4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Intel Corporation.
29 * Virtual Device Labels
30 * ---------------------
32 * The vdev label serves several distinct purposes:
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
37 * 2. Verify that all the devices given in a configuration are present
40 * 3. Determine the uberblock for the pool.
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transaction groups:
68 * +------+ +------+ +------+
70 * | t10 | | t10 | | t10 |
72 * +------+ +------+ +------+
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
78 * In order to identify which labels are valid, the labels are written in the
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
116 * Configuration Information
117 * -------------------------
119 * The nvlist describing the pool and vdev contains the following elements:
121 * version ZFS on-disk version
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
128 * An nvlist of the features necessary for reading the MOS.
130 * Each leaf device label also contains the following:
132 * top_guid Unique ID for top-level vdev in which this is contained
133 * guid Unique ID for the leaf vdev
135 * The 'vs' configuration follows the format described in 'spa_config.c'.
138 #include <sys/zfs_context.h>
140 #include <sys/spa_impl.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/uberblock_impl.h>
146 #include <sys/metaslab.h>
147 #include <sys/metaslab_impl.h>
149 #include <sys/dsl_scan.h>
151 #include <sys/fs/zfs.h>
152 #include <sys/byteorder.h>
153 #include <sys/zfs_bootenv.h>
156 * Basic routines to read and write from a vdev label.
157 * Used throughout the rest of this file.
160 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
162 ASSERT(offset < sizeof (vdev_label_t));
163 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
165 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
166 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
170 * Returns back the vdev label associated with the passed in offset.
173 vdev_label_number(uint64_t psize, uint64_t offset)
177 if (offset >= psize - VDEV_LABEL_END_SIZE) {
178 offset -= psize - VDEV_LABEL_END_SIZE;
179 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
181 l = offset / sizeof (vdev_label_t);
182 return (l < VDEV_LABELS ? l : -1);
186 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
187 uint64_t size, zio_done_func_t *done, void *private, int flags)
190 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
191 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
192 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
194 zio_nowait(zio_read_phys(zio, vd,
195 vdev_label_offset(vd->vdev_psize, l, offset),
196 size, buf, ZIO_CHECKSUM_LABEL, done, private,
197 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
201 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
202 uint64_t size, zio_done_func_t *done, void *private, int flags)
205 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
206 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
207 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
209 zio_nowait(zio_write_phys(zio, vd,
210 vdev_label_offset(vd->vdev_psize, l, offset),
211 size, buf, ZIO_CHECKSUM_LABEL, done, private,
212 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
216 * Generate the nvlist representing this vdev's stats
219 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
225 vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
226 vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
228 vdev_get_stats_ex(vd, vs, vsx);
229 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
230 (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
233 * Add extended stats into a special extended stats nvlist. This keeps
234 * all the extended stats nicely grouped together. The extended stats
235 * nvlist is then added to the main nvlist.
237 nvx = fnvlist_alloc();
239 /* ZIOs in flight to disk */
240 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
241 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
243 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
244 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
246 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
247 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
249 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
250 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
252 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
253 vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
255 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
256 vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
259 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
260 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
262 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
263 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
265 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
266 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
268 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
269 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
271 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
272 vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
274 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
275 vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
278 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
279 vsx->vsx_total_histo[ZIO_TYPE_READ],
280 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
282 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
283 vsx->vsx_total_histo[ZIO_TYPE_WRITE],
284 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
286 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
287 vsx->vsx_disk_histo[ZIO_TYPE_READ],
288 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
290 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
291 vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
292 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
294 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
295 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
296 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
298 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
299 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
300 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
302 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
303 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
304 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
306 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
307 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
308 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
310 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
311 vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
312 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
314 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
315 vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
316 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
319 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
320 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
321 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
323 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
324 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
325 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
327 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
328 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
329 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
331 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
332 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
333 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
335 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
336 vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
337 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
339 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
340 vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
341 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
343 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
344 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
345 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
347 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
348 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
349 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
351 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
352 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
353 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
355 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
356 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
357 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
359 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
360 vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
361 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
363 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
364 vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
365 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
368 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
370 /* Add extended stats nvlist to main nvlist */
371 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
374 kmem_free(vs, sizeof (*vs));
375 kmem_free(vsx, sizeof (*vsx));
379 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
381 spa_t *spa = vd->vdev_spa;
383 if (vd != spa->spa_root_vdev)
386 /* provide either current or previous scan information */
388 if (spa_scan_get_stats(spa, &ps) == 0) {
389 fnvlist_add_uint64_array(nvl,
390 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
391 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
394 pool_removal_stat_t prs;
395 if (spa_removal_get_stats(spa, &prs) == 0) {
396 fnvlist_add_uint64_array(nvl,
397 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
398 sizeof (prs) / sizeof (uint64_t));
401 pool_checkpoint_stat_t pcs;
402 if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
403 fnvlist_add_uint64_array(nvl,
404 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
405 sizeof (pcs) / sizeof (uint64_t));
410 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
412 if (vd == vd->vdev_top) {
413 vdev_rebuild_stat_t vrs;
414 if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
415 fnvlist_add_uint64_array(nvl,
416 ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
417 sizeof (vrs) / sizeof (uint64_t));
423 * Generate the nvlist representing this vdev's config.
