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, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
26 * Copyright (c) 2017, Intel Corporation.
30 * Virtual Device Labels
31 * ---------------------
33 * The vdev label serves several distinct purposes:
35 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
36 * identity within the pool.
38 * 2. Verify that all the devices given in a configuration are present
41 * 3. Determine the uberblock for the pool.
43 * 4. In case of an import operation, determine the configuration of the
44 * toplevel vdev of which it is a part.
46 * 5. If an import operation cannot find all the devices in the pool,
47 * provide enough information to the administrator to determine which
48 * devices are missing.
50 * It is important to note that while the kernel is responsible for writing the
51 * label, it only consumes the information in the first three cases. The
52 * latter information is only consumed in userland when determining the
53 * configuration to import a pool.
59 * Before describing the contents of the label, it's important to understand how
60 * the labels are written and updated with respect to the uberblock.
62 * When the pool configuration is altered, either because it was newly created
63 * or a device was added, we want to update all the labels such that we can deal
64 * with fatal failure at any point. To this end, each disk has two labels which
65 * are updated before and after the uberblock is synced. Assuming we have
66 * labels and an uberblock with the following transaction groups:
69 * +------+ +------+ +------+
71 * | t10 | | t10 | | t10 |
73 * +------+ +------+ +------+
75 * In this stable state, the labels and the uberblock were all updated within
76 * the same transaction group (10). Each label is mirrored and checksummed, so
77 * that we can detect when we fail partway through writing the label.
79 * In order to identify which labels are valid, the labels are written in the
82 * 1. For each vdev, update 'L1' to the new label
83 * 2. Update the uberblock
84 * 3. For each vdev, update 'L2' to the new label
86 * Given arbitrary failure, we can determine the correct label to use based on
87 * the transaction group. If we fail after updating L1 but before updating the
88 * UB, we will notice that L1's transaction group is greater than the uberblock,
89 * so L2 must be valid. If we fail after writing the uberblock but before
90 * writing L2, we will notice that L2's transaction group is less than L1, and
91 * therefore L1 is valid.
93 * Another added complexity is that not every label is updated when the config
94 * is synced. If we add a single device, we do not want to have to re-write
95 * every label for every device in the pool. This means that both L1 and L2 may
96 * be older than the pool uberblock, because the necessary information is stored
103 * The vdev label consists of two distinct parts, and is wrapped within the
104 * vdev_label_t structure. The label includes 8k of padding to permit legacy
105 * VTOC disk labels, but is otherwise ignored.
107 * The first half of the label is a packed nvlist which contains pool wide
108 * properties, per-vdev properties, and configuration information. It is
109 * described in more detail below.
111 * The latter half of the label consists of a redundant array of uberblocks.
112 * These uberblocks are updated whenever a transaction group is committed,
113 * or when the configuration is updated. When a pool is loaded, we scan each
114 * vdev for the 'best' uberblock.
117 * Configuration Information
118 * -------------------------
120 * The nvlist describing the pool and vdev contains the following elements:
122 * version ZFS on-disk version
125 * txg Transaction group in which this label was written
126 * pool_guid Unique identifier for this pool
127 * vdev_tree An nvlist describing vdev tree.
129 * An nvlist of the features necessary for reading the MOS.
131 * Each leaf device label also contains the following:
133 * top_guid Unique ID for top-level vdev in which this is contained
134 * guid Unique ID for the leaf vdev
136 * The 'vs' configuration follows the format described in 'spa_config.c'.
139 #include <sys/zfs_context.h>
141 #include <sys/spa_impl.h>
144 #include <sys/vdev.h>
145 #include <sys/vdev_impl.h>
146 #include <sys/uberblock_impl.h>
147 #include <sys/metaslab.h>
148 #include <sys/metaslab_impl.h>
150 #include <sys/dsl_scan.h>
152 #include <sys/fs/zfs.h>
155 * Basic routines to read and write from a vdev label.
156 * Used throughout the rest of this file.
159 vdev_label_offset(uint64_t psize
, int l
, uint64_t offset
)
161 ASSERT(offset
< sizeof (vdev_label_t
));
162 ASSERT(P2PHASE_TYPED(psize
, sizeof (vdev_label_t
), uint64_t) == 0);
164 return (offset
+ l
* sizeof (vdev_label_t
) + (l
< VDEV_LABELS
/ 2 ?
165 0 : psize
- VDEV_LABELS
* sizeof (vdev_label_t
)));
169 * Returns back the vdev label associated with the passed in offset.
172 vdev_label_number(uint64_t psize
, uint64_t offset
)
176 if (offset
>= psize
- VDEV_LABEL_END_SIZE
) {
177 offset
-= psize
- VDEV_LABEL_END_SIZE
;
178 offset
+= (VDEV_LABELS
/ 2) * sizeof (vdev_label_t
);
180 l
= offset
/ sizeof (vdev_label_t
);
181 return (l
< VDEV_LABELS
? l
: -1);
185 vdev_label_read(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
186 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
189 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
190 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
191 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
193 zio_nowait(zio_read_phys(zio
, vd
,
194 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
195 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
196 ZIO_PRIORITY_SYNC_READ
, flags
, B_TRUE
));
200 vdev_label_write(zio_t
*zio
, vdev_t
*vd
, int l
, abd_t
*buf
, uint64_t offset
,
201 uint64_t size
, zio_done_func_t
*done
, void *private, int flags
)
204 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_READER
) == SCL_STATE
||
205 spa_config_held(zio
->io_spa
, SCL_STATE
, RW_WRITER
) == SCL_STATE
);
206 ASSERT(flags
& ZIO_FLAG_CONFIG_WRITER
);
208 zio_nowait(zio_write_phys(zio
, vd
,
209 vdev_label_offset(vd
->vdev_psize
, l
, offset
),
210 size
, buf
, ZIO_CHECKSUM_LABEL
, done
, private,
211 ZIO_PRIORITY_SYNC_WRITE
, flags
, B_TRUE
));
215 * Generate the nvlist representing this vdev's stats
218 vdev_config_generate_stats(vdev_t
*vd
, nvlist_t
*nv
)
224 vs
= kmem_alloc(sizeof (*vs
), KM_SLEEP
);
225 vsx
= kmem_alloc(sizeof (*vsx
), KM_SLEEP
);
227 vdev_get_stats_ex(vd
, vs
, vsx
);
228 fnvlist_add_uint64_array(nv
, ZPOOL_CONFIG_VDEV_STATS
,
229 (uint64_t *)vs
, sizeof (*vs
) / sizeof (uint64_t));
232 * Add extended stats into a special extended stats nvlist. This keeps
233 * all the extended stats nicely grouped together. The extended stats
234 * nvlist is then added to the main nvlist.
