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.
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.
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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Intel Corporation.
29 * Virtual Device Labels
30 * ---------------------
32 * The vdev label serves several distinct purposes:
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
37 * 2. Verify that all the devices given in a configuration are present
40 * 3. Determine the uberblock for the pool.
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transaction groups:
68 * +------+ +------+ +------+
70 * | t10 | | t10 | | t10 |
72 * +------+ +------+ +------+
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
78 * In order to identify which labels are valid, the labels are written in the
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
116 * Configuration Information
117 * -------------------------
119 * The nvlist describing the pool and vdev contains the following elements:
121 * version ZFS on-disk version
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
128 * An nvlist of the features necessary for reading the MOS.
130 * Each leaf device label also contains the following:
132 * top_guid Unique ID for top-level vdev in which this is contained
133 * guid Unique ID for the leaf vdev
135 * The 'vs' configuration follows the format described in 'spa_config.c'.
138 #include <sys/zfs_context.h>
140 #include <sys/spa_impl.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/uberblock_impl.h>
146 #include <sys/metaslab.h>
147 #include <sys/metaslab_impl.h>
149 #include <sys/dsl_scan.h>
151 #include <sys/fs/zfs.h>
154 * Basic routines to read and write from a vdev label.
155 * Used throughout the rest of this file.
158 vdev_label_offset(uint64_t psize
, int l
, uint64_t offset
)
160 ASSERT(offset
< sizeof (vdev_label_t
));
161 ASSERT(P2PHASE_TYPED(psize
, sizeof (vdev_label_t
), uint64_t) == 0);
163 return (offset
+ l
* sizeof (vdev_label_t
) + (l
< VDEV_LABELS
/ 2 ?
164 0 : psize
- VDEV_LABELS
* sizeof (vdev_label_t
)));
168 * Returns back the vdev label associated with the passed in offset.
171 vdev_label_number(uint64_t psize
, uint64_t offset
)
175 if (offset
>= psize
- VDEV_LABEL_END_SIZE
) {
176 offset
-= psize
- VDEV_LABEL_END_SIZE
;
177 offset
+= (VDEV_LABELS
/ 2) * sizeof (vdev_label_t
);
179 l
= offset
/ sizeof (vdev_label_t
);
180 return (l
< VDEV_LABELS
? l
: -1);
184 vdev_label_read(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
185 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
188 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
189 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
190 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
192 zio_nowait(zio_read_phys(zio
, vd
,
193 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
194 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
195 ZIO_PRIORITY_SYNC_READ
, flags
, B_TRUE
));
199 vdev_label_write(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
200 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
203 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
204 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
205 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
207 zio_nowait(zio_write_phys(zio
, vd
,
208 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
209 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
210 ZIO_PRIORITY_SYNC_WRITE
, flags
, B_TRUE
));
214 * Generate the nvlist representing this vdev's stats
217 vdev_config_generate_stats(vdev_t
*vd
, nvlist_t
*nv
)
223 vs
= kmem_alloc(sizeof (*vs
), KM_SLEEP
);
224 vsx
= kmem_alloc(sizeof (*vsx
), KM_SLEEP
);
226 vdev_get_stats_ex(vd
, vs
, vsx
);
227 fnvlist_add_uint64_array(nv
, ZPOOL_CONFIG_VDEV_STATS
,
228 (uint64_t *)vs
, sizeof (*vs
) / sizeof (uint64_t));
231 * Add extended stats into a special extended stats nvlist. This keeps
232 * all the extended stats nicely grouped together. The extended stats
233 * nvlist is then added to the main nvlist.
235 nvx
= fnvlist_alloc();
237 /* ZIOs in flight to disk */
238 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE
,
239 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_READ
]);
241 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE
,
242 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
244 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE
,
245 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_READ
]);
247 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE
,
248 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
250 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE
,
251 vsx
->vsx_active_queue
[ZIO_PRIORITY_SCRUB
]);
253 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE
,
254 vsx
->vsx_active_queue
[ZIO_PRIORITY_TRIM
]);
257 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE
,
258 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_READ
]);
260 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE
,
261 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
263 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE
,
264 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_READ
]);
266 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE
,
267 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
269 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE
,
270 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SCRUB
]);
272 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE
,
273 vsx
->vsx_pend_queue
[ZIO_PRIORITY_TRIM
]);
276 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO
,
277 vsx
->vsx_total_histo
[ZIO_TYPE_READ
],
278 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_READ
]));
280 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO
,
281 vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
],
282 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
]));
284 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO
,
285 vsx
->vsx_disk_histo
[ZIO_TYPE_READ
],
286 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_READ
]));
288 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO
,
289 vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
],
290 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
]));
292 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO
,
293 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
],
294 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
]));
296 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO
,
297 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
],
298 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
300 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO
,
301 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
],
302 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
]));
304 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO
,
305 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
306 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
308 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO
,
309 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
],
310 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
]));
312 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO
,
313 vsx
->vsx_queue_histo
[ZIO_PRIORITY_TRIM
],
314 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_TRIM
]));
317 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO
,
318 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
],
319 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
]));
321 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO
,
322 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
],
323 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
325 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO
,
326 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
],
327 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
]));
329 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO
,
330 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
331 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
333 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO
,
334 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
],
335 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
]));
337 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO
,
338 vsx
->vsx_ind_histo
[ZIO_PRIORITY_TRIM
],
339 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_TRIM
]));
341 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO
,
342 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
],
343 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
]));
345 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO
,
346 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
],
347 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
349 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO
,
350 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
],
351 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
]));
353 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO
,
354 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
355 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
357 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO
,
358 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
],
359 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
]));
361 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO
,
362 vsx
->vsx_agg_histo
[ZIO_PRIORITY_TRIM
],
363 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_TRIM
]));
366 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SLOW_IOS
, vs
->vs_slow_ios
);
368 /* Add extended stats nvlist to main nvlist */
369 fnvlist_add_nvlist(nv
, ZPOOL_CONFIG_VDEV_STATS_EX
, nvx
);
372 kmem_free(vs
, sizeof (*vs
));
373 kmem_free(vsx
, sizeof (*vsx
));
377 root_vdev_actions_getprogress(vdev_t
*vd
, nvlist_t
*nvl
)
379 spa_t
*spa
= vd
->vdev_spa
;
381 if (vd
!= spa
->spa_root_vdev
)
384 /* provide either current or previous scan information */
386 if (spa_scan_get_stats(spa
, &ps
) == 0) {
387 fnvlist_add_uint64_array(nvl
,
388 ZPOOL_CONFIG_SCAN_STATS
, (uint64_t *)&ps
,
389 sizeof (pool_scan_stat_t
) / sizeof (uint64_t));
392 pool_removal_stat_t prs
;
393 if (spa_removal_get_stats(spa
, &prs
) == 0) {
394 fnvlist_add_uint64_array(nvl
,
395 ZPOOL_CONFIG_REMOVAL_STATS
, (uint64_t *)&prs
,
396 sizeof (prs
) / sizeof (uint64_t));
399 pool_checkpoint_stat_t pcs
;
400 if (spa_checkpoint_get_stats(spa
, &pcs
) == 0) {
401 fnvlist_add_uint64_array(nvl
,
402 ZPOOL_CONFIG_CHECKPOINT_STATS
, (uint64_t *)&pcs
,
403 sizeof (pcs
) / sizeof (uint64_t));
408 top_vdev_actions_getprogress(vdev_t
*vd
, nvlist_t
*nvl
)
410 if (vd
== vd
->vdev_top
) {
411 vdev_rebuild_stat_t vrs
;
412 if (vdev_rebuild_get_stats(vd
, &vrs
) == 0) {
413 fnvlist_add_uint64_array(nvl
,
414 ZPOOL_CONFIG_REBUILD_STATS
, (uint64_t *)&vrs
,
415 sizeof (vrs
) / sizeof (uint64_t));
421 * Generate the nvlist representing this vdev's config.
