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, 2016 by Delphix. All rights reserved.
28 * Virtual Device Labels
29 * ---------------------
31 * The vdev label serves several distinct purposes:
33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
34 * identity within the pool.
36 * 2. Verify that all the devices given in a configuration are present
39 * 3. Determine the uberblock for the pool.
41 * 4. In case of an import operation, determine the configuration of the
42 * toplevel vdev of which it is a part.
44 * 5. If an import operation cannot find all the devices in the pool,
45 * provide enough information to the administrator to determine which
46 * devices are missing.
48 * It is important to note that while the kernel is responsible for writing the
49 * label, it only consumes the information in the first three cases. The
50 * latter information is only consumed in userland when determining the
51 * configuration to import a pool.
57 * Before describing the contents of the label, it's important to understand how
58 * the labels are written and updated with respect to the uberblock.
60 * When the pool configuration is altered, either because it was newly created
61 * or a device was added, we want to update all the labels such that we can deal
62 * with fatal failure at any point. To this end, each disk has two labels which
63 * are updated before and after the uberblock is synced. Assuming we have
64 * labels and an uberblock with the following transaction groups:
67 * +------+ +------+ +------+
69 * | t10 | | t10 | | t10 |
71 * +------+ +------+ +------+
73 * In this stable state, the labels and the uberblock were all updated within
74 * the same transaction group (10). Each label is mirrored and checksummed, so
75 * that we can detect when we fail partway through writing the label.
77 * In order to identify which labels are valid, the labels are written in the
80 * 1. For each vdev, update 'L1' to the new label
81 * 2. Update the uberblock
82 * 3. For each vdev, update 'L2' to the new label
84 * Given arbitrary failure, we can determine the correct label to use based on
85 * the transaction group. If we fail after updating L1 but before updating the
86 * UB, we will notice that L1's transaction group is greater than the uberblock,
87 * so L2 must be valid. If we fail after writing the uberblock but before
88 * writing L2, we will notice that L2's transaction group is less than L1, and
89 * therefore L1 is valid.
91 * Another added complexity is that not every label is updated when the config
92 * is synced. If we add a single device, we do not want to have to re-write
93 * every label for every device in the pool. This means that both L1 and L2 may
94 * be older than the pool uberblock, because the necessary information is stored
101 * The vdev label consists of two distinct parts, and is wrapped within the
102 * vdev_label_t structure. The label includes 8k of padding to permit legacy
103 * VTOC disk labels, but is otherwise ignored.
105 * The first half of the label is a packed nvlist which contains pool wide
106 * properties, per-vdev properties, and configuration information. It is
107 * described in more detail below.
109 * The latter half of the label consists of a redundant array of uberblocks.
110 * These uberblocks are updated whenever a transaction group is committed,
111 * or when the configuration is updated. When a pool is loaded, we scan each
112 * vdev for the 'best' uberblock.
115 * Configuration Information
116 * -------------------------
118 * The nvlist describing the pool and vdev contains the following elements:
120 * version ZFS on-disk version
123 * txg Transaction group in which this label was written
124 * pool_guid Unique identifier for this pool
125 * vdev_tree An nvlist describing vdev tree.
127 * An nvlist of the features necessary for reading the MOS.
129 * Each leaf device label also contains the following:
131 * top_guid Unique ID for top-level vdev in which this is contained
132 * guid Unique ID for the leaf vdev
134 * The 'vs' configuration follows the format described in 'spa_config.c'.
137 #include <sys/zfs_context.h>
139 #include <sys/spa_impl.h>
142 #include <sys/vdev.h>
143 #include <sys/vdev_impl.h>
144 #include <sys/uberblock_impl.h>
145 #include <sys/metaslab.h>
146 #include <sys/metaslab_impl.h>
148 #include <sys/dsl_scan.h>
150 #include <sys/fs/zfs.h>
153 * Basic routines to read and write from a vdev label.
154 * Used throughout the rest of this file.
157 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
159 ASSERT(offset < sizeof (vdev_label_t));
160 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
162 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
163 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
167 * Returns back the vdev label associated with the passed in offset.