236 nvx
= fnvlist_alloc();
238 /* ZIOs in flight to disk */
239 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE
,
240 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_READ
]);
242 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE
,
243 vsx
->vsx_active_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
245 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE
,
246 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_READ
]);
248 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE
,
249 vsx
->vsx_active_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
251 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE
,
252 vsx
->vsx_active_queue
[ZIO_PRIORITY_SCRUB
]);
255 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE
,
256 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_READ
]);
258 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE
,
259 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SYNC_WRITE
]);
261 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE
,
262 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_READ
]);
264 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE
,
265 vsx
->vsx_pend_queue
[ZIO_PRIORITY_ASYNC_WRITE
]);
267 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE
,
268 vsx
->vsx_pend_queue
[ZIO_PRIORITY_SCRUB
]);
271 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO
,
272 vsx
->vsx_total_histo
[ZIO_TYPE_READ
],
273 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_READ
]));
275 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO
,
276 vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
],
277 ARRAY_SIZE(vsx
->vsx_total_histo
[ZIO_TYPE_WRITE
]));
279 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO
,
280 vsx
->vsx_disk_histo
[ZIO_TYPE_READ
],
281 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_READ
]));
283 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO
,
284 vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
],
285 ARRAY_SIZE(vsx
->vsx_disk_histo
[ZIO_TYPE_WRITE
]));
287 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO
,
288 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
],
289 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_READ
]));
291 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO
,
292 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
],
293 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
295 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO
,
296 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
],
297 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_READ
]));
299 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO
,
300 vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
301 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
303 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO
,
304 vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
],
305 ARRAY_SIZE(vsx
->vsx_queue_histo
[ZIO_PRIORITY_SCRUB
]));
308 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO
,
309 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
],
310 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_READ
]));
312 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO
,
313 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
],
314 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
316 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO
,
317 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
],
318 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_READ
]));
320 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO
,
321 vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
322 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
324 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO
,
325 vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
],
326 ARRAY_SIZE(vsx
->vsx_ind_histo
[ZIO_PRIORITY_SCRUB
]));
328 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO
,
329 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
],
330 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_READ
]));
332 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO
,
333 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
],
334 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SYNC_WRITE
]));
336 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO
,
337 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
],
338 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_READ
]));
340 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO
,
341 vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
],
342 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_ASYNC_WRITE
]));
344 fnvlist_add_uint64_array(nvx
, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO
,
345 vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
],
346 ARRAY_SIZE(vsx
->vsx_agg_histo
[ZIO_PRIORITY_SCRUB
]));
349 fnvlist_add_uint64(nvx
, ZPOOL_CONFIG_VDEV_SLOW_IOS
, vs
->vs_slow_ios
);
351 /* Add extended stats nvlist to main nvlist */
352 fnvlist_add_nvlist(nv
, ZPOOL_CONFIG_VDEV_STATS_EX
, nvx
);
355 kmem_free(vs
, sizeof (*vs
));
356 kmem_free(vsx
, sizeof (*vsx
));
360 root_vdev_actions_getprogress(vdev_t
*vd
, nvlist_t
*nvl
)
362 spa_t
*spa
= vd
->vdev_spa
;
364 if (vd
!= spa
->spa_root_vdev
)
367 /* provide either current or previous scan information */
369 if (spa_scan_get_stats(spa
, &ps
) == 0) {
370 fnvlist_add_uint64_array(nvl
,
371 ZPOOL_CONFIG_SCAN_STATS
, (uint64_t *)&ps
,
372 sizeof (pool_scan_stat_t
) / sizeof (uint64_t));
375 pool_removal_stat_t prs
;
376 if (spa_removal_get_stats(spa
, &prs
) == 0) {
377 fnvlist_add_uint64_array(nvl
,
378 ZPOOL_CONFIG_REMOVAL_STATS
, (uint64_t *)&prs
,
379 sizeof (prs
) / sizeof (uint64_t));
382 pool_checkpoint_stat_t pcs
;
383 if (spa_checkpoint_get_stats(spa
, &pcs
) == 0) {
384 fnvlist_add_uint64_array(nvl
,
385 ZPOOL_CONFIG_CHECKPOINT_STATS
, (uint64_t *)&pcs
,
386 sizeof (pcs
) / sizeof (uint64_t));
391 * Generate the nvlist representing this vdev's config.
394 vdev_config_generate(spa_t
*spa
, vdev_t
*vd
, boolean_t getstats
,
395 vdev_config_flag_t flags
)
398 vdev_indirect_config_t
*vic
= &vd
->vdev_indirect_config
;
400 nv
= fnvlist_alloc();
402 fnvlist_add_string(nv
, ZPOOL_CONFIG_TYPE
, vd
->vdev_ops
->vdev_op_type
);
403 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)))
404 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ID
, vd
->vdev_id
);
405 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_GUID
, vd
->vdev_guid
);
407 if (vd
->vdev_path
!= NULL
)
408 fnvlist_add_string(nv
, ZPOOL_CONFIG_PATH
, vd
->vdev_path
);
410 if (vd
->vdev_devid
!= NULL
)
411 fnvlist_add_string(nv
, ZPOOL_CONFIG_DEVID
, vd
->vdev_devid
);
413 if (vd
->vdev_physpath
!= NULL
)
414 fnvlist_add_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
417 if (vd
->vdev_enc_sysfs_path
!= NULL
)
418 fnvlist_add_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
419 vd
->vdev_enc_sysfs_path
);
421 if (vd
->vdev_fru
!= NULL
)
422 fnvlist_add_string(nv
, ZPOOL_CONFIG_FRU
, vd
->vdev_fru
);
424 if (vd
->vdev_nparity
!= 0) {
425 ASSERT(strcmp(vd
->vdev_ops
->vdev_op_type
,
426 VDEV_TYPE_RAIDZ
) == 0);
429 * Make sure someone hasn't managed to sneak a fancy new vdev
430 * into a crufty old storage pool.
432 ASSERT(vd
->vdev_nparity
== 1 ||
433 (vd
->vdev_nparity
<= 2 &&
434 spa_version(spa
) >= SPA_VERSION_RAIDZ2
) ||
435 (vd
->vdev_nparity
<= 3 &&
436 spa_version(spa
) >= SPA_VERSION_RAIDZ3
));
439 * Note that we'll add the nparity tag even on storage pools
440 * that only support a single parity device -- older software
441 * will just ignore it.
