Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/spa.c @ 2082:76b439ec3ac1
PSARC 2006/223 ZFS Hot Spares
PSARC 2006/303 ZFS Clone Promotion
6276916 support for "clone swap"
6288488 du reports misleading size on RAID-Z
6393490 libzfs should be a real library
6397148 fbufs debug code should be removed from buf_hash_insert()
6405966 Hot Spare support in ZFS
6409302 passing a non-root vdev via zpool_create() panics system
6415739 assertion failed: !(zio->io_flags & 0x00040)
6416759 ::dbufs does not find bonus buffers anymore
6417978 double parity RAID-Z a.k.a. RAID6
6424554 full block re-writes need not read data in
6425111 detaching an offline device can result in import confusion
author | eschrock |
---|---|
date | Tue, 30 May 2006 15:47:16 -0700 |
parents | 628267397204 |
children | 73de7a781492 |
rev | line source |
---|---|
789 | 1 /* |
2 * CDDL HEADER START | |
3 * | |
4 * The contents of this file are subject to the terms of the | |
1544 | 5 * Common Development and Distribution License (the "License"). |
6 * You may not use this file except in compliance with the License. | |
789 | 7 * |
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 * or http://www.opensolaris.org/os/licensing. | |
10 * See the License for the specific language governing permissions | |
11 * and limitations under the License. | |
12 * | |
13 * When distributing Covered Code, include this CDDL HEADER in each | |
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 * If applicable, add the following below this CDDL HEADER, with the | |
16 * fields enclosed by brackets "[]" replaced with your own identifying | |
17 * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 * | |
19 * CDDL HEADER END | |
20 */ | |
2082 | 21 |
789 | 22 /* |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
23 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
789 | 24 * Use is subject to license terms. |
25 */ | |
26 | |
27 #pragma ident "%Z%%M% %I% %E% SMI" | |
28 | |
29 /* | |
30 * This file contains all the routines used when modifying on-disk SPA state. | |
31 * This includes opening, importing, destroying, exporting a pool, and syncing a | |
32 * pool. | |
33 */ | |
34 | |
35 #include <sys/zfs_context.h> | |
1544 | 36 #include <sys/fm/fs/zfs.h> |
789 | 37 #include <sys/spa_impl.h> |
38 #include <sys/zio.h> | |
39 #include <sys/zio_checksum.h> | |
40 #include <sys/zio_compress.h> | |
41 #include <sys/dmu.h> | |
42 #include <sys/dmu_tx.h> | |
43 #include <sys/zap.h> | |
44 #include <sys/zil.h> | |
45 #include <sys/vdev_impl.h> | |
46 #include <sys/metaslab.h> | |
47 #include <sys/uberblock_impl.h> | |
48 #include <sys/txg.h> | |
49 #include <sys/avl.h> | |
50 #include <sys/dmu_traverse.h> | |
51 #include <sys/unique.h> | |
52 #include <sys/dsl_pool.h> | |
53 #include <sys/dsl_dir.h> | |
54 #include <sys/dsl_prop.h> | |
55 #include <sys/fs/zfs.h> | |
56 #include <sys/callb.h> | |
57 | |
58 /* | |
59 * ========================================================================== | |
60 * SPA state manipulation (open/create/destroy/import/export) | |
61 * ========================================================================== | |
62 */ | |
63 | |
1544 | 64 static int |
65 spa_error_entry_compare(const void *a, const void *b) | |
66 { | |
67 spa_error_entry_t *sa = (spa_error_entry_t *)a; | |
68 spa_error_entry_t *sb = (spa_error_entry_t *)b; | |
69 int ret; | |
70 | |
71 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, | |
72 sizeof (zbookmark_t)); | |
73 | |
74 if (ret < 0) | |
75 return (-1); | |
76 else if (ret > 0) | |
77 return (1); | |
78 else | |
79 return (0); | |
80 } | |
81 | |
82 /* | |
83 * Utility function which retrieves copies of the current logs and | |
84 * re-initializes them in the process. | |
85 */ | |
86 void | |
87 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
88 { | |
89 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
90 | |
91 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
92 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
93 | |
94 avl_create(&spa->spa_errlist_scrub, | |
95 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
96 offsetof(spa_error_entry_t, se_avl)); | |
97 avl_create(&spa->spa_errlist_last, | |
98 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
99 offsetof(spa_error_entry_t, se_avl)); | |
100 } | |
101 | |
789 | 102 /* |
103 * Activate an uninitialized pool. | |
104 */ | |
105 static void | |
106 spa_activate(spa_t *spa) | |
107 { | |
108 int t; | |
109 | |
110 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
111 | |
112 spa->spa_state = POOL_STATE_ACTIVE; | |
113 | |
114 spa->spa_normal_class = metaslab_class_create(); | |
115 | |
116 for (t = 0; t < ZIO_TYPES; t++) { | |
117 spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue", | |
118 8, maxclsyspri, 50, INT_MAX, | |
119 TASKQ_PREPOPULATE); | |
120 spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr", | |
121 8, maxclsyspri, 50, INT_MAX, | |
122 TASKQ_PREPOPULATE); | |
123 } | |
124 | |
125 rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL); | |
126 | |
127 list_create(&spa->spa_dirty_list, sizeof (vdev_t), | |
128 offsetof(vdev_t, vdev_dirty_node)); | |
129 | |
130 txg_list_create(&spa->spa_vdev_txg_list, | |
131 offsetof(struct vdev, vdev_txg_node)); | |
1544 | 132 |
133 avl_create(&spa->spa_errlist_scrub, | |
134 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
135 offsetof(spa_error_entry_t, se_avl)); | |
136 avl_create(&spa->spa_errlist_last, | |
137 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
138 offsetof(spa_error_entry_t, se_avl)); | |
789 | 139 } |
140 | |
141 /* | |
142 * Opposite of spa_activate(). | |
143 */ | |
144 static void | |
145 spa_deactivate(spa_t *spa) | |
146 { | |
147 int t; | |
148 | |
149 ASSERT(spa->spa_sync_on == B_FALSE); | |
150 ASSERT(spa->spa_dsl_pool == NULL); | |
151 ASSERT(spa->spa_root_vdev == NULL); | |
152 | |
153 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
154 | |
155 txg_list_destroy(&spa->spa_vdev_txg_list); | |
156 | |
157 list_destroy(&spa->spa_dirty_list); | |
158 | |
159 rw_destroy(&spa->spa_traverse_lock); | |
160 | |
161 for (t = 0; t < ZIO_TYPES; t++) { | |
162 taskq_destroy(spa->spa_zio_issue_taskq[t]); | |
163 taskq_destroy(spa->spa_zio_intr_taskq[t]); | |
164 spa->spa_zio_issue_taskq[t] = NULL; | |
165 spa->spa_zio_intr_taskq[t] = NULL; | |
166 } | |
167 | |
168 metaslab_class_destroy(spa->spa_normal_class); | |
169 spa->spa_normal_class = NULL; | |
170 | |
1544 | 171 /* |
172 * If this was part of an import or the open otherwise failed, we may | |
173 * still have errors left in the queues. Empty them just in case. | |
174 */ | |
175 spa_errlog_drain(spa); | |
176 | |
177 avl_destroy(&spa->spa_errlist_scrub); | |
178 avl_destroy(&spa->spa_errlist_last); | |
179 | |
789 | 180 spa->spa_state = POOL_STATE_UNINITIALIZED; |
181 } | |
182 | |
183 /* | |
184 * Verify a pool configuration, and construct the vdev tree appropriately. This | |
185 * will create all the necessary vdevs in the appropriate layout, with each vdev | |
186 * in the CLOSED state. This will prep the pool before open/creation/import. | |
187 * All vdev validation is done by the vdev_alloc() routine. | |
188 */ | |
2082 | 189 static int |
190 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, | |
191 uint_t id, int atype) | |
789 | 192 { |
193 nvlist_t **child; | |
194 uint_t c, children; | |
2082 | 195 int error; |
196 | |
197 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) | |
198 return (error); | |
199 | |
200 if ((*vdp)->vdev_ops->vdev_op_leaf) | |
201 return (0); | |
789 | 202 |
203 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
204 &child, &children) != 0) { | |
2082 | 205 vdev_free(*vdp); |
206 *vdp = NULL; | |
207 return (EINVAL); | |
789 | 208 } |
209 | |
210 for (c = 0; c < children; c++) { | |
2082 | 211 vdev_t *vd; |
212 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, | |
213 atype)) != 0) { | |
214 vdev_free(*vdp); | |
215 *vdp = NULL; | |
216 return (error); | |
789 | 217 } |
218 } | |
219 | |
2082 | 220 ASSERT(*vdp != NULL); |
221 | |
222 return (0); | |
789 | 223 } |
224 | |
225 /* | |
226 * Opposite of spa_load(). | |
227 */ | |
228 static void | |
229 spa_unload(spa_t *spa) | |
230 { | |
2082 | 231 int i; |
232 | |
789 | 233 /* |
1544 | 234 * Stop async tasks. |
235 */ | |
236 spa_async_suspend(spa); | |
237 | |
238 /* | |
789 | 239 * Stop syncing. |
240 */ | |
241 if (spa->spa_sync_on) { | |
242 txg_sync_stop(spa->spa_dsl_pool); | |
243 spa->spa_sync_on = B_FALSE; | |
244 } | |
245 | |
246 /* | |
