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 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/fs/zfs.h>
151 * Basic routines to read and write from a vdev label.
152 * Used throughout the rest of this file.
155 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
157 ASSERT(offset < sizeof (vdev_label_t));
158 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
160 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
161 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
165 * Returns back the vdev label associated with the passed in offset.
168 vdev_label_number(uint64_t psize, uint64_t offset)
172 if (offset >= psize - VDEV_LABEL_END_SIZE) {
173 offset -= psize - VDEV_LABEL_END_SIZE;
174 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
176 l = offset / sizeof (vdev_label_t);
177 return (l < VDEV_LABELS ? l : -1);
181 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
182 uint64_t size, zio_done_func_t *done, void *private, int flags)
184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
188 zio_nowait(zio_read_phys(zio, vd,
189 vdev_label_offset(vd->vdev_psize, l, offset),
190 size, buf, ZIO_CHECKSUM_LABEL, done, private,
191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
195 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
196 uint64_t size, zio_done_func_t *done, void *private, int flags)
198 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
199 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
200 (SCL_CONFIG | SCL_STATE) &&
201 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
202 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
204 zio_nowait(zio_write_phys(zio, vd,
205 vdev_label_offset(vd->vdev_psize, l, offset),
206 size, buf, ZIO_CHECKSUM_LABEL, done, private,
207 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
211 * Generate the nvlist representing this vdev's config.
214 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
215 vdev_config_flag_t flags)
219 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
221 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
222 vd->vdev_ops->vdev_op_type) == 0);
223 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
224 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
226 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
228 if (vd->vdev_path != NULL)
229 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
230 vd->vdev_path) == 0);
232 if (vd->vdev_devid != NULL)
233 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
234 vd->vdev_devid) == 0);
236 if (vd->vdev_physpath != NULL)
237 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
238 vd->vdev_physpath) == 0);
240 if (vd->vdev_fru != NULL)
241 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
244 if (vd->vdev_nparity != 0) {
245 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
246 VDEV_TYPE_RAIDZ) == 0);
249 * Make sure someone hasn't managed to sneak a fancy new vdev
250 * into a crufty old storage pool.
252 ASSERT(vd->vdev_nparity == 1 ||
253 (vd->vdev_nparity <= 2 &&
254 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
255 (vd->vdev_nparity <= 3 &&
256 spa_version(spa) >= SPA_VERSION_RAIDZ3));
259 * Note that we'll add the nparity tag even on storage pools
260 * that only support a single parity device -- older software
261 * will just ignore it.
263 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
264 vd->vdev_nparity) == 0);
267 if (vd->vdev_wholedisk != -1ULL)
268 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
269 vd->vdev_wholedisk) == 0);
271 if (vd->vdev_not_present)
272 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
274 if (vd->vdev_isspare)
275 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
278 vd == vd->vdev_top) {
279 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
280 vd->vdev_ms_array) == 0);
281 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
282 vd->vdev_ms_shift) == 0);
283 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
284 vd->vdev_ashift) == 0);
285 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
286 vd->vdev_asize) == 0);
287 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
288 vd->vdev_islog) == 0);
289 if (vd->vdev_removing)
290 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
291 vd->vdev_removing) == 0);
294 if (vd->vdev_dtl_smo.smo_object != 0)
295 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
296 vd->vdev_dtl_smo.smo_object) == 0);
299 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
300 vd->vdev_crtxg) == 0);
306 vdev_get_stats(vd, &vs);
307 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
308 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
310 /* provide either current or previous scan information */
311 if (spa_scan_get_stats(spa, &ps) == 0) {
312 VERIFY(nvlist_add_uint64_array(nv,
313 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
314 sizeof (pool_scan_stat_t) / sizeof (uint64_t))
319 if (!vd->vdev_ops->vdev_op_leaf) {
323 ASSERT(!vd->vdev_ishole);
325 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
328 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
329 vdev_t *cvd = vd->vdev_child[c];
332 * If we're generating an nvlist of removing
333 * vdevs then skip over any device which is
336 if ((flags & VDEV_CONFIG_REMOVING) &&
340 child[idx++] = vdev_config_generate(spa, cvd,
345 VERIFY(nvlist_add_nvlist_array(nv,