424 vdev_config_generate(spa_t
*spa
, vdev_t
*vd
, boolean_t getstats
,
425 vdev_config_flag_t flags
)
428 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
430 nv
= fnvlist_alloc();
432 fnvlist_add_string(nv
, ZPOOL_CONFIG_TYPE
, vd
->vdev_ops
->vdev_op_type
);
433 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)))
434 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ID
, vd
->vdev_id
);
435 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_GUID
, vd
->vdev_guid
);
437 if (vd
->vdev_path
!= NULL
)
438 fnvlist_add_string(nv
, ZPOOL_CONFIG_PATH
, vd
->vdev_path
);
440 if (vd
->vdev_devid
!= NULL
)
441 fnvlist_add_string(nv
, ZPOOL_CONFIG_DEVID
, vd
->vdev_devid
);
443 if (vd
->vdev_physpath
!= NULL
)
444 fnvlist_add_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
447 if (vd
->vdev_enc_sysfs_path
!= NULL
)
448 fnvlist_add_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
449 vd
->vdev_enc_sysfs_path
);
451 if (vd
->vdev_fru
!= NULL
)
452 fnvlist_add_string(nv
, ZPOOL_CONFIG_FRU
, vd
->vdev_fru
);
454 if (vd
->vdev_nparity
!= 0) {
455 ASSERT(strcmp(vd
->vdev_ops
->vdev_op_type
,
456 VDEV_TYPE_RAIDZ
) == 0);
459 * Make sure someone hasn't managed to sneak a fancy new vdev
460 * into a crufty old storage pool.
462 ASSERT(vd
->vdev_nparity
== 1 ||
463 (vd
->vdev_nparity
<= 2 &&
464 spa_version(spa
) >= SPA_VERSION_RAIDZ2
) ||
465 (vd
->vdev_nparity
<= 3 &&
466 spa_version(spa
) >= SPA_VERSION_RAIDZ3
));
469 * Note that we'll add the nparity tag even on storage pools
470 * that only support a single parity device -- older software
471 * will just ignore it.
473 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NPARITY
, vd
->vdev_nparity
);
476 if (vd
->vdev_wholedisk
!= -1ULL)
477 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
480 if (vd
->vdev_not_present
&& !(flags
& VDEV_CONFIG_MISSING
))
481 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
, 1);
483 if (vd
->vdev_isspare
)
484 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
, 1);
486 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)) &&
487 vd
== vd
->vdev_top
) {
488 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
490 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
492 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, vd
->vdev_ashift
);
493 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
495 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, vd
->vdev_islog
);
496 if (vd
->vdev_removing
) {
497 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
501 /* zpool command expects alloc class data */
502 if (getstats
&& vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
) {
503 const char *bias
= NULL
;
505 switch (vd
->vdev_alloc_bias
) {
507 bias
= VDEV_ALLOC_BIAS_LOG
;
509 case VDEV_BIAS_SPECIAL
:
510 bias
= VDEV_ALLOC_BIAS_SPECIAL
;
512 case VDEV_BIAS_DEDUP
:
513 bias
= VDEV_ALLOC_BIAS_DEDUP
;
516 ASSERT3U(vd
->vdev_alloc_bias
, ==,
519 fnvlist_add_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
524 if (vd
->vdev_dtl_sm
!= NULL
) {
525 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DTL
,
526 space_map_object(vd
->vdev_dtl_sm
));
529 if (vic
->vic_mapping_object
!= 0) {
530 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
531 vic
->vic_mapping_object
);
534 if (vic
->vic_births_object
!= 0) {
535 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
536 vic
->vic_births_object
);
539 if (vic
->vic_prev_indirect_vdev
!= UINT64_MAX
) {
540 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
541 vic
->vic_prev_indirect_vdev
);
545 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
, vd
->vdev_crtxg
);
547 if (vd
->vdev_expansion_time
)
548 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_EXPANSION_TIME
,
549 vd
->vdev_expansion_time
);
551 if (flags
& VDEV_CONFIG_MOS
) {
552 if (vd
->vdev_leaf_zap
!= 0) {
553 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
554 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_LEAF_ZAP
,
558 if (vd
->vdev_top_zap
!= 0) {
559 ASSERT(vd
== vd
->vdev_top
);
560 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
564 if (vd
->vdev_resilver_deferred
) {
565 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
566 ASSERT(spa
->spa_resilver_deferred
);
567 fnvlist_add_boolean(nv
, ZPOOL_CONFIG_RESILVER_DEFER
);
572 vdev_config_generate_stats(vd
, nv
);
574 root_vdev_actions_getprogress(vd
, nv
);
575 top_vdev_actions_getprogress(vd
, nv
);
578 * Note: this can be called from open context
579 * (spa_get_stats()), so we need the rwlock to prevent
580 * the mapping from being changed by condensing.
