1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Intel Corporation.
26 * Copyright 2020 Joyent, Inc.
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/uberblock_impl.h>
147#include <sys/metaslab.h>
148#include <sys/metaslab_impl.h>
149#include <sys/zio.h>
150#include <sys/dsl_scan.h>
151#include <sys/abd.h>
152#include <sys/fs/zfs.h>
153
154/*
155 * Basic routines to read and write from a vdev label.
156 * Used throughout the rest of this file.
157 */
158uint64_t
159vdev_label_offset(uint64_t psize, int l, uint64_t offset)
160{
161	ASSERT(offset < sizeof (vdev_label_t));
162	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
163
164	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
165	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
166}
167
168/*
169 * Returns back the vdev label associated with the passed in offset.
170 */
171int
172vdev_label_number(uint64_t psize, uint64_t offset)
173{
174	int l;
175
176	if (offset >= psize - VDEV_LABEL_END_SIZE) {
177		offset -= psize - VDEV_LABEL_END_SIZE;
178		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
179	}
180	l = offset / sizeof (vdev_label_t);
181	return (l < VDEV_LABELS ? l : -1);
182}
183
184static void
185vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
186    uint64_t size, zio_done_func_t *done, void *private, int flags)
187{
188	ASSERT(
189	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
190	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
191	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
192
193	zio_nowait(zio_read_phys(zio, vd,
194	    vdev_label_offset(vd->vdev_psize, l, offset),
195	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
196	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
197}
198
199void
200vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
201    uint64_t size, zio_done_func_t *done, void *private, int flags)
202{
203	ASSERT(
204	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
205	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
206	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
207
208	zio_nowait(zio_write_phys(zio, vd,
209	    vdev_label_offset(vd->vdev_psize, l, offset),
210	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
211	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
212}
213
214/*
215 * Generate the nvlist representing this vdev's stats
216 */
217void
218vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
219{
220	nvlist_t *nvx;
221	vdev_stat_t *vs;
222	vdev_stat_ex_t *vsx;
223
224	vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
225	vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
226
227	vdev_get_stats_ex(vd, vs, vsx);
228	fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
229	    (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
230
231	/*
232	 * Add extended stats into a special extended stats nvlist.  This keeps
233	 * all the extended stats nicely grouped together.  The extended stats
234	 * nvlist is then added to the main nvlist.
235	 */
236	nvx = fnvlist_alloc();
237
238	/* ZIOs in flight to disk */
239	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
240	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
241
242	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
243	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
244
245	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
246	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
247
248	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
249	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
250
251	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
252	    vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
253
254	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
255	    vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
256
257	/* ZIOs pending */
258	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
259	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
260
261	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
262	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
263
264	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
265	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
266
267	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
268	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
269
270	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
271	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
272
273	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
274	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
275
276	/* Histograms */
277	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
278	    vsx->vsx_total_histo[ZIO_TYPE_READ],
279	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
280
281	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
282	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
283	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
284
285	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
286	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
287	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
288
289	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
290	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
291	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
292
293	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
294	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
295	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
296
297	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
298	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
299	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
300
301	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
302	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
303	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
304
305	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
306	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
307	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
308
309	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
310	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
311	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
312
313	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
314	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
315	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
316
317	/* Request sizes */
318	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
319	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
320	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
321
322	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
323	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
324	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
325
326	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
327	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
328	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
329
330	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
331	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
332	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
333
334	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
335	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
336	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
337
338	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
339	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
340	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
341
342	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
343	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
344	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
345
346	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
347	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
348	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
349
350	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
351	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
352	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
353
354	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
355	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
356	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
357
358	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
359	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
360	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
361
362	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
363	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
364	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
365
366	/* IO delays */
367	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
368
369	/* Add extended stats nvlist to main nvlist */
370	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
371
372	nvlist_free(nvx);
373	kmem_free(vs, sizeof (*vs));
374	kmem_free(vsx, sizeof (*vsx));
375}
376
377static void
378root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
379{
380	spa_t *spa = vd->vdev_spa;
381
382	if (vd != spa->spa_root_vdev)
383		return;
384
385	/* provide either current or previous scan information */
386	pool_scan_stat_t ps;
387	if (spa_scan_get_stats(spa, &ps) == 0) {
388		fnvlist_add_uint64_array(nvl,
389		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
390		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
391	}
392
393	pool_removal_stat_t prs;
394	if (spa_removal_get_stats(spa, &prs) == 0) {
395		fnvlist_add_uint64_array(nvl,
396		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
397		    sizeof (prs) / sizeof (uint64_t));
398	}
399
400	pool_checkpoint_stat_t pcs;
401	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
402		fnvlist_add_uint64_array(nvl,
403		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
404		    sizeof (pcs) / sizeof (uint64_t));
405	}
406}
407
408/*
409 * Generate the nvlist representing this vdev's config.
410 */
411nvlist_t *
412vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
413    vdev_config_flag_t flags)
414{
415	nvlist_t *nv = NULL;
416	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
417
418	nv = fnvlist_alloc();
419
420	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
421	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
422		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
423	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
424
425	if (vd->vdev_path != NULL)
426		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
427
428	if (vd->vdev_devid != NULL)
429		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
430
431	if (vd->vdev_physpath != NULL)
432		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
433		    vd->vdev_physpath);
434
435	if (vd->vdev_fru != NULL)
436		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
437
438	if (vd->vdev_nparity != 0) {
439		ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
440		    VDEV_TYPE_RAIDZ) == 0);
441
442		/*
443		 * Make sure someone hasn't managed to sneak a fancy new vdev
444		 * into a crufty old storage pool.