170 vdev_label_number(uint64_t psize, uint64_t offset)
174 if (offset >= psize - VDEV_LABEL_END_SIZE) {
175 offset -= psize - VDEV_LABEL_END_SIZE;
176 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
178 l = offset / sizeof (vdev_label_t);
179 return (l < VDEV_LABELS ? l : -1);
183 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
184 uint64_t size, zio_done_func_t *done, void *private, int flags)
187 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
188 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
189 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
191 zio_nowait(zio_read_phys(zio, vd,
192 vdev_label_offset(vd->vdev_psize, l, offset),
193 size, buf, ZIO_CHECKSUM_LABEL, done, private,
194 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
198 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
199 uint64_t size, zio_done_func_t *done, void *private, int flags)
202 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
203 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
204 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
206 zio_nowait(zio_write_phys(zio, vd,
207 vdev_label_offset(vd->vdev_psize, l, offset),
208 size, buf, ZIO_CHECKSUM_LABEL, done, private,
209 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
213 * Generate the nvlist representing this vdev's stats
216 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
222 vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
223 vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
225 vdev_get_stats_ex(vd, vs, vsx);
226 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
227 (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
229 kmem_free(vs, sizeof (*vs));
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]));
348 /* Add extended stats nvlist to main nvlist */
349 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
352 kmem_free(vsx, sizeof (*vsx));
356 * Generate the nvlist representing this vdev's config.
359 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
360 vdev_config_flag_t flags)
363 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
365 nv = fnvlist_alloc();
367 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
368 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
369 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
370 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
372 if (vd->vdev_path != NULL)
373 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
375 if (vd->vdev_devid != NULL)
376 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
378 if (vd->vdev_physpath != NULL)
379 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
382 if (vd->vdev_enc_sysfs_path != NULL)
383 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
384 vd->vdev_enc_sysfs_path);
386 if (vd->vdev_fru != NULL)
387 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
389 if (vd->vdev_nparity != 0) {
390 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
391 VDEV_TYPE_RAIDZ) == 0);
394 * Make sure someone hasn't managed to sneak a fancy new vdev
395 * into a crufty old storage pool.
397 ASSERT(vd->vdev_nparity == 1 ||
398 (vd->vdev_nparity <= 2 &&
399 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
400 (vd->vdev_nparity <= 3 &&
401 spa_version(spa) >= SPA_VERSION_RAIDZ3));
404 * Note that we'll add the nparity tag even on storage pools
405 * that only support a single parity device -- older software
406 * will just ignore it.
408 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
411 if (vd->vdev_wholedisk != -1ULL)
412 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
415 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
416 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
418 if (vd->vdev_isspare)
419 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
421 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
422 vd == vd->vdev_top) {
423 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
425 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
427 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
428 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
430 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
431 if (vd->vdev_removing) {
432 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
437 if (vd->vdev_dtl_sm != NULL) {
438 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
439 space_map_object(vd->vdev_dtl_sm));
442 if (vic->vic_mapping_object != 0) {
443 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
444 vic->vic_mapping_object);
447 if (vic->vic_births_object != 0) {
448 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
449 vic->vic_births_object);
452 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
453 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
454 vic->vic_prev_indirect_vdev);
458 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
460 if (flags & VDEV_CONFIG_MOS) {
461 if (vd->vdev_leaf_zap != 0) {
462 ASSERT(vd->vdev_ops->vdev_op_leaf);
463 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
467 if (vd->vdev_top_zap != 0) {
468 ASSERT(vd == vd->vdev_top);
469 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
475 vdev_config_generate_stats(vd, nv);
477 /* provide either current or previous scan information */
479 if (spa_scan_get_stats(spa, &ps) == 0) {
480 fnvlist_add_uint64_array(nv,
481 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
482 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
485 pool_removal_stat_t prs;
486 if (spa_removal_get_stats(spa, &prs) == 0) {
487 fnvlist_add_uint64_array(nv,
488 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
489 sizeof (prs) / sizeof (uint64_t));
493 * Note: this can be called from open context
494 * (spa_get_stats()), so we need the rwlock to prevent
495 * the mapping from being changed by condensing.
