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) 2013 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>
148 #include <sys/trim_map.h>
149 #include <sys/fs/zfs.h>
151 static boolean_t vdev_trim_on_init = B_TRUE;
152 SYSCTL_DECL(_vfs_zfs_vdev);
153 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW,
154 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation");
157 * Basic routines to read and write from a vdev label.
158 * Used throughout the rest of this file.
161 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
163 ASSERT(offset < sizeof (vdev_label_t));
164 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
166 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
167 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
171 * Returns back the vdev label associated with the passed in offset.
174 vdev_label_number(uint64_t psize, uint64_t offset)
178 if (offset >= psize - VDEV_LABEL_END_SIZE) {
179 offset -= psize - VDEV_LABEL_END_SIZE;
180 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
182 l = offset / sizeof (vdev_label_t);
183 return (l < VDEV_LABELS ? l : -1);
187 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
188 uint64_t size, zio_done_func_t *done, void *private, int flags)
190 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
192 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
194 zio_nowait(zio_read_phys(zio, vd,
195 vdev_label_offset(vd->vdev_psize, l, offset),
196 size, buf, ZIO_CHECKSUM_LABEL, done, private,
197 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
201 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
202 uint64_t size, zio_done_func_t *done, void *private, int flags)
204 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
205 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
206 (SCL_CONFIG | SCL_STATE) &&
207 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
208 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
210 zio_nowait(zio_write_phys(zio, vd,
211 vdev_label_offset(vd->vdev_psize, l, offset),
212 size, buf, ZIO_CHECKSUM_LABEL, done, private,
213 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
217 * Generate the nvlist representing this vdev's config.
220 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
221 vdev_config_flag_t flags)
225 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
227 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
228 vd->vdev_ops->vdev_op_type) == 0);
229 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
230 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
232 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
234 if (vd->vdev_path != NULL)
235 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
236 vd->vdev_path) == 0);
238 if (vd->vdev_devid != NULL)
239 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
240 vd->vdev_devid) == 0);
242 if (vd->vdev_physpath != NULL)
243 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
244 vd->vdev_physpath) == 0);
246 if (vd->vdev_fru != NULL)
247 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
250 if (vd->vdev_nparity != 0) {
251 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
252 VDEV_TYPE_RAIDZ) == 0);
255 * Make sure someone hasn't managed to sneak a fancy new vdev
256 * into a crufty old storage pool.
258 ASSERT(vd->vdev_nparity == 1 ||
259 (vd->vdev_nparity <= 2 &&
260 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
261 (vd->vdev_nparity <= 3 &&
262 spa_version(spa) >= SPA_VERSION_RAIDZ3));
265 * Note that we'll add the nparity tag even on storage pools
266 * that only support a single parity device -- older software
267 * will just ignore it.
269 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
270 vd->vdev_nparity) == 0);
273 if (vd->vdev_wholedisk != -1ULL)
274 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
275 vd->vdev_wholedisk) == 0);
277 if (vd->vdev_not_present)
278 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
280 if (vd->vdev_isspare)
281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
283 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
284 vd == vd->vdev_top) {
285 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
286 vd->vdev_ms_array) == 0);
287 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
288 vd->vdev_ms_shift) == 0);
289 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
290 vd->vdev_ashift) == 0);
291 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
292 vd->vdev_asize) == 0);
293 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
294 vd->vdev_islog) == 0);
295 if (vd->vdev_removing)
296 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
297 vd->vdev_removing) == 0);
300 if (vd->vdev_dtl_smo.smo_object != 0)
301 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
302 vd->vdev_dtl_smo.smo_object) == 0);
305 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
306 vd->vdev_crtxg) == 0);
312 vdev_get_stats(vd, &vs);
313 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
314 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
316 /* provide either current or previous scan information */
317 if (spa_scan_get_stats(spa, &ps) == 0) {
318 VERIFY(nvlist_add_uint64_array(nv,
319 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
320 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
325 if (!vd->vdev_ops->vdev_op_leaf) {
329 ASSERT(!vd->vdev_ishole);
331 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
334 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
