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>
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 nv = fnvlist_alloc();
227 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
228 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
229 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
230 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
232 if (vd->vdev_path != NULL)
233 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
235 if (vd->vdev_devid != NULL)
236 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
238 if (vd->vdev_physpath != NULL)
239 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
242 if (vd->vdev_fru != NULL)
243 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
245 if (vd->vdev_nparity != 0) {
246 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
247 VDEV_TYPE_RAIDZ) == 0);
250 * Make sure someone hasn't managed to sneak a fancy new vdev
251 * into a crufty old storage pool.
253 ASSERT(vd->vdev_nparity == 1 ||
254 (vd->vdev_nparity <= 2 &&
255 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
256 (vd->vdev_nparity <= 3 &&
257 spa_version(spa) >= SPA_VERSION_RAIDZ3));
260 * Note that we'll add the nparity tag even on storage pools
261 * that only support a single parity device -- older software
262 * will just ignore it.
264 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
267 if (vd->vdev_wholedisk != -1ULL)
268 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
271 if (vd->vdev_not_present)
272 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
274 if (vd->vdev_isspare)
275 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
278 vd == vd->vdev_top) {
279 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
281 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
283 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
284 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
286 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
287 if (vd->vdev_removing)
288 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
292 if (vd->vdev_dtl_sm != NULL) {
293 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
294 space_map_object(vd->vdev_dtl_sm));
298 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
300 if (flags & VDEV_CONFIG_MOS) {
301 if (vd->vdev_leaf_zap != 0) {
302 ASSERT(vd->vdev_ops->vdev_op_leaf);
303 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
307 if (vd->vdev_top_zap != 0) {
308 ASSERT(vd == vd->vdev_top);
309 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
318 vdev_get_stats(vd, &vs);
319 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
320 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
322 /* provide either current or previous scan information */
323 if (spa_scan_get_stats(spa, &ps) == 0) {
324 fnvlist_add_uint64_array(nv,
325 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
326 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
330 if (!vd->vdev_ops->vdev_op_leaf) {
334 ASSERT(!vd->vdev_ishole);
336 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
339 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
340 vdev_t *cvd = vd->vdev_child[c];
343 * If we're generating an nvlist of removing
344 * vdevs then skip over any device which is
347 if ((flags & VDEV_CONFIG_REMOVING) &&
351 child[idx++] = vdev_config_generate(spa, cvd,
356 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
360 for (c = 0; c < idx; c++)
361 nvlist_free(child[c]);
363 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
366 const char *aux = NULL;
368 if (vd->vdev_offline && !vd->vdev_tmpoffline)
369 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
370 if (vd->vdev_resilver_txg != 0)
371 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
372 vd->vdev_resilver_txg);
373 if (vd->vdev_faulted)
374 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
375 if (vd->vdev_degraded)
376 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
377 if (vd->vdev_removed)
378 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
379 if (vd->vdev_unspare)
380 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
382 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
384 switch (vd->vdev_stat.vs_aux) {
385 case VDEV_AUX_ERR_EXCEEDED:
386 aux = "err_exceeded";
389 case VDEV_AUX_EXTERNAL:
395 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
397 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
398 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
407 * Generate a view of the top-level vdevs. If we currently have holes
408 * in the namespace, then generate an array which contains a list of holey
409 * vdevs. Additionally, add the number of top-level children that currently
413 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
415 vdev_t *rvd = spa->spa_root_vdev;
419 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
421 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
422 vdev_t *tvd = rvd->vdev_child[c];
424 if (tvd->vdev_ishole)
429 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
433 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
434 rvd->vdev_children) == 0);
436 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
440 * Returns the configuration from the label of the given vdev. For vdevs
441 * which don't have a txg value stored on their label (i.e. spares/cache)
442 * or have not been completely initialized (txg = 0) just return
443 * the configuration from the first valid label we find. Otherwise,
444 * find the most up-to-date label that does not exceed the specified
448 vdev_label_read_config(vdev_t *vd, uint64_t txg)
450 spa_t *spa = vd->vdev_spa;
451 nvlist_t *config = NULL;
454 uint64_t best_txg = 0;
456 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
457 ZIO_FLAG_SPECULATIVE;
459 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
461 if (!vdev_readable(vd))
464 vp = zio_buf_alloc(sizeof (vdev_phys_t));
467 for (int l = 0; l < VDEV_LABELS; l++) {
468 nvlist_t *label = NULL;
470 zio = zio_root(spa, NULL, NULL, flags);
472 vdev_label_read(zio, vd, l, vp,
473 offsetof(vdev_label_t, vl_vdev_phys),
474 sizeof (vdev_phys_t), NULL, NULL, flags);
476 if (zio_wait(zio) == 0 &&
477 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
479 uint64_t label_txg = 0;
482 * Auxiliary vdevs won't have txg values in their
483 * labels and newly added vdevs may not have been
484 * completely initialized so just return the
485 * configuration from the first valid label we
488 error = nvlist_lookup_uint64(label,
489 ZPOOL_CONFIG_POOL_TXG, &label_txg);
490 if ((error || label_txg == 0) && !config) {
493 } else if (label_txg <= txg && label_txg > best_txg) {
494 best_txg = label_txg;
496 config = fnvlist_dup(label);
506 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
507 flags |= ZIO_FLAG_TRYHARD;
511 zio_buf_free(vp, sizeof (vdev_phys_t));
517 * Determine if a device is in use. The 'spare_guid' parameter will be filled
518 * in with the device guid if this spare is active elsewhere on the system.