443 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NPARITY
, vd
->vdev_nparity
);
446 if (vd
->vdev_wholedisk
!= -1ULL)
447 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
450 if (vd
->vdev_not_present
&& !(flags
& VDEV_CONFIG_MISSING
))
451 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
, 1);
453 if (vd
->vdev_isspare
)
454 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
, 1);
456 if (!(flags
& (VDEV_CONFIG_SPARE
| VDEV_CONFIG_L2CACHE
)) &&
457 vd
== vd
->vdev_top
) {
458 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
460 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
462 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, vd
->vdev_ashift
);
463 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
465 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, vd
->vdev_islog
);
466 if (vd
->vdev_removing
) {
467 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
471 /* zpool command expects alloc class data */
472 if (getstats
&& vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
) {
473 const char *bias
= NULL
;
475 switch (vd
->vdev_alloc_bias
) {
477 bias
= VDEV_ALLOC_BIAS_LOG
;
479 case VDEV_BIAS_SPECIAL
:
480 bias
= VDEV_ALLOC_BIAS_SPECIAL
;
482 case VDEV_BIAS_DEDUP
:
483 bias
= VDEV_ALLOC_BIAS_DEDUP
;
486 ASSERT3U(vd
->vdev_alloc_bias
, ==,
489 fnvlist_add_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
494 if (vd
->vdev_dtl_sm
!= NULL
) {
495 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DTL
,
496 space_map_object(vd
->vdev_dtl_sm
));
499 if (vic
->vic_mapping_object
!= 0) {
500 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
501 vic
->vic_mapping_object
);
504 if (vic
->vic_births_object
!= 0) {
505 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
506 vic
->vic_births_object
);
509 if (vic
->vic_prev_indirect_vdev
!= UINT64_MAX
) {
510 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
511 vic
->vic_prev_indirect_vdev
);
515 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
, vd
->vdev_crtxg
);
517 if (vd
->vdev_expansion_time
)
518 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_EXPANSION_TIME
,
519 vd
->vdev_expansion_time
);
521 if (flags
& VDEV_CONFIG_MOS
) {
522 if (vd
->vdev_leaf_zap
!= 0) {
523 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
524 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_LEAF_ZAP
,
528 if (vd
->vdev_top_zap
!= 0) {
529 ASSERT(vd
== vd
->vdev_top
);
530 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
534 if (vd
->vdev_resilver_deferred
) {
535 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
536 ASSERT(spa
->spa_resilver_deferred
);
537 fnvlist_add_boolean(nv
, ZPOOL_CONFIG_RESILVER_DEFER
);
542 vdev_config_generate_stats(vd
, nv
);
544 root_vdev_actions_getprogress(vd
, nv
);
547 * Note: this can be called from open context
548 * (spa_get_stats()), so we need the rwlock to prevent
549 * the mapping from being changed by condensing.
551 rw_enter(&vd
->vdev_indirect_rwlock
, RW_READER
);
552 if (vd
->vdev_indirect_mapping
!= NULL
) {
553 ASSERT(vd
->vdev_indirect_births
!= NULL
);
554 vdev_indirect_mapping_t
*vim
=
555 vd
->vdev_indirect_mapping
;
556 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
557 vdev_indirect_mapping_size(vim
));
559 rw_exit(&vd
->vdev_indirect_rwlock
);
560 if (vd
->vdev_mg
!= NULL
&&
561 vd
->vdev_mg
->mg_fragmentation
!= ZFS_FRAG_INVALID
) {
563 * Compute approximately how much memory would be used
564 * for the indirect mapping if this device were to
567 * Note: If the frag metric is invalid, then not
568 * enough metaslabs have been converted to have
571 uint64_t seg_count
= 0;
572 uint64_t to_alloc
= vd
->vdev_stat
.vs_alloc
;
575 * There are the same number of allocated segments
576 * as free segments, so we will have at least one
577 * entry per free segment. However, small free
578 * segments (smaller than vdev_removal_max_span)
579 * will be combined with adjacent allocated segments
580 * as a single mapping.
582 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
583 if (1ULL << (i
+ 1) < vdev_removal_max_span
) {
585 vd
->vdev_mg
->mg_histogram
[i
] <<
589 vd
->vdev_mg
->mg_histogram
[i
];
594 * The maximum length of a mapping is
595 * zfs_remove_max_segment, so we need at least one entry
596 * per zfs_remove_max_segment of allocated data.
598 seg_count
+= to_alloc
/ zfs_remove_max_segment
;
600 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_INDIRECT_SIZE
,
602 sizeof (vdev_indirect_mapping_entry_phys_t
));
606 if (!vd
->vdev_ops
->vdev_op_leaf
) {
610 ASSERT(!vd
->vdev_ishole
);
612 child
= kmem_alloc(vd
->vdev_children
* sizeof (nvlist_t
*),
615 for (c
= 0, idx
= 0; c
< vd
->vdev_children
; c
++) {
616 vdev_t
*cvd
= vd
->vdev_child
[c
];
619 * If we're generating an nvlist of removing
620 * vdevs then skip over any device which is
623 if ((flags
& VDEV_CONFIG_REMOVING
) &&
627 child
[idx
++] = vdev_config_generate(spa
, cvd
,
632 fnvlist_add_nvlist_array(nv
, ZPOOL_CONFIG_CHILDREN
,
636 for (c
= 0; c
< idx
; c
++)
637 nvlist_free(child
[c
]);
639 kmem_free(child
, vd
->vdev_children
* sizeof (nvlist_t
*));
642 const char *aux
= NULL
;
644 if (vd
->vdev_offline
&& !vd
->vdev_tmpoffline
)
645 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_OFFLINE
, B_TRUE
);
646 if (vd
->vdev_resilver_txg
!= 0)
647 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
648 vd
->vdev_resilver_txg
);
649 if (vd
->vdev_faulted
)
650 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_FAULTED
, B_TRUE
);
651 if (vd
->vdev_degraded
)
652 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_DEGRADED
, B_TRUE
);
653 if (vd
->vdev_removed
)
654 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_REMOVED
, B_TRUE
);
655 if (vd
->vdev_unspare
)
656 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_UNSPARE
, B_TRUE
);
658 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_IS_HOLE
, B_TRUE
);
660 /* Set the reason why we're FAULTED/DEGRADED. */
661 switch (vd
->vdev_stat
.vs_aux
) {
662 case VDEV_AUX_ERR_EXCEEDED
:
663 aux
= "err_exceeded";
666 case VDEV_AUX_EXTERNAL
:
671 if (aux
!= NULL
&& !vd
->vdev_tmpoffline
) {
672 fnvlist_add_string(nv
, ZPOOL_CONFIG_AUX_STATE
, aux
);
675 * We're healthy - clear any previous AUX_STATE values.
