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.
26 * Copyright 2019 Joyent, Inc.
30 * Virtual Device Labels
31 * ---------------------
33 * The vdev label serves several distinct purposes:
35 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
36 * identity within the pool.
38 * 2. Verify that all the devices given in a configuration are present
41 * 3. Determine the uberblock for the pool.
43 * 4. In case of an import operation, determine the configuration of the
44 * toplevel vdev of which it is a part.
46 * 5. If an import operation cannot find all the devices in the pool,
47 * provide enough information to the administrator to determine which
48 * devices are missing.
50 * It is important to note that while the kernel is responsible for writing the
51 * label, it only consumes the information in the first three cases. The
52 * latter information is only consumed in userland when determining the
53 * configuration to import a pool.
59 * Before describing the contents of the label, it's important to understand how
60 * the labels are written and updated with respect to the uberblock.
62 * When the pool configuration is altered, either because it was newly created
63 * or a device was added, we want to update all the labels such that we can deal
64 * with fatal failure at any point. To this end, each disk has two labels which
65 * are updated before and after the uberblock is synced. Assuming we have
66 * labels and an uberblock with the following transaction groups:
69 * +------+ +------+ +------+
71 * | t10 | | t10 | | t10 |
73 * +------+ +------+ +------+
75 * In this stable state, the labels and the uberblock were all updated within
76 * the same transaction group (10). Each label is mirrored and checksummed, so
77 * that we can detect when we fail partway through writing the label.
79 * In order to identify which labels are valid, the labels are written in the
82 * 1. For each vdev, update 'L1' to the new label
83 * 2. Update the uberblock
84 * 3. For each vdev, update 'L2' to the new label
86 * Given arbitrary failure, we can determine the correct label to use based on
87 * the transaction group. If we fail after updating L1 but before updating the
88 * UB, we will notice that L1's transaction group is greater than the uberblock,
89 * so L2 must be valid. If we fail after writing the uberblock but before
90 * writing L2, we will notice that L2's transaction group is less than L1, and
91 * therefore L1 is valid.
93 * Another added complexity is that not every label is updated when the config
94 * is synced. If we add a single device, we do not want to have to re-write
95 * every label for every device in the pool. This means that both L1 and L2 may
96 * be older than the pool uberblock, because the necessary information is stored
103 * The vdev label consists of two distinct parts, and is wrapped within the
104 * vdev_label_t structure. The label includes 8k of padding to permit legacy
105 * VTOC disk labels, but is otherwise ignored.
107 * The first half of the label is a packed nvlist which contains pool wide
108 * properties, per-vdev properties, and configuration information. It is
109 * described in more detail below.
111 * The latter half of the label consists of a redundant array of uberblocks.
112 * These uberblocks are updated whenever a transaction group is committed,
113 * or when the configuration is updated. When a pool is loaded, we scan each
114 * vdev for the 'best' uberblock.
117 * Configuration Information
118 * -------------------------
120 * The nvlist describing the pool and vdev contains the following elements:
122 * version ZFS on-disk version
125 * txg Transaction group in which this label was written
126 * pool_guid Unique identifier for this pool
127 * vdev_tree An nvlist describing vdev tree.
129 * An nvlist of the features necessary for reading the MOS.
131 * Each leaf device label also contains the following:
133 * top_guid Unique ID for top-level vdev in which this is contained
134 * guid Unique ID for the leaf vdev
136 * The 'vs' configuration follows the format described in 'spa_config.c'.
139 #include <sys/zfs_context.h>
141 #include <sys/spa_impl.h>
144 #include <sys/vdev.h>
145 #include <sys/vdev_impl.h>
146 #include <sys/uberblock_impl.h>
147 #include <sys/metaslab.h>
148 #include <sys/metaslab_impl.h>
150 #include <sys/dsl_scan.h>
152 #include <sys/fs/zfs.h>
153 #include <sys/trim_map.h>
155 static boolean_t vdev_trim_on_init = B_TRUE;
156 SYSCTL_DECL(_vfs_zfs_vdev);
157 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RWTUN,
158 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation");
161 * Basic routines to read and write from a vdev label.
162 * Used throughout the rest of this file.
165 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
167 ASSERT(offset < sizeof (vdev_label_t));
168 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
170 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
171 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
175 * Returns back the vdev label associated with the passed in offset.