426 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
427 vdev_config_flag_t flags)
430 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
432 nv = fnvlist_alloc();
434 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
435 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
436 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
437 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
439 if (vd->vdev_path != NULL)
440 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
442 if (vd->vdev_devid != NULL)
443 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
445 if (vd->vdev_physpath != NULL)
446 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
449 if (vd->vdev_enc_sysfs_path != NULL)
450 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
451 vd->vdev_enc_sysfs_path);
453 if (vd->vdev_fru != NULL)
454 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
456 if (vd->vdev_nparity != 0) {
457 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
458 VDEV_TYPE_RAIDZ) == 0);
461 * Make sure someone hasn't managed to sneak a fancy new vdev
462 * into a crufty old storage pool.
464 ASSERT(vd->vdev_nparity == 1 ||
465 (vd->vdev_nparity <= 2 &&
466 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
467 (vd->vdev_nparity <= 3 &&
468 spa_version(spa) >= SPA_VERSION_RAIDZ3));
471 * Note that we'll add the nparity tag even on storage pools
472 * that only support a single parity device -- older software
473 * will just ignore it.
475 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
478 if (vd->vdev_wholedisk != -1ULL)
479 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
482 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
483 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
485 if (vd->vdev_isspare)
486 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
488 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
489 vd == vd->vdev_top) {
490 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
492 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
494 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
495 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
497 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
498 if (vd->vdev_removing) {
499 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
503 /* zpool command expects alloc class data */
504 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
505 const char *bias = NULL;
507 switch (vd->vdev_alloc_bias) {
509 bias = VDEV_ALLOC_BIAS_LOG;
511 case VDEV_BIAS_SPECIAL:
512 bias = VDEV_ALLOC_BIAS_SPECIAL;
514 case VDEV_BIAS_DEDUP:
515 bias = VDEV_ALLOC_BIAS_DEDUP;
518 ASSERT3U(vd->vdev_alloc_bias, ==,
521 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
526 if (vd->vdev_dtl_sm != NULL) {
527 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
528 space_map_object(vd->vdev_dtl_sm));
531 if (vic->vic_mapping_object != 0) {
532 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
533 vic->vic_mapping_object);
536 if (vic->vic_births_object != 0) {
537 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
538 vic->vic_births_object);
541 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
542 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
543 vic->vic_prev_indirect_vdev);
547 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
549 if (vd->vdev_expansion_time)
550 fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
551 vd->vdev_expansion_time);
553 if (flags & VDEV_CONFIG_MOS) {
554 if (vd->vdev_leaf_zap != 0) {
555 ASSERT(vd->vdev_ops->vdev_op_leaf);
556 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
560 if (vd->vdev_top_zap != 0) {
561 ASSERT(vd == vd->vdev_top);
562 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
566 if (vd->vdev_resilver_deferred) {
567 ASSERT(vd->vdev_ops->vdev_op_leaf);
568 ASSERT(spa->spa_resilver_deferred);
569 fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
574 vdev_config_generate_stats(vd, nv);
576 root_vdev_actions_getprogress(vd, nv);
577 top_vdev_actions_getprogress(vd, nv);
580 * Note: this can be called from open context
581 * (spa_get_stats()), so we need the rwlock to prevent
582 * the mapping from being changed by condensing.
584 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
585 if (vd->vdev_indirect_mapping != NULL) {
586 ASSERT(vd->vdev_indirect_births != NULL);
587 vdev_indirect_mapping_t *vim =
588 vd->vdev_indirect_mapping;
589 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
590 vdev_indirect_mapping_size(vim));
592 rw_exit(&vd->vdev_indirect_rwlock);
593 if (vd->vdev_mg != NULL &&
594 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
596 * Compute approximately how much memory would be used
597 * for the indirect mapping if this device were to
600 * Note: If the frag metric is invalid, then not
601 * enough metaslabs have been converted to have
604 uint64_t seg_count = 0;
605 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
608 * There are the same number of allocated segments
609 * as free segments, so we will have at least one
610 * entry per free segment. However, small free
611 * segments (smaller than vdev_removal_max_span)
612 * will be combined with adjacent allocated segments
613 * as a single mapping.
615 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
616 if (i + 1 < highbit64(vdev_removal_max_span)
619 vd->vdev_mg->mg_histogram[i] <<
623 vd->vdev_mg->mg_histogram[i];
628 * The maximum length of a mapping is
629 * zfs_remove_max_segment, so we need at least one entry
630 * per zfs_remove_max_segment of allocated data.