346 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
349 for (c = 0; c < idx; c++)
350 nvlist_free(child[c]);
352 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
355 const char *aux = NULL;
357 if (vd->vdev_offline && !vd->vdev_tmpoffline)
358 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
360 if (vd->vdev_resilvering)
361 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
363 if (vd->vdev_faulted)
364 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
366 if (vd->vdev_degraded)
367 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
369 if (vd->vdev_removed)
370 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
372 if (vd->vdev_unspare)
373 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
376 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
379 switch (vd->vdev_stat.vs_aux) {
380 case VDEV_AUX_ERR_EXCEEDED:
381 aux = "err_exceeded";
384 case VDEV_AUX_EXTERNAL:
390 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
393 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
394 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
395 vd->vdev_orig_guid) == 0);
403 * Generate a view of the top-level vdevs. If we currently have holes
404 * in the namespace, then generate an array which contains a list of holey
405 * vdevs. Additionally, add the number of top-level children that currently
409 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
411 vdev_t *rvd = spa->spa_root_vdev;
415 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
417 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
418 vdev_t *tvd = rvd->vdev_child[c];
420 if (tvd->vdev_ishole)
425 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
429 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
430 rvd->vdev_children) == 0);
432 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
436 * Returns the configuration from the label of the given vdev. For vdevs
437 * which don't have a txg value stored on their label (i.e. spares/cache)
438 * or have not been completely initialized (txg = 0) just return
439 * the configuration from the first valid label we find. Otherwise,
440 * find the most up-to-date label that does not exceed the specified
444 vdev_label_read_config(vdev_t *vd, uint64_t txg)
446 spa_t *spa = vd->vdev_spa;
447 nvlist_t *config = NULL;
450 uint64_t best_txg = 0;
452 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
453 ZIO_FLAG_SPECULATIVE;
455 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
457 if (!vdev_readable(vd))
460 vp = zio_buf_alloc(sizeof (vdev_phys_t));
463 for (int l = 0; l < VDEV_LABELS; l++) {
464 nvlist_t *label = NULL;
466 zio = zio_root(spa, NULL, NULL, flags);
468 vdev_label_read(zio, vd, l, vp,
469 offsetof(vdev_label_t, vl_vdev_phys),
470 sizeof (vdev_phys_t), NULL, NULL, flags);
472 if (zio_wait(zio) == 0 &&
473 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
475 uint64_t label_txg = 0;
478 * Auxiliary vdevs won't have txg values in their
479 * labels and newly added vdevs may not have been
480 * completely initialized so just return the
481 * configuration from the first valid label we
484 error = nvlist_lookup_uint64(label,
485 ZPOOL_CONFIG_POOL_TXG, &label_txg);
486 if ((error || label_txg == 0) && !config) {
489 } else if (label_txg <= txg && label_txg > best_txg) {
490 best_txg = label_txg;
492 config = fnvlist_dup(label);
502 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
503 flags |= ZIO_FLAG_TRYHARD;
507 zio_buf_free(vp, sizeof (vdev_phys_t));
513 * Determine if a device is in use. The 'spare_guid' parameter will be filled
514 * in with the device guid if this spare is active elsewhere on the system.
517 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
518 uint64_t *spare_guid, uint64_t *l2cache_guid)
520 spa_t *spa = vd->vdev_spa;
521 uint64_t state, pool_guid, device_guid, txg, spare_pool;
528 *l2cache_guid = 0ULL;
531 * Read the label, if any, and perform some basic sanity checks.
533 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
536 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
539 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
541 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
542 &device_guid) != 0) {
547 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
548 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
550 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
559 * Check to see if this device indeed belongs to the pool it claims to
560 * be a part of. The only way this is allowed is if the device is a hot
561 * spare (which we check for later on).
563 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
564 !spa_guid_exists(pool_guid, device_guid) &&
565 !spa_spare_exists(device_guid, NULL, NULL) &&
566 !spa_l2cache_exists(device_guid, NULL))
570 * If the transaction group is zero, then this an initialized (but
571 * unused) label. This is only an error if the create transaction
572 * on-disk is the same as the one we're using now, in which case the
573 * user has attempted to add the same vdev multiple times in the same
576 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
577 txg == 0 && vdtxg == crtxg)
581 * Check to see if this is a spare device. We do an explicit check for
582 * spa_has_spare() here because it may be on our pending list of spares
583 * to add. We also check if it is an l2cache device.