178 vdev_label_number(uint64_t psize, uint64_t offset)
182 if (offset >= psize - VDEV_LABEL_END_SIZE) {
183 offset -= psize - VDEV_LABEL_END_SIZE;
184 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
186 l = offset / sizeof (vdev_label_t);
187 return (l < VDEV_LABELS ? l : -1);
191 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
192 uint64_t size, zio_done_func_t *done, void *private, int flags)
195 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
196 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
197 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
199 zio_nowait(zio_read_phys(zio, vd,
200 vdev_label_offset(vd->vdev_psize, l, offset),
201 size, buf, ZIO_CHECKSUM_LABEL, done, private,
202 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
206 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
207 uint64_t size, zio_done_func_t *done, void *private, int flags)
210 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
211 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
212 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
214 zio_nowait(zio_write_phys(zio, vd,
215 vdev_label_offset(vd->vdev_psize, l, offset),
216 size, buf, ZIO_CHECKSUM_LABEL, done, private,
217 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
221 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
223 spa_t *spa = vd->vdev_spa;
225 if (vd != spa->spa_root_vdev)
228 /* provide either current or previous scan information */
230 if (spa_scan_get_stats(spa, &ps) == 0) {
231 fnvlist_add_uint64_array(nvl,
232 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
233 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
236 pool_removal_stat_t prs;
237 if (spa_removal_get_stats(spa, &prs) == 0) {
238 fnvlist_add_uint64_array(nvl,
239 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
240 sizeof (prs) / sizeof (uint64_t));
243 pool_checkpoint_stat_t pcs;
244 if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
245 fnvlist_add_uint64_array(nvl,
246 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
247 sizeof (pcs) / sizeof (uint64_t));
252 * Generate the nvlist representing this vdev's config.
255 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
256 vdev_config_flag_t flags)
259 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
261 nv = fnvlist_alloc();
263 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
264 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
265 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
266 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
268 if (vd->vdev_path != NULL)
269 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
271 if (vd->vdev_devid != NULL)
272 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
274 if (vd->vdev_physpath != NULL)
275 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
278 if (vd->vdev_fru != NULL)
279 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
281 if (vd->vdev_nparity != 0) {
282 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
283 VDEV_TYPE_RAIDZ) == 0);
286 * Make sure someone hasn't managed to sneak a fancy new vdev
287 * into a crufty old storage pool.
289 ASSERT(vd->vdev_nparity == 1 ||
290 (vd->vdev_nparity <= 2 &&
291 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
292 (vd->vdev_nparity <= 3 &&
293 spa_version(spa) >= SPA_VERSION_RAIDZ3));
296 * Note that we'll add the nparity tag even on storage pools
297 * that only support a single parity device -- older software
298 * will just ignore it.
300 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
303 if (vd->vdev_wholedisk != -1ULL)
304 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
307 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
308 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
310 if (vd->vdev_isspare)
311 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
313 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
314 vd == vd->vdev_top) {
315 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
317 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
319 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
320 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
322 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
323 if (vd->vdev_removing) {
324 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
328 /* zpool command expects alloc class data */
329 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
330 const char *bias = NULL;
332 switch (vd->vdev_alloc_bias) {
334 bias = VDEV_ALLOC_BIAS_LOG;
336 case VDEV_BIAS_SPECIAL:
337 bias = VDEV_ALLOC_BIAS_SPECIAL;
339 case VDEV_BIAS_DEDUP:
340 bias = VDEV_ALLOC_BIAS_DEDUP;
343 ASSERT3U(vd->vdev_alloc_bias, ==,
346 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
351 if (vd->vdev_dtl_sm != NULL) {
352 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
353 space_map_object(vd->vdev_dtl_sm));
356 if (vic->vic_mapping_object != 0) {
357 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
358 vic->vic_mapping_object);
361 if (vic->vic_births_object != 0) {
362 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
363 vic->vic_births_object);
366 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
367 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
368 vic->vic_prev_indirect_vdev);
372 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
374 if (flags & VDEV_CONFIG_MOS) {
375 if (vd->vdev_leaf_zap != 0) {
376 ASSERT(vd->vdev_ops->vdev_op_leaf);
377 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
381 if (vd->vdev_top_zap != 0) {
382 ASSERT(vd == vd->vdev_top);
383 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
391 vdev_get_stats(vd, &vs);
392 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
393 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
395 root_vdev_actions_getprogress(vd, nv);
398 * Note: this can be called from open context
399 * (spa_get_stats()), so we need the rwlock to prevent
400 * the mapping from being changed by condensing.
402 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
403 if (vd->vdev_indirect_mapping != NULL) {
404 ASSERT(vd->vdev_indirect_births != NULL);
405 vdev_indirect_mapping_t *vim =
406 vd->vdev_indirect_mapping;
407 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
408 vdev_indirect_mapping_size(vim));
410 rw_exit(&vd->vdev_indirect_rwlock);
411 if (vd->vdev_mg != NULL &&
412 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
414 * Compute approximately how much memory would be used
415 * for the indirect mapping if this device were to
418 * Note: If the frag metric is invalid, then not
419 * enough metaslabs have been converted to have
422 uint64_t seg_count = 0;
423 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
426 * There are the same number of allocated segments
427 * as free segments, so we will have at least one
428 * entry per free segment. However, small free
429 * segments (smaller than vdev_removal_max_span)
430 * will be combined with adjacent allocated segments
431 * as a single mapping.
