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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
26 * Virtual Device Labels
27 * ---------------------
29 * The vdev label serves several distinct purposes:
31 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
32 * identity within the pool.
34 * 2. Verify that all the devices given in a configuration are present
37 * 3. Determine the uberblock for the pool.
39 * 4. In case of an import operation, determine the configuration of the
40 * toplevel vdev of which it is a part.
42 * 5. If an import operation cannot find all the devices in the pool,
43 * provide enough information to the administrator to determine which
44 * devices are missing.
46 * It is important to note that while the kernel is responsible for writing the
47 * label, it only consumes the information in the first three cases. The
48 * latter information is only consumed in userland when determining the
49 * configuration to import a pool.
55 * Before describing the contents of the label, it's important to understand how
56 * the labels are written and updated with respect to the uberblock.
58 * When the pool configuration is altered, either because it was newly created
59 * or a device was added, we want to update all the labels such that we can deal
60 * with fatal failure at any point. To this end, each disk has two labels which
61 * are updated before and after the uberblock is synced. Assuming we have
62 * labels and an uberblock with the following transaction groups:
65 * +------+ +------+ +------+
67 * | t10 | | t10 | | t10 |
69 * +------+ +------+ +------+
71 * In this stable state, the labels and the uberblock were all updated within
72 * the same transaction group (10). Each label is mirrored and checksummed, so
73 * that we can detect when we fail partway through writing the label.
75 * In order to identify which labels are valid, the labels are written in the
78 * 1. For each vdev, update 'L1' to the new label
79 * 2. Update the uberblock
80 * 3. For each vdev, update 'L2' to the new label
82 * Given arbitrary failure, we can determine the correct label to use based on
83 * the transaction group. If we fail after updating L1 but before updating the
84 * UB, we will notice that L1's transaction group is greater than the uberblock,
85 * so L2 must be valid. If we fail after writing the uberblock but before
86 * writing L2, we will notice that L2's transaction group is less than L1, and
87 * therefore L1 is valid.
89 * Another added complexity is that not every label is updated when the config
90 * is synced. If we add a single device, we do not want to have to re-write
91 * every label for every device in the pool. This means that both L1 and L2 may
92 * be older than the pool uberblock, because the necessary information is stored
99 * The vdev label consists of two distinct parts, and is wrapped within the
100 * vdev_label_t structure. The label includes 8k of padding to permit legacy
101 * VTOC disk labels, but is otherwise ignored.
103 * The first half of the label is a packed nvlist which contains pool wide
104 * properties, per-vdev properties, and configuration information. It is
105 * described in more detail below.
107 * The latter half of the label consists of a redundant array of uberblocks.
108 * These uberblocks are updated whenever a transaction group is committed,
109 * or when the configuration is updated. When a pool is loaded, we scan each
110 * vdev for the 'best' uberblock.
113 * Configuration Information
114 * -------------------------
116 * The nvlist describing the pool and vdev contains the following elements:
118 * version ZFS on-disk version
121 * txg Transaction group in which this label was written
122 * pool_guid Unique identifier for this pool
123 * vdev_tree An nvlist describing vdev tree.
125 * Each leaf device label also contains the following:
127 * top_guid Unique ID for top-level vdev in which this is contained
128 * guid Unique ID for the leaf vdev
130 * The 'vs' configuration follows the format described in 'spa_config.c'.
133 #include <sys/zfs_context.h>
135 #include <sys/spa_impl.h>
138 #include <sys/vdev.h>
139 #include <sys/vdev_impl.h>
140 #include <sys/uberblock_impl.h>
141 #include <sys/metaslab.h>
143 #include <sys/dsl_scan.h>
144 #include <sys/fs/zfs.h>
147 * Basic routines to read and write from a vdev label.
148 * Used throughout the rest of this file.
151 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
153 ASSERT(offset < sizeof (vdev_label_t));
154 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
156 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
157 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
161 * Returns back the vdev label associated with the passed in offset.
