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, 2015 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>
147 #include <sys/dsl_scan.h>
149 #include <sys/fs/zfs.h>
150 #include <sys/trim_map.h>
152 static boolean_t vdev_trim_on_init = B_TRUE;
153 SYSCTL_DECL(_vfs_zfs_vdev);
154 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW,
155 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation");
158 * Basic routines to read and write from a vdev label.
159 * Used throughout the rest of this file.
162 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
164 ASSERT(offset < sizeof (vdev_label_t));
165 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
167 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
168 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
172 * Returns back the vdev label associated with the passed in offset.
175 vdev_label_number(uint64_t psize, uint64_t offset)
179 if (offset >= psize - VDEV_LABEL_END_SIZE) {
180 offset -= psize - VDEV_LABEL_END_SIZE;
181 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
183 l = offset / sizeof (vdev_label_t);
184 return (l < VDEV_LABELS ? l : -1);
188 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
189 uint64_t size, zio_done_func_t *done, void *private, int flags)
191 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
193 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
195 zio_nowait(zio_read_phys(zio, vd,
196 vdev_label_offset(vd->vdev_psize, l, offset),
197 size, buf, ZIO_CHECKSUM_LABEL, done, private,
198 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
202 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
203 uint64_t size, zio_done_func_t *done, void *private, int flags)
205 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
206 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
207 (SCL_CONFIG | SCL_STATE) &&
208 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
209 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
211 zio_nowait(zio_write_phys(zio, vd,
212 vdev_label_offset(vd->vdev_psize, l, offset),
213 size, buf, ZIO_CHECKSUM_LABEL, done, private,
214 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
218 * Generate the nvlist representing this vdev's config.
221 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
222 vdev_config_flag_t flags)
226 nv = fnvlist_alloc();
228 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
229 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
230 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
231 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
233 if (vd->vdev_path != NULL)
234 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
236 if (vd->vdev_devid != NULL)
237 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
239 if (vd->vdev_physpath != NULL)
240 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
243 if (vd->vdev_fru != NULL)
244 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
246 if (vd->vdev_nparity != 0) {
247 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
248 VDEV_TYPE_RAIDZ) == 0);
251 * Make sure someone hasn't managed to sneak a fancy new vdev
252 * into a crufty old storage pool.
254 ASSERT(vd->vdev_nparity == 1 ||
255 (vd->vdev_nparity <= 2 &&
256 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
257 (vd->vdev_nparity <= 3 &&
258 spa_version(spa) >= SPA_VERSION_RAIDZ3));
261 * Note that we'll add the nparity tag even on storage pools
262 * that only support a single parity device -- older software
263 * will just ignore it.
265 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
268 if (vd->vdev_wholedisk != -1ULL)
269 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
272 if (vd->vdev_not_present)
273 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
275 if (vd->vdev_isspare)
276 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
278 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
279 vd == vd->vdev_top) {
280 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
282 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
284 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
285 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
287 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
288 if (vd->vdev_removing)
289 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
293 if (vd->vdev_dtl_sm != NULL) {
294 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
295 space_map_object(vd->vdev_dtl_sm));
299 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
301 if (flags & VDEV_CONFIG_MOS) {
302 if (vd->vdev_leaf_zap != 0) {
303 ASSERT(vd->vdev_ops->vdev_op_leaf);
304 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
308 if (vd->vdev_top_zap != 0) {
309 ASSERT(vd == vd->vdev_top);
310 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
319 vdev_get_stats(vd, &vs);
320 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
321 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
323 /* provide either current or previous scan information */
324 if (spa_scan_get_stats(spa, &ps) == 0) {
325 fnvlist_add_uint64_array(nv,
326 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
327 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
331 if (!vd->vdev_ops->vdev_op_leaf) {
335 ASSERT(!vd->vdev_ishole);
337 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
340 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
341 vdev_t *cvd = vd->vdev_child[c];
344 * If we're generating an nvlist of removing
345 * vdevs then skip over any device which is
348 if ((flags & VDEV_CONFIG_REMOVING) &&
352 child[idx++] = vdev_config_generate(spa, cvd,
357 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