433 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
434 if (1ULL << (i + 1) < vdev_removal_max_span) {
436 vd->vdev_mg->mg_histogram[i] <<
440 vd->vdev_mg->mg_histogram[i];
445 * The maximum length of a mapping is
446 * zfs_remove_max_segment, so we need at least one entry
447 * per zfs_remove_max_segment of allocated data.
449 seg_count += to_alloc / zfs_remove_max_segment;
451 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
453 sizeof (vdev_indirect_mapping_entry_phys_t));
457 if (!vd->vdev_ops->vdev_op_leaf) {
461 ASSERT(!vd->vdev_ishole);
463 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
466 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
467 vdev_t *cvd = vd->vdev_child[c];
470 * If we're generating an nvlist of removing
471 * vdevs then skip over any device which is
474 if ((flags & VDEV_CONFIG_REMOVING) &&
478 child[idx++] = vdev_config_generate(spa, cvd,
483 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
487 for (c = 0; c < idx; c++)
488 nvlist_free(child[c]);
490 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
493 const char *aux = NULL;
495 if (vd->vdev_offline && !vd->vdev_tmpoffline)
496 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
497 if (vd->vdev_resilver_txg != 0)
498 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
499 vd->vdev_resilver_txg);
500 if (vd->vdev_faulted)
501 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
502 if (vd->vdev_degraded)
503 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
504 if (vd->vdev_removed)
505 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
506 if (vd->vdev_unspare)
507 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
509 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
511 switch (vd->vdev_stat.vs_aux) {
512 case VDEV_AUX_ERR_EXCEEDED:
513 aux = "err_exceeded";
516 case VDEV_AUX_EXTERNAL:
522 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
524 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
525 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
534 * Generate a view of the top-level vdevs. If we currently have holes
535 * in the namespace, then generate an array which contains a list of holey
536 * vdevs. Additionally, add the number of top-level children that currently
540 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
542 vdev_t *rvd = spa->spa_root_vdev;
546 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
548 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
549 vdev_t *tvd = rvd->vdev_child[c];
551 if (tvd->vdev_ishole) {
557 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
561 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
562 rvd->vdev_children) == 0);
564 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
568 * Returns the configuration from the label of the given vdev. For vdevs
569 * which don't have a txg value stored on their label (i.e. spares/cache)
570 * or have not been completely initialized (txg = 0) just return
571 * the configuration from the first valid label we find. Otherwise,
572 * find the most up-to-date label that does not exceed the specified
576 vdev_label_read_config(vdev_t *vd, uint64_t txg)
578 spa_t *spa = vd->vdev_spa;
579 nvlist_t *config = NULL;
583 uint64_t best_txg = 0;
584 uint64_t label_txg = 0;
586 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
587 ZIO_FLAG_SPECULATIVE;
589 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
591 if (!vdev_readable(vd))
594 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
595 vp = abd_to_buf(vp_abd);
598 for (int l = 0; l < VDEV_LABELS; l++) {
599 nvlist_t *label = NULL;
601 zio = zio_root(spa, NULL, NULL, flags);
603 vdev_label_read(zio, vd, l, vp_abd,
604 offsetof(vdev_label_t, vl_vdev_phys),
605 sizeof (vdev_phys_t), NULL, NULL, flags);
607 if (zio_wait(zio) == 0 &&
608 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
611 * Auxiliary vdevs won't have txg values in their
612 * labels and newly added vdevs may not have been
613 * completely initialized so just return the
614 * configuration from the first valid label we
617 error = nvlist_lookup_uint64(label,
618 ZPOOL_CONFIG_POOL_TXG, &label_txg);
619 if ((error || label_txg == 0) && !config) {
622 } else if (label_txg <= txg && label_txg > best_txg) {
623 best_txg = label_txg;
625 config = fnvlist_dup(label);
635 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
636 flags |= ZIO_FLAG_TRYHARD;
641 * We found a valid label but it didn't pass txg restrictions.
643 if (config == NULL && label_txg != 0) {
644 vdev_dbgmsg(vd, "label discarded as txg is too large "
645 "(%llu > %llu)", (u_longlong_t)label_txg,
655 * Determine if a device is in use. The 'spare_guid' parameter will be filled
656 * in with the device guid if this spare is active elsewhere on the system.
659 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
660 uint64_t *spare_guid, uint64_t *l2cache_guid)
662 spa_t *spa = vd->vdev_spa;
663 uint64_t state, pool_guid, device_guid, txg, spare_pool;
670 *l2cache_guid = 0ULL;
673 * Read the label, if any, and perform some basic sanity checks.