247 * Wait for any outstanding prefetch I/O to complete. | |
248 */ | |
1544 | 249 spa_config_enter(spa, RW_WRITER, FTAG); |
250 spa_config_exit(spa, FTAG); | |
789 | 251 |
252 /* | |
253 * Close the dsl pool. | |
254 */ | |
255 if (spa->spa_dsl_pool) { | |
256 dsl_pool_close(spa->spa_dsl_pool); | |
257 spa->spa_dsl_pool = NULL; | |
258 } | |
259 | |
260 /* | |
261 * Close all vdevs. | |
262 */ | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
263 if (spa->spa_root_vdev) |
789 | 264 vdev_free(spa->spa_root_vdev); |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
265 ASSERT(spa->spa_root_vdev == NULL); |
1544 | 266 |
2082 | 267 for (i = 0; i < spa->spa_nspares; i++) |
268 vdev_free(spa->spa_spares[i]); | |
269 if (spa->spa_spares) { | |
270 kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *)); | |
271 spa->spa_spares = NULL; | |
272 } | |
273 if (spa->spa_sparelist) { | |
274 nvlist_free(spa->spa_sparelist); | |
275 spa->spa_sparelist = NULL; | |
276 } | |
277 | |
1544 | 278 spa->spa_async_suspended = 0; |
789 | 279 } |
280 | |
281 /* | |
2082 | 282 * Load (or re-load) the current list of vdevs describing the active spares for |
283 * this pool. When this is called, we have some form of basic information in | |
284 * 'spa_sparelist'. We parse this into vdevs, try to open them, and then | |
285 * re-generate a more complete list including status information. | |
286 */ | |
287 static void | |
288 spa_load_spares(spa_t *spa) | |
289 { | |
290 nvlist_t **spares; | |
291 uint_t nspares; | |
292 int i; | |
293 | |
294 /* | |
295 * First, close and free any existing spare vdevs. | |
296 */ | |
297 for (i = 0; i < spa->spa_nspares; i++) { | |
298 vdev_close(spa->spa_spares[i]); | |
299 vdev_free(spa->spa_spares[i]); | |
300 } | |
301 if (spa->spa_spares) | |
302 kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *)); | |
303 | |
304 if (spa->spa_sparelist == NULL) | |
305 nspares = 0; | |
306 else | |
307 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
308 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
309 | |
310 spa->spa_nspares = (int)nspares; | |
311 spa->spa_spares = NULL; | |
312 | |
313 if (nspares == 0) | |
314 return; | |
315 | |
316 /* | |
317 * Construct the array of vdevs, opening them to get status in the | |
318 * process. | |
319 */ | |
320 spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP); | |
321 for (i = 0; i < spa->spa_nspares; i++) { | |
322 vdev_t *vd; | |
323 | |
324 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
325 VDEV_ALLOC_SPARE) == 0); | |
326 ASSERT(vd != NULL); | |
327 | |
328 spa->spa_spares[i] = vd; | |
329 | |
330 if (vdev_open(vd) != 0) | |
331 continue; | |
332 | |
333 vd->vdev_top = vd; | |
334 (void) vdev_validate_spare(vd); | |
335 } | |
336 | |
337 /* | |
338 * Recompute the stashed list of spares, with status information | |
339 * this time. | |
340 */ | |
341 VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
342 DATA_TYPE_NVLIST_ARRAY) == 0); | |
343 | |
344 spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP); | |
345 for (i = 0; i < spa->spa_nspares; i++) | |
346 spares[i] = vdev_config_generate(spa, spa->spa_spares[i], | |
347 B_TRUE, B_TRUE); | |
348 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
349 spares, spa->spa_nspares) == 0); | |
350 for (i = 0; i < spa->spa_nspares; i++) | |
351 nvlist_free(spares[i]); | |
352 kmem_free(spares, spa->spa_nspares * sizeof (void *)); | |
353 } | |
354 | |
355 static int | |
356 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) | |
357 { | |
358 dmu_buf_t *db; | |
359 char *packed = NULL; | |
360 size_t nvsize = 0; | |
361 int error; | |
362 *value = NULL; | |
363 | |
364 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
365 nvsize = *(uint64_t *)db->db_data; | |
366 dmu_buf_rele(db, FTAG); | |
367 | |
368 packed = kmem_alloc(nvsize, KM_SLEEP); | |
369 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed); | |
370 if (error == 0) | |
371 error = nvlist_unpack(packed, nvsize, value, 0); | |
372 kmem_free(packed, nvsize); | |
373 | |
374 return (error); | |
375 } | |
376 | |
377 /* | |
789 | 378 * Load an existing storage pool, using the pool's builtin spa_config as a |
1544 | 379 * source of configuration information. |
789 | 380 */ |
381 static int | |
1544 | 382 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) |
789 | 383 { |
384 int error = 0; | |
385 nvlist_t *nvroot = NULL; | |
386 vdev_t *rvd; | |
387 uberblock_t *ub = &spa->spa_uberblock; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
388 uint64_t config_cache_txg = spa->spa_config_txg; |
789 | 389 uint64_t pool_guid; |
2082 | 390 uint64_t version; |
789 | 391 zio_t *zio; |
392 | |
1544 | 393 spa->spa_load_state = state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
394 |
789 | 395 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || |
1733 | 396 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { |
1544 | 397 error = EINVAL; |
398 goto out; | |
399 } | |
789 | 400 |
2082 | 401 /* |
402 * Versioning wasn't explicitly added to the label until later, so if | |
403 * it's not present treat it as the initial version. | |
404 */ | |
405 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0) | |
406 version = ZFS_VERSION_INITIAL; | |
407 | |
1733 | 408 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
409 &spa->spa_config_txg); | |
410 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
411 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && |
1544 | 412 spa_guid_exists(pool_guid, 0)) { |
413 error = EEXIST; | |
414 goto out; | |
415 } | |
789 | 416 |
417 /* | |
2082 | 418 * Parse the configuration into a vdev tree. We explicitly set the |
419 * value that will be returned by spa_version() since parsing the | |
420 * configuration requires knowing the version number. | |
789 | 421 */ |
1544 | 422 spa_config_enter(spa, RW_WRITER, FTAG); |
2082 | 423 spa->spa_ubsync.ub_version = version; |
424 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD); | |
1544 | 425 spa_config_exit(spa, FTAG); |
789 | 426 |
2082 | 427 if (error != 0) |
1544 | 428 goto out; |
789 | 429 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
430 ASSERT(spa->spa_root_vdev == rvd); |
789 | 431 ASSERT(spa_guid(spa) == pool_guid); |
432 | |
433 /* | |
434 * Try to open all vdevs, loading each label in the process. | |
435 */ | |
1544 | 436 if (vdev_open(rvd) != 0) { |
437 error = ENXIO; | |
438 goto out; | |
439 } | |
789 | 440 |
441 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
442 * Validate the labels for all leaf vdevs. We need to grab the config |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
443 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
444 * flag. |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
445 */ |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
446 spa_config_enter(spa, RW_READER, FTAG); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
447 error = vdev_validate(rvd); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
448 spa_config_exit(spa, FTAG); |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
449 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
450 if (error != 0) { |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
451 error = EBADF; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
452 goto out; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
453 } |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
454 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
455 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
456 error = ENXIO; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
457 goto out; |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
458 } |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
459 |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
460 /* |
789 | 461 * Find the best uberblock. |
462 */ | |
463 bzero(ub, sizeof (uberblock_t)); | |
464 | |
465 zio = zio_root(spa, NULL, NULL, | |
466 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
467 vdev_uberblock_load(zio, rvd, ub); | |
468 error = zio_wait(zio); | |
469 | |
470 /* | |
471 * If we weren't able to find a single valid uberblock, return failure. | |
472 */ | |
473 if (ub->ub_txg == 0) { | |
1760 | 474 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
475 VDEV_AUX_CORRUPT_DATA); | |
1544 | 476 error = ENXIO; |
477 goto out; | |
478 } | |
479 | |
480 /* | |
481 * If the pool is newer than the code, we can't open it. | |
482 */ | |
1760 | 483 if (ub->ub_version > ZFS_VERSION) { |