164 vdev_label_number(uint64_t psize, uint64_t offset)
168 if (offset >= psize - VDEV_LABEL_END_SIZE) {
169 offset -= psize - VDEV_LABEL_END_SIZE;
170 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
172 l = offset / sizeof (vdev_label_t);
173 return (l < VDEV_LABELS ? l : -1);
177 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
178 uint64_t size, zio_done_func_t *done, void *private, int flags)
180 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
182 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
184 zio_nowait(zio_read_phys(zio, vd,
185 vdev_label_offset(vd->vdev_psize, l, offset),
186 size, buf, ZIO_CHECKSUM_LABEL, done, private,
187 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
191 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
192 uint64_t size, zio_done_func_t *done, void *private, int flags)
194 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
195 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
196 (SCL_CONFIG | SCL_STATE) &&
197 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
198 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
200 zio_nowait(zio_write_phys(zio, vd,
201 vdev_label_offset(vd->vdev_psize, l, offset),
202 size, buf, ZIO_CHECKSUM_LABEL, done, private,
203 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
207 * Generate the nvlist representing this vdev's config.
210 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
211 vdev_config_flag_t flags)
215 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
217 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
218 vd->vdev_ops->vdev_op_type) == 0);
219 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
220 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
222 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
224 if (vd->vdev_path != NULL)
225 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
226 vd->vdev_path) == 0);
228 if (vd->vdev_devid != NULL)
229 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
230 vd->vdev_devid) == 0);
232 if (vd->vdev_physpath != NULL)
233 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 vd->vdev_physpath) == 0);
236 if (vd->vdev_fru != NULL)
237 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
240 if (vd->vdev_nparity != 0) {
241 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
242 VDEV_TYPE_RAIDZ) == 0);
245 * Make sure someone hasn't managed to sneak a fancy new vdev
246 * into a crufty old storage pool.
248 ASSERT(vd->vdev_nparity == 1 ||
249 (vd->vdev_nparity <= 2 &&
250 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
251 (vd->vdev_nparity <= 3 &&
252 spa_version(spa) >= SPA_VERSION_RAIDZ3));
255 * Note that we'll add the nparity tag even on storage pools
256 * that only support a single parity device -- older software
257 * will just ignore it.
259 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
260 vd->vdev_nparity) == 0);
263 if (vd->vdev_wholedisk != -1ULL)
264 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
265 vd->vdev_wholedisk) == 0);
267 if (vd->vdev_not_present)
268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
270 if (vd->vdev_isspare)
271 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
273 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
274 vd == vd->vdev_top) {
275 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
276 vd->vdev_ms_array) == 0);
277 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
278 vd->vdev_ms_shift) == 0);
279 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
280 vd->vdev_ashift) == 0);
281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
282 vd->vdev_asize) == 0);
283 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
284 vd->vdev_islog) == 0);
285 if (vd->vdev_removing)
286 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
287 vd->vdev_removing) == 0);
290 if (vd->vdev_dtl_smo.smo_object != 0)
291 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
292 vd->vdev_dtl_smo.smo_object) == 0);
295 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
296 vd->vdev_crtxg) == 0);
302 vdev_get_stats(vd, &vs);
303 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
304 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
306 /* provide either current or previous scan information */
307 if (spa_scan_get_stats(spa, &ps) == 0) {
308 VERIFY(nvlist_add_uint64_array(nv,
309 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
310 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
315 if (!vd->vdev_ops->vdev_op_leaf) {
319 ASSERT(!vd->vdev_ishole);
321 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
324 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
325 vdev_t *cvd = vd->vdev_child[c];
328 * If we're generating an nvlist of removing
329 * vdevs then skip over any device which is
332 if ((flags & VDEV_CONFIG_REMOVING) &&
336 child[idx++] = vdev_config_generate(spa, cvd,
