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 https://opensource.org/licenses/CDDL-1.0.
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/vdev_raidz.h>
146 #include <sys/vdev_draid.h>
147 #include <sys/uberblock_impl.h>
148 #include <sys/metaslab.h>
149 #include <sys/metaslab_impl.h>
151 #include <sys/dsl_scan.h>
153 #include <sys/fs/zfs.h>
154 #include <sys/byteorder.h>
155 #include <sys/zfs_bootenv.h>
158 * Basic routines to read and write from a vdev label.
159 * Used throughout the rest of this file.
162 vdev_label_offset(uint64_t psize
, int l
, uint64_t offset
)
164 ASSERT(offset
< sizeof (vdev_label_t
));
165 ASSERT(P2PHASE_TYPED(psize
, sizeof (vdev_label_t
), uint64_t) == 0);
167 return (offset
+ l
* sizeof (vdev_label_t
) + (l
< VDEV_LABELS
/ 2 ?
168 0 : psize
- VDEV_LABELS
* sizeof (vdev_label_t
)));
172 * Returns back the vdev label associated with the passed in offset.
175 vdev_label_number(uint64_t psize
, uint64_t offset
)
179 if (offset
>= psize
- VDEV_LABEL_END_SIZE
) {
180 offset
-= psize
- VDEV_LABEL_END_SIZE
;
181 offset
+= (VDEV_LABELS
/ 2) * sizeof (vdev_label_t
);
183 l
= offset
/ sizeof (vdev_label_t
);
184 return (l
< VDEV_LABELS
? l
: -1);
188 vdev_label_read(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
189 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
192 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
193 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
194 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
196 zio_nowait(zio_read_phys(zio
, vd
,
197 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
198 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
199 ZIO_PRIORITY_SYNC_READ
, flags
, B_TRUE
));
203 vdev_label_write(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
204 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
207 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
208 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
209 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
211 zio_nowait(zio_write_phys(zio
, vd
,
212 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
213 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
214 ZIO_PRIORITY_SYNC_WRITE
, flags
, B_TRUE
));
218 * Generate the nvlist representing this vdev's stats
221 vdev_config_generate_stats(vdev_t
*vd
, nvlist_t
*nv
)
227 vs
= kmem_alloc(sizeof (*vs
), KM_SLEEP
);
228 vsx
= kmem_alloc(sizeof (*vsx
), KM_SLEEP
);
230 vdev_get_stats_ex(vd
, vs
, vsx
);
231 fnvlist_add_uint64_array(nv
, ZPOOL_CONFIG_VDEV_STATS
,
232 (uint64_t *)vs
, sizeof (*vs
) / sizeof (uint64_t));
235 * Add extended stats into a special extended stats nvlist. This keeps
236 * all the extended stats nicely grouped together. The extended stats
237 * nvlist is then added to the main nvlist.
239 nvx
= fnvlist_alloc();
241 /* ZIOs in flight to disk */
242 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE
,
243 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_READ
]);
245 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE
,
246 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
248 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE
,
249 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_READ
]);
251 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE
,
252 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
254 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE
,
255 vsx
->vsx_active_queue
[ZIO_PRIORITY_SCRUB
]);
257 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE
,
258 vsx
->vsx_active_queue
[ZIO_PRIORITY_TRIM
]);
260 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE
,
261 vsx
->vsx_active_queue
[ZIO_PRIORITY_REBUILD
]);
264 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE
,
265 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_READ
]);
267 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE
,
268 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
270 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE
,
271 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_READ
]);
273 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE
,
274 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
276 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE
,
277 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SCRUB
]);
279 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE
,
280 vsx
->vsx_pend_queue
[ZIO_PRIORITY_TRIM
]);
282 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE
,
283 vsx
->vsx_pend_queue
[ZIO_PRIORITY_REBUILD
]);
286 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO
,
287 vsx
->vsx_total_histo
[ZIO_TYPE_READ
],
288 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_READ
]));
290 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO
,
291 vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
],
292 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
]));
294 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO
,
295 vsx
->vsx_disk_histo
[ZIO_TYPE_READ
],
296 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_READ
]));
298 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO
,
299 vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
],
300 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
]));
302 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO
,
303 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
],
304 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
]));
306 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO
,
307 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
],
308 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
310 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO
,
311 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
],
312 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
]));
314 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO
,
315 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
316 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
318 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO
,
319 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
],
320 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
]));
322 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO
,
323 vsx
->vsx_queue_histo
[ZIO_PRIORITY_TRIM
],
324 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_TRIM
]));
326 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO
,
327 vsx
->vsx_queue_histo
[ZIO_PRIORITY_REBUILD
],
328 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_REBUILD
]));
331 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO
,
332 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
],
333 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
]));
335 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO
,
336 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
],
337 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
339 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO
,
340 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
],
341 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
]));
343 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO
,
344 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
345 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
347 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO
,
348 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
],
349 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
]));
351 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO
,
352 vsx
->vsx_ind_histo
[ZIO_PRIORITY_TRIM
],
353 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_TRIM
]));
355 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO
,
356 vsx
->vsx_ind_histo
[ZIO_PRIORITY_REBUILD
],
357 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_REBUILD
]));
359 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO
,
360 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
],
361 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
]));
363 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO
,
364 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
],
365 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
367 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO
,
368 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
],
369 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
]));
371 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO
,
372 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
373 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
375 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO
,
376 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
],
377 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
]));
379 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO
,
380 vsx
->vsx_agg_histo
[ZIO_PRIORITY_TRIM
],
381 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_TRIM
]));
383 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO
,
384 vsx
->vsx_agg_histo
[ZIO_PRIORITY_REBUILD
],
385 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_REBUILD
]));
388 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SLOW_IOS
, vs
->vs_slow_ios
);
390 /* Add extended stats nvlist to main nvlist */
391 fnvlist_add_nvlist(nv
, ZPOOL_CONFIG_VDEV_STATS_EX
, nvx
);
394 kmem_free(vs
, sizeof (*vs
));
395 kmem_free(vsx
, sizeof (*vsx
));
399 root_vdev_actions_getprogress(vdev_t
*vd
, nvlist_t
*nvl
)
401 spa_t
*spa
= vd
->vdev_spa
;
403 if (vd
!= spa
->spa_root_vdev
)
406 /* provide either current or previous scan information */
408 if (spa_scan_get_stats(spa
, &ps
) == 0) {
409 fnvlist_add_uint64_array(nvl
,
410 ZPOOL_CONFIG_SCAN_STATS
, (uint64_t *)&ps
,
411 sizeof (pool_scan_stat_t
) / sizeof (uint64_t));
414 pool_removal_stat_t prs
;
415 if (spa_removal_get_stats(spa
, &prs
) == 0) {
416 fnvlist_add_uint64_array(nvl
,
417 ZPOOL_CONFIG_REMOVAL_STATS
, (uint64_t *)&prs
,
418 sizeof (prs
) / sizeof (uint64_t));
421 pool_checkpoint_stat_t pcs
;
422 if (spa_checkpoint_get_stats(spa
, &pcs
) == 0) {
423 fnvlist_add_uint64_array(nvl
,
424 ZPOOL_CONFIG_CHECKPOINT_STATS
, (uint64_t *)&pcs
,
425 sizeof (pcs
) / sizeof (uint64_t));
428 pool_raidz_expand_stat_t pres
;
429 if (spa_raidz_expand_get_stats(spa
, &pres
) == 0) {
430 fnvlist_add_uint64_array(nvl
,
431 ZPOOL_CONFIG_RAIDZ_EXPAND_STATS
, (uint64_t *)&pres
,
432 sizeof (pres
) / sizeof (uint64_t));
437 top_vdev_actions_getprogress(vdev_t
*vd
, nvlist_t
*nvl
)
439 if (vd
== vd
->vdev_top
) {
440 vdev_rebuild_stat_t vrs
;
441 if (vdev_rebuild_get_stats(vd
, &vrs
) == 0) {
442 fnvlist_add_uint64_array(nvl
,
443 ZPOOL_CONFIG_REBUILD_STATS
, (uint64_t *)&vrs
,
444 sizeof (vrs
) / sizeof (uint64_t));
450 * Generate the nvlist representing this vdev's config.
