]> git.proxmox.com Git - mirror_zfs-debian.git/blame - module/zfs/vdev_label.c
Imported Upstream version 0.6.2+git20140204
[mirror_zfs-debian.git] / module / zfs / vdev_label.c
CommitLineData
34dc7c2f
BB
1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
9ae529ec 21
34dc7c2f 22/*
428870ff 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
a08ee875 24 * Copyright (c) 2013 by Delphix. All rights reserved.
34dc7c2f
BB
25 */
26
34dc7c2f
BB
27/*
28 * Virtual Device Labels
29 * ---------------------
30 *
31 * The vdev label serves several distinct purposes:
32 *
33 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
34 * identity within the pool.
35 *
36 * 2. Verify that all the devices given in a configuration are present
37 * within the pool.
38 *
39 * 3. Determine the uberblock for the pool.
40 *
41 * 4. In case of an import operation, determine the configuration of the
42 * toplevel vdev of which it is a part.
43 *
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.
47 *
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.
52 *
53 *
54 * Label Organization
55 * ------------------
56 *
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.
59 *
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:
65 *
66 * L1 UB L2
67 * +------+ +------+ +------+
68 * | | | | | |
69 * | t10 | | t10 | | t10 |
70 * | | | | | |
71 * +------+ +------+ +------+
72 *
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.
76 *
77 * In order to identify which labels are valid, the labels are written in the
78 * following manner:
79 *
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
83 *
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.
90 *
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
95 * on another vdev.
96 *
97 *
98 * On-disk Format
99 * --------------
100 *
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.
104 *
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.
108 *
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.
113 *
114 *
115 * Configuration Information
116 * -------------------------
117 *
118 * The nvlist describing the pool and vdev contains the following elements:
119 *
120 * version ZFS on-disk version
121 * name Pool name
122 * state Pool state
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.
9ae529ec
CS
126 * features_for_read
127 * An nvlist of the features necessary for reading the MOS.
34dc7c2f
BB
128 *
129 * Each leaf device label also contains the following:
130 *
131 * top_guid Unique ID for top-level vdev in which this is contained
132 * guid Unique ID for the leaf vdev
133 *
134 * The 'vs' configuration follows the format described in 'spa_config.c'.
135 */
136
137#include <sys/zfs_context.h>
138#include <sys/spa.h>
139#include <sys/spa_impl.h>
140#include <sys/dmu.h>
141#include <sys/zap.h>
142#include <sys/vdev.h>
143#include <sys/vdev_impl.h>
144#include <sys/uberblock_impl.h>
145#include <sys/metaslab.h>
146#include <sys/zio.h>
428870ff 147#include <sys/dsl_scan.h>
34dc7c2f
BB
148#include <sys/fs/zfs.h>
149
150/*
151 * Basic routines to read and write from a vdev label.
152 * Used throughout the rest of this file.
153 */
154uint64_t
155vdev_label_offset(uint64_t psize, int l, uint64_t offset)
156{
157 ASSERT(offset < sizeof (vdev_label_t));
158 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
159
160 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
161 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
162}
163
b128c09f
BB
164/*
165 * Returns back the vdev label associated with the passed in offset.
166 */
167int
168vdev_label_number(uint64_t psize, uint64_t offset)
169{
170 int l;
171
172 if (offset >= psize - VDEV_LABEL_END_SIZE) {
173 offset -= psize - VDEV_LABEL_END_SIZE;
174 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
175 }
176 l = offset / sizeof (vdev_label_t);
177 return (l < VDEV_LABELS ? l : -1);
178}
179
34dc7c2f
BB
180static void
181vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
b128c09f 182 uint64_t size, zio_done_func_t *done, void *private, int flags)
34dc7c2f 183{
b128c09f
BB
184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
185 SCL_STATE_ALL);
186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
34dc7c2f
BB
187
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,
b128c09f 191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
34dc7c2f
BB
192}
193
194static void
195vdev_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)
197{
b128c09f
BB
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);
34dc7c2f
BB
203
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));
208}
209
210/*
211 * Generate the nvlist representing this vdev's config.
212 */
213nvlist_t *
214vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
428870ff 215 vdev_config_flag_t flags)
34dc7c2f
BB
216{
217 nvlist_t *nv = NULL;
218
b8d06fca 219 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
34dc7c2f 220
a08ee875 221 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
428870ff 222 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
a08ee875
LG
223 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
224 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
34dc7c2f
BB
225
226 if (vd->vdev_path != NULL)
a08ee875 227 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
34dc7c2f
BB
228
229 if (vd->vdev_devid != NULL)
a08ee875 230 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
34dc7c2f
BB
231
232 if (vd->vdev_physpath != NULL)
a08ee875
LG
233 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 vd->vdev_physpath);
34dc7c2f 235
9babb374 236 if (vd->vdev_fru != NULL)
a08ee875 237 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
9babb374 238
34dc7c2f
BB
239 if (vd->vdev_nparity != 0) {
240 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
241 VDEV_TYPE_RAIDZ) == 0);
242
243 /*
244 * Make sure someone hasn't managed to sneak a fancy new vdev
245 * into a crufty old storage pool.