677 if (nvlist_exists(nv
, ZPOOL_CONFIG_AUX_STATE
))
678 nvlist_remove_all(nv
, ZPOOL_CONFIG_AUX_STATE
);
681 if (vd
->vdev_splitting
&& vd
->vdev_orig_guid
!= 0LL) {
682 fnvlist_add_uint64(nv
, ZPOOL_CONFIG_ORIG_GUID
,
691 * Generate a view of the top-level vdevs. If we currently have holes
692 * in the namespace, then generate an array which contains a list of holey
693 * vdevs. Additionally, add the number of top-level children that currently
697 vdev_top_config_generate(spa_t
*spa
, nvlist_t
*config
)
699 vdev_t
*rvd
= spa
->spa_root_vdev
;
703 array
= kmem_alloc(rvd
->vdev_children
* sizeof (uint64_t), KM_SLEEP
);
705 for (c
= 0, idx
= 0; c
< rvd
->vdev_children
; c
++) {
706 vdev_t
*tvd
= rvd
->vdev_child
[c
];
708 if (tvd
->vdev_ishole
) {
714 VERIFY(nvlist_add_uint64_array(config
, ZPOOL_CONFIG_HOLE_ARRAY
,
718 VERIFY(nvlist_add_uint64(config
, ZPOOL_CONFIG_VDEV_CHILDREN
,
719 rvd
->vdev_children
) == 0);
721 kmem_free(array
, rvd
->vdev_children
* sizeof (uint64_t));
725 * Returns the configuration from the label of the given vdev. For vdevs
726 * which don't have a txg value stored on their label (i.e. spares/cache)
727 * or have not been completely initialized (txg = 0) just return
728 * the configuration from the first valid label we find. Otherwise,
729 * find the most up-to-date label that does not exceed the specified
733 vdev_label_read_config(vdev_t
*vd
, uint64_t txg
)
735 spa_t
*spa
= vd
->vdev_spa
;
736 nvlist_t
*config
= NULL
;
740 uint64_t best_txg
= 0;
741 uint64_t label_txg
= 0;
743 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
744 ZIO_FLAG_SPECULATIVE
;
746 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
748 if (!vdev_readable(vd
))
751 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
752 vp
= abd_to_buf(vp_abd
);
755 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
756 nvlist_t
*label
= NULL
;
758 zio
= zio_root(spa
, NULL
, NULL
, flags
);
760 vdev_label_read(zio
, vd
, l
, vp_abd
,
761 offsetof(vdev_label_t
, vl_vdev_phys
),
762 sizeof (vdev_phys_t
), NULL
, NULL
, flags
);
764 if (zio_wait(zio
) == 0 &&
765 nvlist_unpack(vp
->vp_nvlist
, sizeof (vp
->vp_nvlist
),
768 * Auxiliary vdevs won't have txg values in their
769 * labels and newly added vdevs may not have been
770 * completely initialized so just return the
771 * configuration from the first valid label we
774 error
= nvlist_lookup_uint64(label
,
775 ZPOOL_CONFIG_POOL_TXG
, &label_txg
);
776 if ((error
|| label_txg
== 0) && !config
) {
779 } else if (label_txg
<= txg
&& label_txg
> best_txg
) {
780 best_txg
= label_txg
;
782 config
= fnvlist_dup(label
);
792 if (config
== NULL
&& !(flags
& ZIO_FLAG_TRYHARD
)) {
793 flags
|= ZIO_FLAG_TRYHARD
;
798 * We found a valid label but it didn't pass txg restrictions.
800 if (config
== NULL
&& label_txg
!= 0) {
801 vdev_dbgmsg(vd
, "label discarded as txg is too large "
802 "(%llu > %llu)", (u_longlong_t
)label_txg
,
812 * Determine if a device is in use. The 'spare_guid' parameter will be filled
813 * in with the device guid if this spare is active elsewhere on the system.
816 vdev_inuse(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
,
817 uint64_t *spare_guid
, uint64_t *l2cache_guid
)
819 spa_t
*spa
= vd
->vdev_spa
;
820 uint64_t state
, pool_guid
, device_guid
, txg
, spare_pool
;
827 *l2cache_guid
= 0ULL;
830 * Read the label, if any, and perform some basic sanity checks.
832 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
)
835 (void) nvlist_lookup_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
838 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
840 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
,
841 &device_guid
) != 0) {
846 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
847 (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
,
849 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_TXG
,
858 * Check to see if this device indeed belongs to the pool it claims to
859 * be a part of. The only way this is allowed is if the device is a hot
860 * spare (which we check for later on).
862 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
863 !spa_guid_exists(pool_guid
, device_guid
) &&
864 !spa_spare_exists(device_guid
, NULL
, NULL
) &&
865 !spa_l2cache_exists(device_guid
, NULL
))
869 * If the transaction group is zero, then this an initialized (but
870 * unused) label. This is only an error if the create transaction
871 * on-disk is the same as the one we're using now, in which case the
872 * user has attempted to add the same vdev multiple times in the same
875 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
876 txg
== 0 && vdtxg
== crtxg
)
880 * Check to see if this is a spare device. We do an explicit check for
881 * spa_has_spare() here because it may be on our pending list of spares
882 * to add. We also check if it is an l2cache device.
884 if (spa_spare_exists(device_guid
, &spare_pool
, NULL
) ||
885 spa_has_spare(spa
, device_guid
)) {
887 *spare_guid
= device_guid
;
890 case VDEV_LABEL_CREATE
:
891 case VDEV_LABEL_L2CACHE
:
894 case VDEV_LABEL_REPLACE
:
895 return (!spa_has_spare(spa
, device_guid
) ||
898 case VDEV_LABEL_SPARE
:
899 return (spa_has_spare(spa
, device_guid
));
906 * Check to see if this is an l2cache device.
908 if (spa_l2cache_exists(device_guid
, NULL
))
912 * We can't rely on a pool's state if it's been imported
913 * read-only. Instead we look to see if the pools is marked
914 * read-only in the namespace and set the state to active.
