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 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
26 */
27
28 /*
29 * Virtual Device Labels
30 * ---------------------
31 *
32 * The vdev label serves several distinct purposes:
33 *
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
36 *
37 * 2. Verify that all the devices given in a configuration are present
38 * within the pool.
39 *
40 * 3. Determine the uberblock for the pool.
41 *
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
44 *
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
48 *
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
53 *
54 *
55 * Label Organization
56 * ------------------
57 *
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
60 *
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transaction groups:
66 *
67 * L1 UB L2
68 * +------+ +------+ +------+
69 * | | | | | |
70 * | t10 | | t10 | | t10 |
71 * | | | | | |
72 * +------+ +------+ +------+
73 *
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
77 *
78 * In order to identify which labels are valid, the labels are written in the
79 * following manner:
80 *
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
84 *
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
91 *
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
96 * on another vdev.
97 *
98 *
99 * On-disk Format
100 * --------------
101 *
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
105 *
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
109 *
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
114 *
115 *
116 * Configuration Information
117 * -------------------------
118 *
119 * The nvlist describing the pool and vdev contains the following elements:
120 *
121 * version ZFS on-disk version
122 * name Pool name
123 * state Pool state
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
127 * features_for_read
128 * An nvlist of the features necessary for reading the MOS.
129 *
130 * Each leaf device label also contains the following:
131 *
132 * top_guid Unique ID for top-level vdev in which this is contained
133 * guid Unique ID for the leaf vdev
134 *
135 * The 'vs' configuration follows the format described in 'spa_config.c'.
136 */
137
138 #include <sys/zfs_context.h>
139 #include <sys/spa.h>
140 #include <sys/spa_impl.h>
141 #include <sys/dmu.h>
142 #include <sys/zap.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/uberblock_impl.h>
146 #include <sys/metaslab.h>
147 #include <sys/zio.h>
148 #include <sys/dsl_scan.h>
149 #include <sys/fs/zfs.h>
150
151 /*
152 * Basic routines to read and write from a vdev label.
153 * Used throughout the rest of this file.
154 */
155 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)156 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
157 {
158 ASSERT(offset < sizeof (vdev_label_t));
159 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
160
161 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
162 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
163 }
164
165 /*
166 * Returns back the vdev label associated with the passed in offset.
167 */
168 int
vdev_label_number(uint64_t psize,uint64_t offset)169 vdev_label_number(uint64_t psize, uint64_t offset)
170 {
171 int l;
172
173 if (offset >= psize - VDEV_LABEL_END_SIZE) {
174 offset -= psize - VDEV_LABEL_END_SIZE;
175 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
176 }
177 l = offset / sizeof (vdev_label_t);
178 return (l < VDEV_LABELS ? l : -1);
179 }
180
181 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)182 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
183 uint64_t size, zio_done_func_t *done, void *private, int flags)
184 {
185 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
186 SCL_STATE_ALL);
187 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
188
189 zio_nowait(zio_read_phys(zio, vd,
190 vdev_label_offset(vd->vdev_psize, l, offset),
191 size, buf, ZIO_CHECKSUM_LABEL, done, private,
192 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
193 }
194
195 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)196 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
197 uint64_t size, zio_done_func_t *done, void *private, int flags)
198 {
199 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
200 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
201 (SCL_CONFIG | SCL_STATE) &&
202 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
203 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
204
205 zio_nowait(zio_write_phys(zio, vd,
206 vdev_label_offset(vd->vdev_psize, l, offset),
207 size, buf, ZIO_CHECKSUM_LABEL, done, private,
208 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
209 }
210
211 /*
212 * Generate the nvlist representing this vdev's config.
213 */
214 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)215 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
216 vdev_config_flag_t flags)
217 {
218 nvlist_t *nv = NULL;
219
220 nv = fnvlist_alloc();
221
222 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
223 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
224 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
225 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
226
227 if (vd->vdev_path != NULL)
228 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
229
230 if (vd->vdev_devid != NULL)
231 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
232
233 if (vd->vdev_physpath != NULL)
234 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
235 vd->vdev_physpath);
236
237 if (vd->vdev_fru != NULL)
238 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
239
240 if (vd->vdev_nparity != 0) {
241 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
242 VDEV_TYPE_RAIDZ) == 0);
243
244 /*
245 * Make sure someone hasn't managed to sneak a fancy new vdev
246 * into a crufty old storage pool.