335 vdev_t *cvd = vd->vdev_child[c];
338 * If we're generating an nvlist of removing
339 * vdevs then skip over any device which is
342 if ((flags & VDEV_CONFIG_REMOVING) &&
346 child[idx++] = vdev_config_generate(spa, cvd,
351 VERIFY(nvlist_add_nvlist_array(nv,
352 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
355 for (c = 0; c < idx; c++)
356 nvlist_free(child[c]);
358 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
361 const char *aux = NULL;
363 if (vd->vdev_offline && !vd->vdev_tmpoffline)
364 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
366 if (vd->vdev_resilvering)
367 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
369 if (vd->vdev_faulted)
370 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
372 if (vd->vdev_degraded)
373 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
375 if (vd->vdev_removed)
376 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
378 if (vd->vdev_unspare)
379 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
382 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
385 switch (vd->vdev_stat.vs_aux) {
386 case VDEV_AUX_ERR_EXCEEDED:
387 aux = "err_exceeded";
390 case VDEV_AUX_EXTERNAL:
396 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
399 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
400 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
401 vd->vdev_orig_guid) == 0);
409 * Generate a view of the top-level vdevs. If we currently have holes
410 * in the namespace, then generate an array which contains a list of holey
411 * vdevs. Additionally, add the number of top-level children that currently
415 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
417 vdev_t *rvd = spa->spa_root_vdev;
421 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
423 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
424 vdev_t *tvd = rvd->vdev_child[c];
426 if (tvd->vdev_ishole)
431 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
435 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
436 rvd->vdev_children) == 0);
438 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
442 * Returns the configuration from the label of the given vdev. For vdevs
443 * which don't have a txg value stored on their label (i.e. spares/cache)
444 * or have not been completely initialized (txg = 0) just return
445 * the configuration from the first valid label we find. Otherwise,
446 * find the most up-to-date label that does not exceed the specified
450 vdev_label_read_config(vdev_t *vd, uint64_t txg)
452 spa_t *spa = vd->vdev_spa;
453 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 = zio_buf_alloc(sizeof (vdev_phys_t));
469 for (int l = 0; l < VDEV_LABELS; l++) {
470 nvlist_t *label = NULL;
472 zio = zio_root(spa, NULL, NULL, flags);
474 vdev_label_read(zio, vd, l, vp,
475 offsetof(vdev_label_t, vl_vdev_phys),
476 sizeof (vdev_phys_t), NULL, NULL, flags);
478 if (zio_wait(zio) == 0 &&
479 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
481 uint64_t label_txg = 0;
484 * Auxiliary vdevs won't have txg values in their
485 * labels and newly added vdevs may not have been
486 * completely initialized so just return the
487 * configuration from the first valid label we
490 error = nvlist_lookup_uint64(label,
491 ZPOOL_CONFIG_POOL_TXG, &label_txg);
492 if ((error || label_txg == 0) && !config) {
495 } else if (label_txg <= txg && label_txg > best_txg) {
496 best_txg = label_txg;
498 config = fnvlist_dup(label);
508 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
509 flags |= ZIO_FLAG_TRYHARD;
513 zio_buf_free(vp, sizeof (vdev_phys_t));
519 * Determine if a device is in use. The 'spare_guid' parameter will be filled
520 * in with the device guid if this spare is active elsewhere on the system.
523 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
524 uint64_t *spare_guid, uint64_t *l2cache_guid)
526 spa_t *spa = vd->vdev_spa;
527 uint64_t state, pool_guid, device_guid, txg, spare_pool;
534 *l2cache_guid = 0ULL;
537 * Read the label, if any, and perform some basic sanity checks.
539 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
542 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
545 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
547 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
548 &device_guid) != 0) {
553 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
554 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
556 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
565 * Check to see if this device indeed belongs to the pool it claims to
566 * be a part of. The only way this is allowed is if the device is a hot
567 * spare (which we check for later on).
569 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
570 !spa_guid_exists(pool_guid, device_guid) &&
571 !spa_spare_exists(device_guid, NULL, NULL) &&
572 !spa_l2cache_exists(device_guid, NULL))
576 * If the transaction group is zero, then this an initialized (but
577 * unused) label. This is only an error if the create transaction
578 * on-disk is the same as the one we're using now, in which case the
579 * user has attempted to add the same vdev multiple times in the same
582 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
583 txg == 0 && vdtxg == crtxg)
587 * Check to see if this is a spare device. We do an explicit check for
588 * spa_has_spare() here because it may be on our pending list of spares
589 * to add. We also check if it is an l2cache device.
