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 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "%Z%%M% %I% %E% SMI"
29 * Virtual Device Labels
30 * ---------------------
32 * The vdev label serves several distinct purposes:
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
37 * 2. Verify that all the devices given in a configuration are present
40 * 3. Determine the uberblock for the pool.
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transacation groups:
68 * +------+ +------+ +------+
70 * | t10 | | t10 | | t10 |
72 * +------+ +------+ +------+
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
78 * In order to identify which labels are valid, the labels are written in the
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
116 * Configuration Information
117 * -------------------------
119 * The nvlist describing the pool and vdev contains the following elements:
121 * version ZFS on-disk version
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
128 * Each leaf device label also contains the following:
130 * top_guid Unique ID for top-level vdev in which this is contained
131 * guid Unique ID for the leaf vdev
133 * The 'vs' configuration follows the format described in 'spa_config.c'.
136 #include <sys/zfs_context.h>
138 #include <sys/spa_impl.h>
141 #include <sys/vdev.h>
142 #include <sys/vdev_impl.h>
143 #include <sys/uberblock_impl.h>
144 #include <sys/metaslab.h>
146 #include <sys/fs/zfs.h>
149 * Basic routines to read and write from a vdev label.
150 * Used throughout the rest of this file.
153 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
155 ASSERT(offset < sizeof (vdev_label_t));
157 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
158 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
162 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
163 uint64_t size, zio_done_func_t *done, void *private)
165 ASSERT(vd->vdev_children == 0);
167 zio_nowait(zio_read_phys(zio, vd,
168 vdev_label_offset(vd->vdev_psize, l, offset),
169 size, buf, ZIO_CHECKSUM_LABEL, done, private,
170 ZIO_PRIORITY_SYNC_READ,
171 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE));
175 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
176 uint64_t size, zio_done_func_t *done, void *private)
178 ASSERT(vd->vdev_children == 0);
180 zio_nowait(zio_write_phys(zio, vd,
181 vdev_label_offset(vd->vdev_psize, l, offset),
182 size, buf, ZIO_CHECKSUM_LABEL, done, private,
183 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL));
187 * Generate the nvlist representing this vdev's config.
190 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
195 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
197 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
198 vd->vdev_ops->vdev_op_type) == 0);
200 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
202 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
204 if (vd->vdev_path != NULL)
205 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
206 vd->vdev_path) == 0);
208 if (vd->vdev_devid != NULL)
209 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
210 vd->vdev_devid) == 0);
212 if (vd->vdev_nparity != 0) {
213 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
214 VDEV_TYPE_RAIDZ) == 0);
217 * Make sure someone hasn't managed to sneak a fancy new vdev
218 * into a crufty old storage pool.
220 ASSERT(vd->vdev_nparity == 1 ||
221 (vd->vdev_nparity == 2 &&
222 spa_version(spa) >= ZFS_VERSION_RAID6));
225 * Note that we'll add the nparity tag even on storage pools
226 * that only support a single parity device -- older software
227 * will just ignore it.
229 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
230 vd->vdev_nparity) == 0);
233 if (vd->vdev_wholedisk != -1ULL)
234 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
235 vd->vdev_wholedisk) == 0);
237 if (vd->vdev_not_present)
238 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
240 if (vd->vdev_isspare)
241 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
243 if (!isspare && vd == vd->vdev_top) {
244 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
245 vd->vdev_ms_array) == 0);
246 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
247 vd->vdev_ms_shift) == 0);
248 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
249 vd->vdev_ashift) == 0);
250 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
251 vd->vdev_asize) == 0);
254 if (vd->vdev_dtl.smo_object != 0)
255 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
256 vd->vdev_dtl.smo_object) == 0);
260 vdev_get_stats(vd, &vs);
261 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
262 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
265 if (!vd->vdev_ops->vdev_op_leaf) {
269 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
272 for (c = 0; c < vd->vdev_children; c++)
273 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
276 VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
277 child, vd->vdev_children) == 0);
279 for (c = 0; c < vd->vdev_children; c++)
280 nvlist_free(child[c]);
282 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
285 if (vd->vdev_offline && !vd->vdev_tmpoffline)
286 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
289 (void) nvlist_remove(nv, ZPOOL_CONFIG_OFFLINE,
297 vdev_label_read_config(vdev_t *vd)
299 spa_t *spa = vd->vdev_spa;
300 nvlist_t *config = NULL;
305 ASSERT(spa_config_held(spa, RW_READER));
307 if (vdev_is_dead(vd))
310 vp = zio_buf_alloc(sizeof (vdev_phys_t));
312 for (l = 0; l < VDEV_LABELS; l++) {
314 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL |
315 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD);
317 vdev_label_read(zio, vd, l, vp,
318 offsetof(vdev_label_t, vl_vdev_phys),
319 sizeof (vdev_phys_t), NULL, NULL);
321 if (zio_wait(zio) == 0 &&
322 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
326 if (config != NULL) {
332 zio_buf_free(vp, sizeof (vdev_phys_t));
338 * Determine if a device is in use. The 'spare_guid' parameter will be filled
339 * in with the device guid if this spare is active elsewhere on the system.