445		 */
446		ASSERT(vd->vdev_nparity == 1 ||
447		    (vd->vdev_nparity <= 2 &&
448		    spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
449		    (vd->vdev_nparity <= 3 &&
450		    spa_version(spa) >= SPA_VERSION_RAIDZ3));
451
452		/*
453		 * Note that we'll add the nparity tag even on storage pools
454		 * that only support a single parity device -- older software
455		 * will just ignore it.
456		 */
457		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
458	}
459
460	if (vd->vdev_wholedisk != -1ULL)
461		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
462		    vd->vdev_wholedisk);
463
464	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
465		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
466
467	if (vd->vdev_isspare)
468		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
469
470	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
471	    vd == vd->vdev_top) {
472		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
473		    vd->vdev_ms_array);
474		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
475		    vd->vdev_ms_shift);
476		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
477		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
478		    vd->vdev_asize);
479		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
480		if (vd->vdev_removing) {
481			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
482			    vd->vdev_removing);
483		}
484
485		/* zpool command expects alloc class data */
486		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
487			const char *bias = NULL;
488
489			switch (vd->vdev_alloc_bias) {
490			case VDEV_BIAS_LOG:
491				bias = VDEV_ALLOC_BIAS_LOG;
492				break;
493			case VDEV_BIAS_SPECIAL:
494				bias = VDEV_ALLOC_BIAS_SPECIAL;
495				break;
496			case VDEV_BIAS_DEDUP:
497				bias = VDEV_ALLOC_BIAS_DEDUP;
498				break;
499			default:
500				ASSERT3U(vd->vdev_alloc_bias, ==,
501				    VDEV_BIAS_NONE);
502			}
503			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
504			    bias);
505		}
506	}
507
508	if (vd->vdev_dtl_sm != NULL) {
509		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
510		    space_map_object(vd->vdev_dtl_sm));
511	}
512
513	if (vic->vic_mapping_object != 0) {
514		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
515		    vic->vic_mapping_object);
516	}
517
518	if (vic->vic_births_object != 0) {
519		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
520		    vic->vic_births_object);
521	}
522
523	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
524		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
525		    vic->vic_prev_indirect_vdev);
526	}
527
528	if (vd->vdev_crtxg)
529		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
530
531	if (flags & VDEV_CONFIG_MOS) {
532		if (vd->vdev_leaf_zap != 0) {
533			ASSERT(vd->vdev_ops->vdev_op_leaf);
534			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
535			    vd->vdev_leaf_zap);
536		}
537
538		if (vd->vdev_top_zap != 0) {
539			ASSERT(vd == vd->vdev_top);
540			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
541			    vd->vdev_top_zap);
542		}
543
544		if (vd->vdev_resilver_deferred) {
545			ASSERT(vd->vdev_ops->vdev_op_leaf);
546			ASSERT(spa->spa_resilver_deferred);
547			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
548		}
549	}
550
551	if (getstats) {
552		vdev_config_generate_stats(vd, nv);
553
554		root_vdev_actions_getprogress(vd, nv);
555
556		/*
557		 * Note: this can be called from open context
558		 * (spa_get_stats()), so we need the rwlock to prevent
559		 * the mapping from being changed by condensing.
560		 */
561		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
562		if (vd->vdev_indirect_mapping != NULL) {
563			ASSERT(vd->vdev_indirect_births != NULL);
564			vdev_indirect_mapping_t *vim =
565			    vd->vdev_indirect_mapping;
566			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
567			    vdev_indirect_mapping_size(vim));
568		}
569		rw_exit(&vd->vdev_indirect_rwlock);
570		if (vd->vdev_mg != NULL &&
571		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
572			/*
573			 * Compute approximately how much memory would be used
574			 * for the indirect mapping if this device were to
575			 * be removed.
576			 *
577			 * Note: If the frag metric is invalid, then not
578			 * enough metaslabs have been converted to have
579			 * histograms.
580			 */
581			uint64_t seg_count = 0;
582			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
583
584			/*
585			 * There are the same number of allocated segments
586			 * as free segments, so we will have at least one
587			 * entry per free segment.  However, small free
588			 * segments (smaller than vdev_removal_max_span)
589			 * will be combined with adjacent allocated segments
590			 * as a single mapping.
591			 */
592			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
593				if (1ULL << (i + 1) < vdev_removal_max_span) {
594					to_alloc +=
595					    vd->vdev_mg->mg_histogram[i] <<
596					    i + 1;
597				} else {
598					seg_count +=
599					    vd->vdev_mg->mg_histogram[i];
600				}
601			}
602
603			/*
604			 * The maximum length of a mapping is
605			 * zfs_remove_max_segment, so we need at least one entry
606			 * per zfs_remove_max_segment of allocated data.