484 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
485 VDEV_AUX_VERSION_NEWER); | |
1544 | 486 error = ENOTSUP; |
487 goto out; | |
789 | 488 } |
489 | |
490 /* | |
491 * If the vdev guid sum doesn't match the uberblock, we have an | |
492 * incomplete configuration. | |
493 */ | |
1732 | 494 if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { |
1544 | 495 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
496 VDEV_AUX_BAD_GUID_SUM); | |
497 error = ENXIO; | |
498 goto out; | |
789 | 499 } |
500 | |
501 /* | |
502 * Initialize internal SPA structures. | |
503 */ | |
504 spa->spa_state = POOL_STATE_ACTIVE; | |
505 spa->spa_ubsync = spa->spa_uberblock; | |
506 spa->spa_first_txg = spa_last_synced_txg(spa) + 1; | |
1544 | 507 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
508 if (error) { | |
509 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
510 VDEV_AUX_CORRUPT_DATA); | |
511 goto out; | |
512 } | |
789 | 513 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
514 | |
1544 | 515 if (zap_lookup(spa->spa_meta_objset, |
789 | 516 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 517 sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { |
518 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
519 VDEV_AUX_CORRUPT_DATA); | |
520 error = EIO; | |
521 goto out; | |
522 } | |
789 | 523 |
524 if (!mosconfig) { | |
2082 | 525 nvlist_t *newconfig; |
526 | |
527 if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) { | |
1544 | 528 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
529 VDEV_AUX_CORRUPT_DATA); | |
530 error = EIO; | |
531 goto out; | |
532 } | |
789 | 533 |
534 spa_config_set(spa, newconfig); | |
535 spa_unload(spa); | |
536 spa_deactivate(spa); | |
537 spa_activate(spa); | |
538 | |
1544 | 539 return (spa_load(spa, newconfig, state, B_TRUE)); |
540 } | |
541 | |
542 if (zap_lookup(spa->spa_meta_objset, | |
543 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, | |
544 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { | |
545 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
546 VDEV_AUX_CORRUPT_DATA); | |
547 error = EIO; | |
548 goto out; | |
789 | 549 } |
550 | |
1544 | 551 /* |
2082 | 552 * Load the bit that tells us to use the new accounting function |
553 * (raid-z deflation). If we have an older pool, this will not | |
554 * be present. | |
555 */ | |
556 error = zap_lookup(spa->spa_meta_objset, | |
557 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
558 sizeof (uint64_t), 1, &spa->spa_deflate); | |
559 if (error != 0 && error != ENOENT) { | |
560 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
561 VDEV_AUX_CORRUPT_DATA); | |
562 error = EIO; | |
563 goto out; | |
564 } | |
565 | |
566 /* | |
1544 | 567 * Load the persistent error log. If we have an older pool, this will |
568 * not be present. | |
569 */ | |
570 error = zap_lookup(spa->spa_meta_objset, | |
571 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, | |
572 sizeof (uint64_t), 1, &spa->spa_errlog_last); | |
1807
35c8b566d7af
6410711 intent log blocks don't get invited to pool parties
bonwick
parents:
1775
diff
changeset
|
573 if (error != 0 && error != ENOENT) { |
1544 | 574 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
575 VDEV_AUX_CORRUPT_DATA); | |
576 error = EIO; | |
577 goto out; | |
578 } | |
579 | |
580 error = zap_lookup(spa->spa_meta_objset, | |
581 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, | |
582 sizeof (uint64_t), 1, &spa->spa_errlog_scrub); | |
583 if (error != 0 && error != ENOENT) { | |
584 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
585 VDEV_AUX_CORRUPT_DATA); | |
586 error = EIO; | |
587 goto out; | |
588 } | |
789 | 589 |
590 /* | |
2082 | 591 * Load any hot spares for this pool. |
592 */ | |
593 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
594 DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object); | |
595 if (error != 0 && error != ENOENT) { | |
596 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
597 VDEV_AUX_CORRUPT_DATA); | |
598 error = EIO; | |
599 goto out; | |
600 } | |
601 if (error == 0) { | |
602 ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES); | |
603 if (load_nvlist(spa, spa->spa_spares_object, | |
604 &spa->spa_sparelist) != 0) { | |
605 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
606 VDEV_AUX_CORRUPT_DATA); | |
607 error = EIO; | |
608 goto out; | |
609 } | |
610 | |
611 spa_config_enter(spa, RW_WRITER, FTAG); | |
612 spa_load_spares(spa); | |
613 spa_config_exit(spa, FTAG); | |
614 } | |
615 | |
616 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
617 * Load the vdev state for all toplevel vdevs. |
789 | 618 */ |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
619 vdev_load(rvd); |
789 | 620 |
621 /* | |
622 * Propagate the leaf DTLs we just loaded all the way up the tree. | |
623 */ | |
1544 | 624 spa_config_enter(spa, RW_WRITER, FTAG); |
789 | 625 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); |
1544 | 626 spa_config_exit(spa, FTAG); |
789 | 627 |
628 /* | |
629 * Check the state of the root vdev. If it can't be opened, it | |
630 * indicates one or more toplevel vdevs are faulted. | |
631 */ | |
1544 | 632 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
633 error = ENXIO; | |
634 goto out; | |
635 } | |
789 | 636 |
1544 | 637 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
|
638 dmu_tx_t *tx; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
639 int need_update = B_FALSE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
640 int c; |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
641 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
642 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
643 * 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
|
644 * 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
|
645 */ |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
646 tx = dmu_tx_create_assigned(spa_get_dsl(spa), |
789 | 647 spa_first_txg(spa)); |
648 dmu_objset_find(spa->spa_name, zil_claim, tx, 0); | |
649 dmu_tx_commit(tx); | |
650 | |
651 spa->spa_sync_on = B_TRUE; | |
652 txg_sync_start(spa->spa_dsl_pool); | |
653 | |
654 /* | |
655 * Wait for all claims to sync. | |
656 */ | |
657 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
|
658 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
659 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
660 * 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
|
661 * 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
|
662 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
663 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
|
664 state == SPA_LOAD_IMPORT) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
665 need_update = B_TRUE; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
666 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
667 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
|
668 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
|
669 need_update = B_TRUE; |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
670 |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
671 /* |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
672 * 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
|
673 * 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
|
674 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
675 if (need_update) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
676 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
789 | 677 } |
678 | |
1544 | 679 error = 0; |
680 out: | |
2082 | 681 if (error && error != EBADF) |
1544 | 682 zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0); |
683 spa->spa_load_state = SPA_LOAD_NONE; | |
684 spa->spa_ena = 0; | |
685 | |
686 return (error); | |
789 | 687 } |
688 | |
689 /* | |
690 * Pool Open/Import | |
691 * | |
692 * The import case is identical to an open except that the configuration is sent | |
693 * down from userland, instead of grabbed from the configuration cache. For the | |
694 * case of an open, the pool configuration will exist in the | |
695 * POOL_STATE_UNITIALIZED state. | |
696 * | |
697 * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
698 * the same time open the pool, without having to keep around the spa_t in some | |
699 * ambiguous state. | |
700 */ | |
701 static int | |
702 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) | |
703 { | |
704 spa_t *spa; | |
705 int error; | |
706 int loaded = B_FALSE; | |
707 int locked = B_FALSE; | |
708 | |
709 *spapp = NULL; | |
710 | |
711 /* | |
712 * As disgusting as this is, we need to support recursive calls to this | |