246 */
247 ASSERT(vd->vdev_nparity == 1 ||
45d1cae3
BB
248 (vd->vdev_nparity <= 2 &&
249 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
250 (vd->vdev_nparity <= 3 &&
251 spa_version(spa) >= SPA_VERSION_RAIDZ3));
34dc7c2f
BB
252
253 /*
254 * Note that we'll add the nparity tag even on storage pools
255 * that only support a single parity device -- older software
256 * will just ignore it.
257 */
a08ee875 258 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
34dc7c2f
BB
259 }
260
261 if (vd->vdev_wholedisk != -1ULL)
a08ee875
LG
262 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
263 vd->vdev_wholedisk);
34dc7c2f
BB
264
265 if (vd->vdev_not_present)
a08ee875 266 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
34dc7c2f
BB
267
268 if (vd->vdev_isspare)
a08ee875 269 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
34dc7c2f 270
428870ff
BB
271 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
272 vd == vd->vdev_top) {
a08ee875
LG
273 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
274 vd->vdev_ms_array);
275 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
276 vd->vdev_ms_shift);
277 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
278 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
279 vd->vdev_asize);
280 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
428870ff 281 if (vd->vdev_removing)
a08ee875
LG
282 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
283 vd->vdev_removing);
34dc7c2f
BB
284 }
285
fb5f0bc8 286 if (vd->vdev_dtl_smo.smo_object != 0)
a08ee875
LG
287 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 vd->vdev_dtl_smo.smo_object);
34dc7c2f 289
428870ff 290 if (vd->vdev_crtxg)
a08ee875 291 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
428870ff 292
34dc7c2f
BB
293 if (getstats) {
294 vdev_stat_t vs;
428870ff
BB
295 pool_scan_stat_t ps;
296
34dc7c2f 297 vdev_get_stats(vd, &vs);
a08ee875
LG
298 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
299 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
428870ff
BB
300
301 /* provide either current or previous scan information */
302 if (spa_scan_get_stats(spa, &ps) == 0) {
a08ee875 303 fnvlist_add_uint64_array(nv,
428870ff 304 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
a08ee875 305 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
428870ff 306 }
34dc7c2f
BB
307 }
308
309 if (!vd->vdev_ops->vdev_op_leaf) {
310 nvlist_t **child;
428870ff
BB
311 int c, idx;
312
313 ASSERT(!vd->vdev_ishole);
34dc7c2f
BB
314
315 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
b8d06fca 316 KM_PUSHPAGE);
34dc7c2f 317
428870ff
BB
318 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
319 vdev_t *cvd = vd->vdev_child[c];
34dc7c2f 320
428870ff
BB
321 /*
322 * If we're generating an nvlist of removing
323 * vdevs then skip over any device which is
324 * not being removed.
325 */
326 if ((flags & VDEV_CONFIG_REMOVING) &&
327 !cvd->vdev_removing)
328 continue;
34dc7c2f 329
428870ff
BB
330 child[idx++] = vdev_config_generate(spa, cvd,
331 getstats, flags);
332 }
333
334 if (idx) {
a08ee875
LG
335 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
336 child, idx);
428870ff
BB
337 }
338
339 for (c = 0; c < idx; c++)
34dc7c2f
BB
340 nvlist_free(child[c]);
341
342 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
343
344 } else {
428870ff
BB
345 const char *aux = NULL;
346
34dc7c2f 347 if (vd->vdev_offline && !vd->vdev_tmpoffline)
a08ee875
LG
348 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
349 if (vd->vdev_resilver_txg != 0)
350 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
351 vd->vdev_resilver_txg);
34dc7c2f 352 if (vd->vdev_faulted)
a08ee875 353 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
34dc7c2f 354 if (vd->vdev_degraded)
a08ee875 355 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
34dc7c2f 356 if (vd->vdev_removed)
a08ee875 357 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
34dc7c2f 358 if (vd->vdev_unspare)
a08ee875 359 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
428870ff 360 if (vd->vdev_ishole)
a08ee875 361 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
428870ff
BB
362
363 switch (vd->vdev_stat.vs_aux) {
364 case VDEV_AUX_ERR_EXCEEDED:
365 aux = "err_exceeded";
366 break;
367
368 case VDEV_AUX_EXTERNAL:
369 aux = "external";
370 break;
371 }
372
373 if (aux != NULL)
a08ee875 374 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
428870ff
BB
375
376 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
a08ee875
LG
377 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
378 vd->vdev_orig_guid);
428870ff 379 }
34dc7c2f
BB
380 }
381
382 return (nv);
383}
384
428870ff
BB
385/*
386 * Generate a view of the top-level vdevs. If we currently have holes
387 * in the namespace, then generate an array which contains a list of holey
388 * vdevs. Additionally, add the number of top-level children that currently
389 * exist.