521 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
522 uint64_t *spare_guid, uint64_t *l2cache_guid)
524 spa_t *spa = vd->vdev_spa;
525 uint64_t state, pool_guid, device_guid, txg, spare_pool;
532 *l2cache_guid = 0ULL;
535 * Read the label, if any, and perform some basic sanity checks.
537 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
540 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
543 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
545 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
546 &device_guid) != 0) {
551 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
552 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
554 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
563 * Check to see if this device indeed belongs to the pool it claims to
564 * be a part of. The only way this is allowed is if the device is a hot
565 * spare (which we check for later on).
567 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
568 !spa_guid_exists(pool_guid, device_guid) &&
569 !spa_spare_exists(device_guid, NULL, NULL) &&
570 !spa_l2cache_exists(device_guid, NULL))
574 * If the transaction group is zero, then this an initialized (but
575 * unused) label. This is only an error if the create transaction
576 * on-disk is the same as the one we're using now, in which case the
577 * user has attempted to add the same vdev multiple times in the same
580 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
581 txg == 0 && vdtxg == crtxg)
585 * Check to see if this is a spare device. We do an explicit check for
586 * spa_has_spare() here because it may be on our pending list of spares
587 * to add. We also check if it is an l2cache device.
589 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
590 spa_has_spare(spa, device_guid)) {
592 *spare_guid = device_guid;
595 case VDEV_LABEL_CREATE:
596 case VDEV_LABEL_L2CACHE:
599 case VDEV_LABEL_REPLACE:
600 return (!spa_has_spare(spa, device_guid) ||
603 case VDEV_LABEL_SPARE:
604 return (spa_has_spare(spa, device_guid));
609 * Check to see if this is an l2cache device.
611 if (spa_l2cache_exists(device_guid, NULL))
615 * We can't rely on a pool's state if it's been imported
616 * read-only. Instead we look to see if the pools is marked
617 * read-only in the namespace and set the state to active.
619 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
620 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
621 spa_mode(spa) == FREAD)
622 state = POOL_STATE_ACTIVE;
625 * If the device is marked ACTIVE, then this device is in use by another
626 * pool on the system.
628 return (state == POOL_STATE_ACTIVE);
632 * Initialize a vdev label. We check to make sure each leaf device is not in
633 * use, and writable. We put down an initial label which we will later
634 * overwrite with a complete label. Note that it's important to do this
635 * sequentially, not in parallel, so that we catch cases of multiple use of the
636 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
640 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
642 spa_t *spa = vd->vdev_spa;
651 uint64_t spare_guid, l2cache_guid;
652 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
654 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
656 for (int c = 0; c < vd->vdev_children; c++)
657 if ((error = vdev_label_init(vd->vdev_child[c],
658 crtxg, reason)) != 0)
661 /* Track the creation time for this vdev */
662 vd->vdev_crtxg = crtxg;
664 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
668 * Dead vdevs cannot be initialized.
670 if (vdev_is_dead(vd))
671 return (SET_ERROR(EIO));
674 * Determine if the vdev is in use.