585 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
586 spa_has_spare(spa, device_guid)) {
588 *spare_guid = device_guid;
591 case VDEV_LABEL_CREATE:
592 case VDEV_LABEL_L2CACHE:
595 case VDEV_LABEL_REPLACE:
596 return (!spa_has_spare(spa, device_guid) ||
599 case VDEV_LABEL_SPARE:
600 return (spa_has_spare(spa, device_guid));
605 * Check to see if this is an l2cache device.
607 if (spa_l2cache_exists(device_guid, NULL))
611 * We can't rely on a pool's state if it's been imported
612 * read-only. Instead we look to see if the pools is marked
613 * read-only in the namespace and set the state to active.
615 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
616 spa_mode(spa) == FREAD)
617 state = POOL_STATE_ACTIVE;
620 * If the device is marked ACTIVE, then this device is in use by another
621 * pool on the system.
623 return (state == POOL_STATE_ACTIVE);
627 * Initialize a vdev label. We check to make sure each leaf device is not in
628 * use, and writable. We put down an initial label which we will later
629 * overwrite with a complete label. Note that it's important to do this
630 * sequentially, not in parallel, so that we catch cases of multiple use of the
631 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
635 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
637 spa_t *spa = vd->vdev_spa;
646 uint64_t spare_guid, l2cache_guid;
647 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
649 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
651 for (int c = 0; c < vd->vdev_children; c++)
652 if ((error = vdev_label_init(vd->vdev_child[c],
653 crtxg, reason)) != 0)
656 /* Track the creation time for this vdev */
657 vd->vdev_crtxg = crtxg;
659 if (!vd->vdev_ops->vdev_op_leaf)
663 * Dead vdevs cannot be initialized.
665 if (vdev_is_dead(vd))
669 * Determine if the vdev is in use.
671 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
672 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
676 * If this is a request to add or replace a spare or l2cache device
677 * that is in use elsewhere on the system, then we must update the
678 * guid (which was initialized to a random value) to reflect the
679 * actual GUID (which is shared between multiple pools).
681 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
682 spare_guid != 0ULL) {
683 uint64_t guid_delta = spare_guid - vd->vdev_guid;
685 vd->vdev_guid += guid_delta;
687 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
688 pvd->vdev_guid_sum += guid_delta;
691 * If this is a replacement, then we want to fallthrough to the
692 * rest of the code. If we're adding a spare, then it's already
693 * labeled appropriately and we can just return.
695 if (reason == VDEV_LABEL_SPARE)
697 ASSERT(reason == VDEV_LABEL_REPLACE ||
698 reason == VDEV_LABEL_SPLIT);
701 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
702 l2cache_guid != 0ULL) {
703 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
705 vd->vdev_guid += guid_delta;
707 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
708 pvd->vdev_guid_sum += guid_delta;
711 * If this is a replacement, then we want to fallthrough to the
712 * rest of the code. If we're adding an l2cache, then it's
713 * already labeled appropriately and we can just return.
715 if (reason == VDEV_LABEL_L2CACHE)
717 ASSERT(reason == VDEV_LABEL_REPLACE);
721 * Initialize its label.
723 vp = zio_buf_alloc(sizeof (vdev_phys_t));
724 bzero(vp, sizeof (vdev_phys_t));
727 * Generate a label describing the pool and our top-level vdev.
728 * We mark it as being from txg 0 to indicate that it's not
729 * really part of an active pool just yet. The labels will
730 * be written again with a meaningful txg by spa_sync().
732 if (reason == VDEV_LABEL_SPARE ||
733 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
735 * For inactive hot spares, we generate a special label that
736 * identifies as a mutually shared hot spare. We write the
737 * label if we are adding a hot spare, or if we are removing an
738 * active hot spare (in which case we want to revert the
741 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
743 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
744 spa_version(spa)) == 0);
745 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
746 POOL_STATE_SPARE) == 0);
747 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
748 vd->vdev_guid) == 0);
749 } else if (reason == VDEV_LABEL_L2CACHE ||
750 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
752 * For level 2 ARC devices, add a special label.