247 */
248 ASSERT(vd->vdev_nparity == 1 ||
249 (vd->vdev_nparity <= 2 &&
250 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
251 (vd->vdev_nparity <= 3 &&
252 spa_version(spa) >= SPA_VERSION_RAIDZ3));
253
254 /*
255 * Note that we'll add the nparity tag even on storage pools
256 * that only support a single parity device -- older software
257 * will just ignore it.
258 */
259 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
260 }
261
262 if (vd->vdev_wholedisk != -1ULL)
263 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
264 vd->vdev_wholedisk);
265
266 if (vd->vdev_not_present)
267 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
268
269 if (vd->vdev_isspare)
270 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
271
272 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
273 vd == vd->vdev_top) {
274 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
275 vd->vdev_ms_array);
276 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
277 vd->vdev_ms_shift);
278 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
279 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
280 vd->vdev_asize);
281 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
282 if (vd->vdev_removing)
283 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
284 vd->vdev_removing);
285 }
286
287 if (flags & VDEV_CONFIG_L2CACHE)
288 /* indicate that we support L2ARC persistency */
289 VERIFY(nvlist_add_boolean_value(nv,
290 ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
291
292 if (vd->vdev_dtl_sm != NULL) {
293 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
294 space_map_object(vd->vdev_dtl_sm));
295 }
296
297 if (vd->vdev_crtxg)
298 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
299
300 if (getstats) {
301 vdev_stat_t vs;
302 pool_scan_stat_t ps;
303
304 vdev_get_stats(vd, &vs);
305 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
306 (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
307
308 /* provide either current or previous scan information */
309 if (spa_scan_get_stats(spa, &ps) == 0) {
310 fnvlist_add_uint64_array(nv,
311 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
312 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
313 }
314 }
315
316 if (!vd->vdev_ops->vdev_op_leaf) {
317 nvlist_t **child;
318 int c, idx;
319
320 ASSERT(!vd->vdev_ishole);
321
322 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
323 KM_SLEEP);
324
325 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
326 vdev_t *cvd = vd->vdev_child[c];
327
328 /*
329 * If we're generating an nvlist of removing
330 * vdevs then skip over any device which is
331 * not being removed.
332 */
333 if ((flags & VDEV_CONFIG_REMOVING) &&
334 !cvd->vdev_removing)
335 continue;
336
337 child[idx++] = vdev_config_generate(spa, cvd,
338 getstats, flags);
339 }
340
341 if (idx) {
342 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
343 child, idx);
344 }
345
346 for (c = 0; c < idx; c++)
347 nvlist_free(child[c]);
348
349 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
350
351 } else {
352 const char *aux = NULL;
353
354 if (vd->vdev_offline && !vd->vdev_tmpoffline)
355 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
356 if (vd->vdev_resilver_txg != 0)
357 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
358 vd->vdev_resilver_txg);
359 if (vd->vdev_faulted)
360 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
361 if (vd->vdev_degraded)
362 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
363 if (vd->vdev_removed)
364 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
365 if (vd->vdev_unspare)
366 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
367 if (vd->vdev_ishole)
368 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
369
370 switch (vd->vdev_stat.vs_aux) {
371 case VDEV_AUX_ERR_EXCEEDED:
372 aux = "err_exceeded";
373 break;
374
375 case VDEV_AUX_EXTERNAL:
376 aux = "external";
377 break;
378 }
379
380 if (aux != NULL)
381 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
382
383 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
384 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
385 vd->vdev_orig_guid);
386 }
387 }
388
389 return (nv);
390 }
391
392 /*
393 * Generate a view of the top-level vdevs. If we currently have holes
394 * in the namespace, then generate an array which contains a list of holey
395 * vdevs. Additionally, add the number of top-level children that currently
396 * exist.
397 */
398 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)399 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
400 {
401 vdev_t *rvd = spa->spa_root_vdev;
402 uint64_t *array;
403 uint_t c, idx;
404
405 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
406
407 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
408 vdev_t *tvd = rvd->vdev_child[c];
409
410 if (tvd->vdev_ishole)
411 array[idx++] = c;
412 }
413
414 if (idx) {
415 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
416 array, idx) == 0);
417 }
418
419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
420 rvd->vdev_children) == 0);
421
422 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
423 }
424
425 /*
426 * Returns the configuration from the label of the given vdev. For vdevs
427 * which don't have a txg value stored on their label (i.e. spares/cache)
428 * or have not been completely initialized (txg = 0) just return
429 * the configuration from the first valid label we find. Otherwise,
430 * find the most up-to-date label that does not exceed the specified
431 * 'txg' value.
