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 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 */
26
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.
126 * features_for_read
127 * An nvlist of the features necessary for reading the MOS.
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>
147 #include <sys/dsl_scan.h>
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 */
154 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)155 vdev_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
164 /*
165 * Returns back the vdev label associated with the passed in offset.
166 */
167 int
vdev_label_number(uint64_t psize,uint64_t offset)168 vdev_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
180 static void
vdev_label_read(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)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)
183 {
184 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
185 SCL_STATE_ALL);
186 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
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,
191 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
192 }
193
194 static void
vdev_label_write(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)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)
197 {
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);
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 */
213 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)214 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
215 vdev_config_flag_t flags)
216 {
217 nvlist_t *nv = NULL;
218
219 nv = fnvlist_alloc();
220
221 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
222 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
223 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
224 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
225
226 if (vd->vdev_path != NULL)
227 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
228
229 if (vd->vdev_devid != NULL)
230 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
231
232 if (vd->vdev_physpath != NULL)
233 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
234 vd->vdev_physpath);
235
236 if (vd->vdev_fru != NULL)
237 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
238
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 ||
248 (vd->vdev_nparity <= 2 &&
249 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
250 (vd->vdev_nparity <= 3 &&
251 spa_version(spa) >= SPA_VERSION_RAIDZ3));
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 */
258 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
259 }
260
261 if (vd->vdev_wholedisk != -1ULL)
262 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
263 vd->vdev_wholedisk);
264
265 if (vd->vdev_not_present)
266 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
267
268 if (vd->vdev_isspare)
269 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
270
271 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
272 vd == vd->vdev_top) {
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);
281 if (vd->vdev_removing)
282 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
283 vd->vdev_removing);
284 }
285
286 if (vd->vdev_dtl_sm != NULL) {
287 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
288 space_map_object(vd->vdev_dtl_sm));
289 }
290
291 if (vd->vdev_crtxg)
292 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
293
294 if (getstats) {
295 vdev_stat_t vs;
296 pool_scan_stat_t ps;
297
298 vdev_get_stats(vd, &vs);
299 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
300 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
301
302 /* provide either current or previous scan information */
303 if (spa_scan_get_stats(spa, &ps) == 0) {
304 fnvlist_add_uint64_array(nv,
305 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
306 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
307 }
308 }
309
310 if (!vd->vdev_ops->vdev_op_leaf) {
311 nvlist_t **child;
312 int c, idx;
313
314 ASSERT(!vd->vdev_ishole);
315
316 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
317 KM_SLEEP);
318
319 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
320 vdev_t *cvd = vd->vdev_child[c];
321
322 /*
323 * If we're generating an nvlist of removing
324 * vdevs then skip over any device which is
325 * not being removed.
326 */
327 if ((flags & VDEV_CONFIG_REMOVING) &&
328 !cvd->vdev_removing)
329 continue;
330
331 child[idx++] = vdev_config_generate(spa, cvd,
332 getstats, flags);
333 }
334
335 if (idx) {
336 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
337 child, idx);
338 }
339
340 for (c = 0; c < idx; c++)
341 nvlist_free(child[c]);
342
343 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
344
345 } else {
346 const char *aux = NULL;
347
348 if (vd->vdev_offline && !vd->vdev_tmpoffline)
349 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
350 if (vd->vdev_resilver_txg != 0)
351 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
352 vd->vdev_resilver_txg);
353 if (vd->vdev_faulted)
354 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
355 if (vd->vdev_degraded)
356 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
357 if (vd->vdev_removed)
358 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
359 if (vd->vdev_unspare)
360 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
361 if (vd->vdev_ishole)
362 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
363
364 switch (vd->vdev_stat.vs_aux) {
365 case VDEV_AUX_ERR_EXCEEDED:
366 aux = "err_exceeded";
367 break;
368
369 case VDEV_AUX_EXTERNAL:
370 aux = "external";
371 break;
372 }
373
374 if (aux != NULL)
375 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
376
377 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
378 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
379 vd->vdev_orig_guid);
380 }
381 }
382
383 return (nv);
384 }
385
386 /*
387 * Generate a view of the top-level vdevs. If we currently have holes
388 * in the namespace, then generate an array which contains a list of holey
389 * vdevs. Additionally, add the number of top-level children that currently
390 * exist.