497 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
498 if (vd->vdev_indirect_mapping != NULL) {
499 ASSERT(vd->vdev_indirect_births != NULL);
500 vdev_indirect_mapping_t *vim =
501 vd->vdev_indirect_mapping;
502 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
503 vdev_indirect_mapping_size(vim));
505 rw_exit(&vd->vdev_indirect_rwlock);
506 if (vd->vdev_mg != NULL &&
507 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
509 * Compute approximately how much memory would be used
510 * for the indirect mapping if this device were to
513 * Note: If the frag metric is invalid, then not
514 * enough metaslabs have been converted to have
517 uint64_t seg_count = 0;
518 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
521 * There are the same number of allocated segments
522 * as free segments, so we will have at least one
523 * entry per free segment. However, small free
524 * segments (smaller than vdev_removal_max_span)
525 * will be combined with adjacent allocated segments
526 * as a single mapping.
528 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
529 if (1ULL << (i + 1) < vdev_removal_max_span) {
531 vd->vdev_mg->mg_histogram[i] <<
535 vd->vdev_mg->mg_histogram[i];
540 * The maximum length of a mapping is
541 * zfs_remove_max_segment, so we need at least one entry
542 * per zfs_remove_max_segment of allocated data.
544 seg_count += to_alloc / zfs_remove_max_segment;
546 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
548 sizeof (vdev_indirect_mapping_entry_phys_t));
552 if (!vd->vdev_ops->vdev_op_leaf) {
556 ASSERT(!vd->vdev_ishole);
558 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
561 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
562 vdev_t *cvd = vd->vdev_child[c];
565 * If we're generating an nvlist of removing
566 * vdevs then skip over any device which is
569 if ((flags & VDEV_CONFIG_REMOVING) &&
573 child[idx++] = vdev_config_generate(spa, cvd,
578 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
582 for (c = 0; c < idx; c++)
583 nvlist_free(child[c]);
585 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
588 const char *aux = NULL;
590 if (vd->vdev_offline && !vd->vdev_tmpoffline)
591 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
592 if (vd->vdev_resilver_txg != 0)
593 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
594 vd->vdev_resilver_txg);
595 if (vd->vdev_faulted)
596 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
597 if (vd->vdev_degraded)
598 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
599 if (vd->vdev_removed)
600 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
601 if (vd->vdev_unspare)
602 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
604 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
606 /* Set the reason why we're FAULTED/DEGRADED. */
607 switch (vd->vdev_stat.vs_aux) {
608 case VDEV_AUX_ERR_EXCEEDED:
609 aux = "err_exceeded";
612 case VDEV_AUX_EXTERNAL:
617 if (aux != NULL && !vd->vdev_tmpoffline) {
618 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
621 * We're healthy - clear any previous AUX_STATE values.
623 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
624 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
627 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
628 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
637 * Generate a view of the top-level vdevs. If we currently have holes
638 * in the namespace, then generate an array which contains a list of holey
639 * vdevs. Additionally, add the number of top-level children that currently
643 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
645 vdev_t *rvd = spa->spa_root_vdev;
649 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
651 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
652 vdev_t *tvd = rvd->vdev_child[c];
654 if (tvd->vdev_ishole) {
660 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
664 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
665 rvd->vdev_children) == 0);
667 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
671 * Returns the configuration from the label of the given vdev. For vdevs
672 * which don't have a txg value stored on their label (i.e. spares/cache)
673 * or have not been completely initialized (txg = 0) just return
674 * the configuration from the first valid label we find. Otherwise,
675 * find the most up-to-date label that does not exceed the specified
679 vdev_label_read_config(vdev_t *vd, uint64_t txg)
681 spa_t *spa = vd->vdev_spa;
682 nvlist_t *config = NULL;
686 uint64_t best_txg = 0;
687 uint64_t label_txg = 0;
689 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
690 ZIO_FLAG_SPECULATIVE;
692 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
694 if (!vdev_readable(vd))
697 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
698 vp = abd_to_buf(vp_abd);
701 for (int l = 0; l < VDEV_LABELS; l++) {
702 nvlist_t *label = NULL;
704 zio = zio_root(spa, NULL, NULL, flags);
706 vdev_label_read(zio, vd, l, vp_abd,
707 offsetof(vdev_label_t, vl_vdev_phys),
708 sizeof (vdev_phys_t), NULL, NULL, flags);
710 if (zio_wait(zio) == 0 &&
711 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
714 * Auxiliary vdevs won't have txg values in their
715 * labels and newly added vdevs may not have been
716 * completely initialized so just return the
717 * configuration from the first valid label we
720 error = nvlist_lookup_uint64(label,
721 ZPOOL_CONFIG_POOL_TXG, &label_txg);
722 if ((error || label_txg == 0) && !config) {
725 } else if (label_txg <= txg && label_txg > best_txg) {
726 best_txg = label_txg;
728 config = fnvlist_dup(label);
738 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
739 flags |= ZIO_FLAG_TRYHARD;
744 * We found a valid label but it didn't pass txg restrictions.