582 rw_enter(&vd
->vdev_indirect_rwlock
, RW_READER
);
583 if (vd
->vdev_indirect_mapping
!= NULL
) {
584 ASSERT(vd
->vdev_indirect_births
!= NULL
);
585 vdev_indirect_mapping_t
*vim
=
586 vd
->vdev_indirect_mapping
;
587 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
588 vdev_indirect_mapping_size(vim
));
590 rw_exit(&vd
->vdev_indirect_rwlock
);
591 if (vd
->vdev_mg
!= NULL
&&
592 vd
->vdev_mg
->mg_fragmentation
!= ZFS_FRAG_INVALID
) {
594 * Compute approximately how much memory would be used
595 * for the indirect mapping if this device were to
598 * Note: If the frag metric is invalid, then not
599 * enough metaslabs have been converted to have
602 uint64_t seg_count
= 0;
603 uint64_t to_alloc
= vd
->vdev_stat
.vs_alloc
;
606 * There are the same number of allocated segments
607 * as free segments, so we will have at least one
608 * entry per free segment. However, small free
609 * segments (smaller than vdev_removal_max_span)
610 * will be combined with adjacent allocated segments
611 * as a single mapping.
613 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
614 if (1ULL << (i
+ 1) < vdev_removal_max_span
) {
616 vd
->vdev_mg
->mg_histogram
[i
] <<
620 vd
->vdev_mg
->mg_histogram
[i
];
625 * The maximum length of a mapping is
626 * zfs_remove_max_segment, so we need at least one entry
627 * per zfs_remove_max_segment of allocated data.
629 seg_count
+= to_alloc
/ spa_remove_max_segment(spa
);
631 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
633 sizeof (vdev_indirect_mapping_entry_phys_t
));
637 if (!vd
->vdev_ops
->vdev_op_leaf
) {
641 ASSERT(!vd
->vdev_ishole
);
643 child
= kmem_alloc(vd
->vdev_children
* sizeof (nvlist_t
*),
646 for (c
= 0, idx
= 0; c
< vd
->vdev_children
; c
++) {
647 vdev_t
*cvd
= vd
->vdev_child
[c
];
650 * If we're generating an nvlist of removing
651 * vdevs then skip over any device which is
654 if ((flags
& VDEV_CONFIG_REMOVING
) &&
658 child
[idx
++] = vdev_config_generate(spa
, cvd
,
663 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
667 for (c
= 0; c
< idx
; c
++)
668 nvlist_free(child
[c
]);
670 kmem_free(child
, vd
->vdev_children
* sizeof (nvlist_t
*));
673 const char *aux
= NULL
;
675 if (vd
->vdev_offline
&& !vd
->vdev_tmpoffline
)
676 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_OFFLINE
, B_TRUE
);
677 if (vd
->vdev_resilver_txg
!= 0)
678 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
679 vd
->vdev_resilver_txg
);
680 if (vd
->vdev_rebuild_txg
!= 0)
681 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
682 vd
->vdev_rebuild_txg
);
683 if (vd
->vdev_faulted
)
684 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_FAULTED
, B_TRUE
);
685 if (vd
->vdev_degraded
)
686 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DEGRADED
, B_TRUE
);
687 if (vd
->vdev_removed
)
688 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVED
, B_TRUE
);
689 if (vd
->vdev_unspare
)
690 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_UNSPARE
, B_TRUE
);
692 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_HOLE
, B_TRUE
);
694 /* Set the reason why we're FAULTED/DEGRADED. */
695 switch (vd
->vdev_stat
.vs_aux
) {
696 case VDEV_AUX_ERR_EXCEEDED
:
697 aux
= "err_exceeded";
700 case VDEV_AUX_EXTERNAL
:
705 if (aux
!= NULL
&& !vd
->vdev_tmpoffline
) {
706 fnvlist_add_string(nv
, ZPOOL_CONFIG_AUX_STATE
, aux
);
709 * We're healthy - clear any previous AUX_STATE values.
711 if (nvlist_exists(nv
, ZPOOL_CONFIG_AUX_STATE
))
712 nvlist_remove_all(nv
, ZPOOL_CONFIG_AUX_STATE
);
715 if (vd
->vdev_splitting
&& vd
->vdev_orig_guid
!= 0LL) {
716 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ORIG_GUID
,
725 * Generate a view of the top-level vdevs. If we currently have holes
726 * in the namespace, then generate an array which contains a list of holey
727 * vdevs. Additionally, add the number of top-level children that currently
731 vdev_top_config_generate(spa_t
*spa
, nvlist_t
*config
)
733 vdev_t
*rvd
= spa
->spa_root_vdev
;
737 array
= kmem_alloc(rvd
->vdev_children
* sizeof (uint64_t), KM_SLEEP
);
739 for (c
= 0, idx
= 0; c
< rvd
->vdev_children
; c
++) {
740 vdev_t
*tvd
= rvd
->vdev_child
[c
];
742 if (tvd
->vdev_ishole
) {
748 VERIFY(nvlist_add_uint64_array(config
, ZPOOL_CONFIG_HOLE_ARRAY
,
752 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
753 rvd
->vdev_children
) == 0);
755 kmem_free(array
, rvd
->vdev_children
* sizeof (uint64_t));
759 * Returns the configuration from the label of the given vdev. For vdevs
760 * which don't have a txg value stored on their label (i.e. spares/cache)
761 * or have not been completely initialized (txg = 0) just return
762 * the configuration from the first valid label we find. Otherwise,
763 * find the most up-to-date label that does not exceed the specified
767 vdev_label_read_config(vdev_t
*vd
, uint64_t txg
)
769 spa_t
*spa
= vd
->vdev_spa
;
770 nvlist_t
*config
= NULL
;
774 uint64_t best_txg
= 0;
775 uint64_t label_txg
= 0;
777 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
778 ZIO_FLAG_SPECULATIVE
;
780 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
782 if (!vdev_readable(vd
))
785 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
786 vp
= abd_to_buf(vp_abd
);
789 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
790 nvlist_t
*label
= NULL
;
792 zio
= zio_root(spa
, NULL
, NULL
, flags
);
794 vdev_label_read(zio
, vd
, l
, vp_abd
,
795 offsetof(vdev_label_t
, vl_vdev_phys
),
796 sizeof (vdev_phys_t
), NULL
, NULL
, flags
);
798 if (zio_wait(zio
) == 0 &&
799 nvlist_unpack(vp
->vp_nvlist
, sizeof (vp
->vp_nvlist
),
802 * Auxiliary vdevs won't have txg values in their
803 * labels and newly added vdevs may not have been
804 * completely initialized so just return the
805 * configuration from the first valid label we
808 error
= nvlist_lookup_uint64(label
,
809 ZPOOL_CONFIG_POOL_TXG
, &label_txg
);
810 if ((error
|| label_txg
== 0) && !config
) {
813 } else if (label_txg
<= txg
&& label_txg
> best_txg
) {
814 best_txg
= label_txg
;
816 config
= fnvlist_dup(label
);
826 if (config
== NULL
&& !(flags
& ZIO_FLAG_TRYHARD
)) {
827 flags
|= ZIO_FLAG_TRYHARD
;
832 * We found a valid label but it didn't pass txg restrictions.