342 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
343 uint64_t *spare_guid)
345 spa_t *spa = vd->vdev_spa;
346 uint64_t state, pool_guid, device_guid, txg, spare_pool;
354 * Read the label, if any, and perform some basic sanity checks.
356 if ((label = vdev_label_read_config(vd)) == NULL)
359 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
362 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
364 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
365 &device_guid) != 0) {
370 if (state != POOL_STATE_SPARE &&
371 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
373 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
382 * Check to see if this device indeed belongs to the pool it claims to
383 * be a part of. The only way this is allowed is if the device is a hot
384 * spare (which we check for later on).
386 if (state != POOL_STATE_SPARE &&
387 !spa_guid_exists(pool_guid, device_guid) &&
388 !spa_spare_exists(device_guid, NULL))
392 * If the transaction group is zero, then this an initialized (but
393 * unused) label. This is only an error if the create transaction
394 * on-disk is the same as the one we're using now, in which case the
395 * user has attempted to add the same vdev multiple times in the same
398 if (state != POOL_STATE_SPARE && txg == 0 && vdtxg == crtxg)
402 * Check to see if this is a spare device. We do an explicit check for
403 * spa_has_spare() here because it may be on our pending list of spares
406 if (spa_spare_exists(device_guid, &spare_pool) ||
407 spa_has_spare(spa, device_guid)) {
409 *spare_guid = device_guid;
412 case VDEV_LABEL_CREATE:
415 case VDEV_LABEL_REPLACE:
416 return (!spa_has_spare(spa, device_guid) ||
419 case VDEV_LABEL_SPARE:
420 return (spa_has_spare(spa, device_guid));
425 * If the device is marked ACTIVE, then this device is in use by another
426 * pool on the system.
428 return (state == POOL_STATE_ACTIVE);
432 * Initialize a vdev label. We check to make sure each leaf device is not in
433 * use, and writable. We put down an initial label which we will later
434 * overwrite with a complete label. Note that it's important to do this
435 * sequentially, not in parallel, so that we catch cases of multiple use of the
436 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
440 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
442 spa_t *spa = vd->vdev_spa;
445 vdev_boot_header_t *vb;
454 ASSERT(spa_config_held(spa, RW_WRITER));
456 for (c = 0; c < vd->vdev_children; c++)
457 if ((error = vdev_label_init(vd->vdev_child[c],
458 crtxg, reason)) != 0)
461 if (!vd->vdev_ops->vdev_op_leaf)
465 * Dead vdevs cannot be initialized.
467 if (vdev_is_dead(vd))
471 * Determine if the vdev is in use.
473 if (reason != VDEV_LABEL_REMOVE &&
474 vdev_inuse(vd, crtxg, reason, &spare_guid))
477 ASSERT(reason != VDEV_LABEL_REMOVE ||
478 vdev_inuse(vd, crtxg, reason, NULL));
481 * If this is a request to add or replace a spare that is in use
482 * elsewhere on the system, then we must update the guid (which was
483 * initialized to a random value) to reflect the actual GUID (which is
484 * shared between multiple pools).
486 if (reason != VDEV_LABEL_REMOVE && spare_guid != 0ULL) {
487 vdev_t *pvd = vd->vdev_parent;
489 for (; pvd != NULL; pvd = pvd->vdev_parent) {
490 pvd->vdev_guid_sum -= vd->vdev_guid;
491 pvd->vdev_guid_sum += spare_guid;
494 vd->vdev_guid = vd->vdev_guid_sum = spare_guid;
497 * If this is a replacement, then we want to fallthrough to the
498 * rest of the code. If we're adding a spare, then it's already
499 * labelled appropriately and we can just return.