390 */
391void
392vdev_top_config_generate(spa_t *spa, nvlist_t *config)
393{
394 vdev_t *rvd = spa->spa_root_vdev;
395 uint64_t *array;
396 uint_t c, idx;
397
b8d06fca 398 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_PUSHPAGE);
428870ff
BB
399
400 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
401 vdev_t *tvd = rvd->vdev_child[c];
402
403 if (tvd->vdev_ishole)
404 array[idx++] = c;
405 }
406
407 if (idx) {
408 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
409 array, idx) == 0);
410 }
411
412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
413 rvd->vdev_children) == 0);
414
415 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
416}
417
9ae529ec 418/*
3bc7e0fb
GW
419 * Returns the configuration from the label of the given vdev. For vdevs
420 * which don't have a txg value stored on their label (i.e. spares/cache)
421 * or have not been completely initialized (txg = 0) just return
422 * the configuration from the first valid label we find. Otherwise,
423 * find the most up-to-date label that does not exceed the specified
424 * 'txg' value.
9ae529ec 425 */
34dc7c2f 426nvlist_t *
3bc7e0fb 427vdev_label_read_config(vdev_t *vd, uint64_t txg)
34dc7c2f
BB
428{
429 spa_t *spa = vd->vdev_spa;
430 nvlist_t *config = NULL;
431 vdev_phys_t *vp;
432 zio_t *zio;
3bc7e0fb
GW
433 uint64_t best_txg = 0;
434 int error = 0;
9babb374
BB
435 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
436 ZIO_FLAG_SPECULATIVE;
d6320ddb 437 int l;
34dc7c2f 438
b128c09f 439 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
34dc7c2f
BB
440
441 if (!vdev_readable(vd))
442 return (NULL);
443
444 vp = zio_buf_alloc(sizeof (vdev_phys_t));
445
9babb374 446retry:
d6320ddb 447 for (l = 0; l < VDEV_LABELS; l++) {
3bc7e0fb 448 nvlist_t *label = NULL;
34dc7c2f 449
b128c09f 450 zio = zio_root(spa, NULL, NULL, flags);
34dc7c2f
BB
451
452 vdev_label_read(zio, vd, l, vp,
453 offsetof(vdev_label_t, vl_vdev_phys),
b128c09f 454 sizeof (vdev_phys_t), NULL, NULL, flags);
34dc7c2f
BB
455
456 if (zio_wait(zio) == 0 &&
457 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
3bc7e0fb
GW
458 &label, 0) == 0) {
459 uint64_t label_txg = 0;
460
461 /*
462 * Auxiliary vdevs won't have txg values in their
463 * labels and newly added vdevs may not have been
464 * completely initialized so just return the
465 * configuration from the first valid label we
466 * encounter.
467 */
468 error = nvlist_lookup_uint64(label,
469 ZPOOL_CONFIG_POOL_TXG, &label_txg);
470 if ((error || label_txg == 0) && !config) {
471 config = label;
472 break;
473 } else if (label_txg <= txg && label_txg > best_txg) {
474 best_txg = label_txg;
475 nvlist_free(config);
476 config = fnvlist_dup(label);
477 }
478 }
34dc7c2f 479
3bc7e0fb
GW
480 if (label != NULL) {
481 nvlist_free(label);
482 label = NULL;
34dc7c2f
BB
483 }
484 }
485
9babb374
BB
486 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
487 flags |= ZIO_FLAG_TRYHARD;
488 goto retry;
489 }
490
34dc7c2f
BB
491 zio_buf_free(vp, sizeof (vdev_phys_t));
492
493 return (config);
494}
495
496/*
497 * Determine if a device is in use. The 'spare_guid' parameter will be filled
498 * in with the device guid if this spare is active elsewhere on the system.
499 */
500static boolean_t
501vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
502 uint64_t *spare_guid, uint64_t *l2cache_guid)
503{
504 spa_t *spa = vd->vdev_spa;
505 uint64_t state, pool_guid, device_guid, txg, spare_pool;
506 uint64_t vdtxg = 0;
507 nvlist_t *label;
508
509 if (spare_guid)
510 *spare_guid = 0ULL;
511 if (l2cache_guid)
512 *l2cache_guid = 0ULL;
513
514 /*
515 * Read the label, if any, and perform some basic sanity checks.
516 */
3bc7e0fb 517 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
34dc7c2f
BB
518 return (B_FALSE);
519
520 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
521 &vdtxg);
522
523 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
524 &state) != 0 ||
525 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
526 &device_guid) != 0) {
527 nvlist_free(label);
528 return (B_FALSE);
529 }
530
531 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
532 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
533 &pool_guid) != 0 ||
534 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
535 &txg) != 0)) {
536 nvlist_free(label);
537 return (B_FALSE);
538 }
539
540 nvlist_free(label);
541
542 /*
543 * Check to see if this device indeed belongs to the pool it claims to
544 * be a part of. The only way this is allowed is if the device is a hot
545 * spare (which we check for later on).
546 */
547 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
548 !spa_guid_exists(pool_guid, device_guid) &&
b128c09f 549 !spa_spare_exists(device_guid, NULL, NULL) &&
34dc7c2f
BB
550 !spa_l2cache_exists(device_guid, NULL))
551 return (B_FALSE);
552
553 /*
554 * If the transaction group is zero, then this an initialized (but
555 * unused) label. This is only an error if the create transaction
556 * on-disk is the same as the one we're using now, in which case the
557 * user has attempted to add the same vdev multiple times in the same
558 * transaction.