432 */
433 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)434 vdev_label_read_config(vdev_t *vd, uint64_t txg)
435 {
436 spa_t *spa = vd->vdev_spa;
437 nvlist_t *config = NULL;
438 vdev_phys_t *vp;
439 zio_t *zio;
440 uint64_t best_txg = 0;
441 int error = 0;
442 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
443 ZIO_FLAG_SPECULATIVE;
444
445 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
446
447 if (!vdev_readable(vd))
448 return (NULL);
449
450 vp = zio_buf_alloc(sizeof (vdev_phys_t));
451
452 retry:
453 for (int l = 0; l < VDEV_LABELS; l++) {
454 nvlist_t *label = NULL;
455
456 zio = zio_root(spa, NULL, NULL, flags);
457
458 vdev_label_read(zio, vd, l, vp,
459 offsetof(vdev_label_t, vl_vdev_phys),
460 sizeof (vdev_phys_t), NULL, NULL, flags);
461
462 if (zio_wait(zio) == 0 &&
463 nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
464 &label, 0) == 0) {
465 uint64_t label_txg = 0;
466
467 /*
468 * Auxiliary vdevs won't have txg values in their
469 * labels and newly added vdevs may not have been
470 * completely initialized so just return the
471 * configuration from the first valid label we
472 * encounter.
473 */
474 error = nvlist_lookup_uint64(label,
475 ZPOOL_CONFIG_POOL_TXG, &label_txg);
476 if ((error || label_txg == 0) && !config) {
477 config = label;
478 break;
479 } else if (label_txg <= txg && label_txg > best_txg) {
480 best_txg = label_txg;
481 nvlist_free(config);
482 config = fnvlist_dup(label);
483 }
484 }
485
486 if (label != NULL) {
487 nvlist_free(label);
488 label = NULL;
489 }
490 }
491
492 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
493 flags |= ZIO_FLAG_TRYHARD;
494 goto retry;
495 }
496
497 zio_buf_free(vp, sizeof (vdev_phys_t));
498
499 return (config);
500 }
501
502 /*
503 * Determine if a device is in use. The 'spare_guid' parameter will be filled
504 * in with the device guid if this spare is active elsewhere on the system.
505 */
506 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)507 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
508 uint64_t *spare_guid, uint64_t *l2cache_guid)
509 {
510 spa_t *spa = vd->vdev_spa;
511 uint64_t state, pool_guid, device_guid, txg, spare_pool;
512 uint64_t vdtxg = 0;
513 nvlist_t *label;
514
515 if (spare_guid)
516 *spare_guid = 0ULL;
517 if (l2cache_guid)
518 *l2cache_guid = 0ULL;
519
520 /*
521 * Read the label, if any, and perform some basic sanity checks.
522 */
523 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
524 return (B_FALSE);
525
526 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
527 &vdtxg);
528
529 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
530 &state) != 0 ||
531 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
532 &device_guid) != 0) {
533 nvlist_free(label);
534 return (B_FALSE);
535 }
536
537 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
538 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
539 &pool_guid) != 0 ||
540 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
541 &txg) != 0)) {
542 nvlist_free(label);
543 return (B_FALSE);
544 }
545
546 nvlist_free(label);
547
548 /*
549 * Check to see if this device indeed belongs to the pool it claims to
550 * be a part of. The only way this is allowed is if the device is a hot
551 * spare (which we check for later on).
552 */
553 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
554 !spa_guid_exists(pool_guid, device_guid) &&
555 !spa_spare_exists(device_guid, NULL, NULL) &&
556 !spa_l2cache_exists(device_guid, NULL))
557 return (B_FALSE);
558
559 /*
560 * If the transaction group is zero, then this an initialized (but
561 * unused) label. This is only an error if the create transaction
562 * on-disk is the same as the one we're using now, in which case the
563 * user has attempted to add the same vdev multiple times in the same
564 * transaction.
565 */
566 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
567 txg == 0 && vdtxg == crtxg)
568 return (B_TRUE);
569
570 /*
571 * Check to see if this is a spare device. We do an explicit check for
572 * spa_has_spare() here because it may be on our pending list of spares
573 * to add. We also check if it is an l2cache device.