607			 */
608			seg_count += to_alloc / zfs_remove_max_segment;
609
610			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
611			    seg_count *
612			    sizeof (vdev_indirect_mapping_entry_phys_t));
613		}
614	}
615
616	if (!vd->vdev_ops->vdev_op_leaf) {
617		nvlist_t **child;
618		int c, idx;
619
620		ASSERT(!vd->vdev_ishole);
621
622		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
623		    KM_SLEEP);
624
625		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
626			vdev_t *cvd = vd->vdev_child[c];
627
628			/*
629			 * If we're generating an nvlist of removing
630			 * vdevs then skip over any device which is
631			 * not being removed.
632			 */
633			if ((flags & VDEV_CONFIG_REMOVING) &&
634			    !cvd->vdev_removing)
635				continue;
636
637			child[idx++] = vdev_config_generate(spa, cvd,
638			    getstats, flags);
639		}
640
641		if (idx) {
642			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
643			    child, idx);
644		}
645
646		for (c = 0; c < idx; c++)
647			nvlist_free(child[c]);
648
649		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
650
651	} else {
652		const char *aux = NULL;
653
654		if (vd->vdev_offline && !vd->vdev_tmpoffline)
655			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
656		if (vd->vdev_resilver_txg != 0)
657			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
658			    vd->vdev_resilver_txg);
659		if (vd->vdev_faulted)
660			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
661		if (vd->vdev_degraded)
662			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
663		if (vd->vdev_removed)
664			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
665		if (vd->vdev_unspare)
666			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
667		if (vd->vdev_ishole)
668			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
669
670		switch (vd->vdev_stat.vs_aux) {
671		case VDEV_AUX_ERR_EXCEEDED:
672			aux = "err_exceeded";
673			break;
674
675		case VDEV_AUX_EXTERNAL:
676			aux = "external";
677			break;
678		}
679
680		if (aux != NULL)
681			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
682
683		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
684			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
685			    vd->vdev_orig_guid);
686		}
687	}
688
689	return (nv);
690}
691
692/*
693 * Generate a view of the top-level vdevs.  If we currently have holes
694 * in the namespace, then generate an array which contains a list of holey
695 * vdevs.  Additionally, add the number of top-level children that currently
696 * exist.
697 */
698void
699vdev_top_config_generate(spa_t *spa, nvlist_t *config)
700{
701	vdev_t *rvd = spa->spa_root_vdev;
702	uint64_t *array;
703	uint_t c, idx;
704
705	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
706
707	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
708		vdev_t *tvd = rvd->vdev_child[c];
709
710		if (tvd->vdev_ishole) {
711			array[idx++] = c;
712		}
713	}
714
715	if (idx) {
716		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
717		    array, idx) == 0);
718	}
719
720	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
721	    rvd->vdev_children) == 0);
722
723	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
724}
725
726/*
727 * Returns the configuration from the label of the given vdev. For vdevs
728 * which don't have a txg value stored on their label (i.e. spares/cache)
729 * or have not been completely initialized (txg = 0) just return
730 * the configuration from the first valid label we find. Otherwise,
731 * find the most up-to-date label that does not exceed the specified
732 * 'txg' value.
733 */
734nvlist_t *
735vdev_label_read_config(vdev_t *vd, uint64_t txg)
736{
737	spa_t *spa = vd->vdev_spa;
738	nvlist_t *config = NULL;
739	vdev_phys_t *vp;
740	abd_t *vp_abd;
741	zio_t *zio;
742	uint64_t best_txg = 0;
743	uint64_t label_txg = 0;
744	int error = 0;
745	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
746	    ZIO_FLAG_SPECULATIVE;
747
748	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
749
750	if (!vdev_readable(vd))
751		return (NULL);
752
753	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
754	vp = abd_to_buf(vp_abd);
755
756retry:
757	for (int l = 0; l < VDEV_LABELS; l++) {
758		nvlist_t *label = NULL;
759
760		zio = zio_root(spa, NULL, NULL, flags);
761
762		vdev_label_read(zio, vd, l, vp_abd,
763		    offsetof(vdev_label_t, vl_vdev_phys),
764		    sizeof (vdev_phys_t), NULL, NULL, flags);
765
766		if (zio_wait(zio) == 0 &&
767		    nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
768		    &label, 0) == 0) {
769			/*
770			 * Auxiliary vdevs won't have txg values in their
771			 * labels and newly added vdevs may not have been
772			 * completely initialized so just return the
773			 * configuration from the first valid label we
774			 * encounter.
775			 */
776			error = nvlist_lookup_uint64(label,
777			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
778			if ((error || label_txg == 0) && !config) {
779				config = label;
780				break;
781			} else if (label_txg <= txg && label_txg > best_txg) {
782				best_txg = label_txg;
783				nvlist_free(config);
784				config = fnvlist_dup(label);
785			}
786		}
787
788		if (label != NULL) {
789			nvlist_free(label);
790			label = NULL;
791		}
792	}
793
794	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
795		flags |= ZIO_FLAG_TRYHARD;
796		goto retry;
797	}
798
799	/*
800	 * We found a valid label but it didn't pass txg restrictions.
