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) 2013 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/trim_map.h>
149 #include <sys/fs/zfs.h>
150
151 static boolean_t vdev_trim_on_init = B_TRUE;
152 SYSCTL_DECL(_vfs_zfs_vdev);
153 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, trim_on_init, CTLFLAG_RW,
154 &vdev_trim_on_init, 0, "Enable/disable full vdev trim on initialisation");
155
156 /*
157 * Basic routines to read and write from a vdev label.
158 * Used throughout the rest of this file.
159 */
160 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)161 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
162 {
163 ASSERT(offset < sizeof (vdev_label_t));
164 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
165
166 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
167 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
168 }
169
170 /*
171 * Returns back the vdev label associated with the passed in offset.
172 */
173 int
vdev_label_number(uint64_t psize,uint64_t offset)174 vdev_label_number(uint64_t psize, uint64_t offset)
175 {
176 int l;
177
178 if (offset >= psize - VDEV_LABEL_END_SIZE) {
179 offset -= psize - VDEV_LABEL_END_SIZE;
180 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
181 }
182 l = offset / sizeof (vdev_label_t);
183 return (l < VDEV_LABELS ? l : -1);
184 }
185
186 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)187 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
188 uint64_t size, zio_done_func_t *done, void *private, int flags)
189 {
190 ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
191 SCL_STATE_ALL);
192 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
193
194 zio_nowait(zio_read_phys(zio, vd,
195 vdev_label_offset(vd->vdev_psize, l, offset),
196 size, buf, ZIO_CHECKSUM_LABEL, done, private,
197 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
198 }
199
200 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)201 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
202 uint64_t size, zio_done_func_t *done, void *private, int flags)
203 {
204 ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
205 (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
206 (SCL_CONFIG | SCL_STATE) &&
207 dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
208 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
209
210 zio_nowait(zio_write_phys(zio, vd,
211 vdev_label_offset(vd->vdev_psize, l, offset),
212 size, buf, ZIO_CHECKSUM_LABEL, done, private,
213 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
214 }
215
216 /*
217 * Generate the nvlist representing this vdev's config.
218 */
219 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)220 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
221 vdev_config_flag_t flags)
222 {
223 nvlist_t *nv = NULL;
224
225 nv = fnvlist_alloc();
226
227 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
228 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
229 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
230 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
231
232 if (vd->vdev_path != NULL)
233 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
234
235 if (vd->vdev_devid != NULL)
236 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
237
238 if (vd->vdev_physpath != NULL)
239 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
240 vd->vdev_physpath);
241
242 if (vd->vdev_fru != NULL)
243 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
244
245 if (vd->vdev_nparity != 0) {
246 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
247 VDEV_TYPE_RAIDZ) == 0);
248
249 /*
250 * Make sure someone hasn't managed to sneak a fancy new vdev
251 * into a crufty old storage pool.
252 */
253 ASSERT(vd->vdev_nparity == 1 ||
254 (vd->vdev_nparity <= 2 &&
255 spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
256 (vd->vdev_nparity <= 3 &&
257 spa_version(spa) >= SPA_VERSION_RAIDZ3));
258
259 /*
260 * Note that we'll add the nparity tag even on storage pools
261 * that only support a single parity device -- older software
262 * will just ignore it.
263 */
264 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
265 }
266
267 if (vd->vdev_wholedisk != -1ULL)
268 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
269 vd->vdev_wholedisk);
270
271 if (vd->vdev_not_present)
272 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
273
274 if (vd->vdev_isspare)
275 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
276
277 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
278 vd == vd->vdev_top) {
279 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
280 vd->vdev_ms_array);
281 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
282 vd->vdev_ms_shift);
283 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
284 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
285 vd->vdev_asize);
286 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
287 if (vd->vdev_removing)
288 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
289 vd->vdev_removing);
290 }
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 ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
606 spa_mode(spa) == FREAD)
607 state = POOL_STATE_ACTIVE;
608
609 /*
610 * If the device is marked ACTIVE, then this device is in use by another
611 * pool on the system.
612 */
613 return (state == POOL_STATE_ACTIVE);
614 }
615
616 /*
617 * Initialize a vdev label. We check to make sure each leaf device is not in
618 * use, and writable. We put down an initial label which we will later
619 * overwrite with a complete label. Note that it's important to do this
620 * sequentially, not in parallel, so that we catch cases of multiple use of the
621 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
622 * itself.