559 */
560 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
561 txg == 0 && vdtxg == crtxg)
562 return (B_TRUE);
563
564 /*
565 * Check to see if this is a spare device. We do an explicit check for
566 * spa_has_spare() here because it may be on our pending list of spares
567 * to add. We also check if it is an l2cache device.
568 */
b128c09f 569 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
34dc7c2f
BB
570 spa_has_spare(spa, device_guid)) {
571 if (spare_guid)
572 *spare_guid = device_guid;
573
574 switch (reason) {
575 case VDEV_LABEL_CREATE:
576 case VDEV_LABEL_L2CACHE:
577 return (B_TRUE);
578
579 case VDEV_LABEL_REPLACE:
580 return (!spa_has_spare(spa, device_guid) ||
581 spare_pool != 0ULL);
582
583 case VDEV_LABEL_SPARE:
584 return (spa_has_spare(spa, device_guid));
e75c13c3
BB
585 default:
586 break;
34dc7c2f
BB
587 }
588 }
589
590 /*
591 * Check to see if this is an l2cache device.
592 */
593 if (spa_l2cache_exists(device_guid, NULL))
594 return (B_TRUE);
595
572e2857
BB
596 /*
597 * We can't rely on a pool's state if it's been imported
598 * read-only. Instead we look to see if the pools is marked
599 * read-only in the namespace and set the state to active.
600 */
601 if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
602 spa_mode(spa) == FREAD)
603 state = POOL_STATE_ACTIVE;
604
34dc7c2f
BB
605 /*
606 * If the device is marked ACTIVE, then this device is in use by another
607 * pool on the system.
608 */
609 return (state == POOL_STATE_ACTIVE);
610}
611
612/*
613 * Initialize a vdev label. We check to make sure each leaf device is not in
614 * use, and writable. We put down an initial label which we will later
615 * overwrite with a complete label. Note that it's important to do this
616 * sequentially, not in parallel, so that we catch cases of multiple use of the
617 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
618 * itself.
619 */
620int
621vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
622{
623 spa_t *spa = vd->vdev_spa;
624 nvlist_t *label;
625 vdev_phys_t *vp;
9babb374 626 char *pad2;
34dc7c2f
BB
627 uberblock_t *ub;
628 zio_t *zio;
34dc7c2f
BB
629 char *buf;
630 size_t buflen;
631 int error;
d4ed6673 632 uint64_t spare_guid = 0, l2cache_guid = 0;
b128c09f 633 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
d6320ddb
BB
634 int c, l;
635 vdev_t *pvd;
34dc7c2f 636
b128c09f 637 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
34dc7c2f 638
d6320ddb 639 for (c = 0; c < vd->vdev_children; c++)
34dc7c2f
BB
640 if ((error = vdev_label_init(vd->vdev_child[c],
641 crtxg, reason)) != 0)
642 return (error);
643
428870ff
BB
644 /* Track the creation time for this vdev */
645 vd->vdev_crtxg = crtxg;
646
a08ee875 647 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
34dc7c2f
BB
648 return (0);
649
650 /*
651 * Dead vdevs cannot be initialized.
652 */
653 if (vdev_is_dead(vd))
a08ee875 654 return (SET_ERROR(EIO));
34dc7c2f
BB
655
656 /*
657 * Determine if the vdev is in use.
658 */
428870ff 659 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
34dc7c2f 660 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
a08ee875 661 return (SET_ERROR(EBUSY));
34dc7c2f 662
34dc7c2f
BB
663 /*
664 * If this is a request to add or replace a spare or l2cache device
665 * that is in use elsewhere on the system, then we must update the
666 * guid (which was initialized to a random value) to reflect the
667 * actual GUID (which is shared between multiple pools).
668 */
669 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
670 spare_guid != 0ULL) {
b128c09f 671 uint64_t guid_delta = spare_guid - vd->vdev_guid;
34dc7c2f 672
b128c09f 673 vd->vdev_guid += guid_delta;
34dc7c2f 674
d6320ddb 675 for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
b128c09f 676 pvd->vdev_guid_sum += guid_delta;
34dc7c2f
BB
677
678 /*
679 * If this is a replacement, then we want to fallthrough to the
680 * rest of the code. If we're adding a spare, then it's already
681 * labeled appropriately and we can just return.
682 */
683 if (reason == VDEV_LABEL_SPARE)
684 return (0);
428870ff
BB
685 ASSERT(reason == VDEV_LABEL_REPLACE ||
686 reason == VDEV_LABEL_SPLIT);
34dc7c2f
BB
687 }
688
689 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
690 l2cache_guid != 0ULL) {
b128c09f 691 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
34dc7c2f 692
b128c09f 693 vd->vdev_guid += guid_delta;
34dc7c2f 694
d6320ddb 695 for (pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
b128c09f 696 pvd->vdev_guid_sum += guid_delta;
34dc7c2f
BB
697
698 /*
699 * If this is a replacement, then we want to fallthrough to the
700 * rest of the code. If we're adding an l2cache, then it's
701 * already labeled appropriately and we can just return.
702 */
703 if (reason == VDEV_LABEL_L2CACHE)
704 return (0);
705 ASSERT(reason == VDEV_LABEL_REPLACE);
706 }
707
708 /*
709 * Initialize its label.