453 vdev_config_generate(spa_t
*spa
, vdev_t
*vd
, boolean_t getstats
,
454 vdev_config_flag_t flags
)
457 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
459 nv
= fnvlist_alloc();
461 fnvlist_add_string(nv
, ZPOOL_CONFIG_TYPE
, vd
->vdev_ops
->vdev_op_type
);
462 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)))
463 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ID
, vd
->vdev_id
);
464 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_GUID
, vd
->vdev_guid
);
466 if (vd
->vdev_path
!= NULL
)
467 fnvlist_add_string(nv
, ZPOOL_CONFIG_PATH
, vd
->vdev_path
);
469 if (vd
->vdev_devid
!= NULL
)
470 fnvlist_add_string(nv
, ZPOOL_CONFIG_DEVID
, vd
->vdev_devid
);
472 if (vd
->vdev_physpath
!= NULL
)
473 fnvlist_add_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
476 if (vd
->vdev_enc_sysfs_path
!= NULL
)
477 fnvlist_add_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
478 vd
->vdev_enc_sysfs_path
);
480 if (vd
->vdev_fru
!= NULL
)
481 fnvlist_add_string(nv
, ZPOOL_CONFIG_FRU
, vd
->vdev_fru
);
483 if (vd
->vdev_ops
->vdev_op_config_generate
!= NULL
)
484 vd
->vdev_ops
->vdev_op_config_generate(vd
, nv
);
486 if (vd
->vdev_wholedisk
!= -1ULL) {
487 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
491 if (vd
->vdev_not_present
&& !(flags
& VDEV_CONFIG_MISSING
))
492 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
, 1);
494 if (vd
->vdev_isspare
)
495 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
, 1);
497 if (flags
& VDEV_CONFIG_L2CACHE
)
498 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, vd
->vdev_ashift
);
500 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)) &&
501 vd
== vd
->vdev_top
) {
502 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
504 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
506 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, vd
->vdev_ashift
);
507 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
509 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, vd
->vdev_islog
);
510 if (vd
->vdev_noalloc
) {
511 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NONALLOCATING
,
516 * Slog devices are removed synchronously so don't
517 * persist the vdev_removing flag to the label.
519 if (vd
->vdev_removing
&& !vd
->vdev_islog
) {
520 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
524 /* zpool command expects alloc class data */
525 if (getstats
&& vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
) {
526 const char *bias
= NULL
;
528 switch (vd
->vdev_alloc_bias
) {
530 bias
= VDEV_ALLOC_BIAS_LOG
;
532 case VDEV_BIAS_SPECIAL
:
533 bias
= VDEV_ALLOC_BIAS_SPECIAL
;
535 case VDEV_BIAS_DEDUP
:
536 bias
= VDEV_ALLOC_BIAS_DEDUP
;
539 ASSERT3U(vd
->vdev_alloc_bias
, ==,
542 fnvlist_add_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
547 if (vd
->vdev_dtl_sm
!= NULL
) {
548 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DTL
,
549 space_map_object(vd
->vdev_dtl_sm
));
552 if (vic
->vic_mapping_object
!= 0) {
553 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
554 vic
->vic_mapping_object
);
557 if (vic
->vic_births_object
!= 0) {
558 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
559 vic
->vic_births_object
);
562 if (vic
->vic_prev_indirect_vdev
!= UINT64_MAX
) {
563 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
564 vic
->vic_prev_indirect_vdev
);
568 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
, vd
->vdev_crtxg
);
570 if (vd
->vdev_expansion_time
)
571 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_EXPANSION_TIME
,
572 vd
->vdev_expansion_time
);
574 if (flags
& VDEV_CONFIG_MOS
) {
575 if (vd
->vdev_leaf_zap
!= 0) {
576 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
577 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_LEAF_ZAP
,
581 if (vd
->vdev_top_zap
!= 0) {
582 ASSERT(vd
== vd
->vdev_top
);
583 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
587 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_root_zap
!= 0 &&
588 spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
)) {
589 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_ROOT_ZAP
,
593 if (vd
->vdev_resilver_deferred
) {
594 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
595 ASSERT(spa
->spa_resilver_deferred
);
596 fnvlist_add_boolean(nv
, ZPOOL_CONFIG_RESILVER_DEFER
);
601 vdev_config_generate_stats(vd
, nv
);
603 root_vdev_actions_getprogress(vd
, nv
);
604 top_vdev_actions_getprogress(vd
, nv
);
607 * Note: this can be called from open context
608 * (spa_get_stats()), so we need the rwlock to prevent
609 * the mapping from being changed by condensing.
611 rw_enter(&vd
->vdev_indirect_rwlock
, RW_READER
);
612 if (vd
->vdev_indirect_mapping
!= NULL
) {
613 ASSERT(vd
->vdev_indirect_births
!= NULL
);
614 vdev_indirect_mapping_t
*vim
=
615 vd
->vdev_indirect_mapping
;
616 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
617 vdev_indirect_mapping_size(vim
));
619 rw_exit(&vd
->vdev_indirect_rwlock
);
620 if (vd
->vdev_mg
!= NULL
&&
621 vd
->vdev_mg
->mg_fragmentation
!= ZFS_FRAG_INVALID
) {
623 * Compute approximately how much memory would be used
624 * for the indirect mapping if this device were to
627 * Note: If the frag metric is invalid, then not
628 * enough metaslabs have been converted to have
631 uint64_t seg_count
= 0;
632 uint64_t to_alloc
= vd
->vdev_stat
.vs_alloc
;
635 * There are the same number of allocated segments
636 * as free segments, so we will have at least one
637 * entry per free segment. However, small free
638 * segments (smaller than vdev_removal_max_span)
639 * will be combined with adjacent allocated segments
640 * as a single mapping.
642 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
643 if (i
+ 1 < highbit64(vdev_removal_max_span
)
646 vd
->vdev_mg
->mg_histogram
[i
] <<
650 vd
->vdev_mg
->mg_histogram
[i
];
655 * The maximum length of a mapping is
656 * zfs_remove_max_segment, so we need at least one entry
657 * per zfs_remove_max_segment of allocated data.