675 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
678 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
681 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
683 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
684 &device_guid) != 0) {
689 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
690 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
692 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
701 * Check to see if this device indeed belongs to the pool it claims to
702 * be a part of. The only way this is allowed is if the device is a hot
703 * spare (which we check for later on).
705 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
706 !spa_guid_exists(pool_guid, device_guid) &&
707 !spa_spare_exists(device_guid, NULL, NULL) &&
708 !spa_l2cache_exists(device_guid, NULL))
712 * If the transaction group is zero, then this an initialized (but
713 * unused) label. This is only an error if the create transaction
714 * on-disk is the same as the one we're using now, in which case the
715 * user has attempted to add the same vdev multiple times in the same
718 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
719 txg == 0 && vdtxg == crtxg)
723 * Check to see if this is a spare device. We do an explicit check for
724 * spa_has_spare() here because it may be on our pending list of spares
725 * to add. We also check if it is an l2cache device.
727 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
728 spa_has_spare(spa, device_guid)) {
730 *spare_guid = device_guid;
733 case VDEV_LABEL_CREATE:
734 case VDEV_LABEL_L2CACHE:
737 case VDEV_LABEL_REPLACE:
738 return (!spa_has_spare(spa, device_guid) ||
741 case VDEV_LABEL_SPARE:
742 return (spa_has_spare(spa, device_guid));
747 * Check to see if this is an l2cache device.
749 if (spa_l2cache_exists(device_guid, NULL))
753 * We can't rely on a pool's state if it's been imported
754 * read-only. Instead we look to see if the pools is marked
755 * read-only in the namespace and set the state to active.
757 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
758 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
759 spa_mode(spa) == FREAD)
760 state = POOL_STATE_ACTIVE;
763 * If the device is marked ACTIVE, then this device is in use by another
764 * pool on the system.
766 return (state == POOL_STATE_ACTIVE);
770 * Initialize a vdev label. We check to make sure each leaf device is not in
771 * use, and writable. We put down an initial label which we will later
772 * overwrite with a complete label. Note that it's important to do this
773 * sequentially, not in parallel, so that we catch cases of multiple use of the
774 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
778 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
780 spa_t *spa = vd->vdev_spa;
791 uint64_t spare_guid, l2cache_guid;
792 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
794 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
796 for (int c = 0; c < vd->vdev_children; c++)
797 if ((error = vdev_label_init(vd->vdev_child[c],
798 crtxg, reason)) != 0)
801 /* Track the creation time for this vdev */
802 vd->vdev_crtxg = crtxg;
804 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
808 * Dead vdevs cannot be initialized.
810 if (vdev_is_dead(vd))
811 return (SET_ERROR(EIO));
814 * Determine if the vdev is in use.
816 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
817 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
818 return (SET_ERROR(EBUSY));
821 * If this is a request to add or replace a spare or l2cache device
822 * that is in use elsewhere on the system, then we must update the
823 * guid (which was initialized to a random value) to reflect the
824 * actual GUID (which is shared between multiple pools).
826 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
827 spare_guid != 0ULL) {
828 uint64_t guid_delta = spare_guid - vd->vdev_guid;
830 vd->vdev_guid += guid_delta;
832 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
833 pvd->vdev_guid_sum += guid_delta;
836 * If this is a replacement, then we want to fallthrough to the
837 * rest of the code. If we're adding a spare, then it's already
838 * labeled appropriately and we can just return.
840 if (reason == VDEV_LABEL_SPARE)
842 ASSERT(reason == VDEV_LABEL_REPLACE ||
843 reason == VDEV_LABEL_SPLIT);
846 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
847 l2cache_guid != 0ULL) {
848 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
850 vd->vdev_guid += guid_delta;
852 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
853 pvd->vdev_guid_sum += guid_delta;
856 * If this is a replacement, then we want to fallthrough to the
857 * rest of the code. If we're adding an l2cache, then it's
858 * already labeled appropriately and we can just return.
860 if (reason == VDEV_LABEL_L2CACHE)
862 ASSERT(reason == VDEV_LABEL_REPLACE);
866 * TRIM the whole thing, excluding the blank space and boot header
867 * as specified by ZFS On-Disk Specification (section 1.3), so that
868 * we start with a clean slate.
869 * It's just an optimization, so we don't care if it fails.
870 * Don't TRIM if removing so that we don't interfere with zpool
873 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim &&
874 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE ||
875 reason == VDEV_LABEL_L2CACHE))
876 zio_wait(zio_trim(NULL, spa, vd, VDEV_SKIP_SIZE,
877 vd->vdev_psize - VDEV_SKIP_SIZE));
880 * Initialize its label.