710 */
711 vp = zio_buf_alloc(sizeof (vdev_phys_t));
712 bzero(vp, sizeof (vdev_phys_t));
713
714 /*
715 * Generate a label describing the pool and our top-level vdev.
716 * We mark it as being from txg 0 to indicate that it's not
717 * really part of an active pool just yet. The labels will
718 * be written again with a meaningful txg by spa_sync().
719 */
720 if (reason == VDEV_LABEL_SPARE ||
721 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
722 /*
723 * For inactive hot spares, we generate a special label that
724 * identifies as a mutually shared hot spare. We write the
725 * label if we are adding a hot spare, or if we are removing an
726 * active hot spare (in which case we want to revert the
727 * labels).
728 */
b8d06fca 729 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
34dc7c2f
BB
730
731 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
732 spa_version(spa)) == 0);
733 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
734 POOL_STATE_SPARE) == 0);
735 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
736 vd->vdev_guid) == 0);
737 } else if (reason == VDEV_LABEL_L2CACHE ||
738 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
739 /*
740 * For level 2 ARC devices, add a special label.
741 */
b8d06fca 742 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_PUSHPAGE) == 0);
34dc7c2f
BB
743
744 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
745 spa_version(spa)) == 0);
746 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
747 POOL_STATE_L2CACHE) == 0);
748 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
749 vd->vdev_guid) == 0);
750 } else {
428870ff
BB
751 uint64_t txg = 0ULL;
752
753 if (reason == VDEV_LABEL_SPLIT)
754 txg = spa->spa_uberblock.ub_txg;
755 label = spa_config_generate(spa, vd, txg, B_FALSE);
34dc7c2f
BB
756
757 /*
758 * Add our creation time. This allows us to detect multiple
759 * vdev uses as described above, and automatically expires if we
760 * fail.
761 */
762 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
763 crtxg) == 0);
764 }
765
766 buf = vp->vp_nvlist;
767 buflen = sizeof (vp->vp_nvlist);
768
b8d06fca 769 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_PUSHPAGE);
34dc7c2f
BB
770 if (error != 0) {
771 nvlist_free(label);
772 zio_buf_free(vp, sizeof (vdev_phys_t));
773 /* EFAULT means nvlist_pack ran out of room */
774 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
775 }
776
34dc7c2f
BB
777 /*
778 * Initialize uberblock template.
779 */
45d1cae3
BB
780 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
781 bzero(ub, VDEV_UBERBLOCK_RING);
34dc7c2f
BB
782 *ub = spa->spa_uberblock;
783 ub->ub_txg = 0;
784
9babb374
BB
785 /* Initialize the 2nd padding area. */
786 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
787 bzero(pad2, VDEV_PAD_SIZE);
788
34dc7c2f
BB
789 /*
790 * Write everything in parallel.
791 */
9babb374 792retry:
34dc7c2f
BB
793 zio = zio_root(spa, NULL, NULL, flags);
794
d6320ddb 795 for (l = 0; l < VDEV_LABELS; l++) {
34dc7c2f
BB
796
797 vdev_label_write(zio, vd, l, vp,
798 offsetof(vdev_label_t, vl_vdev_phys),
799 sizeof (vdev_phys_t), NULL, NULL, flags);
800
9babb374
BB
801 /*
802 * Skip the 1st padding area.
803 * Zero out the 2nd padding area where it might have
804 * left over data from previous filesystem format.
805 */
806 vdev_label_write(zio, vd, l, pad2,
807 offsetof(vdev_label_t, vl_pad2),
808 VDEV_PAD_SIZE, NULL, NULL, flags);
34dc7c2f 809
45d1cae3
BB
810 vdev_label_write(zio, vd, l, ub,
811 offsetof(vdev_label_t, vl_uberblock),
812 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
34dc7c2f
BB
813 }
814
815 error = zio_wait(zio);
816
9babb374
BB
817 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
818 flags |= ZIO_FLAG_TRYHARD;
819 goto retry;
820 }
821
34dc7c2f 822 nvlist_free(label);
9babb374 823 zio_buf_free(pad2, VDEV_PAD_SIZE);
45d1cae3 824 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
34dc7c2f
BB
825 zio_buf_free(vp, sizeof (vdev_phys_t));
826
827 /*
828 * If this vdev hasn't been previously identified as a spare, then we
829 * mark it as such only if a) we are labeling it as a spare, or b) it
830 * exists as a spare elsewhere in the system. Do the same for
831 * level 2 ARC devices.