659 seg_count
+= to_alloc
/ spa_remove_max_segment(spa
);
661 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
663 sizeof (vdev_indirect_mapping_entry_phys_t
));
667 if (!vd
->vdev_ops
->vdev_op_leaf
) {
671 ASSERT(!vd
->vdev_ishole
);
673 child
= kmem_alloc(vd
->vdev_children
* sizeof (nvlist_t
*),
676 for (c
= 0; c
< vd
->vdev_children
; c
++) {
677 child
[c
] = vdev_config_generate(spa
, vd
->vdev_child
[c
],
681 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
682 (const nvlist_t
* const *)child
, vd
->vdev_children
);
684 for (c
= 0; c
< vd
->vdev_children
; c
++)
685 nvlist_free(child
[c
]);
687 kmem_free(child
, vd
->vdev_children
* sizeof (nvlist_t
*));
690 const char *aux
= NULL
;
692 if (vd
->vdev_offline
&& !vd
->vdev_tmpoffline
)
693 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_OFFLINE
, B_TRUE
);
694 if (vd
->vdev_resilver_txg
!= 0)
695 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
696 vd
->vdev_resilver_txg
);
697 if (vd
->vdev_rebuild_txg
!= 0)
698 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
699 vd
->vdev_rebuild_txg
);
700 if (vd
->vdev_faulted
)
701 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_FAULTED
, B_TRUE
);
702 if (vd
->vdev_degraded
)
703 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DEGRADED
, B_TRUE
);
704 if (vd
->vdev_removed
)
705 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVED
, B_TRUE
);
706 if (vd
->vdev_unspare
)
707 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_UNSPARE
, B_TRUE
);
709 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_HOLE
, B_TRUE
);
711 /* Set the reason why we're FAULTED/DEGRADED. */
712 switch (vd
->vdev_stat
.vs_aux
) {
713 case VDEV_AUX_ERR_EXCEEDED
:
714 aux
= "err_exceeded";
717 case VDEV_AUX_EXTERNAL
:
722 if (aux
!= NULL
&& !vd
->vdev_tmpoffline
) {
723 fnvlist_add_string(nv
, ZPOOL_CONFIG_AUX_STATE
, aux
);
726 * We're healthy - clear any previous AUX_STATE values.
728 if (nvlist_exists(nv
, ZPOOL_CONFIG_AUX_STATE
))
729 nvlist_remove_all(nv
, ZPOOL_CONFIG_AUX_STATE
);
732 if (vd
->vdev_splitting
&& vd
->vdev_orig_guid
!= 0LL) {
733 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ORIG_GUID
,
742 * Generate a view of the top-level vdevs. If we currently have holes
743 * in the namespace, then generate an array which contains a list of holey
744 * vdevs. Additionally, add the number of top-level children that currently
748 vdev_top_config_generate(spa_t
*spa
, nvlist_t
*config
)
750 vdev_t
*rvd
= spa
->spa_root_vdev
;
754 array
= kmem_alloc(rvd
->vdev_children
* sizeof (uint64_t), KM_SLEEP
);
756 for (c
= 0, idx
= 0; c
< rvd
->vdev_children
; c
++) {
757 vdev_t
*tvd
= rvd
->vdev_child
[c
];
759 if (tvd
->vdev_ishole
) {
765 VERIFY(nvlist_add_uint64_array(config
, ZPOOL_CONFIG_HOLE_ARRAY
,
769 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
770 rvd
->vdev_children
) == 0);
772 kmem_free(array
, rvd
->vdev_children
* sizeof (uint64_t));
776 * Returns the configuration from the label of the given vdev. For vdevs
777 * which don't have a txg value stored on their label (i.e. spares/cache)
778 * or have not been completely initialized (txg = 0) just return
779 * the configuration from the first valid label we find. Otherwise,
780 * find the most up-to-date label that does not exceed the specified
784 vdev_label_read_config(vdev_t
*vd
, uint64_t txg
)
786 spa_t
*spa
= vd
->vdev_spa
;
787 nvlist_t
*config
= NULL
;
788 vdev_phys_t
*vp
[VDEV_LABELS
];
789 abd_t
*vp_abd
[VDEV_LABELS
];
790 zio_t
*zio
[VDEV_LABELS
];
791 uint64_t best_txg
= 0;
792 uint64_t label_txg
= 0;
794 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
795 ZIO_FLAG_SPECULATIVE
;
797 ASSERT(vd
->vdev_validate_thread
== curthread
||
798 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
800 if (!vdev_readable(vd
))
804 * The label for a dRAID distributed spare is not stored on disk.
805 * Instead it is generated when needed which allows us to bypass
806 * the pipeline when reading the config from the label.
808 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
809 return (vdev_draid_read_config_spare(vd
));
811 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
812 vp_abd
[l
] = abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
813 vp
[l
] = abd_to_buf(vp_abd
[l
]);
817 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
818 zio
[l
] = zio_root(spa
, NULL
, NULL
, flags
);
820 vdev_label_read(zio
[l
], vd
, l
, vp_abd
[l
],
821 offsetof(vdev_label_t
, vl_vdev_phys
), sizeof (vdev_phys_t
),
824 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
825 nvlist_t
*label
= NULL
;
827 if (zio_wait(zio
[l
]) == 0 &&
828 nvlist_unpack(vp
[l
]->vp_nvlist
, sizeof (vp
[l
]->vp_nvlist
),
831 * Auxiliary vdevs won't have txg values in their
832 * labels and newly added vdevs may not have been
833 * completely initialized so just return the
834 * configuration from the first valid label we
837 error
= nvlist_lookup_uint64(label
,
838 ZPOOL_CONFIG_POOL_TXG
, &label_txg
);
839 if ((error
|| label_txg
== 0) && !config
) {
841 for (l
++; l
< VDEV_LABELS
; l
++)
844 } else if (label_txg
<= txg
&& label_txg
> best_txg
) {
845 best_txg
= label_txg
;
847 config
= fnvlist_dup(label
);
857 if (config
== NULL
&& !(flags
& ZIO_FLAG_TRYHARD
)) {
858 flags
|= ZIO_FLAG_TRYHARD
;
863 * We found a valid label but it didn't pass txg restrictions.
865 if (config
== NULL
&& label_txg
!= 0) {
866 vdev_dbgmsg(vd
, "label discarded as txg is too large "
867 "(%llu > %llu)", (u_longlong_t
)label_txg
,
871 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
879 * Determine if a device is in use. The 'spare_guid' parameter will be filled
880 * in with the device guid if this spare is active elsewhere on the system.
883 vdev_inuse(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
,
884 uint64_t *spare_guid
, uint64_t *l2cache_guid
)
886 spa_t
*spa
= vd
->vdev_spa
;
887 uint64_t state
, pool_guid
, device_guid
, txg
, spare_pool
;
894 *l2cache_guid
= 0ULL;
897 * Read the label, if any, and perform some basic sanity checks.