746 if (config == NULL && label_txg != 0) {
747 vdev_dbgmsg(vd, "label discarded as txg is too large "
748 "(%llu > %llu)", (u_longlong_t)label_txg,
758 * Determine if a device is in use. The 'spare_guid' parameter will be filled
759 * in with the device guid if this spare is active elsewhere on the system.
762 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
763 uint64_t *spare_guid, uint64_t *l2cache_guid)
765 spa_t *spa = vd->vdev_spa;
766 uint64_t state, pool_guid, device_guid, txg, spare_pool;
773 *l2cache_guid = 0ULL;
776 * Read the label, if any, and perform some basic sanity checks.
778 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
781 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
784 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
786 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
787 &device_guid) != 0) {
792 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
793 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
795 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
804 * Check to see if this device indeed belongs to the pool it claims to
805 * be a part of. The only way this is allowed is if the device is a hot
806 * spare (which we check for later on).
808 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
809 !spa_guid_exists(pool_guid, device_guid) &&
810 !spa_spare_exists(device_guid, NULL, NULL) &&
811 !spa_l2cache_exists(device_guid, NULL))
815 * If the transaction group is zero, then this an initialized (but
816 * unused) label. This is only an error if the create transaction
817 * on-disk is the same as the one we're using now, in which case the
818 * user has attempted to add the same vdev multiple times in the same
821 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
822 txg == 0 && vdtxg == crtxg)
826 * Check to see if this is a spare device. We do an explicit check for
827 * spa_has_spare() here because it may be on our pending list of spares
828 * to add. We also check if it is an l2cache device.
830 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
831 spa_has_spare(spa, device_guid)) {
833 *spare_guid = device_guid;
836 case VDEV_LABEL_CREATE:
837 case VDEV_LABEL_L2CACHE:
840 case VDEV_LABEL_REPLACE:
841 return (!spa_has_spare(spa, device_guid) ||
844 case VDEV_LABEL_SPARE:
845 return (spa_has_spare(spa, device_guid));
852 * Check to see if this is an l2cache device.
854 if (spa_l2cache_exists(device_guid, NULL))
858 * We can't rely on a pool's state if it's been imported
859 * read-only. Instead we look to see if the pools is marked
860 * read-only in the namespace and set the state to active.
862 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
863 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
864 spa_mode(spa) == FREAD)
865 state = POOL_STATE_ACTIVE;
868 * If the device is marked ACTIVE, then this device is in use by another
869 * pool on the system.
871 return (state == POOL_STATE_ACTIVE);
875 * Initialize a vdev label. We check to make sure each leaf device is not in
876 * use, and writable. We put down an initial label which we will later
877 * overwrite with a complete label. Note that it's important to do this
878 * sequentially, not in parallel, so that we catch cases of multiple use of the
879 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
883 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
885 spa_t *spa = vd->vdev_spa;
896 uint64_t spare_guid = 0, l2cache_guid = 0;
897 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
899 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
901 for (int c = 0; c < vd->vdev_children; c++)
902 if ((error = vdev_label_init(vd->vdev_child[c],
903 crtxg, reason)) != 0)
906 /* Track the creation time for this vdev */
907 vd->vdev_crtxg = crtxg;
909 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
913 * Dead vdevs cannot be initialized.
915 if (vdev_is_dead(vd))
916 return (SET_ERROR(EIO));
919 * Determine if the vdev is in use.
921 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
922 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
923 return (SET_ERROR(EBUSY));
926 * If this is a request to add or replace a spare or l2cache device
927 * that is in use elsewhere on the system, then we must update the
928 * guid (which was initialized to a random value) to reflect the
929 * actual GUID (which is shared between multiple pools).