591 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
592 spa_has_spare(spa, device_guid)) {
594 *spare_guid = device_guid;
597 case VDEV_LABEL_CREATE:
598 case VDEV_LABEL_L2CACHE:
601 case VDEV_LABEL_REPLACE:
602 return (!spa_has_spare(spa, device_guid) ||
605 case VDEV_LABEL_SPARE:
606 return (spa_has_spare(spa, device_guid));
611 * Check to see if this is an l2cache device.
613 if (spa_l2cache_exists(device_guid, NULL))
617 * We can't rely on a pool's state if it's been imported
618 * read-only. Instead we look to see if the pools is marked
619 * read-only in the namespace and set the state to active.
621 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
622 spa_mode(spa) == FREAD)
623 state = POOL_STATE_ACTIVE;
626 * If the device is marked ACTIVE, then this device is in use by another
627 * pool on the system.
629 return (state == POOL_STATE_ACTIVE);
633 * Initialize a vdev label. We check to make sure each leaf device is not in
634 * use, and writable. We put down an initial label which we will later
635 * overwrite with a complete label. Note that it's important to do this
636 * sequentially, not in parallel, so that we catch cases of multiple use of the
637 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
641 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
643 spa_t *spa = vd->vdev_spa;
652 uint64_t spare_guid, l2cache_guid;
653 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
655 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
657 for (int c = 0; c < vd->vdev_children; c++)
658 if ((error = vdev_label_init(vd->vdev_child[c],
659 crtxg, reason)) != 0)
662 /* Track the creation time for this vdev */
663 vd->vdev_crtxg = crtxg;
665 if (!vd->vdev_ops->vdev_op_leaf)
669 * Dead vdevs cannot be initialized.
671 if (vdev_is_dead(vd))
672 return (SET_ERROR(EIO));
675 * Determine if the vdev is in use.
677 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
678 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
679 return (SET_ERROR(EBUSY));
682 * If this is a request to add or replace a spare or l2cache device
683 * that is in use elsewhere on the system, then we must update the
684 * guid (which was initialized to a random value) to reflect the
685 * actual GUID (which is shared between multiple pools).
687 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
688 spare_guid != 0ULL) {
689 uint64_t guid_delta = spare_guid - vd->vdev_guid;
691 vd->vdev_guid += guid_delta;
693 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
694 pvd->vdev_guid_sum += guid_delta;
697 * If this is a replacement, then we want to fallthrough to the
698 * rest of the code. If we're adding a spare, then it's already
699 * labeled appropriately and we can just return.
701 if (reason == VDEV_LABEL_SPARE)
703 ASSERT(reason == VDEV_LABEL_REPLACE ||
704 reason == VDEV_LABEL_SPLIT);
707 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
708 l2cache_guid != 0ULL) {
709 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
711 vd->vdev_guid += guid_delta;
713 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
714 pvd->vdev_guid_sum += guid_delta;
717 * If this is a replacement, then we want to fallthrough to the
718 * rest of the code. If we're adding an l2cache, then it's
719 * already labeled appropriately and we can just return.
721 if (reason == VDEV_LABEL_L2CACHE)
723 ASSERT(reason == VDEV_LABEL_REPLACE);
727 * TRIM the whole thing so that we start with a clean slate.
728 * It's just an optimization, so we don't care if it fails.
729 * Don't TRIM if removing so that we don't interfere with zpool
732 if (zfs_trim_enabled && vdev_trim_on_init && (reason == VDEV_LABEL_CREATE ||
733 reason == VDEV_LABEL_SPARE || reason == VDEV_LABEL_L2CACHE))
734 zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize));
737 * Initialize its label.
739 vp = zio_buf_alloc(sizeof (vdev_phys_t));
740 bzero(vp, sizeof (vdev_phys_t));
743 * Generate a label describing the pool and our top-level vdev.
744 * We mark it as being from txg 0 to indicate that it's not
745 * really part of an active pool just yet. The labels will
746 * be written again with a meaningful txg by spa_sync().
748 if (reason == VDEV_LABEL_SPARE ||
749 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
751 * For inactive hot spares, we generate a special label that
752 * identifies as a mutually shared hot spare. We write the
753 * label if we are adding a hot spare, or if we are removing an
754 * active hot spare (in which case we want to revert the
757 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
759 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
760 spa_version(spa)) == 0);
761 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
762 POOL_STATE_SPARE) == 0);
763 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
764 vd->vdev_guid) == 0);
765 } else if (reason == VDEV_LABEL_L2CACHE ||
766 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
768 * For level 2 ARC devices, add a special label.