391 */
392 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)393 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
394 {
395 vdev_t *rvd = spa->spa_root_vdev;
396 uint64_t *array;
397 uint_t c, idx;
398
399 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
400
401 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
402 vdev_t *tvd = rvd->vdev_child[c];
403
404 if (tvd->vdev_ishole)
405 array[idx++] = c;
406 }
407
408 if (idx) {
409 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
410 array, idx) == 0);
411 }
412
413 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
414 rvd->vdev_children) == 0);
415
416 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
417 }
418
419 /*
420 * Returns the configuration from the label of the given vdev. For vdevs
421 * which don't have a txg value stored on their label (i.e. spares/cache)
422 * or have not been completely initialized (txg = 0) just return
423 * the configuration from the first valid label we find. Otherwise,
424 * find the most up-to-date label that does not exceed the specified
425 * 'txg' value.
426 */
427 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)428 vdev_label_read_config(vdev_t *vd, uint64_t txg)
429 {
430 spa_t *spa = vd->vdev_spa;
431 nvlist_t *config = NULL;
432 vdev_phys_t *vp;
433 zio_t *zio;
434 uint64_t best_txg = 0;
435 int error = 0;
436 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
437 ZIO_FLAG_SPECULATIVE;
438
439 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
440
441 if (!vdev_readable(vd))
442 return (NULL);
443
444 vp = zio_buf_alloc(sizeof (vdev_phys_t));
445
446 retry:
447 for (int l = 0; l < VDEV_LABELS; l++) {
448 nvlist_t *label = NULL;
449
450 zio = zio_root(spa, NULL, NULL, flags);
451
452 vdev_label_read(zio, vd, l, vp,
453 offsetof(vdev_label_t, vl_vdev_phys),
454 sizeof (vdev_phys_t), NULL, NULL, flags);
455
456 if (zio_wait(zio) == 0 &&
457 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
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 }
479
480 if (label != NULL) {
481 nvlist_free(label);
482 label = NULL;
483 }
484 }
485
486 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
487 flags |= ZIO_FLAG_TRYHARD;
488 goto retry;
489 }
490
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 */
500 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)501 vdev_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 */
517 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
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) &&
549 !spa_spare_exists(device_guid, NULL, NULL) &&
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 */
569 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
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));
585 }
586 }
587
588 /*
589 * Check to see if this is an l2cache device.
590 */
591 if (spa_l2cache_exists(device_guid, NULL))
592 return (B_TRUE);
593
594 /*
595 * We can't rely on a pool's state if it's been imported
596 * read-only. Instead we look to see if the pools is marked
597 * read-only in the namespace and set the state to active.
598 */
599 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
600 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
601 spa_mode(spa) == FREAD)
602 state = POOL_STATE_ACTIVE;
603
604 /*
605 * If the device is marked ACTIVE, then this device is in use by another
606 * pool on the system.
607 */
608 return (state == POOL_STATE_ACTIVE);
609 }
610
611 /*
612 * Initialize a vdev label. We check to make sure each leaf device is not in
613 * use, and writable. We put down an initial label which we will later
614 * overwrite with a complete label. Note that it's important to do this
615 * sequentially, not in parallel, so that we catch cases of multiple use of the
616 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
617 * itself.
618 */
619 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)620 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
621 {
622 spa_t *spa = vd->vdev_spa;
623 nvlist_t *label;
624 vdev_phys_t *vp;
625 char *pad2;
626 uberblock_t *ub;
627 zio_t *zio;
628 char *buf;
629 size_t buflen;
630 int error;
631 uint64_t spare_guid, l2cache_guid;
632 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
633
634 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
635
636 for (int c = 0; c < vd->vdev_children; c++)
637 if ((error = vdev_label_init(vd->vdev_child[c],
638 crtxg, reason)) != 0)
639 return (error);
640
641 /* Track the creation time for this vdev */
642 vd->vdev_crtxg = crtxg;
643
644 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
645 return (0);
646
647 /*
648 * Dead vdevs cannot be initialized.