801	 */
802	if (config == NULL && label_txg != 0) {
803		vdev_dbgmsg(vd, "label discarded as txg is too large "
804		    "(%llu > %llu)", (u_longlong_t)label_txg,
805		    (u_longlong_t)txg);
806	}
807
808	abd_free(vp_abd);
809
810	return (config);
811}
812
813/*
814 * Determine if a device is in use.  The 'spare_guid' parameter will be filled
815 * in with the device guid if this spare is active elsewhere on the system.
816 */
817static boolean_t
818vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
819    uint64_t *spare_guid, uint64_t *l2cache_guid)
820{
821	spa_t *spa = vd->vdev_spa;
822	uint64_t state, pool_guid, device_guid, txg, spare_pool;
823	uint64_t vdtxg = 0;
824	nvlist_t *label;
825
826	if (spare_guid)
827		*spare_guid = 0ULL;
828	if (l2cache_guid)
829		*l2cache_guid = 0ULL;
830
831	/*
832	 * Read the label, if any, and perform some basic sanity checks.
833	 */
834	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
835		return (B_FALSE);
836
837	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
838	    &vdtxg);
839
840	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
841	    &state) != 0 ||
842	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
843	    &device_guid) != 0) {
844		nvlist_free(label);
845		return (B_FALSE);
846	}
847
848	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
849	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
850	    &pool_guid) != 0 ||
851	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
852	    &txg) != 0)) {
853		nvlist_free(label);
854		return (B_FALSE);
855	}
856
857	nvlist_free(label);
858
859	/*
860	 * Check to see if this device indeed belongs to the pool it claims to
861	 * be a part of.  The only way this is allowed is if the device is a hot
862	 * spare (which we check for later on).
863	 */
864	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
865	    !spa_guid_exists(pool_guid, device_guid) &&
866	    !spa_spare_exists(device_guid, NULL, NULL) &&
867	    !spa_l2cache_exists(device_guid, NULL))
868		return (B_FALSE);
869
870	/*
871	 * If the transaction group is zero, then this an initialized (but
872	 * unused) label.  This is only an error if the create transaction
873	 * on-disk is the same as the one we're using now, in which case the
874	 * user has attempted to add the same vdev multiple times in the same
875	 * transaction.
876	 */
877	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
878	    txg == 0 && vdtxg == crtxg)
879		return (B_TRUE);
880
881	/*
882	 * Check to see if this is a spare device.  We do an explicit check for
883	 * spa_has_spare() here because it may be on our pending list of spares
884	 * to add.  We also check if it is an l2cache device.
885	 */
886	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
887	    spa_has_spare(spa, device_guid)) {
888		if (spare_guid)
889			*spare_guid = device_guid;
890
891		switch (reason) {
892		case VDEV_LABEL_CREATE:
893		case VDEV_LABEL_L2CACHE:
894			return (B_TRUE);
895
896		case VDEV_LABEL_REPLACE:
897			return (!spa_has_spare(spa, device_guid) ||
898			    spare_pool != 0ULL);
899
900		case VDEV_LABEL_SPARE:
901			return (spa_has_spare(spa, device_guid));
902		}
903	}
904
905	/*
906	 * Check to see if this is an l2cache device.
907	 */
908	if (spa_l2cache_exists(device_guid, NULL))
909		return (B_TRUE);
910
911	/*
912	 * We can't rely on a pool's state if it's been imported
913	 * read-only.  Instead we look to see if the pools is marked
914	 * read-only in the namespace and set the state to active.
915	 */
916	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
917	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
918	    spa_mode(spa) == FREAD)
919		state = POOL_STATE_ACTIVE;
920
921	/*
922	 * If the device is marked ACTIVE, then this device is in use by another
923	 * pool on the system.
924	 */
925	return (state == POOL_STATE_ACTIVE);
926}
927
928/*
929 * Initialize a vdev label.  We check to make sure each leaf device is not in
930 * use, and writable.  We put down an initial label which we will later
931 * overwrite with a complete label.  Note that it's important to do this
932 * sequentially, not in parallel, so that we catch cases of multiple use of the
933 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
934 * itself.
935 */
936int
937vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
938{
939	spa_t *spa = vd->vdev_spa;
940	nvlist_t *label;
941	vdev_phys_t *vp;
942	abd_t *vp_abd;
943	abd_t *pad2;
944	uberblock_t *ub;
945	abd_t *ub_abd;
946	zio_t *zio;
947	char *buf;
948	size_t buflen;
949	int error;
950	uint64_t spare_guid, l2cache_guid;
951	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
952
953	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
954
955	for (int c = 0; c < vd->vdev_children; c++)
956		if ((error = vdev_label_init(vd->vdev_child[c],
957		    crtxg, reason)) != 0)
958			return (error);
959
960	/* Track the creation time for this vdev */
961	vd->vdev_crtxg = crtxg;
962
963	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
964		return (0);
965
966	/*
967	 * Dead vdevs cannot be initialized.
968	 */
969	if (vdev_is_dead(vd))
970		return (SET_ERROR(EIO));
971
972	/*
973	 * Determine if the vdev is in use.
974	 */
975	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
976	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
977		return (SET_ERROR(EBUSY));
978
979	/*
980	 * If this is a request to add or replace a spare or l2cache device
981	 * that is in use elsewhere on the system, then we must update the
982	 * guid (which was initialized to a random value) to reflect the
983	 * actual GUID (which is shared between multiple pools).
