Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/spa.c @ 2199:712a788c2dfd
PSARC 2006/388 snapshot -r
6373978 want to take lots of snapshots quickly ('zfs snapshot -r')
author | ahrens |
---|---|
date | Wed, 14 Jun 2006 23:16:39 -0700 |
parents | 73de7a781492 |
children | e640bebc73b3 |
rev | line source |
---|---|
789 | 1 /* |
2 * CDDL HEADER START | |
3 * | |
4 * The contents of this file are subject to the terms of the | |
1544 | 5 * Common Development and Distribution License (the "License"). |
6 * You may not use this file except in compliance with the License. | |
789 | 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 */ | |
2082 | 21 |
789 | 22 /* |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
23 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
789 | 24 * Use is subject to license terms. |
25 */ | |
26 | |
27 #pragma ident "%Z%%M% %I% %E% SMI" | |
28 | |
29 /* | |
30 * This file contains all the routines used when modifying on-disk SPA state. | |
31 * This includes opening, importing, destroying, exporting a pool, and syncing a | |
32 * pool. | |
33 */ | |
34 | |
35 #include <sys/zfs_context.h> | |
1544 | 36 #include <sys/fm/fs/zfs.h> |
789 | 37 #include <sys/spa_impl.h> |
38 #include <sys/zio.h> | |
39 #include <sys/zio_checksum.h> | |
40 #include <sys/zio_compress.h> | |
41 #include <sys/dmu.h> | |
42 #include <sys/dmu_tx.h> | |
43 #include <sys/zap.h> | |
44 #include <sys/zil.h> | |
45 #include <sys/vdev_impl.h> | |
46 #include <sys/metaslab.h> | |
47 #include <sys/uberblock_impl.h> | |
48 #include <sys/txg.h> | |
49 #include <sys/avl.h> | |
50 #include <sys/dmu_traverse.h> | |
51 #include <sys/unique.h> | |
52 #include <sys/dsl_pool.h> | |
53 #include <sys/dsl_dir.h> | |
54 #include <sys/dsl_prop.h> | |
55 #include <sys/fs/zfs.h> | |
56 #include <sys/callb.h> | |
57 | |
58 /* | |
59 * ========================================================================== | |
60 * SPA state manipulation (open/create/destroy/import/export) | |
61 * ========================================================================== | |
62 */ | |
63 | |
1544 | 64 static int |
65 spa_error_entry_compare(const void *a, const void *b) | |
66 { | |
67 spa_error_entry_t *sa = (spa_error_entry_t *)a; | |
68 spa_error_entry_t *sb = (spa_error_entry_t *)b; | |
69 int ret; | |
70 | |
71 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, | |
72 sizeof (zbookmark_t)); | |
73 | |
74 if (ret < 0) | |
75 return (-1); | |
76 else if (ret > 0) | |
77 return (1); | |
78 else | |
79 return (0); | |
80 } | |
81 | |
82 /* | |
83 * Utility function which retrieves copies of the current logs and | |
84 * re-initializes them in the process. | |
85 */ | |
86 void | |
87 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
88 { | |
89 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
90 | |
91 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
92 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
93 | |
94 avl_create(&spa->spa_errlist_scrub, | |
95 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
96 offsetof(spa_error_entry_t, se_avl)); | |
97 avl_create(&spa->spa_errlist_last, | |
98 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
99 offsetof(spa_error_entry_t, se_avl)); | |
100 } | |
101 | |
789 | 102 /* |
103 * Activate an uninitialized pool. | |
104 */ | |
105 static void | |
106 spa_activate(spa_t *spa) | |
107 { | |
108 int t; | |
109 | |
110 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
111 | |
112 spa->spa_state = POOL_STATE_ACTIVE; | |
113 | |
114 spa->spa_normal_class = metaslab_class_create(); | |
115 | |
116 for (t = 0; t < ZIO_TYPES; t++) { | |
117 spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue", | |
118 8, maxclsyspri, 50, INT_MAX, | |
119 TASKQ_PREPOPULATE); | |
120 spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr", | |
121 8, maxclsyspri, 50, INT_MAX, | |
122 TASKQ_PREPOPULATE); | |
123 } | |
124 | |
125 rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL); | |
126 | |
127 list_create(&spa->spa_dirty_list, sizeof (vdev_t), | |
128 offsetof(vdev_t, vdev_dirty_node)); | |
129 | |
130 txg_list_create(&spa->spa_vdev_txg_list, | |
131 offsetof(struct vdev, vdev_txg_node)); | |
1544 | 132 |
133 avl_create(&spa->spa_errlist_scrub, | |
134 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
135 offsetof(spa_error_entry_t, se_avl)); | |
136 avl_create(&spa->spa_errlist_last, | |
137 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
138 offsetof(spa_error_entry_t, se_avl)); | |
789 | 139 } |
140 | |
141 /* | |
142 * Opposite of spa_activate(). | |
143 */ | |
144 static void | |
145 spa_deactivate(spa_t *spa) | |
146 { | |
147 int t; | |
148 | |
149 ASSERT(spa->spa_sync_on == B_FALSE); | |
150 ASSERT(spa->spa_dsl_pool == NULL); | |
151 ASSERT(spa->spa_root_vdev == NULL); | |
152 | |
153 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
154 | |
155 txg_list_destroy(&spa->spa_vdev_txg_list); | |
156 | |
157 list_destroy(&spa->spa_dirty_list); | |
158 | |
159 rw_destroy(&spa->spa_traverse_lock); | |
160 | |
161 for (t = 0; t < ZIO_TYPES; t++) { | |
162 taskq_destroy(spa->spa_zio_issue_taskq[t]); | |
163 taskq_destroy(spa->spa_zio_intr_taskq[t]); | |
164 spa->spa_zio_issue_taskq[t] = NULL; | |
165 spa->spa_zio_intr_taskq[t] = NULL; | |
166 } | |
167 | |
168 metaslab_class_destroy(spa->spa_normal_class); | |
169 spa->spa_normal_class = NULL; | |
170 | |
1544 | 171 /* |
172 * If this was part of an import or the open otherwise failed, we may | |
173 * still have errors left in the queues. Empty them just in case. | |
174 */ | |
175 spa_errlog_drain(spa); | |
176 | |
177 avl_destroy(&spa->spa_errlist_scrub); | |
178 avl_destroy(&spa->spa_errlist_last); | |
179 | |
789 | 180 spa->spa_state = POOL_STATE_UNINITIALIZED; |
181 } | |
182 | |
183 /* | |
184 * Verify a pool configuration, and construct the vdev tree appropriately. This | |
185 * will create all the necessary vdevs in the appropriate layout, with each vdev | |
186 * in the CLOSED state. This will prep the pool before open/creation/import. | |
187 * All vdev validation is done by the vdev_alloc() routine. | |
188 */ | |
2082 | 189 static int |
190 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
191 uint_t id, int atype) | |
789 | 192 { |
193 nvlist_t **child; | |
194 uint_t c, children; | |
2082 | 195 int error; |
196 | |
197 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
198 return (error); | |
199 | |
200 if ((*vdp)->vdev_ops->vdev_op_leaf) | |
201 return (0); | |
789 | 202 |
203 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
204 &child, &children) != 0) { | |
2082 | 205 vdev_free(*vdp); |
206 *vdp = NULL; | |
207 return (EINVAL); | |
789 | 208 } |
209 | |
210 for (c = 0; c < children; c++) { | |
2082 | 211 vdev_t *vd; |
212 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
213 atype)) != 0) { | |
214 vdev_free(*vdp); | |
215 *vdp = NULL; | |
216 return (error); | |
789 | 217 } |
218 } | |
219 | |
2082 | 220 ASSERT(*vdp != NULL); |
221 | |
222 return (0); | |
789 | 223 } |
224 | |
225 /* | |
226 * Opposite of spa_load(). | |
227 */ | |
228 static void | |
229 spa_unload(spa_t *spa) | |
230 { | |
2082 | 231 int i; |
232 | |
789 | 233 /* |
1544 | 234 * Stop async tasks. |
235 */ | |
236 spa_async_suspend(spa); | |
237 | |
238 /* | |
789 | 239 * Stop syncing. |
240 */ | |
241 if (spa->spa_sync_on) { | |
242 txg_sync_stop(spa->spa_dsl_pool); | |
243 spa->spa_sync_on = B_FALSE; | |
244 } | |
245 | |
246 /* | |
247 * Wait for any outstanding prefetch I/O to complete. | |
248 */ | |
1544 | 249 spa_config_enter(spa, RW_WRITER, FTAG); |
250 spa_config_exit(spa, FTAG); | |
789 | 251 |
252 /* | |
253 * Close the dsl pool. | |
254 */ | |
255 if (spa->spa_dsl_pool) { | |
256 dsl_pool_close(spa->spa_dsl_pool); | |
257 spa->spa_dsl_pool = NULL; | |
258 } | |
259 | |
260 /* | |
261 * Close all vdevs. | |
262 */ | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
263 if (spa->spa_root_vdev) |
789 | 264 vdev_free(spa->spa_root_vdev); |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
265 ASSERT(spa->spa_root_vdev == NULL); |
1544 | 266 |
2082 | 267 for (i = 0; i < spa->spa_nspares; i++) |
268 vdev_free(spa->spa_spares[i]); | |
269 if (spa->spa_spares) { | |
270 kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *)); | |
271 spa->spa_spares = NULL; | |
272 } | |
273 if (spa->spa_sparelist) { | |
274 nvlist_free(spa->spa_sparelist); | |
275 spa->spa_sparelist = NULL; | |
276 } | |
277 | |
1544 | 278 spa->spa_async_suspended = 0; |
789 | 279 } |
280 | |
281 /* | |
2082 | 282 * Load (or re-load) the current list of vdevs describing the active spares for |
283 * this pool. When this is called, we have some form of basic information in | |
284 * 'spa_sparelist'. We parse this into vdevs, try to open them, and then | |
285 * re-generate a more complete list including status information. | |
286 */ | |
287 static void | |
288 spa_load_spares(spa_t *spa) | |
289 { | |
290 nvlist_t **spares; | |
291 uint_t nspares; | |
292 int i; | |
293 | |
294 /* | |
295 * First, close and free any existing spare vdevs. | |
296 */ | |
297 for (i = 0; i < spa->spa_nspares; i++) { | |
298 vdev_close(spa->spa_spares[i]); | |
299 vdev_free(spa->spa_spares[i]); | |
300 } | |
301 if (spa->spa_spares) | |
302 kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *)); | |
303 | |
304 if (spa->spa_sparelist == NULL) | |
305 nspares = 0; | |
306 else | |
307 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
308 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
309 | |
310 spa->spa_nspares = (int)nspares; | |
311 spa->spa_spares = NULL; | |
312 | |
313 if (nspares == 0) | |
314 return; | |
315 | |
316 /* | |
317 * Construct the array of vdevs, opening them to get status in the | |
318 * process. | |
319 */ | |
320 spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP); | |
321 for (i = 0; i < spa->spa_nspares; i++) { | |
322 vdev_t *vd; | |
323 | |
324 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
325 VDEV_ALLOC_SPARE) == 0); | |
326 ASSERT(vd != NULL); | |
327 | |
328 spa->spa_spares[i] = vd; | |
329 | |
330 if (vdev_open(vd) != 0) | |
331 continue; | |
332 | |
333 vd->vdev_top = vd; | |
334 (void) vdev_validate_spare(vd); | |
335 } | |
336 | |
337 /* | |
338 * Recompute the stashed list of spares, with status information | |
339 * this time. | |
340 */ | |
341 VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
342 DATA_TYPE_NVLIST_ARRAY) == 0); | |
343 | |
344 spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP); | |
345 for (i = 0; i < spa->spa_nspares; i++) | |
346 spares[i] = vdev_config_generate(spa, spa->spa_spares[i], | |
347 B_TRUE, B_TRUE); | |
348 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
349 spares, spa->spa_nspares) == 0); | |
350 for (i = 0; i < spa->spa_nspares; i++) | |
351 nvlist_free(spares[i]); | |
352 kmem_free(spares, spa->spa_nspares * sizeof (void *)); | |
353 } | |
354 | |
355 static int | |
356 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
357 { | |
358 dmu_buf_t *db; | |
359 char *packed = NULL; | |
360 size_t nvsize = 0; | |
361 int error; | |
362 *value = NULL; | |
363 | |
364 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
365 nvsize = *(uint64_t *)db->db_data; | |
366 dmu_buf_rele(db, FTAG); | |
367 | |
368 packed = kmem_alloc(nvsize, KM_SLEEP); | |
369 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed); | |
370 if (error == 0) | |
371 error = nvlist_unpack(packed, nvsize, value, 0); | |
372 kmem_free(packed, nvsize); | |
373 | |
374 return (error); | |
375 } | |
376 | |
377 /* | |
789 | 378 * Load an existing storage pool, using the pool's builtin spa_config as a |
1544 | 379 * source of configuration information. |
789 | 380 */ |
381 static int | |
1544 | 382 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) |
789 | 383 { |
384 int error = 0; | |
385 nvlist_t *nvroot = NULL; | |
386 vdev_t *rvd; | |
387 uberblock_t *ub = &spa->spa_uberblock; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
388 uint64_t config_cache_txg = spa->spa_config_txg; |
789 | 389 uint64_t pool_guid; |
2082 | 390 uint64_t version; |
789 | 391 zio_t *zio; |
392 | |
1544 | 393 spa->spa_load_state = state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
394 |
789 | 395 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || |
1733 | 396 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { |
1544 | 397 error = EINVAL; |
398 goto out; | |
399 } | |
789 | 400 |
2082 | 401 /* |
402 * Versioning wasn't explicitly added to the label until later, so if | |
403 * it's not present treat it as the initial version. | |
404 */ | |
405 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0) | |
406 version = ZFS_VERSION_INITIAL; | |
407 | |
1733 | 408 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
409 &spa->spa_config_txg); | |
410 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
411 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && |
1544 | 412 spa_guid_exists(pool_guid, 0)) { |
413 error = EEXIST; | |
414 goto out; | |
415 } | |
789 | 416 |
2174
73de7a781492
6433717 offline devices should not be marked persistently unavailble
eschrock
parents:
2082
diff
changeset
|
417 spa->spa_load_guid = pool_guid; |
73de7a781492
6433717 offline devices should not be marked persistently unavailble
eschrock
parents:
2082
diff
changeset
|
418 |
789 | 419 /* |
2082 | 420 * Parse the configuration into a vdev tree. We explicitly set the |
421 * value that will be returned by spa_version() since parsing the | |
422 * configuration requires knowing the version number. | |
789 | 423 */ |
1544 | 424 spa_config_enter(spa, RW_WRITER, FTAG); |
2082 | 425 spa->spa_ubsync.ub_version = version; |
426 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD); | |
1544 | 427 spa_config_exit(spa, FTAG); |
789 | 428 |
2082 | 429 if (error != 0) |
1544 | 430 goto out; |
789 | 431 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
