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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * Virtual Device Labels
28 * ---------------------
30 * The vdev label serves several distinct purposes:
32 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
33 * identity within the pool.
35 * 2. Verify that all the devices given in a configuration are present
38 * 3. Determine the uberblock for the pool.
40 * 4. In case of an import operation, determine the configuration of the
41 * toplevel vdev of which it is a part.
43 * 5. If an import operation cannot find all the devices in the pool,
44 * provide enough information to the administrator to determine which
45 * devices are missing.
47 * It is important to note that while the kernel is responsible for writing the
48 * label, it only consumes the information in the first three cases. The
49 * latter information is only consumed in userland when determining the
50 * configuration to import a pool.
56 * Before describing the contents of the label, it's important to understand how
57 * the labels are written and updated with respect to the uberblock.
59 * When the pool configuration is altered, either because it was newly created
60 * or a device was added, we want to update all the labels such that we can deal
61 * with fatal failure at any point. To this end, each disk has two labels which
62 * are updated before and after the uberblock is synced. Assuming we have
63 * labels and an uberblock with the following transaction groups:
66 * +------+ +------+ +------+
68 * | t10 | | t10 | | t10 |
70 * +------+ +------+ +------+
72 * In this stable state, the labels and the uberblock were all updated within
73 * the same transaction group (10). Each label is mirrored and checksummed, so
74 * that we can detect when we fail partway through writing the label.
76 * In order to identify which labels are valid, the labels are written in the
79 * 1. For each vdev, update 'L1' to the new label
80 * 2. Update the uberblock
81 * 3. For each vdev, update 'L2' to the new label
83 * Given arbitrary failure, we can determine the correct label to use based on
84 * the transaction group. If we fail after updating L1 but before updating the
85 * UB, we will notice that L1's transaction group is greater than the uberblock,
86 * so L2 must be valid. If we fail after writing the uberblock but before
87 * writing L2, we will notice that L2's transaction group is less than L1, and
88 * therefore L1 is valid.
90 * Another added complexity is that not every label is updated when the config
91 * is synced. If we add a single device, we do not want to have to re-write
92 * every label for every device in the pool. This means that both L1 and L2 may
93 * be older than the pool uberblock, because the necessary information is stored
100 * The vdev label consists of two distinct parts, and is wrapped within the
101 * vdev_label_t structure. The label includes 8k of padding to permit legacy
102 * VTOC disk labels, but is otherwise ignored.
104 * The first half of the label is a packed nvlist which contains pool wide
105 * properties, per-vdev properties, and configuration information. It is
106 * described in more detail below.
108 * The latter half of the label consists of a redundant array of uberblocks.
109 * These uberblocks are updated whenever a transaction group is committed,
110 * or when the configuration is updated. When a pool is loaded, we scan each
111 * vdev for the 'best' uberblock.
114 * Configuration Information
115 * -------------------------
117 * The nvlist describing the pool and vdev contains the following elements:
119 * version ZFS on-disk version
122 * txg Transaction group in which this label was written
123 * pool_guid Unique identifier for this pool
124 * vdev_tree An nvlist describing vdev tree.
126 * Each leaf device label also contains the following:
128 * top_guid Unique ID for top-level vdev in which this is contained
129 * guid Unique ID for the leaf vdev
131 * The 'vs' configuration follows the format described in 'spa_config.c'.
134 #include <sys/zfs_context.h>
136 #include <sys/spa_impl.h>
139 #include <sys/vdev.h>
140 #include <sys/vdev_impl.h>
141 #include <sys/uberblock_impl.h>
142 #include <sys/metaslab.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 boolean_t isspare, boolean_t isl2cache)
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 (!isspare && !isl2cache)
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_RAID6));
253 * Note that we'll add the nparity tag even on storage pools
254 * that only support a single parity device -- older software
255 * will just ignore it.