649 */
650 if (vdev_is_dead(vd))
651 return (SET_ERROR(EIO));
652
653 /*
654 * Determine if the vdev is in use.
655 */
656 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
657 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
658 return (SET_ERROR(EBUSY));
659
660 /*
661 * If this is a request to add or replace a spare or l2cache device
662 * that is in use elsewhere on the system, then we must update the
663 * guid (which was initialized to a random value) to reflect the
664 * actual GUID (which is shared between multiple pools).
665 */
666 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
667 spare_guid != 0ULL) {
668 uint64_t guid_delta = spare_guid - vd->vdev_guid;
669
670 vd->vdev_guid += guid_delta;
671
672 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
673 pvd->vdev_guid_sum += guid_delta;
674
675 /*
676 * If this is a replacement, then we want to fallthrough to the
677 * rest of the code. If we're adding a spare, then it's already
678 * labeled appropriately and we can just return.
679 */
680 if (reason == VDEV_LABEL_SPARE)
681 return (0);
682 ASSERT(reason == VDEV_LABEL_REPLACE ||
683 reason == VDEV_LABEL_SPLIT);
684 }
685
686 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
687 l2cache_guid != 0ULL) {
688 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
689
690 vd->vdev_guid += guid_delta;
691
692 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
693 pvd->vdev_guid_sum += guid_delta;
694
695 /*
696 * If this is a replacement, then we want to fallthrough to the
697 * rest of the code. If we're adding an l2cache, then it's
698 * already labeled appropriately and we can just return.
699 */
700 if (reason == VDEV_LABEL_L2CACHE)
701 return (0);
702 ASSERT(reason == VDEV_LABEL_REPLACE);
703 }
704
705 /*
706 * Initialize its label.
707 */
708 vp = zio_buf_alloc(sizeof (vdev_phys_t));
709 bzero(vp, sizeof (vdev_phys_t));
710
711 /*
712 * Generate a label describing the pool and our top-level vdev.
713 * We mark it as being from txg 0 to indicate that it's not
714 * really part of an active pool just yet. The labels will
715 * be written again with a meaningful txg by spa_sync().
716 */
717 if (reason == VDEV_LABEL_SPARE ||
718 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
719 /*
720 * For inactive hot spares, we generate a special label that
721 * identifies as a mutually shared hot spare. We write the
722 * label if we are adding a hot spare, or if we are removing an
723 * active hot spare (in which case we want to revert the
724 * labels).
725 */
726 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
727
728 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
729 spa_version(spa)) == 0);
730 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
731 POOL_STATE_SPARE) == 0);
732 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
733 vd->vdev_guid) == 0);
734 } else if (reason == VDEV_LABEL_L2CACHE ||
735 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
736 /*
737 * For level 2 ARC devices, add a special label.
738 */
739 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
740
741 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
742 spa_version(spa)) == 0);
743 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
744 POOL_STATE_L2CACHE) == 0);
745 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
746 vd->vdev_guid) == 0);
747 } else {
748 uint64_t txg = 0ULL;
749
750 if (reason == VDEV_LABEL_SPLIT)
751 txg = spa->spa_uberblock.ub_txg;
752 label = spa_config_generate(spa, vd, txg, B_FALSE);
753
754 /*
755 * Add our creation time. This allows us to detect multiple
756 * vdev uses as described above, and automatically expires if we
757 * fail.
758 */
759 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
760 crtxg) == 0);
761 }
762
763 buf = vp->vp_nvlist;
764 buflen = sizeof (vp->vp_nvlist);
765
766 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
767 if (error != 0) {
768 nvlist_free(label);
769 zio_buf_free(vp, sizeof (vdev_phys_t));
770 /* EFAULT means nvlist_pack ran out of room */
771 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
772 }
773
774 /*
775 * Initialize uberblock template.