713 * function because dsl_dir_open() is called during spa_load(), and ends | |
714 * up calling spa_open() again. The real fix is to figure out how to | |
715 * avoid dsl_dir_open() calling this in the first place. | |
716 */ | |
717 if (mutex_owner(&spa_namespace_lock) != curthread) { | |
718 mutex_enter(&spa_namespace_lock); | |
719 locked = B_TRUE; | |
720 } | |
721 | |
722 if ((spa = spa_lookup(pool)) == NULL) { | |
723 if (locked) | |
724 mutex_exit(&spa_namespace_lock); | |
725 return (ENOENT); | |
726 } | |
727 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
728 | |
729 spa_activate(spa); | |
730 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
731 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); |
789 | 732 |
733 if (error == EBADF) { | |
734 /* | |
1986
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
735 * If vdev_validate() returns failure (indicated by |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
736 * EBADF), it indicates that one of the vdevs indicates |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
737 * that the pool has been exported or destroyed. If |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
738 * this is the case, the config cache is out of sync and |
628267397204
6424405 zpool import destroyed_pool can damage existing pool using same devices
eschrock
parents:
1807
diff
changeset
|
739 * we should remove the pool from the namespace. |
789 | 740 */ |
2082 | 741 zfs_post_ok(spa, NULL); |
789 | 742 spa_unload(spa); |
743 spa_deactivate(spa); | |
744 spa_remove(spa); | |
745 spa_config_sync(); | |
746 if (locked) | |
747 mutex_exit(&spa_namespace_lock); | |
748 return (ENOENT); | |
1544 | 749 } |
750 | |
751 if (error) { | |
789 | 752 /* |
753 * We can't open the pool, but we still have useful | |
754 * information: the state of each vdev after the | |
755 * attempted vdev_open(). Return this to the user. | |
756 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
757 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
|
758 spa_config_enter(spa, RW_READER, FTAG); |
789 | 759 *config = spa_config_generate(spa, NULL, -1ULL, |
760 B_TRUE); | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
761 spa_config_exit(spa, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
762 } |
789 | 763 spa_unload(spa); |
764 spa_deactivate(spa); | |
1544 | 765 spa->spa_last_open_failed = B_TRUE; |
789 | 766 if (locked) |
767 mutex_exit(&spa_namespace_lock); | |
768 *spapp = NULL; | |
769 return (error); | |
1544 | 770 } else { |
771 zfs_post_ok(spa, NULL); | |
772 spa->spa_last_open_failed = B_FALSE; | |
789 | 773 } |
774 | |
775 loaded = B_TRUE; | |
776 } | |
777 | |
778 spa_open_ref(spa, tag); | |
779 if (locked) | |
780 mutex_exit(&spa_namespace_lock); | |
781 | |
782 *spapp = spa; | |
783 | |
784 if (config != NULL) { | |
1544 | 785 spa_config_enter(spa, RW_READER, FTAG); |
789 | 786 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
1544 | 787 spa_config_exit(spa, FTAG); |
789 | 788 } |
789 | |
790 /* | |
791 * If we just loaded the pool, resilver anything that's out of date. | |
792 */ | |
793 if (loaded && (spa_mode & FWRITE)) | |
794 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
795 | |
796 return (0); | |
797 } | |
798 | |
799 int | |
800 spa_open(const char *name, spa_t **spapp, void *tag) | |
801 { | |
802 return (spa_open_common(name, spapp, tag, NULL)); | |
803 } | |
804 | |
1544 | 805 /* |
806 * Lookup the given spa_t, incrementing the inject count in the process, | |
807 * preventing it from being exported or destroyed. | |
808 */ | |
809 spa_t * | |
810 spa_inject_addref(char *name) | |
811 { | |
812 spa_t *spa; | |
813 | |
814 mutex_enter(&spa_namespace_lock); | |
815 if ((spa = spa_lookup(name)) == NULL) { | |
816 mutex_exit(&spa_namespace_lock); | |
817 return (NULL); | |
818 } | |
819 spa->spa_inject_ref++; | |
820 mutex_exit(&spa_namespace_lock); | |
821 | |
822 return (spa); | |
823 } | |
824 | |
825 void | |
826 spa_inject_delref(spa_t *spa) | |
827 { | |
828 mutex_enter(&spa_namespace_lock); | |
829 spa->spa_inject_ref--; | |
830 mutex_exit(&spa_namespace_lock); | |
831 } | |
832 | |
2082 | 833 static void |
834 spa_add_spares(spa_t *spa, nvlist_t *config) | |
835 { | |
836 nvlist_t **spares; | |
837 uint_t i, nspares; | |
838 nvlist_t *nvroot; | |
839 uint64_t guid; | |
840 vdev_stat_t *vs; | |
841 uint_t vsc; | |
842 | |
843 if (spa->spa_nspares == 0) | |
844 return; | |
845 | |
846 VERIFY(nvlist_lookup_nvlist(config, | |
847 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); | |
848 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
849 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
850 if (nspares != 0) { | |
851 VERIFY(nvlist_add_nvlist_array(nvroot, | |
852 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
853 VERIFY(nvlist_lookup_nvlist_array(nvroot, | |
854 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
855 | |
856 /* | |
857 * Go through and find any spares which have since been | |
858 * repurposed as an active spare. If this is the case, update | |
859 * their status appropriately. | |
860 */ | |
861 for (i = 0; i < nspares; i++) { | |
862 VERIFY(nvlist_lookup_uint64(spares[i], | |
863 ZPOOL_CONFIG_GUID, &guid) == 0); | |
864 if (spa_spare_inuse(guid)) { | |
865 VERIFY(nvlist_lookup_uint64_array( | |
866 spares[i], ZPOOL_CONFIG_STATS, | |
867 (uint64_t **)&vs, &vsc) == 0); | |
868 vs->vs_state = VDEV_STATE_CANT_OPEN; | |
869 vs->vs_aux = VDEV_AUX_SPARED; | |
870 } | |
871 } | |
872 } | |
873 } | |
874 | |
789 | 875 int |
1544 | 876 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) |
789 | 877 { |
878 int error; | |
879 spa_t *spa; | |
880 | |
881 *config = NULL; | |
882 error = spa_open_common(name, &spa, FTAG, config); | |
883 | |
2082 | 884 if (spa && *config != NULL) { |
1544 | 885 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, |
886 spa_get_errlog_size(spa)) == 0); | |
887 | |
2082 | 888 spa_add_spares(spa, *config); |
889 } | |
890 | |
1544 | 891 /* |
892 * We want to get the alternate root even for faulted pools, so we cheat | |
893 * and call spa_lookup() directly. | |
894 */ | |
895 if (altroot) { | |
896 if (spa == NULL) { | |
897 mutex_enter(&spa_namespace_lock); | |
898 spa = spa_lookup(name); | |
899 if (spa) | |
900 spa_altroot(spa, altroot, buflen); | |
901 else | |
902 altroot[0] = '\0'; | |
903 spa = NULL; | |
904 mutex_exit(&spa_namespace_lock); | |
905 } else { | |
906 spa_altroot(spa, altroot, buflen); | |
907 } | |
908 } | |
909 | |
789 | 910 if (spa != NULL) |
911 spa_close(spa, FTAG); | |
912 | |
913 return (error); | |
914 } | |
915 | |
916 /* | |
2082 | 917 * Validate that the 'spares' array is well formed. We must have an array of |
918 * nvlists, each which describes a valid leaf vdev. | |
919 */ | |
920 static int | |
921 spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) | |
922 { | |
923 nvlist_t **spares; | |
924 uint_t i, nspares; | |
925 vdev_t *vd; | |
926 int error; | |
927 | |
928 /* | |
929 * It's acceptable to have no spares specified. | |
930 */ | |
931 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
932 &spares, &nspares) != 0) | |
933 return (0); | |
934 | |
935 if (nspares == 0) | |
936 return (EINVAL); | |
937 | |
938 /* | |
939 * Make sure the pool is formatted with a version that supports hot | |
940 * spares. | |
941 */ | |
942 if (spa_version(spa) < ZFS_VERSION_SPARES) | |
943 return (ENOTSUP); | |
944 | |
945 for (i = 0; i < nspares; i++) { | |
946 if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0, | |
947 mode)) != 0) | |
948 return (error); | |
949 | |
950 if (!vd->vdev_ops->vdev_op_leaf) { | |
951 vdev_free(vd); | |
952 return (EINVAL); | |
953 } | |
954 | |
955 if ((error = vdev_open(vd)) != 0) { | |
956 vdev_free(vd); | |
957 return (error); | |
958 } | |
959 | |
960 vd->vdev_top = vd; | |
961 if ((error = vdev_label_spare(vd, crtxg)) != 0) { | |
962 vdev_free(vd); | |
963 return (error); | |
964 } | |
965 | |
966 VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID, | |
967 vd->vdev_guid) == 0); | |
968 | |
969 vdev_free(vd); | |
970 } | |
971 | |
972 return (0); | |
973 } | |
974 | |
975 /* | |
789 | 976 * Pool Creation |
977 */ | |
978 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
979 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot) |
789 | 980 { |
981 spa_t *spa; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
982 vdev_t *rvd; |
789 | 983 dsl_pool_t *dp; |
984 dmu_tx_t *tx; | |
2082 | 985 int c, error = 0; |
