Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/spa.c @ 1775:e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
6410700 zdb should support reading raw blocks out of storage pool
6410709 ztest: spa config can change before pool export
author | billm |
---|---|
date | Mon, 10 Apr 2006 05:03:38 -0700 |
parents | e1ad2821c30d |
children | 35c8b566d7af |
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 */ | |
21 /* | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
789 | 23 * Use is subject to license terms. |
24 */ | |
25 | |
26 #pragma ident "%Z%%M% %I% %E% SMI" | |
27 | |
28 /* | |
29 * This file contains all the routines used when modifying on-disk SPA state. | |
30 * This includes opening, importing, destroying, exporting a pool, and syncing a | |
31 * pool. | |
32 */ | |
33 | |
34 #include <sys/zfs_context.h> | |
1544 | 35 #include <sys/fm/fs/zfs.h> |
789 | 36 #include <sys/spa_impl.h> |
37 #include <sys/zio.h> | |
38 #include <sys/zio_checksum.h> | |
39 #include <sys/zio_compress.h> | |
40 #include <sys/dmu.h> | |
41 #include <sys/dmu_tx.h> | |
42 #include <sys/zap.h> | |
43 #include <sys/zil.h> | |
44 #include <sys/vdev_impl.h> | |
45 #include <sys/metaslab.h> | |
46 #include <sys/uberblock_impl.h> | |
47 #include <sys/txg.h> | |
48 #include <sys/avl.h> | |
49 #include <sys/dmu_traverse.h> | |
50 #include <sys/unique.h> | |
51 #include <sys/dsl_pool.h> | |
52 #include <sys/dsl_dir.h> | |
53 #include <sys/dsl_prop.h> | |
54 #include <sys/fs/zfs.h> | |
55 #include <sys/callb.h> | |
56 | |
57 /* | |
58 * ========================================================================== | |
59 * SPA state manipulation (open/create/destroy/import/export) | |
60 * ========================================================================== | |
61 */ | |
62 | |
1544 | 63 static int |
64 spa_error_entry_compare(const void *a, const void *b) | |
65 { | |
66 spa_error_entry_t *sa = (spa_error_entry_t *)a; | |
67 spa_error_entry_t *sb = (spa_error_entry_t *)b; | |
68 int ret; | |
69 | |
70 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, | |
71 sizeof (zbookmark_t)); | |
72 | |
73 if (ret < 0) | |
74 return (-1); | |
75 else if (ret > 0) | |
76 return (1); | |
77 else | |
78 return (0); | |
79 } | |
80 | |
81 /* | |
82 * Utility function which retrieves copies of the current logs and | |
83 * re-initializes them in the process. | |
84 */ | |
85 void | |
86 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
87 { | |
88 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
89 | |
90 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
91 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
92 | |
93 avl_create(&spa->spa_errlist_scrub, | |
94 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
95 offsetof(spa_error_entry_t, se_avl)); | |
96 avl_create(&spa->spa_errlist_last, | |
97 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
98 offsetof(spa_error_entry_t, se_avl)); | |
99 } | |
100 | |
789 | 101 /* |
102 * Activate an uninitialized pool. | |
103 */ | |
104 static void | |
105 spa_activate(spa_t *spa) | |
106 { | |
107 int t; | |
108 | |
109 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
110 | |
111 spa->spa_state = POOL_STATE_ACTIVE; | |
112 | |
113 spa->spa_normal_class = metaslab_class_create(); | |
114 | |
115 for (t = 0; t < ZIO_TYPES; t++) { | |
116 spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue", | |
117 8, maxclsyspri, 50, INT_MAX, | |
118 TASKQ_PREPOPULATE); | |
119 spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr", | |
120 8, maxclsyspri, 50, INT_MAX, | |
121 TASKQ_PREPOPULATE); | |
122 } | |
123 | |
124 rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL); | |
125 | |
126 list_create(&spa->spa_dirty_list, sizeof (vdev_t), | |
127 offsetof(vdev_t, vdev_dirty_node)); | |
128 | |
129 txg_list_create(&spa->spa_vdev_txg_list, | |
130 offsetof(struct vdev, vdev_txg_node)); | |
1544 | 131 |
132 avl_create(&spa->spa_errlist_scrub, | |
133 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
134 offsetof(spa_error_entry_t, se_avl)); | |
135 avl_create(&spa->spa_errlist_last, | |
136 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
137 offsetof(spa_error_entry_t, se_avl)); | |
789 | 138 } |
139 | |
140 /* | |
141 * Opposite of spa_activate(). | |
142 */ | |
143 static void | |
144 spa_deactivate(spa_t *spa) | |
145 { | |
146 int t; | |
147 | |
148 ASSERT(spa->spa_sync_on == B_FALSE); | |
149 ASSERT(spa->spa_dsl_pool == NULL); | |
150 ASSERT(spa->spa_root_vdev == NULL); | |
151 | |
152 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
153 | |
154 txg_list_destroy(&spa->spa_vdev_txg_list); | |
155 | |
156 list_destroy(&spa->spa_dirty_list); | |
157 | |
158 rw_destroy(&spa->spa_traverse_lock); | |
159 | |
160 for (t = 0; t < ZIO_TYPES; t++) { | |
161 taskq_destroy(spa->spa_zio_issue_taskq[t]); | |
162 taskq_destroy(spa->spa_zio_intr_taskq[t]); | |
163 spa->spa_zio_issue_taskq[t] = NULL; | |
164 spa->spa_zio_intr_taskq[t] = NULL; | |
165 } | |
166 | |
167 metaslab_class_destroy(spa->spa_normal_class); | |
168 spa->spa_normal_class = NULL; | |
169 | |
1544 | 170 /* |
171 * If this was part of an import or the open otherwise failed, we may | |
172 * still have errors left in the queues. Empty them just in case. | |
173 */ | |
174 spa_errlog_drain(spa); | |
175 | |
176 avl_destroy(&spa->spa_errlist_scrub); | |
177 avl_destroy(&spa->spa_errlist_last); | |
178 | |
789 | 179 spa->spa_state = POOL_STATE_UNINITIALIZED; |
180 } | |
181 | |
182 /* | |
183 * Verify a pool configuration, and construct the vdev tree appropriately. This | |
184 * will create all the necessary vdevs in the appropriate layout, with each vdev | |
185 * in the CLOSED state. This will prep the pool before open/creation/import. | |
186 * All vdev validation is done by the vdev_alloc() routine. | |
187 */ | |
188 static vdev_t * | |
189 spa_config_parse(spa_t *spa, nvlist_t *nv, vdev_t *parent, uint_t id, int atype) | |
190 { | |
191 nvlist_t **child; | |
192 uint_t c, children; | |
193 vdev_t *vd; | |
194 | |
195 if ((vd = vdev_alloc(spa, nv, parent, id, atype)) == NULL) | |
196 return (NULL); | |
197 | |
198 if (vd->vdev_ops->vdev_op_leaf) | |
199 return (vd); | |
200 | |
201 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
202 &child, &children) != 0) { | |
203 vdev_free(vd); | |
204 return (NULL); | |
205 } | |
206 | |
207 for (c = 0; c < children; c++) { | |
208 if (spa_config_parse(spa, child[c], vd, c, atype) == NULL) { | |
209 vdev_free(vd); | |
210 return (NULL); | |
211 } | |
212 } | |
213 | |
214 return (vd); | |
215 } | |
216 | |
217 /* | |
218 * Opposite of spa_load(). | |
219 */ | |
220 static void | |
221 spa_unload(spa_t *spa) | |
222 { | |
223 /* | |
1544 | 224 * Stop async tasks. |
225 */ | |
226 spa_async_suspend(spa); | |
227 | |
228 /* | |
789 | 229 * Stop syncing. |
230 */ | |
231 if (spa->spa_sync_on) { | |
232 txg_sync_stop(spa->spa_dsl_pool); | |
233 spa->spa_sync_on = B_FALSE; | |
234 } | |
235 | |
236 /* | |
237 * Wait for any outstanding prefetch I/O to complete. | |
238 */ | |
1544 | 239 spa_config_enter(spa, RW_WRITER, FTAG); |
240 spa_config_exit(spa, FTAG); | |
789 | 241 |
242 /* | |
243 * Close the dsl pool. | |
244 */ | |
245 if (spa->spa_dsl_pool) { | |
246 dsl_pool_close(spa->spa_dsl_pool); | |
247 spa->spa_dsl_pool = NULL; | |
248 } | |
249 | |
250 /* | |
251 * Close all vdevs. | |
252 */ | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
253 if (spa->spa_root_vdev) |
789 | 254 vdev_free(spa->spa_root_vdev); |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
255 ASSERT(spa->spa_root_vdev == NULL); |
1544 | 256 |
257 spa->spa_async_suspended = 0; | |
789 | 258 } |
259 | |
260 /* | |
261 * Load an existing storage pool, using the pool's builtin spa_config as a | |
1544 | 262 * source of configuration information. |
789 | 263 */ |
264 static int | |
1544 | 265 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) |
789 | 266 { |
267 int error = 0; | |
268 nvlist_t *nvroot = NULL; | |
269 vdev_t *rvd; | |
270 uberblock_t *ub = &spa->spa_uberblock; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
271 uint64_t config_cache_txg = spa->spa_config_txg; |
789 | 272 uint64_t pool_guid; |
273 zio_t *zio; | |
274 | |
1544 | 275 spa->spa_load_state = state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
276 |
789 | 277 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || |
1733 | 278 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { |
1544 | 279 error = EINVAL; |
280 goto out; | |
281 } | |
789 | 282 |
1733 | 283 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
284 &spa->spa_config_txg); | |
285 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
286 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && |
1544 | 287 spa_guid_exists(pool_guid, 0)) { |
288 error = EEXIST; | |
289 goto out; | |
290 } | |
789 | 291 |
292 /* | |
293 * Parse the configuration into a vdev tree. | |
294 */ | |
1544 | 295 spa_config_enter(spa, RW_WRITER, FTAG); |
789 | 296 rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_LOAD); |
1544 | 297 spa_config_exit(spa, FTAG); |
789 | 298 |
1544 | 299 if (rvd == NULL) { |
300 error = EINVAL; | |
301 goto out; | |
302 } | |
789 | 303 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
304 ASSERT(spa->spa_root_vdev == rvd); |
789 | 305 ASSERT(spa_guid(spa) == pool_guid); |
306 | |
307 /* | |
308 * Try to open all vdevs, loading each label in the process. | |
309 */ | |
1544 | 310 if (vdev_open(rvd) != 0) { |
311 error = ENXIO; | |
312 goto out; | |
313 } | |
789 | 314 |
315 /* | |
316 * Find the best uberblock. | |
317 */ | |
318 bzero(ub, sizeof (uberblock_t)); | |
319 | |
320 zio = zio_root(spa, NULL, NULL, | |
321 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
322 vdev_uberblock_load(zio, rvd, ub); | |
323 error = zio_wait(zio); | |
324 | |
325 /* | |
326 * If we weren't able to find a single valid uberblock, return failure. | |
327 */ | |
328 if (ub->ub_txg == 0) { | |
1760 | 329 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
330 VDEV_AUX_CORRUPT_DATA); | |
1544 | 331 error = ENXIO; |
332 goto out; | |
333 } | |
334 | |
335 /* | |
336 * If the pool is newer than the code, we can't open it. | |
337 */ | |
1760 | 338 if (ub->ub_version > ZFS_VERSION) { |
339 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
340 VDEV_AUX_VERSION_NEWER); | |
1544 | 341 error = ENOTSUP; |
342 goto out; | |
789 | 343 } |
344 | |
345 /* | |
346 * If the vdev guid sum doesn't match the uberblock, we have an | |
347 * incomplete configuration. | |
348 */ | |