432 ASSERT(spa->spa_root_vdev == rvd); |
789 | 433 ASSERT(spa_guid(spa) == pool_guid); |
434 | |
435 /* | |
436 * Try to open all vdevs, loading each label in the process. | |
437 */ | |
1544 | 438 if (vdev_open(rvd) != 0) { |
439 error = ENXIO; | |
440 goto out; | |
441 } | |
789 | 442 |
443 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
444 * Validate the labels for all leaf vdevs. We need to grab the config |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
445 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
446 * flag. |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
447 */ |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
448 spa_config_enter(spa, RW_READER, FTAG); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
449 error = vdev_validate(rvd); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
450 spa_config_exit(spa, FTAG); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
451 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
452 if (error != 0) { |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
453 error = EBADF; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
454 goto out; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
455 } |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
456 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
457 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
458 error = ENXIO; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
459 goto out; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
460 } |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
461 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
462 /* |
789 | 463 * Find the best uberblock. |
464 */ | |
465 bzero(ub, sizeof (uberblock_t)); | |
466 | |
467 zio = zio_root(spa, NULL, NULL, | |
468 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
469 vdev_uberblock_load(zio, rvd, ub); | |
470 error = zio_wait(zio); | |
471 | |
472 /* | |
473 * If we weren't able to find a single valid uberblock, return failure. | |
474 */ | |
475 if (ub->ub_txg == 0) { | |
1760 | 476 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
477 VDEV_AUX_CORRUPT_DATA); | |
1544 | 478 error = ENXIO; |
479 goto out; | |
480 } | |
481 | |
482 /* | |
483 * If the pool is newer than the code, we can't open it. | |
484 */ | |
1760 | 485 if (ub->ub_version > ZFS_VERSION) { |
486 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
487 VDEV_AUX_VERSION_NEWER); | |
1544 | 488 error = ENOTSUP; |
489 goto out; | |
789 | 490 } |
491 | |
492 /* | |
493 * If the vdev guid sum doesn't match the uberblock, we have an | |
494 * incomplete configuration. | |
495 */ | |
1732 | 496 if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { |
1544 | 497 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
498 VDEV_AUX_BAD_GUID_SUM); | |
499 error = ENXIO; | |
500 goto out; | |
789 | 501 } |
502 | |
503 /* | |
504 * Initialize internal SPA structures. | |
505 */ | |
506 spa->spa_state = POOL_STATE_ACTIVE; | |
507 spa->spa_ubsync = spa->spa_uberblock; | |
508 spa->spa_first_txg = spa_last_synced_txg(spa) + 1; | |
1544 | 509 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
510 if (error) { | |
511 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
512 VDEV_AUX_CORRUPT_DATA); | |
513 goto out; | |
514 } | |
789 | 515 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
516 | |
1544 | 517 if (zap_lookup(spa->spa_meta_objset, |
789 | 518 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 519 sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { |
520 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
521 VDEV_AUX_CORRUPT_DATA); | |
522 error = EIO; | |
523 goto out; | |
524 } | |
789 | 525 |
526 if (!mosconfig) { | |
2082 | 527 nvlist_t *newconfig; |
528 | |
529 if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) { | |
1544 | 530 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
531 VDEV_AUX_CORRUPT_DATA); | |
532 error = EIO; | |
533 goto out; | |
534 } | |
789 | 535 |
536 spa_config_set(spa, newconfig); | |
537 spa_unload(spa); | |
538 spa_deactivate(spa); | |
539 spa_activate(spa); | |
540 | |
1544 | 541 return (spa_load(spa, newconfig, state, B_TRUE)); |
542 } | |
543 | |
544 if (zap_lookup(spa->spa_meta_objset, | |
545 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, | |
546 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { | |
547 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
548 VDEV_AUX_CORRUPT_DATA); | |
549 error = EIO; | |
550 goto out; | |
789 | 551 } |
552 | |
1544 | 553 /* |
2082 | 554 * Load the bit that tells us to use the new accounting function |
555 * (raid-z deflation). If we have an older pool, this will not | |
556 * be present. | |
557 */ | |
558 error = zap_lookup(spa->spa_meta_objset, | |
559 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
560 sizeof (uint64_t), 1, &spa->spa_deflate); | |
561 if (error != 0 && error != ENOENT) { | |
562 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
563 VDEV_AUX_CORRUPT_DATA); | |
564 error = EIO; | |
565 goto out; | |
566 } | |
567 | |
568 /* | |
1544 | 569 * Load the persistent error log. If we have an older pool, this will |
570 * not be present. | |
571 */ | |
572 error = zap_lookup(spa->spa_meta_objset, | |
573 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, | |
574 sizeof (uint64_t), 1, &spa->spa_errlog_last); | |
1807
35c8b566d7af
6410711 intent log blocks don't get invited to pool parties
bonwick
parents:
1775
diff
changeset
|
575 if (error != 0 && error != ENOENT) { |
1544 | 576 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
577 VDEV_AUX_CORRUPT_DATA); | |
578 error = EIO; | |
579 goto out; | |
580 } | |
581 | |
582 error = zap_lookup(spa->spa_meta_objset, | |
583 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, | |
584 sizeof (uint64_t), 1, &spa->spa_errlog_scrub); | |
585 if (error != 0 && error != ENOENT) { | |
586 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
587 VDEV_AUX_CORRUPT_DATA); | |
588 error = EIO; | |
589 goto out; | |
590 } | |
789 | 591 |
592 /* | |
2082 | 593 * Load any hot spares for this pool. |
594 */ | |
595 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
596 DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object); | |
597 if (error != 0 && error != ENOENT) { | |
598 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
599 VDEV_AUX_CORRUPT_DATA); | |
600 error = EIO; | |
601 goto out; | |
602 } | |
603 if (error == 0) { | |
604 ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES); | |
605 if (load_nvlist(spa, spa->spa_spares_object, | |
606 &spa->spa_sparelist) != 0) { | |
607 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
608 VDEV_AUX_CORRUPT_DATA); | |
609 error = EIO; | |
610 goto out; | |
611 } | |
612 | |
613 spa_config_enter(spa, RW_WRITER, FTAG); | |
614 spa_load_spares(spa); | |
615 spa_config_exit(spa, FTAG); | |
616 } | |
617 | |
618 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
619 * Load the vdev state for all toplevel vdevs. |
789 | 620 */ |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
621 vdev_load(rvd); |
789 | 622 |
623 /* | |
624 * Propagate the leaf DTLs we just loaded all the way up the tree. | |
625 */ | |
1544 | 626 spa_config_enter(spa, RW_WRITER, FTAG); |
789 | 627 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); |
1544 | 628 spa_config_exit(spa, FTAG); |
789 | 629 |
630 /* | |
631 * Check the state of the root vdev. If it can't be opened, it | |
632 * indicates one or more toplevel vdevs are faulted. | |
633 */ | |
1544 | 634 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
635 error = ENXIO; | |
636 goto out; | |
637 } | |
789 | 638 |
1544 | 639 if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) { |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
640 dmu_tx_t *tx; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
641 int need_update = B_FALSE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
642 int c; |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
643 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
644 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
645 * Claim log blocks that haven't been committed yet. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
646 * This must all happen in a single txg. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
647 */ |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
648 tx = dmu_tx_create_assigned(spa_get_dsl(spa), |
789 | 649 spa_first_txg(spa)); |
2199 | 650 (void) dmu_objset_find(spa->spa_name, zil_claim, tx, 0); |
789 | 651 dmu_tx_commit(tx); |
652 | |
653 spa->spa_sync_on = B_TRUE; | |
654 txg_sync_start(spa->spa_dsl_pool); | |
655 | |
656 /* | |
657 * Wait for all claims to sync. | |
658 */ | |
659 txg_wait_synced(spa->spa_dsl_pool, 0); | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
660 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
661 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
662 * If the config cache is stale, or we have uninitialized |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
663 * metaslabs (see spa_vdev_add()), then update the config. |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
664 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
665 if (config_cache_txg != spa->spa_config_txg || |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
666 state == SPA_LOAD_IMPORT) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
667 need_update = B_TRUE; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
668 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
669 for (c = 0; c < rvd->vdev_children; c++) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
670 if (rvd->vdev_child[c]->vdev_ms_array == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
671 need_update = B_TRUE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
672 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
673 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
674 * Update the config cache asychronously in case we're the |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
675 * root pool, in which case the config cache isn't writable yet. |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
676 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
677 if (need_update) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
678 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
789 | 679 } |
680 | |
1544 | 681 error = 0; |
682 out: | |
2082 | 683 if (error && error != EBADF) |
1544 | 684 zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0); |
685 spa->spa_load_state = SPA_LOAD_NONE; | |
686 spa->spa_ena = 0; | |
687 | |
688 return (error); | |
789 | 689 } |
690 | |
691 /* | |
692 * Pool Open/Import | |
693 * | |
694 * The import case is identical to an open except that the configuration is sent | |
695 * down from userland, instead of grabbed from the configuration cache. For the | |
696 * case of an open, the pool configuration will exist in the | |
697 * POOL_STATE_UNITIALIZED state. | |
698 * | |
699 * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
700 * the same time open the pool, without having to keep around the spa_t in some | |
701 * ambiguous state. | |
702 */ | |
703 static int | |
704 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) | |
705 { | |
706 spa_t *spa; | |
707 int error; | |
708 int loaded = B_FALSE; | |
709 int locked = B_FALSE; | |
710 | |
711 *spapp = NULL; | |
712 | |
713 /* | |
714 * As disgusting as this is, we need to support recursive calls to this | |
715 * function because dsl_dir_open() is called during spa_load(), and ends | |
716 * up calling spa_open() again. The real fix is to figure out how to | |
717 * avoid dsl_dir_open() calling this in the first place. | |
718 */ | |
719 if (mutex_owner(&spa_namespace_lock) != curthread) { | |
720 mutex_enter(&spa_namespace_lock); | |
721 locked = B_TRUE; | |
722 } | |
723 | |
724 if ((spa = spa_lookup(pool)) == NULL) { | |
725 if (locked) | |
726 mutex_exit(&spa_namespace_lock); | |
727 return (ENOENT); | |
728 } | |
729 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
730 | |
731 spa_activate(spa); | |
732 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
733 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); |
789 | 734 |
735 if (error == EBADF) { | |
736 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
737 * If vdev_validate() returns failure (indicated by |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
738 * EBADF), it indicates that one of the vdevs indicates |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
739 * that the pool has been exported or destroyed. If |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
740 * this is the case, the config cache is out of sync and |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
741 * we should remove the pool from the namespace. |
789 | 742 */ |
2082 | 743 zfs_post_ok(spa, NULL); |
789 | 744 spa_unload(spa); |
745 spa_deactivate(spa); | |
746 spa_remove(spa); | |
747 spa_config_sync(); | |
748 if (locked) | |
749 mutex_exit(&spa_namespace_lock); | |
750 return (ENOENT); | |
1544 | 751 } |
752 | |
753 if (error) { | |
789 | 754 /* |
755 * We can't open the pool, but we still have useful | |
756 * information: the state of each vdev after the | |
757 * attempted vdev_open(). Return this to the user. | |
758 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
759 if (config != NULL && spa->spa_root_vdev != NULL) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
760 spa_config_enter(spa, RW_READER, FTAG); |
789 | 761 *config = spa_config_generate(spa, NULL, -1ULL, |
762 B_TRUE); | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
763 spa_config_exit(spa, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
764 } |
789 | 765 spa_unload(spa); |
766 spa_deactivate(spa); | |
1544 | 767 spa->spa_last_open_failed = B_TRUE; |
789 | 768 if (locked) |
769 mutex_exit(&spa_namespace_lock); | |