623 */
624 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)625 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
626 {
627 spa_t *spa = vd->vdev_spa;
628 nvlist_t *label;
629 vdev_phys_t *vp;
630 char *pad2;
631 uberblock_t *ub;
632 zio_t *zio;
633 char *buf;
634 size_t buflen;
635 int error;
636 uint64_t spare_guid, l2cache_guid;
637 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
638
639 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
640
641 for (int c = 0; c < vd->vdev_children; c++)
642 if ((error = vdev_label_init(vd->vdev_child[c],
643 crtxg, reason)) != 0)
644 return (error);
645
646 /* Track the creation time for this vdev */
647 vd->vdev_crtxg = crtxg;
648
649 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
650 return (0);
651
652 /*
653 * Dead vdevs cannot be initialized.
654 */
655 if (vdev_is_dead(vd))
656 return (SET_ERROR(EIO));
657
658 /*
659 * Determine if the vdev is in use.
660 */
661 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
662 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
663 return (SET_ERROR(EBUSY));
664
665 /*
666 * If this is a request to add or replace a spare or l2cache device
667 * that is in use elsewhere on the system, then we must update the
668 * guid (which was initialized to a random value) to reflect the
669 * actual GUID (which is shared between multiple pools).
670 */
671 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
672 spare_guid != 0ULL) {
673 uint64_t guid_delta = spare_guid - vd->vdev_guid;
674
675 vd->vdev_guid += guid_delta;
676
677 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
678 pvd->vdev_guid_sum += guid_delta;
679
680 /*
681 * If this is a replacement, then we want to fallthrough to the
682 * rest of the code. If we're adding a spare, then it's already
683 * labeled appropriately and we can just return.
684 */
685 if (reason == VDEV_LABEL_SPARE)
686 return (0);
687 ASSERT(reason == VDEV_LABEL_REPLACE ||
688 reason == VDEV_LABEL_SPLIT);
689 }
690
691 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
692 l2cache_guid != 0ULL) {
693 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
694
695 vd->vdev_guid += guid_delta;
696
697 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
698 pvd->vdev_guid_sum += guid_delta;
699
700 /*
701 * If this is a replacement, then we want to fallthrough to the
702 * rest of the code. If we're adding an l2cache, then it's
703 * already labeled appropriately and we can just return.
704 */
705 if (reason == VDEV_LABEL_L2CACHE)
706 return (0);
707 ASSERT(reason == VDEV_LABEL_REPLACE);
708 }
709
710 /*
711 * TRIM the whole thing so that we start with a clean slate.
712 * It's just an optimization, so we don't care if it fails.
713 * Don't TRIM if removing so that we don't interfere with zpool
714 * disaster recovery.
715 */
716 if (zfs_trim_enabled && vdev_trim_on_init && !vd->vdev_notrim &&
717 (reason == VDEV_LABEL_CREATE || reason == VDEV_LABEL_SPARE ||
718 reason == VDEV_LABEL_L2CACHE))
719 zio_wait(zio_trim(NULL, spa, vd, 0, vd->vdev_psize));
720
721 /*
722 * Initialize its label.
723 */
724 vp = zio_buf_alloc(sizeof (vdev_phys_t));
725 bzero(vp, sizeof (vdev_phys_t));
726
727 /*
728 * Generate a label describing the pool and our top-level vdev.
729 * We mark it as being from txg 0 to indicate that it's not
730 * really part of an active pool just yet. The labels will
731 * be written again with a meaningful txg by spa_sync().
732 */
733 if (reason == VDEV_LABEL_SPARE ||
734 (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
735 /*
736 * For inactive hot spares, we generate a special label that
737 * identifies as a mutually shared hot spare. We write the
738 * label if we are adding a hot spare, or if we are removing an
739 * active hot spare (in which case we want to revert the
740 * labels).
741 */
742 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
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_SPARE) == 0);
748 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
749 vd->vdev_guid) == 0);
750 } else if (reason == VDEV_LABEL_L2CACHE ||
751 (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
752 /*
753 * For level 2 ARC devices, add a special label.