1732 | 349 if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { |
1544 | 350 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
351 VDEV_AUX_BAD_GUID_SUM); | |
352 error = ENXIO; | |
353 goto out; | |
789 | 354 } |
355 | |
356 /* | |
357 * Initialize internal SPA structures. | |
358 */ | |
359 spa->spa_state = POOL_STATE_ACTIVE; | |
360 spa->spa_ubsync = spa->spa_uberblock; | |
361 spa->spa_first_txg = spa_last_synced_txg(spa) + 1; | |
1544 | 362 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
363 if (error) { | |
364 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
365 VDEV_AUX_CORRUPT_DATA); | |
366 goto out; | |
367 } | |
789 | 368 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
369 | |
1544 | 370 if (zap_lookup(spa->spa_meta_objset, |
789 | 371 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 372 sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { |
373 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
374 VDEV_AUX_CORRUPT_DATA); | |
375 error = EIO; | |
376 goto out; | |
377 } | |
789 | 378 |
379 if (!mosconfig) { | |
380 dmu_buf_t *db; | |
381 char *packed = NULL; | |
382 size_t nvsize = 0; | |
383 nvlist_t *newconfig = NULL; | |
384 | |
1544 | 385 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, |
386 spa->spa_config_object, FTAG, &db)); | |
789 | 387 nvsize = *(uint64_t *)db->db_data; |
1544 | 388 dmu_buf_rele(db, FTAG); |
789 | 389 |
390 packed = kmem_alloc(nvsize, KM_SLEEP); | |
1544 | 391 error = dmu_read(spa->spa_meta_objset, |
789 | 392 spa->spa_config_object, 0, nvsize, packed); |
393 if (error == 0) | |
394 error = nvlist_unpack(packed, nvsize, &newconfig, 0); | |
395 kmem_free(packed, nvsize); | |
396 | |
1544 | 397 if (error) { |
398 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
399 VDEV_AUX_CORRUPT_DATA); | |
400 error = EIO; | |
401 goto out; | |
402 } | |
789 | 403 |
404 spa_config_set(spa, newconfig); | |
405 | |
406 spa_unload(spa); | |
407 spa_deactivate(spa); | |
408 spa_activate(spa); | |
409 | |
1544 | 410 return (spa_load(spa, newconfig, state, B_TRUE)); |
411 } | |
412 | |
413 if (zap_lookup(spa->spa_meta_objset, | |
414 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, | |
415 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { | |
416 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
417 VDEV_AUX_CORRUPT_DATA); | |
418 error = EIO; | |
419 goto out; | |
789 | 420 } |
421 | |
1544 | 422 /* |
423 * Load the persistent error log. If we have an older pool, this will | |
424 * not be present. | |
425 */ | |
426 error = zap_lookup(spa->spa_meta_objset, | |
427 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, | |
428 sizeof (uint64_t), 1, &spa->spa_errlog_last); | |
429 if (error != 0 &&error != ENOENT) { | |
430 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
431 VDEV_AUX_CORRUPT_DATA); | |
432 error = EIO; | |
433 goto out; | |
434 } | |
435 | |
436 error = zap_lookup(spa->spa_meta_objset, | |
437 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, | |
438 sizeof (uint64_t), 1, &spa->spa_errlog_scrub); | |
439 if (error != 0 && error != ENOENT) { | |
440 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
441 VDEV_AUX_CORRUPT_DATA); | |
442 error = EIO; | |
443 goto out; | |
444 } | |
789 | 445 |
446 /* | |
1544 | 447 * Load the vdev state for all top level vdevs. We need to grab the |
448 * config lock because all label I/O is done with the | |
449 * ZIO_FLAG_CONFIG_HELD flag. | |
789 | 450 */ |
1544 | 451 spa_config_enter(spa, RW_READER, FTAG); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
452 error = vdev_load(rvd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
453 spa_config_exit(spa, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
454 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
455 if (error) |
1544 | 456 goto out; |
789 | 457 |
458 /* | |
459 * Propagate the leaf DTLs we just loaded all the way up the tree. | |
460 */ | |
1544 | 461 spa_config_enter(spa, RW_WRITER, FTAG); |
789 | 462 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); |
1544 | 463 spa_config_exit(spa, FTAG); |
789 | 464 |
465 /* | |
466 * Check the state of the root vdev. If it can't be opened, it | |
467 * indicates one or more toplevel vdevs are faulted. | |
468 */ | |
1544 | 469 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
470 error = ENXIO; | |
471 goto out; | |
472 } | |
789 | 473 |
1544 | 474 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
|
475 dmu_tx_t *tx; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
476 int need_update = B_FALSE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
477 int c; |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
478 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
479 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
480 * 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
|
481 * 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
|
482 */ |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
483 tx = dmu_tx_create_assigned(spa_get_dsl(spa), |
789 | 484 spa_first_txg(spa)); |
485 dmu_objset_find(spa->spa_name, zil_claim, tx, 0); | |
486 dmu_tx_commit(tx); | |
487 | |
488 spa->spa_sync_on = B_TRUE; | |
489 txg_sync_start(spa->spa_dsl_pool); | |
490 | |
491 /* | |
492 * Wait for all claims to sync. | |
493 */ | |
494 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
|
495 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
496 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
497 * 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
|
498 * 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
|
499 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
500 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
|
501 state == SPA_LOAD_IMPORT) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
502 need_update = B_TRUE; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
503 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
504 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
|
505 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
|
506 need_update = B_TRUE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
507 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
508 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
509 * 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
|
510 * 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
|
511 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
512 if (need_update) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
513 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
789 | 514 } |
515 | |
1544 | 516 error = 0; |
517 out: | |
518 if (error) | |
519 zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0); | |
520 spa->spa_load_state = SPA_LOAD_NONE; | |
521 spa->spa_ena = 0; | |
522 | |
523 return (error); | |
789 | 524 } |
525 | |
526 /* | |
527 * Pool Open/Import | |
528 * | |
529 * The import case is identical to an open except that the configuration is sent | |
530 * down from userland, instead of grabbed from the configuration cache. For the | |
531 * case of an open, the pool configuration will exist in the | |
532 * POOL_STATE_UNITIALIZED state. | |
533 * | |
534 * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
535 * the same time open the pool, without having to keep around the spa_t in some | |
536 * ambiguous state. | |
537 */ | |
538 static int | |
539 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) | |
540 { | |
541 spa_t *spa; | |
542 int error; | |
543 int loaded = B_FALSE; | |
544 int locked = B_FALSE; | |
545 | |
546 *spapp = NULL; | |
547 | |
548 /* | |
549 * As disgusting as this is, we need to support recursive calls to this | |
550 * function because dsl_dir_open() is called during spa_load(), and ends | |
551 * up calling spa_open() again. The real fix is to figure out how to | |
552 * avoid dsl_dir_open() calling this in the first place. | |
553 */ | |
554 if (mutex_owner(&spa_namespace_lock) != curthread) { | |
555 mutex_enter(&spa_namespace_lock); | |
556 locked = B_TRUE; | |
557 } | |
558 | |
559 if ((spa = spa_lookup(pool)) == NULL) { | |
560 if (locked) | |
561 mutex_exit(&spa_namespace_lock); | |
562 return (ENOENT); | |
563 } | |
564 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
565 | |
566 spa_activate(spa); | |
567 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
568 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); |
789 | 569 |
570 if (error == EBADF) { | |
571 /* | |
572 * If vdev_load() returns EBADF, it indicates that one | |
573 * of the vdevs indicates that the pool has been | |
574 * exported or destroyed. If this is the case, the | |
575 * config cache is out of sync and we should remove the | |
576 * pool from the namespace. | |
577 */ | |
578 spa_unload(spa); | |
579 spa_deactivate(spa); | |
580 spa_remove(spa); | |
581 spa_config_sync(); | |
582 if (locked) | |
583 mutex_exit(&spa_namespace_lock); | |
584 return (ENOENT); | |
1544 | 585 } |
586 | |
587 if (error) { | |
789 | 588 /* |
589 * We can't open the pool, but we still have useful | |
590 * information: the state of each vdev after the | |
591 * attempted vdev_open(). Return this to the user. | |
592 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
593 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
|
594 spa_config_enter(spa, RW_READER, FTAG); |
789 | 595 *config = spa_config_generate(spa, NULL, -1ULL, |
596 B_TRUE); | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
597 spa_config_exit(spa, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
598 } |
789 | 599 spa_unload(spa); |
600 spa_deactivate(spa); | |
1544 | 601 spa->spa_last_open_failed = B_TRUE; |
789 | 602 if (locked) |
603 mutex_exit(&spa_namespace_lock); | |
604 *spapp = NULL; | |
605 return (error); | |
1544 | 606 } else { |
607 zfs_post_ok(spa, NULL); | |
608 spa->spa_last_open_failed = B_FALSE; | |
789 | 609 } |
610 | |
611 loaded = B_TRUE; | |
612 } | |
613 | |
614 spa_open_ref(spa, tag); | |
615 if (locked) | |
616 mutex_exit(&spa_namespace_lock); | |
617 | |
618 *spapp = spa; | |
619 | |
620 if (config != NULL) { | |
1544 | 621 spa_config_enter(spa, RW_READER, FTAG); |
789 | 622 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
1544 | 623 spa_config_exit(spa, FTAG); |
789 | 624 } |
625 | |
626 /* | |
627 * If we just loaded the pool, resilver anything that's out of date. | |
628 */ | |
629 if (loaded && (spa_mode & FWRITE)) | |
630 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
631 | |
632 return (0); | |
633 } | |
634 | |
635 int | |
636 spa_open(const char *name, spa_t **spapp, void *tag) | |