931 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
932 spare_guid != 0ULL) {
933 uint64_t guid_delta = spare_guid - vd->vdev_guid;
935 vd->vdev_guid += guid_delta;
937 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
938 pvd->vdev_guid_sum += guid_delta;
941 * If this is a replacement, then we want to fallthrough to the
942 * rest of the code. If we're adding a spare, then it's already
943 * labeled appropriately and we can just return.
945 if (reason == VDEV_LABEL_SPARE)
947 ASSERT(reason == VDEV_LABEL_REPLACE ||
948 reason == VDEV_LABEL_SPLIT);
951 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
952 l2cache_guid != 0ULL) {
953 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
955 vd->vdev_guid += guid_delta;
957 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
958 pvd->vdev_guid_sum += guid_delta;
961 * If this is a replacement, then we want to fallthrough to the
962 * rest of the code. If we're adding an l2cache, then it's
963 * already labeled appropriately and we can just return.
965 if (reason == VDEV_LABEL_L2CACHE)
967 ASSERT(reason == VDEV_LABEL_REPLACE);
971 * Initialize its label.
973 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
974 abd_zero(vp_abd, sizeof (vdev_phys_t));
975 vp = abd_to_buf(vp_abd);
978 * Generate a label describing the pool and our top-level vdev.
979 * We mark it as being from txg 0 to indicate that it's not
980 * really part of an active pool just yet. The labels will
981 * be written again with a meaningful txg by spa_sync().
983 if (reason == VDEV_LABEL_SPARE ||
984 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
986 * For inactive hot spares, we generate a special label that
987 * identifies as a mutually shared hot spare. We write the
988 * label if we are adding a hot spare, or if we are removing an
989 * active hot spare (in which case we want to revert the
992 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
994 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
995 spa_version(spa)) == 0);
996 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
997 POOL_STATE_SPARE) == 0);
998 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
999 vd->vdev_guid) == 0);
1000 } else if (reason == VDEV_LABEL_L2CACHE ||
1001 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1003 * For level 2 ARC devices, add a special label.
1005 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1007 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1008 spa_version(spa)) == 0);
1009 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1010 POOL_STATE_L2CACHE) == 0);
1011 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1012 vd->vdev_guid) == 0);
1014 uint64_t txg = 0ULL;
1016 if (reason == VDEV_LABEL_SPLIT)
1017 txg = spa->spa_uberblock.ub_txg;
1018 label = spa_config_generate(spa, vd, txg, B_FALSE);
1021 * Add our creation time. This allows us to detect multiple
1022 * vdev uses as described above, and automatically expires if we
1025 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1029 buf = vp->vp_nvlist;
1030 buflen = sizeof (vp->vp_nvlist);
1032 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1036 /* EFAULT means nvlist_pack ran out of room */
1037 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1041 * Initialize uberblock template.
1043 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1044 abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1045 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1046 ub = abd_to_buf(ub_abd);
1049 /* Initialize the 2nd padding area. */
1050 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1051 abd_zero(pad2, VDEV_PAD_SIZE);
1054 * Write everything in parallel.
1057 zio = zio_root(spa, NULL, NULL, flags);
1059 for (int l = 0; l < VDEV_LABELS; l++) {
1061 vdev_label_write(zio, vd, l, vp_abd,
1062 offsetof(vdev_label_t, vl_vdev_phys),
1063 sizeof (vdev_phys_t), NULL, NULL, flags);
1066 * Skip the 1st padding area.
1067 * Zero out the 2nd padding area where it might have
1068 * left over data from previous filesystem format.
1070 vdev_label_write(zio, vd, l, pad2,
1071 offsetof(vdev_label_t, vl_pad2),
1072 VDEV_PAD_SIZE, NULL, NULL, flags);
1074 vdev_label_write(zio, vd, l, ub_abd,
1075 offsetof(vdev_label_t, vl_uberblock),
1076 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1079 error = zio_wait(zio);
1081 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1082 flags |= ZIO_FLAG_TRYHARD;
1092 * If this vdev hasn't been previously identified as a spare, then we
1093 * mark it as such only if a) we are labeling it as a spare, or b) it
1094 * exists as a spare elsewhere in the system. Do the same for
1095 * level 2 ARC devices.