770 *spapp = NULL; | |
771 return (error); | |
1544 | 772 } else { |
773 zfs_post_ok(spa, NULL); | |
774 spa->spa_last_open_failed = B_FALSE; | |
789 | 775 } |
776 | |
777 loaded = B_TRUE; | |
778 } | |
779 | |
780 spa_open_ref(spa, tag); | |
781 if (locked) | |
782 mutex_exit(&spa_namespace_lock); | |
783 | |
784 *spapp = spa; | |
785 | |
786 if (config != NULL) { | |
1544 | 787 spa_config_enter(spa, RW_READER, FTAG); |
789 | 788 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
1544 | 789 spa_config_exit(spa, FTAG); |
789 | 790 } |
791 | |
792 /* | |
793 * If we just loaded the pool, resilver anything that's out of date. | |
794 */ | |
795 if (loaded && (spa_mode & FWRITE)) | |
796 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
797 | |
798 return (0); | |
799 } | |
800 | |
801 int | |
802 spa_open(const char *name, spa_t **spapp, void *tag) | |
803 { | |
804 return (spa_open_common(name, spapp, tag, NULL)); | |
805 } | |
806 | |
1544 | 807 /* |
808 * Lookup the given spa_t, incrementing the inject count in the process, | |
809 * preventing it from being exported or destroyed. | |
810 */ | |
811 spa_t * | |
812 spa_inject_addref(char *name) | |
813 { | |
814 spa_t *spa; | |
815 | |
816 mutex_enter(&spa_namespace_lock); | |
817 if ((spa = spa_lookup(name)) == NULL) { | |
818 mutex_exit(&spa_namespace_lock); | |
819 return (NULL); | |
820 } | |
821 spa->spa_inject_ref++; | |
822 mutex_exit(&spa_namespace_lock); | |
823 | |
824 return (spa); | |
825 } | |
826 | |
827 void | |
828 spa_inject_delref(spa_t *spa) | |
829 { | |
830 mutex_enter(&spa_namespace_lock); | |
831 spa->spa_inject_ref--; | |
832 mutex_exit(&spa_namespace_lock); | |
833 } | |
834 | |
2082 | 835 static void |
836 spa_add_spares(spa_t *spa, nvlist_t *config) | |
837 { | |
838 nvlist_t **spares; | |
839 uint_t i, nspares; | |
840 nvlist_t *nvroot; | |
841 uint64_t guid; | |
842 vdev_stat_t *vs; | |
843 uint_t vsc; | |
844 | |
845 if (spa->spa_nspares == 0) | |
846 return; | |
847 | |
848 VERIFY(nvlist_lookup_nvlist(config, | |
849 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
850 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
851 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
852 if (nspares != 0) { | |
853 VERIFY(nvlist_add_nvlist_array(nvroot, | |
854 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
855 VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
856 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
857 | |
858 /* | |
859 * Go through and find any spares which have since been | |
860 * repurposed as an active spare. If this is the case, update | |
861 * their status appropriately. | |
862 */ | |
863 for (i = 0; i < nspares; i++) { | |
864 VERIFY(nvlist_lookup_uint64(spares[i], | |
865 ZPOOL_CONFIG_GUID, &guid) == 0); | |
866 if (spa_spare_inuse(guid)) { | |
867 VERIFY(nvlist_lookup_uint64_array( | |
868 spares[i], ZPOOL_CONFIG_STATS, | |
869 (uint64_t **)&vs, &vsc) == 0); | |
870 vs->vs_state = VDEV_STATE_CANT_OPEN; | |
871 vs->vs_aux = VDEV_AUX_SPARED; | |
872 } | |
873 } | |
874 } | |
875 } | |
876 | |
789 | 877 int |
1544 | 878 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) |
789 | 879 { |
880 int error; | |
881 spa_t *spa; | |
882 | |
883 *config = NULL; | |
884 error = spa_open_common(name, &spa, FTAG, config); | |
885 | |
2082 | 886 if (spa && *config != NULL) { |
1544 | 887 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, |
888 spa_get_errlog_size(spa)) == 0); | |
889 | |
2082 | 890 spa_add_spares(spa, *config); |
891 } | |
892 | |
1544 | 893 /* |
894 * We want to get the alternate root even for faulted pools, so we cheat | |
895 * and call spa_lookup() directly. | |
896 */ | |
897 if (altroot) { | |
898 if (spa == NULL) { | |
899 mutex_enter(&spa_namespace_lock); | |
900 spa = spa_lookup(name); | |
901 if (spa) | |
902 spa_altroot(spa, altroot, buflen); | |
903 else | |
904 altroot[0] = '\0'; | |
905 spa = NULL; | |
906 mutex_exit(&spa_namespace_lock); | |
907 } else { | |
908 spa_altroot(spa, altroot, buflen); | |
909 } | |
910 } | |
911 | |
789 | 912 if (spa != NULL) |
913 spa_close(spa, FTAG); | |
914 | |
915 return (error); | |
916 } | |
917 | |
918 /* | |
2082 | 919 * Validate that the 'spares' array is well formed. We must have an array of |
920 * nvlists, each which describes a valid leaf vdev. | |
921 */ | |
922 static int | |
923 spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
924 { | |
925 nvlist_t **spares; | |
926 uint_t i, nspares; | |
927 vdev_t *vd; | |
928 int error; | |
929 | |
930 /* | |
931 * It's acceptable to have no spares specified. | |
932 */ | |
933 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
934 &spares, &nspares) != 0) | |
935 return (0); | |
936 | |
937 if (nspares == 0) | |
938 return (EINVAL); | |
939 | |
940 /* | |
941 * Make sure the pool is formatted with a version that supports hot | |
942 * spares. | |
943 */ | |
944 if (spa_version(spa) < ZFS_VERSION_SPARES) | |
945 return (ENOTSUP); | |
946 | |
947 for (i = 0; i < nspares; i++) { | |
948 if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
949 mode)) != 0) | |
950 return (error); | |
951 | |
952 if (!vd->vdev_ops->vdev_op_leaf) { | |
953 vdev_free(vd); | |
954 return (EINVAL); | |
955 } | |
956 | |
957 if ((error = vdev_open(vd)) != 0) { | |
958 vdev_free(vd); | |
959 return (error); | |
960 } | |
961 | |
962 vd->vdev_top = vd; | |
963 if ((error = vdev_label_spare(vd, crtxg)) != 0) { | |
964 vdev_free(vd); | |
965 return (error); | |
966 } | |
967 | |
968 VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID, | |
969 vd->vdev_guid) == 0); | |
970 | |
971 vdev_free(vd); | |
972 } | |
973 | |
974 return (0); | |
975 } | |
976 | |
977 /* | |
789 | 978 * Pool Creation |
979 */ | |
980 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
981 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot) |
789 | 982 { |
983 spa_t *spa; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
984 vdev_t *rvd; |
789 | 985 dsl_pool_t *dp; |
986 dmu_tx_t *tx; | |
2082 | 987 int c, error = 0; |
789 | 988 uint64_t txg = TXG_INITIAL; |
2082 | 989 nvlist_t **spares; |
990 uint_t nspares; | |
789 | 991 |
992 /* | |
993 * If this pool already exists, return failure. | |
994 */ | |
995 mutex_enter(&spa_namespace_lock); | |
996 if (spa_lookup(pool) != NULL) { | |
997 mutex_exit(&spa_namespace_lock); | |
998 return (EEXIST); | |
999 } | |
1000 | |
1001 /* | |
1002 * Allocate a new spa_t structure. | |
1003 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1004 spa = spa_add(pool, altroot); |
789 | 1005 spa_activate(spa); |
1006 | |
1007 spa->spa_uberblock.ub_txg = txg - 1; | |
1760 | 1008 spa->spa_uberblock.ub_version = ZFS_VERSION; |
789 | 1009 spa->spa_ubsync = spa->spa_uberblock; |
1010 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1011 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1012 * Create the root vdev. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1013 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1014 spa_config_enter(spa, RW_WRITER, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1015 |
2082 | 1016 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); |
1017 | |
1018 ASSERT(error != 0 || rvd != NULL); | |
1019 ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
1020 | |
1021 if (error == 0 && rvd->vdev_children == 0) | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1022 error = EINVAL; |
2082 | 1023 |
1024 if (error == 0 && | |
1025 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
1026 (error = spa_validate_spares(spa, nvroot, txg, | |
1027 VDEV_ALLOC_ADD)) == 0) { | |
1028 for (c = 0; c < rvd->vdev_children; c++) | |
1029 vdev_init(rvd->vdev_child[c], txg); | |
1030 vdev_config_dirty(rvd); | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1031 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1032 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1033 spa_config_exit(spa, FTAG); |
789 | 1034 |
2082 | 1035 if (error != 0) { |
789 | 1036 spa_unload(spa); |
1037 spa_deactivate(spa); | |
1038 spa_remove(spa); | |
1039 mutex_exit(&spa_namespace_lock); | |
1040 return (error); | |
1041 } | |
1042 | |
2082 | 1043 /* |
1044 * Get the list of spares, if specified. | |
1045 */ | |
1046 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
1047 &spares, &nspares) == 0) { | |
1048 VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME, | |
1049 KM_SLEEP) == 0); | |
1050 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1051 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1052 spa_config_enter(spa, RW_WRITER, FTAG); | |
1053 spa_load_spares(spa); | |
1054 spa_config_exit(spa, FTAG); | |
1055 spa->spa_sync_spares = B_TRUE; | |
1056 } | |
1057 | |
789 | 1058 spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg); |
1059 spa->spa_meta_objset = dp->dp_meta_objset; | |
1060 | |
1061 tx = dmu_tx_create_assigned(dp, txg); | |
1062 | |
1063 /* | |
1064 * Create the pool config object. | |
1065 */ | |
1066 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
1067 DMU_OT_PACKED_NVLIST, 1 << 14, | |
1068 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
1069 | |
1544 | 1070 if (zap_add(spa->spa_meta_objset, |
789 | 1071 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 1072 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { |
1073 cmn_err(CE_PANIC, "failed to add pool config"); | |
1074 } | |
789 | 1075 |
2082 | 1076 /* Newly created pools are always deflated. */ |
1077 spa->spa_deflate = TRUE; | |
1078 if (zap_add(spa->spa_meta_objset, | |
1079 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
1080 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
1081 cmn_err(CE_PANIC, "failed to add deflate"); | |
1082 } | |
1083 | |
789 | 1084 /* |
1085 * Create the deferred-free bplist object. Turn off compression | |
1086 * because sync-to-convergence takes longer if the blocksize | |
1087 * keeps changing. | |
1088 */ | |
1089 spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, | |
1090 1 << 14, tx); | |
1091 dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, | |
1092 ZIO_COMPRESS_OFF, tx); | |
1093 | |
1544 | 1094 if (zap_add(spa->spa_meta_objset, |
789 | 1095 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, |
1544 | 1096 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { |
1097 cmn_err(CE_PANIC, "failed to add bplist"); | |
1098 } | |
789 | 1099 |
1100 dmu_tx_commit(tx); | |
1101 | |
1102 spa->spa_sync_on = B_TRUE; | |
1103 txg_sync_start(spa->spa_dsl_pool); | |
1104 | |
1105 /* | |
1106 * We explicitly wait for the first transaction to complete so that our | |
1107 * bean counters are appropriately updated. | |
1108 */ | |
1109 txg_wait_synced(spa->spa_dsl_pool, txg); | |
1110 | |
1111 spa_config_sync(); | |
1112 | |
1113 mutex_exit(&spa_namespace_lock); | |
1114 | |
1115 return (0); | |
1116 } | |
1117 | |
1118 /* | |
1119 * Import the given pool into the system. We set up the necessary spa_t and | |
1120 * then call spa_load() to do the dirty work. | |
1121 */ | |
1122 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1123 spa_import(const char *pool, nvlist_t *config, const char *altroot) |
789 | 1124 { |
1125 spa_t *spa; | |
1126 int error; | |
2082 | 1127 nvlist_t *nvroot; |
1128 nvlist_t **spares; | |
1129 uint_t nspares; | |
789 | 1130 |
1131 if (!(spa_mode & FWRITE)) | |
1132 return (EROFS); | |
1133 | |
1134 /* | |
1135 * If a pool with this name exists, return failure. | |
1136 */ | |
1137 mutex_enter(&spa_namespace_lock); | |
1138 if (spa_lookup(pool) != NULL) { | |
1139 mutex_exit(&spa_namespace_lock); | |
1140 return (EEXIST); | |
1141 } | |
1142 | |
1143 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1144 * Create and initialize the spa structure. |
789 | 1145 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1146 spa = spa_add(pool, altroot); |
789 | 1147 spa_activate(spa); |
1148 | |
1149 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1150 * Pass off the heavy lifting to spa_load(). |
1732 | 1151 * Pass TRUE for mosconfig because the user-supplied config |
1152 * is actually the one to trust when doing an import. | |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1153 */ |
1732 | 1154 error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE); |
789 | 1155 |
2082 | 1156 spa_config_enter(spa, RW_WRITER, FTAG); |
1157 /* | |
1158 * Toss any existing sparelist, as it doesn't have any validity anymore, | |
1159 * and conflicts with spa_has_spare(). | |
1160 */ | |
1161 if (spa->spa_sparelist) { | |
1162 nvlist_free(spa->spa_sparelist); | |
1163 spa->spa_sparelist = NULL; | |
1164 spa_load_spares(spa); | |
1165 } | |
1166 | |
1167 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, | |
1168 &nvroot) == 0); | |
1169 if (error == 0) | |
1170 error = spa_validate_spares(spa, nvroot, -1ULL, | |
1171 VDEV_ALLOC_SPARE); | |
1172 spa_config_exit(spa, FTAG); | |
1173 | |
1174 if (error != 0) { | |
789 | 1175 spa_unload(spa); |
1176 spa_deactivate(spa); | |
1177 spa_remove(spa); | |
1178 mutex_exit(&spa_namespace_lock); | |
1179 return (error); | |
1180 } | |
1181 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1182 /* |
2082 | 1183 * Override any spares as specified by the user, as these may have |
1184 * correct device names/devids, etc. | |
1185 */ | |
1186 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
1187 &spares, &nspares) == 0) { | |
1188 if (spa->spa_sparelist) | |
1189 VERIFY(nvlist_remove(spa->spa_sparelist, | |
1190 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
1191 else | |
1192 VERIFY(nvlist_alloc(&spa->spa_sparelist, | |
1193 NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1194 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1195 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1196 spa_config_enter(spa, RW_WRITER, FTAG); | |
1197 spa_load_spares(spa); | |
1198 spa_config_exit(spa, FTAG); | |
1199 spa->spa_sync_spares = B_TRUE; | |
1200 } | |
1201 | |