632 seg_count += to_alloc / spa_remove_max_segment(spa);
634 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
636 sizeof (vdev_indirect_mapping_entry_phys_t));
640 if (!vd->vdev_ops->vdev_op_leaf) {
644 ASSERT(!vd->vdev_ishole);
646 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
649 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
650 vdev_t *cvd = vd->vdev_child[c];
653 * If we're generating an nvlist of removing
654 * vdevs then skip over any device which is
657 if ((flags & VDEV_CONFIG_REMOVING) &&
661 child[idx++] = vdev_config_generate(spa, cvd,
666 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
670 for (c = 0; c < idx; c++)
671 nvlist_free(child[c]);
673 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
676 const char *aux = NULL;
678 if (vd->vdev_offline && !vd->vdev_tmpoffline)
679 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
680 if (vd->vdev_resilver_txg != 0)
681 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
682 vd->vdev_resilver_txg);
683 if (vd->vdev_rebuild_txg != 0)
684 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
685 vd->vdev_rebuild_txg);
686 if (vd->vdev_faulted)
687 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
688 if (vd->vdev_degraded)
689 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
690 if (vd->vdev_removed)
691 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
692 if (vd->vdev_unspare)
693 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
695 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
697 /* Set the reason why we're FAULTED/DEGRADED. */
698 switch (vd->vdev_stat.vs_aux) {
699 case VDEV_AUX_ERR_EXCEEDED:
700 aux = "err_exceeded";
703 case VDEV_AUX_EXTERNAL:
708 if (aux != NULL && !vd->vdev_tmpoffline) {
709 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
712 * We're healthy - clear any previous AUX_STATE values.
714 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
715 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
718 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
719 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
728 * Generate a view of the top-level vdevs. If we currently have holes
729 * in the namespace, then generate an array which contains a list of holey
730 * vdevs. Additionally, add the number of top-level children that currently
734 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
736 vdev_t *rvd = spa->spa_root_vdev;
740 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
742 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
743 vdev_t *tvd = rvd->vdev_child[c];
745 if (tvd->vdev_ishole) {
751 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
755 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
756 rvd->vdev_children) == 0);
758 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
762 * Returns the configuration from the label of the given vdev. For vdevs
763 * which don't have a txg value stored on their label (i.e. spares/cache)
764 * or have not been completely initialized (txg = 0) just return
765 * the configuration from the first valid label we find. Otherwise,
766 * find the most up-to-date label that does not exceed the specified
770 vdev_label_read_config(vdev_t *vd, uint64_t txg)
772 spa_t *spa = vd->vdev_spa;
773 nvlist_t *config = NULL;
777 uint64_t best_txg = 0;
778 uint64_t label_txg = 0;
780 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
781 ZIO_FLAG_SPECULATIVE;
783 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
785 if (!vdev_readable(vd))
788 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
789 vp = abd_to_buf(vp_abd);
792 for (int l = 0; l < VDEV_LABELS; l++) {
793 nvlist_t *label = NULL;
795 zio = zio_root(spa, NULL, NULL, flags);
797 vdev_label_read(zio, vd, l, vp_abd,
798 offsetof(vdev_label_t, vl_vdev_phys),
799 sizeof (vdev_phys_t), NULL, NULL, flags);
801 if (zio_wait(zio) == 0 &&
802 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
805 * Auxiliary vdevs won't have txg values in their
806 * labels and newly added vdevs may not have been
807 * completely initialized so just return the
808 * configuration from the first valid label we
811 error = nvlist_lookup_uint64(label,
812 ZPOOL_CONFIG_POOL_TXG, &label_txg);
813 if ((error || label_txg == 0) && !config) {
816 } else if (label_txg <= txg && label_txg > best_txg) {
817 best_txg = label_txg;
819 config = fnvlist_dup(label);
829 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
830 flags |= ZIO_FLAG_TRYHARD;
835 * We found a valid label but it didn't pass txg restrictions.
837 if (config == NULL && label_txg != 0) {
838 vdev_dbgmsg(vd, "label discarded as txg is too large "
839 "(%llu > %llu)", (u_longlong_t)label_txg,
849 * Determine if a device is in use. The 'spare_guid' parameter will be filled
850 * in with the device guid if this spare is active elsewhere on the system.
853 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
854 uint64_t *spare_guid, uint64_t *l2cache_guid)
856 spa_t *spa = vd->vdev_spa;
857 uint64_t state, pool_guid, device_guid, txg, spare_pool;
864 *l2cache_guid = 0ULL;
867 * Read the label, if any, and perform some basic sanity checks.
869 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
872 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
875 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
877 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
878 &device_guid) != 0) {
883 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
884 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
886 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
895 * Check to see if this device indeed belongs to the pool it claims to
896 * be a part of. The only way this is allowed is if the device is a hot
897 * spare (which we check for later on).