754 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
756 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
757 spa_version(spa)) == 0);
758 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
759 POOL_STATE_L2CACHE) == 0);
760 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
761 vd->vdev_guid) == 0);
765 if (reason == VDEV_LABEL_SPLIT)
766 txg = spa->spa_uberblock.ub_txg;
767 label = spa_config_generate(spa, vd, txg, B_FALSE);
770 * Add our creation time. This allows us to detect multiple
771 * vdev uses as described above, and automatically expires if we
774 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
779 buflen = sizeof (vp->vp_nvlist);
781 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
784 zio_buf_free(vp, sizeof (vdev_phys_t));
785 /* EFAULT means nvlist_pack ran out of room */
786 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
790 * Initialize uberblock template.
792 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
793 bzero(ub, VDEV_UBERBLOCK_RING);
794 *ub = spa->spa_uberblock;
797 /* Initialize the 2nd padding area. */
798 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
799 bzero(pad2, VDEV_PAD_SIZE);
802 * Write everything in parallel.
805 zio = zio_root(spa, NULL, NULL, flags);
807 for (int l = 0; l < VDEV_LABELS; l++) {
809 vdev_label_write(zio, vd, l, vp,
810 offsetof(vdev_label_t, vl_vdev_phys),
811 sizeof (vdev_phys_t), NULL, NULL, flags);
814 * Skip the 1st padding area.
815 * Zero out the 2nd padding area where it might have
816 * left over data from previous filesystem format.
818 vdev_label_write(zio, vd, l, pad2,
819 offsetof(vdev_label_t, vl_pad2),
820 VDEV_PAD_SIZE, NULL, NULL, flags);
822 vdev_label_write(zio, vd, l, ub,
823 offsetof(vdev_label_t, vl_uberblock),
824 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
827 error = zio_wait(zio);
829 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
830 flags |= ZIO_FLAG_TRYHARD;
835 zio_buf_free(pad2, VDEV_PAD_SIZE);
836 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
837 zio_buf_free(vp, sizeof (vdev_phys_t));
840 * If this vdev hasn't been previously identified as a spare, then we
841 * mark it as such only if a) we are labeling it as a spare, or b) it
842 * exists as a spare elsewhere in the system. Do the same for
843 * level 2 ARC devices.
845 if (error == 0 && !vd->vdev_isspare &&
846 (reason == VDEV_LABEL_SPARE ||
847 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
850 if (error == 0 && !vd->vdev_isl2cache &&
851 (reason == VDEV_LABEL_L2CACHE ||
852 spa_l2cache_exists(vd->vdev_guid, NULL)))
859 * ==========================================================================
860 * uberblock load/sync
861 * ==========================================================================
865 * Consider the following situation: txg is safely synced to disk. We've
866 * written the first uberblock for txg + 1, and then we lose power. When we
867 * come back up, we fail to see the uberblock for txg + 1 because, say,
868 * it was on a mirrored device and the replica to which we wrote txg + 1
869 * is now offline. If we then make some changes and sync txg + 1, and then
870 * the missing replica comes back, then for a few seconds we'll have two
871 * conflicting uberblocks on disk with the same txg. The solution is simple:
872 * among uberblocks with equal txg, choose the one with the latest timestamp.
875 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
877 if (ub1->ub_txg < ub2->ub_txg)
879 if (ub1->ub_txg > ub2->ub_txg)
882 if (ub1->ub_timestamp < ub2->ub_timestamp)
884 if (ub1->ub_timestamp > ub2->ub_timestamp)
891 uberblock_t *ubl_ubbest; /* Best uberblock */
892 vdev_t *ubl_vd; /* vdev associated with the above */
896 vdev_uberblock_load_done(zio_t *zio)
898 vdev_t *vd = zio->io_vd;
899 spa_t *spa = zio->io_spa;
900 zio_t *rio = zio->io_private;
901 uberblock_t *ub = zio->io_data;
902 struct ubl_cbdata *cbp = rio->io_private;
904 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
906 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
907 mutex_enter(&rio->io_lock);
908 if (ub->ub_txg <= spa->spa_load_max_txg &&
909 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
911 * Keep track of the vdev in which this uberblock
912 * was found. We will use this information later
913 * to obtain the config nvlist associated with
916 *cbp->ubl_ubbest = *ub;
919 mutex_exit(&rio->io_lock);
922 zio_buf_free(zio->io_data, zio->io_size);
926 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
927 struct ubl_cbdata *cbp)
929 for (int c = 0; c < vd->vdev_children; c++)
930 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
932 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
933 for (int l = 0; l < VDEV_LABELS; l++) {
934 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
935 vdev_label_read(zio, vd, l,
936 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
937 VDEV_UBERBLOCK_OFFSET(vd, n),
938 VDEV_UBERBLOCK_SIZE(vd),
939 vdev_uberblock_load_done, zio, flags);
946 * Reads the 'best' uberblock from disk along with its associated
947 * configuration. First, we read the uberblock array of each label of each
948 * vdev, keeping track of the uberblock with the highest txg in each array.