776 */
777 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
778 bzero(ub, VDEV_UBERBLOCK_RING);
779 *ub = spa->spa_uberblock;
780 ub->ub_txg = 0;
781
782 /* Initialize the 2nd padding area. */
783 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
784 bzero(pad2, VDEV_PAD_SIZE);
785
786 /*
787 * Write everything in parallel.
788 */
789 retry:
790 zio = zio_root(spa, NULL, NULL, flags);
791
792 for (int l = 0; l < VDEV_LABELS; l++) {
793
794 vdev_label_write(zio, vd, l, vp,
795 offsetof(vdev_label_t, vl_vdev_phys),
796 sizeof (vdev_phys_t), NULL, NULL, flags);
797
798 /*
799 * Skip the 1st padding area.
800 * Zero out the 2nd padding area where it might have
801 * left over data from previous filesystem format.
802 */
803 vdev_label_write(zio, vd, l, pad2,
804 offsetof(vdev_label_t, vl_pad2),
805 VDEV_PAD_SIZE, NULL, NULL, flags);
806
807 vdev_label_write(zio, vd, l, ub,
808 offsetof(vdev_label_t, vl_uberblock),
809 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
810 }
811
812 error = zio_wait(zio);
813
814 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
815 flags |= ZIO_FLAG_TRYHARD;
816 goto retry;
817 }
818
819 nvlist_free(label);
820 zio_buf_free(pad2, VDEV_PAD_SIZE);
821 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
822 zio_buf_free(vp, sizeof (vdev_phys_t));
823
824 /*
825 * If this vdev hasn't been previously identified as a spare, then we
826 * mark it as such only if a) we are labeling it as a spare, or b) it
827 * exists as a spare elsewhere in the system. Do the same for
828 * level 2 ARC devices.
829 */
830 if (error == 0 && !vd->vdev_isspare &&
831 (reason == VDEV_LABEL_SPARE ||
832 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
833 spa_spare_add(vd);
834
835 if (error == 0 && !vd->vdev_isl2cache &&
836 (reason == VDEV_LABEL_L2CACHE ||
837 spa_l2cache_exists(vd->vdev_guid, NULL)))
838 spa_l2cache_add(vd);
839
840 return (error);
841 }
842
843 /*
844 * ==========================================================================
845 * uberblock load/sync
846 * ==========================================================================
847 */
848
849 /*
850 * Consider the following situation: txg is safely synced to disk. We've
851 * written the first uberblock for txg + 1, and then we lose power. When we
852 * come back up, we fail to see the uberblock for txg + 1 because, say,
853 * it was on a mirrored device and the replica to which we wrote txg + 1
854 * is now offline. If we then make some changes and sync txg + 1, and then
855 * the missing replica comes back, then for a few seconds we'll have two
856 * conflicting uberblocks on disk with the same txg. The solution is simple:
857 * among uberblocks with equal txg, choose the one with the latest timestamp.
858 */
859 static int
vdev_uberblock_compare(uberblock_t * ub1,uberblock_t * ub2)860 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
861 {
862 if (ub1->ub_txg < ub2->ub_txg)
863 return (-1);
864 if (ub1->ub_txg > ub2->ub_txg)
865 return (1);
866
867 if (ub1->ub_timestamp < ub2->ub_timestamp)
868 return (-1);
869 if (ub1->ub_timestamp > ub2->ub_timestamp)
870 return (1);
871
872 return (0);
873 }
874
875 struct ubl_cbdata {
876 uberblock_t *ubl_ubbest; /* Best uberblock */
877 vdev_t *ubl_vd; /* vdev associated with the above */
878 };
879
880 static void
vdev_uberblock_load_done(zio_t * zio)881 vdev_uberblock_load_done(zio_t *zio)
882 {
883 vdev_t *vd = zio->io_vd;
884 spa_t *spa = zio->io_spa;
885 zio_t *rio = zio->io_private;
886 uberblock_t *ub = zio->io_data;
887 struct ubl_cbdata *cbp = rio->io_private;
888
889 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
890
891 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
892 mutex_enter(&rio->io_lock);
893 if (ub->ub_txg <= spa->spa_load_max_txg &&
894 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
895 /*
896 * Keep track of the vdev in which this uberblock
897 * was found. We will use this information later
898 * to obtain the config nvlist associated with
899 * this uberblock.