637 { | |
638 return (spa_open_common(name, spapp, tag, NULL)); | |
639 } | |
640 | |
1544 | 641 /* |
642 * Lookup the given spa_t, incrementing the inject count in the process, | |
643 * preventing it from being exported or destroyed. | |
644 */ | |
645 spa_t * | |
646 spa_inject_addref(char *name) | |
647 { | |
648 spa_t *spa; | |
649 | |
650 mutex_enter(&spa_namespace_lock); | |
651 if ((spa = spa_lookup(name)) == NULL) { | |
652 mutex_exit(&spa_namespace_lock); | |
653 return (NULL); | |
654 } | |
655 spa->spa_inject_ref++; | |
656 mutex_exit(&spa_namespace_lock); | |
657 | |
658 return (spa); | |
659 } | |
660 | |
661 void | |
662 spa_inject_delref(spa_t *spa) | |
663 { | |
664 mutex_enter(&spa_namespace_lock); | |
665 spa->spa_inject_ref--; | |
666 mutex_exit(&spa_namespace_lock); | |
667 } | |
668 | |
789 | 669 int |
1544 | 670 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) |
789 | 671 { |
672 int error; | |
673 spa_t *spa; | |
674 | |
675 *config = NULL; | |
676 error = spa_open_common(name, &spa, FTAG, config); | |
677 | |
1544 | 678 if (spa && *config != NULL) |
679 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, | |
680 spa_get_errlog_size(spa)) == 0); | |
681 | |
682 /* | |
683 * We want to get the alternate root even for faulted pools, so we cheat | |
684 * and call spa_lookup() directly. | |
685 */ | |
686 if (altroot) { | |
687 if (spa == NULL) { | |
688 mutex_enter(&spa_namespace_lock); | |
689 spa = spa_lookup(name); | |
690 if (spa) | |
691 spa_altroot(spa, altroot, buflen); | |
692 else | |
693 altroot[0] = '\0'; | |
694 spa = NULL; | |
695 mutex_exit(&spa_namespace_lock); | |
696 } else { | |
697 spa_altroot(spa, altroot, buflen); | |
698 } | |
699 } | |
700 | |
789 | 701 if (spa != NULL) |
702 spa_close(spa, FTAG); | |
703 | |
704 return (error); | |
705 } | |
706 | |
707 /* | |
708 * Pool Creation | |
709 */ | |
710 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
711 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot) |
789 | 712 { |
713 spa_t *spa; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
714 vdev_t *rvd; |
789 | 715 dsl_pool_t *dp; |
716 dmu_tx_t *tx; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
717 int c, error; |
789 | 718 uint64_t txg = TXG_INITIAL; |
719 | |
720 /* | |
721 * If this pool already exists, return failure. | |
722 */ | |
723 mutex_enter(&spa_namespace_lock); | |
724 if (spa_lookup(pool) != NULL) { | |
725 mutex_exit(&spa_namespace_lock); | |
726 return (EEXIST); | |
727 } | |
728 | |
729 /* | |
730 * Allocate a new spa_t structure. | |
731 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
732 spa = spa_add(pool, altroot); |
789 | 733 spa_activate(spa); |
734 | |
735 spa->spa_uberblock.ub_txg = txg - 1; | |
1760 | 736 spa->spa_uberblock.ub_version = ZFS_VERSION; |
789 | 737 spa->spa_ubsync = spa->spa_uberblock; |
738 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
739 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
740 * Create the root vdev. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
741 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
742 spa_config_enter(spa, RW_WRITER, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
743 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
744 rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
745 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
746 ASSERT(spa->spa_root_vdev == rvd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
747 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
748 if (rvd == NULL) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
749 error = EINVAL; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
750 } else { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
751 if ((error = vdev_create(rvd, txg)) == 0) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
752 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
|
753 vdev_init(rvd->vdev_child[c], txg); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
754 vdev_config_dirty(rvd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
755 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
756 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
757 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
758 spa_config_exit(spa, FTAG); |
789 | 759 |
760 if (error) { | |
761 spa_unload(spa); | |
762 spa_deactivate(spa); | |
763 spa_remove(spa); | |
764 mutex_exit(&spa_namespace_lock); | |
765 return (error); | |
766 } | |
767 | |
768 spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg); | |
769 spa->spa_meta_objset = dp->dp_meta_objset; | |
770 | |
771 tx = dmu_tx_create_assigned(dp, txg); | |
772 | |
773 /* | |
774 * Create the pool config object. | |
775 */ | |
776 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
777 DMU_OT_PACKED_NVLIST, 1 << 14, | |
778 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
779 | |
1544 | 780 if (zap_add(spa->spa_meta_objset, |
789 | 781 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 782 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { |
783 cmn_err(CE_PANIC, "failed to add pool config"); | |
784 } | |
789 | 785 |
786 /* | |
787 * Create the deferred-free bplist object. Turn off compression | |
788 * because sync-to-convergence takes longer if the blocksize | |
789 * keeps changing. | |
790 */ | |
791 spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, | |
792 1 << 14, tx); | |
793 dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, | |
794 ZIO_COMPRESS_OFF, tx); | |
795 | |
1544 | 796 if (zap_add(spa->spa_meta_objset, |
789 | 797 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, |
1544 | 798 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { |
799 cmn_err(CE_PANIC, "failed to add bplist"); | |
800 } | |
789 | 801 |
802 dmu_tx_commit(tx); | |
803 | |
804 spa->spa_sync_on = B_TRUE; | |
805 txg_sync_start(spa->spa_dsl_pool); | |
806 | |
807 /* | |
808 * We explicitly wait for the first transaction to complete so that our | |
809 * bean counters are appropriately updated. | |
810 */ | |
811 txg_wait_synced(spa->spa_dsl_pool, txg); | |
812 | |
813 spa_config_sync(); | |
814 | |
815 mutex_exit(&spa_namespace_lock); | |
816 | |
817 return (0); | |
818 } | |
819 | |
820 /* | |
821 * Import the given pool into the system. We set up the necessary spa_t and | |
822 * then call spa_load() to do the dirty work. | |
823 */ | |
824 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
825 spa_import(const char *pool, nvlist_t *config, const char *altroot) |
789 | 826 { |
827 spa_t *spa; | |
828 int error; | |
829 | |
830 if (!(spa_mode & FWRITE)) | |
831 return (EROFS); | |
832 | |
833 /* | |
834 * If a pool with this name exists, return failure. | |
835 */ | |
836 mutex_enter(&spa_namespace_lock); | |
837 if (spa_lookup(pool) != NULL) { | |
838 mutex_exit(&spa_namespace_lock); | |
839 return (EEXIST); | |
840 } | |
841 | |
842 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
843 * Create and initialize the spa structure. |
789 | 844 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
845 spa = spa_add(pool, altroot); |
789 | 846 spa_activate(spa); |
847 | |
848 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
849 * Pass off the heavy lifting to spa_load(). |
1732 | 850 * Pass TRUE for mosconfig because the user-supplied config |
851 * 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
|
852 */ |
1732 | 853 error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE); |
789 | 854 |
855 if (error) { | |
856 spa_unload(spa); | |
857 spa_deactivate(spa); | |
858 spa_remove(spa); | |
859 mutex_exit(&spa_namespace_lock); | |
860 return (error); | |
861 } | |
862 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
863 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
864 * 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
|
865 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
866 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
|
867 |
789 | 868 mutex_exit(&spa_namespace_lock); |
869 | |
870 /* | |
871 * Resilver anything that's out of date. | |
872 */ | |
873 if (spa_mode & FWRITE) | |
874 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
875 | |
876 return (0); | |
877 } | |
878 | |
879 /* | |
880 * This (illegal) pool name is used when temporarily importing a spa_t in order | |
881 * to get the vdev stats associated with the imported devices. | |
882 */ | |
883 #define TRYIMPORT_NAME "$import" | |
884 | |
885 nvlist_t * | |
886 spa_tryimport(nvlist_t *tryconfig) | |
887 { | |
888 nvlist_t *config = NULL; | |
889 char *poolname; | |
890 spa_t *spa; | |
891 uint64_t state; | |
892 | |
893 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
894 return (NULL); | |
895 | |
896 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
897 return (NULL); | |
898 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
899 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
900 * Create and initialize the spa structure. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
901 */ |
789 | 902 mutex_enter(&spa_namespace_lock); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
903 spa = spa_add(TRYIMPORT_NAME, NULL); |
789 | 904 spa_activate(spa); |
905 | |
906 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
907 * Pass off the heavy lifting to spa_load(). |
1732 | 908 * Pass TRUE for mosconfig because the user-supplied config |
909 * is actually the one to trust when doing an import. | |
789 | 910 */ |
1732 | 911 (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); |
789 | 912 |
913 /* | |
914 * If 'tryconfig' was at least parsable, return the current config. | |
915 */ | |
916 if (spa->spa_root_vdev != NULL) { | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
917 spa_config_enter(spa, RW_READER, FTAG); |
789 | 918 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
|
919 spa_config_exit(spa, FTAG); |
789 | 920 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, |
921 poolname) == 0); | |
922 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
923 state) == 0); | |
924 } | |
925 | |
926 spa_unload(spa); | |
927 spa_deactivate(spa); | |
928 spa_remove(spa); | |
929 mutex_exit(&spa_namespace_lock); | |
930 | |
931 return (config); | |
932 } | |
933 | |
934 /* | |
935 * Pool export/destroy | |
936 * | |
937 * The act of destroying or exporting a pool is very simple. We make sure there | |
938 * is no more pending I/O and any references to the pool are gone. Then, we | |
939 * update the pool state and sync all the labels to disk, removing the | |
940 * configuration from the cache afterwards. | |
941 */ | |