949 * Then, we read the configuration from the same vdev as the best uberblock.
952 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
955 spa_t *spa = rvd->vdev_spa;
956 struct ubl_cbdata cb;
957 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
958 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
963 bzero(ub, sizeof (uberblock_t));
969 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
970 zio = zio_root(spa, NULL, &cb, flags);
971 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
972 (void) zio_wait(zio);
975 * It's possible that the best uberblock was discovered on a label
976 * that has a configuration which was written in a future txg.
977 * Search all labels on this vdev to find the configuration that
978 * matches the txg for our uberblock.
980 if (cb.ubl_vd != NULL)
981 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
982 spa_config_exit(spa, SCL_ALL, FTAG);
986 * On success, increment root zio's count of good writes.
987 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
990 vdev_uberblock_sync_done(zio_t *zio)
992 uint64_t *good_writes = zio->io_private;
994 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
995 atomic_add_64(good_writes, 1);
999 * Write the uberblock to all labels of all leaves of the specified vdev.
1002 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1007 for (int c = 0; c < vd->vdev_children; c++)
1008 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1010 if (!vd->vdev_ops->vdev_op_leaf)
1013 if (!vdev_writeable(vd))
1016 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1018 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1019 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1022 for (int l = 0; l < VDEV_LABELS; l++)
1023 vdev_label_write(zio, vd, l, ubbuf,
1024 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1025 vdev_uberblock_sync_done, zio->io_private,
1026 flags | ZIO_FLAG_DONT_PROPAGATE);
1028 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1032 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1034 spa_t *spa = svd[0]->vdev_spa;
1036 uint64_t good_writes = 0;
1038 zio = zio_root(spa, NULL, &good_writes, flags);
1040 for (int v = 0; v < svdcount; v++)
1041 vdev_uberblock_sync(zio, ub, svd[v], flags);
1043 (void) zio_wait(zio);
1046 * Flush the uberblocks to disk. This ensures that the odd labels
1047 * are no longer needed (because the new uberblocks and the even
1048 * labels are safely on disk), so it is safe to overwrite them.
1050 zio = zio_root(spa, NULL, NULL, flags);
1052 for (int v = 0; v < svdcount; v++)
1053 zio_flush(zio, svd[v]);
1055 (void) zio_wait(zio);
1057 return (good_writes >= 1 ? 0 : EIO);
1061 * On success, increment the count of good writes for our top-level vdev.
1064 vdev_label_sync_done(zio_t *zio)
1066 uint64_t *good_writes = zio->io_private;
1068 if (zio->io_error == 0)
1069 atomic_add_64(good_writes, 1);
1073 * If there weren't enough good writes, indicate failure to the parent.
1076 vdev_label_sync_top_done(zio_t *zio)
1078 uint64_t *good_writes = zio->io_private;
1080 if (*good_writes == 0)
1081 zio->io_error = EIO;
1083 kmem_free(good_writes, sizeof (uint64_t));
1087 * We ignore errors for log and cache devices, simply free the private data.
1090 vdev_label_sync_ignore_done(zio_t *zio)
1092 kmem_free(zio->io_private, sizeof (uint64_t));
1096 * Write all even or odd labels to all leaves of the specified vdev.
1099 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1106 for (int c = 0; c < vd->vdev_children; c++)
1107 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1109 if (!vd->vdev_ops->vdev_op_leaf)
1112 if (!vdev_writeable(vd))
1116 * Generate a label describing the top-level config to which we belong.