916 if (state
!= POOL_STATE_SPARE
&& state
!= POOL_STATE_L2CACHE
&&
917 (spa
= spa_by_guid(pool_guid
, device_guid
)) != NULL
&&
918 spa_mode(spa
) == FREAD
)
919 state
= POOL_STATE_ACTIVE
;
922 * If the device is marked ACTIVE, then this device is in use by another
923 * pool on the system.
925 return (state
== POOL_STATE_ACTIVE
);
929 * Initialize a vdev label. We check to make sure each leaf device is not in
930 * use, and writable. We put down an initial label which we will later
931 * overwrite with a complete label. Note that it's important to do this
932 * sequentially, not in parallel, so that we catch cases of multiple use of the
933 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
937 vdev_label_init(vdev_t
*vd
, uint64_t crtxg
, vdev_labeltype_t reason
)
939 spa_t
*spa
= vd
->vdev_spa
;
950 uint64_t spare_guid
= 0, l2cache_guid
= 0;
951 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
953 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
955 for (int c
= 0; c
< vd
->vdev_children
; c
++)
956 if ((error
= vdev_label_init(vd
->vdev_child
[c
],
957 crtxg
, reason
)) != 0)
960 /* Track the creation time for this vdev */
961 vd
->vdev_crtxg
= crtxg
;
963 if (!vd
->vdev_ops
->vdev_op_leaf
|| !spa_writeable(spa
))
967 * Dead vdevs cannot be initialized.
969 if (vdev_is_dead(vd
))
970 return (SET_ERROR(EIO
));
973 * Determine if the vdev is in use.
975 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPLIT
&&
976 vdev_inuse(vd
, crtxg
, reason
, &spare_guid
, &l2cache_guid
))
977 return (SET_ERROR(EBUSY
));
980 * If this is a request to add or replace a spare or l2cache device
981 * that is in use elsewhere on the system, then we must update the
982 * guid (which was initialized to a random value) to reflect the
983 * actual GUID (which is shared between multiple pools).
985 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_L2CACHE
&&
986 spare_guid
!= 0ULL) {
987 uint64_t guid_delta
= spare_guid
- vd
->vdev_guid
;
989 vd
->vdev_guid
+= guid_delta
;
991 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
992 pvd
->vdev_guid_sum
+= guid_delta
;
995 * If this is a replacement, then we want to fallthrough to the
996 * rest of the code. If we're adding a spare, then it's already
997 * labeled appropriately and we can just return.
999 if (reason
== VDEV_LABEL_SPARE
)
1001 ASSERT(reason
== VDEV_LABEL_REPLACE
||
1002 reason
== VDEV_LABEL_SPLIT
);
1005 if (reason
!= VDEV_LABEL_REMOVE
&& reason
!= VDEV_LABEL_SPARE
&&
1006 l2cache_guid
!= 0ULL) {
1007 uint64_t guid_delta
= l2cache_guid
- vd
->vdev_guid
;
1009 vd
->vdev_guid
+= guid_delta
;
1011 for (vdev_t
*pvd
= vd
; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
1012 pvd
->vdev_guid_sum
+= guid_delta
;
1015 * If this is a replacement, then we want to fallthrough to the
1016 * rest of the code. If we're adding an l2cache, then it's
1017 * already labeled appropriately and we can just return.
1019 if (reason
== VDEV_LABEL_L2CACHE
)
1021 ASSERT(reason
== VDEV_LABEL_REPLACE
);
1025 * Initialize its label.
1027 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1028 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1029 vp
= abd_to_buf(vp_abd
);
1032 * Generate a label describing the pool and our top-level vdev.
1033 * We mark it as being from txg 0 to indicate that it's not
1034 * really part of an active pool just yet. The labels will
1035 * be written again with a meaningful txg by spa_sync().
1037 if (reason
== VDEV_LABEL_SPARE
||
1038 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isspare
)) {
1040 * For inactive hot spares, we generate a special label that
1041 * identifies as a mutually shared hot spare. We write the
1042 * label if we are adding a hot spare, or if we are removing an
1043 * active hot spare (in which case we want to revert the
1046 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1048 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1049 spa_version(spa
)) == 0);
1050 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1051 POOL_STATE_SPARE
) == 0);
1052 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1053 vd
->vdev_guid
) == 0);
1054 } else if (reason
== VDEV_LABEL_L2CACHE
||
1055 (reason
== VDEV_LABEL_REMOVE
&& vd
->vdev_isl2cache
)) {
1057 * For level 2 ARC devices, add a special label.
1059 VERIFY(nvlist_alloc(&label
, NV_UNIQUE_NAME
, KM_SLEEP
) == 0);
1061 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_VERSION
,
1062 spa_version(spa
)) == 0);
1063 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1064 POOL_STATE_L2CACHE
) == 0);
1065 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_GUID
,
1066 vd
->vdev_guid
) == 0);
1068 uint64_t txg
= 0ULL;
1070 if (reason
== VDEV_LABEL_SPLIT
)
1071 txg
= spa
->spa_uberblock
.ub_txg
;
1072 label
= spa_config_generate(spa
, vd
, txg
, B_FALSE
);
1075 * Add our creation time. This allows us to detect multiple
1076 * vdev uses as described above, and automatically expires if we
1079 VERIFY(nvlist_add_uint64(label
, ZPOOL_CONFIG_CREATE_TXG
,
1083 buf
= vp
->vp_nvlist
;
1084 buflen
= sizeof (vp
->vp_nvlist
);
1086 error
= nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
);
1090 /* EFAULT means nvlist_pack ran out of room */
1091 return (SET_ERROR(error
== EFAULT
? ENAMETOOLONG
: EINVAL
));
1095 * Initialize uberblock template.
1097 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_RING
, B_TRUE
);
1098 abd_zero(ub_abd
, VDEV_UBERBLOCK_RING
);
1099 abd_copy_from_buf(ub_abd
, &spa
->spa_uberblock
, sizeof (uberblock_t
));
1100 ub
= abd_to_buf(ub_abd
);
1103 /* Initialize the 2nd padding area. */
1104 pad2
= abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
);
1105 abd_zero(pad2
, VDEV_PAD_SIZE
);
1108 * Write everything in parallel.
1111 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1113 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1115 vdev_label_write(zio
, vd
, l
, vp_abd
,
1116 offsetof(vdev_label_t
, vl_vdev_phys
),
1117 sizeof (vdev_phys_t
), NULL
, NULL
, flags
);
1120 * Skip the 1st padding area.
1121 * Zero out the 2nd padding area where it might have
1122 * left over data from previous filesystem format.