574 */
575 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
576 spa_has_spare(spa, device_guid)) {
577 if (spare_guid)
578 *spare_guid = device_guid;
579
580 switch (reason) {
581 case VDEV_LABEL_CREATE:
582 case VDEV_LABEL_L2CACHE:
583 return (B_TRUE);
584
585 case VDEV_LABEL_REPLACE:
586 return (!spa_has_spare(spa, device_guid) ||
587 spare_pool != 0ULL);
588
589 case VDEV_LABEL_SPARE:
590 return (spa_has_spare(spa, device_guid));
591 }
592 }
593
594 /*
595 * Check to see if this is an l2cache device.
596 */
597 if (spa_l2cache_exists(device_guid, NULL))
598 return (B_TRUE);
599
600 /*
601 * We can't rely on a pool's state if it's been imported
602 * read-only. Instead we look to see if the pools is marked
603 * read-only in the namespace and set the state to active.
604 */
605 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
606 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
607 spa_mode(spa) == FREAD)
608 state = POOL_STATE_ACTIVE;
609
610 /*
611 * If the device is marked ACTIVE, then this device is in use by another
612 * pool on the system.
613 */
614 return (state == POOL_STATE_ACTIVE);
615 }
616
617 /*
618 * Initialize a vdev label. We check to make sure each leaf device is not in
619 * use, and writable. We put down an initial label which we will later
620 * overwrite with a complete label. Note that it's important to do this
621 * sequentially, not in parallel, so that we catch cases of multiple use of the
622 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
623 * itself.
624 */
625 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)626 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
627 {
628 spa_t *spa = vd->vdev_spa;
629 nvlist_t *label;
630 vdev_phys_t *vp;
631 char *pad2;
632 uberblock_t *ub;
633 zio_t *zio;
634 char *buf;
635 size_t buflen;
636 int error;
637 uint64_t spare_guid, l2cache_guid;
638 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
639
640 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
641
642 for (int c = 0; c < vd->vdev_children; c++)
643 if ((error = vdev_label_init(vd->vdev_child[c],
644 crtxg, reason)) != 0)
645 return (error);
646
647 /* Track the creation time for this vdev */
648 vd->vdev_crtxg = crtxg;
649
650 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
651 return (0);
652
653 /*
654 * Dead vdevs cannot be initialized.
655 */
656 if (vdev_is_dead(vd))
657 return (SET_ERROR(EIO));
658
659 /*
660 * Determine if the vdev is in use.
661 */
662 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
663 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
664 return (SET_ERROR(EBUSY));
665
666 /*
667 * If this is a request to add or replace a spare or l2cache device
668 * that is in use elsewhere on the system, then we must update the
669 * guid (which was initialized to a random value) to reflect the
670 * actual GUID (which is shared between multiple pools).
671 */
672 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
673 spare_guid != 0ULL) {
674 uint64_t guid_delta = spare_guid - vd->vdev_guid;
675
676 vd->vdev_guid += guid_delta;
677
678 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
679 pvd->vdev_guid_sum += guid_delta;
680
681 /*
682 * If this is a replacement, then we want to fallthrough to the
683 * rest of the code. If we're adding a spare, then it's already
684 * labeled appropriately and we can just return.
685 */
686 if (reason == VDEV_LABEL_SPARE)
687 return (0);
688 ASSERT(reason == VDEV_LABEL_REPLACE ||
689 reason == VDEV_LABEL_SPLIT);
690 }
691
692 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
693 l2cache_guid != 0ULL) {
694 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
695
696 vd->vdev_guid += guid_delta;
697
698 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
699 pvd->vdev_guid_sum += guid_delta;
700
701 /*
702 * If this is a replacement, then we want to fallthrough to the
703 * rest of the code. If we're adding an l2cache, then it's
704 * already labeled appropriately and we can just return.
705 */
706 if (reason == VDEV_LABEL_L2CACHE)
707 return (0);
708 ASSERT(reason == VDEV_LABEL_REPLACE);
709 }
710
711 /*
712 * Initialize its label.
713 */
714 vp = zio_buf_alloc(sizeof (vdev_phys_t));
715 bzero(vp, sizeof (vdev_phys_t));
716
717 /*
718 * Generate a label describing the pool and our top-level vdev.