882 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
883 abd_zero(vp_abd, sizeof (vdev_phys_t));
884 vp = abd_to_buf(vp_abd);
887 * Generate a label describing the pool and our top-level vdev.
888 * We mark it as being from txg 0 to indicate that it's not
889 * really part of an active pool just yet. The labels will
890 * be written again with a meaningful txg by spa_sync().
892 if (reason == VDEV_LABEL_SPARE ||
893 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
895 * For inactive hot spares, we generate a special label that
896 * identifies as a mutually shared hot spare. We write the
897 * label if we are adding a hot spare, or if we are removing an
898 * active hot spare (in which case we want to revert the
901 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
903 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
904 spa_version(spa)) == 0);
905 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
906 POOL_STATE_SPARE) == 0);
907 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
908 vd->vdev_guid) == 0);
909 } else if (reason == VDEV_LABEL_L2CACHE ||
910 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
912 * For level 2 ARC devices, add a special label.
914 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
916 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
917 spa_version(spa)) == 0);
918 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
919 POOL_STATE_L2CACHE) == 0);
920 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
921 vd->vdev_guid) == 0);
925 if (reason == VDEV_LABEL_SPLIT)
926 txg = spa->spa_uberblock.ub_txg;
927 label = spa_config_generate(spa, vd, txg, B_FALSE);
930 * Add our creation time. This allows us to detect multiple
931 * vdev uses as described above, and automatically expires if we
934 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
939 buflen = sizeof (vp->vp_nvlist);
941 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
945 /* EFAULT means nvlist_pack ran out of room */
946 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
950 * Initialize uberblock template.
952 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
953 abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
954 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
955 ub = abd_to_buf(ub_abd);
958 /* Initialize the 2nd padding area. */
959 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
960 abd_zero(bootenv, VDEV_PAD_SIZE);
963 * Write everything in parallel.
966 zio = zio_root(spa, NULL, NULL, flags);
968 for (int l = 0; l < VDEV_LABELS; l++) {
970 vdev_label_write(zio, vd, l, vp_abd,
971 offsetof(vdev_label_t, vl_vdev_phys),
972 sizeof (vdev_phys_t), NULL, NULL, flags);
975 * Skip the 1st padding area.
976 * Zero out the 2nd padding area where it might have
977 * left over data from previous filesystem format.
979 vdev_label_write(zio, vd, l, bootenv,
980 offsetof(vdev_label_t, vl_be),
981 VDEV_PAD_SIZE, NULL, NULL, flags);
983 vdev_label_write(zio, vd, l, ub_abd,
984 offsetof(vdev_label_t, vl_uberblock),
985 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
988 error = zio_wait(zio);
990 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
991 flags |= ZIO_FLAG_TRYHARD;
1001 * If this vdev hasn't been previously identified as a spare, then we
1002 * mark it as such only if a) we are labeling it as a spare, or b) it
1003 * exists as a spare elsewhere in the system. Do the same for
1004 * level 2 ARC devices.
1006 if (error == 0 && !vd->vdev_isspare &&
1007 (reason == VDEV_LABEL_SPARE ||
1008 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1011 if (error == 0 && !vd->vdev_isl2cache &&
1012 (reason == VDEV_LABEL_L2CACHE ||
1013 spa_l2cache_exists(vd->vdev_guid, NULL)))
1014 spa_l2cache_add(vd);
1020 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1021 * callback to finish, store our abd in the callback pointer. Otherwise, we
1022 * just free our abd and return.
1025 vdev_label_read_bootenv_done(zio_t *zio)
1027 zio_t *rio = zio->io_private;
1028 abd_t **cbp = rio->io_private;
1030 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1032 if (zio->io_error == 0) {
1033 mutex_enter(&rio->io_lock);
1035 /* Will free this buffer in vdev_label_read_bootenv. */
1038 abd_free(zio->io_abd);
1040 mutex_exit(&rio->io_lock);
1042 abd_free(zio->io_abd);
1047 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1049 for (int c = 0; c < vd->vdev_children; c++)
1050 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1053 * We just use the first label that has a correct checksum; the
1054 * bootloader should have rewritten them all to be the same on boot,
1055 * and any changes we made since boot have been the same across all
1058 * While grub supports writing to all four labels, other bootloaders
1059 * don't, so we only use the first two labels to store boot
1062 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1063 for (int l = 0; l < VDEV_LABELS / 2; l++) {
1064 vdev_label_read(zio, vd, l,
1065 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1066 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1067 vdev_label_read_bootenv_done, zio, flags);
1073 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *command)
1075 spa_t *spa = rvd->vdev_spa;
1077 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1078 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1081 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1083 zio_t *zio = zio_root(spa, NULL, &abd, flags);
1084 vdev_label_read_bootenv_impl(zio, rvd, flags);
1085 int err = zio_wait(zio);
1088 vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1089 if (vbe->vbe_version != VB_RAW) {
1091 return (SET_ERROR(ENOTSUP));
1093 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1094 fnvlist_add_string(command, "envmap", vbe->vbe_bootenv);
1095 /* abd was allocated in vdev_label_read_bootenv_impl() */
1097 /* If we managed to read any successfully, return success. */
1104 vdev_label_write_bootenv(vdev_t *vd, char *envmap)
1107 spa_t *spa = vd->vdev_spa;
1108 vdev_boot_envblock_t *bootenv;
1109 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1112 if (strlen(envmap) >= sizeof (bootenv->vbe_bootenv)) {
1113 return (SET_ERROR(E2BIG));
1116 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1118 for (int c = 0; c < vd->vdev_children; c++) {
1119 int child_err = vdev_label_write_bootenv(vd->vdev_child[c],
1122 * As long as any of the disks managed to write all of their
1123 * labels successfully, return success.
