
1805 	 * current uberblock (it has the same txg).  However, the current
1806 	 * state is committed, so we want to write it to a different slot. If
1807 	 * we overwrote the same slot, and we lose power during the uberblock
1808 	 * write, and the disk does not do single-sector overwrites
1809 	 * atomically (even though it is required to - i.e. we should see
1810 	 * either the old or the new uberblock), then we could lose this
1811 	 * txg's uberblock. Rewinding to the previous txg's uberblock may not
1812 	 * be possible because RAIDZ expansion may have already overwritten
1813 	 * some of the data, so we need the progress indicator in the
1814 	 * uberblock.
1815 	 */
1816 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1817 	int n = (ub->ub_txg - (RRSS_GET_STATE(ub) == RRSS_SCRATCH_VALID)) %
1818 	    (VDEV_UBERBLOCK_COUNT(vd) - m);
1819 
1820 	/* Copy the uberblock_t into the ABD */
1821 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1822 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1823 	abd_zero_off(ub_abd, sizeof (uberblock_t),
1824 	    VDEV_UBERBLOCK_SIZE(vd) - sizeof (uberblock_t));
1825 
1826 	for (int l = 0; l < VDEV_LABELS; l++)
1827 		vdev_label_write(zio, vd, l, ub_abd,
1828 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1829 		    vdev_uberblock_sync_done, good_writes,
1830 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1831 
1832 	abd_free(ub_abd);
1833 }
1834 
1835 /* Sync the uberblocks to all vdevs in svd[] */
1836 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1837 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1838 {
1839 	spa_t *spa = svd[0]->vdev_spa;
1840 	zio_t *zio;
1841 	uint64_t good_writes = 0;
1842 
1843 	zio = zio_root(spa, NULL, NULL, flags);
1844 
1845 	for (int v = 0; v < svdcount; v++)
1846 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1847 
1848 	if (spa->spa_aux_sync_uber) {
1849 		for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1850 			vdev_uberblock_sync(zio, &good_writes, ub,
1851 			    spa->spa_spares.sav_vdevs[v], flags);
1852 		}
1853 		for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1854 			vdev_uberblock_sync(zio, &good_writes, ub,
1855 			    spa->spa_l2cache.sav_vdevs[v], flags);
1856 		}
1857 	}
1858 	(void) zio_wait(zio);
1859 
1860 	/*
1861 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1862 	 * are no longer needed (because the new uberblocks and the even
1863 	 * labels are safely on disk), so it is safe to overwrite them.
1864 	 */
1865 	zio = zio_root(spa, NULL, NULL, flags);
1866 
1867 	for (int v = 0; v < svdcount; v++) {
1868 		if (vdev_writeable(svd[v])) {
1869 			zio_flush(zio, svd[v]);
1870 		}
1871 	}
1872 	if (spa->spa_aux_sync_uber) {
1873 		spa->spa_aux_sync_uber = B_FALSE;
1874 		for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1875 			if (vdev_writeable(spa->spa_spares.sav_vdevs[v])) {
1876 				zio_flush(zio, spa->spa_spares.sav_vdevs[v]);
1877 			}
1878 		}
1879 		for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1880 			if (vdev_writeable(spa->spa_l2cache.sav_vdevs[v])) {
1881 				zio_flush(zio, spa->spa_l2cache.sav_vdevs[v]);
1882 			}
1883 		}
1884 	}
1885 
1886 	(void) zio_wait(zio);
1887 
1888 	return (good_writes >= 1 ? 0 : EIO);
1889 }
1890 
1891 /*
1892  * On success, increment the count of good writes for our top-level vdev.
1893  */
1894 static void
vdev_label_sync_done(zio_t * zio)1895 vdev_label_sync_done(zio_t *zio)
1896 {
1897 	uint64_t *good_writes = zio->io_private;
1898 
1899 	if (zio->io_error == 0)
1900 		atomic_inc_64(good_writes);
1901 }
1902 
1903 /*
1904  * If there weren't enough good writes, indicate failure to the parent.
1905  */
1906 static void
vdev_label_sync_top_done(zio_t * zio)1907 vdev_label_sync_top_done(zio_t *zio)
1908 {
1909 	uint64_t *good_writes = zio->io_private;
1910 
1911 	if (*good_writes == 0)
1912 		zio->io_error = SET_ERROR(EIO);
1913 
1914 	kmem_free(good_writes, sizeof (uint64_t));
1915 }
1916 
1917 /*
1918  * We ignore errors for log and cache devices, simply free the private data.
1919  */
1920 static void
vdev_label_sync_ignore_done(zio_t * zio)1921 vdev_label_sync_ignore_done(zio_t *zio)
1922 {
1923 	kmem_free(zio->io_private, sizeof (uint64_t));
1924 }
1925 
1926 /*
1927  * Write all even or odd labels to all leaves of the specified vdev.