361 for (c = 0; c < idx; c++)
362 nvlist_free(child[c]);
364 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
367 const char *aux = NULL;
369 if (vd->vdev_offline && !vd->vdev_tmpoffline)
370 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
371 if (vd->vdev_resilver_txg != 0)
372 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
373 vd->vdev_resilver_txg);
374 if (vd->vdev_faulted)
375 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
376 if (vd->vdev_degraded)
377 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
378 if (vd->vdev_removed)
379 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
380 if (vd->vdev_unspare)
381 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
383 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
385 switch (vd->vdev_stat.vs_aux) {
386 case VDEV_AUX_ERR_EXCEEDED:
387 aux = "err_exceeded";
390 case VDEV_AUX_EXTERNAL:
396 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
398 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
399 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
408 * Generate a view of the top-level vdevs. If we currently have holes
409 * in the namespace, then generate an array which contains a list of holey
410 * vdevs. Additionally, add the number of top-level children that currently
414 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
416 vdev_t *rvd = spa->spa_root_vdev;
420 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
422 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
423 vdev_t *tvd = rvd->vdev_child[c];
425 if (tvd->vdev_ishole)
430 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
434 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
435 rvd->vdev_children) == 0);
437 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
441 * Returns the configuration from the label of the given vdev. For vdevs
442 * which don't have a txg value stored on their label (i.e. spares/cache)
443 * or have not been completely initialized (txg = 0) just return
444 * the configuration from the first valid label we find. Otherwise,
445 * find the most up-to-date label that does not exceed the specified
449 vdev_label_read_config(vdev_t *vd, uint64_t txg)
451 spa_t *spa = vd->vdev_spa;
452 nvlist_t *config = NULL;
456 uint64_t best_txg = 0;
458 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
459 ZIO_FLAG_SPECULATIVE;
461 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
463 if (!vdev_readable(vd))
466 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
467 vp = abd_to_buf(vp_abd);
470 for (int l = 0; l < VDEV_LABELS; l++) {
471 nvlist_t *label = NULL;
473 zio = zio_root(spa, NULL, NULL, flags);
475 vdev_label_read(zio, vd, l, vp_abd,
476 offsetof(vdev_label_t, vl_vdev_phys),
477 sizeof (vdev_phys_t), NULL, NULL, flags);
479 if (zio_wait(zio) == 0 &&
480 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
482 uint64_t label_txg = 0;
485 * Auxiliary vdevs won't have txg values in their
486 * labels and newly added vdevs may not have been
487 * completely initialized so just return the
488 * configuration from the first valid label we
491 error = nvlist_lookup_uint64(label,
492 ZPOOL_CONFIG_POOL_TXG, &label_txg);
493 if ((error || label_txg == 0) && !config) {
496 } else if (label_txg <= txg && label_txg > best_txg) {
497 best_txg = label_txg;
499 config = fnvlist_dup(label);
509 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
510 flags |= ZIO_FLAG_TRYHARD;
520 * Determine if a device is in use. The 'spare_guid' parameter will be filled
521 * in with the device guid if this spare is active elsewhere on the system.
524 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
525 uint64_t *spare_guid, uint64_t *l2cache_guid)
527 spa_t *spa = vd->vdev_spa;
528 uint64_t state, pool_guid, device_guid, txg, spare_pool;
535 *l2cache_guid = 0ULL;
538 * Read the label, if any, and perform some basic sanity checks.
540 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
543 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
546 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
548 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
549 &device_guid) != 0) {
554 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
555 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
557 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
566 * Check to see if this device indeed belongs to the pool it claims to
567 * be a part of. The only way this is allowed is if the device is a hot
568 * spare (which we check for later on).
570 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
571 !spa_guid_exists(pool_guid, device_guid) &&
572 !spa_spare_exists(device_guid, NULL, NULL) &&
573 !spa_l2cache_exists(device_guid, NULL))
577 * If the transaction group is zero, then this an initialized (but
578 * unused) label. This is only an error if the create transaction
579 * on-disk is the same as the one we're using now, in which case the
580 * user has attempted to add the same vdev multiple times in the same
583 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
584 txg == 0 && vdtxg == crtxg)
588 * Check to see if this is a spare device. We do an explicit check for
589 * spa_has_spare() here because it may be on our pending list of spares
590 * to add. We also check if it is an l2cache device.
592 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
593 spa_has_spare(spa, device_guid)) {
595 *spare_guid = device_guid;
598 case VDEV_LABEL_CREATE:
599 case VDEV_LABEL_L2CACHE:
602 case VDEV_LABEL_REPLACE:
603 return (!spa_has_spare(spa, device_guid) ||
606 case VDEV_LABEL_SPARE:
607 return (spa_has_spare(spa, device_guid));
612 * Check to see if this is an l2cache device.