754 */
755 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
756
757 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
758 spa_version(spa)) == 0);
759 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
760 POOL_STATE_L2CACHE) == 0);
761 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
762 vd->vdev_guid) == 0);
763 } else {
764 uint64_t txg = 0ULL;
765
766 if (reason == VDEV_LABEL_SPLIT)
767 txg = spa->spa_uberblock.ub_txg;
768 label = spa_config_generate(spa, vd, txg, B_FALSE);
769
770 /*
771 * Add our creation time. This allows us to detect multiple
772 * vdev uses as described above, and automatically expires if we
773 * fail.
774 */
775 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
776 crtxg) == 0);
777 }
778
779 buf = vp->vp_nvlist;
780 buflen = sizeof (vp->vp_nvlist);
781
782 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
783 if (error != 0) {
784 nvlist_free(label);
785 zio_buf_free(vp, sizeof (vdev_phys_t));
786 /* EFAULT means nvlist_pack ran out of room */
787 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
788 }
789
790 /*
791 * Initialize uberblock template.
792 */
793 ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
794 bzero(ub, VDEV_UBERBLOCK_RING);
795 *ub = spa->spa_uberblock;
796 ub->ub_txg = 0;
797
798 /* Initialize the 2nd padding area. */
799 pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
800 bzero(pad2, VDEV_PAD_SIZE);
801
802 /*
803 * Write everything in parallel.
804 */
805 retry:
806 zio = zio_root(spa, NULL, NULL, flags);
807
808 for (int l = 0; l < VDEV_LABELS; l++) {
809
810 vdev_label_write(zio, vd, l, vp,
811 offsetof(vdev_label_t, vl_vdev_phys),
812 sizeof (vdev_phys_t), NULL, NULL, flags);
813
814 /*
815 * Skip the 1st padding area.
816 * Zero out the 2nd padding area where it might have
817 * left over data from previous filesystem format.
818 */
819 vdev_label_write(zio, vd, l, pad2,
820 offsetof(vdev_label_t, vl_pad2),
821 VDEV_PAD_SIZE, NULL, NULL, flags);
822
823 vdev_label_write(zio, vd, l, ub,
824 offsetof(vdev_label_t, vl_uberblock),
825 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
826 }
827
828 error = zio_wait(zio);
829
830 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
831 flags |= ZIO_FLAG_TRYHARD;
832 goto retry;
833 }
834
835 nvlist_free(label);
836 zio_buf_free(pad2, VDEV_PAD_SIZE);
837 zio_buf_free(ub, VDEV_UBERBLOCK_RING);
838 zio_buf_free(vp, sizeof (vdev_phys_t));
839
840 /*
841 * If this vdev hasn't been previously identified as a spare, then we
842 * mark it as such only if a) we are labeling it as a spare, or b) it
843 * exists as a spare elsewhere in the system. Do the same for
844 * level 2 ARC devices.
845 */
846 if (error == 0 && !vd->vdev_isspare &&
847 (reason == VDEV_LABEL_SPARE ||
848 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
849 spa_spare_add(vd);
850
851 if (error == 0 && !vd->vdev_isl2cache &&
852 (reason == VDEV_LABEL_L2CACHE ||
853 spa_l2cache_exists(vd->vdev_guid, NULL)))
854 spa_l2cache_add(vd);
855
856 return (error);
857 }
858
859 /*
860 * ==========================================================================
861 * uberblock load/sync
862 * ==========================================================================
863 */
864
865 /*
866 * Consider the following situation: txg is safely synced to disk. We've
867 * written the first uberblock for txg + 1, and then we lose power. When we
868 * come back up, we fail to see the uberblock for txg + 1 because, say,
869 * it was on a mirrored device and the replica to which we wrote txg + 1
870 * is now offline. If we then make some changes and sync txg + 1, and then
871 * the missing replica comes back, then for a few seconds we'll have two
872 * conflicting uberblocks on disk with the same txg. The solution is simple:
873 * among uberblocks with equal txg, choose the one with the latest timestamp.