676 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
677 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
678 return (SET_ERROR(EBUSY));
681 * If this is a request to add or replace a spare or l2cache device
682 * that is in use elsewhere on the system, then we must update the
683 * guid (which was initialized to a random value) to reflect the
684 * actual GUID (which is shared between multiple pools).
686 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
687 spare_guid != 0ULL) {
688 uint64_t guid_delta = spare_guid - vd->vdev_guid;
690 vd->vdev_guid += guid_delta;
692 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
693 pvd->vdev_guid_sum += guid_delta;
696 * If this is a replacement, then we want to fallthrough to the
697 * rest of the code. If we're adding a spare, then it's already
698 * labeled appropriately and we can just return.
700 if (reason == VDEV_LABEL_SPARE)
702 ASSERT(reason == VDEV_LABEL_REPLACE ||
703 reason == VDEV_LABEL_SPLIT);
706 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
707 l2cache_guid != 0ULL) {
708 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
710 vd->vdev_guid += guid_delta;
712 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
713 pvd->vdev_guid_sum += guid_delta;
716 * If this is a replacement, then we want to fallthrough to the
717 * rest of the code. If we're adding an l2cache, then it's
718 * already labeled appropriately and we can just return.
720 if (reason == VDEV_LABEL_L2CACHE)
722 ASSERT(reason == VDEV_LABEL_REPLACE);
726 * TRIM the whole thing so that we start with a clean slate.
727 * It's just an optimization, so we don't care if it fails.
728 * Don't TRIM if removing so that we don't interfere with zpool
731 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim &&
732 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE ||
733 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 vdev_label_write_pad2(vdev_t *vd, const char *buf, size_t size)
877 spa_t *spa = vd->vdev_spa;
880 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
883 if (size > VDEV_PAD_SIZE)
886 if (!vd->vdev_ops->vdev_op_leaf)
888 if (vdev_is_dead(vd))
891 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
893 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
894 bzero(pad2, VDEV_PAD_SIZE);
895 memcpy(pad2, buf, size);
898 zio = zio_root(spa, NULL, NULL, flags);
899 vdev_label_write(zio, vd, 0, pad2,
900 offsetof(vdev_label_t, vl_pad2),
901 VDEV_PAD_SIZE, NULL, NULL, flags);
902 error = zio_wait(zio);
903 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
904 flags |= ZIO_FLAG_TRYHARD;
908 zio_buf_free(pad2, VDEV_PAD_SIZE);
913 * ==========================================================================
914 * uberblock load/sync
915 * ==========================================================================
919 * Consider the following situation: txg is safely synced to disk. We've
920 * written the first uberblock for txg + 1, and then we lose power. When we
921 * come back up, we fail to see the uberblock for txg + 1 because, say,
922 * it was on a mirrored device and the replica to which we wrote txg + 1
923 * is now offline. If we then make some changes and sync txg + 1, and then
924 * the missing replica comes back, then for a few seconds we'll have two
925 * conflicting uberblocks on disk with the same txg. The solution is simple:
926 * among uberblocks with equal txg, choose the one with the latest timestamp.
929 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
931 if (ub1->ub_txg < ub2->ub_txg)
933 if (ub1->ub_txg > ub2->ub_txg)
936 if (ub1->ub_timestamp < ub2->ub_timestamp)
938 if (ub1->ub_timestamp > ub2->ub_timestamp)
945 uberblock_t *ubl_ubbest; /* Best uberblock */
946 vdev_t *ubl_vd; /* vdev associated with the above */
950 vdev_uberblock_load_done(zio_t *zio)
952 vdev_t *vd = zio->io_vd;
953 spa_t *spa = zio->io_spa;
954 zio_t *rio = zio->io_private;
955 uberblock_t *ub = zio->io_data;
956 struct ubl_cbdata *cbp = rio->io_private;
958 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
960 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
961 mutex_enter(&rio->io_lock);
962 if (ub->ub_txg <= spa->spa_load_max_txg &&
963 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
965 * Keep track of the vdev in which this uberblock
966 * was found. We will use this information later
967 * to obtain the config nvlist associated with
970 *cbp->ubl_ubbest = *ub;
973 mutex_exit(&rio->io_lock);
976 zio_buf_free(zio->io_data, zio->io_size);
980 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
981 struct ubl_cbdata *cbp)
983 for (int c = 0; c < vd->vdev_children; c++)
984 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
986 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
987 for (int l = 0; l < VDEV_LABELS; l++) {
988 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
989 vdev_label_read(zio, vd, l,
990 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
991 VDEV_UBERBLOCK_OFFSET(vd, n),
992 VDEV_UBERBLOCK_SIZE(vd),
993 vdev_uberblock_load_done, zio, flags);
1000 * Reads the 'best' uberblock from disk along with its associated
1001 * configuration. First, we read the uberblock array of each label of each
1002 * vdev, keeping track of the uberblock with the highest txg in each array.