1097 if (error == 0 && !vd->vdev_isspare &&
1098 (reason == VDEV_LABEL_SPARE ||
1099 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1102 if (error == 0 && !vd->vdev_isl2cache &&
1103 (reason == VDEV_LABEL_L2CACHE ||
1104 spa_l2cache_exists(vd->vdev_guid, NULL)))
1105 spa_l2cache_add(vd);
1111 * ==========================================================================
1112 * uberblock load/sync
1113 * ==========================================================================
1117 * Consider the following situation: txg is safely synced to disk. We've
1118 * written the first uberblock for txg + 1, and then we lose power. When we
1119 * come back up, we fail to see the uberblock for txg + 1 because, say,
1120 * it was on a mirrored device and the replica to which we wrote txg + 1
1121 * is now offline. If we then make some changes and sync txg + 1, and then
1122 * the missing replica comes back, then for a few seconds we'll have two
1123 * conflicting uberblocks on disk with the same txg. The solution is simple:
1124 * among uberblocks with equal txg, choose the one with the latest timestamp.
1127 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1129 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1133 return (AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp));
1137 uberblock_t *ubl_ubbest; /* Best uberblock */
1138 vdev_t *ubl_vd; /* vdev associated with the above */
1142 vdev_uberblock_load_done(zio_t *zio)
1144 vdev_t *vd = zio->io_vd;
1145 spa_t *spa = zio->io_spa;
1146 zio_t *rio = zio->io_private;
1147 uberblock_t *ub = abd_to_buf(zio->io_abd);
1148 struct ubl_cbdata *cbp = rio->io_private;
1150 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1152 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1153 mutex_enter(&rio->io_lock);
1154 if (ub->ub_txg <= spa->spa_load_max_txg &&
1155 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1157 * Keep track of the vdev in which this uberblock
1158 * was found. We will use this information later
1159 * to obtain the config nvlist associated with
1162 *cbp->ubl_ubbest = *ub;
1165 mutex_exit(&rio->io_lock);
1168 abd_free(zio->io_abd);
1172 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1173 struct ubl_cbdata *cbp)
1175 for (int c = 0; c < vd->vdev_children; c++)
1176 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1178 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1179 for (int l = 0; l < VDEV_LABELS; l++) {
1180 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1181 vdev_label_read(zio, vd, l,
1182 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1183 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1184 VDEV_UBERBLOCK_SIZE(vd),
1185 vdev_uberblock_load_done, zio, flags);
1192 * Reads the 'best' uberblock from disk along with its associated
1193 * configuration. First, we read the uberblock array of each label of each
1194 * vdev, keeping track of the uberblock with the highest txg in each array.
1195 * Then, we read the configuration from the same vdev as the best uberblock.
1198 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1201 spa_t *spa = rvd->vdev_spa;
1202 struct ubl_cbdata cb;
1203 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1204 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1209 bzero(ub, sizeof (uberblock_t));
1215 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1216 zio = zio_root(spa, NULL, &cb, flags);
1217 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1218 (void) zio_wait(zio);
1221 * It's possible that the best uberblock was discovered on a label
1222 * that has a configuration which was written in a future txg.
1223 * Search all labels on this vdev to find the configuration that
1224 * matches the txg for our uberblock.
1226 if (cb.ubl_vd != NULL) {
1227 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1228 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1230 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1231 if (*config == NULL && spa->spa_extreme_rewind) {
1232 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1233 "Trying again without txg restrictions.");
1234 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1236 if (*config == NULL) {
1237 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1240 spa_config_exit(spa, SCL_ALL, FTAG);
1244 * For use when a leaf vdev is expanded.
1245 * The location of labels 2 and 3 changed, and at the new location the
1246 * uberblock rings are either empty or contain garbage. The sync will write
1247 * new configs there because the vdev is dirty, but expansion also needs the
1248 * uberblock rings copied. Read them from label 0 which did not move.
1250 * Since the point is to populate labels {2,3} with valid uberblocks,
1251 * we zero uberblocks we fail to read or which are not valid.