1202 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1203 * Update the config cache to include the newly-imported pool. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1204 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1205 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1206 |
789 | 1207 mutex_exit(&spa_namespace_lock); |
1208 | |
1209 /* | |
1210 * Resilver anything that's out of date. | |
1211 */ | |
1212 if (spa_mode & FWRITE) | |
1213 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1214 | |
1215 return (0); | |
1216 } | |
1217 | |
1218 /* | |
1219 * This (illegal) pool name is used when temporarily importing a spa_t in order | |
1220 * to get the vdev stats associated with the imported devices. | |
1221 */ | |
1222 #define TRYIMPORT_NAME "$import" | |
1223 | |
1224 nvlist_t * | |
1225 spa_tryimport(nvlist_t *tryconfig) | |
1226 { | |
1227 nvlist_t *config = NULL; | |
1228 char *poolname; | |
1229 spa_t *spa; | |
1230 uint64_t state; | |
1231 | |
1232 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
1233 return (NULL); | |
1234 | |
1235 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
1236 return (NULL); | |
1237 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1238 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1239 * Create and initialize the spa structure. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1240 */ |
789 | 1241 mutex_enter(&spa_namespace_lock); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1242 spa = spa_add(TRYIMPORT_NAME, NULL); |
789 | 1243 spa_activate(spa); |
1244 | |
1245 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1246 * Pass off the heavy lifting to spa_load(). |
1732 | 1247 * Pass TRUE for mosconfig because the user-supplied config |
1248 * is actually the one to trust when doing an import. | |
789 | 1249 */ |
1732 | 1250 (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); |
789 | 1251 |
1252 /* | |
1253 * If 'tryconfig' was at least parsable, return the current config. | |
1254 */ | |
1255 if (spa->spa_root_vdev != NULL) { | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1256 spa_config_enter(spa, RW_READER, FTAG); |
789 | 1257 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1258 spa_config_exit(spa, FTAG); |
789 | 1259 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, |
1260 poolname) == 0); | |
1261 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
1262 state) == 0); | |
2082 | 1263 |
1264 /* | |
1265 * Add the list of hot spares. | |
1266 */ | |
1267 spa_add_spares(spa, config); | |
789 | 1268 } |
1269 | |
1270 spa_unload(spa); | |
1271 spa_deactivate(spa); | |
1272 spa_remove(spa); | |
1273 mutex_exit(&spa_namespace_lock); | |
1274 | |
1275 return (config); | |
1276 } | |
1277 | |
1278 /* | |
1279 * Pool export/destroy | |
1280 * | |
1281 * The act of destroying or exporting a pool is very simple. We make sure there | |
1282 * is no more pending I/O and any references to the pool are gone. Then, we | |
1283 * update the pool state and sync all the labels to disk, removing the | |
1284 * configuration from the cache afterwards. | |
1285 */ | |
1286 static int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1287 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig) |
789 | 1288 { |
1289 spa_t *spa; | |
1290 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1291 if (oldconfig) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1292 *oldconfig = NULL; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1293 |
789 | 1294 if (!(spa_mode & FWRITE)) |
1295 return (EROFS); | |
1296 | |
1297 mutex_enter(&spa_namespace_lock); | |
1298 if ((spa = spa_lookup(pool)) == NULL) { | |
1299 mutex_exit(&spa_namespace_lock); | |
1300 return (ENOENT); | |
1301 } | |
1302 | |
1303 /* | |
1544 | 1304 * Put a hold on the pool, drop the namespace lock, stop async tasks, |
1305 * reacquire the namespace lock, and see if we can export. | |
1306 */ | |
1307 spa_open_ref(spa, FTAG); | |
1308 mutex_exit(&spa_namespace_lock); | |
1309 spa_async_suspend(spa); | |
1310 mutex_enter(&spa_namespace_lock); | |
1311 spa_close(spa, FTAG); | |
1312 | |
1313 /* | |
789 | 1314 * The pool will be in core if it's openable, |
1315 * in which case we can modify its state. | |
1316 */ | |
1317 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { | |
1318 /* | |
1319 * Objsets may be open only because they're dirty, so we | |
1320 * have to force it to sync before checking spa_refcnt. | |
1321 */ | |
1322 spa_scrub_suspend(spa); | |
1323 txg_wait_synced(spa->spa_dsl_pool, 0); | |
1324 | |
1544 | 1325 /* |
1326 * A pool cannot be exported or destroyed if there are active | |
1327 * references. If we are resetting a pool, allow references by | |
1328 * fault injection handlers. | |
1329 */ | |
1330 if (!spa_refcount_zero(spa) || | |
1331 (spa->spa_inject_ref != 0 && | |
1332 new_state != POOL_STATE_UNINITIALIZED)) { | |
789 | 1333 spa_scrub_resume(spa); |
1544 | 1334 spa_async_resume(spa); |
789 | 1335 mutex_exit(&spa_namespace_lock); |
1336 return (EBUSY); | |
1337 } | |
1338 | |
1339 spa_scrub_resume(spa); | |
1340 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); | |
1341 | |
1342 /* | |
1343 * We want this to be reflected on every label, | |
1344 * so mark them all dirty. spa_unload() will do the | |
1345 * final sync that pushes these changes out. | |
1346 */ | |
1544 | 1347 if (new_state != POOL_STATE_UNINITIALIZED) { |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1348 spa_config_enter(spa, RW_WRITER, FTAG); |
1544 | 1349 spa->spa_state = new_state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1350 spa->spa_final_txg = spa_last_synced_txg(spa) + 1; |
1544 | 1351 vdev_config_dirty(spa->spa_root_vdev); |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1352 spa_config_exit(spa, FTAG); |
1544 | 1353 } |
789 | 1354 } |
1355 | |
1356 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
1357 spa_unload(spa); | |
1358 spa_deactivate(spa); | |
1359 } | |
1360 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1361 if (oldconfig && spa->spa_config) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1362 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1363 |
1544 | 1364 if (new_state != POOL_STATE_UNINITIALIZED) { |
1365 spa_remove(spa); | |
1366 spa_config_sync(); | |
1367 } | |
789 | 1368 mutex_exit(&spa_namespace_lock); |
1369 | |
1370 return (0); | |
1371 } | |
1372 | |
1373 /* | |
1374 * Destroy a storage pool. | |
1375 */ | |
1376 int | |
1377 spa_destroy(char *pool) | |
1378 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1379 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL)); |
789 | 1380 } |
1381 | |
1382 /* | |
1383 * Export a storage pool. | |
1384 */ | |
1385 int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1386 spa_export(char *pool, nvlist_t **oldconfig) |
789 | 1387 { |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1388 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig)); |
789 | 1389 } |
1390 | |
1391 /* | |
1544 | 1392 * Similar to spa_export(), this unloads the spa_t without actually removing it |
1393 * from the namespace in any way. | |
1394 */ | |
1395 int | |
1396 spa_reset(char *pool) | |
1397 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1398 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL)); |
1544 | 1399 } |
1400 | |
1401 | |
1402 /* | |
789 | 1403 * ========================================================================== |
1404 * Device manipulation | |
1405 * ========================================================================== | |
1406 */ | |
1407 | |
1408 /* | |
1409 * Add capacity to a storage pool. | |
1410 */ | |
1411 int | |
1412 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
1413 { | |
1414 uint64_t txg; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1415 int c, error; |
789 | 1416 vdev_t *rvd = spa->spa_root_vdev; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1417 vdev_t *vd, *tvd; |
2082 | 1418 nvlist_t **spares; |
1419 uint_t i, nspares; | |
789 | 1420 |
1421 txg = spa_vdev_enter(spa); | |
1422 | |
2082 | 1423 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, |
1424 VDEV_ALLOC_ADD)) != 0) | |
1425 return (spa_vdev_exit(spa, NULL, txg, error)); | |
1426 | |
1427 if ((error = spa_validate_spares(spa, nvroot, txg, | |
1428 VDEV_ALLOC_ADD)) != 0) | |
789 | 1429 return (spa_vdev_exit(spa, vd, txg, error)); |
1430 | |
2082 | 1431 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
1432 &spares, &nspares) != 0) | |
1433 nspares = 0; | |
1434 | |
1435 if (vd->vdev_children == 0 && nspares == 0) | |
1436 return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
1437 | |
1438 if (vd->vdev_children != 0) { | |
1439 if ((error = vdev_create(vd, txg, B_FALSE)) != 0) | |
1440 return (spa_vdev_exit(spa, vd, txg, error)); | |
1441 | |
1442 /* | |
1443 * Transfer each new top-level vdev from vd to rvd. | |
1444 */ | |
1445 for (c = 0; c < vd->vdev_children; c++) { | |
1446 tvd = vd->vdev_child[c]; | |
1447 vdev_remove_child(vd, tvd); | |
1448 tvd->vdev_id = rvd->vdev_children; | |
1449 vdev_add_child(rvd, tvd); | |
1450 vdev_config_dirty(tvd); | |
1451 } | |
1452 } | |
1453 | |
1454 if (nspares != 0) { | |
1455 if (spa->spa_sparelist != NULL) { | |
1456 nvlist_t **oldspares; | |
1457 uint_t oldnspares; | |
1458 nvlist_t **newspares; | |
1459 | |
1460 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
1461 ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0); | |
1462 | |
1463 newspares = kmem_alloc(sizeof (void *) * | |
1464 (nspares + oldnspares), KM_SLEEP); | |
1465 for (i = 0; i < oldnspares; i++) | |
1466 VERIFY(nvlist_dup(oldspares[i], | |
1467 &newspares[i], KM_SLEEP) == 0); | |
1468 for (i = 0; i < nspares; i++) | |
1469 VERIFY(nvlist_dup(spares[i], | |
1470 &newspares[i + oldnspares], | |
1471 KM_SLEEP) == 0); | |
1472 | |
1473 VERIFY(nvlist_remove(spa->spa_sparelist, | |
1474 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
1475 | |
1476 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1477 ZPOOL_CONFIG_SPARES, newspares, | |
1478 nspares + oldnspares) == 0); | |
1479 for (i = 0; i < oldnspares + nspares; i++) | |
1480 nvlist_free(newspares[i]); | |
1481 kmem_free(newspares, (oldnspares + nspares) * | |
1482 sizeof (void *)); | |
1483 } else { | |
1484 VERIFY(nvlist_alloc(&spa->spa_sparelist, | |
1485 NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1486 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1487 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1488 } | |
1489 | |
1490 spa_load_spares(spa); | |
1491 spa->spa_sync_spares = B_TRUE; | |
789 | 1492 } |
1493 | |
1494 /* | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1495 * We have to be careful when adding new vdevs to an existing pool. |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1496 * If other threads start allocating from these vdevs before we |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1497 * sync the config cache, and we lose power, then upon reboot we may |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1498 * fail to open the pool because there are DVAs that the config cache |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1499 * can't translate. Therefore, we first add the vdevs without |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1500 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1501 * and then let spa_config_update() initialize the new metaslabs. |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1502 * |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1503 * spa_load() checks for added-but-not-initialized vdevs, so that |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1504 * if we lose power at any point in this sequence, the remaining |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1505 * steps will be completed the next time we load the pool. |
789 | 1506 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1507 (void) spa_vdev_exit(spa, vd, txg, 0); |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1508 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1509 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1510 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1511 mutex_exit(&spa_namespace_lock); |
789 | 1512 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1513 return (0); |
789 | 1514 } |
1515 | |
1516 /* | |
1517 * Attach a device to a mirror. The arguments are the path to any device | |
1518 * in the mirror, and the nvroot for the new device. If the path specifies | |
1519 * a device that is not mirrored, we automatically insert the mirror vdev. | |
1520 * | |
1521 * If 'replacing' is specified, the new device is intended to replace the | |
1522 * existing device; in this case the two devices are made into their own | |
1523 * mirror using the 'replacing' vdev, which is functionally idendical to | |
1524 * the mirror vdev (it actually reuses all the same ops) but has a few | |
1525 * extra rules: you can't attach to it after it's been created, and upon | |
1526 * completion of resilvering, the first disk (the one being replaced) | |
1527 * is automatically detached. | |
1528 */ | |
1529 int | |
1544 | 1530 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) |
789 | 1531 { |
1532 uint64_t txg, open_txg; | |
1533 int error; | |
1534 vdev_t *rvd = spa->spa_root_vdev; | |
1535 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
2082 | 1536 vdev_ops_t *pvops; |
789 | 1537 |
1538 txg = spa_vdev_enter(spa); | |
1539 | |
1544 | 1540 oldvd = vdev_lookup_by_guid(rvd, guid); |
789 | 1541 |
1542 if (oldvd == NULL) | |
1543 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1544 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1545 if (!oldvd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1546 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1547 |
789 | 1548 pvd = oldvd->vdev_parent; |
1549 | |
2082 | 1550 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, |