614 if (spa_l2cache_exists(device_guid, NULL))
618 * We can't rely on a pool's state if it's been imported
619 * read-only. Instead we look to see if the pools is marked
620 * read-only in the namespace and set the state to active.
622 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
623 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
624 spa_mode(spa) == FREAD)
625 state = POOL_STATE_ACTIVE;
628 * If the device is marked ACTIVE, then this device is in use by another
629 * pool on the system.
631 return (state == POOL_STATE_ACTIVE);
635 * Initialize a vdev label. We check to make sure each leaf device is not in
636 * use, and writable. We put down an initial label which we will later
637 * overwrite with a complete label. Note that it's important to do this
638 * sequentially, not in parallel, so that we catch cases of multiple use of the
639 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
643 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
645 spa_t *spa = vd->vdev_spa;
656 uint64_t spare_guid, l2cache_guid;
657 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
659 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
661 for (int c = 0; c < vd->vdev_children; c++)
662 if ((error = vdev_label_init(vd->vdev_child[c],
663 crtxg, reason)) != 0)
666 /* Track the creation time for this vdev */
667 vd->vdev_crtxg = crtxg;
669 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
673 * Dead vdevs cannot be initialized.
675 if (vdev_is_dead(vd))
676 return (SET_ERROR(EIO));
679 * Determine if the vdev is in use.
681 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
682 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
683 return (SET_ERROR(EBUSY));
686 * If this is a request to add or replace a spare or l2cache device
687 * that is in use elsewhere on the system, then we must update the
688 * guid (which was initialized to a random value) to reflect the
689 * actual GUID (which is shared between multiple pools).
691 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
692 spare_guid != 0ULL) {
693 uint64_t guid_delta = spare_guid - vd->vdev_guid;
695 vd->vdev_guid += guid_delta;
697 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
698 pvd->vdev_guid_sum += guid_delta;
701 * If this is a replacement, then we want to fallthrough to the
702 * rest of the code. If we're adding a spare, then it's already
703 * labeled appropriately and we can just return.
705 if (reason == VDEV_LABEL_SPARE)
707 ASSERT(reason == VDEV_LABEL_REPLACE ||
708 reason == VDEV_LABEL_SPLIT);
711 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
712 l2cache_guid != 0ULL) {
713 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
715 vd->vdev_guid += guid_delta;
717 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
718 pvd->vdev_guid_sum += guid_delta;
721 * If this is a replacement, then we want to fallthrough to the
722 * rest of the code. If we're adding an l2cache, then it's
723 * already labeled appropriately and we can just return.
725 if (reason == VDEV_LABEL_L2CACHE)
727 ASSERT(reason == VDEV_LABEL_REPLACE);
731 * TRIM the whole thing so that we start with a clean slate.
732 * It's just an optimization, so we don't care if it fails.
733 * Don't TRIM if removing so that we don't interfere with zpool
736 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim &&
737 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE ||
738 reason == VDEV_LABEL_L2CACHE))
739 zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize));
742 * Initialize its label.
744 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
745 abd_zero(vp_abd, sizeof (vdev_phys_t));
746 vp = abd_to_buf(vp_abd);
749 * Generate a label describing the pool and our top-level vdev.
750 * We mark it as being from txg 0 to indicate that it's not
751 * really part of an active pool just yet. The labels will
752 * be written again with a meaningful txg by spa_sync().
754 if (reason == VDEV_LABEL_SPARE ||
755 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
757 * For inactive hot spares, we generate a special label that
758 * identifies as a mutually shared hot spare. We write the
759 * label if we are adding a hot spare, or if we are removing an
760 * active hot spare (in which case we want to revert the
763 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
765 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
766 spa_version(spa)) == 0);
767 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
768 POOL_STATE_SPARE) == 0);
769 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
770 vd->vdev_guid) == 0);
771 } else if (reason == VDEV_LABEL_L2CACHE ||
772 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
774 * For level 2 ARC devices, add a special label.
776 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
778 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
779 spa_version(spa)) == 0);
780 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
781 POOL_STATE_L2CACHE) == 0);
782 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
783 vd->vdev_guid) == 0);
787 if (reason == VDEV_LABEL_SPLIT)
788 txg = spa->spa_uberblock.ub_txg;
789 label = spa_config_generate(spa, vd, txg, B_FALSE);
792 * Add our creation time. This allows us to detect multiple
793 * vdev uses as described above, and automatically expires if we
796 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
801 buflen = sizeof (vp->vp_nvlist);
803 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
807 /* EFAULT means nvlist_pack ran out of room */
808 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
812 * Initialize uberblock template.