834 if (config
== NULL
&& label_txg
!= 0) {
835 vdev_dbgmsg(vd
, "label discarded as txg is too large "
836 "(%llu > %llu)", (u_longlong_t
)label_txg
,
846 * Determine if a device is in use. The 'spare_guid' parameter will be filled
847 * in with the device guid if this spare is active elsewhere on the system.
850 vdev_inuse(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
,
851 uint64_t *spare_guid
, uint64_t *l2cache_guid
)
853 spa_t
*spa
= vd
->vdev_spa
;
854 uint64_t state
, pool_guid
, device_guid
, txg
, spare_pool
;
861 *l2cache_guid
= 0ULL;
864 * Read the label, if any, and perform some basic sanity checks.
866 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
)
869 (void) nvlist_lookup_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
872 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
874 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
875 &device_guid
) != 0) {
880 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
881 (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
,
883 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
892 * Check to see if this device indeed belongs to the pool it claims to
893 * be a part of. The only way this is allowed is if the device is a hot
894 * spare (which we check for later on).
896 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
897 !spa_guid_exists(pool_guid
, device_guid
) &&
898 !spa_spare_exists(device_guid
, NULL
, NULL
) &&
899 !spa_l2cache_exists(device_guid
, NULL
))
903 * If the transaction group is zero, then this an initialized (but
904 * unused) label. This is only an error if the create transaction
905 * on-disk is the same as the one we're using now, in which case the
906 * user has attempted to add the same vdev multiple times in the same
909 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
910 txg
== 0 && vdtxg
== crtxg
)
914 * Check to see if this is a spare device. We do an explicit check for
915 * spa_has_spare() here because it may be on our pending list of spares
916 * to add. We also check if it is an l2cache device.
918 if (spa_spare_exists(device_guid
, &spare_pool
, NULL
) ||
919 spa_has_spare(spa
, device_guid
)) {
921 *spare_guid
= device_guid
;
924 case VDEV_LABEL_CREATE
:
925 case VDEV_LABEL_L2CACHE
:
928 case VDEV_LABEL_REPLACE
:
929 return (!spa_has_spare(spa
, device_guid
) ||
932 case VDEV_LABEL_SPARE
:
933 return (spa_has_spare(spa
, device_guid
));
940 * Check to see if this is an l2cache device.
942 if (spa_l2cache_exists(device_guid
, NULL
))
946 * We can't rely on a pool's state if it's been imported
947 * read-only. Instead we look to see if the pools is marked
948 * read-only in the namespace and set the state to active.
950 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
951 (spa
= spa_by_guid(pool_guid
, device_guid
)) != NULL
&&
952 spa_mode(spa
) == SPA_MODE_READ
)
953 state
= POOL_STATE_ACTIVE
;
956 * If the device is marked ACTIVE, then this device is in use by another
957 * pool on the system.
959 return (state
== POOL_STATE_ACTIVE
);
963 * Initialize a vdev label. We check to make sure each leaf device is not in
964 * use, and writable. We put down an initial label which we will later
965 * overwrite with a complete label. Note that it's important to do this
966 * sequentially, not in parallel, so that we catch cases of multiple use of the
967 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
971 vdev_label_init(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
)
973 spa_t
*spa
= vd
->vdev_spa
;
984 uint64_t spare_guid
= 0, l2cache_guid
= 0;
985 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
987 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
989 for (int c
= 0; c
< vd
->vdev_children
; c
++)
990 if ((error
= vdev_label_init(vd
->vdev_child
[c
],
991 crtxg
, reason
)) != 0)
994 /* Track the creation time for this vdev */
995 vd
->vdev_crtxg
= crtxg
;
997 if (!vd
->vdev_ops
->vdev_op_leaf
|| !spa_writeable(spa
))
1001 * Dead vdevs cannot be initialized.
1003 if (vdev_is_dead(vd
))
1004 return (SET_ERROR(EIO
));
1007 * Determine if the vdev is in use.
1009 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPLIT
&&
1010 vdev_inuse(vd
, crtxg
, reason
, &spare_guid
, &l2cache_guid
))
1011 return (SET_ERROR(EBUSY
));
1014 * If this is a request to add or replace a spare or l2cache device
1015 * that is in use elsewhere on the system, then we must update the
1016 * guid (which was initialized to a random value) to reflect the
1017 * actual GUID (which is shared between multiple pools).
1019 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_L2CACHE
&&
1020 spare_guid
!= 0ULL) {
1021 uint64_t guid_delta
= spare_guid
- vd
->vdev_guid
;
1023 vd
->vdev_guid
+= guid_delta
;
1025 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
1026 pvd
->vdev_guid_sum
+= guid_delta
;
1029 * If this is a replacement, then we want to fallthrough to the
1030 * rest of the code. If we're adding a spare, then it's already
1031 * labeled appropriately and we can just return.
1033 if (reason
== VDEV_LABEL_SPARE
)
1035 ASSERT(reason
== VDEV_LABEL_REPLACE
||
1036 reason
== VDEV_LABEL_SPLIT
);
1039 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPARE
&&
1040 l2cache_guid
!= 0ULL) {
1041 uint64_t guid_delta
= l2cache_guid
- vd
->vdev_guid
;
1043 vd
->vdev_guid
+= guid_delta
;
1045 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
1046 pvd
->vdev_guid_sum
+= guid_delta
;
1049 * If this is a replacement, then we want to fallthrough to the
1050 * rest of the code. If we're adding an l2cache, then it's
1051 * already labeled appropriately and we can just return.
1053 if (reason
== VDEV_LABEL_L2CACHE
)
1055 ASSERT(reason
== VDEV_LABEL_REPLACE
);
1059 * Initialize its label.
1061 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1062 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1063 vp
= abd_to_buf(vp_abd
);
1066 * Generate a label describing the pool and our top-level vdev.