257 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
258 vd->vdev_nparity) == 0);
261 if (vd->vdev_wholedisk != -1ULL)
262 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
263 vd->vdev_wholedisk) == 0);
265 if (vd->vdev_not_present)
266 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
268 if (vd->vdev_isspare)
269 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
271 if (!isspare && !isl2cache && vd == vd->vdev_top) {
272 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
273 vd->vdev_ms_array) == 0);
274 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
275 vd->vdev_ms_shift) == 0);
276 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
277 vd->vdev_ashift) == 0);
278 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
279 vd->vdev_asize) == 0);
280 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
281 vd->vdev_islog) == 0);
284 if (vd->vdev_dtl_smo.smo_object != 0)
285 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
286 vd->vdev_dtl_smo.smo_object) == 0);
290 vdev_get_stats(vd, &vs);
291 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
292 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
295 if (!vd->vdev_ops->vdev_op_leaf) {
299 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
302 for (c = 0; c < vd->vdev_children; c++)
303 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
304 getstats, isspare, isl2cache);
306 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
307 child, vd->vdev_children) == 0);
309 for (c = 0; c < vd->vdev_children; c++)
310 nvlist_free(child[c]);
312 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
315 if (vd->vdev_offline && !vd->vdev_tmpoffline)
316 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
318 if (vd->vdev_faulted)
319 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
321 if (vd->vdev_degraded)
322 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
324 if (vd->vdev_removed)
325 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
327 if (vd->vdev_unspare)
328 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
336 vdev_label_read_config(vdev_t *vd)
338 spa_t *spa = vd->vdev_spa;
339 nvlist_t *config = NULL;
342 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
343 ZIO_FLAG_SPECULATIVE;
345 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
347 if (!vdev_readable(vd))
350 vp = zio_buf_alloc(sizeof (vdev_phys_t));
353 for (int l = 0; l < VDEV_LABELS; l++) {
355 zio = zio_root(spa, NULL, NULL, flags);
357 vdev_label_read(zio, vd, l, vp,
358 offsetof(vdev_label_t, vl_vdev_phys),
359 sizeof (vdev_phys_t), NULL, NULL, flags);
361 if (zio_wait(zio) == 0 &&
362 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
366 if (config != NULL) {
372 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
373 flags |= ZIO_FLAG_TRYHARD;
377 zio_buf_free(vp, sizeof (vdev_phys_t));
383 * Determine if a device is in use. The 'spare_guid' parameter will be filled
384 * in with the device guid if this spare is active elsewhere on the system.
387 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
388 uint64_t *spare_guid, uint64_t *l2cache_guid)
390 spa_t *spa = vd->vdev_spa;
391 uint64_t state, pool_guid, device_guid, txg, spare_pool;
398 *l2cache_guid = 0ULL;
401 * Read the label, if any, and perform some basic sanity checks.
403 if ((label = vdev_label_read_config(vd)) == NULL)
406 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
409 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
411 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
412 &device_guid) != 0) {
417 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
418 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
420 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
429 * Check to see if this device indeed belongs to the pool it claims to
430 * be a part of. The only way this is allowed is if the device is a hot
431 * spare (which we check for later on).
433 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
434 !spa_guid_exists(pool_guid, device_guid) &&
435 !spa_spare_exists(device_guid, NULL, NULL) &&
436 !spa_l2cache_exists(device_guid, NULL))
440 * If the transaction group is zero, then this an initialized (but
441 * unused) label. This is only an error if the create transaction
442 * on-disk is the same as the one we're using now, in which case the
443 * user has attempted to add the same vdev multiple times in the same
446 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
447 txg == 0 && vdtxg == crtxg)
451 * Check to see if this is a spare device. We do an explicit check for
452 * spa_has_spare() here because it may be on our pending list of spares
453 * to add. We also check if it is an l2cache device.
455 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
456 spa_has_spare(spa, device_guid)) {
458 *spare_guid = device_guid;
461 case VDEV_LABEL_CREATE:
462 case VDEV_LABEL_L2CACHE:
465 case VDEV_LABEL_REPLACE:
466 return (!spa_has_spare(spa, device_guid) ||
469 case VDEV_LABEL_SPARE:
470 return (spa_has_spare(spa, device_guid));
475 * Check to see if this is an l2cache device.
477 if (spa_l2cache_exists(device_guid, NULL))
481 * If the device is marked ACTIVE, then this device is in use by another
482 * pool on the system.
484 return (state == POOL_STATE_ACTIVE);
488 * Initialize a vdev label. We check to make sure each leaf device is not in
489 * use, and writable. We put down an initial label which we will later
490 * overwrite with a complete label. Note that it's important to do this
491 * sequentially, not in parallel, so that we catch cases of multiple use of the
492 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
496 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
498 spa_t *spa = vd->vdev_spa;
507 uint64_t spare_guid, l2cache_guid;
508 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
510 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
512 for (int c = 0; c < vd->vdev_children; c++)
513 if ((error = vdev_label_init(vd->vdev_child[c],
514 crtxg, reason)) != 0)
517 if (!vd->vdev_ops->vdev_op_leaf)
521 * Dead vdevs cannot be initialized.
523 if (vdev_is_dead(vd))
527 * Determine if the vdev is in use.
529 if (reason != VDEV_LABEL_REMOVE &&
530 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
534 * If this is a request to add or replace a spare or l2cache device
535 * that is in use elsewhere on the system, then we must update the
536 * guid (which was initialized to a random value) to reflect the
537 * actual GUID (which is shared between multiple pools).