899 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
)
902 (void) nvlist_lookup_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
905 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
907 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
908 &device_guid
) != 0) {
913 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
914 (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
,
916 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
925 * Check to see if this device indeed belongs to the pool it claims to
926 * be a part of. The only way this is allowed is if the device is a hot
927 * spare (which we check for later on).
929 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
930 !spa_guid_exists(pool_guid
, device_guid
) &&
931 !spa_spare_exists(device_guid
, NULL
, NULL
) &&
932 !spa_l2cache_exists(device_guid
, NULL
))
936 * If the transaction group is zero, then this an initialized (but
937 * unused) label. This is only an error if the create transaction
938 * on-disk is the same as the one we're using now, in which case the
939 * user has attempted to add the same vdev multiple times in the same
942 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
943 txg
== 0 && vdtxg
== crtxg
)
947 * Check to see if this is a spare device. We do an explicit check for
948 * spa_has_spare() here because it may be on our pending list of spares
951 if (spa_spare_exists(device_guid
, &spare_pool
, NULL
) ||
952 spa_has_spare(spa
, device_guid
)) {
954 *spare_guid
= device_guid
;
957 case VDEV_LABEL_CREATE
:
960 case VDEV_LABEL_REPLACE
:
961 return (!spa_has_spare(spa
, device_guid
) ||
964 case VDEV_LABEL_SPARE
:
965 return (spa_has_spare(spa
, device_guid
));
972 * Check to see if this is an l2cache device.
974 if (spa_l2cache_exists(device_guid
, NULL
) ||
975 spa_has_l2cache(spa
, device_guid
)) {
977 *l2cache_guid
= device_guid
;
980 case VDEV_LABEL_CREATE
:
983 case VDEV_LABEL_REPLACE
:
984 return (!spa_has_l2cache(spa
, device_guid
));
986 case VDEV_LABEL_L2CACHE
:
987 return (spa_has_l2cache(spa
, device_guid
));
994 * We can't rely on a pool's state if it's been imported
995 * read-only. Instead we look to see if the pools is marked
996 * read-only in the namespace and set the state to active.
998 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
999 (spa
= spa_by_guid(pool_guid
, device_guid
)) != NULL
&&
1000 spa_mode(spa
) == SPA_MODE_READ
)
1001 state
= POOL_STATE_ACTIVE
;
1004 * If the device is marked ACTIVE, then this device is in use by another
1005 * pool on the system.
1007 return (state
== POOL_STATE_ACTIVE
);
1011 * Initialize a vdev label. We check to make sure each leaf device is not in
1012 * use, and writable. We put down an initial label which we will later
1013 * overwrite with a complete label. Note that it's important to do this
1014 * sequentially, not in parallel, so that we catch cases of multiple use of the
1015 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1019 vdev_label_init(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
)
1021 spa_t
*spa
= vd
->vdev_spa
;
1032 uint64_t spare_guid
= 0, l2cache_guid
= 0;
1033 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1035 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1037 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1038 if ((error
= vdev_label_init(vd
->vdev_child
[c
],
1039 crtxg
, reason
)) != 0)
1042 /* Track the creation time for this vdev */
1043 vd
->vdev_crtxg
= crtxg
;
1045 if (!vd
->vdev_ops
->vdev_op_leaf
|| !spa_writeable(spa
))
1049 * Dead vdevs cannot be initialized.
1051 if (vdev_is_dead(vd
))
1052 return (SET_ERROR(EIO
));
1055 * Determine if the vdev is in use.
1057 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPLIT
&&
1058 vdev_inuse(vd
, crtxg
, reason
, &spare_guid
, &l2cache_guid
))
1059 return (SET_ERROR(EBUSY
));
1062 * If this is a request to add or replace a spare or l2cache device
1063 * that is in use elsewhere on the system, then we must update the
1064 * guid (which was initialized to a random value) to reflect the
1065 * actual GUID (which is shared between multiple pools).
1067 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_L2CACHE
&&
1068 spare_guid
!= 0ULL) {
1069 uint64_t guid_delta
= spare_guid
- vd
->vdev_guid
;
1071 vd
->vdev_guid
+= guid_delta
;
1073 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
1074 pvd
->vdev_guid_sum
+= guid_delta
;
1077 * If this is a replacement, then we want to fallthrough to the
1078 * rest of the code. If we're adding a spare, then it's already
1079 * labeled appropriately and we can just return.
1081 if (reason
== VDEV_LABEL_SPARE
)
1083 ASSERT(reason
== VDEV_LABEL_REPLACE
||
1084 reason
== VDEV_LABEL_SPLIT
);
1087 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPARE
&&
1088 l2cache_guid
!= 0ULL) {
1089 uint64_t guid_delta
= l2cache_guid
- vd
->vdev_guid
;
1091 vd
->vdev_guid
+= guid_delta
;
1093 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
1094 pvd
->vdev_guid_sum
+= guid_delta
;
1097 * If this is a replacement, then we want to fallthrough to the
1098 * rest of the code. If we're adding an l2cache, then it's
1099 * already labeled appropriately and we can just return.
1101 if (reason
== VDEV_LABEL_L2CACHE
)
1103 ASSERT(reason
== VDEV_LABEL_REPLACE
);
1107 * Initialize its label.
1109 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1110 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1111 vp
= abd_to_buf(vp_abd
);
1114 * Generate a label describing the pool and our top-level vdev.
1115 * We mark it as being from txg 0 to indicate that it's not
1116 * really part of an active pool just yet. The labels will
1117 * be written again with a meaningful txg by spa_sync().
1119 if (reason
== VDEV_LABEL_SPARE
||
1120 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isspare
)) {
1122 * For inactive hot spares, we generate a special label that
1123 * identifies as a mutually shared hot spare. We write the
1124 * label if we are adding a hot spare, or if we are removing an
1125 * active hot spare (in which case we want to revert the
1128 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1130 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1131 spa_version(spa
)) == 0);
1132 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1133 POOL_STATE_SPARE
) == 0);
1134 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1135 vd
->vdev_guid
) == 0);
1136 } else if (reason
== VDEV_LABEL_L2CACHE
||
1137 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isl2cache
)) {
1139 * For level 2 ARC devices, add a special label.
1141 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1143 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1144 spa_version(spa
)) == 0);
1145 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1146 POOL_STATE_L2CACHE
) == 0);
1147 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1148 vd
->vdev_guid
) == 0);
1151 * This is merely to facilitate reporting the ashift of the
1152 * cache device through zdb. The actual retrieval of the
1153 * ashift (in vdev_alloc()) uses the nvlist
1154 * spa->spa_l2cache->sav_config (populated in
1155 * spa_ld_open_aux_vdevs()).
1157 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_ASHIFT
,
1158 vd
->vdev_ashift
) == 0);
1160 uint64_t txg
= 0ULL;
1162 if (reason
== VDEV_LABEL_SPLIT
)
1163 txg
= spa
->spa_uberblock
.ub_txg
;
1164 label
= spa_config_generate(spa
, vd
, txg
, B_FALSE
);
1167 * Add our creation time. This allows us to detect multiple
1168 * vdev uses as described above, and automatically expires if we
1171 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
1175 buf
= vp
->vp_nvlist
;
1176 buflen
= sizeof (vp
->vp_nvlist
);
1178 error
= nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
);
1182 /* EFAULT means nvlist_pack ran out of room */
1183 return (SET_ERROR(error
== EFAULT
? ENAMETOOLONG
: EINVAL
));
1187 * Initialize uberblock template.