770 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
772 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
773 spa_version(spa)) == 0);
774 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
775 POOL_STATE_L2CACHE) == 0);
776 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
777 vd->vdev_guid) == 0);
781 if (reason == VDEV_LABEL_SPLIT)
782 txg = spa->spa_uberblock.ub_txg;
783 label = spa_config_generate(spa, vd, txg, B_FALSE);
786 * Add our creation time. This allows us to detect multiple
787 * vdev uses as described above, and automatically expires if we
790 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
795 buflen = sizeof (vp->vp_nvlist);
797 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
800 zio_buf_free(vp, sizeof (vdev_phys_t));
801 /* EFAULT means nvlist_pack ran out of room */
802 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
806 * Initialize uberblock template.
808 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
809 bzero(ub, VDEV_UBERBLOCK_RING);
810 *ub = spa->spa_uberblock;
813 /* Initialize the 2nd padding area. */
814 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
815 bzero(pad2, VDEV_PAD_SIZE);
818 * Write everything in parallel.
821 zio = zio_root(spa, NULL, NULL, flags);
823 for (int l = 0; l < VDEV_LABELS; l++) {
825 vdev_label_write(zio, vd, l, vp,
826 offsetof(vdev_label_t, vl_vdev_phys),
827 sizeof (vdev_phys_t), NULL, NULL, flags);
830 * Skip the 1st padding area.
831 * Zero out the 2nd padding area where it might have
832 * left over data from previous filesystem format.
834 vdev_label_write(zio, vd, l, pad2,
835 offsetof(vdev_label_t, vl_pad2),
836 VDEV_PAD_SIZE, NULL, NULL, flags);
838 vdev_label_write(zio, vd, l, ub,
839 offsetof(vdev_label_t, vl_uberblock),
840 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
843 error = zio_wait(zio);
845 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
846 flags |= ZIO_FLAG_TRYHARD;
851 zio_buf_free(pad2, VDEV_PAD_SIZE);
852 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
853 zio_buf_free(vp, sizeof (vdev_phys_t));
856 * If this vdev hasn't been previously identified as a spare, then we
857 * mark it as such only if a) we are labeling it as a spare, or b) it
858 * exists as a spare elsewhere in the system. Do the same for
859 * level 2 ARC devices.
861 if (error == 0 && !vd->vdev_isspare &&
862 (reason == VDEV_LABEL_SPARE ||
863 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
866 if (error == 0 && !vd->vdev_isl2cache &&
867 (reason == VDEV_LABEL_L2CACHE ||
868 spa_l2cache_exists(vd->vdev_guid, NULL)))
875 * ==========================================================================
876 * uberblock load/sync
877 * ==========================================================================
881 * Consider the following situation: txg is safely synced to disk. We've
882 * written the first uberblock for txg + 1, and then we lose power. When we
883 * come back up, we fail to see the uberblock for txg + 1 because, say,
884 * it was on a mirrored device and the replica to which we wrote txg + 1
885 * is now offline. If we then make some changes and sync txg + 1, and then
886 * the missing replica comes back, then for a few seconds we'll have two
887 * conflicting uberblocks on disk with the same txg. The solution is simple:
888 * among uberblocks with equal txg, choose the one with the latest timestamp.
891 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
893 if (ub1->ub_txg < ub2->ub_txg)
895 if (ub1->ub_txg > ub2->ub_txg)
898 if (ub1->ub_timestamp < ub2->ub_timestamp)
900 if (ub1->ub_timestamp > ub2->ub_timestamp)
907 uberblock_t *ubl_ubbest; /* Best uberblock */
908 vdev_t *ubl_vd; /* vdev associated with the above */
912 vdev_uberblock_load_done(zio_t *zio)
914 vdev_t *vd = zio->io_vd;
915 spa_t *spa = zio->io_spa;
916 zio_t *rio = zio->io_private;
917 uberblock_t *ub = zio->io_data;
918 struct ubl_cbdata *cbp = rio->io_private;
920 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
922 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
923 mutex_enter(&rio->io_lock);
924 if (ub->ub_txg <= spa->spa_load_max_txg &&
925 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
927 * Keep track of the vdev in which this uberblock
928 * was found. We will use this information later
929 * to obtain the config nvlist associated with
932 *cbp->ubl_ubbest = *ub;
935 mutex_exit(&rio->io_lock);
938 zio_buf_free(zio->io_data, zio->io_size);
942 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
943 struct ubl_cbdata *cbp)
945 for (int c = 0; c < vd->vdev_children; c++)
946 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
948 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
949 for (int l = 0; l < VDEV_LABELS; l++) {
950 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
951 vdev_label_read(zio, vd, l,
952 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
953 VDEV_UBERBLOCK_OFFSET(vd, n),
954 VDEV_UBERBLOCK_SIZE(vd),
955 vdev_uberblock_load_done, zio, flags);
962 * Reads the 'best' uberblock from disk along with its associated
963 * configuration. First, we read the uberblock array of each label of each
964 * vdev, keeping track of the uberblock with the highest txg in each array.