719 * We mark it as being from txg 0 to indicate that it's not
720 * really part of an active pool just yet. The labels will
721 * be written again with a meaningful txg by spa_sync().
722 */
723 if (reason == VDEV_LABEL_SPARE ||
724 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
725 /*
726 * For inactive hot spares, we generate a special label that
727 * identifies as a mutually shared hot spare. We write the
728 * label if we are adding a hot spare, or if we are removing an
729 * active hot spare (in which case we want to revert the
730 * labels).
731 */
732 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
733
734 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
735 spa_version(spa)) == 0);
736 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
737 POOL_STATE_SPARE) == 0);
738 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
739 vd->vdev_guid) == 0);
740 } else if (reason == VDEV_LABEL_L2CACHE ||
741 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
742 /*
743 * For level 2 ARC devices, add a special label.
744 */
745 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
746
747 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
748 spa_version(spa)) == 0);
749 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
750 POOL_STATE_L2CACHE) == 0);
751 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
752 vd->vdev_guid) == 0);
753 } else {
754 uint64_t txg = 0ULL;
755
756 if (reason == VDEV_LABEL_SPLIT)
757 txg = spa->spa_uberblock.ub_txg;
758 label = spa_config_generate(spa, vd, txg, B_FALSE);
759
760 /*
761 * Add our creation time. This allows us to detect multiple
762 * vdev uses as described above, and automatically expires if we
763 * fail.
764 */
765 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
766 crtxg) == 0);
767 }
768
769 buf = vp->vp_nvlist;
770 buflen = sizeof (vp->vp_nvlist);
771
772 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
773 if (error != 0) {
774 nvlist_free(label);
775 zio_buf_free(vp, sizeof (vdev_phys_t));
776 /* EFAULT means nvlist_pack ran out of room */
777 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
778 }
779
780 /*
781 * Initialize uberblock template.
782 */
783 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
784 bzero(ub, VDEV_UBERBLOCK_RING);
785 *ub = spa->spa_uberblock;
786 ub->ub_txg = 0;
787
788 /* Initialize the 2nd padding area. */
789 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
790 bzero(pad2, VDEV_PAD_SIZE);
791
792 /*
793 * Write everything in parallel.
794 */
795 retry:
796 zio = zio_root(spa, NULL, NULL, flags);
797
798 for (int l = 0; l < VDEV_LABELS; l++) {
799
800 vdev_label_write(zio, vd, l, vp,
801 offsetof(vdev_label_t, vl_vdev_phys),
802 sizeof (vdev_phys_t), NULL, NULL, flags);
803
804 /*
805 * Skip the 1st padding area.
806 * Zero out the 2nd padding area where it might have
807 * left over data from previous filesystem format.
808 */
809 vdev_label_write(zio, vd, l, pad2,
810 offsetof(vdev_label_t, vl_pad2),
811 VDEV_PAD_SIZE, NULL, NULL, flags);
812
813 vdev_label_write(zio, vd, l, ub,
814 offsetof(vdev_label_t, vl_uberblock),
815 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
816 }
817
818 error = zio_wait(zio);
819
820 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
821 flags |= ZIO_FLAG_TRYHARD;
822 goto retry;
823 }
824
825 nvlist_free(label);
826 zio_buf_free(pad2, VDEV_PAD_SIZE);
827 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
828 zio_buf_free(vp, sizeof (vdev_phys_t));
829
830 /*
831 * If this vdev hasn't been previously identified as a spare, then we
832 * mark it as such only if a) we are labeling it as a spare, or b) it
833 * exists as a spare elsewhere in the system. Do the same for
834 * level 2 ARC devices.
835 */
836 if (error == 0 && !vd->vdev_isspare &&
837 (reason == VDEV_LABEL_SPARE ||
838 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
839 spa_spare_add(vd);
840
841 if (error == 0 && !vd->vdev_isl2cache &&
842 (reason == VDEV_LABEL_L2CACHE ||
843 spa_l2cache_exists(vd->vdev_guid, NULL)))
844 spa_l2cache_add(vd);
845
846 return (error);
847 }
848
849 /*
850 * ==========================================================================
851 * uberblock load/sync
852 * ==========================================================================
853 */
854
855 /*
856 * Consider the following situation: txg is safely synced to disk. We've
857 * written the first uberblock for txg + 1, and then we lose power. When we
858 * come back up, we fail to see the uberblock for txg + 1 because, say,
859 * it was on a mirrored device and the replica to which we wrote txg + 1
860 * is now offline. If we then make some changes and sync txg + 1, and then
861 * the missing replica comes back, then for a few seconds we'll have two
862 * conflicting uberblocks on disk with the same txg. The solution is simple:
863 * among uberblocks with equal txg, choose the one with the latest timestamp.