899 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
900 !spa_guid_exists(pool_guid, device_guid) &&
901 !spa_spare_exists(device_guid, NULL, NULL) &&
902 !spa_l2cache_exists(device_guid, NULL))
906 * If the transaction group is zero, then this an initialized (but
907 * unused) label. This is only an error if the create transaction
908 * on-disk is the same as the one we're using now, in which case the
909 * user has attempted to add the same vdev multiple times in the same
912 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
913 txg == 0 && vdtxg == crtxg)
917 * Check to see if this is a spare device. We do an explicit check for
918 * spa_has_spare() here because it may be on our pending list of spares
919 * to add. We also check if it is an l2cache device.
921 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
922 spa_has_spare(spa, device_guid)) {
924 *spare_guid = device_guid;
927 case VDEV_LABEL_CREATE:
928 case VDEV_LABEL_L2CACHE:
931 case VDEV_LABEL_REPLACE:
932 return (!spa_has_spare(spa, device_guid) ||
935 case VDEV_LABEL_SPARE:
936 return (spa_has_spare(spa, device_guid));
943 * Check to see if this is an l2cache device.
945 if (spa_l2cache_exists(device_guid, NULL))
949 * We can't rely on a pool's state if it's been imported
950 * read-only. Instead we look to see if the pools is marked
951 * read-only in the namespace and set the state to active.
953 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
954 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
955 spa_mode(spa) == SPA_MODE_READ)
956 state = POOL_STATE_ACTIVE;
959 * If the device is marked ACTIVE, then this device is in use by another
960 * pool on the system.
962 return (state == POOL_STATE_ACTIVE);
966 * Initialize a vdev label. We check to make sure each leaf device is not in
967 * use, and writable. We put down an initial label which we will later
968 * overwrite with a complete label. Note that it's important to do this
969 * sequentially, not in parallel, so that we catch cases of multiple use of the
970 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
974 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
976 spa_t *spa = vd->vdev_spa;
987 uint64_t spare_guid = 0, l2cache_guid = 0;
988 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
990 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
992 for (int c = 0; c < vd->vdev_children; c++)
993 if ((error = vdev_label_init(vd->vdev_child[c],
994 crtxg, reason)) != 0)
997 /* Track the creation time for this vdev */
998 vd->vdev_crtxg = crtxg;
1000 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1004 * Dead vdevs cannot be initialized.
1006 if (vdev_is_dead(vd))
1007 return (SET_ERROR(EIO));
1010 * Determine if the vdev is in use.
1012 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1013 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1014 return (SET_ERROR(EBUSY));
1017 * If this is a request to add or replace a spare or l2cache device
1018 * that is in use elsewhere on the system, then we must update the
1019 * guid (which was initialized to a random value) to reflect the
1020 * actual GUID (which is shared between multiple pools).
1022 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1023 spare_guid != 0ULL) {
1024 uint64_t guid_delta = spare_guid - vd->vdev_guid;
1026 vd->vdev_guid += guid_delta;
1028 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1029 pvd->vdev_guid_sum += guid_delta;
1032 * If this is a replacement, then we want to fallthrough to the
1033 * rest of the code. If we're adding a spare, then it's already
1034 * labeled appropriately and we can just return.
1036 if (reason == VDEV_LABEL_SPARE)
1038 ASSERT(reason == VDEV_LABEL_REPLACE ||
1039 reason == VDEV_LABEL_SPLIT);
1042 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1043 l2cache_guid != 0ULL) {
1044 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1046 vd->vdev_guid += guid_delta;
1048 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1049 pvd->vdev_guid_sum += guid_delta;
1052 * If this is a replacement, then we want to fallthrough to the
1053 * rest of the code. If we're adding an l2cache, then it's
1054 * already labeled appropriately and we can just return.
1056 if (reason == VDEV_LABEL_L2CACHE)
1058 ASSERT(reason == VDEV_LABEL_REPLACE);
1062 * Initialize its label.
1064 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1065 abd_zero(vp_abd, sizeof (vdev_phys_t));
1066 vp = abd_to_buf(vp_abd);
1069 * Generate a label describing the pool and our top-level vdev.
1070 * We mark it as being from txg 0 to indicate that it's not
1071 * really part of an active pool just yet. The labels will
1072 * be written again with a meaningful txg by spa_sync().
1074 if (reason == VDEV_LABEL_SPARE ||
1075 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1077 * For inactive hot spares, we generate a special label that
1078 * identifies as a mutually shared hot spare. We write the
1079 * label if we are adding a hot spare, or if we are removing an
1080 * active hot spare (in which case we want to revert the
1083 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1085 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1086 spa_version(spa)) == 0);
1087 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1088 POOL_STATE_SPARE) == 0);
1089 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1090 vd->vdev_guid) == 0);
1091 } else if (reason == VDEV_LABEL_L2CACHE ||
1092 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1094 * For level 2 ARC devices, add a special label.