1129 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1130 !vdev_writeable(vd)) {
1133 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1134 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1135 abd_zero(abd, VDEV_PAD_SIZE);
1136 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1138 char *buf = bootenv->vbe_bootenv;
1139 (void) strlcpy(buf, envmap, sizeof (bootenv->vbe_bootenv));
1140 bootenv->vbe_version = VB_RAW;
1141 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1144 zio = zio_root(spa, NULL, NULL, flags);
1145 for (int l = 0; l < VDEV_LABELS / 2; l++) {
1146 vdev_label_write(zio, vd, l, abd,
1147 offsetof(vdev_label_t, vl_be),
1148 VDEV_PAD_SIZE, NULL, NULL, flags);
1151 error = zio_wait(zio);
1152 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1153 flags |= ZIO_FLAG_TRYHARD;
1162 vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size)
1164 spa_t *spa = vd->vdev_spa;
1167 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1170 if (size > VDEV_PAD_SIZE)
1173 if (!vd->vdev_ops->vdev_op_leaf)
1175 if (vdev_is_dead(vd))
1178 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1180 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1181 abd_zero(pad2, VDEV_PAD_SIZE);
1182 abd_copy_from_buf(pad2, buf, size);
1185 zio = zio_root(spa, NULL, NULL, flags);
1186 vdev_label_write(zio, vd, 0, pad2,
1187 offsetof(vdev_label_t, vl_be),
1188 VDEV_PAD_SIZE, NULL, NULL, flags);
1189 error = zio_wait(zio);
1190 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1191 flags |= ZIO_FLAG_TRYHARD;
1200 * ==========================================================================
1201 * uberblock load/sync
1202 * ==========================================================================
1206 * Consider the following situation: txg is safely synced to disk. We've
1207 * written the first uberblock for txg + 1, and then we lose power. When we
1208 * come back up, we fail to see the uberblock for txg + 1 because, say,
1209 * it was on a mirrored device and the replica to which we wrote txg + 1
1210 * is now offline. If we then make some changes and sync txg + 1, and then
1211 * the missing replica comes back, then for a few seconds we'll have two
1212 * conflicting uberblocks on disk with the same txg. The solution is simple:
1213 * among uberblocks with equal txg, choose the one with the latest timestamp.
1216 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1218 int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1223 cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1228 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1229 * ZFS, e.g. zfsonlinux >= 0.7.
1231 * If one ub has MMP and the other does not, they were written by
1232 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1235 * Since timestamp and txg are the same if we get this far, either is
1236 * acceptable for importing the pool.
1238 unsigned int seq1 = 0;
1239 unsigned int seq2 = 0;
1241 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1242 seq1 = MMP_SEQ(ub1);
1244 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1245 seq2 = MMP_SEQ(ub2);
1247 return (AVL_CMP(seq1, seq2));
1251 uberblock_t *ubl_ubbest; /* Best uberblock */
1252 vdev_t *ubl_vd; /* vdev associated with the above */
1256 vdev_uberblock_load_done(zio_t *zio)
1258 vdev_t *vd = zio->io_vd;
1259 spa_t *spa = zio->io_spa;
1260 zio_t *rio = zio->io_private;
1261 uberblock_t *ub = abd_to_buf(zio->io_abd);
1262 struct ubl_cbdata *cbp = rio->io_private;
1264 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1266 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1267 mutex_enter(&rio->io_lock);
1268 if (ub->ub_txg <= spa->spa_load_max_txg &&
1269 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1271 * Keep track of the vdev in which this uberblock
1272 * was found. We will use this information later
1273 * to obtain the config nvlist associated with
1276 *cbp->ubl_ubbest = *ub;
1279 mutex_exit(&rio->io_lock);
1282 abd_free(zio->io_abd);
1286 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1287 struct ubl_cbdata *cbp)
1289 for (int c = 0; c < vd->vdev_children; c++)
1290 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1292 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1293 for (int l = 0; l < VDEV_LABELS; l++) {
1294 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1295 vdev_label_read(zio, vd, l,
1296 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1297 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1298 VDEV_UBERBLOCK_SIZE(vd),
1299 vdev_uberblock_load_done, zio, flags);
1306 * Reads the 'best' uberblock from disk along with its associated
1307 * configuration. First, we read the uberblock array of each label of each
1308 * vdev, keeping track of the uberblock with the highest txg in each array.