864 */
865 static int
vdev_uberblock_compare(uberblock_t * ub1,uberblock_t * ub2)866 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
867 {
868 if (ub1->ub_txg < ub2->ub_txg)
869 return (-1);
870 if (ub1->ub_txg > ub2->ub_txg)
871 return (1);
872
873 if (ub1->ub_timestamp < ub2->ub_timestamp)
874 return (-1);
875 if (ub1->ub_timestamp > ub2->ub_timestamp)
876 return (1);
877
878 return (0);
879 }
880
881 struct ubl_cbdata {
882 uberblock_t *ubl_ubbest; /* Best uberblock */
883 vdev_t *ubl_vd; /* vdev associated with the above */
884 };
885
886 static void
vdev_uberblock_load_done(zio_t * zio)887 vdev_uberblock_load_done(zio_t *zio)
888 {
889 vdev_t *vd = zio->io_vd;
890 spa_t *spa = zio->io_spa;
891 zio_t *rio = zio->io_private;
892 uberblock_t *ub = zio->io_data;
893 struct ubl_cbdata *cbp = rio->io_private;
894
895 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
896
897 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
898 mutex_enter(&rio->io_lock);
899 if (ub->ub_txg <= spa->spa_load_max_txg &&
900 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
901 /*
902 * Keep track of the vdev in which this uberblock
903 * was found. We will use this information later
904 * to obtain the config nvlist associated with
905 * this uberblock.
906 */
907 *cbp->ubl_ubbest = *ub;
908 cbp->ubl_vd = vd;
909 }
910 mutex_exit(&rio->io_lock);
911 }
912
913 zio_buf_free(zio->io_data, zio->io_size);
914 }
915
916 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)917 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
918 struct ubl_cbdata *cbp)
919 {
920 for (int c = 0; c < vd->vdev_children; c++)
921 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
922
923 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
924 for (int l = 0; l < VDEV_LABELS; l++) {
925 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
926 vdev_label_read(zio, vd, l,
927 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
928 VDEV_UBERBLOCK_OFFSET(vd, n),
929 VDEV_UBERBLOCK_SIZE(vd),
930 vdev_uberblock_load_done, zio, flags);
931 }
932 }
933 }
934 }
935
936 /*
937 * Reads the 'best' uberblock from disk along with its associated
938 * configuration. First, we read the uberblock array of each label of each
939 * vdev, keeping track of the uberblock with the highest txg in each array.
940 * Then, we read the configuration from the same vdev as the best uberblock.
941 */
942 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)943 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
944 {
945 zio_t *zio;
946 spa_t *spa = rvd->vdev_spa;
947 struct ubl_cbdata cb;
948 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
949 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
950
951 ASSERT(ub);
952 ASSERT(config);
953
954 bzero(ub, sizeof (uberblock_t));
955 *config = NULL;
956
957 cb.ubl_ubbest = ub;
958 cb.ubl_vd = NULL;
959
960 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
961 zio = zio_root(spa, NULL, &cb, flags);
962 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
963 (void) zio_wait(zio);
964
965 /*
966 * It's possible that the best uberblock was discovered on a label
967 * that has a configuration which was written in a future txg.
968 * Search all labels on this vdev to find the configuration that
969 * matches the txg for our uberblock.
970 */
971 if (cb.ubl_vd != NULL)
972 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
973 spa_config_exit(spa, SCL_ALL, FTAG);
974 }
975
976 /*
977 * On success, increment root zio's count of good writes.
978 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
979 */
980 static void
vdev_uberblock_sync_done(zio_t * zio)981 vdev_uberblock_sync_done(zio_t *zio)
982 {
983 uint64_t *good_writes = zio->io_private;
984
985 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
986 atomic_inc_64(good_writes);
987 }
988
989 /*
990 * Write the uberblock to all labels of all leaves of the specified vdev.