832 */
833 if (error == 0 && !vd->vdev_isspare &&
834 (reason == VDEV_LABEL_SPARE ||
b128c09f 835 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
34dc7c2f
BB
836 spa_spare_add(vd);
837
838 if (error == 0 && !vd->vdev_isl2cache &&
839 (reason == VDEV_LABEL_L2CACHE ||
840 spa_l2cache_exists(vd->vdev_guid, NULL)))
841 spa_l2cache_add(vd);
842
843 return (error);
844}
845
846/*
847 * ==========================================================================
848 * uberblock load/sync
849 * ==========================================================================
850 */
851
852/*
853 * Consider the following situation: txg is safely synced to disk. We've
854 * written the first uberblock for txg + 1, and then we lose power. When we
855 * come back up, we fail to see the uberblock for txg + 1 because, say,
856 * it was on a mirrored device and the replica to which we wrote txg + 1
857 * is now offline. If we then make some changes and sync txg + 1, and then
9ae529ec 858 * the missing replica comes back, then for a few seconds we'll have two
34dc7c2f
BB
859 * conflicting uberblocks on disk with the same txg. The solution is simple:
860 * among uberblocks with equal txg, choose the one with the latest timestamp.
861 */
862static int
863vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
864{
865 if (ub1->ub_txg < ub2->ub_txg)
866 return (-1);
867 if (ub1->ub_txg > ub2->ub_txg)
868 return (1);
869
870 if (ub1->ub_timestamp < ub2->ub_timestamp)
871 return (-1);
872 if (ub1->ub_timestamp > ub2->ub_timestamp)
873 return (1);
874
875 return (0);
876}
877
9ae529ec
CS
878struct ubl_cbdata {
879 uberblock_t *ubl_ubbest; /* Best uberblock */
880 vdev_t *ubl_vd; /* vdev associated with the above */
9ae529ec
CS
881};
882
34dc7c2f
BB
883static void
884vdev_uberblock_load_done(zio_t *zio)
885{
9ae529ec 886 vdev_t *vd = zio->io_vd;
428870ff 887 spa_t *spa = zio->io_spa;
b128c09f 888 zio_t *rio = zio->io_private;
34dc7c2f 889 uberblock_t *ub = zio->io_data;
9ae529ec 890 struct ubl_cbdata *cbp = rio->io_private;
34dc7c2f 891
9ae529ec 892 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
34dc7c2f
BB
893
894 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
b128c09f 895 mutex_enter(&rio->io_lock);
428870ff 896 if (ub->ub_txg <= spa->spa_load_max_txg &&
9ae529ec
CS
897 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
898 /*
3bc7e0fb
GW
899 * Keep track of the vdev in which this uberblock
900 * was found. We will use this information later
901 * to obtain the config nvlist associated with
9ae529ec
CS
902 * this uberblock.
903 */
904 *cbp->ubl_ubbest = *ub;
905 cbp->ubl_vd = vd;
9ae529ec 906 }
b128c09f 907 mutex_exit(&rio->io_lock);
34dc7c2f
BB
908 }
909
910 zio_buf_free(zio->io_data, zio->io_size);
911}
912
9ae529ec
CS
913static void
914vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
915 struct ubl_cbdata *cbp)
34dc7c2f 916{
d6320ddb 917 int c, l, n;
b128c09f 918
d6320ddb 919 for (c = 0; c < vd->vdev_children; c++)
9ae529ec 920 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
34dc7c2f 921
b128c09f 922 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
d6320ddb
BB
923 for (l = 0; l < VDEV_LABELS; l++) {
924 for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
b128c09f
BB
925 vdev_label_read(zio, vd, l,
926 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
927 VDEV_UBERBLOCK_OFFSET(vd, n),
928 VDEV_UBERBLOCK_SIZE(vd),
929 vdev_uberblock_load_done, zio, flags);
930 }
34dc7c2f
BB
931 }
932 }
9ae529ec
CS
933}
934
935/*
936 * Reads the 'best' uberblock from disk along with its associated
937 * configuration. First, we read the uberblock array of each label of each
938 * vdev, keeping track of the uberblock with the highest txg in each array.
3bc7e0fb 939 * Then, we read the configuration from the same vdev as the best uberblock.
9ae529ec
CS
940 */
941void
942vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
943{
9ae529ec
CS
944 zio_t *zio;
945 spa_t *spa = rvd->vdev_spa;
946 struct ubl_cbdata cb;
947 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
948 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
949
950 ASSERT(ub);
951 ASSERT(config);
b128c09f 952
9ae529ec
CS
953 bzero(ub, sizeof (uberblock_t));
954 *config = NULL;
955
956 cb.ubl_ubbest = ub;
957 cb.ubl_vd = NULL;
958
959 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
960 zio = zio_root(spa, NULL, &cb, flags);
961 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
962 (void) zio_wait(zio);
3bc7e0fb
GW
963
964 /*
965 * It's possible that the best uberblock was discovered on a label
966 * that has a configuration which was written in a future txg.
967 * Search all labels on this vdev to find the configuration that
968 * matches the txg for our uberblock.
969 */
970 if (cb.ubl_vd != NULL)
971 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
9ae529ec 972 spa_config_exit(spa, SCL_ALL, FTAG);
34dc7c2f
BB
973}
974
975/*
976 * On success, increment root zio's count of good writes.
977 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
978 */
979static void
980vdev_uberblock_sync_done(zio_t *zio)
981{
982 uint64_t *good_writes = zio->io_private;
983
984 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
985 atomic_add_64(good_writes, 1);
986}
987
988/*
989 * Write the uberblock to all labels of all leaves of the specified vdev.