1189 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_RING
, B_TRUE
);
1190 abd_copy_from_buf(ub_abd
, &spa
->spa_uberblock
, sizeof (uberblock_t
));
1191 abd_zero_off(ub_abd
, sizeof (uberblock_t
),
1192 VDEV_UBERBLOCK_RING
- sizeof (uberblock_t
));
1193 ub
= abd_to_buf(ub_abd
);
1196 /* Initialize the 2nd padding area. */
1197 bootenv
= abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
);
1198 abd_zero(bootenv
, VDEV_PAD_SIZE
);
1201 * Write everything in parallel.
1204 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1206 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1208 vdev_label_write(zio
, vd
, l
, vp_abd
,
1209 offsetof(vdev_label_t
, vl_vdev_phys
),
1210 sizeof (vdev_phys_t
), NULL
, NULL
, flags
);
1213 * Skip the 1st padding area.
1214 * Zero out the 2nd padding area where it might have
1215 * left over data from previous filesystem format.
1217 vdev_label_write(zio
, vd
, l
, bootenv
,
1218 offsetof(vdev_label_t
, vl_be
),
1219 VDEV_PAD_SIZE
, NULL
, NULL
, flags
);
1221 vdev_label_write(zio
, vd
, l
, ub_abd
,
1222 offsetof(vdev_label_t
, vl_uberblock
),
1223 VDEV_UBERBLOCK_RING
, NULL
, NULL
, flags
);
1226 error
= zio_wait(zio
);
1228 if (error
!= 0 && !(flags
& ZIO_FLAG_TRYHARD
)) {
1229 flags
|= ZIO_FLAG_TRYHARD
;
1239 * If this vdev hasn't been previously identified as a spare, then we
1240 * mark it as such only if a) we are labeling it as a spare, or b) it
1241 * exists as a spare elsewhere in the system. Do the same for
1242 * level 2 ARC devices.
1244 if (error
== 0 && !vd
->vdev_isspare
&&
1245 (reason
== VDEV_LABEL_SPARE
||
1246 spa_spare_exists(vd
->vdev_guid
, NULL
, NULL
)))
1249 if (error
== 0 && !vd
->vdev_isl2cache
&&
1250 (reason
== VDEV_LABEL_L2CACHE
||
1251 spa_l2cache_exists(vd
->vdev_guid
, NULL
)))
1252 spa_l2cache_add(vd
);
1258 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1259 * callback to finish, store our abd in the callback pointer. Otherwise, we
1260 * just free our abd and return.
1263 vdev_label_read_bootenv_done(zio_t
*zio
)
1265 zio_t
*rio
= zio
->io_private
;
1266 abd_t
**cbp
= rio
->io_private
;
1268 ASSERT3U(zio
->io_size
, ==, VDEV_PAD_SIZE
);
1270 if (zio
->io_error
== 0) {
1271 mutex_enter(&rio
->io_lock
);
1273 /* Will free this buffer in vdev_label_read_bootenv. */
1276 abd_free(zio
->io_abd
);
1278 mutex_exit(&rio
->io_lock
);
1280 abd_free(zio
->io_abd
);
1285 vdev_label_read_bootenv_impl(zio_t
*zio
, vdev_t
*vd
, int flags
)
1287 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1288 vdev_label_read_bootenv_impl(zio
, vd
->vdev_child
[c
], flags
);
1291 * We just use the first label that has a correct checksum; the
1292 * bootloader should have rewritten them all to be the same on boot,
1293 * and any changes we made since boot have been the same across all
1296 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1297 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1298 vdev_label_read(zio
, vd
, l
,
1299 abd_alloc_linear(VDEV_PAD_SIZE
, B_FALSE
),
1300 offsetof(vdev_label_t
, vl_be
), VDEV_PAD_SIZE
,
1301 vdev_label_read_bootenv_done
, zio
, flags
);
1307 vdev_label_read_bootenv(vdev_t
*rvd
, nvlist_t
*bootenv
)
1310 spa_t
*spa
= rvd
->vdev_spa
;
1312 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1313 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_TRYHARD
;
1316 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1318 zio_t
*zio
= zio_root(spa
, NULL
, &abd
, flags
);
1319 vdev_label_read_bootenv_impl(zio
, rvd
, flags
);
1320 int err
= zio_wait(zio
);
1324 vdev_boot_envblock_t
*vbe
= abd_to_buf(abd
);
1326 vbe
->vbe_version
= ntohll(vbe
->vbe_version
);
1327 switch (vbe
->vbe_version
) {
1330 * if we have textual data in vbe_bootenv, create nvlist
1331 * with key "envmap".
1333 fnvlist_add_uint64(bootenv
, BOOTENV_VERSION
, VB_RAW
);
1334 vbe
->vbe_bootenv
[sizeof (vbe
->vbe_bootenv
) - 1] = '\0';
1335 fnvlist_add_string(bootenv
, GRUB_ENVMAP
,
1340 err
= nvlist_unpack(vbe
->vbe_bootenv
,
1341 sizeof (vbe
->vbe_bootenv
), &config
, 0);
1343 fnvlist_merge(bootenv
, config
);
1344 nvlist_free(config
);
1349 /* Check for FreeBSD zfs bootonce command string */
1350 buf
= abd_to_buf(abd
);
1352 fnvlist_add_uint64(bootenv
, BOOTENV_VERSION
,
1356 fnvlist_add_string(bootenv
, FREEBSD_BOOTONCE
, buf
);
1360 * abd was allocated in vdev_label_read_bootenv_impl()
1364 * If we managed to read any successfully,
1373 vdev_label_write_bootenv(vdev_t
*vd
, nvlist_t
*env
)
1376 spa_t
*spa
= vd
->vdev_spa
;
1377 vdev_boot_envblock_t
*bootenv
;
1378 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1384 error
= nvlist_size(env
, &nvsize
, NV_ENCODE_XDR
);
1386 return (SET_ERROR(error
));
1388 if (nvsize
>= sizeof (bootenv
->vbe_bootenv
)) {
1389 return (SET_ERROR(E2BIG
));
1392 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1395 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1398 child_err
= vdev_label_write_bootenv(vd
->vdev_child
[c
], env
);
1400 * As long as any of the disks managed to write all of their
1401 * labels successfully, return success.