984	 */
985	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
986	    spare_guid != 0ULL) {
987		uint64_t guid_delta = spare_guid - vd->vdev_guid;
988
989		vd->vdev_guid += guid_delta;
990
991		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
992			pvd->vdev_guid_sum += guid_delta;
993
994		/*
995		 * If this is a replacement, then we want to fallthrough to the
996		 * rest of the code.  If we're adding a spare, then it's already
997		 * labeled appropriately and we can just return.
998		 */
999		if (reason == VDEV_LABEL_SPARE)
1000			return (0);
1001		ASSERT(reason == VDEV_LABEL_REPLACE ||
1002		    reason == VDEV_LABEL_SPLIT);
1003	}
1004
1005	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1006	    l2cache_guid != 0ULL) {
1007		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1008
1009		vd->vdev_guid += guid_delta;
1010
1011		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1012			pvd->vdev_guid_sum += guid_delta;
1013
1014		/*
1015		 * If this is a replacement, then we want to fallthrough to the
1016		 * rest of the code.  If we're adding an l2cache, then it's
1017		 * already labeled appropriately and we can just return.
1018		 */
1019		if (reason == VDEV_LABEL_L2CACHE)
1020			return (0);
1021		ASSERT(reason == VDEV_LABEL_REPLACE);
1022	}
1023
1024	/*
1025	 * Initialize its label.
1026	 */
1027	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1028	abd_zero(vp_abd, sizeof (vdev_phys_t));
1029	vp = abd_to_buf(vp_abd);
1030
1031	/*
1032	 * Generate a label describing the pool and our top-level vdev.
1033	 * We mark it as being from txg 0 to indicate that it's not
1034	 * really part of an active pool just yet.  The labels will
1035	 * be written again with a meaningful txg by spa_sync().
1036	 */
1037	if (reason == VDEV_LABEL_SPARE ||
1038	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1039		/*
1040		 * For inactive hot spares, we generate a special label that
1041		 * identifies as a mutually shared hot spare.  We write the
1042		 * label if we are adding a hot spare, or if we are removing an
1043		 * active hot spare (in which case we want to revert the
1044		 * labels).
1045		 */
1046		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1047
1048		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1049		    spa_version(spa)) == 0);
1050		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1051		    POOL_STATE_SPARE) == 0);
1052		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1053		    vd->vdev_guid) == 0);
1054	} else if (reason == VDEV_LABEL_L2CACHE ||
1055	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1056		/*
1057		 * For level 2 ARC devices, add a special label.
1058		 */
1059		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1060
1061		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1062		    spa_version(spa)) == 0);
1063		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1064		    POOL_STATE_L2CACHE) == 0);
1065		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1066		    vd->vdev_guid) == 0);
1067	} else {
1068		uint64_t txg = 0ULL;
1069
1070		if (reason == VDEV_LABEL_SPLIT)
1071			txg = spa->spa_uberblock.ub_txg;
1072		label = spa_config_generate(spa, vd, txg, B_FALSE);
1073
1074		/*
1075		 * Add our creation time.  This allows us to detect multiple
1076		 * vdev uses as described above, and automatically expires if we
1077		 * fail.
1078		 */
1079		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1080		    crtxg) == 0);
1081	}
1082
1083	buf = vp->vp_nvlist;
1084	buflen = sizeof (vp->vp_nvlist);
1085
1086	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1087	if (error != 0) {
1088		nvlist_free(label);
1089		abd_free(vp_abd);
1090		/* EFAULT means nvlist_pack ran out of room */
1091		return (error == EFAULT ? ENAMETOOLONG : EINVAL);
1092	}
1093
1094	/*
1095	 * Initialize uberblock template.
1096	 */
1097	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1098	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1099	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1100	ub = abd_to_buf(ub_abd);
1101	ub->ub_txg = 0;
1102
1103	/* Initialize the 2nd padding area. */
1104	pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1105	abd_zero(pad2, VDEV_PAD_SIZE);
1106
1107	/*
1108	 * Write everything in parallel.
1109	 */
1110retry:
1111	zio = zio_root(spa, NULL, NULL, flags);
1112
1113	for (int l = 0; l < VDEV_LABELS; l++) {
1114
1115		vdev_label_write(zio, vd, l, vp_abd,
1116		    offsetof(vdev_label_t, vl_vdev_phys),
1117		    sizeof (vdev_phys_t), NULL, NULL, flags);
1118
1119		/*
1120		 * Skip the 1st padding area.
1121		 * Zero out the 2nd padding area where it might have
1122		 * left over data from previous filesystem format.
1123		 */
1124		vdev_label_write(zio, vd, l, pad2,
1125		    offsetof(vdev_label_t, vl_pad2),
1126		    VDEV_PAD_SIZE, NULL, NULL, flags);
1127
1128		vdev_label_write(zio, vd, l, ub_abd,
1129		    offsetof(vdev_label_t, vl_uberblock),
1130		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1131	}
1132
1133	error = zio_wait(zio);
1134
1135	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1136		flags |= ZIO_FLAG_TRYHARD;
1137		goto retry;
1138	}
1139
1140	nvlist_free(label);
1141	abd_free(pad2);
1142	abd_free(ub_abd);
1143	abd_free(vp_abd);
1144
1145	/*
1146	 * If this vdev hasn't been previously identified as a spare, then we
1147	 * mark it as such only if a) we are labeling it as a spare, or b) it
1148	 * exists as a spare elsewhere in the system.  Do the same for
1149	 * level 2 ARC devices.