539 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
540 spare_guid != 0ULL) {
541 uint64_t guid_delta = spare_guid - vd->vdev_guid;
543 vd->vdev_guid += guid_delta;
545 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
546 pvd->vdev_guid_sum += guid_delta;
549 * If this is a replacement, then we want to fallthrough to the
550 * rest of the code. If we're adding a spare, then it's already
551 * labeled appropriately and we can just return.
553 if (reason == VDEV_LABEL_SPARE)
555 ASSERT(reason == VDEV_LABEL_REPLACE);
558 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
559 l2cache_guid != 0ULL) {
560 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
562 vd->vdev_guid += guid_delta;
564 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
565 pvd->vdev_guid_sum += guid_delta;
568 * If this is a replacement, then we want to fallthrough to the
569 * rest of the code. If we're adding an l2cache, then it's
570 * already labeled appropriately and we can just return.
572 if (reason == VDEV_LABEL_L2CACHE)
574 ASSERT(reason == VDEV_LABEL_REPLACE);
578 * Initialize its label.
580 vp = zio_buf_alloc(sizeof (vdev_phys_t));
581 bzero(vp, sizeof (vdev_phys_t));
584 * Generate a label describing the pool and our top-level vdev.
585 * We mark it as being from txg 0 to indicate that it's not
586 * really part of an active pool just yet. The labels will
587 * be written again with a meaningful txg by spa_sync().
589 if (reason == VDEV_LABEL_SPARE ||
590 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
592 * For inactive hot spares, we generate a special label that
593 * identifies as a mutually shared hot spare. We write the
594 * label if we are adding a hot spare, or if we are removing an
595 * active hot spare (in which case we want to revert the
598 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
600 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
601 spa_version(spa)) == 0);
602 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
603 POOL_STATE_SPARE) == 0);
604 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
605 vd->vdev_guid) == 0);
606 } else if (reason == VDEV_LABEL_L2CACHE ||
607 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
609 * For level 2 ARC devices, add a special label.
611 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
613 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
614 spa_version(spa)) == 0);
615 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
616 POOL_STATE_L2CACHE) == 0);
617 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
618 vd->vdev_guid) == 0);
620 label = spa_config_generate(spa, vd, 0ULL, B_FALSE);
623 * Add our creation time. This allows us to detect multiple
624 * vdev uses as described above, and automatically expires if we
627 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
632 buflen = sizeof (vp->vp_nvlist);
634 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
637 zio_buf_free(vp, sizeof (vdev_phys_t));
638 /* EFAULT means nvlist_pack ran out of room */
639 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
643 * Initialize uberblock template.
645 ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
646 bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
647 *ub = spa->spa_uberblock;
650 /* Initialize the 2nd padding area. */
651 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
652 bzero(pad2, VDEV_PAD_SIZE);
655 * Write everything in parallel.
658 zio = zio_root(spa, NULL, NULL, flags);
660 for (int l = 0; l < VDEV_LABELS; l++) {
662 vdev_label_write(zio, vd, l, vp,
663 offsetof(vdev_label_t, vl_vdev_phys),
664 sizeof (vdev_phys_t), NULL, NULL, flags);
667 * Skip the 1st padding area.
668 * Zero out the 2nd padding area where it might have
669 * left over data from previous filesystem format.
671 vdev_label_write(zio, vd, l, pad2,
672 offsetof(vdev_label_t, vl_pad2),
673 VDEV_PAD_SIZE, NULL, NULL, flags);
675 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
676 vdev_label_write(zio, vd, l, ub,
677 VDEV_UBERBLOCK_OFFSET(vd, n),
678 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
682 error = zio_wait(zio);
684 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
685 flags |= ZIO_FLAG_TRYHARD;
690 zio_buf_free(pad2, VDEV_PAD_SIZE);
691 zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
692 zio_buf_free(vp, sizeof (vdev_phys_t));
695 * If this vdev hasn't been previously identified as a spare, then we
696 * mark it as such only if a) we are labeling it as a spare, or b) it
697 * exists as a spare elsewhere in the system. Do the same for
698 * level 2 ARC devices.