942 static int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
943 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig) |
789 | 944 { |
945 spa_t *spa; | |
946 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
947 if (oldconfig) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
948 *oldconfig = NULL; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
949 |
789 | 950 if (!(spa_mode & FWRITE)) |
951 return (EROFS); | |
952 | |
953 mutex_enter(&spa_namespace_lock); | |
954 if ((spa = spa_lookup(pool)) == NULL) { | |
955 mutex_exit(&spa_namespace_lock); | |
956 return (ENOENT); | |
957 } | |
958 | |
959 /* | |
1544 | 960 * Put a hold on the pool, drop the namespace lock, stop async tasks, |
961 * reacquire the namespace lock, and see if we can export. | |
962 */ | |
963 spa_open_ref(spa, FTAG); | |
964 mutex_exit(&spa_namespace_lock); | |
965 spa_async_suspend(spa); | |
966 mutex_enter(&spa_namespace_lock); | |
967 spa_close(spa, FTAG); | |
968 | |
969 /* | |
789 | 970 * The pool will be in core if it's openable, |
971 * in which case we can modify its state. | |
972 */ | |
973 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { | |
974 /* | |
975 * Objsets may be open only because they're dirty, so we | |
976 * have to force it to sync before checking spa_refcnt. | |
977 */ | |
978 spa_scrub_suspend(spa); | |
979 txg_wait_synced(spa->spa_dsl_pool, 0); | |
980 | |
1544 | 981 /* |
982 * A pool cannot be exported or destroyed if there are active | |
983 * references. If we are resetting a pool, allow references by | |
984 * fault injection handlers. | |
985 */ | |
986 if (!spa_refcount_zero(spa) || | |
987 (spa->spa_inject_ref != 0 && | |
988 new_state != POOL_STATE_UNINITIALIZED)) { | |
789 | 989 spa_scrub_resume(spa); |
1544 | 990 spa_async_resume(spa); |
789 | 991 mutex_exit(&spa_namespace_lock); |
992 return (EBUSY); | |
993 } | |
994 | |
995 spa_scrub_resume(spa); | |
996 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); | |
997 | |
998 /* | |
999 * We want this to be reflected on every label, | |
1000 * so mark them all dirty. spa_unload() will do the | |
1001 * final sync that pushes these changes out. | |
1002 */ | |
1544 | 1003 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
|
1004 spa_config_enter(spa, RW_WRITER, FTAG); |
1544 | 1005 spa->spa_state = new_state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1006 spa->spa_final_txg = spa_last_synced_txg(spa) + 1; |
1544 | 1007 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
|
1008 spa_config_exit(spa, FTAG); |
1544 | 1009 } |
789 | 1010 } |
1011 | |
1012 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
1013 spa_unload(spa); | |
1014 spa_deactivate(spa); | |
1015 } | |
1016 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1017 if (oldconfig && spa->spa_config) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1018 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
|
1019 |
1544 | 1020 if (new_state != POOL_STATE_UNINITIALIZED) { |
1021 spa_remove(spa); | |
1022 spa_config_sync(); | |
1023 } | |
789 | 1024 mutex_exit(&spa_namespace_lock); |
1025 | |
1026 return (0); | |
1027 } | |
1028 | |
1029 /* | |
1030 * Destroy a storage pool. | |
1031 */ | |
1032 int | |
1033 spa_destroy(char *pool) | |
1034 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1035 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL)); |
789 | 1036 } |
1037 | |
1038 /* | |
1039 * Export a storage pool. | |
1040 */ | |
1041 int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1042 spa_export(char *pool, nvlist_t **oldconfig) |
789 | 1043 { |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1044 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig)); |
789 | 1045 } |
1046 | |
1047 /* | |
1544 | 1048 * Similar to spa_export(), this unloads the spa_t without actually removing it |
1049 * from the namespace in any way. | |
1050 */ | |
1051 int | |
1052 spa_reset(char *pool) | |
1053 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1054 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL)); |
1544 | 1055 } |
1056 | |
1057 | |
1058 /* | |
789 | 1059 * ========================================================================== |
1060 * Device manipulation | |
1061 * ========================================================================== | |
1062 */ | |
1063 | |
1064 /* | |
1065 * Add capacity to a storage pool. | |
1066 */ | |
1067 int | |
1068 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
1069 { | |
1070 uint64_t txg; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1071 int c, error; |
789 | 1072 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
|
1073 vdev_t *vd, *tvd; |
789 | 1074 |
1075 txg = spa_vdev_enter(spa); | |
1076 | |
1077 vd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
1078 | |
1079 if (vd == NULL) | |
1080 return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
1081 | |
1082 if ((error = vdev_create(vd, txg)) != 0) | |
1083 return (spa_vdev_exit(spa, vd, txg, error)); | |
1084 | |
1085 /* | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1086 * Transfer each new top-level vdev from vd to rvd. |
789 | 1087 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1088 for (c = 0; c < vd->vdev_children; c++) { |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1089 tvd = vd->vdev_child[c]; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1090 vdev_remove_child(vd, tvd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1091 tvd->vdev_id = rvd->vdev_children; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1092 vdev_add_child(rvd, tvd); |
789 | 1093 vdev_config_dirty(tvd); |
1094 } | |
1095 | |
1096 /* | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1097 * 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
|
1098 * 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
|
1099 * 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
|
1100 * 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
|
1101 * 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
|
1102 * 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
|
1103 * 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
|
1104 * |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1105 * 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
|
1106 * 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
|
1107 * steps will be completed the next time we load the pool. |
789 | 1108 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1109 (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
|
1110 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1111 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1112 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
|
1113 mutex_exit(&spa_namespace_lock); |
789 | 1114 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1115 return (0); |
789 | 1116 } |
1117 | |
1118 /* | |
1119 * Attach a device to a mirror. The arguments are the path to any device | |
1120 * in the mirror, and the nvroot for the new device. If the path specifies | |
1121 * a device that is not mirrored, we automatically insert the mirror vdev. | |
1122 * | |
1123 * If 'replacing' is specified, the new device is intended to replace the | |
1124 * existing device; in this case the two devices are made into their own | |
1125 * mirror using the 'replacing' vdev, which is functionally idendical to | |
1126 * the mirror vdev (it actually reuses all the same ops) but has a few | |
1127 * extra rules: you can't attach to it after it's been created, and upon | |
1128 * completion of resilvering, the first disk (the one being replaced) | |
1129 * is automatically detached. | |
1130 */ | |
1131 int | |
1544 | 1132 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) |
789 | 1133 { |
1134 uint64_t txg, open_txg; | |
1135 int error; | |
1136 vdev_t *rvd = spa->spa_root_vdev; | |
1137 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
1138 vdev_ops_t *pvops = replacing ? &vdev_replacing_ops : &vdev_mirror_ops; | |
1139 | |
1140 txg = spa_vdev_enter(spa); | |
1141 | |
1544 | 1142 oldvd = vdev_lookup_by_guid(rvd, guid); |
789 | 1143 |
1144 if (oldvd == NULL) | |
1145 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1146 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1147 if (!oldvd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1148 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
|
1149 |
789 | 1150 pvd = oldvd->vdev_parent; |
1151 | |
1152 /* | |
1153 * The parent must be a mirror or the root, unless we're replacing; | |
1154 * in that case, the parent can be anything but another replacing vdev. | |
1155 */ | |
1156 if (pvd->vdev_ops != &vdev_mirror_ops && | |
1157 pvd->vdev_ops != &vdev_root_ops && | |
1158 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) | |
1159 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1160 | |
1161 newrootvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
1162 | |
1163 if (newrootvd == NULL || newrootvd->vdev_children != 1) | |
1164 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
1165 | |
1166 newvd = newrootvd->vdev_child[0]; | |
1167 | |
1168 if (!newvd->vdev_ops->vdev_op_leaf) | |
1169 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
1170 | |
1171 if ((error = vdev_create(newrootvd, txg)) != 0) | |
1172 return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
1173 | |
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
|
1174 /* |
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
|
1175 * 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
|
1176 * 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
|
1177 */ |
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
|
1178 if (newvd->vdev_psize < vdev_get_rsize(oldvd)) |
789 | 1179 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
1180 | |
1732 | 1181 /* |
1182 * The new device cannot have a higher alignment requirement | |
1183 * than the top-level vdev. | |
1184 */ | |
1185 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
789 | 1186 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); |
1187 | |
1188 /* | |
1189 * If this is an in-place replacement, update oldvd's path and devid | |
1190 * to make it distinguishable from newvd, and unopenable from now on. | |
1191 */ | |
1192 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
1193 spa_strfree(oldvd->vdev_path); | |
1194 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
1195 KM_SLEEP); | |
1196 (void) sprintf(oldvd->vdev_path, "%s/%s", | |
1197 newvd->vdev_path, "old"); | |
1198 if (oldvd->vdev_devid != NULL) { | |
1199 spa_strfree(oldvd->vdev_devid); | |