1551 VDEV_ALLOC_ADD)) != 0 || newrootvd->vdev_children != 1) | |
789 | 1552 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); |
1553 | |
1554 newvd = newrootvd->vdev_child[0]; | |
1555 | |
1556 if (!newvd->vdev_ops->vdev_op_leaf) | |
1557 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
1558 | |
2082 | 1559 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) |
789 | 1560 return (spa_vdev_exit(spa, newrootvd, txg, error)); |
1561 | |
2082 | 1562 if (!replacing) { |
1563 /* | |
1564 * For attach, the only allowable parent is a mirror or the root | |
1565 * vdev. | |
1566 */ | |
1567 if (pvd->vdev_ops != &vdev_mirror_ops && | |
1568 pvd->vdev_ops != &vdev_root_ops) | |
1569 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1570 | |
1571 pvops = &vdev_mirror_ops; | |
1572 } else { | |
1573 /* | |
1574 * Active hot spares can only be replaced by inactive hot | |
1575 * spares. | |
1576 */ | |
1577 if (pvd->vdev_ops == &vdev_spare_ops && | |
1578 pvd->vdev_child[1] == oldvd && | |
1579 !spa_has_spare(spa, newvd->vdev_guid)) | |
1580 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1581 | |
1582 /* | |
1583 * If the source is a hot spare, and the parent isn't already a | |
1584 * spare, then we want to create a new hot spare. Otherwise, we | |
1585 * want to create a replacing vdev. | |
1586 */ | |
1587 if (pvd->vdev_ops == &vdev_replacing_ops) | |
1588 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1589 else if (pvd->vdev_ops != &vdev_spare_ops && | |
1590 newvd->vdev_isspare) | |
1591 pvops = &vdev_spare_ops; | |
1592 else | |
1593 pvops = &vdev_replacing_ops; | |
1594 } | |
1595 | |
1175
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1596 /* |
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1597 * Compare the new device size with the replaceable/attachable |
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1598 * device size. |
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1599 */ |
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1600 if (newvd->vdev_psize < vdev_get_rsize(oldvd)) |
789 | 1601 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
1602 | |
1732 | 1603 /* |
1604 * The new device cannot have a higher alignment requirement | |
1605 * than the top-level vdev. | |
1606 */ | |
1607 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
789 | 1608 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); |
1609 | |
1610 /* | |
1611 * If this is an in-place replacement, update oldvd's path and devid | |
1612 * to make it distinguishable from newvd, and unopenable from now on. | |
1613 */ | |
1614 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
1615 spa_strfree(oldvd->vdev_path); | |
1616 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
1617 KM_SLEEP); | |
1618 (void) sprintf(oldvd->vdev_path, "%s/%s", | |
1619 newvd->vdev_path, "old"); | |
1620 if (oldvd->vdev_devid != NULL) { | |
1621 spa_strfree(oldvd->vdev_devid); | |
1622 oldvd->vdev_devid = NULL; | |
1623 } | |
1624 } | |
1625 | |
1626 /* | |
2082 | 1627 * If the parent is not a mirror, or if we're replacing, insert the new |
1628 * mirror/replacing/spare vdev above oldvd. | |
789 | 1629 */ |
1630 if (pvd->vdev_ops != pvops) | |
1631 pvd = vdev_add_parent(oldvd, pvops); | |
1632 | |
1633 ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
1634 ASSERT(pvd->vdev_ops == pvops); | |
1635 ASSERT(oldvd->vdev_parent == pvd); | |
1636 | |
1637 /* | |
1638 * Extract the new device from its root and add it to pvd. | |
1639 */ | |
1640 vdev_remove_child(newrootvd, newvd); | |
1641 newvd->vdev_id = pvd->vdev_children; | |
1642 vdev_add_child(pvd, newvd); | |
1643 | |
1544 | 1644 /* |
1645 * If newvd is smaller than oldvd, but larger than its rsize, | |
1646 * the addition of newvd may have decreased our parent's asize. | |
1647 */ | |
1648 pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); | |
1649 | |
789 | 1650 tvd = newvd->vdev_top; |
1651 ASSERT(pvd->vdev_top == tvd); | |
1652 ASSERT(tvd->vdev_parent == rvd); | |
1653 | |
1654 vdev_config_dirty(tvd); | |
1655 | |
1656 /* | |
1657 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate | |
1658 * upward when spa_vdev_exit() calls vdev_dtl_reassess(). | |
1659 */ | |
1660 open_txg = txg + TXG_CONCURRENT_STATES - 1; | |
1661 | |
1662 mutex_enter(&newvd->vdev_dtl_lock); | |
1663 space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, | |
1664 open_txg - TXG_INITIAL + 1); | |
1665 mutex_exit(&newvd->vdev_dtl_lock); | |
1666 | |
1544 | 1667 dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg); |
1668 | |
789 | 1669 /* |
1670 * Mark newvd's DTL dirty in this txg. | |
1671 */ | |
1732 | 1672 vdev_dirty(tvd, VDD_DTL, newvd, txg); |
789 | 1673 |
1674 (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); | |
1675 | |
1676 /* | |
1677 * Kick off a resilver to update newvd. | |
1678 */ | |
1679 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1680 | |
1681 return (0); | |
1682 } | |
1683 | |
1684 /* | |
1685 * Detach a device from a mirror or replacing vdev. | |
1686 * If 'replace_done' is specified, only detach if the parent | |
1687 * is a replacing vdev. | |
1688 */ | |
1689 int | |
1544 | 1690 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) |
789 | 1691 { |
1692 uint64_t txg; | |
1693 int c, t, error; | |
1694 vdev_t *rvd = spa->spa_root_vdev; | |
1695 vdev_t *vd, *pvd, *cvd, *tvd; | |
2082 | 1696 boolean_t unspare = B_FALSE; |
1697 uint64_t unspare_guid; | |
789 | 1698 |
1699 txg = spa_vdev_enter(spa); | |
1700 | |
1544 | 1701 vd = vdev_lookup_by_guid(rvd, guid); |
789 | 1702 |
1703 if (vd == NULL) | |
1704 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1705 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1706 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1707 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1708 |
789 | 1709 pvd = vd->vdev_parent; |
1710 | |
1711 /* | |
1712 * If replace_done is specified, only remove this device if it's | |
2082 | 1713 * the first child of a replacing vdev. For the 'spare' vdev, either |
1714 * disk can be removed. | |
789 | 1715 */ |
2082 | 1716 if (replace_done) { |
1717 if (pvd->vdev_ops == &vdev_replacing_ops) { | |
1718 if (vd->vdev_id != 0) | |
1719 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1720 } else if (pvd->vdev_ops != &vdev_spare_ops) { | |
1721 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1722 } | |
1723 } | |
1724 | |
1725 ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
1726 spa_version(spa) >= ZFS_VERSION_SPARES); | |
789 | 1727 |
1728 /* | |
2082 | 1729 * Only mirror, replacing, and spare vdevs support detach. |
789 | 1730 */ |
1731 if (pvd->vdev_ops != &vdev_replacing_ops && | |
2082 | 1732 pvd->vdev_ops != &vdev_mirror_ops && |
1733 pvd->vdev_ops != &vdev_spare_ops) | |
789 | 1734 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
1735 | |
1736 /* | |
1737 * If there's only one replica, you can't detach it. | |
1738 */ | |
1739 if (pvd->vdev_children <= 1) | |
1740 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
1741 | |
1742 /* | |
1743 * If all siblings have non-empty DTLs, this device may have the only | |
1744 * valid copy of the data, which means we cannot safely detach it. | |
1745 * | |
1746 * XXX -- as in the vdev_offline() case, we really want a more | |
1747 * precise DTL check. | |
1748 */ | |
1749 for (c = 0; c < pvd->vdev_children; c++) { | |
1750 uint64_t dirty; | |
1751 | |
1752 cvd = pvd->vdev_child[c]; | |
1753 if (cvd == vd) | |
1754 continue; | |
1755 if (vdev_is_dead(cvd)) | |
1756 continue; | |
1757 mutex_enter(&cvd->vdev_dtl_lock); | |
1758 dirty = cvd->vdev_dtl_map.sm_space | | |
1759 cvd->vdev_dtl_scrub.sm_space; | |
1760 mutex_exit(&cvd->vdev_dtl_lock); | |
1761 if (!dirty) | |
1762 break; | |
1763 } | |
2082 | 1764 |
1765 /* | |
1766 * If we are a replacing or spare vdev, then we can always detach the | |
1767 * latter child, as that is how one cancels the operation. | |
1768 */ | |
1769 if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) && | |
1770 c == pvd->vdev_children) | |
789 | 1771 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
1772 | |
1773 /* | |
2082 | 1774 * If we are detaching the original disk from a spare, then it implies |
1775 * that the spare should become a real disk, and be removed from the | |
1776 * active spare list for the pool. | |
1777 */ | |
1778 if (pvd->vdev_ops == &vdev_spare_ops && | |
1779 vd->vdev_id == 0) | |
1780 unspare = B_TRUE; | |
1781 | |
1782 /* | |
789 | 1783 * Erase the disk labels so the disk can be used for other things. |
1784 * This must be done after all other error cases are handled, | |
1785 * but before we disembowel vd (so we can still do I/O to it). | |
1786 * But if we can't do it, don't treat the error as fatal -- | |
1787 * it may be that the unwritability of the disk is the reason | |
1788 * it's being detached! | |
1789 */ | |
2082 | 1790 error = vdev_label_init(vd, 0, B_FALSE); |
789 | 1791 if (error) |
1792 dprintf("unable to erase labels on %s\n", vdev_description(vd)); | |
1793 | |
1794 /* | |
1795 * Remove vd from its parent and compact the parent's children. | |
1796 */ | |
1797 vdev_remove_child(pvd, vd); | |
1798 vdev_compact_children(pvd); | |
1799 | |
1800 /* | |
1801 * Remember one of the remaining children so we can get tvd below. | |
1802 */ | |
1803 cvd = pvd->vdev_child[0]; | |
1804 | |
1805 /* | |
2082 | 1806 * If we need to remove the remaining child from the list of hot spares, |
1807 * do it now, marking the vdev as no longer a spare in the process. We | |
1808 * must do this before vdev_remove_parent(), because that can change the | |
1809 * GUID if it creates a new toplevel GUID. | |
1810 */ | |
1811 if (unspare) { | |
1812 ASSERT(cvd->vdev_isspare); | |
1813 spa_spare_remove(cvd->vdev_guid); | |
1814 cvd->vdev_isspare = B_FALSE; | |
1815 unspare_guid = cvd->vdev_guid; | |
1816 } | |
1817 | |
1818 /* | |
789 | 1819 * If the parent mirror/replacing vdev only has one child, |
1820 * the parent is no longer needed. Remove it from the tree. | |
1821 */ | |
1822 if (pvd->vdev_children == 1) | |
1823 vdev_remove_parent(cvd); | |
1824 | |
1825 /* | |
1826 * We don't set tvd until now because the parent we just removed | |
1827 * may have been the previous top-level vdev. | |
1828 */ | |
1829 tvd = cvd->vdev_top; | |
1830 ASSERT(tvd->vdev_parent == rvd); | |
1831 | |
1832 /* | |
1833 * Reopen this top-level vdev to reassess health after detach. | |
1834 */ | |
1544 | 1835 vdev_reopen(tvd); |
789 | 1836 |
1837 /* | |
1838 * If the device we just detached was smaller than the others, | |
1732 | 1839 * it may be possible to add metaslabs (i.e. grow the pool). |
1840 * vdev_metaslab_init() can't fail because the existing metaslabs | |
1841 * are already in core, so there's nothing to read from disk. | |
789 | 1842 */ |
1732 | 1843 VERIFY(vdev_metaslab_init(tvd, txg) == 0); |
789 | 1844 |
1845 vdev_config_dirty(tvd); | |
1846 | |
1847 /* | |
1848 * Mark vd's DTL as dirty in this txg. | |
1849 * vdev_dtl_sync() will see that vd->vdev_detached is set | |
1850 * and free vd's DTL object in syncing context. | |
1851 * But first make sure we're not on any *other* txg's DTL list, | |
1852 * to prevent vd from being accessed after it's freed. | |
1853 */ | |
1854 for (t = 0; t < TXG_SIZE; t++) | |
1855 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
1732 | 1856 vd->vdev_detached = B_TRUE; |
1857 vdev_dirty(tvd, VDD_DTL, vd, txg); | |
789 | 1858 |
1544 | 1859 dprintf("detached %s in txg %llu\n", vd->vdev_path, txg); |
789 | 1860 |
2082 | 1861 error = spa_vdev_exit(spa, vd, txg, 0); |
1862 | |
1863 /* | |
1864 * If we are supposed to remove the given vdev from the list of spares, | |
1865 * iterate over all pools in the system and replace it if it's present. | |
1866 */ | |
1867 if (unspare) { | |
1868 spa = NULL; | |
1869 mutex_enter(&spa_namespace_lock); | |
1870 while ((spa = spa_next(spa)) != NULL) { | |
1871 if (spa->spa_state != POOL_STATE_ACTIVE) | |
1872 continue; | |
1873 | |
1874 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); | |
1875 } | |
1876 mutex_exit(&spa_namespace_lock); | |
1877 } | |
1878 | |
1879 return (error); | |
1880 } | |
1881 | |
1882 /* | |
1883 * Remove a device from the pool. Currently, this supports removing only hot | |
1884 * spares. | |
1885 */ | |
1886 int | |
1887 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) | |
1888 { | |
1889 vdev_t *vd; | |
1890 nvlist_t **spares, *nv, **newspares; | |
1891 uint_t i, j, nspares; | |
1892 int ret = 0; | |
1893 | |
1894 spa_config_enter(spa, RW_WRITER, FTAG); | |
1895 | |
1896 vd = spa_lookup_by_guid(spa, guid); | |
1897 | |
1898 nv = NULL; | |
1899 if (spa->spa_spares != NULL && | |
1900 nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1901 &spares, &nspares) == 0) { | |
1902 for (i = 0; i < nspares; i++) { | |
1903 uint64_t theguid; | |
1904 | |
1905 VERIFY(nvlist_lookup_uint64(spares[i], | |
1906 ZPOOL_CONFIG_GUID, &theguid) == 0); | |
1907 if (theguid == guid) { | |
1908 nv = spares[i]; | |
1909 break; | |
1910 } | |
1911 } | |
1912 } | |
1913 | |
1914 /* | |
1915 * We only support removing a hot spare, and only if it's not currently | |
1916 * in use in this pool. | |
1917 */ | |
1918 if (nv == NULL && vd == NULL) { | |
1919 ret = ENOENT; | |
1920 goto out; | |
1921 } | |
1922 | |
1923 if (nv == NULL && vd != NULL) { | |
1924 ret = ENOTSUP; | |
1925 goto out; | |
1926 } | |
1927 | |
1928 if (!unspare && nv != NULL && vd != NULL) { | |
1929 ret = EBUSY; | |
1930 goto out; | |
1931 } | |
1932 | |
1933 if (nspares == 1) { | |
1934 newspares = NULL; | |
1935 } else { | |
1936 newspares = kmem_alloc((nspares - 1) * sizeof (void *), | |
1937 KM_SLEEP); | |
1938 for (i = 0, j = 0; i < nspares; i++) { | |
1939 if (spares[i] != nv) | |
1940 VERIFY(nvlist_dup(spares[i], | |
1941 &newspares[j++], KM_SLEEP) == 0); | |
1942 } | |
1943 } | |
1944 | |