1067 * We mark it as being from txg 0 to indicate that it's not
1068 * really part of an active pool just yet. The labels will
1069 * be written again with a meaningful txg by spa_sync().
1071 if (reason
== VDEV_LABEL_SPARE
||
1072 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isspare
)) {
1074 * For inactive hot spares, we generate a special label that
1075 * identifies as a mutually shared hot spare. We write the
1076 * label if we are adding a hot spare, or if we are removing an
1077 * active hot spare (in which case we want to revert the
1080 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1082 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1083 spa_version(spa
)) == 0);
1084 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1085 POOL_STATE_SPARE
) == 0);
1086 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1087 vd
->vdev_guid
) == 0);
1088 } else if (reason
== VDEV_LABEL_L2CACHE
||
1089 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isl2cache
)) {
1091 * For level 2 ARC devices, add a special label.
1093 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1095 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1096 spa_version(spa
)) == 0);
1097 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1098 POOL_STATE_L2CACHE
) == 0);
1099 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1100 vd
->vdev_guid
) == 0);
1102 uint64_t txg
= 0ULL;
1104 if (reason
== VDEV_LABEL_SPLIT
)
1105 txg
= spa
->spa_uberblock
.ub_txg
;
1106 label
= spa_config_generate(spa
, vd
, txg
, B_FALSE
);
1109 * Add our creation time. This allows us to detect multiple
1110 * vdev uses as described above, and automatically expires if we
1113 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
1117 buf
= vp
->vp_nvlist
;
1118 buflen
= sizeof (vp
->vp_nvlist
);
1120 error
= nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
);
1124 /* EFAULT means nvlist_pack ran out of room */
1125 return (SET_ERROR(error
== EFAULT
? ENAMETOOLONG
: EINVAL
));
1129 * Initialize uberblock template.
1131 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_RING
, B_TRUE
);
1132 abd_zero(ub_abd
, VDEV_UBERBLOCK_RING
);
1133 abd_copy_from_buf(ub_abd
, &spa
->spa_uberblock
, sizeof (uberblock_t
));
1134 ub
= abd_to_buf(ub_abd
);
1137 /* Initialize the 2nd padding area. */
1138 bootenv
= abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
);
1139 abd_zero(bootenv
, VDEV_PAD_SIZE
);
1142 * Write everything in parallel.
1145 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1147 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1149 vdev_label_write(zio
, vd
, l
, vp_abd
,
1150 offsetof(vdev_label_t
, vl_vdev_phys
),
1151 sizeof (vdev_phys_t
), NULL
, NULL
, flags
);
1154 * Skip the 1st padding area.
1155 * Zero out the 2nd padding area where it might have
1156 * left over data from previous filesystem format.
1158 vdev_label_write(zio
, vd
, l
, bootenv
,
1159 offsetof(vdev_label_t
, vl_be
),
1160 VDEV_PAD_SIZE
, NULL
, NULL
, flags
);
1162 vdev_label_write(zio
, vd
, l
, ub_abd
,
1163 offsetof(vdev_label_t
, vl_uberblock
),
1164 VDEV_UBERBLOCK_RING
, NULL
, NULL
, flags
);
1167 error
= zio_wait(zio
);
1169 if (error
!= 0 && !(flags
& ZIO_FLAG_TRYHARD
)) {
1170 flags
|= ZIO_FLAG_TRYHARD
;
1180 * If this vdev hasn't been previously identified as a spare, then we
1181 * mark it as such only if a) we are labeling it as a spare, or b) it
1182 * exists as a spare elsewhere in the system. Do the same for
1183 * level 2 ARC devices.
1185 if (error
== 0 && !vd
->vdev_isspare
&&
1186 (reason
== VDEV_LABEL_SPARE
||
1187 spa_spare_exists(vd
->vdev_guid
, NULL
, NULL
)))
1190 if (error
== 0 && !vd
->vdev_isl2cache
&&
1191 (reason
== VDEV_LABEL_L2CACHE
||
1192 spa_l2cache_exists(vd
->vdev_guid
, NULL
)))
1193 spa_l2cache_add(vd
);
1199 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1200 * callback to finish, store our abd in the callback pointer. Otherwise, we
1201 * just free our abd and return.
1204 vdev_label_read_bootenv_done(zio_t
*zio
)
1206 zio_t
*rio
= zio
->io_private
;
1207 abd_t
**cbp
= rio
->io_private
;
1209 ASSERT3U(zio
->io_size
, ==, VDEV_PAD_SIZE
);
1211 if (zio
->io_error
== 0) {
1212 mutex_enter(&rio
->io_lock
);
1214 /* Will free this buffer in vdev_label_read_bootenv. */
1217 abd_free(zio
->io_abd
);
1219 mutex_exit(&rio
->io_lock
);
1221 abd_free(zio
->io_abd
);
1226 vdev_label_read_bootenv_impl(zio_t
*zio
, vdev_t
*vd
, int flags
)
1228 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1229 vdev_label_read_bootenv_impl(zio
, vd
->vdev_child
[c
], flags
);
1232 * We just use the first label that has a correct checksum; the
1233 * bootloader should have rewritten them all to be the same on boot,
1234 * and any changes we made since boot have been the same across all
1237 * While grub supports writing to all four labels, other bootloaders
1238 * don't, so we only use the first two labels to store boot
1241 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1242 for (int l
= 0; l
< VDEV_LABELS
/ 2; l
++) {
1243 vdev_label_read(zio
, vd
, l
,
1244 abd_alloc_linear(VDEV_PAD_SIZE
, B_FALSE
),
1245 offsetof(vdev_label_t
, vl_be
), VDEV_PAD_SIZE
,
1246 vdev_label_read_bootenv_done
, zio
, flags
);
1252 vdev_label_read_bootenv(vdev_t
*rvd
, nvlist_t
*command
)
1254 spa_t
*spa
= rvd
->vdev_spa
;
1256 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1257 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_TRYHARD
;
1260 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1262 zio_t
*zio
= zio_root(spa
, NULL
, &abd
, flags
);
1263 vdev_label_read_bootenv_impl(zio
, rvd
, flags
);
1264 int err
= zio_wait(zio
);
1267 vdev_boot_envblock_t
*vbe
= abd_to_buf(abd
);
1268 if (vbe
->vbe_version
!= VB_RAW
) {
1270 return (SET_ERROR(ENOTSUP
));
1272 vbe
->vbe_bootenv
[sizeof (vbe
->vbe_bootenv
) - 1] = '\0';
1273 fnvlist_add_string(command
, "envmap", vbe
->vbe_bootenv
);
1274 /* abd was allocated in vdev_label_read_bootenv_impl() */
1276 /* If we managed to read any successfully, return success. */
1283 vdev_label_write_bootenv(vdev_t
*vd
, char *envmap
)
1286 spa_t
*spa
= vd
->vdev_spa
;
1287 vdev_boot_envblock_t
*bootenv
;
1288 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1291 if (strlen(envmap
) >= sizeof (bootenv
->vbe_bootenv
)) {
1292 return (SET_ERROR(E2BIG
));
1295 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1297 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1298 int child_err
= vdev_label_write_bootenv(vd
->vdev_child
[c
],
1301 * As long as any of the disks managed to write all of their
1302 * labels successfully, return success.