341 VERIFY(nvlist_add_nvlist_array(nv,
342 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
345 for (c = 0; c < idx; c++)
346 nvlist_free(child[c]);
348 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
351 const char *aux = NULL;
353 if (vd->vdev_offline && !vd->vdev_tmpoffline)
354 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
356 if (vd->vdev_resilvering)
357 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
359 if (vd->vdev_faulted)
360 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
362 if (vd->vdev_degraded)
363 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
365 if (vd->vdev_removed)
366 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
368 if (vd->vdev_unspare)
369 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
372 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
375 switch (vd->vdev_stat.vs_aux) {
376 case VDEV_AUX_ERR_EXCEEDED:
377 aux = "err_exceeded";
380 case VDEV_AUX_EXTERNAL:
386 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
389 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
390 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
391 vd->vdev_orig_guid) == 0);
399 * Generate a view of the top-level vdevs. If we currently have holes
400 * in the namespace, then generate an array which contains a list of holey
401 * vdevs. Additionally, add the number of top-level children that currently
405 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
407 vdev_t *rvd = spa->spa_root_vdev;
411 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
413 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
414 vdev_t *tvd = rvd->vdev_child[c];
416 if (tvd->vdev_ishole)
421 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
425 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
426 rvd->vdev_children) == 0);
428 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
432 vdev_label_read_config(vdev_t *vd)
434 spa_t *spa = vd->vdev_spa;
435 nvlist_t *config = NULL;
438 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
439 ZIO_FLAG_SPECULATIVE;
442 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
444 if (!vdev_readable(vd))
447 vp = zio_buf_alloc(sizeof (vdev_phys_t));
450 for (l = 0; l < VDEV_LABELS; l++) {
452 zio = zio_root(spa, NULL, NULL, flags);
454 vdev_label_read(zio, vd, l, vp,
455 offsetof(vdev_label_t, vl_vdev_phys),
456 sizeof (vdev_phys_t), NULL, NULL, flags);
458 if (zio_wait(zio) == 0 &&
459 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
463 if (config != NULL) {
469 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
470 flags |= ZIO_FLAG_TRYHARD;
474 zio_buf_free(vp, sizeof (vdev_phys_t));
480 * Determine if a device is in use. The 'spare_guid' parameter will be filled
481 * in with the device guid if this spare is active elsewhere on the system.
484 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
485 uint64_t *spare_guid, uint64_t *l2cache_guid)
487 spa_t *spa = vd->vdev_spa;
488 uint64_t state, pool_guid, device_guid, txg, spare_pool;
495 *l2cache_guid = 0ULL;
498 * Read the label, if any, and perform some basic sanity checks.
500 if ((label = vdev_label_read_config(vd)) == NULL)
503 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
506 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
508 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
509 &device_guid) != 0) {
514 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
515 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
517 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
526 * Check to see if this device indeed belongs to the pool it claims to
527 * be a part of. The only way this is allowed is if the device is a hot
528 * spare (which we check for later on).
530 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
531 !spa_guid_exists(pool_guid, device_guid) &&
532 !spa_spare_exists(device_guid, NULL, NULL) &&
533 !spa_l2cache_exists(device_guid, NULL))
537 * If the transaction group is zero, then this an initialized (but
538 * unused) label. This is only an error if the create transaction
539 * on-disk is the same as the one we're using now, in which case the
540 * user has attempted to add the same vdev multiple times in the same
543 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
544 txg == 0 && vdtxg == crtxg)
548 * Check to see if this is a spare device. We do an explicit check for
549 * spa_has_spare() here because it may be on our pending list of spares
550 * to add. We also check if it is an l2cache device.
552 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
553 spa_has_spare(spa, device_guid)) {
555 *spare_guid = device_guid;
558 case VDEV_LABEL_CREATE:
559 case VDEV_LABEL_L2CACHE:
562 case VDEV_LABEL_REPLACE:
563 return (!spa_has_spare(spa, device_guid) ||
566 case VDEV_LABEL_SPARE:
567 return (spa_has_spare(spa, device_guid));
574 * Check to see if this is an l2cache device.