900 */
901 *cbp->ubl_ubbest = *ub;
902 cbp->ubl_vd = vd;
903 }
904 mutex_exit(&rio->io_lock);
905 }
906
907 zio_buf_free(zio->io_data, zio->io_size);
908 }
909
910 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)911 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
912 struct ubl_cbdata *cbp)
913 {
914 for (int c = 0; c < vd->vdev_children; c++)
915 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
916
917 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
918 for (int l = 0; l < VDEV_LABELS; l++) {
919 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
920 vdev_label_read(zio, vd, l,
921 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
922 VDEV_UBERBLOCK_OFFSET(vd, n),
923 VDEV_UBERBLOCK_SIZE(vd),
924 vdev_uberblock_load_done, zio, flags);
925 }
926 }
927 }
928 }
929
930 /*
931 * Reads the 'best' uberblock from disk along with its associated
932 * configuration. First, we read the uberblock array of each label of each
933 * vdev, keeping track of the uberblock with the highest txg in each array.
934 * Then, we read the configuration from the same vdev as the best uberblock.
935 */
936 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)937 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
938 {
939 zio_t *zio;
940 spa_t *spa = rvd->vdev_spa;
941 struct ubl_cbdata cb;
942 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
943 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
944
945 ASSERT(ub);
946 ASSERT(config);
947
948 bzero(ub, sizeof (uberblock_t));
949 *config = NULL;
950
951 cb.ubl_ubbest = ub;
952 cb.ubl_vd = NULL;
953
954 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
955 zio = zio_root(spa, NULL, &cb, flags);
956 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
957 (void) zio_wait(zio);
958
959 /*
960 * It's possible that the best uberblock was discovered on a label
961 * that has a configuration which was written in a future txg.
962 * Search all labels on this vdev to find the configuration that
963 * matches the txg for our uberblock.
964 */
965 if (cb.ubl_vd != NULL)
966 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
967 spa_config_exit(spa, SCL_ALL, FTAG);
968 }
969
970 /*
971 * On success, increment root zio's count of good writes.
972 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
973 */
974 static void
vdev_uberblock_sync_done(zio_t * zio)975 vdev_uberblock_sync_done(zio_t *zio)
976 {
977 uint64_t *good_writes = zio->io_private;
978
979 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
980 atomic_inc_64(good_writes);
981 }
982
983 /*
984 * Write the uberblock to all labels of all leaves of the specified vdev.
985 */
986 static void
vdev_uberblock_sync(zio_t * zio,uberblock_t * ub,vdev_t * vd,int flags)987 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
988 {
989 uberblock_t *ubbuf;
990 int n;
991
992 for (int c = 0; c < vd->vdev_children; c++)
993 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
994
995 if (!vd->vdev_ops->vdev_op_leaf)
996 return;
997
998 if (!vdev_writeable(vd))
999 return;
1000
1001 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1002
1003 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1004 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1005 *ubbuf = *ub;
1006
1007 for (int l = 0; l < VDEV_LABELS; l++)
1008 vdev_label_write(zio, vd, l, ubbuf,
1009 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1010 vdev_uberblock_sync_done, zio->io_private,
1011 flags | ZIO_FLAG_DONT_PROPAGATE);
1012
1013 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1014 }
1015
1016 /* Sync the uberblocks to all vdevs in svd[] */
1017 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1018 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1019 {
1020 spa_t *spa = svd[0]->vdev_spa;
1021 zio_t *zio;
1022 uint64_t good_writes = 0;
1023
1024 zio = zio_root(spa, NULL, &good_writes, flags);
1025
1026 for (int v = 0; v < svdcount; v++)
1027 vdev_uberblock_sync(zio, ub, svd[v], flags);
1028
1029 (void) zio_wait(zio);
1030
1031 /*
1032 * Flush the uberblocks to disk. This ensures that the odd labels
1033 * are no longer needed (because the new uberblocks and the even
1034 * labels are safely on disk), so it is safe to overwrite them.