1096 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1098 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1099 spa_version(spa)) == 0);
1100 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1101 POOL_STATE_L2CACHE) == 0);
1102 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1103 vd->vdev_guid) == 0);
1105 uint64_t txg = 0ULL;
1107 if (reason == VDEV_LABEL_SPLIT)
1108 txg = spa->spa_uberblock.ub_txg;
1109 label = spa_config_generate(spa, vd, txg, B_FALSE);
1112 * Add our creation time. This allows us to detect multiple
1113 * vdev uses as described above, and automatically expires if we
1116 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1120 buf = vp->vp_nvlist;
1121 buflen = sizeof (vp->vp_nvlist);
1123 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1127 /* EFAULT means nvlist_pack ran out of room */
1128 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1132 * Initialize uberblock template.
1134 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1135 abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1136 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1137 ub = abd_to_buf(ub_abd);
1140 /* Initialize the 2nd padding area. */
1141 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1142 abd_zero(bootenv, VDEV_PAD_SIZE);
1145 * Write everything in parallel.
1148 zio = zio_root(spa, NULL, NULL, flags);
1150 for (int l = 0; l < VDEV_LABELS; l++) {
1152 vdev_label_write(zio, vd, l, vp_abd,
1153 offsetof(vdev_label_t, vl_vdev_phys),
1154 sizeof (vdev_phys_t), NULL, NULL, flags);
1157 * Skip the 1st padding area.
1158 * Zero out the 2nd padding area where it might have
1159 * left over data from previous filesystem format.
1161 vdev_label_write(zio, vd, l, bootenv,
1162 offsetof(vdev_label_t, vl_be),
1163 VDEV_PAD_SIZE, NULL, NULL, flags);
1165 vdev_label_write(zio, vd, l, ub_abd,
1166 offsetof(vdev_label_t, vl_uberblock),
1167 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1170 error = zio_wait(zio);
1172 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1173 flags |= ZIO_FLAG_TRYHARD;
1183 * If this vdev hasn't been previously identified as a spare, then we
1184 * mark it as such only if a) we are labeling it as a spare, or b) it
1185 * exists as a spare elsewhere in the system. Do the same for
1186 * level 2 ARC devices.
1188 if (error == 0 && !vd->vdev_isspare &&
1189 (reason == VDEV_LABEL_SPARE ||
1190 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1193 if (error == 0 && !vd->vdev_isl2cache &&
1194 (reason == VDEV_LABEL_L2CACHE ||
1195 spa_l2cache_exists(vd->vdev_guid, NULL)))
1196 spa_l2cache_add(vd);
1202 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1203 * callback to finish, store our abd in the callback pointer. Otherwise, we
1204 * just free our abd and return.
1207 vdev_label_read_bootenv_done(zio_t *zio)
1209 zio_t *rio = zio->io_private;
1210 abd_t **cbp = rio->io_private;
1212 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1214 if (zio->io_error == 0) {
1215 mutex_enter(&rio->io_lock);
1217 /* Will free this buffer in vdev_label_read_bootenv. */
1220 abd_free(zio->io_abd);
1222 mutex_exit(&rio->io_lock);
1224 abd_free(zio->io_abd);
1229 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1231 for (int c = 0; c < vd->vdev_children; c++)
1232 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1235 * We just use the first label that has a correct checksum; the
1236 * bootloader should have rewritten them all to be the same on boot,
1237 * and any changes we made since boot have been the same across all
1240 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1241 for (int l = 0; l < VDEV_LABELS; l++) {
1242 vdev_label_read(zio, vd, l,
1243 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1244 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1245 vdev_label_read_bootenv_done, zio, flags);
1251 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1254 spa_t *spa = rvd->vdev_spa;
1256 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1257 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1260 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1262 zio_t *zio = zio_root(spa, NULL, &abd, flags);
1263 vdev_label_read_bootenv_impl(zio, rvd, flags);
1264 int err = zio_wait(zio);
1268 vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1270 vbe->vbe_version = ntohll(vbe->vbe_version);
1271 switch (vbe->vbe_version) {
1274 * if we have textual data in vbe_bootenv, create nvlist
1275 * with key "envmap".
1277 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1278 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1279 fnvlist_add_string(bootenv, GRUB_ENVMAP,
1284 err = nvlist_unpack(vbe->vbe_bootenv,
1285 sizeof (vbe->vbe_bootenv), &config, 0);
1287 fnvlist_merge(bootenv, config);
1288 nvlist_free(config);
1293 /* Check for FreeBSD zfs bootonce command string */
1294 buf = abd_to_buf(abd);
1296 fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1300 fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1304 * abd was allocated in vdev_label_read_bootenv_impl()
1308 * If we managed to read any successfully,
1317 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1320 spa_t *spa = vd->vdev_spa;
1321 vdev_boot_envblock_t *bootenv;
1322 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1327 error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1329 return (SET_ERROR(error));
1331 if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1332 return (SET_ERROR(E2BIG));
1335 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1338 for (int c = 0; c < vd->vdev_children; c++) {
1341 child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1343 * As long as any of the disks managed to write all of their
1344 * labels successfully, return success.