1003 * Then, we read the configuration from the same vdev as the best uberblock.
1006 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1009 spa_t *spa = rvd->vdev_spa;
1010 struct ubl_cbdata cb;
1011 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1012 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1017 bzero(ub, sizeof (uberblock_t));
1023 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1024 zio = zio_root(spa, NULL, &cb, flags);
1025 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1026 (void) zio_wait(zio);
1029 * It's possible that the best uberblock was discovered on a label
1030 * that has a configuration which was written in a future txg.
1031 * Search all labels on this vdev to find the configuration that
1032 * matches the txg for our uberblock.
1034 if (cb.ubl_vd != NULL)
1035 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1036 spa_config_exit(spa, SCL_ALL, FTAG);
1040 * On success, increment root zio's count of good writes.
1041 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1044 vdev_uberblock_sync_done(zio_t *zio)
1046 uint64_t *good_writes = zio->io_private;
1048 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1049 atomic_inc_64(good_writes);
1053 * Write the uberblock to all labels of all leaves of the specified vdev.
1056 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1061 for (int c = 0; c < vd->vdev_children; c++)
1062 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1064 if (!vd->vdev_ops->vdev_op_leaf)
1067 if (!vdev_writeable(vd))
1070 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1072 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1073 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1076 for (int l = 0; l < VDEV_LABELS; l++)
1077 vdev_label_write(zio, vd, l, ubbuf,
1078 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1079 vdev_uberblock_sync_done, zio->io_private,
1080 flags | ZIO_FLAG_DONT_PROPAGATE);
1082 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1085 /* Sync the uberblocks to all vdevs in svd[] */
1087 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1089 spa_t *spa = svd[0]->vdev_spa;
1091 uint64_t good_writes = 0;
1093 zio = zio_root(spa, NULL, &good_writes, flags);
1095 for (int v = 0; v < svdcount; v++)
1096 vdev_uberblock_sync(zio, ub, svd[v], flags);
1098 (void) zio_wait(zio);
1101 * Flush the uberblocks to disk. This ensures that the odd labels
1102 * are no longer needed (because the new uberblocks and the even
1103 * labels are safely on disk), so it is safe to overwrite them.
1105 zio = zio_root(spa, NULL, NULL, flags);
1107 for (int v = 0; v < svdcount; v++)
1108 zio_flush(zio, svd[v]);
1110 (void) zio_wait(zio);
1112 return (good_writes >= 1 ? 0 : EIO);
1116 * On success, increment the count of good writes for our top-level vdev.
1119 vdev_label_sync_done(zio_t *zio)
1121 uint64_t *good_writes = zio->io_private;
1123 if (zio->io_error == 0)
1124 atomic_inc_64(good_writes);
1128 * If there weren't enough good writes, indicate failure to the parent.
1131 vdev_label_sync_top_done(zio_t *zio)
1133 uint64_t *good_writes = zio->io_private;
1135 if (*good_writes == 0)
1136 zio->io_error = SET_ERROR(EIO);
1138 kmem_free(good_writes, sizeof (uint64_t));
1142 * We ignore errors for log and cache devices, simply free the private data.
1145 vdev_label_sync_ignore_done(zio_t *zio)
1147 kmem_free(zio->io_private, sizeof (uint64_t));
1151 * Write all even or odd labels to all leaves of the specified vdev.
1154 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1161 for (int c = 0; c < vd->vdev_children; c++)
1162 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1164 if (!vd->vdev_ops->vdev_op_leaf)
1167 if (!vdev_writeable(vd))
1171 * Generate a label describing the top-level config to which we belong.