1945 VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1946 DATA_TYPE_NVLIST_ARRAY) == 0); | |
1947 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1948 newspares, nspares - 1) == 0); | |
1949 for (i = 0; i < nspares - 1; i++) | |
1950 nvlist_free(newspares[i]); | |
1951 kmem_free(newspares, (nspares - 1) * sizeof (void *)); | |
1952 spa_load_spares(spa); | |
1953 spa->spa_sync_spares = B_TRUE; | |
1954 | |
1955 out: | |
1956 spa_config_exit(spa, FTAG); | |
1957 | |
1958 return (ret); | |
789 | 1959 } |
1960 | |
1961 /* | |
1544 | 1962 * Find any device that's done replacing, so we can detach it. |
789 | 1963 */ |
1544 | 1964 static vdev_t * |
1965 spa_vdev_replace_done_hunt(vdev_t *vd) | |
789 | 1966 { |
1544 | 1967 vdev_t *newvd, *oldvd; |
789 | 1968 int c; |
1969 | |
1544 | 1970 for (c = 0; c < vd->vdev_children; c++) { |
1971 oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]); | |
1972 if (oldvd != NULL) | |
1973 return (oldvd); | |
1974 } | |
789 | 1975 |
1976 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { | |
1544 | 1977 oldvd = vd->vdev_child[0]; |
1978 newvd = vd->vdev_child[1]; | |
789 | 1979 |
1544 | 1980 mutex_enter(&newvd->vdev_dtl_lock); |
1981 if (newvd->vdev_dtl_map.sm_space == 0 && | |
1982 newvd->vdev_dtl_scrub.sm_space == 0) { | |
1983 mutex_exit(&newvd->vdev_dtl_lock); | |
1984 return (oldvd); | |
1985 } | |
1986 mutex_exit(&newvd->vdev_dtl_lock); | |
1987 } | |
789 | 1988 |
1544 | 1989 return (NULL); |
789 | 1990 } |
1991 | |
1544 | 1992 static void |
789 | 1993 spa_vdev_replace_done(spa_t *spa) |
1994 { | |
1544 | 1995 vdev_t *vd; |
2082 | 1996 vdev_t *pvd; |
1544 | 1997 uint64_t guid; |
2082 | 1998 uint64_t pguid = 0; |
789 | 1999 |
1544 | 2000 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2001 |
1544 | 2002 while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) { |
2003 guid = vd->vdev_guid; | |
2082 | 2004 /* |
2005 * If we have just finished replacing a hot spared device, then | |
2006 * we need to detach the parent's first child (the original hot | |
2007 * spare) as well. | |
2008 */ | |
2009 pvd = vd->vdev_parent; | |
2010 if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2011 pvd->vdev_id == 0) { | |
2012 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); | |
2013 ASSERT(pvd->vdev_parent->vdev_children == 2); | |
2014 pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid; | |
2015 } | |
1544 | 2016 spa_config_exit(spa, FTAG); |
2017 if (spa_vdev_detach(spa, guid, B_TRUE) != 0) | |
2018 return; | |
2082 | 2019 if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0) |
2020 return; | |
1544 | 2021 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2022 } |
2023 | |
1544 | 2024 spa_config_exit(spa, FTAG); |
789 | 2025 } |
2026 | |
2027 /* | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2028 * Update the stored path for this vdev. Dirty the vdev configuration, relying |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2029 * on spa_vdev_enter/exit() to synchronize the labels and cache. |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2030 */ |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2031 int |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2032 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2033 { |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2034 vdev_t *rvd, *vd; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2035 uint64_t txg; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2036 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2037 rvd = spa->spa_root_vdev; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2038 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2039 txg = spa_vdev_enter(spa); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2040 |
2082 | 2041 if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) { |
2042 /* | |
2043 * Determine if this is a reference to a hot spare. In that | |
2044 * case, update the path as stored in the spare list. | |
2045 */ | |
2046 nvlist_t **spares; | |
2047 uint_t i, nspares; | |
2048 if (spa->spa_sparelist != NULL) { | |
2049 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
2050 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
2051 for (i = 0; i < nspares; i++) { | |
2052 uint64_t theguid; | |
2053 VERIFY(nvlist_lookup_uint64(spares[i], | |
2054 ZPOOL_CONFIG_GUID, &theguid) == 0); | |
2055 if (theguid == guid) | |
2056 break; | |
2057 } | |
2058 | |
2059 if (i == nspares) | |
2060 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); | |
2061 | |
2062 VERIFY(nvlist_add_string(spares[i], | |
2063 ZPOOL_CONFIG_PATH, newpath) == 0); | |
2064 spa_load_spares(spa); | |
2065 spa->spa_sync_spares = B_TRUE; | |
2066 return (spa_vdev_exit(spa, NULL, txg, 0)); | |
2067 } else { | |
2068 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); | |
2069 } | |
2070 } | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2071 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
2072 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
2073 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
2074 |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2075 spa_strfree(vd->vdev_path); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2076 vd->vdev_path = spa_strdup(newpath); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2077 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2078 vdev_config_dirty(vd->vdev_top); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2079 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2080 return (spa_vdev_exit(spa, NULL, txg, 0)); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2081 } |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2082 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2083 /* |
789 | 2084 * ========================================================================== |
2085 * SPA Scrubbing | |
2086 * ========================================================================== | |
2087 */ | |
2088 | |
1544 | 2089 void |
2090 spa_scrub_throttle(spa_t *spa, int direction) | |
2091 { | |
2092 mutex_enter(&spa->spa_scrub_lock); | |
2093 spa->spa_scrub_throttled += direction; | |
2094 ASSERT(spa->spa_scrub_throttled >= 0); | |
2095 if (spa->spa_scrub_throttled == 0) | |
2096 cv_broadcast(&spa->spa_scrub_io_cv); | |
2097 mutex_exit(&spa->spa_scrub_lock); | |
2098 } | |
789 | 2099 |
2100 static void | |
2101 spa_scrub_io_done(zio_t *zio) | |
2102 { | |
2103 spa_t *spa = zio->io_spa; | |
2104 | |
2105 zio_buf_free(zio->io_data, zio->io_size); | |
2106 | |
2107 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2108 if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2109 vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev; |
789 | 2110 spa->spa_scrub_errors++; |
2111 mutex_enter(&vd->vdev_stat_lock); | |
2112 vd->vdev_stat.vs_scrub_errors++; | |
2113 mutex_exit(&vd->vdev_stat_lock); | |
2114 } | |
1544 | 2115 if (--spa->spa_scrub_inflight == 0) { |
2116 cv_broadcast(&spa->spa_scrub_io_cv); | |
2117 ASSERT(spa->spa_scrub_throttled == 0); | |
2118 } | |
2119 mutex_exit(&spa->spa_scrub_lock); | |
789 | 2120 } |
2121 | |
2122 static void | |
1544 | 2123 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags, |
2124 zbookmark_t *zb) | |
789 | 2125 { |
2126 size_t size = BP_GET_LSIZE(bp); | |
2127 void *data = zio_buf_alloc(size); | |
2128 | |
2129 mutex_enter(&spa->spa_scrub_lock); | |
2130 spa->spa_scrub_inflight++; | |
2131 mutex_exit(&spa->spa_scrub_lock); | |
2132 | |
1544 | 2133 if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET) |
2134 flags |= ZIO_FLAG_SPECULATIVE; /* intent log block */ | |
2135 | |
1807
35c8b566d7af
6410711 intent log blocks don't get invited to pool parties
bonwick
parents:
1775
diff
changeset
|
2136 flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL; |
1544 | 2137 |
789 | 2138 zio_nowait(zio_read(NULL, spa, bp, data, size, |
1544 | 2139 spa_scrub_io_done, NULL, priority, flags, zb)); |
789 | 2140 } |
2141 | |
2142 /* ARGSUSED */ | |
2143 static int | |
2144 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a) | |
2145 { | |
2146 blkptr_t *bp = &bc->bc_blkptr; | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2147 vdev_t *vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2148 dva_t *dva = bp->blk_dva; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2149 int needs_resilver = B_FALSE; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2150 int d; |
789 | 2151 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2152 if (bc->bc_errno) { |
789 | 2153 /* |
2154 * We can't scrub this block, but we can continue to scrub | |
2155 * the rest of the pool. Note the error and move along. | |
2156 */ | |
2157 mutex_enter(&spa->spa_scrub_lock); | |
2158 spa->spa_scrub_errors++; | |
2159 mutex_exit(&spa->spa_scrub_lock); | |
2160 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2161 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2162 vd->vdev_stat.vs_scrub_errors++; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2163 mutex_exit(&vd->vdev_stat_lock); |
789 | 2164 |
2165 return (ERESTART); | |
2166 } | |
2167 | |
2168 ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg); | |
2169 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2170 for (d = 0; d < BP_GET_NDVAS(bp); d++) { |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2171 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d])); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2172 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2173 ASSERT(vd != NULL); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2174 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2175 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2176 * Keep track of how much data we've examined so that |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2177 * zpool(1M) status can make useful progress reports. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2178 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2179 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2180 vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2181 mutex_exit(&vd->vdev_stat_lock); |
789 | 2182 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2183 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) { |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2184 if (DVA_GET_GANG(&dva[d])) { |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2185 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2186 * Gang members may be spread across multiple |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2187 * vdevs, so the best we can do is look at the |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2188 * pool-wide DTL. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2189 * XXX -- it would be better to change our |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2190 * allocation policy to ensure that this can't |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2191 * happen. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2192 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2193 vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2194 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2195 if (vdev_dtl_contains(&vd->vdev_dtl_map, |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2196 bp->blk_birth, 1)) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2197 needs_resilver = B_TRUE; |
789 | 2198 } |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2199 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2200 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2201 if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING) |
789 | 2202 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB, |
1544 | 2203 ZIO_FLAG_SCRUB, &bc->bc_bookmark); |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2204 else if (needs_resilver) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2205 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER, |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2206 ZIO_FLAG_RESILVER, &bc->bc_bookmark); |
789 | 2207 |
2208 return (0); | |
2209 } | |
2210 | |
2211 static void | |
2212 spa_scrub_thread(spa_t *spa) | |
2213 { | |
2214 callb_cpr_t cprinfo; | |
2215 traverse_handle_t *th = spa->spa_scrub_th; | |
2216 vdev_t *rvd = spa->spa_root_vdev; | |
2217 pool_scrub_type_t scrub_type = spa->spa_scrub_type; | |
2218 int error = 0; | |
2219 boolean_t complete; | |
2220 | |
2221 CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG); | |
2222 | |
797
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2223 /* |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2224 * If we're restarting due to a snapshot create/delete, |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2225 * wait for that to complete. |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2226 */ |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2227 txg_wait_synced(spa_get_dsl(spa), 0); |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2228 |
1544 | 2229 dprintf("start %s mintxg=%llu maxtxg=%llu\n", |
2230 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
2231 spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg); | |
2232 | |
2233 spa_config_enter(spa, RW_WRITER, FTAG); | |
2234 vdev_reopen(rvd); /* purge all vdev caches */ | |
789 | 2235 vdev_config_dirty(rvd); /* rewrite all disk labels */ |
2236 vdev_scrub_stat_update(rvd, scrub_type, B_FALSE); | |
1544 | 2237 spa_config_exit(spa, FTAG); |
789 | 2238 |
2239 mutex_enter(&spa->spa_scrub_lock); | |
2240 spa->spa_scrub_errors = 0; | |
2241 spa->spa_scrub_active = 1; | |
1544 | 2242 ASSERT(spa->spa_scrub_inflight == 0); |
2243 ASSERT(spa->spa_scrub_throttled == 0); | |
789 | 2244 |
2245 while (!spa->spa_scrub_stop) { | |
2246 CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1544 | 2247 while (spa->spa_scrub_suspended) { |
789 | 2248 spa->spa_scrub_active = 0; |
2249 cv_broadcast(&spa->spa_scrub_cv); | |
2250 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2251 spa->spa_scrub_active = 1; | |