874 */
875 static int
vdev_uberblock_compare(uberblock_t * ub1,uberblock_t * ub2)876 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
877 {
878 if (ub1->ub_txg < ub2->ub_txg)
879 return (-1);
880 if (ub1->ub_txg > ub2->ub_txg)
881 return (1);
882
883 if (ub1->ub_timestamp < ub2->ub_timestamp)
884 return (-1);
885 if (ub1->ub_timestamp > ub2->ub_timestamp)
886 return (1);
887
888 return (0);
889 }
890
891 struct ubl_cbdata {
892 uberblock_t *ubl_ubbest; /* Best uberblock */
893 vdev_t *ubl_vd; /* vdev associated with the above */
894 };
895
896 static void
vdev_uberblock_load_done(zio_t * zio)897 vdev_uberblock_load_done(zio_t *zio)
898 {
899 vdev_t *vd = zio->io_vd;
900 spa_t *spa = zio->io_spa;
901 zio_t *rio = zio->io_private;
902 uberblock_t *ub = zio->io_data;
903 struct ubl_cbdata *cbp = rio->io_private;
904
905 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
906
907 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
908 mutex_enter(&rio->io_lock);
909 if (ub->ub_txg <= spa->spa_load_max_txg &&
910 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
911 /*
912 * Keep track of the vdev in which this uberblock
913 * was found. We will use this information later
914 * to obtain the config nvlist associated with
915 * this uberblock.
916 */
917 *cbp->ubl_ubbest = *ub;
918 cbp->ubl_vd = vd;
919 }
920 mutex_exit(&rio->io_lock);
921 }
922
923 zio_buf_free(zio->io_data, zio->io_size);
924 }
925
926 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)927 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
928 struct ubl_cbdata *cbp)
929 {
930 for (int c = 0; c < vd->vdev_children; c++)
931 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
932
933 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
934 for (int l = 0; l < VDEV_LABELS; l++) {
935 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
936 vdev_label_read(zio, vd, l,
937 zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
938 VDEV_UBERBLOCK_OFFSET(vd, n),
939 VDEV_UBERBLOCK_SIZE(vd),
940 vdev_uberblock_load_done, zio, flags);
941 }
942 }
943 }
944 }
945
946 /*
947 * Reads the 'best' uberblock from disk along with its associated
948 * configuration. First, we read the uberblock array of each label of each
949 * vdev, keeping track of the uberblock with the highest txg in each array.
950 * Then, we read the configuration from the same vdev as the best uberblock.
951 */
952 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)953 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
954 {
955 zio_t *zio;
956 spa_t *spa = rvd->vdev_spa;
957 struct ubl_cbdata cb;
958 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
959 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
960
961 ASSERT(ub);
962 ASSERT(config);
963
964 bzero(ub, sizeof (uberblock_t));
965 *config = NULL;
966
967 cb.ubl_ubbest = ub;
968 cb.ubl_vd = NULL;
969
970 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
971 zio = zio_root(spa, NULL, &cb, flags);
972 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
973 (void) zio_wait(zio);
974
975 /*
976 * It's possible that the best uberblock was discovered on a label
977 * that has a configuration which was written in a future txg.
978 * Search all labels on this vdev to find the configuration that
979 * matches the txg for our uberblock.
980 */
981 if (cb.ubl_vd != NULL)
982 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
983 spa_config_exit(spa, SCL_ALL, FTAG);
984 }
985
986 /*
987 * On success, increment root zio's count of good writes.
988 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
989 */
990 static void
vdev_uberblock_sync_done(zio_t * zio)991 vdev_uberblock_sync_done(zio_t *zio)
992 {
993 uint64_t *good_writes = zio->io_private;
994
995 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
996 atomic_inc_64(good_writes);
997 }
998
999 /*
1000 * Write the uberblock to all labels of all leaves of the specified vdev.
1001 */
1002 static void
vdev_uberblock_sync(zio_t * zio,uberblock_t * ub,vdev_t * vd,int flags)1003 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1004 {
1005 uberblock_t *ubbuf;
1006 int n;
1007
1008 for (int c = 0; c < vd->vdev_children; c++)
1009 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1010
1011 if (!vd->vdev_ops->vdev_op_leaf)
1012 return;
1013
1014 if (!vdev_writeable(vd))
1015 return;
1016
1017 n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1018
1019 ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1020 bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1021 *ubbuf = *ub;
1022
1023 for (int l = 0; l < VDEV_LABELS; l++)
1024 vdev_label_write(zio, vd, l, ubbuf,
1025 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1026 vdev_uberblock_sync_done, zio->io_private,
1027 flags | ZIO_FLAG_DONT_PROPAGATE);
1028
1029 zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1030 }
1031
1032 /* Sync the uberblocks to all vdevs in svd[] */
1033 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1034 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1035 {
1036 spa_t *spa = svd[0]->vdev_spa;
1037 zio_t *zio;
1038 uint64_t good_writes = 0;
1039
1040 zio = zio_root(spa, NULL, &good_writes, flags);
1041
1042 for (int v = 0; v < svdcount; v++)
1043 vdev_uberblock_sync(zio, ub, svd[v], flags);
1044
1045 (void) zio_wait(zio);
1046
1047 /*
1048 * Flush the uberblocks to disk. This ensures that the odd labels
1049 * are no longer needed (because the new uberblocks and the even
1050 * labels are safely on disk), so it is safe to overwrite them.