576 if (spa_l2cache_exists(device_guid, NULL))
580 * We can't rely on a pool's state if it's been imported
581 * read-only. Instead we look to see if the pools is marked
582 * read-only in the namespace and set the state to active.
584 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
585 spa_mode(spa) == FREAD)
586 state = POOL_STATE_ACTIVE;
589 * If the device is marked ACTIVE, then this device is in use by another
590 * pool on the system.
592 return (state == POOL_STATE_ACTIVE);
596 * Initialize a vdev label. We check to make sure each leaf device is not in
597 * use, and writable. We put down an initial label which we will later
598 * overwrite with a complete label. Note that it's important to do this
599 * sequentially, not in parallel, so that we catch cases of multiple use of the
600 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
604 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
606 spa_t *spa = vd->vdev_spa;
615 uint64_t spare_guid, l2cache_guid;
616 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
620 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
622 for (c = 0; c < vd->vdev_children; c++)
623 if ((error = vdev_label_init(vd->vdev_child[c],
624 crtxg, reason)) != 0)
627 /* Track the creation time for this vdev */
628 vd->vdev_crtxg = crtxg;
630 if (!vd->vdev_ops->vdev_op_leaf)
634 * Dead vdevs cannot be initialized.
636 if (vdev_is_dead(vd))
640 * Determine if the vdev is in use.
642 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
643 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
647 * If this is a request to add or replace a spare or l2cache device
648 * that is in use elsewhere on the system, then we must update the
649 * guid (which was initialized to a random value) to reflect the
650 * actual GUID (which is shared between multiple pools).
652 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
653 spare_guid != 0ULL) {
654 uint64_t guid_delta = spare_guid - vd->vdev_guid;
656 vd->vdev_guid += guid_delta;
658 for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
659 pvd->vdev_guid_sum += guid_delta;
662 * If this is a replacement, then we want to fallthrough to the
663 * rest of the code. If we're adding a spare, then it's already
664 * labeled appropriately and we can just return.
666 if (reason == VDEV_LABEL_SPARE)
668 ASSERT(reason == VDEV_LABEL_REPLACE ||
669 reason == VDEV_LABEL_SPLIT);
672 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
673 l2cache_guid != 0ULL) {
674 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
676 vd->vdev_guid += guid_delta;
678 for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
679 pvd->vdev_guid_sum += guid_delta;
682 * If this is a replacement, then we want to fallthrough to the
683 * rest of the code. If we're adding an l2cache, then it's
684 * already labeled appropriately and we can just return.
686 if (reason == VDEV_LABEL_L2CACHE)
688 ASSERT(reason == VDEV_LABEL_REPLACE);
692 * Initialize its label.
694 vp = zio_buf_alloc(sizeof (vdev_phys_t));
695 bzero(vp, sizeof (vdev_phys_t));
698 * Generate a label describing the pool and our top-level vdev.
699 * We mark it as being from txg 0 to indicate that it's not
700 * really part of an active pool just yet. The labels will
701 * be written again with a meaningful txg by spa_sync().
703 if (reason == VDEV_LABEL_SPARE ||
704 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
706 * For inactive hot spares, we generate a special label that
707 * identifies as a mutually shared hot spare. We write the
708 * label if we are adding a hot spare, or if we are removing an
709 * active hot spare (in which case we want to revert the
712 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
714 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
715 spa_version(spa)) == 0);
716 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
717 POOL_STATE_SPARE) == 0);
718 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
719 vd->vdev_guid) == 0);
720 } else if (reason == VDEV_LABEL_L2CACHE ||
721 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
723 * For level 2 ARC devices, add a special label.