789 | 986 uint64_t txg = TXG_INITIAL; |
2082 | 987 nvlist_t **spares; |
988 uint_t nspares; | |
789 | 989 |
990 /* | |
991 * If this pool already exists, return failure. | |
992 */ | |
993 mutex_enter(&spa_namespace_lock); | |
994 if (spa_lookup(pool) != NULL) { | |
995 mutex_exit(&spa_namespace_lock); | |
996 return (EEXIST); | |
997 } | |
998 | |
999 /* | |
1000 * Allocate a new spa_t structure. | |
1001 */ | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1002 spa = spa_add(pool, altroot); |
789 | 1003 spa_activate(spa); |
1004 | |
1005 spa->spa_uberblock.ub_txg = txg - 1; | |
1760 | 1006 spa->spa_uberblock.ub_version = ZFS_VERSION; |
789 | 1007 spa->spa_ubsync = spa->spa_uberblock; |
1008 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1009 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1010 * Create the root vdev. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1011 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1012 spa_config_enter(spa, RW_WRITER, FTAG); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1013 |
2082 | 1014 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); |
1015 | |
1016 ASSERT(error != 0 || rvd != NULL); | |
1017 ASSERT(error != 0 || spa->spa_root_vdev == rvd); | |
1018 | |
1019 if (error == 0 && rvd->vdev_children == 0) | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1020 error = EINVAL; |
2082 | 1021 |
1022 if (error == 0 && | |
1023 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && | |
1024 (error = spa_validate_spares(spa, nvroot, txg, | |
1025 VDEV_ALLOC_ADD)) == 0) { | |
1026 for (c = 0; c < rvd->vdev_children; c++) | |
1027 vdev_init(rvd->vdev_child[c], txg); | |
1028 vdev_config_dirty(rvd); | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1029 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1030 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1031 spa_config_exit(spa, FTAG); |
789 | 1032 |
2082 | 1033 if (error != 0) { |
789 | 1034 spa_unload(spa); |
1035 spa_deactivate(spa); | |
1036 spa_remove(spa); | |
1037 mutex_exit(&spa_namespace_lock); | |
1038 return (error); | |
1039 } | |
1040 | |
2082 | 1041 /* |
1042 * Get the list of spares, if specified. | |
1043 */ | |
1044 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
1045 &spares, &nspares) == 0) { | |
1046 VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME, | |
1047 KM_SLEEP) == 0); | |
1048 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1049 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1050 spa_config_enter(spa, RW_WRITER, FTAG); | |
1051 spa_load_spares(spa); | |
1052 spa_config_exit(spa, FTAG); | |
1053 spa->spa_sync_spares = B_TRUE; | |
1054 } | |
1055 | |
789 | 1056 spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg); |
1057 spa->spa_meta_objset = dp->dp_meta_objset; | |
1058 | |
1059 tx = dmu_tx_create_assigned(dp, txg); | |
1060 | |
1061 /* | |
1062 * Create the pool config object. | |
1063 */ | |
1064 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
1065 DMU_OT_PACKED_NVLIST, 1 << 14, | |
1066 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
1067 | |
1544 | 1068 if (zap_add(spa->spa_meta_objset, |
789 | 1069 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
1544 | 1070 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { |
1071 cmn_err(CE_PANIC, "failed to add pool config"); | |
1072 } | |
789 | 1073 |
2082 | 1074 /* Newly created pools are always deflated. */ |
1075 spa->spa_deflate = TRUE; | |
1076 if (zap_add(spa->spa_meta_objset, | |
1077 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
1078 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { | |
1079 cmn_err(CE_PANIC, "failed to add deflate"); | |
1080 } | |
1081 | |
789 | 1082 /* |
1083 * Create the deferred-free bplist object. Turn off compression | |
1084 * because sync-to-convergence takes longer if the blocksize | |
1085 * keeps changing. | |
1086 */ | |
1087 spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, | |
1088 1 << 14, tx); | |
1089 dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, | |
1090 ZIO_COMPRESS_OFF, tx); | |
1091 | |
1544 | 1092 if (zap_add(spa->spa_meta_objset, |
789 | 1093 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, |
1544 | 1094 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { |
1095 cmn_err(CE_PANIC, "failed to add bplist"); | |
1096 } | |
789 | 1097 |
1098 dmu_tx_commit(tx); | |
1099 | |
1100 spa->spa_sync_on = B_TRUE; | |
1101 txg_sync_start(spa->spa_dsl_pool); | |
1102 | |
1103 /* | |
1104 * We explicitly wait for the first transaction to complete so that our | |
1105 * bean counters are appropriately updated. | |
1106 */ | |
1107 txg_wait_synced(spa->spa_dsl_pool, txg); | |
1108 | |
1109 spa_config_sync(); | |
1110 | |
1111 mutex_exit(&spa_namespace_lock); | |
1112 | |
1113 return (0); | |
1114 } | |
1115 | |
1116 /* | |
1117 * Import the given pool into the system. We set up the necessary spa_t and | |
1118 * then call spa_load() to do the dirty work. | |
1119 */ | |
1120 int | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1121 spa_import(const char *pool, nvlist_t *config, const char *altroot) |
789 | 1122 { |
1123 spa_t *spa; | |
1124 int error; | |
2082 | 1125 nvlist_t *nvroot; |
1126 nvlist_t **spares; | |
1127 uint_t nspares; | |
789 | 1128 |
1129 if (!(spa_mode & FWRITE)) | |
1130 return (EROFS); | |
1131 | |
1132 /* | |
1133 * If a pool with this name exists, return failure. | |
1134 */ | |
1135 mutex_enter(&spa_namespace_lock); | |
1136 if (spa_lookup(pool) != NULL) { | |
1137 mutex_exit(&spa_namespace_lock); | |
1138 return (EEXIST); | |
1139 } | |
1140 | |
1141 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1142 * Create and initialize the spa structure. |
789 | 1143 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1144 spa = spa_add(pool, altroot); |
789 | 1145 spa_activate(spa); |
1146 | |
1147 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1148 * Pass off the heavy lifting to spa_load(). |
1732 | 1149 * Pass TRUE for mosconfig because the user-supplied config |
1150 * 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
|
1151 */ |
1732 | 1152 error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE); |
789 | 1153 |
2082 | 1154 spa_config_enter(spa, RW_WRITER, FTAG); |
1155 /* | |
1156 * Toss any existing sparelist, as it doesn't have any validity anymore, | |
1157 * and conflicts with spa_has_spare(). | |
1158 */ | |
1159 if (spa->spa_sparelist) { | |
1160 nvlist_free(spa->spa_sparelist); | |
1161 spa->spa_sparelist = NULL; | |
1162 spa_load_spares(spa); | |
1163 } | |
1164 | |
1165 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, | |
1166 &nvroot) == 0); | |
1167 if (error == 0) | |
1168 error = spa_validate_spares(spa, nvroot, -1ULL, | |
1169 VDEV_ALLOC_SPARE); | |
1170 spa_config_exit(spa, FTAG); | |
1171 | |
1172 if (error != 0) { | |
789 | 1173 spa_unload(spa); |
1174 spa_deactivate(spa); | |
1175 spa_remove(spa); | |
1176 mutex_exit(&spa_namespace_lock); | |
1177 return (error); | |
1178 } | |
1179 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1180 /* |
2082 | 1181 * Override any spares as specified by the user, as these may have |
1182 * correct device names/devids, etc. | |
1183 */ | |
1184 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
1185 &spares, &nspares) == 0) { | |
1186 if (spa->spa_sparelist) | |
1187 VERIFY(nvlist_remove(spa->spa_sparelist, | |
1188 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
1189 else | |
1190 VERIFY(nvlist_alloc(&spa->spa_sparelist, | |
1191 NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1192 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1193 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1194 spa_config_enter(spa, RW_WRITER, FTAG); | |
1195 spa_load_spares(spa); | |
1196 spa_config_exit(spa, FTAG); | |
1197 spa->spa_sync_spares = B_TRUE; | |
1198 } | |
1199 | |
1200 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1201 * 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
|
1202 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1203 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
|
1204 |
789 | 1205 mutex_exit(&spa_namespace_lock); |
1206 | |
1207 /* | |
1208 * Resilver anything that's out of date. | |
1209 */ | |
1210 if (spa_mode & FWRITE) | |
1211 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1212 | |
1213 return (0); | |
1214 } | |
1215 | |
1216 /* | |
1217 * This (illegal) pool name is used when temporarily importing a spa_t in order | |
1218 * to get the vdev stats associated with the imported devices. | |
1219 */ | |
1220 #define TRYIMPORT_NAME "$import" | |