1051 */
1052 zio = zio_root(spa, NULL, NULL, flags);
1053
1054 for (int v = 0; v < svdcount; v++)
1055 zio_flush(zio, svd[v]);
1056
1057 (void) zio_wait(zio);
1058
1059 return (good_writes >= 1 ? 0 : EIO);
1060 }
1061
1062 /*
1063 * On success, increment the count of good writes for our top-level vdev.
1064 */
1065 static void
vdev_label_sync_done(zio_t * zio)1066 vdev_label_sync_done(zio_t *zio)
1067 {
1068 uint64_t *good_writes = zio->io_private;
1069
1070 if (zio->io_error == 0)
1071 atomic_inc_64(good_writes);
1072 }
1073
1074 /*
1075 * If there weren't enough good writes, indicate failure to the parent.
1076 */
1077 static void
vdev_label_sync_top_done(zio_t * zio)1078 vdev_label_sync_top_done(zio_t *zio)
1079 {
1080 uint64_t *good_writes = zio->io_private;
1081
1082 if (*good_writes == 0)
1083 zio->io_error = SET_ERROR(EIO);
1084
1085 kmem_free(good_writes, sizeof (uint64_t));
1086 }
1087
1088 /*
1089 * We ignore errors for log and cache devices, simply free the private data.
1090 */
1091 static void
vdev_label_sync_ignore_done(zio_t * zio)1092 vdev_label_sync_ignore_done(zio_t *zio)
1093 {
1094 kmem_free(zio->io_private, sizeof (uint64_t));
1095 }
1096
1097 /*
1098 * Write all even or odd labels to all leaves of the specified vdev.
1099 */
1100 static void
vdev_label_sync(zio_t * zio,vdev_t * vd,int l,uint64_t txg,int flags)1101 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1102 {
1103 nvlist_t *label;
1104 vdev_phys_t *vp;
1105 char *buf;
1106 size_t buflen;
1107
1108 for (int c = 0; c < vd->vdev_children; c++)
1109 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1110
1111 if (!vd->vdev_ops->vdev_op_leaf)
1112 return;
1113
1114 if (!vdev_writeable(vd))
1115 return;
1116
1117 /*
1118 * Generate a label describing the top-level config to which we belong.
1119 */
1120 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1121
1122 vp = zio_buf_alloc(sizeof (vdev_phys_t));
1123 bzero(vp, sizeof (vdev_phys_t));
1124
1125 buf = vp->vp_nvlist;
1126 buflen = sizeof (vp->vp_nvlist);
1127
1128 if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1129 for (; l < VDEV_LABELS; l += 2) {
1130 vdev_label_write(zio, vd, l, vp,
1131 offsetof(vdev_label_t, vl_vdev_phys),
1132 sizeof (vdev_phys_t),
1133 vdev_label_sync_done, zio->io_private,
1134 flags | ZIO_FLAG_DONT_PROPAGATE);
1135 }
1136 }
1137
1138 zio_buf_free(vp, sizeof (vdev_phys_t));
1139 nvlist_free(label);
1140 }
1141
1142 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1143 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1144 {
1145 list_t *dl = &spa->spa_config_dirty_list;
1146 vdev_t *vd;
1147 zio_t *zio;
1148 int error;
1149
1150 /*
1151 * Write the new labels to disk.
1152 */
1153 zio = zio_root(spa, NULL, NULL, flags);
1154
1155 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1156 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1157 KM_SLEEP);
1158
1159 ASSERT(!vd->vdev_ishole);
1160
1161 zio_t *vio = zio_null(zio, spa, NULL,
1162 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1163 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1164 good_writes, flags);
1165 vdev_label_sync(vio, vd, l, txg, flags);
1166 zio_nowait(vio);
1167 }
1168
1169 error = zio_wait(zio);
1170
1171 /*
1172 * Flush the new labels to disk.