1308 if (!vd
->vdev_ops
->vdev_op_leaf
|| vdev_is_dead(vd
) ||
1309 !vdev_writeable(vd
)) {
1312 ASSERT3U(sizeof (*bootenv
), ==, VDEV_PAD_SIZE
);
1313 abd_t
*abd
= abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
);
1314 abd_zero(abd
, VDEV_PAD_SIZE
);
1315 bootenv
= abd_borrow_buf_copy(abd
, VDEV_PAD_SIZE
);
1317 char *buf
= bootenv
->vbe_bootenv
;
1318 (void) strlcpy(buf
, envmap
, sizeof (bootenv
->vbe_bootenv
));
1319 bootenv
->vbe_version
= VB_RAW
;
1320 abd_return_buf_copy(abd
, bootenv
, VDEV_PAD_SIZE
);
1323 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1324 for (int l
= 0; l
< VDEV_LABELS
/ 2; l
++) {
1325 vdev_label_write(zio
, vd
, l
, abd
,
1326 offsetof(vdev_label_t
, vl_be
),
1327 VDEV_PAD_SIZE
, NULL
, NULL
, flags
);
1330 error
= zio_wait(zio
);
1331 if (error
!= 0 && !(flags
& ZIO_FLAG_TRYHARD
)) {
1332 flags
|= ZIO_FLAG_TRYHARD
;
1341 * ==========================================================================
1342 * uberblock load/sync
1343 * ==========================================================================
1347 * Consider the following situation: txg is safely synced to disk. We've
1348 * written the first uberblock for txg + 1, and then we lose power. When we
1349 * come back up, we fail to see the uberblock for txg + 1 because, say,
1350 * it was on a mirrored device and the replica to which we wrote txg + 1
1351 * is now offline. If we then make some changes and sync txg + 1, and then
1352 * the missing replica comes back, then for a few seconds we'll have two
1353 * conflicting uberblocks on disk with the same txg. The solution is simple:
1354 * among uberblocks with equal txg, choose the one with the latest timestamp.
1357 vdev_uberblock_compare(const uberblock_t
*ub1
, const uberblock_t
*ub2
)
1359 int cmp
= TREE_CMP(ub1
->ub_txg
, ub2
->ub_txg
);
1364 cmp
= TREE_CMP(ub1
->ub_timestamp
, ub2
->ub_timestamp
);
1369 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1370 * ZFS, e.g. zfsonlinux >= 0.7.
1372 * If one ub has MMP and the other does not, they were written by
1373 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1376 * Since timestamp and txg are the same if we get this far, either is
1377 * acceptable for importing the pool.
1379 unsigned int seq1
= 0;
1380 unsigned int seq2
= 0;
1382 if (MMP_VALID(ub1
) && MMP_SEQ_VALID(ub1
))
1383 seq1
= MMP_SEQ(ub1
);
1385 if (MMP_VALID(ub2
) && MMP_SEQ_VALID(ub2
))
1386 seq2
= MMP_SEQ(ub2
);
1388 return (TREE_CMP(seq1
, seq2
));
1392 uberblock_t
*ubl_ubbest
; /* Best uberblock */
1393 vdev_t
*ubl_vd
; /* vdev associated with the above */
1397 vdev_uberblock_load_done(zio_t
*zio
)
1399 vdev_t
*vd
= zio
->io_vd
;
1400 spa_t
*spa
= zio
->io_spa
;
1401 zio_t
*rio
= zio
->io_private
;
1402 uberblock_t
*ub
= abd_to_buf(zio
->io_abd
);
1403 struct ubl_cbdata
*cbp
= rio
->io_private
;
1405 ASSERT3U(zio
->io_size
, ==, VDEV_UBERBLOCK_SIZE(vd
));
1407 if (zio
->io_error
== 0 && uberblock_verify(ub
) == 0) {
1408 mutex_enter(&rio
->io_lock
);
1409 if (ub
->ub_txg
<= spa
->spa_load_max_txg
&&
1410 vdev_uberblock_compare(ub
, cbp
->ubl_ubbest
) > 0) {
1412 * Keep track of the vdev in which this uberblock
1413 * was found. We will use this information later
1414 * to obtain the config nvlist associated with
1417 *cbp
->ubl_ubbest
= *ub
;
1420 mutex_exit(&rio
->io_lock
);
1423 abd_free(zio
->io_abd
);
1427 vdev_uberblock_load_impl(zio_t
*zio
, vdev_t
*vd
, int flags
,
1428 struct ubl_cbdata
*cbp
)
1430 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1431 vdev_uberblock_load_impl(zio
, vd
->vdev_child
[c
], flags
, cbp
);
1433 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1434 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1435 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1436 vdev_label_read(zio
, vd
, l
,
1437 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
),
1438 B_TRUE
), VDEV_UBERBLOCK_OFFSET(vd
, n
),
1439 VDEV_UBERBLOCK_SIZE(vd
),
1440 vdev_uberblock_load_done
, zio
, flags
);
1447 * Reads the 'best' uberblock from disk along with its associated
1448 * configuration. First, we read the uberblock array of each label of each
1449 * vdev, keeping track of the uberblock with the highest txg in each array.
1450 * Then, we read the configuration from the same vdev as the best uberblock.
1453 vdev_uberblock_load(vdev_t
*rvd
, uberblock_t
*ub
, nvlist_t
**config
)
1456 spa_t
*spa
= rvd
->vdev_spa
;
1457 struct ubl_cbdata cb
;
1458 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1459 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_TRYHARD
;
1464 bzero(ub
, sizeof (uberblock_t
));
1470 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1471 zio
= zio_root(spa
, NULL
, &cb
, flags
);
1472 vdev_uberblock_load_impl(zio
, rvd
, flags
, &cb
);
1473 (void) zio_wait(zio
);
1476 * It's possible that the best uberblock was discovered on a label
1477 * that has a configuration which was written in a future txg.