1035 */
1036 zio = zio_root(spa, NULL, NULL, flags);
1037
1038 for (int v = 0; v < svdcount; v++)
1039 zio_flush(zio, svd[v]);
1040
1041 (void) zio_wait(zio);
1042
1043 return (good_writes >= 1 ? 0 : EIO);
1044 }
1045
1046 /*
1047 * On success, increment the count of good writes for our top-level vdev.
1048 */
1049 static void
vdev_label_sync_done(zio_t * zio)1050 vdev_label_sync_done(zio_t *zio)
1051 {
1052 uint64_t *good_writes = zio->io_private;
1053
1054 if (zio->io_error == 0)
1055 atomic_inc_64(good_writes);
1056 }
1057
1058 /*
1059 * If there weren't enough good writes, indicate failure to the parent.
1060 */
1061 static void
vdev_label_sync_top_done(zio_t * zio)1062 vdev_label_sync_top_done(zio_t *zio)
1063 {
1064 uint64_t *good_writes = zio->io_private;
1065
1066 if (*good_writes == 0)
1067 zio->io_error = SET_ERROR(EIO);
1068
1069 kmem_free(good_writes, sizeof (uint64_t));
1070 }
1071
1072 /*
1073 * We ignore errors for log and cache devices, simply free the private data.
1074 */
1075 static void
vdev_label_sync_ignore_done(zio_t * zio)1076 vdev_label_sync_ignore_done(zio_t *zio)
1077 {
1078 kmem_free(zio->io_private, sizeof (uint64_t));
1079 }
1080
1081 /*
1082 * Write all even or odd labels to all leaves of the specified vdev.
1083 */
1084 static void
vdev_label_sync(zio_t * zio,vdev_t * vd,int l,uint64_t txg,int flags)1085 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1086 {
1087 nvlist_t *label;
1088 vdev_phys_t *vp;
1089 char *buf;
1090 size_t buflen;
1091
1092 for (int c = 0; c < vd->vdev_children; c++)
1093 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1094
1095 if (!vd->vdev_ops->vdev_op_leaf)
1096 return;
1097
1098 if (!vdev_writeable(vd))
1099 return;
1100
1101 /*
1102 * Generate a label describing the top-level config to which we belong.
1103 */
1104 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1105
1106 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1107 bzero(vp, sizeof (vdev_phys_t));
1108
1109 buf = vp->vp_nvlist;
1110 buflen = sizeof (vp->vp_nvlist);
1111
1112 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1113 for (; l < VDEV_LABELS; l += 2) {
1114 vdev_label_write(zio, vd, l, vp,
1115 offsetof(vdev_label_t, vl_vdev_phys),
1116 sizeof (vdev_phys_t),
1117 vdev_label_sync_done, zio->io_private,
1118 flags | ZIO_FLAG_DONT_PROPAGATE);
1119 }
1120 }
1121
1122 zio_buf_free(vp, sizeof (vdev_phys_t));
1123 nvlist_free(label);
1124 }
1125
1126 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1127 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1128 {
1129 list_t *dl = &spa->spa_config_dirty_list;
1130 vdev_t *vd;
1131 zio_t *zio;
1132 int error;
1133
1134 /*
1135 * Write the new labels to disk.
1136 */
1137 zio = zio_root(spa, NULL, NULL, flags);
1138
1139 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1140 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1141 KM_SLEEP);
1142
1143 ASSERT(!vd->vdev_ishole);
1144
1145 zio_t *vio = zio_null(zio, spa, NULL,
1146 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1147 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1148 good_writes, flags);
1149 vdev_label_sync(vio, vd, l, txg, flags);
1150 zio_nowait(vio);
1151 }
1152
1153 error = zio_wait(zio);
1154
1155 /*
1156 * Flush the new labels to disk.