725 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
727 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
728 spa_version(spa)) == 0);
729 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
730 POOL_STATE_L2CACHE) == 0);
731 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
732 vd->vdev_guid) == 0);
736 if (reason == VDEV_LABEL_SPLIT)
737 txg = spa->spa_uberblock.ub_txg;
738 label = spa_config_generate(spa, vd, txg, B_FALSE);
741 * Add our creation time. This allows us to detect multiple
742 * vdev uses as described above, and automatically expires if we
745 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
750 buflen = sizeof (vp->vp_nvlist);
752 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
755 zio_buf_free(vp, sizeof (vdev_phys_t));
756 /* EFAULT means nvlist_pack ran out of room */
757 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
761 * Initialize uberblock template.
763 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
764 bzero(ub, VDEV_UBERBLOCK_RING);
765 *ub = spa->spa_uberblock;
768 /* Initialize the 2nd padding area. */
769 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
770 bzero(pad2, VDEV_PAD_SIZE);
773 * Write everything in parallel.
776 zio = zio_root(spa, NULL, NULL, flags);
778 for (l = 0; l < VDEV_LABELS; l++) {
780 vdev_label_write(zio, vd, l, vp,
781 offsetof(vdev_label_t, vl_vdev_phys),
782 sizeof (vdev_phys_t), NULL, NULL, flags);
785 * Skip the 1st padding area.
786 * Zero out the 2nd padding area where it might have
787 * left over data from previous filesystem format.
789 vdev_label_write(zio, vd, l, pad2,
790 offsetof(vdev_label_t, vl_pad2),
791 VDEV_PAD_SIZE, NULL, NULL, flags);
793 vdev_label_write(zio, vd, l, ub,
794 offsetof(vdev_label_t, vl_uberblock),
795 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
798 error = zio_wait(zio);
800 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
801 flags |= ZIO_FLAG_TRYHARD;
806 zio_buf_free(pad2, VDEV_PAD_SIZE);
807 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
808 zio_buf_free(vp, sizeof (vdev_phys_t));
811 * If this vdev hasn't been previously identified as a spare, then we
812 * mark it as such only if a) we are labeling it as a spare, or b) it
813 * exists as a spare elsewhere in the system. Do the same for
814 * level 2 ARC devices.
816 if (error == 0 && !vd->vdev_isspare &&
817 (reason == VDEV_LABEL_SPARE ||
818 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
821 if (error == 0 && !vd->vdev_isl2cache &&
822 (reason == VDEV_LABEL_L2CACHE ||
823 spa_l2cache_exists(vd->vdev_guid, NULL)))
830 * ==========================================================================
831 * uberblock load/sync
832 * ==========================================================================
836 * Consider the following situation: txg is safely synced to disk. We've
837 * written the first uberblock for txg + 1, and then we lose power. When we
838 * come back up, we fail to see the uberblock for txg + 1 because, say,
839 * it was on a mirrored device and the replica to which we wrote txg + 1
840 * is now offline. If we then make some changes and sync txg + 1, and then
841 * the missing replica comes back, then for a new seconds we'll have two
842 * conflicting uberblocks on disk with the same txg. The solution is simple:
843 * among uberblocks with equal txg, choose the one with the latest timestamp.
846 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
848 if (ub1->ub_txg < ub2->ub_txg)
850 if (ub1->ub_txg > ub2->ub_txg)
853 if (ub1->ub_timestamp < ub2->ub_timestamp)
855 if (ub1->ub_timestamp > ub2->ub_timestamp)
862 vdev_uberblock_load_done(zio_t *zio)
864 spa_t *spa = zio->io_spa;
865 zio_t *rio = zio->io_private;
866 uberblock_t *ub = zio->io_data;
867 uberblock_t *ubbest = rio->io_private;
869 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
871 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
872 mutex_enter(&rio->io_lock);
873 if (ub->ub_txg <= spa->spa_load_max_txg &&
874 vdev_uberblock_compare(ub, ubbest) > 0)
876 mutex_exit(&rio->io_lock);
879 zio_buf_free(zio->io_data, zio->io_size);
883 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
885 spa_t *spa = vd->vdev_spa;
886 vdev_t *rvd = spa->spa_root_vdev;
887 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
888 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
893 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
894 zio = zio_root(spa, NULL, ubbest, flags);
895 bzero(ubbest, sizeof (uberblock_t));
900 for (c = 0; c < vd->vdev_children; c++)
901 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
903 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
904 for (l = 0; l < VDEV_LABELS; l++) {
905 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
906 vdev_label_read(zio, vd, l,
907 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
908 VDEV_UBERBLOCK_OFFSET(vd, n),
909 VDEV_UBERBLOCK_SIZE(vd),
910 vdev_uberblock_load_done, zio, flags);
916 (void) zio_wait(zio);
917 spa_config_exit(spa, SCL_ALL, FTAG);
922 * On success, increment root zio's count of good writes.