1928  */
1929 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1930 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1931     vdev_t *vd, int l, uint64_t txg, int flags)
1932 {
1933 	nvlist_t *label;
1934 	vdev_phys_t *vp;
1935 	abd_t *vp_abd;
1936 	char *buf;
1937 	size_t buflen;
1938 	vdev_t *pvd = vd->vdev_parent;
1939 	boolean_t spare_in_use = B_FALSE;
1940 
1941 	for (int c = 0; c < vd->vdev_children; c++) {
1942 		vdev_label_sync(zio, good_writes,
1943 		    vd->vdev_child[c], l, txg, flags);
1944 	}
1945 
1946 	if (!vd->vdev_ops->vdev_op_leaf)
1947 		return;
1948 
1949 	if (!vdev_writeable(vd))
1950 		return;
1951 
1952 	/*
1953 	 * The top-level config never needs to be written to a distributed
1954 	 * spare.  When read vdev_dspare_label_read_config() will generate
1955 	 * the config for the vdev_label_read_config().
1956 	 */
1957 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1958 		return;
1959 
1960 	if (pvd && pvd->vdev_ops == &vdev_spare_ops)
1961 		spare_in_use = B_TRUE;
1962 
1963 	/*
1964 	 * Generate a label describing the top-level config to which we belong.
1965 	 */
1966 	if ((vd->vdev_isspare && !spare_in_use) || vd->vdev_isl2cache) {
1967 		label = vdev_aux_label_generate(vd, vd->vdev_isspare);
1968 	} else {
1969 		label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1970 	}
1971 
1972 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1973 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1974 	vp = abd_to_buf(vp_abd);
1975 
1976 	buf = vp->vp_nvlist;
1977 	buflen = sizeof (vp->vp_nvlist);
1978 
1979 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1980 		for (; l < VDEV_LABELS; l += 2) {
1981 			vdev_label_write(zio, vd, l, vp_abd,
1982 			    offsetof(vdev_label_t, vl_vdev_phys),
1983 			    sizeof (vdev_phys_t),
1984 			    vdev_label_sync_done, good_writes,
1985 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1986 		}
1987 	}
1988 
1989 	abd_free(vp_abd);
1990 	nvlist_free(label);
1991 }
1992 
1993 static int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1994 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1995 {
1996 	list_t *dl = &spa->spa_config_dirty_list;
1997 	vdev_t *vd;
1998 	zio_t *zio;
1999 	int error;
2000 
2001 	/*
2002 	 * Write the new labels to disk.
2003 	 */
2004 	zio = zio_root(spa, NULL, NULL, flags);
2005 
2006 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
2007 		uint64_t *good_writes;
2008 
2009 		ASSERT(!vd->vdev_ishole);
2010 
2011 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
2012 		zio_t *vio = zio_null(zio, spa, NULL,
2013 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
2014 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
2015 		    good_writes, flags);
2016 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
2017 		zio_nowait(vio);
2018 	}
2019 
2020 	/*
2021 	 * AUX path may have changed during import
2022 	 */
2023 	spa_aux_vdev_t *sav[2] = {&spa->spa_spares, &spa->spa_l2cache};
2024 	for (int i = 0; i < 2; i++) {
2025 		for (int v = 0; v < sav[i]->sav_count; v++) {
2026 			uint64_t *good_writes;
2027 			if (!sav[i]->sav_label_sync)
2028 				continue;
2029 			good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
2030 			zio_t *vio = zio_null(zio, spa, NULL,
2031 			    vdev_label_sync_ignore_done, good_writes, flags);
2032 			vdev_label_sync(vio, good_writes, sav[i]->sav_vdevs[v],
2033 			    l, txg, flags);
2034 			zio_nowait(vio);
2035 		}
2036 	}
2037 
2038 	error = zio_wait(zio);
2039 
2040 	/*
2041 	 * Flush the new labels to disk.
2042 	 */
2043 	zio = zio_root(spa, NULL, NULL, flags);
2044 
2045 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
2046 		zio_flush(zio, vd);
2047 
2048 	for (int i = 0; i < 2; i++) {
2049 		if (!sav[i]->sav_label_sync)
2050 			continue;
2051 		for (int v = 0; v < sav[i]->sav_count; v++)
2052 			zio_flush(zio, sav[i]->sav_vdevs[v]);
2053 		if (l == 1)
2054 			sav[i]->sav_label_sync = B_FALSE;
2055 	}
2056 
2057 	(void) zio_wait(zio);
2058 
2059 	return (error);
2060 }
2061 
2062 /*
2063  * Sync the uberblock and any changes to the vdev configuration.
2064  *
2065  * The order of operations is carefully crafted to ensure that
2066  * if the system panics or loses power at any time, the state on disk
2067  * is still transactionally consistent.  The in-line comments below
2068  * describe the failure semantics at each stage.