814 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
815 abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
816 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
817 ub = abd_to_buf(ub_abd);
820 /* Initialize the 2nd padding area. */
821 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
822 abd_zero(pad2, VDEV_PAD_SIZE);
825 * Write everything in parallel.
828 zio = zio_root(spa, NULL, NULL, flags);
830 for (int l = 0; l < VDEV_LABELS; l++) {
832 vdev_label_write(zio, vd, l, vp_abd,
833 offsetof(vdev_label_t, vl_vdev_phys),
834 sizeof (vdev_phys_t), NULL, NULL, flags);
837 * Skip the 1st padding area.
838 * Zero out the 2nd padding area where it might have
839 * left over data from previous filesystem format.
841 vdev_label_write(zio, vd, l, pad2,
842 offsetof(vdev_label_t, vl_pad2),
843 VDEV_PAD_SIZE, NULL, NULL, flags);
845 vdev_label_write(zio, vd, l, ub_abd,
846 offsetof(vdev_label_t, vl_uberblock),
847 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
850 error = zio_wait(zio);
852 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
853 flags |= ZIO_FLAG_TRYHARD;
863 * If this vdev hasn't been previously identified as a spare, then we
864 * mark it as such only if a) we are labeling it as a spare, or b) it
865 * exists as a spare elsewhere in the system. Do the same for
866 * level 2 ARC devices.
868 if (error == 0 && !vd->vdev_isspare &&
869 (reason == VDEV_LABEL_SPARE ||
870 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
873 if (error == 0 && !vd->vdev_isl2cache &&
874 (reason == VDEV_LABEL_L2CACHE ||
875 spa_l2cache_exists(vd->vdev_guid, NULL)))
882 vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size)
884 spa_t *spa = vd->vdev_spa;
887 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
890 if (size > VDEV_PAD_SIZE)
893 if (!vd->vdev_ops->vdev_op_leaf)
895 if (vdev_is_dead(vd))
898 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
900 pad2 = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
901 abd_zero(pad2, VDEV_PAD_SIZE);
902 abd_copy_from_buf(pad2, buf, size);
905 zio = zio_root(spa, NULL, NULL, flags);
906 vdev_label_write(zio, vd, 0, pad2,
907 offsetof(vdev_label_t, vl_pad2),
908 VDEV_PAD_SIZE, NULL, NULL, flags);
909 error = zio_wait(zio);
910 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
911 flags |= ZIO_FLAG_TRYHARD;
920 * ==========================================================================
921 * uberblock load/sync
922 * ==========================================================================
926 * Consider the following situation: txg is safely synced to disk. We've
927 * written the first uberblock for txg + 1, and then we lose power. When we
928 * come back up, we fail to see the uberblock for txg + 1 because, say,
929 * it was on a mirrored device and the replica to which we wrote txg + 1
930 * is now offline. If we then make some changes and sync txg + 1, and then
931 * the missing replica comes back, then for a few seconds we'll have two
932 * conflicting uberblocks on disk with the same txg. The solution is simple:
933 * among uberblocks with equal txg, choose the one with the latest timestamp.
936 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
938 if (ub1->ub_txg < ub2->ub_txg)
940 if (ub1->ub_txg > ub2->ub_txg)
943 if (ub1->ub_timestamp < ub2->ub_timestamp)
945 if (ub1->ub_timestamp > ub2->ub_timestamp)
952 uberblock_t *ubl_ubbest; /* Best uberblock */
953 vdev_t *ubl_vd; /* vdev associated with the above */
957 vdev_uberblock_load_done(zio_t *zio)
959 vdev_t *vd = zio->io_vd;
960 spa_t *spa = zio->io_spa;
961 zio_t *rio = zio->io_private;
962 uberblock_t *ub = abd_to_buf(zio->io_abd);
963 struct ubl_cbdata *cbp = rio->io_private;
965 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
967 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
968 mutex_enter(&rio->io_lock);
969 if (ub->ub_txg <= spa->spa_load_max_txg &&
970 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
972 * Keep track of the vdev in which this uberblock
973 * was found. We will use this information later
974 * to obtain the config nvlist associated with
977 *cbp->ubl_ubbest = *ub;
980 mutex_exit(&rio->io_lock);
983 abd_free(zio->io_abd);
987 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
988 struct ubl_cbdata *cbp)
990 for (int c = 0; c < vd->vdev_children; c++)
991 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
993 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
994 for (int l = 0; l < VDEV_LABELS; l++) {
995 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
996 vdev_label_read(zio, vd, l,
997 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
998 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
999 VDEV_UBERBLOCK_SIZE(vd),
1000 vdev_uberblock_load_done, zio, flags);
1007 * Reads the 'best' uberblock from disk along with its associated
1008 * configuration. First, we read the uberblock array of each label of each
1009 * vdev, keeping track of the uberblock with the highest txg in each array.