1173 */
1174 zio = zio_root(spa, NULL, NULL, flags);
1175
1176 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1177 zio_flush(zio, vd);
1178
1179 (void) zio_wait(zio);
1180
1181 return (error);
1182 }
1183
1184 /*
1185 * Sync the uberblock and any changes to the vdev configuration.
1186 *
1187 * The order of operations is carefully crafted to ensure that
1188 * if the system panics or loses power at any time, the state on disk
1189 * is still transactionally consistent. The in-line comments below
1190 * describe the failure semantics at each stage.
1191 *
1192 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1193 * at any time, you can just call it again, and it will resume its work.
1194 */
1195 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg,boolean_t tryhard)1196 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1197 {
1198 spa_t *spa = svd[0]->vdev_spa;
1199 uberblock_t *ub = &spa->spa_uberblock;
1200 vdev_t *vd;
1201 zio_t *zio;
1202 int error;
1203 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1204
1205 /*
1206 * Normally, we don't want to try too hard to write every label and
1207 * uberblock. If there is a flaky disk, we don't want the rest of the
1208 * sync process to block while we retry. But if we can't write a
1209 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1210 * bailing out and declaring the pool faulted.
1211 */
1212 if (tryhard)
1213 flags |= ZIO_FLAG_TRYHARD;
1214
1215 ASSERT(ub->ub_txg <= txg);
1216
1217 /*
1218 * If this isn't a resync due to I/O errors,
1219 * and nothing changed in this transaction group,
1220 * and the vdev configuration hasn't changed,
1221 * then there's nothing to do.
1222 */
1223 if (ub->ub_txg < txg &&
1224 uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1225 list_is_empty(&spa->spa_config_dirty_list))
1226 return (0);
1227
1228 if (txg > spa_freeze_txg(spa))
1229 return (0);
1230
1231 ASSERT(txg <= spa->spa_final_txg);
1232
1233 /*
1234 * Flush the write cache of every disk that's been written to
1235 * in this transaction group. This ensures that all blocks
1236 * written in this txg will be committed to stable storage
1237 * before any uberblock that references them.
1238 */
1239 zio = zio_root(spa, NULL, NULL, flags);
1240
1241 for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1242 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1243 zio_flush(zio, vd);
1244
1245 (void) zio_wait(zio);
1246
1247 /*
1248 * Sync out the even labels (L0, L2) for every dirty vdev. If the
1249 * system dies in the middle of this process, that's OK: all of the
1250 * even labels that made it to disk will be newer than any uberblock,
1251 * and will therefore be considered invalid. The odd labels (L1, L3),
1252 * which have not yet been touched, will still be valid. We flush
1253 * the new labels to disk to ensure that all even-label updates
1254 * are committed to stable storage before the uberblock update.
1255 */
1256 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1257 return (error);
1258
1259 /*
1260 * Sync the uberblocks to all vdevs in svd[].
1261 * If the system dies in the middle of this step, there are two cases
1262 * to consider, and the on-disk state is consistent either way:
1263 *
1264 * (1) If none of the new uberblocks made it to disk, then the
1265 * previous uberblock will be the newest, and the odd labels
1266 * (which had not yet been touched) will be valid with respect
1267 * to that uberblock.
1268 *
1269 * (2) If one or more new uberblocks made it to disk, then they
1270 * will be the newest, and the even labels (which had all
1271 * been successfully committed) will be valid with respect
1272 * to the new uberblocks.
1273 */
1274 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1275 return (error);
1276
1277 /*
1278 * Sync out odd labels for every dirty vdev. If the system dies
1279 * in the middle of this process, the even labels and the new
1280 * uberblocks will suffice to open the pool. The next time
1281 * the pool is opened, the first thing we'll do -- before any
1282 * user data is modified -- is mark every vdev dirty so that
1283 * all labels will be brought up to date. We flush the new labels
1284 * to disk to ensure that all odd-label updates are committed to
1285 * stable storage before the next transaction group begins.
1286 */
1287 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0)
1288 return (error);
1289
1290 trim_thread_wakeup(spa);
1291
1292 return (0);
1293 }
1294