1118 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1120 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1121 bzero(vp, sizeof (vdev_phys_t));
1123 buf = vp->vp_nvlist;
1124 buflen = sizeof (vp->vp_nvlist);
1126 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1127 for (; l < VDEV_LABELS; l += 2) {
1128 vdev_label_write(zio, vd, l, vp,
1129 offsetof(vdev_label_t, vl_vdev_phys),
1130 sizeof (vdev_phys_t),
1131 vdev_label_sync_done, zio->io_private,
1132 flags | ZIO_FLAG_DONT_PROPAGATE);
1136 zio_buf_free(vp, sizeof (vdev_phys_t));
1141 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1143 list_t *dl = &spa->spa_config_dirty_list;
1149 * Write the new labels to disk.
1151 zio = zio_root(spa, NULL, NULL, flags);
1153 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1154 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1157 ASSERT(!vd->vdev_ishole);
1159 zio_t *vio = zio_null(zio, spa, NULL,
1160 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1161 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1162 good_writes, flags);
1163 vdev_label_sync(vio, vd, l, txg, flags);
1167 error = zio_wait(zio);
1170 * Flush the new labels to disk.
1172 zio = zio_root(spa, NULL, NULL, flags);
1174 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1177 (void) zio_wait(zio);
1183 * Sync the uberblock and any changes to the vdev configuration.
1185 * The order of operations is carefully crafted to ensure that
1186 * if the system panics or loses power at any time, the state on disk
1187 * is still transactionally consistent. The in-line comments below
1188 * describe the failure semantics at each stage.
1190 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1191 * at any time, you can just call it again, and it will resume its work.
1194 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1196 spa_t *spa = svd[0]->vdev_spa;
1197 uberblock_t *ub = &spa->spa_uberblock;
1201 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1204 * Normally, we don't want to try too hard to write every label and
1205 * uberblock. If there is a flaky disk, we don't want the rest of the
1206 * sync process to block while we retry. But if we can't write a
1207 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1208 * bailing out and declaring the pool faulted.
1211 flags |= ZIO_FLAG_TRYHARD;
1213 ASSERT(ub->ub_txg <= txg);
1216 * If this isn't a resync due to I/O errors,
1217 * and nothing changed in this transaction group,
1218 * and the vdev configuration hasn't changed,
1219 * then there's nothing to do.
1221 if (ub->ub_txg < txg &&
1222 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1223 list_is_empty(&spa->spa_config_dirty_list))
1226 if (txg > spa_freeze_txg(spa))
1229 ASSERT(txg <= spa->spa_final_txg);
1232 * Flush the write cache of every disk that's been written to
1233 * in this transaction group. This ensures that all blocks
1234 * written in this txg will be committed to stable storage
1235 * before any uberblock that references them.
1237 zio = zio_root(spa, NULL, NULL, flags);
1239 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1240 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1243 (void) zio_wait(zio);
1246 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1247 * system dies in the middle of this process, that's OK: all of the
1248 * even labels that made it to disk will be newer than any uberblock,
1249 * and will therefore be considered invalid. The odd labels (L1, L3),
1250 * which have not yet been touched, will still be valid. We flush
1251 * the new labels to disk to ensure that all even-label updates
1252 * are committed to stable storage before the uberblock update.
1254 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1258 * Sync the uberblocks to all vdevs in svd[].
1259 * If the system dies in the middle of this step, there are two cases
1260 * to consider, and the on-disk state is consistent either way:
1262 * (1) If none of the new uberblocks made it to disk, then the
1263 * previous uberblock will be the newest, and the odd labels
1264 * (which had not yet been touched) will be valid with respect
1265 * to that uberblock.
1267 * (2) If one or more new uberblocks made it to disk, then they
1268 * will be the newest, and the even labels (which had all
1269 * been successfully committed) will be valid with respect
1270 * to the new uberblocks.
1272 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1276 * Sync out odd labels for every dirty vdev. If the system dies
1277 * in the middle of this process, the even labels and the new
1278 * uberblocks will suffice to open the pool. The next time
1279 * the pool is opened, the first thing we'll do -- before any
1280 * user data is modified -- is mark every vdev dirty so that
1281 * all labels will be brought up to date. We flush the new labels
1282 * to disk to ensure that all odd-label updates are committed to
1283 * stable storage before the next transaction group begins.
1285 return (vdev_label_sync_list(spa, 1, txg, flags));