1309 * Then, we read the configuration from the same vdev as the best uberblock.
1312 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1315 spa_t *spa = rvd->vdev_spa;
1316 struct ubl_cbdata cb;
1317 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1318 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1323 bzero(ub, sizeof (uberblock_t));
1329 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1330 zio = zio_root(spa, NULL, &cb, flags);
1331 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1332 (void) zio_wait(zio);
1335 * It's possible that the best uberblock was discovered on a label
1336 * that has a configuration which was written in a future txg.
1337 * Search all labels on this vdev to find the configuration that
1338 * matches the txg for our uberblock.
1340 if (cb.ubl_vd != NULL) {
1341 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1342 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1344 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1345 if (*config == NULL && spa->spa_extreme_rewind) {
1346 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1347 "Trying again without txg restrictions.");
1348 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1350 if (*config == NULL) {
1351 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1354 spa_config_exit(spa, SCL_ALL, FTAG);
1358 * On success, increment root zio's count of good writes.
1359 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1362 vdev_uberblock_sync_done(zio_t *zio)
1364 uint64_t *good_writes = zio->io_private;
1366 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1367 atomic_inc_64(good_writes);
1371 * Write the uberblock to all labels of all leaves of the specified vdev.
1374 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1375 uberblock_t *ub, vdev_t *vd, int flags)
1377 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1378 vdev_uberblock_sync(zio, good_writes,
1379 ub, vd->vdev_child[c], flags);
1382 if (!vd->vdev_ops->vdev_op_leaf)
1385 if (!vdev_writeable(vd))
1388 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1389 int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1391 /* Copy the uberblock_t into the ABD */
1392 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1393 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1394 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1396 for (int l = 0; l < VDEV_LABELS; l++)
1397 vdev_label_write(zio, vd, l, ub_abd,
1398 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1399 vdev_uberblock_sync_done, good_writes,
1400 flags | ZIO_FLAG_DONT_PROPAGATE);
1405 /* Sync the uberblocks to all vdevs in svd[] */
1407 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1409 spa_t *spa = svd[0]->vdev_spa;
1411 uint64_t good_writes = 0;
1413 zio = zio_root(spa, NULL, NULL, flags);
1415 for (int v = 0; v < svdcount; v++)
1416 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1418 (void) zio_wait(zio);
1421 * Flush the uberblocks to disk. This ensures that the odd labels
1422 * are no longer needed (because the new uberblocks and the even
1423 * labels are safely on disk), so it is safe to overwrite them.
1425 zio = zio_root(spa, NULL, NULL, flags);
1427 for (int v = 0; v < svdcount; v++) {
1428 if (vdev_writeable(svd[v])) {
1429 zio_flush(zio, svd[v]);
1433 (void) zio_wait(zio);
1435 return (good_writes >= 1 ? 0 : EIO);
1439 * On success, increment the count of good writes for our top-level vdev.
1442 vdev_label_sync_done(zio_t *zio)
1444 uint64_t *good_writes = zio->io_private;
1446 if (zio->io_error == 0)
1447 atomic_inc_64(good_writes);
1451 * If there weren't enough good writes, indicate failure to the parent.
1454 vdev_label_sync_top_done(zio_t *zio)
1456 uint64_t *good_writes = zio->io_private;
1458 if (*good_writes == 0)
1459 zio->io_error = SET_ERROR(EIO);
1461 kmem_free(good_writes, sizeof (uint64_t));
1465 * We ignore errors for log and cache devices, simply free the private data.
1468 vdev_label_sync_ignore_done(zio_t *zio)
1470 kmem_free(zio->io_private, sizeof (uint64_t));
1474 * Write all even or odd labels to all leaves of the specified vdev.
1477 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1478 vdev_t *vd, int l, uint64_t txg, int flags)
1486 for (int c = 0; c < vd->vdev_children; c++) {
1487 vdev_label_sync(zio, good_writes,
1488 vd->vdev_child[c], l, txg, flags);
1491 if (!vd->vdev_ops->vdev_op_leaf)
1494 if (!vdev_writeable(vd))
1498 * Generate a label describing the top-level config to which we belong.