1407 if (!vd
->vdev_ops
->vdev_op_leaf
|| vdev_is_dead(vd
) ||
1408 !vdev_writeable(vd
)) {
1411 ASSERT3U(sizeof (*bootenv
), ==, VDEV_PAD_SIZE
);
1412 abd_t
*abd
= abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
);
1413 abd_zero(abd
, VDEV_PAD_SIZE
);
1415 bootenv
= abd_borrow_buf_copy(abd
, VDEV_PAD_SIZE
);
1416 nvbuf
= bootenv
->vbe_bootenv
;
1417 nvsize
= sizeof (bootenv
->vbe_bootenv
);
1419 bootenv
->vbe_version
= fnvlist_lookup_uint64(env
, BOOTENV_VERSION
);
1420 switch (bootenv
->vbe_version
) {
1422 if (nvlist_lookup_string(env
, GRUB_ENVMAP
, &tmp
) == 0) {
1423 (void) strlcpy(bootenv
->vbe_bootenv
, tmp
, nvsize
);
1429 error
= nvlist_pack(env
, &nvbuf
, &nvsize
, NV_ENCODE_XDR
,
1439 bootenv
->vbe_version
= htonll(bootenv
->vbe_version
);
1440 abd_return_buf_copy(abd
, bootenv
, VDEV_PAD_SIZE
);
1443 return (SET_ERROR(error
));
1447 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1448 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1449 vdev_label_write(zio
, vd
, l
, abd
,
1450 offsetof(vdev_label_t
, vl_be
),
1451 VDEV_PAD_SIZE
, NULL
, NULL
, flags
);
1454 error
= zio_wait(zio
);
1455 if (error
!= 0 && !(flags
& ZIO_FLAG_TRYHARD
)) {
1456 flags
|= ZIO_FLAG_TRYHARD
;
1465 * ==========================================================================
1466 * uberblock load/sync
1467 * ==========================================================================
1471 * Consider the following situation: txg is safely synced to disk. We've
1472 * written the first uberblock for txg + 1, and then we lose power. When we
1473 * come back up, we fail to see the uberblock for txg + 1 because, say,
1474 * it was on a mirrored device and the replica to which we wrote txg + 1
1475 * is now offline. If we then make some changes and sync txg + 1, and then
1476 * the missing replica comes back, then for a few seconds we'll have two
1477 * conflicting uberblocks on disk with the same txg. The solution is simple:
1478 * among uberblocks with equal txg, choose the one with the latest timestamp.
1481 vdev_uberblock_compare(const uberblock_t
*ub1
, const uberblock_t
*ub2
)
1483 int cmp
= TREE_CMP(ub1
->ub_txg
, ub2
->ub_txg
);
1488 cmp
= TREE_CMP(ub1
->ub_timestamp
, ub2
->ub_timestamp
);
1493 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1494 * ZFS, e.g. OpenZFS >= 0.7.
1496 * If one ub has MMP and the other does not, they were written by
1497 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1500 * Since timestamp and txg are the same if we get this far, either is
1501 * acceptable for importing the pool.
1503 unsigned int seq1
= 0;
1504 unsigned int seq2
= 0;
1506 if (MMP_VALID(ub1
) && MMP_SEQ_VALID(ub1
))
1507 seq1
= MMP_SEQ(ub1
);
1509 if (MMP_VALID(ub2
) && MMP_SEQ_VALID(ub2
))
1510 seq2
= MMP_SEQ(ub2
);
1512 return (TREE_CMP(seq1
, seq2
));
1516 uberblock_t ubl_latest
; /* Most recent uberblock */
1517 uberblock_t
*ubl_ubbest
; /* Best uberblock (w/r/t max_txg) */
1518 vdev_t
*ubl_vd
; /* vdev associated with the above */
1522 vdev_uberblock_load_done(zio_t
*zio
)
1524 vdev_t
*vd
= zio
->io_vd
;
1525 spa_t
*spa
= zio
->io_spa
;
1526 zio_t
*rio
= zio
->io_private
;
1527 uberblock_t
*ub
= abd_to_buf(zio
->io_abd
);
1528 struct ubl_cbdata
*cbp
= rio
->io_private
;
1530 ASSERT3U(zio
->io_size
, ==, VDEV_UBERBLOCK_SIZE(vd
));
1532 if (zio
->io_error
== 0 && uberblock_verify(ub
) == 0) {
1533 mutex_enter(&rio
->io_lock
);
1534 if (vdev_uberblock_compare(ub
, &cbp
->ubl_latest
) > 0) {
1535 cbp
->ubl_latest
= *ub
;
1537 if (ub
->ub_txg
<= spa
->spa_load_max_txg
&&
1538 vdev_uberblock_compare(ub
, cbp
->ubl_ubbest
) > 0) {
1540 * Keep track of the vdev in which this uberblock
1541 * was found. We will use this information later
1542 * to obtain the config nvlist associated with
1545 *cbp
->ubl_ubbest
= *ub
;
1548 mutex_exit(&rio
->io_lock
);
1551 abd_free(zio
->io_abd
);
1555 vdev_uberblock_load_impl(zio_t
*zio
, vdev_t
*vd
, int flags
,
1556 struct ubl_cbdata
*cbp
)
1558 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1559 vdev_uberblock_load_impl(zio
, vd
->vdev_child
[c
], flags
, cbp
);
1561 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
) &&
1562 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
1563 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1564 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1565 vdev_label_read(zio
, vd
, l
,
1566 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
),
1567 B_TRUE
), VDEV_UBERBLOCK_OFFSET(vd
, n
),
1568 VDEV_UBERBLOCK_SIZE(vd
),
1569 vdev_uberblock_load_done
, zio
, flags
);
1576 * Reads the 'best' uberblock from disk along with its associated
1577 * configuration. First, we read the uberblock array of each label of each
1578 * vdev, keeping track of the uberblock with the highest txg in each array.
1579 * Then, we read the configuration from the same vdev as the best uberblock.
1582 vdev_uberblock_load(vdev_t
*rvd
, uberblock_t
*ub
, nvlist_t
**config
)
1585 spa_t
*spa
= rvd
->vdev_spa
;
1586 struct ubl_cbdata cb
;
1587 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1588 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_TRYHARD
;
1593 memset(ub
, 0, sizeof (uberblock_t
));
1594 memset(&cb
, 0, sizeof (cb
));
1599 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1600 zio
= zio_root(spa
, NULL
, &cb
, flags
);
1601 vdev_uberblock_load_impl(zio
, rvd
, flags
, &cb
);
1602 (void) zio_wait(zio
);
1605 * It's possible that the best uberblock was discovered on a label
1606 * that has a configuration which was written in a future txg.
1607 * Search all labels on this vdev to find the configuration that
1608 * matches the txg for our uberblock.
1610 if (cb
.ubl_vd
!= NULL
) {
1611 vdev_dbgmsg(cb
.ubl_vd
, "best uberblock found for spa %s. "
1612 "txg %llu", spa
->spa_name
, (u_longlong_t
)ub
->ub_txg
);
1614 if (ub
->ub_raidz_reflow_info
!=
1615 cb
.ubl_latest
.ub_raidz_reflow_info
) {
1616 vdev_dbgmsg(cb
.ubl_vd
,
1617 "spa=%s best uberblock (txg=%llu info=0x%llx) "
1618 "has different raidz_reflow_info than latest "
1619 "uberblock (txg=%llu info=0x%llx)",
1621 (u_longlong_t
)ub
->ub_txg
,
1622 (u_longlong_t
)ub
->ub_raidz_reflow_info
,
1623 (u_longlong_t
)cb
.ubl_latest
.ub_txg
,
1624 (u_longlong_t
)cb
.ubl_latest
.ub_raidz_reflow_info
);
1625 memset(ub
, 0, sizeof (uberblock_t
));
1626 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1630 *config
= vdev_label_read_config(cb
.ubl_vd
, ub
->ub_txg
);
1631 if (*config
== NULL
&& spa
->spa_extreme_rewind
) {
1632 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config. "
1633 "Trying again without txg restrictions.");
1634 *config
= vdev_label_read_config(cb
.ubl_vd
, UINT64_MAX
);
1636 if (*config
== NULL
) {
1637 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config");
1640 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1644 * For use when a leaf vdev is expanded.