1124 vdev_label_write(zio
, vd
, l
, pad2
,
1125 offsetof(vdev_label_t
, vl_pad2
),
1126 VDEV_PAD_SIZE
, NULL
, NULL
, flags
);
1128 vdev_label_write(zio
, vd
, l
, ub_abd
,
1129 offsetof(vdev_label_t
, vl_uberblock
),
1130 VDEV_UBERBLOCK_RING
, NULL
, NULL
, flags
);
1133 error
= zio_wait(zio
);
1135 if (error
!= 0 && !(flags
& ZIO_FLAG_TRYHARD
)) {
1136 flags
|= ZIO_FLAG_TRYHARD
;
1146 * If this vdev hasn't been previously identified as a spare, then we
1147 * mark it as such only if a) we are labeling it as a spare, or b) it
1148 * exists as a spare elsewhere in the system. Do the same for
1149 * level 2 ARC devices.
1151 if (error
== 0 && !vd
->vdev_isspare
&&
1152 (reason
== VDEV_LABEL_SPARE
||
1153 spa_spare_exists(vd
->vdev_guid
, NULL
, NULL
)))
1156 if (error
== 0 && !vd
->vdev_isl2cache
&&
1157 (reason
== VDEV_LABEL_L2CACHE
||
1158 spa_l2cache_exists(vd
->vdev_guid
, NULL
)))
1159 spa_l2cache_add(vd
);
1165 * ==========================================================================
1166 * uberblock load/sync
1167 * ==========================================================================
1171 * Consider the following situation: txg is safely synced to disk. We've
1172 * written the first uberblock for txg + 1, and then we lose power. When we
1173 * come back up, we fail to see the uberblock for txg + 1 because, say,
1174 * it was on a mirrored device and the replica to which we wrote txg + 1
1175 * is now offline. If we then make some changes and sync txg + 1, and then
1176 * the missing replica comes back, then for a few seconds we'll have two
1177 * conflicting uberblocks on disk with the same txg. The solution is simple:
1178 * among uberblocks with equal txg, choose the one with the latest timestamp.
1181 vdev_uberblock_compare(const uberblock_t
*ub1
, const uberblock_t
*ub2
)
1183 int cmp
= AVL_CMP(ub1
->ub_txg
, ub2
->ub_txg
);
1187 return (AVL_CMP(ub1
->ub_timestamp
, ub2
->ub_timestamp
));
1191 uberblock_t
*ubl_ubbest
; /* Best uberblock */
1192 vdev_t
*ubl_vd
; /* vdev associated with the above */
1196 vdev_uberblock_load_done(zio_t
*zio
)
1198 vdev_t
*vd
= zio
->io_vd
;
1199 spa_t
*spa
= zio
->io_spa
;
1200 zio_t
*rio
= zio
->io_private
;
1201 uberblock_t
*ub
= abd_to_buf(zio
->io_abd
);
1202 struct ubl_cbdata
*cbp
= rio
->io_private
;
1204 ASSERT3U(zio
->io_size
, ==, VDEV_UBERBLOCK_SIZE(vd
));
1206 if (zio
->io_error
== 0 && uberblock_verify(ub
) == 0) {
1207 mutex_enter(&rio
->io_lock
);
1208 if (ub
->ub_txg
<= spa
->spa_load_max_txg
&&
1209 vdev_uberblock_compare(ub
, cbp
->ubl_ubbest
) > 0) {
1211 * Keep track of the vdev in which this uberblock
1212 * was found. We will use this information later
1213 * to obtain the config nvlist associated with
1216 *cbp
->ubl_ubbest
= *ub
;
1219 mutex_exit(&rio
->io_lock
);
1222 abd_free(zio
->io_abd
);
1226 vdev_uberblock_load_impl(zio_t
*zio
, vdev_t
*vd
, int flags
,
1227 struct ubl_cbdata
*cbp
)
1229 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1230 vdev_uberblock_load_impl(zio
, vd
->vdev_child
[c
], flags
, cbp
);
1232 if (vd
->vdev_ops
->vdev_op_leaf
&& vdev_readable(vd
)) {
1233 for (int l
= 0; l
< VDEV_LABELS
; l
++) {
1234 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1235 vdev_label_read(zio
, vd
, l
,
1236 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
),
1237 B_TRUE
), VDEV_UBERBLOCK_OFFSET(vd
, n
),
1238 VDEV_UBERBLOCK_SIZE(vd
),
1239 vdev_uberblock_load_done
, zio
, flags
);
1246 * Reads the 'best' uberblock from disk along with its associated
1247 * configuration. First, we read the uberblock array of each label of each
1248 * vdev, keeping track of the uberblock with the highest txg in each array.
1249 * Then, we read the configuration from the same vdev as the best uberblock.
1252 vdev_uberblock_load(vdev_t
*rvd
, uberblock_t
*ub
, nvlist_t
**config
)
1255 spa_t
*spa
= rvd
->vdev_spa
;
1256 struct ubl_cbdata cb
;
1257 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1258 ZIO_FLAG_SPECULATIVE
| ZIO_FLAG_TRYHARD
;
1263 bzero(ub
, sizeof (uberblock_t
));
1269 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1270 zio
= zio_root(spa
, NULL
, &cb
, flags
);
1271 vdev_uberblock_load_impl(zio
, rvd
, flags
, &cb
);
1272 (void) zio_wait(zio
);
1275 * It's possible that the best uberblock was discovered on a label
1276 * that has a configuration which was written in a future txg.
1277 * Search all labels on this vdev to find the configuration that
1278 * matches the txg for our uberblock.
1280 if (cb
.ubl_vd
!= NULL
) {
1281 vdev_dbgmsg(cb
.ubl_vd
, "best uberblock found for spa %s. "
1282 "txg %llu", spa
->spa_name
, (u_longlong_t
)ub
->ub_txg
);
1284 *config
= vdev_label_read_config(cb
.ubl_vd
, ub
->ub_txg
);
1285 if (*config
== NULL
&& spa
->spa_extreme_rewind
) {
1286 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config. "
1287 "Trying again without txg restrictions.");
1288 *config
= vdev_label_read_config(cb
.ubl_vd
, UINT64_MAX
);
1290 if (*config
== NULL
) {
1291 vdev_dbgmsg(cb
.ubl_vd
, "failed to read label config");
1294 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1298 * For use when a leaf vdev is expanded.
1299 * The location of labels 2 and 3 changed, and at the new location the
1300 * uberblock rings are either empty or contain garbage. The sync will write
1301 * new configs there because the vdev is dirty, but expansion also needs the
1302 * uberblock rings copied. Read them from label 0 which did not move.