1478 * Search all labels on this vdev to find the configuration that
1479 * matches the txg for our uberblock.
1481 if (cb
.ubl_vd
!= NULL
) {
1482 vdev_dbgmsg(cb
.ubl_vd
, "best uberblock found for spa %s. "
1483 "txg %llu", spa
->spa_name
, (u_longlong_t
)ub
->ub_txg
);
1485 *config
= vdev_label_read_config(cb
.ubl_vd
, ub
->ub_txg
);
1486 if (*config
== NULL
&& spa
->spa_extreme_rewind
) {
1487 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config. "
1488 "Trying again without txg restrictions.");
1489 *config
= vdev_label_read_config(cb
.ubl_vd
, UINT64_MAX
);
1491 if (*config
== NULL
) {
1492 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config");
1495 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1499 * For use when a leaf vdev is expanded.
1500 * The location of labels 2 and 3 changed, and at the new location the
1501 * uberblock rings are either empty or contain garbage. The sync will write
1502 * new configs there because the vdev is dirty, but expansion also needs the
1503 * uberblock rings copied. Read them from label 0 which did not move.
1505 * Since the point is to populate labels {2,3} with valid uberblocks,
1506 * we zero uberblocks we fail to read or which are not valid.
1510 vdev_copy_uberblocks(vdev_t
*vd
)
1514 int locks
= (SCL_L2ARC
| SCL_ZIO
);
1515 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1516 ZIO_FLAG_SPECULATIVE
;
1518 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_READER
) ==
1520 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1522 spa_config_enter(vd
->vdev_spa
, locks
, FTAG
, RW_READER
);
1524 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1526 write_zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1527 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1528 const int src_label
= 0;
1531 zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1532 vdev_label_read(zio
, vd
, src_label
, ub_abd
,
1533 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1536 if (zio_wait(zio
) || uberblock_verify(abd_to_buf(ub_abd
)))
1537 abd_zero(ub_abd
, VDEV_UBERBLOCK_SIZE(vd
));
1539 for (int l
= 2; l
< VDEV_LABELS
; l
++)
1540 vdev_label_write(write_zio
, vd
, l
, ub_abd
,
1541 VDEV_UBERBLOCK_OFFSET(vd
, n
),
1542 VDEV_UBERBLOCK_SIZE(vd
), NULL
, NULL
,
1543 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1545 (void) zio_wait(write_zio
);
1547 spa_config_exit(vd
->vdev_spa
, locks
, FTAG
);
1553 * On success, increment root zio's count of good writes.
1554 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1557 vdev_uberblock_sync_done(zio_t
*zio
)
1559 uint64_t *good_writes
= zio
->io_private
;
1561 if (zio
->io_error
== 0 && zio
->io_vd
->vdev_top
->vdev_ms_array
!= 0)
1562 atomic_inc_64(good_writes
);
1566 * Write the uberblock to all labels of all leaves of the specified vdev.
1569 vdev_uberblock_sync(zio_t
*zio
, uint64_t *good_writes
,
1570 uberblock_t
*ub
, vdev_t
*vd
, int flags
)
1572 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
1573 vdev_uberblock_sync(zio
, good_writes
,
1574 ub
, vd
->vdev_child
[c
], flags
);
1577 if (!vd
->vdev_ops
->vdev_op_leaf
)
1580 if (!vdev_writeable(vd
))
1583 /* If the vdev was expanded, need to copy uberblock rings. */
1584 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1585 vd
->vdev_copy_uberblocks
== B_TRUE
) {
1586 vdev_copy_uberblocks(vd
);
1587 vd
->vdev_copy_uberblocks
= B_FALSE
;
1590 int m
= spa_multihost(vd
->vdev_spa
) ? MMP_BLOCKS_PER_LABEL
: 0;
1591 int n
= ub
->ub_txg
% (VDEV_UBERBLOCK_COUNT(vd
) - m
);
1593 /* Copy the uberblock_t into the ABD */
1594 abd_t
*ub_abd
= abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1595 abd_zero(ub_abd
, VDEV_UBERBLOCK_SIZE(vd
));
1596 abd_copy_from_buf(ub_abd
, ub
, sizeof (uberblock_t
));
1598 for (int l
= 0; l
< VDEV_LABELS
; l
++)
1599 vdev_label_write(zio
, vd
, l
, ub_abd
,
1600 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1601 vdev_uberblock_sync_done
, good_writes
,
1602 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1607 /* Sync the uberblocks to all vdevs in svd[] */
1609 vdev_uberblock_sync_list(vdev_t
**svd
, int svdcount
, uberblock_t
*ub
, int flags
)
1611 spa_t
*spa
= svd
[0]->vdev_spa
;
1613 uint64_t good_writes
= 0;
1615 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1617 for (int v
= 0; v
< svdcount
; v
++)
1618 vdev_uberblock_sync(zio
, &good_writes
, ub
, svd
[v
], flags
);
1620 (void) zio_wait(zio
);
1623 * Flush the uberblocks to disk. This ensures that the odd labels
1624 * are no longer needed (because the new uberblocks and the even
1625 * labels are safely on disk), so it is safe to overwrite them.
1627 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1629 for (int v
= 0; v
< svdcount
; v
++) {
1630 if (vdev_writeable(svd
[v
])) {
1631 zio_flush(zio
, svd
[v
]);
1635 (void) zio_wait(zio
);
1637 return (good_writes
>= 1 ? 0 : EIO
);
1641 * On success, increment the count of good writes for our top-level vdev.
1644 vdev_label_sync_done(zio_t
*zio
)
1646 uint64_t *good_writes
= zio
->io_private
;
1648 if (zio
->io_error
== 0)
1649 atomic_inc_64(good_writes
);
1653 * If there weren't enough good writes, indicate failure to the parent.
1656 vdev_label_sync_top_done(zio_t
*zio
)
1658 uint64_t *good_writes
= zio
->io_private
;
1660 if (*good_writes
== 0)
1661 zio
->io_error
= SET_ERROR(EIO
);
1663 kmem_free(good_writes
, sizeof (uint64_t));
1667 * We ignore errors for log and cache devices, simply free the private data.
1670 vdev_label_sync_ignore_done(zio_t
*zio
)
1672 kmem_free(zio
->io_private
, sizeof (uint64_t));
1676 * Write all even or odd labels to all leaves of the specified vdev.