1221 | |
1222 nvlist_t * | |
1223 spa_tryimport(nvlist_t *tryconfig) | |
1224 { | |
1225 nvlist_t *config = NULL; | |
1226 char *poolname; | |
1227 spa_t *spa; | |
1228 uint64_t state; | |
1229 | |
1230 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
1231 return (NULL); | |
1232 | |
1233 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
1234 return (NULL); | |
1235 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1236 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1237 * Create and initialize the spa structure. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1238 */ |
789 | 1239 mutex_enter(&spa_namespace_lock); |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1240 spa = spa_add(TRYIMPORT_NAME, NULL); |
789 | 1241 spa_activate(spa); |
1242 | |
1243 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1244 * Pass off the heavy lifting to spa_load(). |
1732 | 1245 * Pass TRUE for mosconfig because the user-supplied config |
1246 * is actually the one to trust when doing an import. | |
789 | 1247 */ |
1732 | 1248 (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); |
789 | 1249 |
1250 /* | |
1251 * If 'tryconfig' was at least parsable, return the current config. | |
1252 */ | |
1253 if (spa->spa_root_vdev != NULL) { | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1254 spa_config_enter(spa, RW_READER, FTAG); |
789 | 1255 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
|
1256 spa_config_exit(spa, FTAG); |
789 | 1257 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, |
1258 poolname) == 0); | |
1259 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
1260 state) == 0); | |
2082 | 1261 |
1262 /* | |
1263 * Add the list of hot spares. | |
1264 */ | |
1265 spa_add_spares(spa, config); | |
789 | 1266 } |
1267 | |
1268 spa_unload(spa); | |
1269 spa_deactivate(spa); | |
1270 spa_remove(spa); | |
1271 mutex_exit(&spa_namespace_lock); | |
1272 | |
1273 return (config); | |
1274 } | |
1275 | |
1276 /* | |
1277 * Pool export/destroy | |
1278 * | |
1279 * The act of destroying or exporting a pool is very simple. We make sure there | |
1280 * is no more pending I/O and any references to the pool are gone. Then, we | |
1281 * update the pool state and sync all the labels to disk, removing the | |
1282 * configuration from the cache afterwards. | |
1283 */ | |
1284 static int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1285 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig) |
789 | 1286 { |
1287 spa_t *spa; | |
1288 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1289 if (oldconfig) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1290 *oldconfig = NULL; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1291 |
789 | 1292 if (!(spa_mode & FWRITE)) |
1293 return (EROFS); | |
1294 | |
1295 mutex_enter(&spa_namespace_lock); | |
1296 if ((spa = spa_lookup(pool)) == NULL) { | |
1297 mutex_exit(&spa_namespace_lock); | |
1298 return (ENOENT); | |
1299 } | |
1300 | |
1301 /* | |
1544 | 1302 * Put a hold on the pool, drop the namespace lock, stop async tasks, |
1303 * reacquire the namespace lock, and see if we can export. | |
1304 */ | |
1305 spa_open_ref(spa, FTAG); | |
1306 mutex_exit(&spa_namespace_lock); | |
1307 spa_async_suspend(spa); | |
1308 mutex_enter(&spa_namespace_lock); | |
1309 spa_close(spa, FTAG); | |
1310 | |
1311 /* | |
789 | 1312 * The pool will be in core if it's openable, |
1313 * in which case we can modify its state. | |
1314 */ | |
1315 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { | |
1316 /* | |
1317 * Objsets may be open only because they're dirty, so we | |
1318 * have to force it to sync before checking spa_refcnt. | |
1319 */ | |
1320 spa_scrub_suspend(spa); | |
1321 txg_wait_synced(spa->spa_dsl_pool, 0); | |
1322 | |
1544 | 1323 /* |
1324 * A pool cannot be exported or destroyed if there are active | |
1325 * references. If we are resetting a pool, allow references by | |
1326 * fault injection handlers. | |
1327 */ | |
1328 if (!spa_refcount_zero(spa) || | |
1329 (spa->spa_inject_ref != 0 && | |
1330 new_state != POOL_STATE_UNINITIALIZED)) { | |
789 | 1331 spa_scrub_resume(spa); |
1544 | 1332 spa_async_resume(spa); |
789 | 1333 mutex_exit(&spa_namespace_lock); |
1334 return (EBUSY); | |
1335 } | |
1336 | |
1337 spa_scrub_resume(spa); | |
1338 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); | |
1339 | |
1340 /* | |
1341 * We want this to be reflected on every label, | |
1342 * so mark them all dirty. spa_unload() will do the | |
1343 * final sync that pushes these changes out. | |
1344 */ | |
1544 | 1345 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
|
1346 spa_config_enter(spa, RW_WRITER, FTAG); |
1544 | 1347 spa->spa_state = new_state; |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1348 spa->spa_final_txg = spa_last_synced_txg(spa) + 1; |
1544 | 1349 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
|
1350 spa_config_exit(spa, FTAG); |
1544 | 1351 } |
789 | 1352 } |
1353 | |
1354 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
1355 spa_unload(spa); | |
1356 spa_deactivate(spa); | |
1357 } | |
1358 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1359 if (oldconfig && spa->spa_config) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1360 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
|
1361 |
1544 | 1362 if (new_state != POOL_STATE_UNINITIALIZED) { |
1363 spa_remove(spa); | |
1364 spa_config_sync(); | |
1365 } | |
789 | 1366 mutex_exit(&spa_namespace_lock); |
1367 | |
1368 return (0); | |
1369 } | |
1370 | |
1371 /* | |
1372 * Destroy a storage pool. | |
1373 */ | |
1374 int | |
1375 spa_destroy(char *pool) | |
1376 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1377 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL)); |
789 | 1378 } |
1379 | |
1380 /* | |
1381 * Export a storage pool. | |
1382 */ | |
1383 int | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1384 spa_export(char *pool, nvlist_t **oldconfig) |
789 | 1385 { |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1386 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig)); |
789 | 1387 } |
1388 | |
1389 /* | |
1544 | 1390 * Similar to spa_export(), this unloads the spa_t without actually removing it |
1391 * from the namespace in any way. | |
1392 */ | |
1393 int | |
1394 spa_reset(char *pool) | |
1395 { | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
1396 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL)); |
1544 | 1397 } |
1398 | |
1399 | |
1400 /* | |
789 | 1401 * ========================================================================== |
1402 * Device manipulation | |
1403 * ========================================================================== | |
1404 */ | |
1405 | |
1406 /* | |
1407 * Add capacity to a storage pool. | |
1408 */ | |
1409 int | |
1410 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
1411 { | |
1412 uint64_t txg; | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1413 int c, error; |
789 | 1414 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
|
1415 vdev_t *vd, *tvd; |
2082 | 1416 nvlist_t **spares; |
1417 uint_t i, nspares; | |
789 | 1418 |
1419 txg = spa_vdev_enter(spa); | |
1420 | |
2082 | 1421 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, |
1422 VDEV_ALLOC_ADD)) != 0) | |
1423 return (spa_vdev_exit(spa, NULL, txg, error)); | |
1424 | |
1425 if ((error = spa_validate_spares(spa, nvroot, txg, | |
1426 VDEV_ALLOC_ADD)) != 0) | |
789 | 1427 return (spa_vdev_exit(spa, vd, txg, error)); |
1428 | |
2082 | 1429 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, |
1430 &spares, &nspares) != 0) | |
1431 nspares = 0; | |
1432 | |
1433 if (vd->vdev_children == 0 && nspares == 0) | |
1434 return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
1435 | |
1436 if (vd->vdev_children != 0) { | |
1437 if ((error = vdev_create(vd, txg, B_FALSE)) != 0) | |
1438 return (spa_vdev_exit(spa, vd, txg, error)); | |
1439 | |
1440 /* | |
1441 * Transfer each new top-level vdev from vd to rvd. | |
1442 */ | |
1443 for (c = 0; c < vd->vdev_children; c++) { | |
1444 tvd = vd->vdev_child[c]; | |
1445 vdev_remove_child(vd, tvd); | |
1446 tvd->vdev_id = rvd->vdev_children; | |
1447 vdev_add_child(rvd, tvd); | |
1448 vdev_config_dirty(tvd); | |
1449 } | |
1450 } | |
1451 | |
1452 if (nspares != 0) { | |
1453 if (spa->spa_sparelist != NULL) { | |
1454 nvlist_t **oldspares; | |
1455 uint_t oldnspares; | |
1456 nvlist_t **newspares; | |
1457 | |
1458 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