1304 * Since the point is to populate labels {2,3} with valid uberblocks,
1305 * we zero uberblocks we fail to read or which are not valid.
1309 vdev_copy_uberblocks(vdev_t
*vd
)
1313 int locks
= (SCL_L2ARC
| SCL_ZIO
);
1314 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
|
1315 ZIO_FLAG_SPECULATIVE
;
1317 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_STATE
, RW_READER
) ==
1319 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1321 spa_config_enter(vd
->vdev_spa
, locks
, FTAG
, RW_READER
);
1323 ub_abd
= abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1325 write_zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1326 for (int n
= 0; n
< VDEV_UBERBLOCK_COUNT(vd
); n
++) {
1327 const int src_label
= 0;
1330 zio
= zio_root(vd
->vdev_spa
, NULL
, NULL
, flags
);
1331 vdev_label_read(zio
, vd
, src_label
, ub_abd
,
1332 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1335 if (zio_wait(zio
) || uberblock_verify(abd_to_buf(ub_abd
)))
1336 abd_zero(ub_abd
, VDEV_UBERBLOCK_SIZE(vd
));
1338 for (int l
= 2; l
< VDEV_LABELS
; l
++)
1339 vdev_label_write(write_zio
, vd
, l
, ub_abd
,
1340 VDEV_UBERBLOCK_OFFSET(vd
, n
),
1341 VDEV_UBERBLOCK_SIZE(vd
), NULL
, NULL
,
1342 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1344 (void) zio_wait(write_zio
);
1346 spa_config_exit(vd
->vdev_spa
, locks
, FTAG
);
1352 * On success, increment root zio's count of good writes.
1353 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1356 vdev_uberblock_sync_done(zio_t
*zio
)
1358 uint64_t *good_writes
= zio
->io_private
;
1360 if (zio
->io_error
== 0 && zio
->io_vd
->vdev_top
->vdev_ms_array
!= 0)
1361 atomic_inc_64(good_writes
);
1365 * Write the uberblock to all labels of all leaves of the specified vdev.
1368 vdev_uberblock_sync(zio_t
*zio
, uint64_t *good_writes
,
1369 uberblock_t
*ub
, vdev_t
*vd
, int flags
)
1371 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++) {
1372 vdev_uberblock_sync(zio
, good_writes
,
1373 ub
, vd
->vdev_child
[c
], flags
);
1376 if (!vd
->vdev_ops
->vdev_op_leaf
)
1379 if (!vdev_writeable(vd
))
1382 /* If the vdev was expanded, need to copy uberblock rings. */
1383 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1384 vd
->vdev_copy_uberblocks
== B_TRUE
) {
1385 vdev_copy_uberblocks(vd
);
1386 vd
->vdev_copy_uberblocks
= B_FALSE
;
1389 int m
= spa_multihost(vd
->vdev_spa
) ? MMP_BLOCKS_PER_LABEL
: 0;
1390 int n
= ub
->ub_txg
% (VDEV_UBERBLOCK_COUNT(vd
) - m
);
1392 /* Copy the uberblock_t into the ABD */
1393 abd_t
*ub_abd
= abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd
), B_TRUE
);
1394 abd_zero(ub_abd
, VDEV_UBERBLOCK_SIZE(vd
));
1395 abd_copy_from_buf(ub_abd
, ub
, sizeof (uberblock_t
));
1397 for (int l
= 0; l
< VDEV_LABELS
; l
++)
1398 vdev_label_write(zio
, vd
, l
, ub_abd
,
1399 VDEV_UBERBLOCK_OFFSET(vd
, n
), VDEV_UBERBLOCK_SIZE(vd
),
1400 vdev_uberblock_sync_done
, good_writes
,
1401 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1406 /* Sync the uberblocks to all vdevs in svd[] */
1408 vdev_uberblock_sync_list(vdev_t
**svd
, int svdcount
, uberblock_t
*ub
, int flags
)
1410 spa_t
*spa
= svd
[0]->vdev_spa
;
1412 uint64_t good_writes
= 0;
1414 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1416 for (int v
= 0; v
< svdcount
; v
++)
1417 vdev_uberblock_sync(zio
, &good_writes
, ub
, svd
[v
], flags
);
1419 (void) zio_wait(zio
);
1422 * Flush the uberblocks to disk. This ensures that the odd labels
1423 * are no longer needed (because the new uberblocks and the even
1424 * labels are safely on disk), so it is safe to overwrite them.
1426 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1428 for (int v
= 0; v
< svdcount
; v
++) {
1429 if (vdev_writeable(svd
[v
])) {
1430 zio_flush(zio
, svd
[v
]);
1434 (void) zio_wait(zio
);
1436 return (good_writes
>= 1 ? 0 : EIO
);
1440 * On success, increment the count of good writes for our top-level vdev.
1443 vdev_label_sync_done(zio_t
*zio
)
1445 uint64_t *good_writes
= zio
->io_private
;
1447 if (zio
->io_error
== 0)
1448 atomic_inc_64(good_writes
);
1452 * If there weren't enough good writes, indicate failure to the parent.
1455 vdev_label_sync_top_done(zio_t
*zio
)
1457 uint64_t *good_writes
= zio
->io_private
;
1459 if (*good_writes
== 0)
1460 zio
->io_error
= SET_ERROR(EIO
);
1462 kmem_free(good_writes
, sizeof (uint64_t));
1466 * We ignore errors for log and cache devices, simply free the private data.
1469 vdev_label_sync_ignore_done(zio_t
*zio
)
1471 kmem_free(zio
->io_private
, sizeof (uint64_t));
1475 * Write all even or odd labels to all leaves of the specified vdev.
1478 vdev_label_sync(zio_t
*zio
, uint64_t *good_writes
,
1479 vdev_t
*vd
, int l
, uint64_t txg
, int flags
)
1487 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1488 vdev_label_sync(zio
, good_writes
,
1489 vd
->vdev_child
[c
], l
, txg
, flags
);
1492 if (!vd
->vdev_ops
->vdev_op_leaf
)
1495 if (!vdev_writeable(vd
))
1499 * Generate a label describing the top-level config to which we belong.