2252 } | |
2253 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock); | |
2254 | |
2255 if (spa->spa_scrub_restart_txg != 0) | |
2256 break; | |
2257 | |
2258 mutex_exit(&spa->spa_scrub_lock); | |
2259 error = traverse_more(th); | |
2260 mutex_enter(&spa->spa_scrub_lock); | |
2261 if (error != EAGAIN) | |
2262 break; | |
1544 | 2263 |
2264 while (spa->spa_scrub_throttled > 0) | |
2265 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
789 | 2266 } |
2267 | |
2268 while (spa->spa_scrub_inflight) | |
2269 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2270 | |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2271 spa->spa_scrub_active = 0; |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2272 cv_broadcast(&spa->spa_scrub_cv); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2273 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2274 mutex_exit(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2275 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2276 spa_config_enter(spa, RW_WRITER, FTAG); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2277 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2278 mutex_enter(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2279 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2280 /* |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2281 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2282 * AND the spa config lock to synchronize with any config changes |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2283 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit(). |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2284 */ |
789 | 2285 if (spa->spa_scrub_restart_txg != 0) |
2286 error = ERESTART; | |
2287 | |
1544 | 2288 if (spa->spa_scrub_stop) |
2289 error = EINTR; | |
2290 | |
789 | 2291 /* |
1544 | 2292 * Even if there were uncorrectable errors, we consider the scrub |
2293 * completed. The downside is that if there is a transient error during | |
2294 * a resilver, we won't resilver the data properly to the target. But | |
2295 * if the damage is permanent (more likely) we will resilver forever, | |
2296 * which isn't really acceptable. Since there is enough information for | |
2297 * the user to know what has failed and why, this seems like a more | |
2298 * tractable approach. | |
789 | 2299 */ |
1544 | 2300 complete = (error == 0); |
789 | 2301 |
1544 | 2302 dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n", |
2303 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
789 | 2304 spa->spa_scrub_maxtxg, complete ? "done" : "FAILED", |
2305 error, spa->spa_scrub_errors, spa->spa_scrub_stop); | |
2306 | |
2307 mutex_exit(&spa->spa_scrub_lock); | |
2308 | |
2309 /* | |
2310 * If the scrub/resilver completed, update all DTLs to reflect this. | |
2311 * Whether it succeeded or not, vacate all temporary scrub DTLs. | |
2312 */ | |
2313 vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1, | |
2314 complete ? spa->spa_scrub_maxtxg : 0, B_TRUE); | |
2315 vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete); | |
1544 | 2316 spa_errlog_rotate(spa); |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2317 |
1544 | 2318 spa_config_exit(spa, FTAG); |
789 | 2319 |
2320 mutex_enter(&spa->spa_scrub_lock); | |
2321 | |
1544 | 2322 /* |
2323 * We may have finished replacing a device. | |
2324 * Let the async thread assess this and handle the detach. | |
2325 */ | |
2326 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); | |
789 | 2327 |
2328 /* | |
2329 * If we were told to restart, our final act is to start a new scrub. | |
2330 */ | |
2331 if (error == ERESTART) | |
1544 | 2332 spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ? |
2333 SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB); | |
789 | 2334 |
1544 | 2335 spa->spa_scrub_type = POOL_SCRUB_NONE; |
2336 spa->spa_scrub_active = 0; | |
2337 spa->spa_scrub_thread = NULL; | |
2338 cv_broadcast(&spa->spa_scrub_cv); | |
789 | 2339 CALLB_CPR_EXIT(&cprinfo); /* drops &spa->spa_scrub_lock */ |
2340 thread_exit(); | |
2341 } | |
2342 | |
2343 void | |
2344 spa_scrub_suspend(spa_t *spa) | |
2345 { | |
2346 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2347 spa->spa_scrub_suspended++; |
789 | 2348 while (spa->spa_scrub_active) { |
2349 cv_broadcast(&spa->spa_scrub_cv); | |
2350 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2351 } | |
2352 while (spa->spa_scrub_inflight) | |
2353 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2354 mutex_exit(&spa->spa_scrub_lock); | |
2355 } | |
2356 | |
2357 void | |
2358 spa_scrub_resume(spa_t *spa) | |
2359 { | |
2360 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2361 ASSERT(spa->spa_scrub_suspended != 0); |
2362 if (--spa->spa_scrub_suspended == 0) | |
789 | 2363 cv_broadcast(&spa->spa_scrub_cv); |
2364 mutex_exit(&spa->spa_scrub_lock); | |
2365 } | |
2366 | |
2367 void | |
2368 spa_scrub_restart(spa_t *spa, uint64_t txg) | |
2369 { | |
2370 /* | |
2371 * Something happened (e.g. snapshot create/delete) that means | |
2372 * we must restart any in-progress scrubs. The itinerary will | |
2373 * fix this properly. | |
2374 */ | |
2375 mutex_enter(&spa->spa_scrub_lock); | |
2376 spa->spa_scrub_restart_txg = txg; | |
2377 mutex_exit(&spa->spa_scrub_lock); | |
2378 } | |
2379 | |
1544 | 2380 int |
2381 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force) | |
789 | 2382 { |
2383 space_seg_t *ss; | |
2384 uint64_t mintxg, maxtxg; | |
2385 vdev_t *rvd = spa->spa_root_vdev; | |
2386 | |
2387 if ((uint_t)type >= POOL_SCRUB_TYPES) | |
2388 return (ENOTSUP); | |
2389 | |
1544 | 2390 mutex_enter(&spa->spa_scrub_lock); |
2391 | |
789 | 2392 /* |
2393 * If there's a scrub or resilver already in progress, stop it. | |
2394 */ | |
2395 while (spa->spa_scrub_thread != NULL) { | |
2396 /* | |
2397 * Don't stop a resilver unless forced. | |
2398 */ | |
1544 | 2399 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) { |
2400 mutex_exit(&spa->spa_scrub_lock); | |
789 | 2401 return (EBUSY); |
1544 | 2402 } |
789 | 2403 spa->spa_scrub_stop = 1; |
2404 cv_broadcast(&spa->spa_scrub_cv); | |
2405 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2406 } | |
2407 | |
2408 /* | |
2409 * Terminate the previous traverse. | |
2410 */ | |
2411 if (spa->spa_scrub_th != NULL) { | |
2412 traverse_fini(spa->spa_scrub_th); | |
2413 spa->spa_scrub_th = NULL; | |
2414 } | |
2415 | |
1544 | 2416 if (rvd == NULL) { |
2417 ASSERT(spa->spa_scrub_stop == 0); | |
2418 ASSERT(spa->spa_scrub_type == type); | |
2419 ASSERT(spa->spa_scrub_restart_txg == 0); | |
2420 mutex_exit(&spa->spa_scrub_lock); | |
2421 return (0); | |
2422 } | |
789 | 2423 |
2424 mintxg = TXG_INITIAL - 1; | |
2425 maxtxg = spa_last_synced_txg(spa) + 1; | |
2426 | |
1544 | 2427 mutex_enter(&rvd->vdev_dtl_lock); |
789 | 2428 |
1544 | 2429 if (rvd->vdev_dtl_map.sm_space == 0) { |
2430 /* | |
2431 * The pool-wide DTL is empty. | |
1732 | 2432 * If this is a resilver, there's nothing to do except |
2433 * check whether any in-progress replacements have completed. | |
1544 | 2434 */ |
1732 | 2435 if (type == POOL_SCRUB_RESILVER) { |
1544 | 2436 type = POOL_SCRUB_NONE; |
1732 | 2437 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); |
2438 } | |
1544 | 2439 } else { |
2440 /* | |
2441 * The pool-wide DTL is non-empty. | |
2442 * If this is a normal scrub, upgrade to a resilver instead. | |
2443 */ | |
2444 if (type == POOL_SCRUB_EVERYTHING) | |
2445 type = POOL_SCRUB_RESILVER; | |
2446 } | |
789 | 2447 |
1544 | 2448 if (type == POOL_SCRUB_RESILVER) { |
789 | 2449 /* |
2450 * Determine the resilvering boundaries. | |
2451 * | |
2452 * Note: (mintxg, maxtxg) is an open interval, | |
2453 * i.e. mintxg and maxtxg themselves are not included. | |
2454 * | |
2455 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1 | |
2456 * so we don't claim to resilver a txg that's still changing. | |
2457 */ | |
2458 ss = avl_first(&rvd->vdev_dtl_map.sm_root); | |
1544 | 2459 mintxg = ss->ss_start - 1; |
789 | 2460 ss = avl_last(&rvd->vdev_dtl_map.sm_root); |
1544 | 2461 maxtxg = MIN(ss->ss_end, maxtxg); |
789 | 2462 } |
2463 | |
1544 | 2464 mutex_exit(&rvd->vdev_dtl_lock); |
2465 | |
2466 spa->spa_scrub_stop = 0; | |
2467 spa->spa_scrub_type = type; | |
2468 spa->spa_scrub_restart_txg = 0; | |
2469 | |
2470 if (type != POOL_SCRUB_NONE) { | |
2471 spa->spa_scrub_mintxg = mintxg; | |
789 | 2472 spa->spa_scrub_maxtxg = maxtxg; |
2473 spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL, | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2474 ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL, |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2475 ZIO_FLAG_CANFAIL); |
789 | 2476 traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg); |
2477 spa->spa_scrub_thread = thread_create(NULL, 0, | |
2478 spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri); | |
2479 } | |
2480 | |
1544 | 2481 mutex_exit(&spa->spa_scrub_lock); |
2482 | |
789 | 2483 return (0); |
2484 } | |
2485 | |
1544 | 2486 /* |
2487 * ========================================================================== | |
2488 * SPA async task processing | |
2489 * ========================================================================== | |
2490 */ | |
2491 | |
2492 static void | |
2493 spa_async_reopen(spa_t *spa) | |
789 | 2494 { |
1544 | 2495 vdev_t *rvd = spa->spa_root_vdev; |
2496 vdev_t *tvd; | |
2497 int c; | |
2498 | |
2499 spa_config_enter(spa, RW_WRITER, FTAG); | |
2500 | |
2501 for (c = 0; c < rvd->vdev_children; c++) { | |
2502 tvd = rvd->vdev_child[c]; | |
2503 if (tvd->vdev_reopen_wanted) { | |
2504 tvd->vdev_reopen_wanted = 0; | |
2505 vdev_reopen(tvd); | |
2506 } | |
2507 } | |
789 | 2508 |
1544 | 2509 spa_config_exit(spa, FTAG); |
2510 } | |
2511 | |
2512 static void | |
2513 spa_async_thread(spa_t *spa) | |
2514 { | |
2515 int tasks; | |
2516 | |
2517 ASSERT(spa->spa_sync_on); | |
789 | 2518 |
1544 | 2519 mutex_enter(&spa->spa_async_lock); |
2520 tasks = spa->spa_async_tasks; | |
2521 spa->spa_async_tasks = 0; | |
2522 mutex_exit(&spa->spa_async_lock); | |
2523 | |
2524 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2525 * See if the config needs to be updated. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2526 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2527 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2528 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2529 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2530 mutex_exit(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2531 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2532 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2533 /* |
1544 | 2534 * See if any devices need to be reopened. |
2535 */ | |
2536 if (tasks & SPA_ASYNC_REOPEN) | |
2537 spa_async_reopen(spa); | |
2538 | |
2539 /* | |
2540 * If any devices are done replacing, detach them. | |
2541 */ | |
2542 if (tasks & SPA_ASYNC_REPLACE_DONE) | |
789 | 2543 spa_vdev_replace_done(spa); |
2544 | |
1544 | 2545 /* |
2546 * Kick off a scrub. | |
2547 */ | |
2548 if (tasks & SPA_ASYNC_SCRUB) | |
2549 VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0); | |
2550 | |
2551 /* | |
2552 * Kick off a resilver. | |
2553 */ | |
2554 if (tasks & SPA_ASYNC_RESILVER) | |
2555 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
2556 | |
2557 /* | |
2558 * Let the world know that we're done. | |
2559 */ | |
2560 mutex_enter(&spa->spa_async_lock); | |
2561 spa->spa_async_thread = NULL; | |
2562 cv_broadcast(&spa->spa_async_cv); | |
2563 mutex_exit(&spa->spa_async_lock); | |
2564 thread_exit(); | |
2565 } | |
2566 | |
2567 void | |
2568 spa_async_suspend(spa_t *spa) | |
2569 { | |
2570 mutex_enter(&spa->spa_async_lock); | |
2571 spa->spa_async_suspended++; | |
2572 while (spa->spa_async_thread != NULL) | |
2573 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
2574 mutex_exit(&spa->spa_async_lock); | |
2575 } | |
2576 | |
2577 void | |
2578 spa_async_resume(spa_t *spa) | |
2579 { | |
2580 mutex_enter(&spa->spa_async_lock); | |
2581 ASSERT(spa->spa_async_suspended != 0); | |
2582 spa->spa_async_suspended--; | |
2583 mutex_exit(&spa->spa_async_lock); | |
2584 } | |
2585 | |
2586 static void | |
2587 spa_async_dispatch(spa_t *spa) | |
2588 { | |
2589 mutex_enter(&spa->spa_async_lock); | |
2590 if (spa->spa_async_tasks && !spa->spa_async_suspended && | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2591 spa->spa_async_thread == NULL && |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2592 rootdir != NULL && !vn_is_readonly(rootdir)) |
1544 | 2593 spa->spa_async_thread = thread_create(NULL, 0, |
2594 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
2595 mutex_exit(&spa->spa_async_lock); | |
2596 } | |
2597 | |
2598 void | |
2599 spa_async_request(spa_t *spa, int task) | |
2600 { | |
2601 mutex_enter(&spa->spa_async_lock); | |
2602 spa->spa_async_tasks |= task; | |
2603 mutex_exit(&spa->spa_async_lock); | |
789 | 2604 } |
2605 | |
2606 /* | |
2607 * ========================================================================== | |
2608 * SPA syncing routines | |
2609 * ========================================================================== | |
2610 */ | |
2611 | |
2612 static void | |
2613 spa_sync_deferred_frees(spa_t *spa, uint64_t txg) | |
2614 { | |
2615 bplist_t *bpl = &spa->spa_sync_bplist; | |
2616 dmu_tx_t *tx; | |
2617 blkptr_t blk; | |
2618 uint64_t itor = 0; | |
2619 zio_t *zio; | |
2620 int error; | |
2621 uint8_t c = 1; | |
2622 | |
2623 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD); | |
2624 | |
2625 while (bplist_iterate(bpl, &itor, &blk) == 0) | |
2626 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL)); | |
2627 | |
2628 error = zio_wait(zio); | |
2629 ASSERT3U(error, ==, 0); | |
2630 | |
2631 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2632 bplist_vacate(bpl, tx); | |
2633 | |