1679 vdev_label_sync(zio_t
*zio
, uint64_t *good_writes
,
1680 vdev_t
*vd
, int l
, uint64_t txg
, int flags
)
1688 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1689 vdev_label_sync(zio
, good_writes
,
1690 vd
->vdev_child
[c
], l
, txg
, flags
);
1693 if (!vd
->vdev_ops
->vdev_op_leaf
)
1696 if (!vdev_writeable(vd
))
1700 * Generate a label describing the top-level config to which we belong.
1702 label
= spa_config_generate(vd
->vdev_spa
, vd
, txg
, B_FALSE
);
1704 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1705 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1706 vp
= abd_to_buf(vp_abd
);
1708 buf
= vp
->vp_nvlist
;
1709 buflen
= sizeof (vp
->vp_nvlist
);
1711 if (!nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
)) {
1712 for (; l
< VDEV_LABELS
; l
+= 2) {
1713 vdev_label_write(zio
, vd
, l
, vp_abd
,
1714 offsetof(vdev_label_t
, vl_vdev_phys
),
1715 sizeof (vdev_phys_t
),
1716 vdev_label_sync_done
, good_writes
,
1717 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1726 vdev_label_sync_list(spa_t
*spa
, int l
, uint64_t txg
, int flags
)
1728 list_t
*dl
= &spa
->spa_config_dirty_list
;
1734 * Write the new labels to disk.
1736 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1738 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
)) {
1739 uint64_t *good_writes
;
1741 ASSERT(!vd
->vdev_ishole
);
1743 good_writes
= kmem_zalloc(sizeof (uint64_t), KM_SLEEP
);
1744 zio_t
*vio
= zio_null(zio
, spa
, NULL
,
1745 (vd
->vdev_islog
|| vd
->vdev_aux
!= NULL
) ?
1746 vdev_label_sync_ignore_done
: vdev_label_sync_top_done
,
1747 good_writes
, flags
);
1748 vdev_label_sync(vio
, good_writes
, vd
, l
, txg
, flags
);
1752 error
= zio_wait(zio
);
1755 * Flush the new labels to disk.
1757 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1759 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
))
1762 (void) zio_wait(zio
);
1768 * Sync the uberblock and any changes to the vdev configuration.
1770 * The order of operations is carefully crafted to ensure that
1771 * if the system panics or loses power at any time, the state on disk
1772 * is still transactionally consistent. The in-line comments below
1773 * describe the failure semantics at each stage.
1775 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1776 * at any time, you can just call it again, and it will resume its work.
1779 vdev_config_sync(vdev_t
**svd
, int svdcount
, uint64_t txg
)
1781 spa_t
*spa
= svd
[0]->vdev_spa
;
1782 uberblock_t
*ub
= &spa
->spa_uberblock
;
1784 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1786 ASSERT(svdcount
!= 0);
1789 * Normally, we don't want to try too hard to write every label and
1790 * uberblock. If there is a flaky disk, we don't want the rest of the
1791 * sync process to block while we retry. But if we can't write a
1792 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1793 * bailing out and declaring the pool faulted.
1796 if ((flags
& ZIO_FLAG_TRYHARD
) != 0)
1798 flags
|= ZIO_FLAG_TRYHARD
;
1801 ASSERT(ub
->ub_txg
<= txg
);
1804 * If this isn't a resync due to I/O errors,
1805 * and nothing changed in this transaction group,
1806 * and the vdev configuration hasn't changed,
1807 * then there's nothing to do.
1809 if (ub
->ub_txg
< txg
) {
1810 boolean_t changed
= uberblock_update(ub
, spa
->spa_root_vdev
,
1811 txg
, spa
->spa_mmp
.mmp_delay
);
1813 if (!changed
&& list_is_empty(&spa
->spa_config_dirty_list
))
1817 if (txg
> spa_freeze_txg(spa
))
1820 ASSERT(txg
<= spa
->spa_final_txg
);
1823 * Flush the write cache of every disk that's been written to
1824 * in this transaction group. This ensures that all blocks
1825 * written in this txg will be committed to stable storage
1826 * before any uberblock that references them.
1828 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, flags
);
1831 txg_list_head(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)); vd
!= NULL
;
1832 vd
= txg_list_next(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
)))
1835 (void) zio_wait(zio
);
1838 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1839 * system dies in the middle of this process, that's OK: all of the
1840 * even labels that made it to disk will be newer than any uberblock,
1841 * and will therefore be considered invalid. The odd labels (L1, L3),
1842 * which have not yet been touched, will still be valid. We flush
1843 * the new labels to disk to ensure that all even-label updates
1844 * are committed to stable storage before the uberblock update.
1846 if ((error
= vdev_label_sync_list(spa
, 0, txg
, flags
)) != 0) {
1847 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1848 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1849 "for pool '%s' when syncing out the even labels "
1850 "of dirty vdevs", error
, spa_name(spa
));
1856 * Sync the uberblocks to all vdevs in svd[].
1857 * If the system dies in the middle of this step, there are two cases
1858 * to consider, and the on-disk state is consistent either way:
1860 * (1) If none of the new uberblocks made it to disk, then the
1861 * previous uberblock will be the newest, and the odd labels
1862 * (which had not yet been touched) will be valid with respect
1863 * to that uberblock.
1865 * (2) If one or more new uberblocks made it to disk, then they
1866 * will be the newest, and the even labels (which had all
1867 * been successfully committed) will be valid with respect
1868 * to the new uberblocks.
1870 if ((error
= vdev_uberblock_sync_list(svd
, svdcount
, ub
, flags
)) != 0) {
1871 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1872 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1873 "%d for pool '%s'", error
, spa_name(spa
));
1878 if (spa_multihost(spa
))
1879 mmp_update_uberblock(spa
, ub
);
1882 * Sync out odd labels for every dirty vdev. If the system dies
1883 * in the middle of this process, the even labels and the new
1884 * uberblocks will suffice to open the pool. The next time
1885 * the pool is opened, the first thing we'll do -- before any
1886 * user data is modified -- is mark every vdev dirty so that
1887 * all labels will be brought up to date. We flush the new labels
1888 * to disk to ensure that all odd-label updates are committed to
1889 * stable storage before the next transaction group begins.
1891 if ((error
= vdev_label_sync_list(spa
, 1, txg
, flags
)) != 0) {
1892 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1893 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1894 "for pool '%s' when syncing out the odd labels of "
1895 "dirty vdevs", error
, spa_name(spa
));