2069  *
2070  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
2071  * at any time, you can just call it again, and it will resume its work.
2072  */
2073 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)2074 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
2075 {
2076 	spa_t *spa = svd[0]->vdev_spa;
2077 	uberblock_t *ub = &spa->spa_uberblock;
2078 	int error = 0;
2079 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
2080 
2081 	ASSERT(svdcount != 0);
2082 retry:
2083 	/*
2084 	 * Normally, we don't want to try too hard to write every label and
2085 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
2086 	 * sync process to block while we retry.  But if we can't write a
2087 	 * single label out, we should retry with ZIO_FLAG_IO_RETRY before
2088 	 * bailing out and declaring the pool faulted.
2089 	 */
2090 	if (error != 0) {
2091 		if ((flags & ZIO_FLAG_IO_RETRY) != 0)
2092 			return (error);
2093 		flags |= ZIO_FLAG_IO_RETRY;
2094 	}
2095 
2096 	ASSERT(ub->ub_txg <= txg);
2097 
2098 	/*
2099 	 * If this isn't a resync due to I/O errors,
2100 	 * and nothing changed in this transaction group,
2101 	 * and multihost protection isn't enabled,
2102 	 * and the vdev configuration hasn't changed,
2103 	 * then there's nothing to do.
2104 	 */
2105 	if (ub->ub_txg < txg) {
2106 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
2107 		    txg, spa->spa_mmp.mmp_delay);
2108 
2109 		if (!changed && list_is_empty(&spa->spa_config_dirty_list) &&
2110 		    !spa_multihost(spa))
2111 			return (0);
2112 	}
2113 
2114 	if (txg > spa_freeze_txg(spa))
2115 		return (0);
2116 
2117 	ASSERT(txg <= spa->spa_final_txg);
2118 
2119 	/*
2120 	 * Flush the write cache of every disk that's been written to
2121 	 * in this transaction group.  This ensures that all blocks
2122 	 * written in this txg will be committed to stable storage
2123 	 * before any uberblock that references them.
2124 	 */
2125 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
2126 
2127 	for (vdev_t *vd =
2128 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
2129 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
2130 		zio_flush(zio, vd);
2131 
2132 	(void) zio_wait(zio);
2133 
2134 	/*
2135 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
2136 	 * system dies in the middle of this process, that's OK: all of the
2137 	 * even labels that made it to disk will be newer than any uberblock,
2138 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
2139 	 * which have not yet been touched, will still be valid.  We flush
2140 	 * the new labels to disk to ensure that all even-label updates
2141 	 * are committed to stable storage before the uberblock update.
2142 	 */
2143 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
2144 		if ((flags & ZIO_FLAG_IO_RETRY) != 0) {
2145 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2146 			    "for pool '%s' when syncing out the even labels "
2147 			    "of dirty vdevs", error, spa_name(spa));
2148 		}
2149 		goto retry;
2150 	}
2151 
2152 	/*
2153 	 * Sync the uberblocks to all vdevs in svd[].
2154 	 * If the system dies in the middle of this step, there are two cases
2155 	 * to consider, and the on-disk state is consistent either way:
2156 	 *
2157 	 * (1)	If none of the new uberblocks made it to disk, then the
2158 	 *	previous uberblock will be the newest, and the odd labels
2159 	 *	(which had not yet been touched) will be valid with respect
2160 	 *	to that uberblock.
2161 	 *
2162 	 * (2)	If one or more new uberblocks made it to disk, then they
2163 	 *	will be the newest, and the even labels (which had all
2164 	 *	been successfully committed) will be valid with respect
2165 	 *	to the new uberblocks.
2166 	 */
2167 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
2168 		if ((flags & ZIO_FLAG_IO_RETRY) != 0) {
2169 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2170 			    "%d for pool '%s'", error, spa_name(spa));
2171 		}
2172 		goto retry;
2173 	}
2174 
2175 	if (spa_multihost(spa))
2176 		mmp_update_uberblock(spa, ub);
2177 
2178 	/*
2179 	 * Sync out odd labels for every dirty vdev.  If the system dies
2180 	 * in the middle of this process, the even labels and the new
2181 	 * uberblocks will suffice to open the pool.  The next time
2182 	 * the pool is opened, the first thing we'll do -- before any
2183 	 * user data is modified -- is mark every vdev dirty so that
2184 	 * all labels will be brought up to date.  We flush the new labels
2185 	 * to disk to ensure that all odd-label updates are committed to
2186 	 * stable storage before the next transaction group begins.
2187 	 */
2188 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2189 		if ((flags & ZIO_FLAG_IO_RETRY) != 0) {
2190 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2191 			    "for pool '%s' when syncing out the odd labels of "
2192 			    "dirty vdevs", error, spa_name(spa));
2193 		}
2194 		goto retry;
2195 	}
2196 
2197 	return (0);
2198 }
2199