1200 oldvd->vdev_devid = NULL; | |
1201 } | |
1202 } | |
1203 | |
1204 /* | |
1205 * If the parent is not a mirror, or if we're replacing, | |
1206 * insert the new mirror/replacing vdev above oldvd. | |
1207 */ | |
1208 if (pvd->vdev_ops != pvops) | |
1209 pvd = vdev_add_parent(oldvd, pvops); | |
1210 | |
1211 ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
1212 ASSERT(pvd->vdev_ops == pvops); | |
1213 ASSERT(oldvd->vdev_parent == pvd); | |
1214 | |
1215 /* | |
1216 * Extract the new device from its root and add it to pvd. | |
1217 */ | |
1218 vdev_remove_child(newrootvd, newvd); | |
1219 newvd->vdev_id = pvd->vdev_children; | |
1220 vdev_add_child(pvd, newvd); | |
1221 | |
1544 | 1222 /* |
1223 * If newvd is smaller than oldvd, but larger than its rsize, | |
1224 * the addition of newvd may have decreased our parent's asize. | |
1225 */ | |
1226 pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); | |
1227 | |
789 | 1228 tvd = newvd->vdev_top; |
1229 ASSERT(pvd->vdev_top == tvd); | |
1230 ASSERT(tvd->vdev_parent == rvd); | |
1231 | |
1232 vdev_config_dirty(tvd); | |
1233 | |
1234 /* | |
1235 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate | |
1236 * upward when spa_vdev_exit() calls vdev_dtl_reassess(). | |
1237 */ | |
1238 open_txg = txg + TXG_CONCURRENT_STATES - 1; | |
1239 | |
1240 mutex_enter(&newvd->vdev_dtl_lock); | |
1241 space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, | |
1242 open_txg - TXG_INITIAL + 1); | |
1243 mutex_exit(&newvd->vdev_dtl_lock); | |
1244 | |
1544 | 1245 dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg); |
1246 | |
789 | 1247 /* |
1248 * Mark newvd's DTL dirty in this txg. | |
1249 */ | |
1732 | 1250 vdev_dirty(tvd, VDD_DTL, newvd, txg); |
789 | 1251 |
1252 (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); | |
1253 | |
1254 /* | |
1255 * Kick off a resilver to update newvd. | |
1256 */ | |
1257 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1258 | |
1259 return (0); | |
1260 } | |
1261 | |
1262 /* | |
1263 * Detach a device from a mirror or replacing vdev. | |
1264 * If 'replace_done' is specified, only detach if the parent | |
1265 * is a replacing vdev. | |
1266 */ | |
1267 int | |
1544 | 1268 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) |
789 | 1269 { |
1270 uint64_t txg; | |
1271 int c, t, error; | |
1272 vdev_t *rvd = spa->spa_root_vdev; | |
1273 vdev_t *vd, *pvd, *cvd, *tvd; | |
1274 | |
1275 txg = spa_vdev_enter(spa); | |
1276 | |
1544 | 1277 vd = vdev_lookup_by_guid(rvd, guid); |
789 | 1278 |
1279 if (vd == NULL) | |
1280 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1281 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1282 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1283 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
|
1284 |
789 | 1285 pvd = vd->vdev_parent; |
1286 | |
1287 /* | |
1288 * If replace_done is specified, only remove this device if it's | |
1289 * the first child of a replacing vdev. | |
1290 */ | |
1291 if (replace_done && | |
1292 (vd->vdev_id != 0 || pvd->vdev_ops != &vdev_replacing_ops)) | |
1293 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1294 | |
1295 /* | |
1296 * Only mirror and replacing vdevs support detach. | |
1297 */ | |
1298 if (pvd->vdev_ops != &vdev_replacing_ops && | |
1299 pvd->vdev_ops != &vdev_mirror_ops) | |
1300 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1301 | |
1302 /* | |
1303 * If there's only one replica, you can't detach it. | |
1304 */ | |
1305 if (pvd->vdev_children <= 1) | |
1306 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
1307 | |
1308 /* | |
1309 * If all siblings have non-empty DTLs, this device may have the only | |
1310 * valid copy of the data, which means we cannot safely detach it. | |
1311 * | |
1312 * XXX -- as in the vdev_offline() case, we really want a more | |
1313 * precise DTL check. | |
1314 */ | |
1315 for (c = 0; c < pvd->vdev_children; c++) { | |
1316 uint64_t dirty; | |
1317 | |
1318 cvd = pvd->vdev_child[c]; | |
1319 if (cvd == vd) | |
1320 continue; | |
1321 if (vdev_is_dead(cvd)) | |
1322 continue; | |
1323 mutex_enter(&cvd->vdev_dtl_lock); | |
1324 dirty = cvd->vdev_dtl_map.sm_space | | |
1325 cvd->vdev_dtl_scrub.sm_space; | |
1326 mutex_exit(&cvd->vdev_dtl_lock); | |
1327 if (!dirty) | |
1328 break; | |
1329 } | |
1330 if (c == pvd->vdev_children) | |
1331 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
1332 | |
1333 /* | |
1334 * Erase the disk labels so the disk can be used for other things. | |
1335 * This must be done after all other error cases are handled, | |
1336 * but before we disembowel vd (so we can still do I/O to it). | |
1337 * But if we can't do it, don't treat the error as fatal -- | |
1338 * it may be that the unwritability of the disk is the reason | |
1339 * it's being detached! | |
1340 */ | |
1341 error = vdev_label_init(vd, 0); | |
1342 if (error) | |
1343 dprintf("unable to erase labels on %s\n", vdev_description(vd)); | |
1344 | |
1345 /* | |
1346 * Remove vd from its parent and compact the parent's children. | |
1347 */ | |
1348 vdev_remove_child(pvd, vd); | |
1349 vdev_compact_children(pvd); | |
1350 | |
1351 /* | |
1352 * Remember one of the remaining children so we can get tvd below. | |
1353 */ | |
1354 cvd = pvd->vdev_child[0]; | |
1355 | |
1356 /* | |
1357 * If the parent mirror/replacing vdev only has one child, | |
1358 * the parent is no longer needed. Remove it from the tree. | |
1359 */ | |
1360 if (pvd->vdev_children == 1) | |
1361 vdev_remove_parent(cvd); | |
1362 | |
1363 /* | |
1364 * We don't set tvd until now because the parent we just removed | |
1365 * may have been the previous top-level vdev. | |
1366 */ | |
1367 tvd = cvd->vdev_top; | |
1368 ASSERT(tvd->vdev_parent == rvd); | |
1369 | |
1370 /* | |
1371 * Reopen this top-level vdev to reassess health after detach. | |
1372 */ | |
1544 | 1373 vdev_reopen(tvd); |
789 | 1374 |
1375 /* | |
1376 * If the device we just detached was smaller than the others, | |
1732 | 1377 * it may be possible to add metaslabs (i.e. grow the pool). |
1378 * vdev_metaslab_init() can't fail because the existing metaslabs | |
1379 * are already in core, so there's nothing to read from disk. | |
789 | 1380 */ |
1732 | 1381 VERIFY(vdev_metaslab_init(tvd, txg) == 0); |
789 | 1382 |
1383 vdev_config_dirty(tvd); | |
1384 | |
1385 /* | |
1386 * Mark vd's DTL as dirty in this txg. | |
1387 * vdev_dtl_sync() will see that vd->vdev_detached is set | |
1388 * and free vd's DTL object in syncing context. | |
1389 * But first make sure we're not on any *other* txg's DTL list, | |
1390 * to prevent vd from being accessed after it's freed. | |
1391 */ | |
1392 for (t = 0; t < TXG_SIZE; t++) | |
1393 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
1732 | 1394 vd->vdev_detached = B_TRUE; |
1395 vdev_dirty(tvd, VDD_DTL, vd, txg); | |
789 | 1396 |
1544 | 1397 dprintf("detached %s in txg %llu\n", vd->vdev_path, txg); |
789 | 1398 |
1399 return (spa_vdev_exit(spa, vd, txg, 0)); | |
1400 } | |
1401 | |
1402 /* | |
1544 | 1403 * Find any device that's done replacing, so we can detach it. |
789 | 1404 */ |
1544 | 1405 static vdev_t * |
1406 spa_vdev_replace_done_hunt(vdev_t *vd) | |
789 | 1407 { |
1544 | 1408 vdev_t *newvd, *oldvd; |
789 | 1409 int c; |
1410 | |
1544 | 1411 for (c = 0; c < vd->vdev_children; c++) { |
1412 oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]); | |
1413 if (oldvd != NULL) | |
1414 return (oldvd); | |
1415 } | |
789 | 1416 |
1417 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { | |
1544 | 1418 oldvd = vd->vdev_child[0]; |
1419 newvd = vd->vdev_child[1]; | |
789 | 1420 |
1544 | 1421 mutex_enter(&newvd->vdev_dtl_lock); |
1422 if (newvd->vdev_dtl_map.sm_space == 0 && | |
1423 newvd->vdev_dtl_scrub.sm_space == 0) { | |
1424 mutex_exit(&newvd->vdev_dtl_lock); | |
1425 return (oldvd); | |
1426 } | |
1427 mutex_exit(&newvd->vdev_dtl_lock); | |
1428 } | |
789 | 1429 |
1544 | 1430 return (NULL); |
789 | 1431 } |
1432 | |
1544 | 1433 static void |
789 | 1434 spa_vdev_replace_done(spa_t *spa) |
1435 { | |
1544 | 1436 vdev_t *vd; |
1437 uint64_t guid; | |
789 | 1438 |
1544 | 1439 spa_config_enter(spa, RW_READER, FTAG); |
789 | 1440 |
1544 | 1441 while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) { |
1442 guid = vd->vdev_guid; | |
1443 spa_config_exit(spa, FTAG); | |
1444 if (spa_vdev_detach(spa, guid, B_TRUE) != 0) | |
1445 return; | |
1446 spa_config_enter(spa, RW_READER, FTAG); | |
789 | 1447 } |
1448 | |
1544 | 1449 spa_config_exit(spa, FTAG); |
789 | 1450 } |
1451 | |
1452 /* | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1453 * 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
|
1454 * on spa_vdev_enter/exit() to synchronize the labels and cache. |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1455 */ |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1456 int |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1457 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
|
1458 { |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1459 vdev_t *rvd, *vd; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1460 uint64_t txg; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1461 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1462 rvd = spa->spa_root_vdev; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1463 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1464 txg = spa_vdev_enter(spa); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1465 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1466 if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1467 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1468 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1469 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1470 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
|
1471 |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1472 spa_strfree(vd->vdev_path); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1473 vd->vdev_path = spa_strdup(newpath); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1474 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1475 vdev_config_dirty(vd->vdev_top); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1476 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1477 return (spa_vdev_exit(spa, NULL, txg, 0)); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1478 } |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1479 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1480 /* |
789 | 1481 * ========================================================================== |