923 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
926 vdev_uberblock_sync_done(zio_t *zio)
928 uint64_t *good_writes = zio->io_private;
930 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
931 atomic_add_64(good_writes, 1);
935 * Write the uberblock to all labels of all leaves of the specified vdev.
938 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
943 for (c = 0; c < vd->vdev_children; c++)
944 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
946 if (!vd->vdev_ops->vdev_op_leaf)
949 if (!vdev_writeable(vd))
952 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
954 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
955 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
958 for (l = 0; l < VDEV_LABELS; l++)
959 vdev_label_write(zio, vd, l, ubbuf,
960 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
961 vdev_uberblock_sync_done, zio->io_private,
962 flags | ZIO_FLAG_DONT_PROPAGATE);
964 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
968 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
970 spa_t *spa = svd[0]->vdev_spa;
972 uint64_t good_writes = 0;
975 zio = zio_root(spa, NULL, &good_writes, flags);
977 for (v = 0; v < svdcount; v++)
978 vdev_uberblock_sync(zio, ub, svd[v], flags);
980 (void) zio_wait(zio);
983 * Flush the uberblocks to disk. This ensures that the odd labels
984 * are no longer needed (because the new uberblocks and the even
985 * labels are safely on disk), so it is safe to overwrite them.
987 zio = zio_root(spa, NULL, NULL, flags);
989 for (v = 0; v < svdcount; v++)
990 zio_flush(zio, svd[v]);
992 (void) zio_wait(zio);
994 return (good_writes >= 1 ? 0 : EIO);
998 * On success, increment the count of good writes for our top-level vdev.
1001 vdev_label_sync_done(zio_t *zio)
1003 uint64_t *good_writes = zio->io_private;
1005 if (zio->io_error == 0)
1006 atomic_add_64(good_writes, 1);
1010 * If there weren't enough good writes, indicate failure to the parent.
1013 vdev_label_sync_top_done(zio_t *zio)
1015 uint64_t *good_writes = zio->io_private;
1017 if (*good_writes == 0)
1018 zio->io_error = EIO;
1020 kmem_free(good_writes, sizeof (uint64_t));
1024 * We ignore errors for log and cache devices, simply free the private data.
1027 vdev_label_sync_ignore_done(zio_t *zio)
1029 kmem_free(zio->io_private, sizeof (uint64_t));
1033 * Write all even or odd labels to all leaves of the specified vdev.
1036 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1044 for (c = 0; c < vd->vdev_children; c++)
1045 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1047 if (!vd->vdev_ops->vdev_op_leaf)
1050 if (!vdev_writeable(vd))
1054 * Generate a label describing the top-level config to which we belong.
1056 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1058 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1059 bzero(vp, sizeof (vdev_phys_t));
1061 buf = vp->vp_nvlist;
1062 buflen = sizeof (vp->vp_nvlist);
1064 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1065 for (; l < VDEV_LABELS; l += 2) {
1066 vdev_label_write(zio, vd, l, vp,
1067 offsetof(vdev_label_t, vl_vdev_phys),
1068 sizeof (vdev_phys_t),
1069 vdev_label_sync_done, zio->io_private,
1070 flags | ZIO_FLAG_DONT_PROPAGATE);
1074 zio_buf_free(vp, sizeof (vdev_phys_t));
1079 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1081 list_t *dl = &spa->spa_config_dirty_list;
1087 * Write the new labels to disk.