1645 * The location of labels 2 and 3 changed, and at the new location the
1646 * uberblock rings are either empty or contain garbage. The sync will write
1647 * new configs there because the vdev is dirty, but expansion also needs the
1648 * uberblock rings copied. Read them from label 0 which did not move.
1650 * Since the point is to populate labels {2,3} with valid uberblocks,
1651 * we zero uberblocks we fail to read or which are not valid.
1655 vdev_copy_uberblocks(vdev_t
*vd
)
1659 int locks
= (SCL_L2ARC
| SCL_ZIO
);
1660 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1661 ZIO_FLAG_SPECULATIVE
;
1663 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_READER
) ==
1665 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1668 * No uberblocks are stored on distributed spares, they may be
1669 * safely skipped when expanding a leaf vdev.
1671 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
1674 spa_config_enter(vd
->vdev_spa
, locks
, FTAG
, RW_READER
);
1676 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1678 write_zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1679 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1680 const int src_label
= 0;
1683 zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1684 vdev_label_read(zio
, vd
, src_label
, ub_abd
,
1685 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1688 if (zio_wait(zio
) || uberblock_verify(abd_to_buf(ub_abd
)))
1689 abd_zero(ub_abd
, VDEV_UBERBLOCK_SIZE(vd
));
1691 for (int l
= 2; l
< VDEV_LABELS
; l
++)
1692 vdev_label_write(write_zio
, vd
, l
, ub_abd
,
1693 VDEV_UBERBLOCK_OFFSET(vd
, n
),
1694 VDEV_UBERBLOCK_SIZE(vd
), NULL
, NULL
,
1695 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1697 (void) zio_wait(write_zio
);
1699 spa_config_exit(vd
->vdev_spa
, locks
, FTAG
);
1705 * On success, increment root zio's count of good writes.
1706 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1709 vdev_uberblock_sync_done(zio_t
*zio
)
1711 uint64_t *good_writes
= zio
->io_private
;
1713 if (zio
->io_error
== 0 && zio
->io_vd
->vdev_top
->vdev_ms_array
!= 0)
1714 atomic_inc_64(good_writes
);
1718 * Write the uberblock to all labels of all leaves of the specified vdev.
1721 vdev_uberblock_sync(zio_t
*zio
, uint64_t *good_writes
,
1722 uberblock_t
*ub
, vdev_t
*vd
, int flags
)
1724 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
1725 vdev_uberblock_sync(zio
, good_writes
,
1726 ub
, vd
->vdev_child
[c
], flags
);
1729 if (!vd
->vdev_ops
->vdev_op_leaf
)
1732 if (!vdev_writeable(vd
))
1736 * There's no need to write uberblocks to a distributed spare, they
1737 * are already stored on all the leaves of the parent dRAID. For
1738 * this same reason vdev_uberblock_load_impl() skips distributed
1739 * spares when reading uberblocks.
1741 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
1744 /* If the vdev was expanded, need to copy uberblock rings. */
1745 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1746 vd
->vdev_copy_uberblocks
== B_TRUE
) {
1747 vdev_copy_uberblocks(vd
);
1748 vd
->vdev_copy_uberblocks
= B_FALSE
;
1752 * We chose a slot based on the txg. If this uberblock has a special
1753 * RAIDZ expansion state, then it is essentially an update of the
1754 * current uberblock (it has the same txg). However, the current
1755 * state is committed, so we want to write it to a different slot. If
1756 * we overwrote the same slot, and we lose power during the uberblock
1757 * write, and the disk does not do single-sector overwrites
1758 * atomically (even though it is required to - i.e. we should see
1759 * either the old or the new uberblock), then we could lose this
1760 * txg's uberblock. Rewinding to the previous txg's uberblock may not
1761 * be possible because RAIDZ expansion may have already overwritten
1762 * some of the data, so we need the progress indicator in the
1765 int m
= spa_multihost(vd
->vdev_spa
) ? MMP_BLOCKS_PER_LABEL
: 0;
1766 int n
= (ub
->ub_txg
- (RRSS_GET_STATE(ub
) == RRSS_SCRATCH_VALID
)) %
1767 (VDEV_UBERBLOCK_COUNT(vd
) - m
);
1769 /* Copy the uberblock_t into the ABD */
1770 abd_t
*ub_abd
= abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1771 abd_copy_from_buf(ub_abd
, ub
, sizeof (uberblock_t
));
1772 abd_zero_off(ub_abd
, sizeof (uberblock_t
),
1773 VDEV_UBERBLOCK_SIZE(vd
) - sizeof (uberblock_t
));
1775 for (int l
= 0; l
< VDEV_LABELS
; l
++)
1776 vdev_label_write(zio
, vd
, l
, ub_abd
,
1777 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1778 vdev_uberblock_sync_done
, good_writes
,
1779 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1784 /* Sync the uberblocks to all vdevs in svd[] */
1786 vdev_uberblock_sync_list(vdev_t
**svd
, int svdcount
, uberblock_t
*ub
, int flags
)
1788 spa_t
*spa
= svd
[0]->vdev_spa
;
1790 uint64_t good_writes
= 0;
1792 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1794 for (int v
= 0; v
< svdcount
; v
++)
1795 vdev_uberblock_sync(zio
, &good_writes
, ub
, svd
[v
], flags
);
1797 (void) zio_wait(zio
);
1800 * Flush the uberblocks to disk. This ensures that the odd labels
1801 * are no longer needed (because the new uberblocks and the even
1802 * labels are safely on disk), so it is safe to overwrite them.
1804 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1806 for (int v
= 0; v
< svdcount
; v
++) {
1807 if (vdev_writeable(svd
[v
])) {
1808 zio_flush(zio
, svd
[v
]);
1812 (void) zio_wait(zio
);
1814 return (good_writes
>= 1 ? 0 : EIO
);
1818 * On success, increment the count of good writes for our top-level vdev.
1821 vdev_label_sync_done(zio_t
*zio
)
1823 uint64_t *good_writes
= zio
->io_private
;
1825 if (zio
->io_error
== 0)
1826 atomic_inc_64(good_writes
);
1830 * If there weren't enough good writes, indicate failure to the parent.
1833 vdev_label_sync_top_done(zio_t
*zio
)
1835 uint64_t *good_writes
= zio
->io_private
;
1837 if (*good_writes
== 0)
1838 zio
->io_error
= SET_ERROR(EIO
);
1840 kmem_free(good_writes
, sizeof (uint64_t));
1844 * We ignore errors for log and cache devices, simply free the private data.