1350 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1351 !vdev_writeable(vd)) {
1354 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1355 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1356 abd_zero(abd, VDEV_PAD_SIZE);
1358 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1359 nvbuf = bootenv->vbe_bootenv;
1360 nvsize = sizeof (bootenv->vbe_bootenv);
1362 bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1363 switch (bootenv->vbe_version) {
1365 if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) {
1366 (void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize);
1372 error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1382 bootenv->vbe_version = htonll(bootenv->vbe_version);
1383 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1386 return (SET_ERROR(error));
1390 zio = zio_root(spa, NULL, NULL, flags);
1391 for (int l = 0; l < VDEV_LABELS; l++) {
1392 vdev_label_write(zio, vd, l, abd,
1393 offsetof(vdev_label_t, vl_be),
1394 VDEV_PAD_SIZE, NULL, NULL, flags);
1397 error = zio_wait(zio);
1398 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1399 flags |= ZIO_FLAG_TRYHARD;
1408 * ==========================================================================
1409 * uberblock load/sync
1410 * ==========================================================================
1414 * Consider the following situation: txg is safely synced to disk. We've
1415 * written the first uberblock for txg + 1, and then we lose power. When we
1416 * come back up, we fail to see the uberblock for txg + 1 because, say,
1417 * it was on a mirrored device and the replica to which we wrote txg + 1
1418 * is now offline. If we then make some changes and sync txg + 1, and then
1419 * the missing replica comes back, then for a few seconds we'll have two
1420 * conflicting uberblocks on disk with the same txg. The solution is simple:
1421 * among uberblocks with equal txg, choose the one with the latest timestamp.
1424 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1426 int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1431 cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1436 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1437 * ZFS, e.g. OpenZFS >= 0.7.
1439 * If one ub has MMP and the other does not, they were written by
1440 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1443 * Since timestamp and txg are the same if we get this far, either is
1444 * acceptable for importing the pool.
1446 unsigned int seq1 = 0;
1447 unsigned int seq2 = 0;
1449 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1450 seq1 = MMP_SEQ(ub1);
1452 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1453 seq2 = MMP_SEQ(ub2);
1455 return (TREE_CMP(seq1, seq2));
1459 uberblock_t *ubl_ubbest; /* Best uberblock */
1460 vdev_t *ubl_vd; /* vdev associated with the above */
1464 vdev_uberblock_load_done(zio_t *zio)
1466 vdev_t *vd = zio->io_vd;
1467 spa_t *spa = zio->io_spa;
1468 zio_t *rio = zio->io_private;
1469 uberblock_t *ub = abd_to_buf(zio->io_abd);
1470 struct ubl_cbdata *cbp = rio->io_private;
1472 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1474 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1475 mutex_enter(&rio->io_lock);
1476 if (ub->ub_txg <= spa->spa_load_max_txg &&
1477 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1479 * Keep track of the vdev in which this uberblock
1480 * was found. We will use this information later
1481 * to obtain the config nvlist associated with
1484 *cbp->ubl_ubbest = *ub;
1487 mutex_exit(&rio->io_lock);
1490 abd_free(zio->io_abd);
1494 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1495 struct ubl_cbdata *cbp)
1497 for (int c = 0; c < vd->vdev_children; c++)
1498 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1500 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1501 for (int l = 0; l < VDEV_LABELS; l++) {
1502 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1503 vdev_label_read(zio, vd, l,
1504 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1505 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1506 VDEV_UBERBLOCK_SIZE(vd),
1507 vdev_uberblock_load_done, zio, flags);
1514 * Reads the 'best' uberblock from disk along with its associated
1515 * configuration. First, we read the uberblock array of each label of each
1516 * vdev, keeping track of the uberblock with the highest txg in each array.
1517 * Then, we read the configuration from the same vdev as the best uberblock.
1520 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1523 spa_t *spa = rvd->vdev_spa;
1524 struct ubl_cbdata cb;
1525 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1526 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1531 bzero(ub, sizeof (uberblock_t));
1537 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1538 zio = zio_root(spa, NULL, &cb, flags);
1539 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1540 (void) zio_wait(zio);
1543 * It's possible that the best uberblock was discovered on a label
1544 * that has a configuration which was written in a future txg.
1545 * Search all labels on this vdev to find the configuration that
1546 * matches the txg for our uberblock.
1548 if (cb.ubl_vd != NULL) {
1549 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1550 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1552 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1553 if (*config == NULL && spa->spa_extreme_rewind) {
1554 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1555 "Trying again without txg restrictions.");
1556 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1558 if (*config == NULL) {
1559 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1562 spa_config_exit(spa, SCL_ALL, FTAG);
1566 * For use when a leaf vdev is expanded.