1150	 */
1151	if (error == 0 && !vd->vdev_isspare &&
1152	    (reason == VDEV_LABEL_SPARE ||
1153	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1154		spa_spare_add(vd);
1155
1156	if (error == 0 && !vd->vdev_isl2cache &&
1157	    (reason == VDEV_LABEL_L2CACHE ||
1158	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1159		spa_l2cache_add(vd);
1160
1161	return (error);
1162}
1163
1164/*
1165 * ==========================================================================
1166 * uberblock load/sync
1167 * ==========================================================================
1168 */
1169
1170/*
1171 * Consider the following situation: txg is safely synced to disk.  We've
1172 * written the first uberblock for txg + 1, and then we lose power.  When we
1173 * come back up, we fail to see the uberblock for txg + 1 because, say,
1174 * it was on a mirrored device and the replica to which we wrote txg + 1
1175 * is now offline.  If we then make some changes and sync txg + 1, and then
1176 * the missing replica comes back, then for a few seconds we'll have two
1177 * conflicting uberblocks on disk with the same txg.  The solution is simple:
1178 * among uberblocks with equal txg, choose the one with the latest timestamp.
1179 */
1180static int
1181vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1182{
1183	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1184
1185	if (likely(cmp))
1186		return (cmp);
1187
1188	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1189	if (likely(cmp))
1190		return (cmp);
1191
1192	/*
1193	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1194	 * ZFS, e.g. zfsonlinux >= 0.7.
1195	 *
1196	 * If one ub has MMP and the other does not, they were written by
1197	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1198	 * a 0 value.
1199	 *
1200	 * Since timestamp and txg are the same if we get this far, either is
1201	 * acceptable for importing the pool.
1202	 */
1203	unsigned int seq1 = 0;
1204	unsigned int seq2 = 0;
1205
1206	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1207		seq1 = MMP_SEQ(ub1);
1208
1209	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1210		seq2 = MMP_SEQ(ub2);
1211
1212	return (TREE_CMP(seq1, seq2));
1213}
1214
1215struct ubl_cbdata {
1216	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1217	vdev_t		*ubl_vd;	/* vdev associated with the above */
1218};
1219
1220static void
1221vdev_uberblock_load_done(zio_t *zio)
1222{
1223	vdev_t *vd = zio->io_vd;
1224	spa_t *spa = zio->io_spa;
1225	zio_t *rio = zio->io_private;
1226	uberblock_t *ub = abd_to_buf(zio->io_abd);
1227	struct ubl_cbdata *cbp = rio->io_private;
1228
1229	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1230
1231	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1232		mutex_enter(&rio->io_lock);
1233		if (ub->ub_txg <= spa->spa_load_max_txg &&
1234		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1235			/*
1236			 * Keep track of the vdev in which this uberblock
1237			 * was found. We will use this information later
1238			 * to obtain the config nvlist associated with
1239			 * this uberblock.
1240			 */
1241			*cbp->ubl_ubbest = *ub;
1242			cbp->ubl_vd = vd;
1243		}
1244		mutex_exit(&rio->io_lock);
1245	}
1246
1247	abd_free(zio->io_abd);
1248}
1249
1250static void
1251vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1252    struct ubl_cbdata *cbp)
1253{
1254	for (int c = 0; c < vd->vdev_children; c++)
1255		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1256
1257	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1258		for (int l = 0; l < VDEV_LABELS; l++) {
1259			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1260				vdev_label_read(zio, vd, l,
1261				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1262				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1263				    VDEV_UBERBLOCK_SIZE(vd),
1264				    vdev_uberblock_load_done, zio, flags);
1265			}
1266		}
1267	}
1268}
1269
1270/*
1271 * Reads the 'best' uberblock from disk along with its associated
1272 * configuration. First, we read the uberblock array of each label of each
1273 * vdev, keeping track of the uberblock with the highest txg in each array.
1274 * Then, we read the configuration from the same vdev as the best uberblock.
1275 */
1276void
1277vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1278{
1279	zio_t *zio;
1280	spa_t *spa = rvd->vdev_spa;
1281	struct ubl_cbdata cb;
1282	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1283	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1284
1285	ASSERT(ub);
1286	ASSERT(config);
1287
1288	bzero(ub, sizeof (uberblock_t));
1289	*config = NULL;
1290
1291	cb.ubl_ubbest = ub;
1292	cb.ubl_vd = NULL;
1293
1294	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1295	zio = zio_root(spa, NULL, &cb, flags);
1296	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1297	(void) zio_wait(zio);
1298
1299	/*
1300	 * It's possible that the best uberblock was discovered on a label
1301	 * that has a configuration which was written in a future txg.
1302	 * Search all labels on this vdev to find the configuration that
1303	 * matches the txg for our uberblock.