1482 * SPA Scrubbing | |
1483 * ========================================================================== | |
1484 */ | |
1485 | |
1544 | 1486 void |
1487 spa_scrub_throttle(spa_t *spa, int direction) | |
1488 { | |
1489 mutex_enter(&spa->spa_scrub_lock); | |
1490 spa->spa_scrub_throttled += direction; | |
1491 ASSERT(spa->spa_scrub_throttled >= 0); | |
1492 if (spa->spa_scrub_throttled == 0) | |
1493 cv_broadcast(&spa->spa_scrub_io_cv); | |
1494 mutex_exit(&spa->spa_scrub_lock); | |
1495 } | |
789 | 1496 |
1497 static void | |
1498 spa_scrub_io_done(zio_t *zio) | |
1499 { | |
1500 spa_t *spa = zio->io_spa; | |
1501 | |
1502 zio_buf_free(zio->io_data, zio->io_size); | |
1503 | |
1504 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 1505 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
|
1506 vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev; |
789 | 1507 spa->spa_scrub_errors++; |
1508 mutex_enter(&vd->vdev_stat_lock); | |
1509 vd->vdev_stat.vs_scrub_errors++; | |
1510 mutex_exit(&vd->vdev_stat_lock); | |
1511 } | |
1544 | 1512 if (--spa->spa_scrub_inflight == 0) { |
1513 cv_broadcast(&spa->spa_scrub_io_cv); | |
1514 ASSERT(spa->spa_scrub_throttled == 0); | |
1515 } | |
1516 mutex_exit(&spa->spa_scrub_lock); | |
789 | 1517 } |
1518 | |
1519 static void | |
1544 | 1520 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags, |
1521 zbookmark_t *zb) | |
789 | 1522 { |
1523 size_t size = BP_GET_LSIZE(bp); | |
1524 void *data = zio_buf_alloc(size); | |
1525 | |
1526 mutex_enter(&spa->spa_scrub_lock); | |
1527 spa->spa_scrub_inflight++; | |
1528 mutex_exit(&spa->spa_scrub_lock); | |
1529 | |
1544 | 1530 if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET) |
1531 flags |= ZIO_FLAG_SPECULATIVE; /* intent log block */ | |
1532 | |
1533 flags |= ZIO_FLAG_CANFAIL; | |
1534 | |
789 | 1535 zio_nowait(zio_read(NULL, spa, bp, data, size, |
1544 | 1536 spa_scrub_io_done, NULL, priority, flags, zb)); |
789 | 1537 } |
1538 | |
1539 /* ARGSUSED */ | |
1540 static int | |
1541 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a) | |
1542 { | |
1543 blkptr_t *bp = &bc->bc_blkptr; | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1544 vdev_t *vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1545 dva_t *dva = bp->blk_dva; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1546 int needs_resilver = B_FALSE; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1547 int d; |
789 | 1548 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1549 if (bc->bc_errno) { |
789 | 1550 /* |
1551 * We can't scrub this block, but we can continue to scrub | |
1552 * the rest of the pool. Note the error and move along. | |
1553 */ | |
1554 mutex_enter(&spa->spa_scrub_lock); | |
1555 spa->spa_scrub_errors++; | |
1556 mutex_exit(&spa->spa_scrub_lock); | |
1557 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1558 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1559 vd->vdev_stat.vs_scrub_errors++; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1560 mutex_exit(&vd->vdev_stat_lock); |
789 | 1561 |
1562 return (ERESTART); | |
1563 } | |
1564 | |
1565 ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg); | |
1566 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1567 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
|
1568 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
|
1569 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1570 ASSERT(vd != NULL); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1571 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1572 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1573 * 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
|
1574 * 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
|
1575 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1576 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1577 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
|
1578 mutex_exit(&vd->vdev_stat_lock); |
789 | 1579 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1580 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
|
1581 if (DVA_GET_GANG(&dva[d])) { |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1582 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1583 * Gang members may be spread across multiple |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1584 * 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
|
1585 * pool-wide DTL. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1586 * 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
|
1587 * 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
|
1588 * happen. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1589 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1590 vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1591 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1592 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
|
1593 bp->blk_birth, 1)) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1594 needs_resilver = B_TRUE; |
789 | 1595 } |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1596 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1597 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1598 if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING) |
789 | 1599 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB, |
1544 | 1600 ZIO_FLAG_SCRUB, &bc->bc_bookmark); |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1601 else if (needs_resilver) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1602 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
|
1603 ZIO_FLAG_RESILVER, &bc->bc_bookmark); |
789 | 1604 |
1605 return (0); | |
1606 } | |
1607 | |
1608 static void | |
1609 spa_scrub_thread(spa_t *spa) | |
1610 { | |
1611 callb_cpr_t cprinfo; | |
1612 traverse_handle_t *th = spa->spa_scrub_th; | |
1613 vdev_t *rvd = spa->spa_root_vdev; | |
1614 pool_scrub_type_t scrub_type = spa->spa_scrub_type; | |
1615 int error = 0; | |
1616 boolean_t complete; | |
1617 | |
1618 CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG); | |
1619 | |
797
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1620 /* |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1621 * 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
|
1622 * wait for that to complete. |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1623 */ |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1624 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
|
1625 |
1544 | 1626 dprintf("start %s mintxg=%llu maxtxg=%llu\n", |
1627 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
1628 spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg); | |
1629 | |
1630 spa_config_enter(spa, RW_WRITER, FTAG); | |
1631 vdev_reopen(rvd); /* purge all vdev caches */ | |
789 | 1632 vdev_config_dirty(rvd); /* rewrite all disk labels */ |
1633 vdev_scrub_stat_update(rvd, scrub_type, B_FALSE); | |
1544 | 1634 spa_config_exit(spa, FTAG); |
789 | 1635 |
1636 mutex_enter(&spa->spa_scrub_lock); | |
1637 spa->spa_scrub_errors = 0; | |
1638 spa->spa_scrub_active = 1; | |
1544 | 1639 ASSERT(spa->spa_scrub_inflight == 0); |
1640 ASSERT(spa->spa_scrub_throttled == 0); | |
789 | 1641 |
1642 while (!spa->spa_scrub_stop) { | |
1643 CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1544 | 1644 while (spa->spa_scrub_suspended) { |
789 | 1645 spa->spa_scrub_active = 0; |
1646 cv_broadcast(&spa->spa_scrub_cv); | |
1647 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
1648 spa->spa_scrub_active = 1; | |
1649 } | |
1650 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock); | |
1651 | |
1652 if (spa->spa_scrub_restart_txg != 0) | |
1653 break; | |
1654 | |
1655 mutex_exit(&spa->spa_scrub_lock); | |
1656 error = traverse_more(th); | |
1657 mutex_enter(&spa->spa_scrub_lock); | |
1658 if (error != EAGAIN) | |
1659 break; | |
1544 | 1660 |
1661 while (spa->spa_scrub_throttled > 0) | |
1662 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
789 | 1663 } |
1664 | |
1665 while (spa->spa_scrub_inflight) | |
1666 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
1667 | |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1668 spa->spa_scrub_active = 0; |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1669 cv_broadcast(&spa->spa_scrub_cv); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1670 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1671 mutex_exit(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1672 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1673 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
|
1674 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1675 mutex_enter(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1676 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1677 /* |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1678 * 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
|
1679 * 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
|
1680 * 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
|
1681 */ |
789 | 1682 if (spa->spa_scrub_restart_txg != 0) |
1683 error = ERESTART; | |
1684 | |
1544 | 1685 if (spa->spa_scrub_stop) |
1686 error = EINTR; | |
1687 | |
789 | 1688 /* |
1544 | 1689 * Even if there were uncorrectable errors, we consider the scrub |
1690 * completed. The downside is that if there is a transient error during | |
1691 * a resilver, we won't resilver the data properly to the target. But | |
1692 * if the damage is permanent (more likely) we will resilver forever, | |
1693 * which isn't really acceptable. Since there is enough information for | |
1694 * the user to know what has failed and why, this seems like a more | |
1695 * tractable approach. | |
789 | 1696 */ |
1544 | 1697 complete = (error == 0); |
789 | 1698 |
1544 | 1699 dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n", |
1700 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
789 | 1701 spa->spa_scrub_maxtxg, complete ? "done" : "FAILED", |
1702 error, spa->spa_scrub_errors, spa->spa_scrub_stop); | |
1703 | |
1704 mutex_exit(&spa->spa_scrub_lock); | |
1705 | |
1706 /* | |
1707 * If the scrub/resilver completed, update all DTLs to reflect this. | |
1708 * Whether it succeeded or not, vacate all temporary scrub DTLs. | |
1709 */ | |
1710 vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1, | |
1711 complete ? spa->spa_scrub_maxtxg : 0, B_TRUE); | |
1712 vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete); | |
1544 | 1713 spa_errlog_rotate(spa); |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1714 |
1544 | 1715 spa_config_exit(spa, FTAG); |