1567 * The location of labels 2 and 3 changed, and at the new location the
1568 * uberblock rings are either empty or contain garbage. The sync will write
1569 * new configs there because the vdev is dirty, but expansion also needs the
1570 * uberblock rings copied. Read them from label 0 which did not move.
1572 * Since the point is to populate labels {2,3} with valid uberblocks,
1573 * we zero uberblocks we fail to read or which are not valid.
1577 vdev_copy_uberblocks(vdev_t *vd)
1581 int locks = (SCL_L2ARC | SCL_ZIO);
1582 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1583 ZIO_FLAG_SPECULATIVE;
1585 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1587 ASSERT(vd->vdev_ops->vdev_op_leaf);
1589 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1591 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1593 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1594 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1595 const int src_label = 0;
1598 zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1599 vdev_label_read(zio, vd, src_label, ub_abd,
1600 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1603 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1604 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1606 for (int l = 2; l < VDEV_LABELS; l++)
1607 vdev_label_write(write_zio, vd, l, ub_abd,
1608 VDEV_UBERBLOCK_OFFSET(vd, n),
1609 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1610 flags | ZIO_FLAG_DONT_PROPAGATE);
1612 (void) zio_wait(write_zio);
1614 spa_config_exit(vd->vdev_spa, locks, FTAG);
1620 * On success, increment root zio's count of good writes.
1621 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1624 vdev_uberblock_sync_done(zio_t *zio)
1626 uint64_t *good_writes = zio->io_private;
1628 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1629 atomic_inc_64(good_writes);
1633 * Write the uberblock to all labels of all leaves of the specified vdev.
1636 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1637 uberblock_t *ub, vdev_t *vd, int flags)
1639 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1640 vdev_uberblock_sync(zio, good_writes,
1641 ub, vd->vdev_child[c], flags);
1644 if (!vd->vdev_ops->vdev_op_leaf)
1647 if (!vdev_writeable(vd))
1650 /* If the vdev was expanded, need to copy uberblock rings. */
1651 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1652 vd->vdev_copy_uberblocks == B_TRUE) {
1653 vdev_copy_uberblocks(vd);
1654 vd->vdev_copy_uberblocks = B_FALSE;
1657 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1658 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1660 /* Copy the uberblock_t into the ABD */
1661 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1662 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1663 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1665 for (int l = 0; l < VDEV_LABELS; l++)
1666 vdev_label_write(zio, vd, l, ub_abd,
1667 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1668 vdev_uberblock_sync_done, good_writes,
1669 flags | ZIO_FLAG_DONT_PROPAGATE);
1674 /* Sync the uberblocks to all vdevs in svd[] */
1676 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1678 spa_t *spa = svd[0]->vdev_spa;
1680 uint64_t good_writes = 0;
1682 zio = zio_root(spa, NULL, NULL, flags);
1684 for (int v = 0; v < svdcount; v++)
1685 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1687 (void) zio_wait(zio);
1690 * Flush the uberblocks to disk. This ensures that the odd labels
1691 * are no longer needed (because the new uberblocks and the even
1692 * labels are safely on disk), so it is safe to overwrite them.
1694 zio = zio_root(spa, NULL, NULL, flags);
1696 for (int v = 0; v < svdcount; v++) {
1697 if (vdev_writeable(svd[v])) {
1698 zio_flush(zio, svd[v]);
1702 (void) zio_wait(zio);
1704 return (good_writes >= 1 ? 0 : EIO);
1708 * On success, increment the count of good writes for our top-level vdev.
1711 vdev_label_sync_done(zio_t *zio)
1713 uint64_t *good_writes = zio->io_private;
1715 if (zio->io_error == 0)
1716 atomic_inc_64(good_writes);
1720 * If there weren't enough good writes, indicate failure to the parent.
1723 vdev_label_sync_top_done(zio_t *zio)
1725 uint64_t *good_writes = zio->io_private;
1727 if (*good_writes == 0)
1728 zio->io_error = SET_ERROR(EIO);
1730 kmem_free(good_writes, sizeof (uint64_t));
1734 * We ignore errors for log and cache devices, simply free the private data.
1737 vdev_label_sync_ignore_done(zio_t *zio)
1739 kmem_free(zio->io_private, sizeof (uint64_t));
1743 * Write all even or odd labels to all leaves of the specified vdev.
1746 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1747 vdev_t *vd, int l, uint64_t txg, int flags)
1755 for (int c = 0; c < vd->vdev_children; c++) {
1756 vdev_label_sync(zio, good_writes,
1757 vd->vdev_child[c], l, txg, flags);
1760 if (!vd->vdev_ops->vdev_op_leaf)
1763 if (!vdev_writeable(vd))
1767 * Generate a label describing the top-level config to which we belong.