1501 label
= spa_config_generate(vd
->vdev_spa
, vd
, txg
, B_FALSE
);
1503 vp_abd
= abd_alloc_linear(sizeof (vdev_phys_t
), B_TRUE
);
1504 abd_zero(vp_abd
, sizeof (vdev_phys_t
));
1505 vp
= abd_to_buf(vp_abd
);
1507 buf
= vp
->vp_nvlist
;
1508 buflen
= sizeof (vp
->vp_nvlist
);
1510 if (!nvlist_pack(label
, &buf
, &buflen
, NV_ENCODE_XDR
, KM_SLEEP
)) {
1511 for (; l
< VDEV_LABELS
; l
+= 2) {
1512 vdev_label_write(zio
, vd
, l
, vp_abd
,
1513 offsetof(vdev_label_t
, vl_vdev_phys
),
1514 sizeof (vdev_phys_t
),
1515 vdev_label_sync_done
, good_writes
,
1516 flags
| ZIO_FLAG_DONT_PROPAGATE
);
1525 vdev_label_sync_list(spa_t
*spa
, int l
, uint64_t txg
, int flags
)
1527 list_t
*dl
= &spa
->spa_config_dirty_list
;
1533 * Write the new labels to disk.
1535 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1537 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
)) {
1538 uint64_t *good_writes
;
1540 ASSERT(!vd
->vdev_ishole
);
1542 good_writes
= kmem_zalloc(sizeof (uint64_t), KM_SLEEP
);
1543 zio_t
*vio
= zio_null(zio
, spa
, NULL
,
1544 (vd
->vdev_islog
|| vd
->vdev_aux
!= NULL
) ?
1545 vdev_label_sync_ignore_done
: vdev_label_sync_top_done
,
1546 good_writes
, flags
);
1547 vdev_label_sync(vio
, good_writes
, vd
, l
, txg
, flags
);
1551 error
= zio_wait(zio
);
1554 * Flush the new labels to disk.
1556 zio
= zio_root(spa
, NULL
, NULL
, flags
);
1558 for (vd
= list_head(dl
); vd
!= NULL
; vd
= list_next(dl
, vd
))
1561 (void) zio_wait(zio
);
1567 * Sync the uberblock and any changes to the vdev configuration.
1569 * The order of operations is carefully crafted to ensure that
1570 * if the system panics or loses power at any time, the state on disk
1571 * is still transactionally consistent. The in-line comments below
1572 * describe the failure semantics at each stage.
1574 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1575 * at any time, you can just call it again, and it will resume its work.
1578 vdev_config_sync(vdev_t
**svd
, int svdcount
, uint64_t txg
)
1580 spa_t
*spa
= svd
[0]->vdev_spa
;
1581 uberblock_t
*ub
= &spa
->spa_uberblock
;
1583 int flags
= ZIO_FLAG_CONFIG_WRITER
| ZIO_FLAG_CANFAIL
;
1585 ASSERT(svdcount
!= 0);
1588 * Normally, we don't want to try too hard to write every label and
1589 * uberblock. If there is a flaky disk, we don't want the rest of the
1590 * sync process to block while we retry. But if we can't write a
1591 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1592 * bailing out and declaring the pool faulted.
1595 if ((flags
& ZIO_FLAG_TRYHARD
) != 0)
1597 flags
|= ZIO_FLAG_TRYHARD
;
1600 ASSERT(ub
->ub_txg
<= txg
);
1603 * If this isn't a resync due to I/O errors,
1604 * and nothing changed in this transaction group,
1605 * and the vdev configuration hasn't changed,
1606 * then there's nothing to do.
1608 if (ub
->ub_txg
< txg
) {
1609 boolean_t changed
= uberblock_update(ub
, spa
->spa_root_vdev
,
1610 txg
, spa
->spa_mmp
.mmp_delay
);
1612 if (!changed
&& list_is_empty(&spa
->spa_config_dirty_list
))
1616 if (txg
> spa_freeze_txg(spa
))
1619 ASSERT(txg
<= spa
->spa_final_txg
);
1622 * Flush the write cache of every disk that's been written to
1623 * in this transaction group. This ensures that all blocks
1624 * written in this txg will be committed to stable storage
1625 * before any uberblock that references them.
1627 zio_t
*zio
= zio_root(spa
, NULL
, NULL
, flags
);
1630 txg_list_head(&spa
->spa_vdev_txg_list
, TXG_CLEAN(txg
)); vd
!= NULL
;
1631 vd
= txg_list_next(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
)))
1634 (void) zio_wait(zio
);
1637 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1638 * system dies in the middle of this process, that's OK: all of the
1639 * even labels that made it to disk will be newer than any uberblock,
1640 * and will therefore be considered invalid. The odd labels (L1, L3),
1641 * which have not yet been touched, will still be valid. We flush
1642 * the new labels to disk to ensure that all even-label updates
1643 * are committed to stable storage before the uberblock update.
1645 if ((error
= vdev_label_sync_list(spa
, 0, txg
, flags
)) != 0) {
1646 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1647 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1648 "for pool '%s' when syncing out the even labels "
1649 "of dirty vdevs", error
, spa_name(spa
));
1655 * Sync the uberblocks to all vdevs in svd[].
1656 * If the system dies in the middle of this step, there are two cases
1657 * to consider, and the on-disk state is consistent either way:
1659 * (1) If none of the new uberblocks made it to disk, then the
1660 * previous uberblock will be the newest, and the odd labels
1661 * (which had not yet been touched) will be valid with respect
1662 * to that uberblock.
1664 * (2) If one or more new uberblocks made it to disk, then they
1665 * will be the newest, and the even labels (which had all
1666 * been successfully committed) will be valid with respect
1667 * to the new uberblocks.
1669 if ((error
= vdev_uberblock_sync_list(svd
, svdcount
, ub
, flags
)) != 0) {
1670 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1671 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1672 "%d for pool '%s'", error
, spa_name(spa
));
1677 if (spa_multihost(spa
))
1678 mmp_update_uberblock(spa
, ub
);
1681 * Sync out odd labels for every dirty vdev. If the system dies
1682 * in the middle of this process, the even labels and the new
1683 * uberblocks will suffice to open the pool. The next time
1684 * the pool is opened, the first thing we'll do -- before any
1685 * user data is modified -- is mark every vdev dirty so that
1686 * all labels will be brought up to date. We flush the new labels
1687 * to disk to ensure that all odd-label updates are committed to
1688 * stable storage before the next transaction group begins.
1690 if ((error
= vdev_label_sync_list(spa
, 1, txg
, flags
)) != 0) {
1691 if ((flags
& ZIO_FLAG_TRYHARD
) != 0) {
1692 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1693 "for pool '%s' when syncing out the odd labels of "
1694 "dirty vdevs", error
, spa_name(spa
));