990 */
991static void
b128c09f 992vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
34dc7c2f 993{
34dc7c2f 994 uberblock_t *ubbuf;
d6320ddb 995 int c, l, n;
34dc7c2f 996
d6320ddb 997 for (c = 0; c < vd->vdev_children; c++)
b128c09f 998 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
34dc7c2f
BB
999
1000 if (!vd->vdev_ops->vdev_op_leaf)
1001 return;
1002
b128c09f 1003 if (!vdev_writeable(vd))
34dc7c2f
BB
1004 return;
1005
1006 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1007
1008 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1009 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1010 *ubbuf = *ub;
1011
d6320ddb 1012 for (l = 0; l < VDEV_LABELS; l++)
34dc7c2f 1013 vdev_label_write(zio, vd, l, ubbuf,
b128c09f 1014 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
34dc7c2f 1015 vdev_uberblock_sync_done, zio->io_private,
b128c09f 1016 flags | ZIO_FLAG_DONT_PROPAGATE);
34dc7c2f
BB
1017
1018 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1019}
1020
a08ee875 1021/* Sync the uberblocks to all vdevs in svd[] */
34dc7c2f
BB
1022int
1023vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1024{
1025 spa_t *spa = svd[0]->vdev_spa;
34dc7c2f
BB
1026 zio_t *zio;
1027 uint64_t good_writes = 0;
d6320ddb 1028 int v;
34dc7c2f
BB
1029
1030 zio = zio_root(spa, NULL, &good_writes, flags);
1031
d6320ddb 1032 for (v = 0; v < svdcount; v++)
b128c09f 1033 vdev_uberblock_sync(zio, ub, svd[v], flags);
34dc7c2f
BB
1034
1035 (void) zio_wait(zio);
1036
1037 /*
1038 * Flush the uberblocks to disk. This ensures that the odd labels
1039 * are no longer needed (because the new uberblocks and the even
1040 * labels are safely on disk), so it is safe to overwrite them.
1041 */
1042 zio = zio_root(spa, NULL, NULL, flags);
1043
d6320ddb 1044 for (v = 0; v < svdcount; v++)
34dc7c2f
BB
1045 zio_flush(zio, svd[v]);
1046
1047 (void) zio_wait(zio);
1048
1049 return (good_writes >= 1 ? 0 : EIO);
1050}
1051
1052/*
1053 * On success, increment the count of good writes for our top-level vdev.
1054 */
1055static void
1056vdev_label_sync_done(zio_t *zio)
1057{
1058 uint64_t *good_writes = zio->io_private;
1059
1060 if (zio->io_error == 0)
1061 atomic_add_64(good_writes, 1);
1062}
1063
1064/*
1065 * If there weren't enough good writes, indicate failure to the parent.
1066 */
1067static void
1068vdev_label_sync_top_done(zio_t *zio)
1069{
1070 uint64_t *good_writes = zio->io_private;
1071
1072 if (*good_writes == 0)
a08ee875 1073 zio->io_error = SET_ERROR(EIO);
34dc7c2f
BB
1074
1075 kmem_free(good_writes, sizeof (uint64_t));
1076}
1077
b128c09f
BB
1078/*
1079 * We ignore errors for log and cache devices, simply free the private data.
1080 */
1081static void
1082vdev_label_sync_ignore_done(zio_t *zio)
1083{
1084 kmem_free(zio->io_private, sizeof (uint64_t));
1085}
1086
34dc7c2f
BB
1087/*
1088 * Write all even or odd labels to all leaves of the specified vdev.
1089 */
1090static void
b128c09f 1091vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
34dc7c2f
BB
1092{
1093 nvlist_t *label;
1094 vdev_phys_t *vp;
1095 char *buf;
1096 size_t buflen;
d6320ddb 1097 int c;
34dc7c2f 1098
d6320ddb 1099 for (c = 0; c < vd->vdev_children; c++)
b128c09f 1100 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
34dc7c2f
BB
1101
1102 if (!vd->vdev_ops->vdev_op_leaf)
1103 return;
1104
b128c09f 1105 if (!vdev_writeable(vd))
34dc7c2f
BB
1106 return;
1107
1108 /*
1109 * Generate a label describing the top-level config to which we belong.
1110 */
1111 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1112
1113 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1114 bzero(vp, sizeof (vdev_phys_t));
1115
1116 buf = vp->vp_nvlist;
1117 buflen = sizeof (vp->vp_nvlist);
1118
a08ee875 1119 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_PUSHPAGE)) {
34dc7c2f
BB
1120 for (; l < VDEV_LABELS; l += 2) {
1121 vdev_label_write(zio, vd, l, vp,
1122 offsetof(vdev_label_t, vl_vdev_phys),
1123 sizeof (vdev_phys_t),
1124 vdev_label_sync_done, zio->io_private,
b128c09f 1125 flags | ZIO_FLAG_DONT_PROPAGATE);
34dc7c2f
BB
1126 }
1127 }
1128
1129 zio_buf_free(vp, sizeof (vdev_phys_t));
1130 nvlist_free(label);
1131}
1132
1133int
b128c09f 1134vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
34dc7c2f 1135{
b128c09f 1136 list_t *dl = &spa->spa_config_dirty_list;
34dc7c2f
BB
1137 vdev_t *vd;
1138 zio_t *zio;
1139 int error;
1140
1141 /*
1142 * Write the new labels to disk.