789 | 1716 |
1717 mutex_enter(&spa->spa_scrub_lock); | |
1718 | |
1544 | 1719 /* |
1720 * We may have finished replacing a device. | |
1721 * Let the async thread assess this and handle the detach. | |
1722 */ | |
1723 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); | |
789 | 1724 |
1725 /* | |
1726 * If we were told to restart, our final act is to start a new scrub. | |
1727 */ | |
1728 if (error == ERESTART) | |
1544 | 1729 spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ? |
1730 SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB); | |
789 | 1731 |
1544 | 1732 spa->spa_scrub_type = POOL_SCRUB_NONE; |
1733 spa->spa_scrub_active = 0; | |
1734 spa->spa_scrub_thread = NULL; | |
1735 cv_broadcast(&spa->spa_scrub_cv); | |
789 | 1736 CALLB_CPR_EXIT(&cprinfo); /* drops &spa->spa_scrub_lock */ |
1737 thread_exit(); | |
1738 } | |
1739 | |
1740 void | |
1741 spa_scrub_suspend(spa_t *spa) | |
1742 { | |
1743 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 1744 spa->spa_scrub_suspended++; |
789 | 1745 while (spa->spa_scrub_active) { |
1746 cv_broadcast(&spa->spa_scrub_cv); | |
1747 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
1748 } | |
1749 while (spa->spa_scrub_inflight) | |
1750 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
1751 mutex_exit(&spa->spa_scrub_lock); | |
1752 } | |
1753 | |
1754 void | |
1755 spa_scrub_resume(spa_t *spa) | |
1756 { | |
1757 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 1758 ASSERT(spa->spa_scrub_suspended != 0); |
1759 if (--spa->spa_scrub_suspended == 0) | |
789 | 1760 cv_broadcast(&spa->spa_scrub_cv); |
1761 mutex_exit(&spa->spa_scrub_lock); | |
1762 } | |
1763 | |
1764 void | |
1765 spa_scrub_restart(spa_t *spa, uint64_t txg) | |
1766 { | |
1767 /* | |
1768 * Something happened (e.g. snapshot create/delete) that means | |
1769 * we must restart any in-progress scrubs. The itinerary will | |
1770 * fix this properly. | |
1771 */ | |
1772 mutex_enter(&spa->spa_scrub_lock); | |
1773 spa->spa_scrub_restart_txg = txg; | |
1774 mutex_exit(&spa->spa_scrub_lock); | |
1775 } | |
1776 | |
1544 | 1777 int |
1778 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force) | |
789 | 1779 { |
1780 space_seg_t *ss; | |
1781 uint64_t mintxg, maxtxg; | |
1782 vdev_t *rvd = spa->spa_root_vdev; | |
1783 | |
1784 if ((uint_t)type >= POOL_SCRUB_TYPES) | |
1785 return (ENOTSUP); | |
1786 | |
1544 | 1787 mutex_enter(&spa->spa_scrub_lock); |
1788 | |
789 | 1789 /* |
1790 * If there's a scrub or resilver already in progress, stop it. | |
1791 */ | |
1792 while (spa->spa_scrub_thread != NULL) { | |
1793 /* | |
1794 * Don't stop a resilver unless forced. | |
1795 */ | |
1544 | 1796 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) { |
1797 mutex_exit(&spa->spa_scrub_lock); | |
789 | 1798 return (EBUSY); |
1544 | 1799 } |
789 | 1800 spa->spa_scrub_stop = 1; |
1801 cv_broadcast(&spa->spa_scrub_cv); | |
1802 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
1803 } | |
1804 | |
1805 /* | |
1806 * Terminate the previous traverse. | |
1807 */ | |
1808 if (spa->spa_scrub_th != NULL) { | |
1809 traverse_fini(spa->spa_scrub_th); | |
1810 spa->spa_scrub_th = NULL; | |
1811 } | |
1812 | |
1544 | 1813 if (rvd == NULL) { |
1814 ASSERT(spa->spa_scrub_stop == 0); | |
1815 ASSERT(spa->spa_scrub_type == type); | |
1816 ASSERT(spa->spa_scrub_restart_txg == 0); | |
1817 mutex_exit(&spa->spa_scrub_lock); | |
1818 return (0); | |
1819 } | |
789 | 1820 |
1821 mintxg = TXG_INITIAL - 1; | |
1822 maxtxg = spa_last_synced_txg(spa) + 1; | |
1823 | |
1544 | 1824 mutex_enter(&rvd->vdev_dtl_lock); |
789 | 1825 |
1544 | 1826 if (rvd->vdev_dtl_map.sm_space == 0) { |
1827 /* | |
1828 * The pool-wide DTL is empty. | |
1732 | 1829 * If this is a resilver, there's nothing to do except |
1830 * check whether any in-progress replacements have completed. | |
1544 | 1831 */ |
1732 | 1832 if (type == POOL_SCRUB_RESILVER) { |
1544 | 1833 type = POOL_SCRUB_NONE; |
1732 | 1834 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); |
1835 } | |
1544 | 1836 } else { |
1837 /* | |
1838 * The pool-wide DTL is non-empty. | |
1839 * If this is a normal scrub, upgrade to a resilver instead. | |
1840 */ | |
1841 if (type == POOL_SCRUB_EVERYTHING) | |
1842 type = POOL_SCRUB_RESILVER; | |
1843 } | |
789 | 1844 |
1544 | 1845 if (type == POOL_SCRUB_RESILVER) { |
789 | 1846 /* |
1847 * Determine the resilvering boundaries. | |
1848 * | |
1849 * Note: (mintxg, maxtxg) is an open interval, | |
1850 * i.e. mintxg and maxtxg themselves are not included. | |
1851 * | |
1852 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1 | |
1853 * so we don't claim to resilver a txg that's still changing. | |
1854 */ | |
1855 ss = avl_first(&rvd->vdev_dtl_map.sm_root); | |
1544 | 1856 mintxg = ss->ss_start - 1; |
789 | 1857 ss = avl_last(&rvd->vdev_dtl_map.sm_root); |
1544 | 1858 maxtxg = MIN(ss->ss_end, maxtxg); |
789 | 1859 } |
1860 | |
1544 | 1861 mutex_exit(&rvd->vdev_dtl_lock); |
1862 | |
1863 spa->spa_scrub_stop = 0; | |
1864 spa->spa_scrub_type = type; | |
1865 spa->spa_scrub_restart_txg = 0; | |
1866 | |
1867 if (type != POOL_SCRUB_NONE) { | |
1868 spa->spa_scrub_mintxg = mintxg; | |
789 | 1869 spa->spa_scrub_maxtxg = maxtxg; |
1870 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
|
1871 ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL, |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1872 ZIO_FLAG_CANFAIL); |
789 | 1873 traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg); |
1874 spa->spa_scrub_thread = thread_create(NULL, 0, | |
1875 spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri); | |
1876 } | |
1877 | |
1544 | 1878 mutex_exit(&spa->spa_scrub_lock); |
1879 | |
789 | 1880 return (0); |
1881 } | |
1882 | |
1544 | 1883 /* |
1884 * ========================================================================== | |
1885 * SPA async task processing | |
1886 * ========================================================================== | |
1887 */ | |
1888 | |
1889 static void | |
1890 spa_async_reopen(spa_t *spa) | |
789 | 1891 { |
1544 | 1892 vdev_t *rvd = spa->spa_root_vdev; |
1893 vdev_t *tvd; | |
1894 int c; | |
1895 | |
1896 spa_config_enter(spa, RW_WRITER, FTAG); | |
1897 | |
1898 for (c = 0; c < rvd->vdev_children; c++) { | |
1899 tvd = rvd->vdev_child[c]; | |
1900 if (tvd->vdev_reopen_wanted) { | |
1901 tvd->vdev_reopen_wanted = 0; | |
1902 vdev_reopen(tvd); | |
1903 } | |
1904 } | |
789 | 1905 |
1544 | 1906 spa_config_exit(spa, FTAG); |
1907 } | |
1908 | |
1909 static void | |
1910 spa_async_thread(spa_t *spa) | |
1911 { | |
1912 int tasks; | |
1913 | |
1914 ASSERT(spa->spa_sync_on); | |
789 | 1915 |
1544 | 1916 mutex_enter(&spa->spa_async_lock); |
1917 tasks = spa->spa_async_tasks; | |
1918 spa->spa_async_tasks = 0; | |
1919 mutex_exit(&spa->spa_async_lock); | |
1920 | |
1921 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1922 * 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
|
1923 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1924 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1925 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1926 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
|
1927 mutex_exit(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1928 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1929 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1930 /* |
1544 | 1931 * See if any devices need to be reopened. |
1932 */ | |
1933 if (tasks & SPA_ASYNC_REOPEN) | |
1934 spa_async_reopen(spa); | |
1935 | |
1936 /* | |
1937 * If any devices are done replacing, detach them. | |
1938 */ | |
1939 if (tasks & SPA_ASYNC_REPLACE_DONE) | |
789 | 1940 spa_vdev_replace_done(spa); |
1941 | |
1544 | 1942 /* |
1943 * Kick off a scrub. | |
1944 */ | |
1945 if (tasks & SPA_ASYNC_SCRUB) | |
1946 VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0); | |
1947 | |
1948 /* | |
1949 * Kick off a resilver. | |
1950 */ | |
1951 if (tasks & SPA_ASYNC_RESILVER) | |
1952 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1953 | |
1954 /* | |
1955 * Let the world know that we're done. | |
1956 */ | |
1957 mutex_enter(&spa->spa_async_lock); | |
1958 spa->spa_async_thread = NULL; | |
1959 cv_broadcast(&spa->spa_async_cv); | |
1960 mutex_exit(&spa->spa_async_lock); | |
1961 thread_exit(); | |
1962 } | |
1963 | |
1964 void | |
1965 spa_async_suspend(spa_t *spa) | |
1966 { | |
1967 mutex_enter(&spa->spa_async_lock); | |
1968 spa->spa_async_suspended++; | |
1969 while (spa->spa_async_thread != NULL) | |
1970 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
1971 mutex_exit(&spa->spa_async_lock); | |
1972 } | |
1973 | |
1974 void | |
1975 spa_async_resume(spa_t *spa) | |
1976 { | |
1977 mutex_enter(&spa->spa_async_lock); | |
1978 ASSERT(spa->spa_async_suspended != 0); | |
1979 spa->spa_async_suspended--; | |
1980 mutex_exit(&spa->spa_async_lock); | |
1981 } | |
1982 | |
1983 static void | |
1984 spa_async_dispatch(spa_t *spa) | |
1985 { | |
1986 mutex_enter(&spa->spa_async_lock); | |
1987 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
|
1988 spa->spa_async_thread == NULL && |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1989 rootdir != NULL && !vn_is_readonly(rootdir)) |
1544 | 1990 spa->spa_async_thread = thread_create(NULL, 0, |
1991 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
1992 mutex_exit(&spa->spa_async_lock); | |
1993 } | |
1994 | |
1995 void | |
1996 spa_async_request(spa_t *spa, int task) | |
1997 { | |
1998 mutex_enter(&spa->spa_async_lock); | |
1999 spa->spa_async_tasks |= task; | |
2000 mutex_exit(&spa->spa_async_lock); | |
789 | 2001 } |
2002 | |
2003 /* | |
2004 * ========================================================================== | |
2005 * SPA syncing routines | |
2006 * ========================================================================== | |
2007 */ | |
2008 | |
2009 static void | |
2010 spa_sync_deferred_frees(spa_t *spa, uint64_t txg) | |
2011 { | |
2012 bplist_t *bpl = &spa->spa_sync_bplist; | |
2013 dmu_tx_t *tx; | |
2014 blkptr_t blk; | |
2015 uint64_t itor = 0; | |
2016 zio_t *zio; | |
2017 int error; | |
2018 uint8_t c = 1; | |
2019 | |
2020 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD); | |
2021 | |
2022 while (bplist_iterate(bpl, &itor, &blk) == 0) | |
2023 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL)); | |
2024 | |
2025 error = zio_wait(zio); | |
2026 ASSERT3U(error, ==, 0); | |
2027 | |
2028 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2029 bplist_vacate(bpl, tx); | |
2030 | |
2031 /* | |
2032 * Pre-dirty the first block so we sync to convergence faster. | |