965 * Then, we read the configuration from the same vdev as the best uberblock.
968 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
971 spa_t *spa = rvd->vdev_spa;
972 struct ubl_cbdata cb;
973 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
974 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
979 bzero(ub, sizeof (uberblock_t));
985 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
986 zio = zio_root(spa, NULL, &cb, flags);
987 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
988 (void) zio_wait(zio);
991 * It's possible that the best uberblock was discovered on a label
992 * that has a configuration which was written in a future txg.
993 * Search all labels on this vdev to find the configuration that
994 * matches the txg for our uberblock.
996 if (cb.ubl_vd != NULL)
997 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
998 spa_config_exit(spa, SCL_ALL, FTAG);
1002 * On success, increment root zio's count of good writes.
1003 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1006 vdev_uberblock_sync_done(zio_t *zio)
1008 uint64_t *good_writes = zio->io_private;
1010 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1011 atomic_add_64(good_writes, 1);
1015 * Write the uberblock to all labels of all leaves of the specified vdev.
1018 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1023 for (int c = 0; c < vd->vdev_children; c++)
1024 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1026 if (!vd->vdev_ops->vdev_op_leaf)
1029 if (!vdev_writeable(vd))
1032 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1034 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1035 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1038 for (int l = 0; l < VDEV_LABELS; l++)
1039 vdev_label_write(zio, vd, l, ubbuf,
1040 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1041 vdev_uberblock_sync_done, zio->io_private,
1042 flags | ZIO_FLAG_DONT_PROPAGATE);
1044 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1048 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1050 spa_t *spa = svd[0]->vdev_spa;
1052 uint64_t good_writes = 0;
1054 zio = zio_root(spa, NULL, &good_writes, flags);
1056 for (int v = 0; v < svdcount; v++)
1057 vdev_uberblock_sync(zio, ub, svd[v], flags);
1059 (void) zio_wait(zio);
1062 * Flush the uberblocks to disk. This ensures that the odd labels
1063 * are no longer needed (because the new uberblocks and the even
1064 * labels are safely on disk), so it is safe to overwrite them.
1066 zio = zio_root(spa, NULL, NULL, flags);
1068 for (int v = 0; v < svdcount; v++)
1069 zio_flush(zio, svd[v]);
1071 (void) zio_wait(zio);
1073 return (good_writes >= 1 ? 0 : EIO);
1077 * On success, increment the count of good writes for our top-level vdev.
1080 vdev_label_sync_done(zio_t *zio)
1082 uint64_t *good_writes = zio->io_private;
1084 if (zio->io_error == 0)
1085 atomic_add_64(good_writes, 1);
1089 * If there weren't enough good writes, indicate failure to the parent.
1092 vdev_label_sync_top_done(zio_t *zio)
1094 uint64_t *good_writes = zio->io_private;
1096 if (*good_writes == 0)
1097 zio->io_error = SET_ERROR(EIO);
1099 kmem_free(good_writes, sizeof (uint64_t));
1103 * We ignore errors for log and cache devices, simply free the private data.
1106 vdev_label_sync_ignore_done(zio_t *zio)
1108 kmem_free(zio->io_private, sizeof (uint64_t));
1112 * Write all even or odd labels to all leaves of the specified vdev.
1115 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1122 for (int c = 0; c < vd->vdev_children; c++)
1123 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1125 if (!vd->vdev_ops->vdev_op_leaf)
1128 if (!vdev_writeable(vd))
1132 * Generate a label describing the top-level config to which we belong.