1143 */
1144 zio = zio_root(spa, NULL, NULL, flags);
1145
1146 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
d6320ddb
BB
1147 uint64_t *good_writes;
1148 zio_t *vio;
428870ff
BB
1149
1150 ASSERT(!vd->vdev_ishole);
1151
b8d06fca 1152 good_writes = kmem_zalloc(sizeof (uint64_t), KM_PUSHPAGE);
d6320ddb 1153 vio = zio_null(zio, spa, NULL,
b128c09f
BB
1154 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1155 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
34dc7c2f 1156 good_writes, flags);
b128c09f 1157 vdev_label_sync(vio, vd, l, txg, flags);
34dc7c2f
BB
1158 zio_nowait(vio);
1159 }
1160
1161 error = zio_wait(zio);
1162
1163 /*
1164 * Flush the new labels to disk.
1165 */
1166 zio = zio_root(spa, NULL, NULL, flags);
1167
1168 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1169 zio_flush(zio, vd);
1170
1171 (void) zio_wait(zio);
1172
1173 return (error);
1174}
1175
1176/*
1177 * Sync the uberblock and any changes to the vdev configuration.
1178 *
1179 * The order of operations is carefully crafted to ensure that
1180 * if the system panics or loses power at any time, the state on disk
1181 * is still transactionally consistent. The in-line comments below
1182 * describe the failure semantics at each stage.
1183 *
1184 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1185 * at any time, you can just call it again, and it will resume its work.
1186 */
1187int
9babb374 1188vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
34dc7c2f
BB
1189{
1190 spa_t *spa = svd[0]->vdev_spa;
1191 uberblock_t *ub = &spa->spa_uberblock;
1192 vdev_t *vd;
1193 zio_t *zio;
1194 int error;
b128c09f 1195 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
34dc7c2f 1196
9babb374
BB
1197 /*
1198 * Normally, we don't want to try too hard to write every label and
1199 * uberblock. If there is a flaky disk, we don't want the rest of the
1200 * sync process to block while we retry. But if we can't write a
1201 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1202 * bailing out and declaring the pool faulted.
1203 */
1204 if (tryhard)
1205 flags |= ZIO_FLAG_TRYHARD;
1206
34dc7c2f
BB
1207 ASSERT(ub->ub_txg <= txg);
1208
1209 /*
1210 * If this isn't a resync due to I/O errors,
1211 * and nothing changed in this transaction group,
1212 * and the vdev configuration hasn't changed,
1213 * then there's nothing to do.
1214 */
1215 if (ub->ub_txg < txg &&
1216 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
b128c09f 1217 list_is_empty(&spa->spa_config_dirty_list))
34dc7c2f
BB
1218 return (0);
1219
1220 if (txg > spa_freeze_txg(spa))
1221 return (0);
1222
1223 ASSERT(txg <= spa->spa_final_txg);
1224
1225 /*
1226 * Flush the write cache of every disk that's been written to
1227 * in this transaction group. This ensures that all blocks
1228 * written in this txg will be committed to stable storage
1229 * before any uberblock that references them.
1230 */
1231 zio = zio_root(spa, NULL, NULL, flags);
1232
1233 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1234 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1235 zio_flush(zio, vd);
1236
1237 (void) zio_wait(zio);
1238
1239 /*
1240 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1241 * system dies in the middle of this process, that's OK: all of the
1242 * even labels that made it to disk will be newer than any uberblock,
1243 * and will therefore be considered invalid. The odd labels (L1, L3),
1244 * which have not yet been touched, will still be valid. We flush
1245 * the new labels to disk to ensure that all even-label updates
1246 * are committed to stable storage before the uberblock update.
1247 */
b128c09f 1248 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
34dc7c2f
BB
1249 return (error);
1250
1251 /*
1252 * Sync the uberblocks to all vdevs in svd[].
1253 * If the system dies in the middle of this step, there are two cases
1254 * to consider, and the on-disk state is consistent either way:
1255 *
1256 * (1) If none of the new uberblocks made it to disk, then the
1257 * previous uberblock will be the newest, and the odd labels
1258 * (which had not yet been touched) will be valid with respect
1259 * to that uberblock.
1260 *
1261 * (2) If one or more new uberblocks made it to disk, then they
1262 * will be the newest, and the even labels (which had all
1263 * been successfully committed) will be valid with respect
1264 * to the new uberblocks.
1265 */
1266 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1267 return (error);
1268
1269 /*
1270 * Sync out odd labels for every dirty vdev. If the system dies
1271 * in the middle of this process, the even labels and the new
1272 * uberblocks will suffice to open the pool. The next time
1273 * the pool is opened, the first thing we'll do -- before any
1274 * user data is modified -- is mark every vdev dirty so that
1275 * all labels will be brought up to date. We flush the new labels
1276 * to disk to ensure that all odd-label updates are committed to
1277 * stable storage before the next transaction group begins.
1278 */
b128c09f 1279 return (vdev_label_sync_list(spa, 1, txg, flags));
34dc7c2f 1280}