2634 /* | |
2635 * Pre-dirty the first block so we sync to convergence faster. | |
2636 * (Usually only the first block is needed.) | |
2637 */ | |
2638 dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); | |
2639 dmu_tx_commit(tx); | |
2640 } | |
2641 | |
2642 static void | |
2082 | 2643 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) |
2644 { | |
2645 char *packed = NULL; | |
2646 size_t nvsize = 0; | |
2647 dmu_buf_t *db; | |
2648 | |
2649 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
2650 | |
2651 packed = kmem_alloc(nvsize, KM_SLEEP); | |
2652 | |
2653 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
2654 KM_SLEEP) == 0); | |
2655 | |
2656 dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx); | |
2657 | |
2658 kmem_free(packed, nvsize); | |
2659 | |
2660 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
2661 dmu_buf_will_dirty(db, tx); | |
2662 *(uint64_t *)db->db_data = nvsize; | |
2663 dmu_buf_rele(db, FTAG); | |
2664 } | |
2665 | |
2666 static void | |
2667 spa_sync_spares(spa_t *spa, dmu_tx_t *tx) | |
2668 { | |
2669 nvlist_t *nvroot; | |
2670 nvlist_t **spares; | |
2671 int i; | |
2672 | |
2673 if (!spa->spa_sync_spares) | |
2674 return; | |
2675 | |
2676 /* | |
2677 * Update the MOS nvlist describing the list of available spares. | |
2678 * spa_validate_spares() will have already made sure this nvlist is | |
2679 * valid and the vdevs are labelled appropriately. | |
2680 */ | |
2681 if (spa->spa_spares_object == 0) { | |
2682 spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset, | |
2683 DMU_OT_PACKED_NVLIST, 1 << 14, | |
2684 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
2685 VERIFY(zap_update(spa->spa_meta_objset, | |
2686 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES, | |
2687 sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0); | |
2688 } | |
2689 | |
2690 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
2691 if (spa->spa_nspares == 0) { | |
2692 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
2693 NULL, 0) == 0); | |
2694 } else { | |
2695 spares = kmem_alloc(spa->spa_nspares * sizeof (void *), | |
2696 KM_SLEEP); | |
2697 for (i = 0; i < spa->spa_nspares; i++) | |
2698 spares[i] = vdev_config_generate(spa, | |
2699 spa->spa_spares[i], B_FALSE, B_TRUE); | |
2700 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
2701 spares, spa->spa_nspares) == 0); | |
2702 for (i = 0; i < spa->spa_nspares; i++) | |
2703 nvlist_free(spares[i]); | |
2704 kmem_free(spares, spa->spa_nspares * sizeof (void *)); | |
2705 } | |
2706 | |
2707 spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx); | |
2708 | |
2709 spa->spa_sync_spares = B_FALSE; | |
2710 } | |
2711 | |
2712 static void | |
789 | 2713 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) |
2714 { | |
2715 nvlist_t *config; | |
2716 | |
2717 if (list_is_empty(&spa->spa_dirty_list)) | |
2718 return; | |
2719 | |
2720 config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE); | |
2721 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2722 if (spa->spa_config_syncing) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2723 nvlist_free(spa->spa_config_syncing); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2724 spa->spa_config_syncing = config; |
789 | 2725 |
2082 | 2726 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); |
789 | 2727 } |
2728 | |
2729 /* | |
2730 * Sync the specified transaction group. New blocks may be dirtied as | |
2731 * part of the process, so we iterate until it converges. | |
2732 */ | |
2733 void | |
2734 spa_sync(spa_t *spa, uint64_t txg) | |
2735 { | |
2736 dsl_pool_t *dp = spa->spa_dsl_pool; | |
2737 objset_t *mos = spa->spa_meta_objset; | |
2738 bplist_t *bpl = &spa->spa_sync_bplist; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2739 vdev_t *rvd = spa->spa_root_vdev; |
789 | 2740 vdev_t *vd; |
2741 dmu_tx_t *tx; | |
2742 int dirty_vdevs; | |
2743 | |
2744 /* | |
2745 * Lock out configuration changes. | |
2746 */ | |
1544 | 2747 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2748 |
2749 spa->spa_syncing_txg = txg; | |
2750 spa->spa_sync_pass = 0; | |
2751 | |
1544 | 2752 VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); |
789 | 2753 |
2082 | 2754 tx = dmu_tx_create_assigned(dp, txg); |
2755 | |
2756 /* | |
2757 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg, | |
2758 * set spa_deflate if we have no raid-z vdevs. | |
2759 */ | |
2760 if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE && | |
2761 spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) { | |
2762 int i; | |
2763 | |
2764 for (i = 0; i < rvd->vdev_children; i++) { | |
2765 vd = rvd->vdev_child[i]; | |
2766 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
2767 break; | |
2768 } | |
2769 if (i == rvd->vdev_children) { | |
2770 spa->spa_deflate = TRUE; | |
2771 VERIFY(0 == zap_add(spa->spa_meta_objset, | |
2772 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
2773 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
2774 } | |
2775 } | |
2776 | |
789 | 2777 /* |
2778 * If anything has changed in this txg, push the deferred frees | |
2779 * from the previous txg. If not, leave them alone so that we | |
2780 * don't generate work on an otherwise idle system. | |
2781 */ | |
2782 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || | |
2783 !txg_list_empty(&dp->dp_dirty_dirs, txg)) | |
2784 spa_sync_deferred_frees(spa, txg); | |
2785 | |
2786 /* | |
2787 * Iterate to convergence. | |
2788 */ | |
2789 do { | |
2790 spa->spa_sync_pass++; | |
2791 | |
2792 spa_sync_config_object(spa, tx); | |
2082 | 2793 spa_sync_spares(spa, tx); |
1544 | 2794 spa_errlog_sync(spa, txg); |
789 | 2795 dsl_pool_sync(dp, txg); |
2796 | |
2797 dirty_vdevs = 0; | |
2798 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { | |
2799 vdev_sync(vd, txg); | |
2800 dirty_vdevs++; | |
2801 } | |
2802 | |
2803 bplist_sync(bpl, tx); | |
2804 } while (dirty_vdevs); | |
2805 | |
2806 bplist_close(bpl); | |
2807 | |
2808 dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); | |
2809 | |
2810 /* | |
2811 * Rewrite the vdev configuration (which includes the uberblock) | |
2812 * to commit the transaction group. | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2813 * |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2814 * If there are any dirty vdevs, sync the uberblock to all vdevs. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2815 * Otherwise, pick a random top-level vdev that's known to be |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2816 * visible in the config cache (see spa_vdev_add() for details). |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2817 * If the write fails, try the next vdev until we're tried them all. |
789 | 2818 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2819 if (!list_is_empty(&spa->spa_dirty_list)) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2820 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2821 } else { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2822 int children = rvd->vdev_children; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2823 int c0 = spa_get_random(children); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2824 int c; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2825 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2826 for (c = 0; c < children; c++) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2827 vd = rvd->vdev_child[(c0 + c) % children]; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2828 if (vd->vdev_ms_array == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2829 continue; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2830 if (vdev_config_sync(vd, txg) == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2831 break; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2832 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2833 if (c == children) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2834 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2835 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2836 |
2082 | 2837 dmu_tx_commit(tx); |
2838 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2839 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2840 * Clear the dirty config list. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2841 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2842 while ((vd = list_head(&spa->spa_dirty_list)) != NULL) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2843 vdev_config_clean(vd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2844 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2845 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2846 * Now that the new config has synced transactionally, |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2847 * let it become visible to the config cache. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2848 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2849 if (spa->spa_config_syncing != NULL) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2850 spa_config_set(spa, spa->spa_config_syncing); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2851 spa->spa_config_txg = txg; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2852 spa->spa_config_syncing = NULL; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2853 } |
789 | 2854 |
2855 /* | |
2856 * Make a stable copy of the fully synced uberblock. | |
2857 * We use this as the root for pool traversals. | |
2858 */ | |
2859 spa->spa_traverse_wanted = 1; /* tells traverse_more() to stop */ | |
2860 | |
2861 spa_scrub_suspend(spa); /* stop scrubbing and finish I/Os */ | |
2862 | |
2863 rw_enter(&spa->spa_traverse_lock, RW_WRITER); | |
2864 spa->spa_traverse_wanted = 0; | |
2865 spa->spa_ubsync = spa->spa_uberblock; | |
2866 rw_exit(&spa->spa_traverse_lock); | |
2867 | |
2868 spa_scrub_resume(spa); /* resume scrub with new ubsync */ | |
2869 | |
2870 /* | |
2871 * Clean up the ZIL records for the synced txg. | |
2872 */ | |
2873 dsl_pool_zil_clean(dp); | |
2874 | |
2875 /* | |
2876 * Update usable space statistics. | |
2877 */ | |
2878 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) | |
2879 vdev_sync_done(vd, txg); | |
2880 | |
2881 /* | |
2882 * It had better be the case that we didn't dirty anything | |
2082 | 2883 * since vdev_config_sync(). |
789 | 2884 */ |
2885 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
2886 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
2887 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
2888 ASSERT(bpl->bpl_queue == NULL); | |
2889 | |
1544 | 2890 spa_config_exit(spa, FTAG); |
2891 | |
2892 /* | |
2893 * If any async tasks have been requested, kick them off. | |
2894 */ | |
2895 spa_async_dispatch(spa); | |
789 | 2896 } |
2897 | |
2898 /* | |
2899 * Sync all pools. We don't want to hold the namespace lock across these | |
2900 * operations, so we take a reference on the spa_t and drop the lock during the | |
2901 * sync. | |
2902 */ | |
2903 void | |
2904 spa_sync_allpools(void) | |
2905 { | |
2906 spa_t *spa = NULL; | |
2907 mutex_enter(&spa_namespace_lock); | |
2908 while ((spa = spa_next(spa)) != NULL) { | |
2909 if (spa_state(spa) != POOL_STATE_ACTIVE) | |
2910 continue; | |
2911 spa_open_ref(spa, FTAG); | |
2912 mutex_exit(&spa_namespace_lock); | |
2913 txg_wait_synced(spa_get_dsl(spa), 0); | |
2914 mutex_enter(&spa_namespace_lock); | |
2915 spa_close(spa, FTAG); | |
2916 } | |
2917 mutex_exit(&spa_namespace_lock); | |
2918 } | |
2919 | |
2920 /* | |
2921 * ========================================================================== | |
2922 * Miscellaneous routines | |
2923 * ========================================================================== | |
2924 */ | |
2925 | |
2926 /* | |
2927 * Remove all pools in the system. | |
2928 */ | |
2929 void | |
2930 spa_evict_all(void) | |
2931 { | |
2932 spa_t *spa; | |
2933 | |
2934 /* | |
2935 * Remove all cached state. All pools should be closed now, | |
2936 * so every spa in the AVL tree should be unreferenced. | |
2937 */ | |
2938 mutex_enter(&spa_namespace_lock); | |
2939 while ((spa = spa_next(NULL)) != NULL) { | |
2940 /* | |
1544 | 2941 * Stop async tasks. The async thread may need to detach |
2942 * a device that's been replaced, which requires grabbing | |
2943 * spa_namespace_lock, so we must drop it here. | |
789 | 2944 */ |
2945 spa_open_ref(spa, FTAG); | |
2946 mutex_exit(&spa_namespace_lock); | |
1544 | 2947 spa_async_suspend(spa); |
789 | 2948 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); |
2949 mutex_enter(&spa_namespace_lock); | |
2950 spa_close(spa, FTAG); | |
2951 | |
2952 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
2953 spa_unload(spa); | |
2954 spa_deactivate(spa); | |
2955 } | |
2956 spa_remove(spa); | |
2957 } | |
2958 mutex_exit(&spa_namespace_lock); | |
2959 } | |
1544 | 2960 |
2961 vdev_t * | |
2962 spa_lookup_by_guid(spa_t *spa, uint64_t guid) | |
2963 { | |
2964 return (vdev_lookup_by_guid(spa->spa_root_vdev, guid)); | |
2965 } | |
1760 | 2966 |
2967 void | |
2968 spa_upgrade(spa_t *spa) | |
2969 { | |
2970 spa_config_enter(spa, RW_WRITER, FTAG); | |
2971 | |
2972 /* | |
2973 * This should only be called for a non-faulted pool, and since a | |
2974 * future version would result in an unopenable pool, this shouldn't be | |
2975 * possible. | |
2976 */ | |
2977 ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION); | |
2978 | |
2979 spa->spa_uberblock.ub_version = ZFS_VERSION; | |
2980 vdev_config_dirty(spa->spa_root_vdev); | |
2981 | |
2982 spa_config_exit(spa, FTAG); | |
2082 | 2983 |
2984 txg_wait_synced(spa_get_dsl(spa), 0); | |
1760 | 2985 } |
2082 | 2986 |
2987 boolean_t | |
2988 spa_has_spare(spa_t *spa, uint64_t guid) | |
2989 { | |
2990 int i; | |
2991 | |
2992 for (i = 0; i < spa->spa_nspares; i++) | |
2993 if (spa->spa_spares[i]->vdev_guid == guid) | |
2994 return (B_TRUE); | |
2995 | |
2996 return (B_FALSE); | |
2997 } |