1255 vdev_copy_uberblocks(vdev_t *vd)
1259 int locks = (SCL_L2ARC | SCL_ZIO);
1260 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1261 ZIO_FLAG_SPECULATIVE;
1263 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1265 ASSERT(vd->vdev_ops->vdev_op_leaf);
1267 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1269 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1271 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1272 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1273 const int src_label = 0;
1276 zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1277 vdev_label_read(zio, vd, src_label, ub_abd,
1278 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1281 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1282 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1284 for (int l = 2; l < VDEV_LABELS; l++)
1285 vdev_label_write(write_zio, vd, l, ub_abd,
1286 VDEV_UBERBLOCK_OFFSET(vd, n),
1287 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1288 flags | ZIO_FLAG_DONT_PROPAGATE);
1290 (void) zio_wait(write_zio);
1292 spa_config_exit(vd->vdev_spa, locks, FTAG);
1298 * On success, increment root zio's count of good writes.
1299 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1302 vdev_uberblock_sync_done(zio_t *zio)
1304 uint64_t *good_writes = zio->io_private;
1306 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1307 atomic_inc_64(good_writes);
1311 * Write the uberblock to all labels of all leaves of the specified vdev.
1314 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1315 uberblock_t *ub, vdev_t *vd, int flags)
1317 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1318 vdev_uberblock_sync(zio, good_writes,
1319 ub, vd->vdev_child[c], flags);
1322 if (!vd->vdev_ops->vdev_op_leaf)
1325 if (!vdev_writeable(vd))
1328 /* If the vdev was expanded, need to copy uberblock rings. */
1329 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1330 vd->vdev_copy_uberblocks == B_TRUE) {
1331 vdev_copy_uberblocks(vd);
1332 vd->vdev_copy_uberblocks = B_FALSE;
1335 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1336 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1338 /* Copy the uberblock_t into the ABD */
1339 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1340 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1341 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1343 for (int l = 0; l < VDEV_LABELS; l++)
1344 vdev_label_write(zio, vd, l, ub_abd,
1345 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1346 vdev_uberblock_sync_done, good_writes,
1347 flags | ZIO_FLAG_DONT_PROPAGATE);
1352 /* Sync the uberblocks to all vdevs in svd[] */
1354 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1356 spa_t *spa = svd[0]->vdev_spa;
1358 uint64_t good_writes = 0;
1360 zio = zio_root(spa, NULL, NULL, flags);
1362 for (int v = 0; v < svdcount; v++)
1363 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1365 (void) zio_wait(zio);
1368 * Flush the uberblocks to disk. This ensures that the odd labels
1369 * are no longer needed (because the new uberblocks and the even
1370 * labels are safely on disk), so it is safe to overwrite them.
1372 zio = zio_root(spa, NULL, NULL, flags);
1374 for (int v = 0; v < svdcount; v++) {
1375 if (vdev_writeable(svd[v])) {
1376 zio_flush(zio, svd[v]);
1380 (void) zio_wait(zio);
1382 return (good_writes >= 1 ? 0 : EIO);
1386 * On success, increment the count of good writes for our top-level vdev.
1389 vdev_label_sync_done(zio_t *zio)
1391 uint64_t *good_writes = zio->io_private;
1393 if (zio->io_error == 0)
1394 atomic_inc_64(good_writes);
1398 * If there weren't enough good writes, indicate failure to the parent.
1401 vdev_label_sync_top_done(zio_t *zio)
1403 uint64_t *good_writes = zio->io_private;
1405 if (*good_writes == 0)
1406 zio->io_error = SET_ERROR(EIO);
1408 kmem_free(good_writes, sizeof (uint64_t));
1412 * We ignore errors for log and cache devices, simply free the private data.
1415 vdev_label_sync_ignore_done(zio_t *zio)
1417 kmem_free(zio->io_private, sizeof (uint64_t));
1421 * Write all even or odd labels to all leaves of the specified vdev.
1424 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1425 vdev_t *vd, int l, uint64_t txg, int flags)
1433 for (int c = 0; c < vd->vdev_children; c++) {
1434 vdev_label_sync(zio, good_writes,
1435 vd->vdev_child[c], l, txg, flags);
1438 if (!vd->vdev_ops->vdev_op_leaf)
1441 if (!vdev_writeable(vd))
1445 * Generate a label describing the top-level config to which we belong.