2033 * (Usually only the first block is needed.) | |
2034 */ | |
2035 dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); | |
2036 dmu_tx_commit(tx); | |
2037 } | |
2038 | |
2039 static void | |
2040 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
2041 { | |
2042 nvlist_t *config; | |
2043 char *packed = NULL; | |
2044 size_t nvsize = 0; | |
2045 dmu_buf_t *db; | |
2046 | |
2047 if (list_is_empty(&spa->spa_dirty_list)) | |
2048 return; | |
2049 | |
2050 config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE); | |
2051 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2052 if (spa->spa_config_syncing) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2053 nvlist_free(spa->spa_config_syncing); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2054 spa->spa_config_syncing = config; |
789 | 2055 |
2056 VERIFY(nvlist_size(config, &nvsize, NV_ENCODE_XDR) == 0); | |
2057 | |
2058 packed = kmem_alloc(nvsize, KM_SLEEP); | |
2059 | |
1544 | 2060 VERIFY(nvlist_pack(config, &packed, &nvsize, NV_ENCODE_XDR, |
2061 KM_SLEEP) == 0); | |
789 | 2062 |
2063 dmu_write(spa->spa_meta_objset, spa->spa_config_object, 0, nvsize, | |
2064 packed, tx); | |
2065 | |
2066 kmem_free(packed, nvsize); | |
2067 | |
1544 | 2068 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, |
2069 spa->spa_config_object, FTAG, &db)); | |
789 | 2070 dmu_buf_will_dirty(db, tx); |
2071 *(uint64_t *)db->db_data = nvsize; | |
1544 | 2072 dmu_buf_rele(db, FTAG); |
789 | 2073 } |
2074 | |
2075 /* | |
2076 * Sync the specified transaction group. New blocks may be dirtied as | |
2077 * part of the process, so we iterate until it converges. | |
2078 */ | |
2079 void | |
2080 spa_sync(spa_t *spa, uint64_t txg) | |
2081 { | |
2082 dsl_pool_t *dp = spa->spa_dsl_pool; | |
2083 objset_t *mos = spa->spa_meta_objset; | |
2084 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
|
2085 vdev_t *rvd = spa->spa_root_vdev; |
789 | 2086 vdev_t *vd; |
2087 dmu_tx_t *tx; | |
2088 int dirty_vdevs; | |
2089 | |
2090 /* | |
2091 * Lock out configuration changes. | |
2092 */ | |
1544 | 2093 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2094 |
2095 spa->spa_syncing_txg = txg; | |
2096 spa->spa_sync_pass = 0; | |
2097 | |
1544 | 2098 VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); |
789 | 2099 |
2100 /* | |
2101 * If anything has changed in this txg, push the deferred frees | |
2102 * from the previous txg. If not, leave them alone so that we | |
2103 * don't generate work on an otherwise idle system. | |
2104 */ | |
2105 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || | |
2106 !txg_list_empty(&dp->dp_dirty_dirs, txg)) | |
2107 spa_sync_deferred_frees(spa, txg); | |
2108 | |
2109 /* | |
2110 * Iterate to convergence. | |
2111 */ | |
2112 do { | |
2113 spa->spa_sync_pass++; | |
2114 | |
2115 tx = dmu_tx_create_assigned(dp, txg); | |
2116 spa_sync_config_object(spa, tx); | |
2117 dmu_tx_commit(tx); | |
2118 | |
1544 | 2119 spa_errlog_sync(spa, txg); |
2120 | |
789 | 2121 dsl_pool_sync(dp, txg); |
2122 | |
2123 dirty_vdevs = 0; | |
2124 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { | |
2125 vdev_sync(vd, txg); | |
2126 dirty_vdevs++; | |
2127 } | |
2128 | |
2129 tx = dmu_tx_create_assigned(dp, txg); | |
2130 bplist_sync(bpl, tx); | |
2131 dmu_tx_commit(tx); | |
2132 | |
2133 } while (dirty_vdevs); | |
2134 | |
2135 bplist_close(bpl); | |
2136 | |
2137 dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); | |
2138 | |
2139 /* | |
2140 * Rewrite the vdev configuration (which includes the uberblock) | |
2141 * to commit the transaction group. | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2142 * |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2143 * 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
|
2144 * 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
|
2145 * 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
|
2146 * If the write fails, try the next vdev until we're tried them all. |
789 | 2147 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2148 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
|
2149 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2150 } else { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2151 int children = rvd->vdev_children; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2152 int c0 = spa_get_random(children); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2153 int c; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2154 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2155 for (c = 0; c < children; c++) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2156 vd = rvd->vdev_child[(c0 + c) % children]; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2157 if (vd->vdev_ms_array == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2158 continue; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2159 if (vdev_config_sync(vd, txg) == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2160 break; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2161 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2162 if (c == children) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2163 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2164 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2165 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2166 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2167 * Clear the dirty config list. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2168 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2169 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
|
2170 vdev_config_clean(vd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2171 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2172 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2173 * 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
|
2174 * 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
|
2175 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2176 if (spa->spa_config_syncing != NULL) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2177 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
|
2178 spa->spa_config_txg = txg; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2179 spa->spa_config_syncing = NULL; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2180 } |
789 | 2181 |
2182 /* | |
2183 * Make a stable copy of the fully synced uberblock. | |
2184 * We use this as the root for pool traversals. | |
2185 */ | |
2186 spa->spa_traverse_wanted = 1; /* tells traverse_more() to stop */ | |
2187 | |
2188 spa_scrub_suspend(spa); /* stop scrubbing and finish I/Os */ | |
2189 | |
2190 rw_enter(&spa->spa_traverse_lock, RW_WRITER); | |
2191 spa->spa_traverse_wanted = 0; | |
2192 spa->spa_ubsync = spa->spa_uberblock; | |
2193 rw_exit(&spa->spa_traverse_lock); | |
2194 | |
2195 spa_scrub_resume(spa); /* resume scrub with new ubsync */ | |
2196 | |
2197 /* | |
2198 * Clean up the ZIL records for the synced txg. | |
2199 */ | |
2200 dsl_pool_zil_clean(dp); | |
2201 | |
2202 /* | |
2203 * Update usable space statistics. | |
2204 */ | |
2205 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) | |
2206 vdev_sync_done(vd, txg); | |
2207 | |
2208 /* | |
2209 * It had better be the case that we didn't dirty anything | |
2210 * since spa_sync_labels(). | |
2211 */ | |
2212 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
2213 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
2214 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
2215 ASSERT(bpl->bpl_queue == NULL); | |
2216 | |
1544 | 2217 spa_config_exit(spa, FTAG); |
2218 | |
2219 /* | |
2220 * If any async tasks have been requested, kick them off. | |
2221 */ | |
2222 spa_async_dispatch(spa); | |
789 | 2223 } |
2224 | |
2225 /* | |
2226 * Sync all pools. We don't want to hold the namespace lock across these | |
2227 * operations, so we take a reference on the spa_t and drop the lock during the | |
2228 * sync. | |
2229 */ | |
2230 void | |
2231 spa_sync_allpools(void) | |
2232 { | |
2233 spa_t *spa = NULL; | |
2234 mutex_enter(&spa_namespace_lock); | |
2235 while ((spa = spa_next(spa)) != NULL) { | |
2236 if (spa_state(spa) != POOL_STATE_ACTIVE) | |
2237 continue; | |
2238 spa_open_ref(spa, FTAG); | |
2239 mutex_exit(&spa_namespace_lock); | |
2240 txg_wait_synced(spa_get_dsl(spa), 0); | |
2241 mutex_enter(&spa_namespace_lock); | |
2242 spa_close(spa, FTAG); | |
2243 } | |
2244 mutex_exit(&spa_namespace_lock); | |
2245 } | |
2246 | |
2247 /* | |
2248 * ========================================================================== | |
2249 * Miscellaneous routines | |
2250 * ========================================================================== | |
2251 */ | |
2252 | |
2253 /* | |
2254 * Remove all pools in the system. | |
2255 */ | |
2256 void | |
2257 spa_evict_all(void) | |
2258 { | |
2259 spa_t *spa; | |
2260 | |
2261 /* | |
2262 * Remove all cached state. All pools should be closed now, | |
2263 * so every spa in the AVL tree should be unreferenced. | |
2264 */ | |
2265 mutex_enter(&spa_namespace_lock); | |
2266 while ((spa = spa_next(NULL)) != NULL) { | |
2267 /* | |
1544 | 2268 * Stop async tasks. The async thread may need to detach |
2269 * a device that's been replaced, which requires grabbing | |
2270 * spa_namespace_lock, so we must drop it here. | |
789 | 2271 */ |
2272 spa_open_ref(spa, FTAG); | |
2273 mutex_exit(&spa_namespace_lock); | |
1544 | 2274 spa_async_suspend(spa); |
789 | 2275 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); |
2276 mutex_enter(&spa_namespace_lock); | |
2277 spa_close(spa, FTAG); | |
2278 | |
2279 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
2280 spa_unload(spa); | |
2281 spa_deactivate(spa); | |
2282 } | |
2283 spa_remove(spa); | |
2284 } | |
2285 mutex_exit(&spa_namespace_lock); | |
2286 } | |
1544 | 2287 |
2288 vdev_t * | |
2289 spa_lookup_by_guid(spa_t *spa, uint64_t guid) | |
2290 { | |
2291 return (vdev_lookup_by_guid(spa->spa_root_vdev, guid)); | |
2292 } | |
1760 | 2293 |
2294 void | |
2295 spa_upgrade(spa_t *spa) | |
2296 { | |
2297 spa_config_enter(spa, RW_WRITER, FTAG); | |
2298 | |
2299 /* | |
2300 * This should only be called for a non-faulted pool, and since a | |
2301 * future version would result in an unopenable pool, this shouldn't be | |
2302 * possible. | |
2303 */ | |
2304 ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION); | |
2305 | |
2306 spa->spa_uberblock.ub_version = ZFS_VERSION; | |
2307 vdev_config_dirty(spa->spa_root_vdev); | |
2308 | |
2309 spa_config_exit(spa, FTAG); | |
2310 } |