1459 ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0); | |
1460 | |
1461 newspares = kmem_alloc(sizeof (void *) * | |
1462 (nspares + oldnspares), KM_SLEEP); | |
1463 for (i = 0; i < oldnspares; i++) | |
1464 VERIFY(nvlist_dup(oldspares[i], | |
1465 &newspares[i], KM_SLEEP) == 0); | |
1466 for (i = 0; i < nspares; i++) | |
1467 VERIFY(nvlist_dup(spares[i], | |
1468 &newspares[i + oldnspares], | |
1469 KM_SLEEP) == 0); | |
1470 | |
1471 VERIFY(nvlist_remove(spa->spa_sparelist, | |
1472 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); | |
1473 | |
1474 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1475 ZPOOL_CONFIG_SPARES, newspares, | |
1476 nspares + oldnspares) == 0); | |
1477 for (i = 0; i < oldnspares + nspares; i++) | |
1478 nvlist_free(newspares[i]); | |
1479 kmem_free(newspares, (oldnspares + nspares) * | |
1480 sizeof (void *)); | |
1481 } else { | |
1482 VERIFY(nvlist_alloc(&spa->spa_sparelist, | |
1483 NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
1484 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, | |
1485 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); | |
1486 } | |
1487 | |
1488 spa_load_spares(spa); | |
1489 spa->spa_sync_spares = B_TRUE; | |
789 | 1490 } |
1491 | |
1492 /* | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1493 * 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
|
1494 * 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
|
1495 * 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
|
1496 * 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
|
1497 * 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
|
1498 * 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
|
1499 * 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
|
1500 * |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1501 * 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
|
1502 * 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
|
1503 * steps will be completed the next time we load the pool. |
789 | 1504 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1505 (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
|
1506 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1507 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1508 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
|
1509 mutex_exit(&spa_namespace_lock); |
789 | 1510 |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1511 return (0); |
789 | 1512 } |
1513 | |
1514 /* | |
1515 * Attach a device to a mirror. The arguments are the path to any device | |
1516 * in the mirror, and the nvroot for the new device. If the path specifies | |
1517 * a device that is not mirrored, we automatically insert the mirror vdev. | |
1518 * | |
1519 * If 'replacing' is specified, the new device is intended to replace the | |
1520 * existing device; in this case the two devices are made into their own | |
1521 * mirror using the 'replacing' vdev, which is functionally idendical to | |
1522 * the mirror vdev (it actually reuses all the same ops) but has a few | |
1523 * extra rules: you can't attach to it after it's been created, and upon | |
1524 * completion of resilvering, the first disk (the one being replaced) | |
1525 * is automatically detached. | |
1526 */ | |
1527 int | |
1544 | 1528 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) |
789 | 1529 { |
1530 uint64_t txg, open_txg; | |
1531 int error; | |
1532 vdev_t *rvd = spa->spa_root_vdev; | |
1533 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
2082 | 1534 vdev_ops_t *pvops; |
789 | 1535 |
1536 txg = spa_vdev_enter(spa); | |
1537 | |
1544 | 1538 oldvd = vdev_lookup_by_guid(rvd, guid); |
789 | 1539 |
1540 if (oldvd == NULL) | |
1541 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1542 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1543 if (!oldvd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1544 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
|
1545 |
789 | 1546 pvd = oldvd->vdev_parent; |
1547 | |
2082 | 1548 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, |
1549 VDEV_ALLOC_ADD)) != 0 || newrootvd->vdev_children != 1) | |
789 | 1550 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); |
1551 | |
1552 newvd = newrootvd->vdev_child[0]; | |
1553 | |
1554 if (!newvd->vdev_ops->vdev_op_leaf) | |
1555 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
1556 | |
2082 | 1557 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) |
789 | 1558 return (spa_vdev_exit(spa, newrootvd, txg, error)); |
1559 | |
2082 | 1560 if (!replacing) { |
1561 /* | |
1562 * For attach, the only allowable parent is a mirror or the root | |
1563 * vdev. | |
1564 */ | |
1565 if (pvd->vdev_ops != &vdev_mirror_ops && | |
1566 pvd->vdev_ops != &vdev_root_ops) | |
1567 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1568 | |
1569 pvops = &vdev_mirror_ops; | |
1570 } else { | |
1571 /* | |
1572 * Active hot spares can only be replaced by inactive hot | |
1573 * spares. | |
1574 */ | |
1575 if (pvd->vdev_ops == &vdev_spare_ops && | |
1576 pvd->vdev_child[1] == oldvd && | |
1577 !spa_has_spare(spa, newvd->vdev_guid)) | |
1578 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1579 | |
1580 /* | |
1581 * If the source is a hot spare, and the parent isn't already a | |
1582 * spare, then we want to create a new hot spare. Otherwise, we | |
1583 * want to create a replacing vdev. | |
1584 */ | |
1585 if (pvd->vdev_ops == &vdev_replacing_ops) | |
1586 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); | |
1587 else if (pvd->vdev_ops != &vdev_spare_ops && | |
1588 newvd->vdev_isspare) | |
1589 pvops = &vdev_spare_ops; | |
1590 else | |
1591 pvops = &vdev_replacing_ops; | |
1592 } | |
1593 | |
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
|
1594 /* |
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
|
1595 * 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
|
1596 * 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
|
1597 */ |
759d20c7e57b
6366265 attach/replace should allow a new device size at least the min of all devs in a mirror/raidz
lling
parents:
797
diff
changeset
|
1598 if (newvd->vdev_psize < vdev_get_rsize(oldvd)) |
789 | 1599 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
1600 | |
1732 | 1601 /* |
1602 * The new device cannot have a higher alignment requirement | |
1603 * than the top-level vdev. | |
1604 */ | |
1605 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
789 | 1606 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); |
1607 | |
1608 /* | |
1609 * If this is an in-place replacement, update oldvd's path and devid | |
1610 * to make it distinguishable from newvd, and unopenable from now on. | |
1611 */ | |
1612 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
1613 spa_strfree(oldvd->vdev_path); | |
1614 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
1615 KM_SLEEP); | |
1616 (void) sprintf(oldvd->vdev_path, "%s/%s", | |
1617 newvd->vdev_path, "old"); | |
1618 if (oldvd->vdev_devid != NULL) { | |
1619 spa_strfree(oldvd->vdev_devid); | |
1620 oldvd->vdev_devid = NULL; | |
1621 } | |
1622 } | |
1623 | |
1624 /* | |
2082 | 1625 * If the parent is not a mirror, or if we're replacing, insert the new |
1626 * mirror/replacing/spare vdev above oldvd. | |
789 | 1627 */ |
1628 if (pvd->vdev_ops != pvops) | |
1629 pvd = vdev_add_parent(oldvd, pvops); | |
1630 | |
1631 ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
1632 ASSERT(pvd->vdev_ops == pvops); | |
1633 ASSERT(oldvd->vdev_parent == pvd); | |
1634 | |
1635 /* | |
1636 * Extract the new device from its root and add it to pvd. | |
1637 */ | |
1638 vdev_remove_child(newrootvd, newvd); | |
1639 newvd->vdev_id = pvd->vdev_children; | |
1640 vdev_add_child(pvd, newvd); | |
1641 | |
1544 | 1642 /* |
1643 * If newvd is smaller than oldvd, but larger than its rsize, | |
1644 * the addition of newvd may have decreased our parent's asize. | |
1645 */ | |
1646 pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); | |
1647 | |
789 | 1648 tvd = newvd->vdev_top; |
1649 ASSERT(pvd->vdev_top == tvd); | |
1650 ASSERT(tvd->vdev_parent == rvd); | |
1651 | |
1652 vdev_config_dirty(tvd); | |
1653 | |
1654 /* | |
1655 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate | |
1656 * upward when spa_vdev_exit() calls vdev_dtl_reassess(). | |
1657 */ | |
1658 open_txg = txg + TXG_CONCURRENT_STATES - 1; | |
1659 | |
1660 mutex_enter(&newvd->vdev_dtl_lock); | |
1661 space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, | |
1662 open_txg - TXG_INITIAL + 1); | |
1663 mutex_exit(&newvd->vdev_dtl_lock); | |
1664 | |