991 */
992 static void
vdev_uberblock_sync(zio_t * zio,uberblock_t * ub,vdev_t * vd,int flags)993 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
994 {
995 uberblock_t *ubbuf;
996 int n;
997
998 for (int c = 0; c < vd->vdev_children; c++)
999 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1000
1001 if (!vd->vdev_ops->vdev_op_leaf)
1002 return;
1003
1004 if (!vdev_writeable(vd))
1005 return;
1006
1007 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1008
1009 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1010 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1011 *ubbuf = *ub;
1012
1013 for (int l = 0; l < VDEV_LABELS; l++)
1014 vdev_label_write(zio, vd, l, ubbuf,
1015 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1016 vdev_uberblock_sync_done, zio->io_private,
1017 flags | ZIO_FLAG_DONT_PROPAGATE);
1018
1019 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1020 }
1021
1022 /* Sync the uberblocks to all vdevs in svd[] */
1023 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1024 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1025 {
1026 spa_t *spa = svd[0]->vdev_spa;
1027 zio_t *zio;
1028 uint64_t good_writes = 0;
1029
1030 zio = zio_root(spa, NULL, &good_writes, flags);
1031
1032 for (int v = 0; v < svdcount; v++)
1033 vdev_uberblock_sync(zio, ub, svd[v], flags);
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
1044 for (int v = 0; v < svdcount; v++)
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 */
1055 static void
vdev_label_sync_done(zio_t * zio)1056 vdev_label_sync_done(zio_t *zio)
1057 {
1058 uint64_t *good_writes = zio->io_private;
1059
1060 if (zio->io_error == 0)
1061 atomic_inc_64(good_writes);
1062 }
1063
1064 /*
1065 * If there weren't enough good writes, indicate failure to the parent.
1066 */
1067 static void
vdev_label_sync_top_done(zio_t * zio)1068 vdev_label_sync_top_done(zio_t *zio)
1069 {
1070 uint64_t *good_writes = zio->io_private;
1071
1072 if (*good_writes == 0)
1073 zio->io_error = SET_ERROR(EIO);
1074
1075 kmem_free(good_writes, sizeof (uint64_t));
1076 }
1077
1078 /*
1079 * We ignore errors for log and cache devices, simply free the private data.
1080 */
1081 static void
vdev_label_sync_ignore_done(zio_t * zio)1082 vdev_label_sync_ignore_done(zio_t *zio)
1083 {
1084 kmem_free(zio->io_private, sizeof (uint64_t));
1085 }
1086
1087 /*
1088 * Write all even or odd labels to all leaves of the specified vdev.
1089 */
1090 static void
vdev_label_sync(zio_t * zio,vdev_t * vd,int l,uint64_t txg,int flags)1091 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1092 {
1093 nvlist_t *label;
1094 vdev_phys_t *vp;
1095 char *buf;
1096 size_t buflen;
1097
1098 for (int c = 0; c < vd->vdev_children; c++)
1099 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1100
1101 if (!vd->vdev_ops->vdev_op_leaf)
1102 return;
1103
1104 if (!vdev_writeable(vd))
1105 return;
1106
1107 /*
1108 * Generate a label describing the top-level config to which we belong.
1109 */
1110 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1111
1112 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1113 bzero(vp, sizeof (vdev_phys_t));
1114
1115 buf = vp->vp_nvlist;
1116 buflen = sizeof (vp->vp_nvlist);
1117
1118 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1119 for (; l < VDEV_LABELS; l += 2) {
1120 vdev_label_write(zio, vd, l, vp,
1121 offsetof(vdev_label_t, vl_vdev_phys),
1122 sizeof (vdev_phys_t),
1123 vdev_label_sync_done, zio->io_private,
1124 flags | ZIO_FLAG_DONT_PROPAGATE);
1125 }
1126 }
1127
1128 zio_buf_free(vp, sizeof (vdev_phys_t));
1129 nvlist_free(label);
1130 }
1131
1132 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1133 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1134 {
1135 list_t *dl = &spa->spa_config_dirty_list;
1136 vdev_t *vd;
1137 zio_t *zio;
1138 int error;
1139
1140 /*
1141 * Write the new labels to disk.
1142 */
1143 zio = zio_root(spa, NULL, NULL, flags);
1144
1145 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1146 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1147 KM_SLEEP);
1148
1149 ASSERT(!vd->vdev_ishole);
1150
1151 zio_t *vio = zio_null(zio, spa, NULL,
1152 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1153 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1154 good_writes, flags);
1155 vdev_label_sync(vio, vd, l, txg, flags);
1156 zio_nowait(vio);
1157 }
1158
1159 error = zio_wait(zio);
1160
1161 /*
1162 * Flush the new labels to disk.