1173 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1175 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1176 bzero(vp, sizeof (vdev_phys_t));
1178 buf = vp->vp_nvlist;
1179 buflen = sizeof (vp->vp_nvlist);
1181 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1182 for (; l < VDEV_LABELS; l += 2) {
1183 vdev_label_write(zio, vd, l, vp,
1184 offsetof(vdev_label_t, vl_vdev_phys),
1185 sizeof (vdev_phys_t),
1186 vdev_label_sync_done, zio->io_private,
1187 flags | ZIO_FLAG_DONT_PROPAGATE);
1191 zio_buf_free(vp, sizeof (vdev_phys_t));
1196 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1198 list_t *dl = &spa->spa_config_dirty_list;
1204 * Write the new labels to disk.
1206 zio = zio_root(spa, NULL, NULL, flags);
1208 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1209 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1212 ASSERT(!vd->vdev_ishole);
1214 zio_t *vio = zio_null(zio, spa, NULL,
1215 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1216 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1217 good_writes, flags);
1218 vdev_label_sync(vio, vd, l, txg, flags);
1222 error = zio_wait(zio);
1225 * Flush the new labels to disk.
1227 zio = zio_root(spa, NULL, NULL, flags);
1229 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1232 (void) zio_wait(zio);
1238 * Sync the uberblock and any changes to the vdev configuration.
1240 * The order of operations is carefully crafted to ensure that
1241 * if the system panics or loses power at any time, the state on disk
1242 * is still transactionally consistent. The in-line comments below
1243 * describe the failure semantics at each stage.
1245 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1246 * at any time, you can just call it again, and it will resume its work.
1249 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1251 spa_t *spa = svd[0]->vdev_spa;
1252 uberblock_t *ub = &spa->spa_uberblock;
1256 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1260 * Normally, we don't want to try too hard to write every label and
1261 * uberblock. If there is a flaky disk, we don't want the rest of the
1262 * sync process to block while we retry. But if we can't write a
1263 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1264 * bailing out and declaring the pool faulted.
1267 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1269 flags |= ZIO_FLAG_TRYHARD;
1272 ASSERT(ub->ub_txg <= txg);
1275 * If this isn't a resync due to I/O errors,
1276 * and nothing changed in this transaction group,
1277 * and the vdev configuration hasn't changed,
1278 * then there's nothing to do.
1280 if (ub->ub_txg < txg &&
1281 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1282 list_is_empty(&spa->spa_config_dirty_list))
1285 if (txg > spa_freeze_txg(spa))
1288 ASSERT(txg <= spa->spa_final_txg);
1291 * Flush the write cache of every disk that's been written to
1292 * in this transaction group. This ensures that all blocks
1293 * written in this txg will be committed to stable storage
1294 * before any uberblock that references them.
1296 zio = zio_root(spa, NULL, NULL, flags);
1298 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1299 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1302 (void) zio_wait(zio);
1305 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1306 * system dies in the middle of this process, that's OK: all of the
1307 * even labels that made it to disk will be newer than any uberblock,
1308 * and will therefore be considered invalid. The odd labels (L1, L3),
1309 * which have not yet been touched, will still be valid. We flush
1310 * the new labels to disk to ensure that all even-label updates
1311 * are committed to stable storage before the uberblock update.
1313 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1317 * Sync the uberblocks to all vdevs in svd[].
1318 * If the system dies in the middle of this step, there are two cases
1319 * to consider, and the on-disk state is consistent either way:
1321 * (1) If none of the new uberblocks made it to disk, then the
1322 * previous uberblock will be the newest, and the odd labels
1323 * (which had not yet been touched) will be valid with respect
1324 * to that uberblock.
1326 * (2) If one or more new uberblocks made it to disk, then they
1327 * will be the newest, and the even labels (which had all
1328 * been successfully committed) will be valid with respect
1329 * to the new uberblocks.
1331 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1335 * Sync out odd labels for every dirty vdev. If the system dies
1336 * in the middle of this process, the even labels and the new
1337 * uberblocks will suffice to open the pool. The next time
1338 * the pool is opened, the first thing we'll do -- before any
1339 * user data is modified -- is mark every vdev dirty so that
1340 * all labels will be brought up to date. We flush the new labels
1341 * to disk to ensure that all odd-label updates are committed to
1342 * stable storage before the next transaction group begins.
1344 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1347 trim_thread_wakeup(spa);