1500 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1502 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1503 abd_zero(vp_abd, sizeof (vdev_phys_t));
1504 vp = abd_to_buf(vp_abd);
1506 buf = vp->vp_nvlist;
1507 buflen = sizeof (vp->vp_nvlist);
1509 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1510 for (; l < VDEV_LABELS; l += 2) {
1511 vdev_label_write(zio, vd, l, vp_abd,
1512 offsetof(vdev_label_t, vl_vdev_phys),
1513 sizeof (vdev_phys_t),
1514 vdev_label_sync_done, good_writes,
1515 flags | ZIO_FLAG_DONT_PROPAGATE);
1524 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1526 list_t *dl = &spa->spa_config_dirty_list;
1532 * Write the new labels to disk.
1534 zio = zio_root(spa, NULL, NULL, flags);
1536 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1537 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1540 ASSERT(!vd->vdev_ishole);
1542 zio_t *vio = zio_null(zio, spa, NULL,
1543 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1544 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1545 good_writes, flags);
1546 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1550 error = zio_wait(zio);
1553 * Flush the new labels to disk.
1555 zio = zio_root(spa, NULL, NULL, flags);
1557 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1560 (void) zio_wait(zio);
1566 * Sync the uberblock and any changes to the vdev configuration.
1568 * The order of operations is carefully crafted to ensure that
1569 * if the system panics or loses power at any time, the state on disk
1570 * is still transactionally consistent. The in-line comments below
1571 * describe the failure semantics at each stage.
1573 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1574 * at any time, you can just call it again, and it will resume its work.
1577 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1579 spa_t *spa = svd[0]->vdev_spa;
1580 uberblock_t *ub = &spa->spa_uberblock;
1582 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1584 ASSERT(svdcount != 0);
1587 * Normally, we don't want to try too hard to write every label and
1588 * uberblock. If there is a flaky disk, we don't want the rest of the
1589 * sync process to block while we retry. But if we can't write a
1590 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1591 * bailing out and declaring the pool faulted.
1594 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1596 flags |= ZIO_FLAG_TRYHARD;
1599 ASSERT(ub->ub_txg <= txg);
1602 * If this isn't a resync due to I/O errors,
1603 * and nothing changed in this transaction group,
1604 * and the vdev configuration hasn't changed,
1605 * then there's nothing to do.
1607 if (ub->ub_txg < txg) {
1608 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1609 txg, spa->spa_mmp.mmp_delay);
1611 if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1615 if (txg > spa_freeze_txg(spa))
1618 ASSERT(txg <= spa->spa_final_txg);
1621 * Flush the write cache of every disk that's been written to
1622 * in this transaction group. This ensures that all blocks
1623 * written in this txg will be committed to stable storage
1624 * before any uberblock that references them.
1626 zio_t *zio = zio_root(spa, NULL, NULL, flags);
1629 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1630 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1633 (void) zio_wait(zio);
1636 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1637 * system dies in the middle of this process, that's OK: all of the
1638 * even labels that made it to disk will be newer than any uberblock,
1639 * and will therefore be considered invalid. The odd labels (L1, L3),
1640 * which have not yet been touched, will still be valid. We flush
1641 * the new labels to disk to ensure that all even-label updates
1642 * are committed to stable storage before the uberblock update.
1644 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1645 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1646 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1647 "for pool '%s' when syncing out the even labels "
1648 "of dirty vdevs", error, spa_name(spa));
1654 * Sync the uberblocks to all vdevs in svd[].
1655 * If the system dies in the middle of this step, there are two cases
1656 * to consider, and the on-disk state is consistent either way:
1658 * (1) If none of the new uberblocks made it to disk, then the
1659 * previous uberblock will be the newest, and the odd labels
1660 * (which had not yet been touched) will be valid with respect
1661 * to that uberblock.
1663 * (2) If one or more new uberblocks made it to disk, then they
1664 * will be the newest, and the even labels (which had all
1665 * been successfully committed) will be valid with respect
1666 * to the new uberblocks.
1668 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1669 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1670 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1671 "%d for pool '%s'", error, spa_name(spa));
1676 if (spa_multihost(spa))
1677 mmp_update_uberblock(spa, ub);
1680 * Sync out odd labels for every dirty vdev. If the system dies
1681 * in the middle of this process, the even labels and the new
1682 * uberblocks will suffice to open the pool. The next time
1683 * the pool is opened, the first thing we'll do -- before any
1684 * user data is modified -- is mark every vdev dirty so that
1685 * all labels will be brought up to date. We flush the new labels
1686 * to disk to ensure that all odd-label updates are committed to
1687 * stable storage before the next transaction group begins.
1689 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1690 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1691 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1692 "for pool '%s' when syncing out the odd labels of "
1693 "dirty vdevs", error, spa_name(spa));
1698 trim_thread_wakeup(spa);