501 if (reason == VDEV_LABEL_SPARE)
503 ASSERT(reason == VDEV_LABEL_REPLACE);
507 * Initialize its label.
509 vp = zio_buf_alloc(sizeof (vdev_phys_t));
510 bzero(vp, sizeof (vdev_phys_t));
513 * Generate a label describing the pool and our top-level vdev.
514 * We mark it as being from txg 0 to indicate that it's not
515 * really part of an active pool just yet. The labels will
516 * be written again with a meaningful txg by spa_sync().
518 if (reason == VDEV_LABEL_SPARE ||
519 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
521 * For inactive hot spares, we generate a special label that
522 * identifies as a mutually shared hot spare. We write the
523 * label if we are adding a hot spare, or if we are removing an
524 * active hot spare (in which case we want to revert the
527 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
529 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
530 spa_version(spa)) == 0);
531 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
532 POOL_STATE_SPARE) == 0);
533 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
534 vd->vdev_guid) == 0);
536 label = spa_config_generate(spa, vd, 0ULL, B_FALSE);
539 * Add our creation time. This allows us to detect multiple
540 * vdev uses as described above, and automatically expires if we
543 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
548 buflen = sizeof (vp->vp_nvlist);
550 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
553 zio_buf_free(vp, sizeof (vdev_phys_t));
554 /* EFAULT means nvlist_pack ran out of room */
555 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
559 * Initialize boot block header.
561 vb = zio_buf_alloc(sizeof (vdev_boot_header_t));
562 bzero(vb, sizeof (vdev_boot_header_t));
563 vb->vb_magic = VDEV_BOOT_MAGIC;
564 vb->vb_version = VDEV_BOOT_VERSION;
565 vb->vb_offset = VDEV_BOOT_OFFSET;
566 vb->vb_size = VDEV_BOOT_SIZE;
569 * Initialize uberblock template.
571 ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
572 bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
573 *ub = spa->spa_uberblock;
577 * Write everything in parallel.
579 zio = zio_root(spa, NULL, NULL,
580 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
582 for (l = 0; l < VDEV_LABELS; l++) {
584 vdev_label_write(zio, vd, l, vp,
585 offsetof(vdev_label_t, vl_vdev_phys),
586 sizeof (vdev_phys_t), NULL, NULL);
588 vdev_label_write(zio, vd, l, vb,
589 offsetof(vdev_label_t, vl_boot_header),
590 sizeof (vdev_boot_header_t), NULL, NULL);
592 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
593 vdev_label_write(zio, vd, l, ub,
594 VDEV_UBERBLOCK_OFFSET(vd, n),
595 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL);
599 error = zio_wait(zio);
602 zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
603 zio_buf_free(vb, sizeof (vdev_boot_header_t));
604 zio_buf_free(vp, sizeof (vdev_phys_t));
607 * If this vdev hasn't been previously identified as a spare, then we
608 * mark it as such only if a) we are labelling it as a spare, or b) it
609 * exists as a spare elsewhere in the system.
611 if (error == 0 && !vd->vdev_isspare &&
612 (reason == VDEV_LABEL_SPARE ||
613 spa_spare_exists(vd->vdev_guid, NULL)))
620 * ==========================================================================
621 * uberblock load/sync
622 * ==========================================================================
626 * Consider the following situation: txg is safely synced to disk. We've
627 * written the first uberblock for txg + 1, and then we lose power. When we
628 * come back up, we fail to see the uberblock for txg + 1 because, say,
629 * it was on a mirrored device and the replica to which we wrote txg + 1
630 * is now offline. If we then make some changes and sync txg + 1, and then
631 * the missing replica comes back, then for a new seconds we'll have two
632 * conflicting uberblocks on disk with the same txg. The solution is simple:
633 * among uberblocks with equal txg, choose the one with the latest timestamp.