1847 vdev_label_sync_ignore_done(zio_t
*zio
)
1849 kmem_free(zio
->io_private
, sizeof (uint64_t));
1853 * Write all even or odd labels to all leaves of the specified vdev.
1856 vdev_label_sync(zio_t
*zio
, uint64_t *good_writes
,
1857 vdev_t
*vd
, int l
, uint64_t txg
, int flags
)
1865 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1866 vdev_label_sync(zio
, good_writes
,
1867 vd
->vdev_child
[c
], l
, txg
, flags
);
1870 if (!vd
->vdev_ops
->vdev_op_leaf
)
1873 if (!vdev_writeable(vd
))
1877 * The top-level config never needs to be written to a distributed
1878 * spare. When read vdev_dspare_label_read_config() will generate
1879 * the config for the vdev_label_read_config().
1881 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
1885 * Generate a label describing the top-level config to which we belong.
1887 label
= spa_config_generate(vd
->vdev_spa
, vd
, txg
, B_FALSE
);
1889 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1890 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1891 vp
= abd_to_buf(vp_abd
);
1893 buf
= vp
->vp_nvlist
;
1894 buflen
= sizeof (vp
->vp_nvlist
);
1896 if (!nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
)) {
1897 for (; l
< VDEV_LABELS
; l
+= 2) {
1898 vdev_label_write(zio
, vd
, l
, vp_abd
,
1899 offsetof(vdev_label_t
, vl_vdev_phys
),
1900 sizeof (vdev_phys_t
),
1901 vdev_label_sync_done
, good_writes
,
1902 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1911 vdev_label_sync_list(spa_t
*spa
, int l
, uint64_t txg
, int flags
)
1913 list_t
*dl
= &spa
->spa_config_dirty_list
;
1919 * Write the new labels to disk.
1921 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1923 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
)) {
1924 uint64_t *good_writes
;
1926 ASSERT(!vd
->vdev_ishole
);
1928 good_writes
= kmem_zalloc(sizeof (uint64_t), KM_SLEEP
);
1929 zio_t
*vio
= zio_null(zio
, spa
, NULL
,
1930 (vd
->vdev_islog
|| vd
->vdev_aux
!= NULL
) ?
1931 vdev_label_sync_ignore_done
: vdev_label_sync_top_done
,
1932 good_writes
, flags
);
1933 vdev_label_sync(vio
, good_writes
, vd
, l
, txg
, flags
);
1937 error
= zio_wait(zio
);
1940 * Flush the new labels to disk.
1942 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1944 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
))
1947 (void) zio_wait(zio
);
1953 * Sync the uberblock and any changes to the vdev configuration.
1955 * The order of operations is carefully crafted to ensure that
1956 * if the system panics or loses power at any time, the state on disk
1957 * is still transactionally consistent. The in-line comments below
1958 * describe the failure semantics at each stage.
1960 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1961 * at any time, you can just call it again, and it will resume its work.
1964 vdev_config_sync(vdev_t
**svd
, int svdcount
, uint64_t txg
)
1966 spa_t
*spa
= svd
[0]->vdev_spa
;
1967 uberblock_t
*ub
= &spa
->spa_uberblock
;
1969 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1971 ASSERT(svdcount
!= 0);
1974 * Normally, we don't want to try too hard to write every label and
1975 * uberblock. If there is a flaky disk, we don't want the rest of the
1976 * sync process to block while we retry. But if we can't write a
1977 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1978 * bailing out and declaring the pool faulted.
1981 if ((flags
& ZIO_FLAG_TRYHARD
) != 0)
1983 flags
|= ZIO_FLAG_TRYHARD
;
1986 ASSERT(ub
->ub_txg
<= txg
);
1989 * If this isn't a resync due to I/O errors,
1990 * and nothing changed in this transaction group,
1991 * and the vdev configuration hasn't changed,
1992 * then there's nothing to do.
1994 if (ub
->ub_txg
< txg
) {
1995 boolean_t changed
= uberblock_update(ub
, spa
->spa_root_vdev
,
1996 txg
, spa
->spa_mmp
.mmp_delay
);
1998 if (!changed
&& list_is_empty(&spa
->spa_config_dirty_list
))
2002 if (txg
> spa_freeze_txg(spa
))
2005 ASSERT(txg
<= spa
->spa_final_txg
);
2008 * Flush the write cache of every disk that's been written to
2009 * in this transaction group. This ensures that all blocks
2010 * written in this txg will be committed to stable storage
2011 * before any uberblock that references them.
2013 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, flags
);
2016 txg_list_head(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)); vd
!= NULL
;
2017 vd
= txg_list_next(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
)))
2020 (void) zio_wait(zio
);
2023 * Sync out the even labels (L0, L2) for every dirty vdev. If the
2024 * system dies in the middle of this process, that's OK: all of the
2025 * even labels that made it to disk will be newer than any uberblock,
2026 * and will therefore be considered invalid. The odd labels (L1, L3),
2027 * which have not yet been touched, will still be valid. We flush
2028 * the new labels to disk to ensure that all even-label updates
2029 * are committed to stable storage before the uberblock update.
2031 if ((error
= vdev_label_sync_list(spa
, 0, txg
, flags
)) != 0) {
2032 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
2033 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2034 "for pool '%s' when syncing out the even labels "
2035 "of dirty vdevs", error
, spa_name(spa
));
2041 * Sync the uberblocks to all vdevs in svd[].
2042 * If the system dies in the middle of this step, there are two cases
2043 * to consider, and the on-disk state is consistent either way:
2045 * (1) If none of the new uberblocks made it to disk, then the
2046 * previous uberblock will be the newest, and the odd labels
2047 * (which had not yet been touched) will be valid with respect
2048 * to that uberblock.
2050 * (2) If one or more new uberblocks made it to disk, then they
2051 * will be the newest, and the even labels (which had all
2052 * been successfully committed) will be valid with respect
2053 * to the new uberblocks.
2055 if ((error
= vdev_uberblock_sync_list(svd
, svdcount
, ub
, flags
)) != 0) {
2056 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
2057 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2058 "%d for pool '%s'", error
, spa_name(spa
));
2063 if (spa_multihost(spa
))
2064 mmp_update_uberblock(spa
, ub
);
2067 * Sync out odd labels for every dirty vdev. If the system dies
2068 * in the middle of this process, the even labels and the new
2069 * uberblocks will suffice to open the pool. The next time
2070 * the pool is opened, the first thing we'll do -- before any
2071 * user data is modified -- is mark every vdev dirty so that
2072 * all labels will be brought up to date. We flush the new labels
2073 * to disk to ensure that all odd-label updates are committed to
2074 * stable storage before the next transaction group begins.
2076 if ((error
= vdev_label_sync_list(spa
, 1, txg
, flags
)) != 0) {
2077 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
2078 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2079 "for pool '%s' when syncing out the odd labels of "
2080 "dirty vdevs", error
, spa_name(spa
));