700 if (error == 0 && !vd->vdev_isspare &&
701 (reason == VDEV_LABEL_SPARE ||
702 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
705 if (error == 0 && !vd->vdev_isl2cache &&
706 (reason == VDEV_LABEL_L2CACHE ||
707 spa_l2cache_exists(vd->vdev_guid, NULL)))
714 * ==========================================================================
715 * uberblock load/sync
716 * ==========================================================================
720 * For use by zdb and debugging purposes only
722 uint64_t ub_max_txg = UINT64_MAX;
725 * Consider the following situation: txg is safely synced to disk. We've
726 * written the first uberblock for txg + 1, and then we lose power. When we
727 * come back up, we fail to see the uberblock for txg + 1 because, say,
728 * it was on a mirrored device and the replica to which we wrote txg + 1
729 * is now offline. If we then make some changes and sync txg + 1, and then
730 * the missing replica comes back, then for a new seconds we'll have two
731 * conflicting uberblocks on disk with the same txg. The solution is simple:
732 * among uberblocks with equal txg, choose the one with the latest timestamp.
735 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
737 if (ub1->ub_txg < ub2->ub_txg)
739 if (ub1->ub_txg > ub2->ub_txg)
742 if (ub1->ub_timestamp < ub2->ub_timestamp)
744 if (ub1->ub_timestamp > ub2->ub_timestamp)
751 vdev_uberblock_load_done(zio_t *zio)
753 zio_t *rio = zio->io_private;
754 uberblock_t *ub = zio->io_data;
755 uberblock_t *ubbest = rio->io_private;
757 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
759 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
760 mutex_enter(&rio->io_lock);
761 if (ub->ub_txg <= ub_max_txg &&
762 vdev_uberblock_compare(ub, ubbest) > 0)
764 mutex_exit(&rio->io_lock);
767 zio_buf_free(zio->io_data, zio->io_size);
771 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
773 spa_t *spa = vd->vdev_spa;
774 vdev_t *rvd = spa->spa_root_vdev;
775 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
776 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
780 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
781 zio = zio_root(spa, NULL, ubbest, flags);
782 bzero(ubbest, sizeof (uberblock_t));
787 for (int c = 0; c < vd->vdev_children; c++)
788 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
790 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
791 for (int l = 0; l < VDEV_LABELS; l++) {
792 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
793 vdev_label_read(zio, vd, l,
794 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
795 VDEV_UBERBLOCK_OFFSET(vd, n),
796 VDEV_UBERBLOCK_SIZE(vd),
797 vdev_uberblock_load_done, zio, flags);
803 (void) zio_wait(zio);
804 spa_config_exit(spa, SCL_ALL, FTAG);
809 * On success, increment root zio's count of good writes.
810 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
813 vdev_uberblock_sync_done(zio_t *zio)
815 uint64_t *good_writes = zio->io_private;
817 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
818 atomic_add_64(good_writes, 1);
822 * Write the uberblock to all labels of all leaves of the specified vdev.
825 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
830 for (int c = 0; c < vd->vdev_children; c++)
831 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
833 if (!vd->vdev_ops->vdev_op_leaf)
836 if (!vdev_writeable(vd))
839 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
841 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
842 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
845 for (int l = 0; l < VDEV_LABELS; l++)
846 vdev_label_write(zio, vd, l, ubbuf,
847 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
848 vdev_uberblock_sync_done, zio->io_private,
849 flags | ZIO_FLAG_DONT_PROPAGATE);
851 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
855 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
857 spa_t *spa = svd[0]->vdev_spa;
859 uint64_t good_writes = 0;
861 zio = zio_root(spa, NULL, &good_writes, flags);
863 for (int v = 0; v < svdcount; v++)
864 vdev_uberblock_sync(zio, ub, svd[v], flags);
866 (void) zio_wait(zio);
869 * Flush the uberblocks to disk. This ensures that the odd labels
870 * are no longer needed (because the new uberblocks and the even
871 * labels are safely on disk), so it is safe to overwrite them.
873 zio = zio_root(spa, NULL, NULL, flags);
875 for (int v = 0; v < svdcount; v++)
876 zio_flush(zio, svd[v]);
878 (void) zio_wait(zio);
880 return (good_writes >= 1 ? 0 : EIO);
884 * On success, increment the count of good writes for our top-level vdev.
887 vdev_label_sync_done(zio_t *zio)
889 uint64_t *good_writes = zio->io_private;
891 if (zio->io_error == 0)
892 atomic_add_64(good_writes, 1);
896 * If there weren't enough good writes, indicate failure to the parent.
899 vdev_label_sync_top_done(zio_t *zio)
901 uint64_t *good_writes = zio->io_private;
903 if (*good_writes == 0)
906 kmem_free(good_writes, sizeof (uint64_t));
910 * We ignore errors for log and cache devices, simply free the private data.
913 vdev_label_sync_ignore_done(zio_t *zio)
915 kmem_free(zio->io_private, sizeof (uint64_t));
919 * Write all even or odd labels to all leaves of the specified vdev.