1089 zio = zio_root(spa, NULL, NULL, flags);
1091 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1092 uint64_t *good_writes;
1095 ASSERT(!vd->vdev_ishole);
1097 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1098 vio = zio_null(zio, spa, NULL,
1099 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1100 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1101 good_writes, flags);
1102 vdev_label_sync(vio, vd, l, txg, flags);
1106 error = zio_wait(zio);
1109 * Flush the new labels to disk.
1111 zio = zio_root(spa, NULL, NULL, flags);
1113 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1116 (void) zio_wait(zio);
1122 * Sync the uberblock and any changes to the vdev configuration.
1124 * The order of operations is carefully crafted to ensure that
1125 * if the system panics or loses power at any time, the state on disk
1126 * is still transactionally consistent. The in-line comments below
1127 * describe the failure semantics at each stage.
1129 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1130 * at any time, you can just call it again, and it will resume its work.
1133 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1135 spa_t *spa = svd[0]->vdev_spa;
1136 uberblock_t *ub = &spa->spa_uberblock;
1140 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1143 * Normally, we don't want to try too hard to write every label and
1144 * uberblock. If there is a flaky disk, we don't want the rest of the
1145 * sync process to block while we retry. But if we can't write a
1146 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1147 * bailing out and declaring the pool faulted.
1150 flags |= ZIO_FLAG_TRYHARD;
1152 ASSERT(ub->ub_txg <= txg);
1155 * If this isn't a resync due to I/O errors,
1156 * and nothing changed in this transaction group,
1157 * and the vdev configuration hasn't changed,
1158 * then there's nothing to do.
1160 if (ub->ub_txg < txg &&
1161 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1162 list_is_empty(&spa->spa_config_dirty_list))
1165 if (txg > spa_freeze_txg(spa))
1168 ASSERT(txg <= spa->spa_final_txg);
1171 * Flush the write cache of every disk that's been written to
1172 * in this transaction group. This ensures that all blocks
1173 * written in this txg will be committed to stable storage
1174 * before any uberblock that references them.
1176 zio = zio_root(spa, NULL, NULL, flags);
1178 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1179 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1182 (void) zio_wait(zio);
1185 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1186 * system dies in the middle of this process, that's OK: all of the
1187 * even labels that made it to disk will be newer than any uberblock,
1188 * and will therefore be considered invalid. The odd labels (L1, L3),
1189 * which have not yet been touched, will still be valid. We flush
1190 * the new labels to disk to ensure that all even-label updates
1191 * are committed to stable storage before the uberblock update.
1193 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1197 * Sync the uberblocks to all vdevs in svd[].
1198 * If the system dies in the middle of this step, there are two cases
1199 * to consider, and the on-disk state is consistent either way:
1201 * (1) If none of the new uberblocks made it to disk, then the
1202 * previous uberblock will be the newest, and the odd labels
1203 * (which had not yet been touched) will be valid with respect
1204 * to that uberblock.
1206 * (2) If one or more new uberblocks made it to disk, then they
1207 * will be the newest, and the even labels (which had all
1208 * been successfully committed) will be valid with respect
1209 * to the new uberblocks.
1211 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1215 * Sync out odd labels for every dirty vdev. If the system dies
1216 * in the middle of this process, the even labels and the new
1217 * uberblocks will suffice to open the pool. The next time
1218 * the pool is opened, the first thing we'll do -- before any
1219 * user data is modified -- is mark every vdev dirty so that
1220 * all labels will be brought up to date. We flush the new labels
1221 * to disk to ensure that all odd-label updates are committed to
1222 * stable storage before the next transaction group begins.
1224 return (vdev_label_sync_list(spa, 1, txg, flags));