1769 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1771 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1772 abd_zero(vp_abd, sizeof (vdev_phys_t));
1773 vp = abd_to_buf(vp_abd);
1775 buf = vp->vp_nvlist;
1776 buflen = sizeof (vp->vp_nvlist);
1778 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1779 for (; l < VDEV_LABELS; l += 2) {
1780 vdev_label_write(zio, vd, l, vp_abd,
1781 offsetof(vdev_label_t, vl_vdev_phys),
1782 sizeof (vdev_phys_t),
1783 vdev_label_sync_done, good_writes,
1784 flags | ZIO_FLAG_DONT_PROPAGATE);
1793 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1795 list_t *dl = &spa->spa_config_dirty_list;
1801 * Write the new labels to disk.
1803 zio = zio_root(spa, NULL, NULL, flags);
1805 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1806 uint64_t *good_writes;
1808 ASSERT(!vd->vdev_ishole);
1810 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1811 zio_t *vio = zio_null(zio, spa, NULL,
1812 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1813 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1814 good_writes, flags);
1815 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1819 error = zio_wait(zio);
1822 * Flush the new labels to disk.
1824 zio = zio_root(spa, NULL, NULL, flags);
1826 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1829 (void) zio_wait(zio);
1835 * Sync the uberblock and any changes to the vdev configuration.
1837 * The order of operations is carefully crafted to ensure that
1838 * if the system panics or loses power at any time, the state on disk
1839 * is still transactionally consistent. The in-line comments below
1840 * describe the failure semantics at each stage.
1842 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1843 * at any time, you can just call it again, and it will resume its work.
1846 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1848 spa_t *spa = svd[0]->vdev_spa;
1849 uberblock_t *ub = &spa->spa_uberblock;
1851 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1853 ASSERT(svdcount != 0);
1856 * Normally, we don't want to try too hard to write every label and
1857 * uberblock. If there is a flaky disk, we don't want the rest of the
1858 * sync process to block while we retry. But if we can't write a
1859 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1860 * bailing out and declaring the pool faulted.
1863 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1865 flags |= ZIO_FLAG_TRYHARD;
1868 ASSERT(ub->ub_txg <= txg);
1871 * If this isn't a resync due to I/O errors,
1872 * and nothing changed in this transaction group,
1873 * and the vdev configuration hasn't changed,
1874 * then there's nothing to do.
1876 if (ub->ub_txg < txg) {
1877 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1878 txg, spa->spa_mmp.mmp_delay);
1880 if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1884 if (txg > spa_freeze_txg(spa))
1887 ASSERT(txg <= spa->spa_final_txg);
1890 * Flush the write cache of every disk that's been written to
1891 * in this transaction group. This ensures that all blocks
1892 * written in this txg will be committed to stable storage
1893 * before any uberblock that references them.
1895 zio_t *zio = zio_root(spa, NULL, NULL, flags);
1898 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1899 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1902 (void) zio_wait(zio);
1905 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1906 * system dies in the middle of this process, that's OK: all of the
1907 * even labels that made it to disk will be newer than any uberblock,
1908 * and will therefore be considered invalid. The odd labels (L1, L3),
1909 * which have not yet been touched, will still be valid. We flush
1910 * the new labels to disk to ensure that all even-label updates
1911 * are committed to stable storage before the uberblock update.
1913 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1914 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1915 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1916 "for pool '%s' when syncing out the even labels "
1917 "of dirty vdevs", error, spa_name(spa));
1923 * Sync the uberblocks to all vdevs in svd[].
1924 * If the system dies in the middle of this step, there are two cases
1925 * to consider, and the on-disk state is consistent either way:
1927 * (1) If none of the new uberblocks made it to disk, then the
1928 * previous uberblock will be the newest, and the odd labels
1929 * (which had not yet been touched) will be valid with respect
1930 * to that uberblock.
1932 * (2) If one or more new uberblocks made it to disk, then they
1933 * will be the newest, and the even labels (which had all
1934 * been successfully committed) will be valid with respect
1935 * to the new uberblocks.
1937 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1938 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1939 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1940 "%d for pool '%s'", error, spa_name(spa));
1945 if (spa_multihost(spa))
1946 mmp_update_uberblock(spa, ub);
1949 * Sync out odd labels for every dirty vdev. If the system dies
1950 * in the middle of this process, the even labels and the new
1951 * uberblocks will suffice to open the pool. The next time
1952 * the pool is opened, the first thing we'll do -- before any
1953 * user data is modified -- is mark every vdev dirty so that
1954 * all labels will be brought up to date. We flush the new labels
1955 * to disk to ensure that all odd-label updates are committed to
1956 * stable storage before the next transaction group begins.
1958 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1959 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1960 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1961 "for pool '%s' when syncing out the odd labels of "
1962 "dirty vdevs", error, spa_name(spa));