1447 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1449 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1450 abd_zero(vp_abd, sizeof (vdev_phys_t));
1451 vp = abd_to_buf(vp_abd);
1453 buf = vp->vp_nvlist;
1454 buflen = sizeof (vp->vp_nvlist);
1456 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1457 for (; l < VDEV_LABELS; l += 2) {
1458 vdev_label_write(zio, vd, l, vp_abd,
1459 offsetof(vdev_label_t, vl_vdev_phys),
1460 sizeof (vdev_phys_t),
1461 vdev_label_sync_done, good_writes,
1462 flags | ZIO_FLAG_DONT_PROPAGATE);
1471 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1473 list_t *dl = &spa->spa_config_dirty_list;
1479 * Write the new labels to disk.
1481 zio = zio_root(spa, NULL, NULL, flags);
1483 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1484 uint64_t *good_writes;
1486 ASSERT(!vd->vdev_ishole);
1488 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1489 zio_t *vio = zio_null(zio, spa, NULL,
1490 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1491 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1492 good_writes, flags);
1493 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1497 error = zio_wait(zio);
1500 * Flush the new labels to disk.
1502 zio = zio_root(spa, NULL, NULL, flags);
1504 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1507 (void) zio_wait(zio);
1513 * Sync the uberblock and any changes to the vdev configuration.
1515 * The order of operations is carefully crafted to ensure that
1516 * if the system panics or loses power at any time, the state on disk
1517 * is still transactionally consistent. The in-line comments below
1518 * describe the failure semantics at each stage.
1520 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1521 * at any time, you can just call it again, and it will resume its work.
1524 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1526 spa_t *spa = svd[0]->vdev_spa;
1527 uberblock_t *ub = &spa->spa_uberblock;
1531 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1535 * Normally, we don't want to try too hard to write every label and
1536 * uberblock. If there is a flaky disk, we don't want the rest of the
1537 * sync process to block while we retry. But if we can't write a
1538 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1539 * bailing out and declaring the pool faulted.
1542 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1544 flags |= ZIO_FLAG_TRYHARD;
1547 ASSERT(ub->ub_txg <= txg);
1550 * If this isn't a resync due to I/O errors,
1551 * and nothing changed in this transaction group,
1552 * and the vdev configuration hasn't changed,
1553 * then there's nothing to do.
1555 if (ub->ub_txg < txg) {
1556 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1557 txg, spa->spa_mmp.mmp_delay);
1559 if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1563 if (txg > spa_freeze_txg(spa))
1566 ASSERT(txg <= spa->spa_final_txg);
1569 * Flush the write cache of every disk that's been written to
1570 * in this transaction group. This ensures that all blocks
1571 * written in this txg will be committed to stable storage
1572 * before any uberblock that references them.
1574 zio = zio_root(spa, NULL, NULL, flags);
1576 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1577 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1580 (void) zio_wait(zio);
1583 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1584 * system dies in the middle of this process, that's OK: all of the
1585 * even labels that made it to disk will be newer than any uberblock,
1586 * and will therefore be considered invalid. The odd labels (L1, L3),
1587 * which have not yet been touched, will still be valid. We flush
1588 * the new labels to disk to ensure that all even-label updates
1589 * are committed to stable storage before the uberblock update.
1591 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1595 * Sync the uberblocks to all vdevs in svd[].
1596 * If the system dies in the middle of this step, there are two cases
1597 * to consider, and the on-disk state is consistent either way:
1599 * (1) If none of the new uberblocks made it to disk, then the
1600 * previous uberblock will be the newest, and the odd labels
1601 * (which had not yet been touched) will be valid with respect
1602 * to that uberblock.
1604 * (2) If one or more new uberblocks made it to disk, then they
1605 * will be the newest, and the even labels (which had all
1606 * been successfully committed) will be valid with respect
1607 * to the new uberblocks.
1609 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1612 if (spa_multihost(spa))
1613 mmp_update_uberblock(spa, ub);
1616 * Sync out odd labels for every dirty vdev. If the system dies
1617 * in the middle of this process, the even labels and the new
1618 * uberblocks will suffice to open the pool. The next time
1619 * the pool is opened, the first thing we'll do -- before any
1620 * user data is modified -- is mark every vdev dirty so that
1621 * all labels will be brought up to date. We flush the new labels
1622 * to disk to ensure that all odd-label updates are committed to
1623 * stable storage before the next transaction group begins.
1625 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)