1010 * Then, we read the configuration from the same vdev as the best uberblock.
1013 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1016 spa_t *spa = rvd->vdev_spa;
1017 struct ubl_cbdata cb;
1018 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1019 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1024 bzero(ub, sizeof (uberblock_t));
1030 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1031 zio = zio_root(spa, NULL, &cb, flags);
1032 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1033 (void) zio_wait(zio);
1036 * It's possible that the best uberblock was discovered on a label
1037 * that has a configuration which was written in a future txg.
1038 * Search all labels on this vdev to find the configuration that
1039 * matches the txg for our uberblock.
1041 if (cb.ubl_vd != NULL)
1042 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1043 spa_config_exit(spa, SCL_ALL, FTAG);
1047 * On success, increment root zio's count of good writes.
1048 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1051 vdev_uberblock_sync_done(zio_t *zio)
1053 uint64_t *good_writes = zio->io_private;
1055 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1056 atomic_inc_64(good_writes);
1060 * Write the uberblock to all labels of all leaves of the specified vdev.
1063 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1065 for (int c = 0; c < vd->vdev_children; c++)
1066 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1068 if (!vd->vdev_ops->vdev_op_leaf)
1071 if (!vdev_writeable(vd))
1074 int n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1076 /* Copy the uberblock_t into the ABD */
1077 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1078 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1079 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1081 for (int l = 0; l < VDEV_LABELS; l++)
1082 vdev_label_write(zio, vd, l, ub_abd,
1083 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1084 vdev_uberblock_sync_done, zio->io_private,
1085 flags | ZIO_FLAG_DONT_PROPAGATE);
1090 /* Sync the uberblocks to all vdevs in svd[] */
1092 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1094 spa_t *spa = svd[0]->vdev_spa;
1096 uint64_t good_writes = 0;
1098 zio = zio_root(spa, NULL, &good_writes, flags);
1100 for (int v = 0; v < svdcount; v++)
1101 vdev_uberblock_sync(zio, ub, svd[v], flags);
1103 (void) zio_wait(zio);
1106 * Flush the uberblocks to disk. This ensures that the odd labels
1107 * are no longer needed (because the new uberblocks and the even
1108 * labels are safely on disk), so it is safe to overwrite them.
1110 zio = zio_root(spa, NULL, NULL, flags);
1112 for (int v = 0; v < svdcount; v++)
1113 zio_flush(zio, svd[v]);
1115 (void) zio_wait(zio);
1117 return (good_writes >= 1 ? 0 : EIO);
1121 * On success, increment the count of good writes for our top-level vdev.
1124 vdev_label_sync_done(zio_t *zio)
1126 uint64_t *good_writes = zio->io_private;
1128 if (zio->io_error == 0)
1129 atomic_inc_64(good_writes);
1133 * If there weren't enough good writes, indicate failure to the parent.
1136 vdev_label_sync_top_done(zio_t *zio)
1138 uint64_t *good_writes = zio->io_private;
1140 if (*good_writes == 0)
1141 zio->io_error = SET_ERROR(EIO);
1143 kmem_free(good_writes, sizeof (uint64_t));
1147 * We ignore errors for log and cache devices, simply free the private data.
1150 vdev_label_sync_ignore_done(zio_t *zio)
1152 kmem_free(zio->io_private, sizeof (uint64_t));
1156 * Write all even or odd labels to all leaves of the specified vdev.
1159 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1167 for (int c = 0; c < vd->vdev_children; c++)
1168 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1170 if (!vd->vdev_ops->vdev_op_leaf)
1173 if (!vdev_writeable(vd))
1177 * Generate a label describing the top-level config to which we belong.