1304	 */
1305	if (cb.ubl_vd != NULL) {
1306		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1307		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1308
1309		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1310		if (*config == NULL && spa->spa_extreme_rewind) {
1311			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1312			    "Trying again without txg restrictions.");
1313			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1314		}
1315		if (*config == NULL) {
1316			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1317		}
1318	}
1319	spa_config_exit(spa, SCL_ALL, FTAG);
1320}
1321
1322/*
1323 * On success, increment root zio's count of good writes.
1324 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1325 */
1326static void
1327vdev_uberblock_sync_done(zio_t *zio)
1328{
1329	uint64_t *good_writes = zio->io_private;
1330
1331	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1332		atomic_inc_64(good_writes);
1333}
1334
1335/*
1336 * Write the uberblock to all labels of all leaves of the specified vdev.
1337 */
1338static void
1339vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1340    uberblock_t *ub, vdev_t *vd, int flags)
1341{
1342	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1343		vdev_uberblock_sync(zio, good_writes,
1344		    ub, vd->vdev_child[c], flags);
1345	}
1346
1347	if (!vd->vdev_ops->vdev_op_leaf)
1348		return;
1349
1350	if (!vdev_writeable(vd))
1351		return;
1352
1353	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1354	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1355
1356	/* Copy the uberblock_t into the ABD */
1357	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1358	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1359	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1360
1361	for (int l = 0; l < VDEV_LABELS; l++)
1362		vdev_label_write(zio, vd, l, ub_abd,
1363		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1364		    vdev_uberblock_sync_done, good_writes,
1365		    flags | ZIO_FLAG_DONT_PROPAGATE);
1366
1367	abd_free(ub_abd);
1368}
1369
1370/* Sync the uberblocks to all vdevs in svd[] */
1371int
1372vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1373{
1374	spa_t *spa = svd[0]->vdev_spa;
1375	zio_t *zio;
1376	uint64_t good_writes = 0;
1377
1378	zio = zio_root(spa, NULL, NULL, flags);
1379
1380	for (int v = 0; v < svdcount; v++)
1381		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1382
1383	(void) zio_wait(zio);
1384
1385	/*
1386	 * Flush the uberblocks to disk.  This ensures that the odd labels
1387	 * are no longer needed (because the new uberblocks and the even
1388	 * labels are safely on disk), so it is safe to overwrite them.
1389	 */
1390	zio = zio_root(spa, NULL, NULL, flags);
1391
1392	for (int v = 0; v < svdcount; v++) {
1393		if (vdev_writeable(svd[v])) {
1394			zio_flush(zio, svd[v]);
1395		}
1396	}
1397
1398	(void) zio_wait(zio);
1399
1400	return (good_writes >= 1 ? 0 : EIO);
1401}
1402
1403/*
1404 * On success, increment the count of good writes for our top-level vdev.
1405 */
1406static void
1407vdev_label_sync_done(zio_t *zio)
1408{
1409	uint64_t *good_writes = zio->io_private;
1410
1411	if (zio->io_error == 0)
1412		atomic_inc_64(good_writes);
1413}
1414
1415/*
1416 * If there weren't enough good writes, indicate failure to the parent.
1417 */
1418static void
1419vdev_label_sync_top_done(zio_t *zio)
1420{
1421	uint64_t *good_writes = zio->io_private;
1422
1423	if (*good_writes == 0)
1424		zio->io_error = SET_ERROR(EIO);
1425
1426	kmem_free(good_writes, sizeof (uint64_t));
1427}
1428
1429/*
1430 * We ignore errors for log and cache devices, simply free the private data.
1431 */
1432static void
1433vdev_label_sync_ignore_done(zio_t *zio)
1434{
1435	kmem_free(zio->io_private, sizeof (uint64_t));
1436}
1437
1438/*
1439 * Write all even or odd labels to all leaves of the specified vdev.
1440 */
1441static void
1442vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1443    vdev_t *vd, int l, uint64_t txg, int flags)
1444{
1445	nvlist_t *label;
1446	vdev_phys_t *vp;
1447	abd_t *vp_abd;
1448	char *buf;
1449	size_t buflen;
1450
1451	for (int c = 0; c < vd->vdev_children; c++) {
1452		vdev_label_sync(zio, good_writes,
1453		    vd->vdev_child[c], l, txg, flags);
1454	}
1455
1456	if (!vd->vdev_ops->vdev_op_leaf)
1457		return;
1458
1459	if (!vdev_writeable(vd))
1460		return;
1461
1462	/*
1463	 * Generate a label describing the top-level config to which we belong.
1464	 */
1465	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1466
1467	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1468	abd_zero(vp_abd, sizeof (vdev_phys_t));
1469	vp = abd_to_buf(vp_abd);
1470
1471	buf = vp->vp_nvlist;
1472	buflen = sizeof (vp->vp_nvlist);
1473
1474	if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1475		for (; l < VDEV_LABELS; l += 2) {
1476			vdev_label_write(zio, vd, l, vp_abd,
1477			    offsetof(vdev_label_t, vl_vdev_phys),
1478			    sizeof (vdev_phys_t),
1479			    vdev_label_sync_done, good_writes,
1480			    flags | ZIO_FLAG_DONT_PROPAGATE);
1481		}
1482	}
1483
1484	abd_free(vp_abd);
1485	nvlist_free(label);
1486}
1487
1488int
1489vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1490{
1491	list_t *dl = &spa->spa_config_dirty_list;
1492	vdev_t *vd;
1493	zio_t *zio;
1494	int error;
1495
1496	/*
1497	 * Write the new labels to disk.