1157 */
1158 zio = zio_root(spa, NULL, NULL, flags);
1159
1160 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1161 zio_flush(zio, vd);
1162
1163 (void) zio_wait(zio);
1164
1165 return (error);
1166 }
1167
1168 /*
1169 * Sync the uberblock and any changes to the vdev configuration.
1170 *
1171 * The order of operations is carefully crafted to ensure that
1172 * if the system panics or loses power at any time, the state on disk
1173 * is still transactionally consistent. The in-line comments below
1174 * describe the failure semantics at each stage.
1175 *
1176 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1177 * at any time, you can just call it again, and it will resume its work.
1178 */
1179 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)1180 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1181 {
1182 spa_t *spa = svd[0]->vdev_spa;
1183 uberblock_t *ub = &spa->spa_uberblock;
1184 vdev_t *vd;
1185 zio_t *zio;
1186 int error = 0;
1187 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1188
1189 retry:
1190 /*
1191 * Normally, we don't want to try too hard to write every label and
1192 * uberblock. If there is a flaky disk, we don't want the rest of the
1193 * sync process to block while we retry. But if we can't write a
1194 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1195 * bailing out and declaring the pool faulted.
1196 */
1197 if (error != 0) {
1198 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1199 return (error);
1200 flags |= ZIO_FLAG_TRYHARD;
1201 }
1202
1203 ASSERT(ub->ub_txg <= txg);
1204
1205 /*
1206 * If this isn't a resync due to I/O errors,
1207 * and nothing changed in this transaction group,
1208 * and the vdev configuration hasn't changed,
1209 * then there's nothing to do.
1210 */
1211 if (ub->ub_txg < txg &&
1212 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1213 list_is_empty(&spa->spa_config_dirty_list))
1214 return (0);
1215
1216 if (txg > spa_freeze_txg(spa))
1217 return (0);
1218
1219 ASSERT(txg <= spa->spa_final_txg);
1220
1221 /*
1222 * Flush the write cache of every disk that's been written to
1223 * in this transaction group. This ensures that all blocks
1224 * written in this txg will be committed to stable storage
1225 * before any uberblock that references them.
1226 */
1227 zio = zio_root(spa, NULL, NULL, flags);
1228
1229 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1230 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1231 zio_flush(zio, vd);
1232
1233 (void) zio_wait(zio);
1234
1235 /*
1236 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1237 * system dies in the middle of this process, that's OK: all of the
1238 * even labels that made it to disk will be newer than any uberblock,
1239 * and will therefore be considered invalid. The odd labels (L1, L3),
1240 * which have not yet been touched, will still be valid. We flush
1241 * the new labels to disk to ensure that all even-label updates
1242 * are committed to stable storage before the uberblock update.
1243 */
1244 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1245 goto retry;
1246
1247 /*
1248 * Sync the uberblocks to all vdevs in svd[].
1249 * If the system dies in the middle of this step, there are two cases
1250 * to consider, and the on-disk state is consistent either way:
1251 *
1252 * (1) If none of the new uberblocks made it to disk, then the
1253 * previous uberblock will be the newest, and the odd labels
1254 * (which had not yet been touched) will be valid with respect
1255 * to that uberblock.
1256 *
1257 * (2) If one or more new uberblocks made it to disk, then they
1258 * will be the newest, and the even labels (which had all
1259 * been successfully committed) will be valid with respect
1260 * to the new uberblocks.
1261 */
1262 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1263 goto retry;
1264
1265 /*
1266 * Sync out odd labels for every dirty vdev. If the system dies
1267 * in the middle of this process, the even labels and the new
1268 * uberblocks will suffice to open the pool. The next time
1269 * the pool is opened, the first thing we'll do -- before any
1270 * user data is modified -- is mark every vdev dirty so that
1271 * all labels will be brought up to date. We flush the new labels
1272 * to disk to ensure that all odd-label updates are committed to
1273 * stable storage before the next transaction group begins.
1274 */
1275 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1276 goto retry;
1277
1278 return (0);
1279 }
1280