1179 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1181 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1182 abd_zero(vp_abd, sizeof (vdev_phys_t));
1183 vp = abd_to_buf(vp_abd);
1185 buf = vp->vp_nvlist;
1186 buflen = sizeof (vp->vp_nvlist);
1188 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1189 for (; l < VDEV_LABELS; l += 2) {
1190 vdev_label_write(zio, vd, l, vp_abd,
1191 offsetof(vdev_label_t, vl_vdev_phys),
1192 sizeof (vdev_phys_t),
1193 vdev_label_sync_done, zio->io_private,
1194 flags | ZIO_FLAG_DONT_PROPAGATE);
1203 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1205 list_t *dl = &spa->spa_config_dirty_list;
1211 * Write the new labels to disk.
1213 zio = zio_root(spa, NULL, NULL, flags);
1215 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1216 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1219 ASSERT(!vd->vdev_ishole);
1221 zio_t *vio = zio_null(zio, spa, NULL,
1222 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1223 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1224 good_writes, flags);
1225 vdev_label_sync(vio, vd, l, txg, flags);
1229 error = zio_wait(zio);
1232 * Flush the new labels to disk.
1234 zio = zio_root(spa, NULL, NULL, flags);
1236 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1239 (void) zio_wait(zio);
1245 * Sync the uberblock and any changes to the vdev configuration.
1247 * The order of operations is carefully crafted to ensure that
1248 * if the system panics or loses power at any time, the state on disk
1249 * is still transactionally consistent. The in-line comments below
1250 * describe the failure semantics at each stage.
1252 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1253 * at any time, you can just call it again, and it will resume its work.
1256 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1258 spa_t *spa = svd[0]->vdev_spa;
1259 uberblock_t *ub = &spa->spa_uberblock;
1263 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1267 * Normally, we don't want to try too hard to write every label and
1268 * uberblock. If there is a flaky disk, we don't want the rest of the
1269 * sync process to block while we retry. But if we can't write a
1270 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1271 * bailing out and declaring the pool faulted.
1274 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1276 flags |= ZIO_FLAG_TRYHARD;
1279 ASSERT(ub->ub_txg <= txg);
1282 * If this isn't a resync due to I/O errors,
1283 * and nothing changed in this transaction group,
1284 * and the vdev configuration hasn't changed,
1285 * then there's nothing to do.
1287 if (ub->ub_txg < txg &&
1288 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1289 list_is_empty(&spa->spa_config_dirty_list))
1292 if (txg > spa_freeze_txg(spa))
1295 ASSERT(txg <= spa->spa_final_txg);
1298 * Flush the write cache of every disk that's been written to
1299 * in this transaction group. This ensures that all blocks
1300 * written in this txg will be committed to stable storage
1301 * before any uberblock that references them.
1303 zio = zio_root(spa, NULL, NULL, flags);
1305 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1306 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1309 (void) zio_wait(zio);
1312 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1313 * system dies in the middle of this process, that's OK: all of the
1314 * even labels that made it to disk will be newer than any uberblock,
1315 * and will therefore be considered invalid. The odd labels (L1, L3),
1316 * which have not yet been touched, will still be valid. We flush
1317 * the new labels to disk to ensure that all even-label updates
1318 * are committed to stable storage before the uberblock update.
1320 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1324 * Sync the uberblocks to all vdevs in svd[].
1325 * If the system dies in the middle of this step, there are two cases
1326 * to consider, and the on-disk state is consistent either way:
1328 * (1) If none of the new uberblocks made it to disk, then the
1329 * previous uberblock will be the newest, and the odd labels
1330 * (which had not yet been touched) will be valid with respect
1331 * to that uberblock.
1333 * (2) If one or more new uberblocks made it to disk, then they
1334 * will be the newest, and the even labels (which had all
1335 * been successfully committed) will be valid with respect
1336 * to the new uberblocks.
1338 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1342 * Sync out odd labels for every dirty vdev. If the system dies
1343 * in the middle of this process, the even labels and the new
1344 * uberblocks will suffice to open the pool. The next time
1345 * the pool is opened, the first thing we'll do -- before any
1346 * user data is modified -- is mark every vdev dirty so that
1347 * all labels will be brought up to date. We flush the new labels
1348 * to disk to ensure that all odd-label updates are committed to
1349 * stable storage before the next transaction group begins.
1351 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1354 trim_thread_wakeup(spa);