636 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
638 if (ub1->ub_txg < ub2->ub_txg)
640 if (ub1->ub_txg > ub2->ub_txg)
643 if (ub1->ub_timestamp < ub2->ub_timestamp)
645 if (ub1->ub_timestamp > ub2->ub_timestamp)
652 vdev_uberblock_load_done(zio_t *zio)
654 uberblock_t *ub = zio->io_data;
655 uberblock_t *ubbest = zio->io_private;
656 spa_t *spa = zio->io_spa;
658 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
660 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
661 mutex_enter(&spa->spa_uberblock_lock);
662 if (vdev_uberblock_compare(ub, ubbest) > 0)
664 mutex_exit(&spa->spa_uberblock_lock);
667 zio_buf_free(zio->io_data, zio->io_size);
671 vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
675 for (c = 0; c < vd->vdev_children; c++)
676 vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
678 if (!vd->vdev_ops->vdev_op_leaf)
681 if (vdev_is_dead(vd))
684 for (l = 0; l < VDEV_LABELS; l++) {
685 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
686 vdev_label_read(zio, vd, l,
687 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
688 VDEV_UBERBLOCK_OFFSET(vd, n),
689 VDEV_UBERBLOCK_SIZE(vd),
690 vdev_uberblock_load_done, ubbest);
696 * Write the uberblock to both labels of all leaves of the specified vdev.
697 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
700 vdev_uberblock_sync_done(zio_t *zio)
702 uint64_t *good_writes = zio->io_root->io_private;
704 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
705 atomic_add_64(good_writes, 1);
709 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, uint64_t txg)
713 for (c = 0; c < vd->vdev_children; c++)
714 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], txg);
716 if (!vd->vdev_ops->vdev_op_leaf)
719 if (vdev_is_dead(vd))
722 n = txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
724 ASSERT(ub->ub_txg == txg);
726 for (l = 0; l < VDEV_LABELS; l++)
727 vdev_label_write(zio, vd, l, ub,
728 VDEV_UBERBLOCK_OFFSET(vd, n),
729 VDEV_UBERBLOCK_SIZE(vd),
730 vdev_uberblock_sync_done, NULL);
732 dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg);
736 vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *vd, uint64_t txg)
739 size_t size = vd->vdev_top ? VDEV_UBERBLOCK_SIZE(vd) : SPA_MAXBLOCKSIZE;
740 uint64_t *good_writes;
744 ubbuf = zio_buf_alloc(size);
748 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
750 zio = zio_root(spa, NULL, good_writes,
751 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
753 vdev_uberblock_sync(zio, ubbuf, vd, txg);
755 error = zio_wait(zio);
757 if (error && *good_writes != 0) {
758 dprintf("partial success: good_writes = %llu\n", *good_writes);
763 * It's possible to have no good writes and no error if every vdev is in
764 * the CANT_OPEN state.
766 if (*good_writes == 0 && error == 0)
769 kmem_free(good_writes, sizeof (uint64_t));
770 zio_buf_free(ubbuf, size);
776 * Sync out an individual vdev.
779 vdev_sync_label_done(zio_t *zio)
781 uint64_t *good_writes = zio->io_root->io_private;
783 if (zio->io_error == 0)
784 atomic_add_64(good_writes, 1);
788 vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg)
796 for (c = 0; c < vd->vdev_children; c++)
797 vdev_sync_label(zio, vd->vdev_child[c], l, txg);
799 if (!vd->vdev_ops->vdev_op_leaf)
802 if (vdev_is_dead(vd))
806 * Generate a label describing the top-level config to which we belong.
808 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
810 vp = zio_buf_alloc(sizeof (vdev_phys_t));
811 bzero(vp, sizeof (vdev_phys_t));
814 buflen = sizeof (vp->vp_nvlist);
816 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0)
817 vdev_label_write(zio, vd, l, vp,
818 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
819 vdev_sync_label_done, NULL);
821 zio_buf_free(vp, sizeof (vdev_phys_t));
824 dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg);
828 vdev_sync_labels(vdev_t *vd, int l, uint64_t txg)
830 uint64_t *good_writes;
834 ASSERT(vd == vd->vdev_top);
836 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
838 zio = zio_root(vd->vdev_spa, NULL, good_writes,
839 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
842 * Recursively kick off writes to all labels.
844 vdev_sync_label(zio, vd, l, txg);
846 error = zio_wait(zio);
848 if (error && *good_writes != 0) {
849 dprintf("partial success: good_writes = %llu\n", *good_writes);
853 if (*good_writes == 0 && error == 0)
856 kmem_free(good_writes, sizeof (uint64_t));
862 * Sync the entire vdev configuration.