1163 */
1164 zio = zio_root(spa, NULL, NULL, flags);
1165
1166 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1167 zio_flush(zio, vd);
1168
1169 (void) zio_wait(zio);
1170
1171 return (error);
1172 }
1173
1174 /*
1175 * Sync the uberblock and any changes to the vdev configuration.
1176 *
1177 * The order of operations is carefully crafted to ensure that
1178 * if the system panics or loses power at any time, the state on disk
1179 * is still transactionally consistent. The in-line comments below
1180 * describe the failure semantics at each stage.
1181 *
1182 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1183 * at any time, you can just call it again, and it will resume its work.
1184 */
1185 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)1186 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1187 {
1188 spa_t *spa = svd[0]->vdev_spa;
1189 uberblock_t *ub = &spa->spa_uberblock;
1190 vdev_t *vd;
1191 zio_t *zio;
1192 int error = 0;
1193 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1194
1195 retry:
1196 /*
1197 * Normally, we don't want to try too hard to write every label and
1198 * uberblock. If there is a flaky disk, we don't want the rest of the
1199 * sync process to block while we retry. But if we can't write a
1200 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1201 * bailing out and declaring the pool faulted.
1202 */
1203 if (error != 0) {
1204 if ((flags & ZIO_FLAG_TRYHARD) != 0)
1205 return (error);
1206 flags |= ZIO_FLAG_TRYHARD;
1207 }
1208
1209 ASSERT(ub->ub_txg <= txg);
1210
1211 /*
1212 * If this isn't a resync due to I/O errors,
1213 * and nothing changed in this transaction group,
1214 * and the vdev configuration hasn't changed,
1215 * then there's nothing to do.
1216 */
1217 if (ub->ub_txg < txg &&
1218 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1219 list_is_empty(&spa->spa_config_dirty_list))
1220 return (0);
1221
1222 if (txg > spa_freeze_txg(spa))
1223 return (0);
1224
1225 ASSERT(txg <= spa->spa_final_txg);
1226
1227 /*
1228 * Flush the write cache of every disk that's been written to
1229 * in this transaction group. This ensures that all blocks
1230 * written in this txg will be committed to stable storage
1231 * before any uberblock that references them.
1232 */
1233 zio = zio_root(spa, NULL, NULL, flags);
1234
1235 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1236 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1237 zio_flush(zio, vd);
1238
1239 (void) zio_wait(zio);
1240
1241 /*
1242 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1243 * system dies in the middle of this process, that's OK: all of the
1244 * even labels that made it to disk will be newer than any uberblock,
1245 * and will therefore be considered invalid. The odd labels (L1, L3),
1246 * which have not yet been touched, will still be valid. We flush
1247 * the new labels to disk to ensure that all even-label updates
1248 * are committed to stable storage before the uberblock update.
1249 */
1250 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1251 goto retry;
1252
1253 /*
1254 * Sync the uberblocks to all vdevs in svd[].
1255 * If the system dies in the middle of this step, there are two cases
1256 * to consider, and the on-disk state is consistent either way:
1257 *
1258 * (1) If none of the new uberblocks made it to disk, then the
1259 * previous uberblock will be the newest, and the odd labels
1260 * (which had not yet been touched) will be valid with respect
1261 * to that uberblock.
1262 *
1263 * (2) If one or more new uberblocks made it to disk, then they
1264 * will be the newest, and the even labels (which had all
1265 * been successfully committed) will be valid with respect
1266 * to the new uberblocks.
1267 */
1268 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1269 goto retry;
1270
1271 /*
1272 * Sync out odd labels for every dirty vdev. If the system dies
1273 * in the middle of this process, the even labels and the new
1274 * uberblocks will suffice to open the pool. The next time
1275 * the pool is opened, the first thing we'll do -- before any
1276 * user data is modified -- is mark every vdev dirty so that
1277 * all labels will be brought up to date. We flush the new labels
1278 * to disk to ensure that all odd-label updates are committed to
1279 * stable storage before the next transaction group begins.
1280 */
1281 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1282 goto retry;
1283
1284 return (0);
1285 }
1286