922 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
929 for (int c = 0; c < vd->vdev_children; c++)
930 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
932 if (!vd->vdev_ops->vdev_op_leaf)
935 if (!vdev_writeable(vd))
939 * Generate a label describing the top-level config to which we belong.
941 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
943 vp = zio_buf_alloc(sizeof (vdev_phys_t));
944 bzero(vp, sizeof (vdev_phys_t));
947 buflen = sizeof (vp->vp_nvlist);
949 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
950 for (; l < VDEV_LABELS; l += 2) {
951 vdev_label_write(zio, vd, l, vp,
952 offsetof(vdev_label_t, vl_vdev_phys),
953 sizeof (vdev_phys_t),
954 vdev_label_sync_done, zio->io_private,
955 flags | ZIO_FLAG_DONT_PROPAGATE);
959 zio_buf_free(vp, sizeof (vdev_phys_t));
964 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
966 list_t *dl = &spa->spa_config_dirty_list;
972 * Write the new labels to disk.
974 zio = zio_root(spa, NULL, NULL, flags);
976 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
977 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
979 zio_t *vio = zio_null(zio, spa, NULL,
980 (vd->vdev_islog || vd->vdev_aux != NULL) ?
981 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
983 vdev_label_sync(vio, vd, l, txg, flags);
987 error = zio_wait(zio);
990 * Flush the new labels to disk.
992 zio = zio_root(spa, NULL, NULL, flags);
994 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
997 (void) zio_wait(zio);
1003 * Sync the uberblock and any changes to the vdev configuration.
1005 * The order of operations is carefully crafted to ensure that
1006 * if the system panics or loses power at any time, the state on disk
1007 * is still transactionally consistent. The in-line comments below
1008 * describe the failure semantics at each stage.
1010 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1011 * at any time, you can just call it again, and it will resume its work.
1014 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1016 spa_t *spa = svd[0]->vdev_spa;
1017 uberblock_t *ub = &spa->spa_uberblock;
1021 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1024 * Normally, we don't want to try too hard to write every label and
1025 * uberblock. If there is a flaky disk, we don't want the rest of the
1026 * sync process to block while we retry. But if we can't write a
1027 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1028 * bailing out and declaring the pool faulted.
1031 flags |= ZIO_FLAG_TRYHARD;
1033 ASSERT(ub->ub_txg <= txg);
1036 * If this isn't a resync due to I/O errors,
1037 * and nothing changed in this transaction group,
1038 * and the vdev configuration hasn't changed,
1039 * then there's nothing to do.
1041 if (ub->ub_txg < txg &&
1042 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1043 list_is_empty(&spa->spa_config_dirty_list))
1046 if (txg > spa_freeze_txg(spa))
1049 ASSERT(txg <= spa->spa_final_txg);
1052 * Flush the write cache of every disk that's been written to
1053 * in this transaction group. This ensures that all blocks
1054 * written in this txg will be committed to stable storage
1055 * before any uberblock that references them.
1057 zio = zio_root(spa, NULL, NULL, flags);
1059 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1060 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1063 (void) zio_wait(zio);
1066 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1067 * system dies in the middle of this process, that's OK: all of the
1068 * even labels that made it to disk will be newer than any uberblock,
1069 * and will therefore be considered invalid. The odd labels (L1, L3),
1070 * which have not yet been touched, will still be valid. We flush
1071 * the new labels to disk to ensure that all even-label updates
1072 * are committed to stable storage before the uberblock update.
1074 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1078 * Sync the uberblocks to all vdevs in svd[].
1079 * If the system dies in the middle of this step, there are two cases
1080 * to consider, and the on-disk state is consistent either way:
1082 * (1) If none of the new uberblocks made it to disk, then the
1083 * previous uberblock will be the newest, and the odd labels
1084 * (which had not yet been touched) will be valid with respect
1085 * to that uberblock.
1087 * (2) If one or more new uberblocks made it to disk, then they
1088 * will be the newest, and the even labels (which had all
1089 * been successfully committed) will be valid with respect
1090 * to the new uberblocks.
1092 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1096 * Sync out odd labels for every dirty vdev. If the system dies
1097 * in the middle of this process, the even labels and the new
1098 * uberblocks will suffice to open the pool. The next time
1099 * the pool is opened, the first thing we'll do -- before any
1100 * user data is modified -- is mark every vdev dirty so that
1101 * all labels will be brought up to date. We flush the new labels
1102 * to disk to ensure that all odd-label updates are committed to
1103 * stable storage before the next transaction group begins.
1105 return (vdev_label_sync_list(spa, 1, txg, flags));