864 * The order of operations is carefully crafted to ensure that
865 * if the system panics or loses power at any time, the state on disk
866 * is still transactionally consistent. The in-line comments below
867 * describe the failure semantics at each stage.
869 * Moreover, it is designed to be idempotent: if spa_sync_labels() fails
870 * at any time, you can just call it again, and it will resume its work.
873 vdev_config_sync(vdev_t *uvd, uint64_t txg)
875 spa_t *spa = uvd->vdev_spa;
876 uberblock_t *ub = &spa->spa_uberblock;
877 vdev_t *rvd = spa->spa_root_vdev;
882 ASSERT(ub->ub_txg <= txg);
885 * If this isn't a resync due to I/O errors, and nothing changed
886 * in this transaction group, and the vdev configuration hasn't changed,
887 * then there's nothing to do.
889 if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE &&
890 list_is_empty(&spa->spa_dirty_list)) {
891 dprintf("nothing to sync in %s in txg %llu\n",
896 if (txg > spa_freeze_txg(spa))
899 ASSERT(txg <= spa->spa_final_txg);
901 dprintf("syncing %s txg %llu\n", spa_name(spa), txg);
904 * Flush the write cache of every disk that's been written to
905 * in this transaction group. This ensures that all blocks
906 * written in this txg will be committed to stable storage
907 * before any uberblock that references them.
909 zio = zio_root(spa, NULL, NULL,
910 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
911 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
912 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) {
913 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
914 NULL, NULL, ZIO_PRIORITY_NOW,
915 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
917 (void) zio_wait(zio);
920 * Sync out the even labels (L0, L2) for every dirty vdev. If the
921 * system dies in the middle of this process, that's OK: all of the
922 * even labels that made it to disk will be newer than any uberblock,
923 * and will therefore be considered invalid. The odd labels (L1, L3),
924 * which have not yet been touched, will still be valid.
926 for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
927 vd = list_next(&spa->spa_dirty_list, vd)) {
928 for (l = 0; l < VDEV_LABELS; l++) {
931 if ((error = vdev_sync_labels(vd, l, txg)) != 0)
937 * Flush the new labels to disk. This ensures that all even-label
938 * updates are committed to stable storage before the uberblock update.
940 zio = zio_root(spa, NULL, NULL,
941 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
942 for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
943 vd = list_next(&spa->spa_dirty_list, vd)) {
944 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
945 NULL, NULL, ZIO_PRIORITY_NOW,
946 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
948 (void) zio_wait(zio);
951 * Sync the uberblocks to all vdevs in the tree specified by uvd.
952 * If the system dies in the middle of this step, there are two cases
953 * to consider, and the on-disk state is consistent either way:
955 * (1) If none of the new uberblocks made it to disk, then the
956 * previous uberblock will be the newest, and the odd labels
957 * (which had not yet been touched) will be valid with respect
960 * (2) If one or more new uberblocks made it to disk, then they
961 * will be the newest, and the even labels (which had all
962 * been successfully committed) will be valid with respect
963 * to the new uberblocks.
965 if ((error = vdev_uberblock_sync_tree(spa, ub, uvd, txg)) != 0)
969 * Flush the uberblocks to disk. This ensures that the odd labels
970 * are no longer needed (because the new uberblocks and the even
971 * labels are safely on disk), so it is safe to overwrite them.
973 (void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE,
974 NULL, NULL, ZIO_PRIORITY_NOW,
975 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
978 * Sync out odd labels for every dirty vdev. If the system dies
979 * in the middle of this process, the even labels and the new
980 * uberblocks will suffice to open the pool. The next time
981 * the pool is opened, the first thing we'll do -- before any
982 * user data is modified -- is mark every vdev dirty so that
983 * all labels will be brought up to date.
985 for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
986 vd = list_next(&spa->spa_dirty_list, vd)) {
987 for (l = 0; l < VDEV_LABELS; l++) {
990 if ((error = vdev_sync_labels(vd, l, txg)) != 0)
996 * Flush the new labels to disk. This ensures that all odd-label
997 * updates are committed to stable storage before the next
998 * transaction group begins.
1000 zio = zio_root(spa, NULL, NULL,
1001 ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
1002 for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
1003 vd = list_next(&spa->spa_dirty_list, vd)) {
1004 zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
1005 NULL, NULL, ZIO_PRIORITY_NOW,
1006 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
1008 (void) zio_wait(zio);