1134 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1136 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1137 bzero(vp, sizeof (vdev_phys_t));
1139 buf = vp->vp_nvlist;
1140 buflen = sizeof (vp->vp_nvlist);
1142 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1143 for (; l < VDEV_LABELS; l += 2) {
1144 vdev_label_write(zio, vd, l, vp,
1145 offsetof(vdev_label_t, vl_vdev_phys),
1146 sizeof (vdev_phys_t),
1147 vdev_label_sync_done, zio->io_private,
1148 flags | ZIO_FLAG_DONT_PROPAGATE);
1152 zio_buf_free(vp, sizeof (vdev_phys_t));
1157 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1159 list_t *dl = &spa->spa_config_dirty_list;
1165 * Write the new labels to disk.
1167 zio = zio_root(spa, NULL, NULL, flags);
1169 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1170 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1173 ASSERT(!vd->vdev_ishole);
1175 zio_t *vio = zio_null(zio, spa, NULL,
1176 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1177 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1178 good_writes, flags);
1179 vdev_label_sync(vio, vd, l, txg, flags);
1183 error = zio_wait(zio);
1186 * Flush the new labels to disk.
1188 zio = zio_root(spa, NULL, NULL, flags);
1190 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1193 (void) zio_wait(zio);
1199 * Sync the uberblock and any changes to the vdev configuration.
1201 * The order of operations is carefully crafted to ensure that
1202 * if the system panics or loses power at any time, the state on disk
1203 * is still transactionally consistent. The in-line comments below
1204 * describe the failure semantics at each stage.
1206 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1207 * at any time, you can just call it again, and it will resume its work.
1210 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1212 spa_t *spa = svd[0]->vdev_spa;
1213 uberblock_t *ub = &spa->spa_uberblock;
1217 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1220 * Normally, we don't want to try too hard to write every label and
1221 * uberblock. If there is a flaky disk, we don't want the rest of the
1222 * sync process to block while we retry. But if we can't write a
1223 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1224 * bailing out and declaring the pool faulted.
1227 flags |= ZIO_FLAG_TRYHARD;
1229 ASSERT(ub->ub_txg <= txg);
1232 * If this isn't a resync due to I/O errors,
1233 * and nothing changed in this transaction group,
1234 * and the vdev configuration hasn't changed,
1235 * then there's nothing to do.
1237 if (ub->ub_txg < txg &&
1238 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1239 list_is_empty(&spa->spa_config_dirty_list))
1242 if (txg > spa_freeze_txg(spa))
1245 ASSERT(txg <= spa->spa_final_txg);
1248 * Flush the write cache of every disk that's been written to
1249 * in this transaction group. This ensures that all blocks
1250 * written in this txg will be committed to stable storage
1251 * before any uberblock that references them.
1253 zio = zio_root(spa, NULL, NULL, flags);
1255 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1256 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1259 (void) zio_wait(zio);
1262 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1263 * system dies in the middle of this process, that's OK: all of the
1264 * even labels that made it to disk will be newer than any uberblock,
1265 * and will therefore be considered invalid. The odd labels (L1, L3),
1266 * which have not yet been touched, will still be valid. We flush
1267 * the new labels to disk to ensure that all even-label updates
1268 * are committed to stable storage before the uberblock update.
1270 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1274 * Sync the uberblocks to all vdevs in svd[].
1275 * If the system dies in the middle of this step, there are two cases
1276 * to consider, and the on-disk state is consistent either way:
1278 * (1) If none of the new uberblocks made it to disk, then the
1279 * previous uberblock will be the newest, and the odd labels
1280 * (which had not yet been touched) will be valid with respect
1281 * to that uberblock.
1283 * (2) If one or more new uberblocks made it to disk, then they
1284 * will be the newest, and the even labels (which had all
1285 * been successfully committed) will be valid with respect
1286 * to the new uberblocks.
1288 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1292 * Sync out odd labels for every dirty vdev. If the system dies
1293 * in the middle of this process, the even labels and the new
1294 * uberblocks will suffice to open the pool. The next time
1295 * the pool is opened, the first thing we'll do -- before any
1296 * user data is modified -- is mark every vdev dirty so that
1297 * all labels will be brought up to date. We flush the new labels
1298 * to disk to ensure that all odd-label updates are committed to
1299 * stable storage before the next transaction group begins.
1301 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1304 trim_thread_wakeup(spa);