1498	 */
1499	zio = zio_root(spa, NULL, NULL, flags);
1500
1501	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1502		uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1503		    KM_SLEEP);
1504
1505		ASSERT(!vd->vdev_ishole);
1506
1507		zio_t *vio = zio_null(zio, spa, NULL,
1508		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1509		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1510		    good_writes, flags);
1511		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1512		zio_nowait(vio);
1513	}
1514
1515	error = zio_wait(zio);
1516
1517	/*
1518	 * Flush the new labels to disk.
1519	 */
1520	zio = zio_root(spa, NULL, NULL, flags);
1521
1522	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1523		zio_flush(zio, vd);
1524
1525	(void) zio_wait(zio);
1526
1527	return (error);
1528}
1529
1530/*
1531 * Sync the uberblock and any changes to the vdev configuration.
1532 *
1533 * The order of operations is carefully crafted to ensure that
1534 * if the system panics or loses power at any time, the state on disk
1535 * is still transactionally consistent.  The in-line comments below
1536 * describe the failure semantics at each stage.
1537 *
1538 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1539 * at any time, you can just call it again, and it will resume its work.
1540 */
1541int
1542vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1543{
1544	spa_t *spa = svd[0]->vdev_spa;
1545	uberblock_t *ub = &spa->spa_uberblock;
1546	int error = 0;
1547	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1548
1549	ASSERT(svdcount != 0);
1550retry:
1551	/*
1552	 * Normally, we don't want to try too hard to write every label and
1553	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1554	 * sync process to block while we retry.  But if we can't write a
1555	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1556	 * bailing out and declaring the pool faulted.
1557	 */
1558	if (error != 0) {
1559		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1560			return (error);
1561		flags |= ZIO_FLAG_TRYHARD;
1562	}
1563
1564	ASSERT(ub->ub_txg <= txg);
1565
1566	/*
1567	 * If this isn't a resync due to I/O errors,
1568	 * and nothing changed in this transaction group,
1569	 * and the vdev configuration hasn't changed,
1570	 * then there's nothing to do.
1571	 */
1572	if (ub->ub_txg < txg) {
1573		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1574		    txg, spa->spa_mmp.mmp_delay);
1575
1576		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1577			return (0);
1578	}
1579
1580	if (txg > spa_freeze_txg(spa))
1581		return (0);
1582
1583	ASSERT(txg <= spa->spa_final_txg);
1584
1585	/*
1586	 * Flush the write cache of every disk that's been written to
1587	 * in this transaction group.  This ensures that all blocks
1588	 * written in this txg will be committed to stable storage
1589	 * before any uberblock that references them.
1590	 */
1591	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1592
1593	for (vdev_t *vd =
1594	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1595	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1596		zio_flush(zio, vd);
1597
1598	(void) zio_wait(zio);
1599
1600	/*
1601	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1602	 * system dies in the middle of this process, that's OK: all of the
1603	 * even labels that made it to disk will be newer than any uberblock,
1604	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1605	 * which have not yet been touched, will still be valid.  We flush
1606	 * the new labels to disk to ensure that all even-label updates
1607	 * are committed to stable storage before the uberblock update.
1608	 */
1609	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1610		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1611			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1612			    "for pool '%s' when syncing out the even labels "
1613			    "of dirty vdevs", error, spa_name(spa));
1614		}
1615		goto retry;
1616	}
1617
1618	/*
1619	 * Sync the uberblocks to all vdevs in svd[].
1620	 * If the system dies in the middle of this step, there are two cases
1621	 * to consider, and the on-disk state is consistent either way:
1622	 *
1623	 * (1)	If none of the new uberblocks made it to disk, then the
1624	 *	previous uberblock will be the newest, and the odd labels
1625	 *	(which had not yet been touched) will be valid with respect
1626	 *	to that uberblock.
1627	 *
1628	 * (2)	If one or more new uberblocks made it to disk, then they
1629	 *	will be the newest, and the even labels (which had all
1630	 *	been successfully committed) will be valid with respect
1631	 *	to the new uberblocks.
1632	 */
1633	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1634		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1635			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1636			    "%d for pool '%s'", error, spa_name(spa));
1637		}
1638		goto retry;
1639	}
1640
1641	if (spa_multihost(spa))
1642		mmp_update_uberblock(spa, ub);
1643
1644	/*
1645	 * Sync out odd labels for every dirty vdev.  If the system dies
1646	 * in the middle of this process, the even labels and the new
1647	 * uberblocks will suffice to open the pool.  The next time
1648	 * the pool is opened, the first thing we'll do -- before any
1649	 * user data is modified -- is mark every vdev dirty so that
1650	 * all labels will be brought up to date.  We flush the new labels
1651	 * to disk to ensure that all odd-label updates are committed to
1652	 * stable storage before the next transaction group begins.
1653	 */
1654	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1655		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1656			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1657			    "for pool '%s' when syncing out the odd labels of "
1658			    "dirty vdevs", error, spa_name(spa));
1659		}
1660		goto retry;
1661	}
1662
1663	return (0);
1664}
1665