1544 | 1665 dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg); |
1666 | |
789 | 1667 /* |
1668 * Mark newvd's DTL dirty in this txg. | |
1669 */ | |
1732 | 1670 vdev_dirty(tvd, VDD_DTL, newvd, txg); |
789 | 1671 |
1672 (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); | |
1673 | |
1674 /* | |
1675 * Kick off a resilver to update newvd. | |
1676 */ | |
1677 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
1678 | |
1679 return (0); | |
1680 } | |
1681 | |
1682 /* | |
1683 * Detach a device from a mirror or replacing vdev. | |
1684 * If 'replace_done' is specified, only detach if the parent | |
1685 * is a replacing vdev. | |
1686 */ | |
1687 int | |
1544 | 1688 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) |
789 | 1689 { |
1690 uint64_t txg; | |
1691 int c, t, error; | |
1692 vdev_t *rvd = spa->spa_root_vdev; | |
1693 vdev_t *vd, *pvd, *cvd, *tvd; | |
2082 | 1694 boolean_t unspare = B_FALSE; |
1695 uint64_t unspare_guid; | |
789 | 1696 |
1697 txg = spa_vdev_enter(spa); | |
1698 | |
1544 | 1699 vd = vdev_lookup_by_guid(rvd, guid); |
789 | 1700 |
1701 if (vd == NULL) | |
1702 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
1703 | |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1704 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1705 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
|
1706 |
789 | 1707 pvd = vd->vdev_parent; |
1708 | |
1709 /* | |
1710 * If replace_done is specified, only remove this device if it's | |
2082 | 1711 * the first child of a replacing vdev. For the 'spare' vdev, either |
1712 * disk can be removed. | |
789 | 1713 */ |
2082 | 1714 if (replace_done) { |
1715 if (pvd->vdev_ops == &vdev_replacing_ops) { | |
1716 if (vd->vdev_id != 0) | |
1717 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1718 } else if (pvd->vdev_ops != &vdev_spare_ops) { | |
1719 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
1720 } | |
1721 } | |
1722 | |
1723 ASSERT(pvd->vdev_ops != &vdev_spare_ops || | |
1724 spa_version(spa) >= ZFS_VERSION_SPARES); | |
789 | 1725 |
1726 /* | |
2082 | 1727 * Only mirror, replacing, and spare vdevs support detach. |
789 | 1728 */ |
1729 if (pvd->vdev_ops != &vdev_replacing_ops && | |
2082 | 1730 pvd->vdev_ops != &vdev_mirror_ops && |
1731 pvd->vdev_ops != &vdev_spare_ops) | |
789 | 1732 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); |
1733 | |
1734 /* | |
1735 * If there's only one replica, you can't detach it. | |
1736 */ | |
1737 if (pvd->vdev_children <= 1) | |
1738 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
1739 | |
1740 /* | |
1741 * If all siblings have non-empty DTLs, this device may have the only | |
1742 * valid copy of the data, which means we cannot safely detach it. | |
1743 * | |
1744 * XXX -- as in the vdev_offline() case, we really want a more | |
1745 * precise DTL check. | |
1746 */ | |
1747 for (c = 0; c < pvd->vdev_children; c++) { | |
1748 uint64_t dirty; | |
1749 | |
1750 cvd = pvd->vdev_child[c]; | |
1751 if (cvd == vd) | |
1752 continue; | |
1753 if (vdev_is_dead(cvd)) | |
1754 continue; | |
1755 mutex_enter(&cvd->vdev_dtl_lock); | |
1756 dirty = cvd->vdev_dtl_map.sm_space | | |
1757 cvd->vdev_dtl_scrub.sm_space; | |
1758 mutex_exit(&cvd->vdev_dtl_lock); | |
1759 if (!dirty) | |
1760 break; | |
1761 } | |
2082 | 1762 |
1763 /* | |
1764 * If we are a replacing or spare vdev, then we can always detach the | |
1765 * latter child, as that is how one cancels the operation. | |
1766 */ | |
1767 if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) && | |
1768 c == pvd->vdev_children) | |
789 | 1769 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); |
1770 | |
1771 /* | |
2082 | 1772 * If we are detaching the original disk from a spare, then it implies |
1773 * that the spare should become a real disk, and be removed from the | |
1774 * active spare list for the pool. | |
1775 */ | |
1776 if (pvd->vdev_ops == &vdev_spare_ops && | |
1777 vd->vdev_id == 0) | |
1778 unspare = B_TRUE; | |
1779 | |
1780 /* | |
789 | 1781 * Erase the disk labels so the disk can be used for other things. |
1782 * This must be done after all other error cases are handled, | |
1783 * but before we disembowel vd (so we can still do I/O to it). | |
1784 * But if we can't do it, don't treat the error as fatal -- | |
1785 * it may be that the unwritability of the disk is the reason | |
1786 * it's being detached! | |
1787 */ | |
2082 | 1788 error = vdev_label_init(vd, 0, B_FALSE); |
789 | 1789 if (error) |
1790 dprintf("unable to erase labels on %s\n", vdev_description(vd)); | |
1791 | |
1792 /* | |
1793 * Remove vd from its parent and compact the parent's children. | |
1794 */ | |
1795 vdev_remove_child(pvd, vd); | |
1796 vdev_compact_children(pvd); | |
1797 | |
1798 /* | |
1799 * Remember one of the remaining children so we can get tvd below. | |
1800 */ | |
1801 cvd = pvd->vdev_child[0]; | |
1802 | |
1803 /* | |
2082 | 1804 * If we need to remove the remaining child from the list of hot spares, |
1805 * do it now, marking the vdev as no longer a spare in the process. We | |
1806 * must do this before vdev_remove_parent(), because that can change the | |
1807 * GUID if it creates a new toplevel GUID. | |
1808 */ | |
1809 if (unspare) { | |
1810 ASSERT(cvd->vdev_isspare); | |
1811 spa_spare_remove(cvd->vdev_guid); | |
1812 cvd->vdev_isspare = B_FALSE; | |
1813 unspare_guid = cvd->vdev_guid; | |
1814 } | |
1815 | |
1816 /* | |
789 | 1817 * If the parent mirror/replacing vdev only has one child, |
1818 * the parent is no longer needed. Remove it from the tree. | |
1819 */ | |
1820 if (pvd->vdev_children == 1) | |
1821 vdev_remove_parent(cvd); | |
1822 | |
1823 /* | |
1824 * We don't set tvd until now because the parent we just removed | |
1825 * may have been the previous top-level vdev. | |
1826 */ | |
1827 tvd = cvd->vdev_top; | |
1828 ASSERT(tvd->vdev_parent == rvd); | |
1829 | |
1830 /* | |
1831 * Reopen this top-level vdev to reassess health after detach. | |
1832 */ | |
1544 | 1833 vdev_reopen(tvd); |
789 | 1834 |
1835 /* | |
1836 * If the device we just detached was smaller than the others, | |
1732 | 1837 * it may be possible to add metaslabs (i.e. grow the pool). |
1838 * vdev_metaslab_init() can't fail because the existing metaslabs | |
1839 * are already in core, so there's nothing to read from disk. | |
789 | 1840 */ |
1732 | 1841 VERIFY(vdev_metaslab_init(tvd, txg) == 0); |
789 | 1842 |
1843 vdev_config_dirty(tvd); | |
1844 | |
1845 /* | |
1846 * Mark vd's DTL as dirty in this txg. | |
1847 * vdev_dtl_sync() will see that vd->vdev_detached is set | |
1848 * and free vd's DTL object in syncing context. | |
1849 * But first make sure we're not on any *other* txg's DTL list, | |
1850 * to prevent vd from being accessed after it's freed. | |
1851 */ | |
1852 for (t = 0; t < TXG_SIZE; t++) | |
1853 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
1732 | 1854 vd->vdev_detached = B_TRUE; |
1855 vdev_dirty(tvd, VDD_DTL, vd, txg); | |
789 | 1856 |
1544 | 1857 dprintf("detached %s in txg %llu\n", vd->vdev_path, txg); |
789 | 1858 |
2082 | 1859 error = spa_vdev_exit(spa, vd, txg, 0); |
1860 | |
1861 /* | |
1862 * If we are supposed to remove the given vdev from the list of spares, | |
1863 * iterate over all pools in the system and replace it if it's present. | |
1864 */ | |
1865 if (unspare) { | |
1866 spa = NULL; | |
1867 mutex_enter(&spa_namespace_lock); | |
1868 while ((spa = spa_next(spa)) != NULL) { | |
1869 if (spa->spa_state != POOL_STATE_ACTIVE) | |
1870 continue; | |
1871 | |
1872 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); | |
1873 } | |
1874 mutex_exit(&spa_namespace_lock); | |
1875 } | |
1876 | |
1877 return (error); | |
1878 } | |
1879 | |
1880 /* | |
1881 * Remove a device from the pool. Currently, this supports removing only hot | |
1882 * spares. | |
1883 */ | |
1884 int | |
1885 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) | |
1886 { | |
1887 vdev_t *vd; | |
1888 nvlist_t **spares, *nv, **newspares; | |
1889 uint_t i, j, nspares; | |
1890 int ret = 0; | |
1891 | |
1892 spa_config_enter(spa, RW_WRITER, FTAG); | |
1893 | |
1894 vd = spa_lookup_by_guid(spa, guid); | |
1895 | |
1896 nv = NULL; | |
1897 if (spa->spa_spares != NULL && | |
1898 nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1899 &spares, &nspares) == 0) { | |
1900 for (i = 0; i < nspares; i++) { | |
1901 uint64_t theguid; | |
1902 | |
1903 VERIFY(nvlist_lookup_uint64(spares[i], | |
1904 ZPOOL_CONFIG_GUID, &theguid) == 0); | |
1905 if (theguid == guid) { | |
1906 nv = spares[i]; | |
1907 break; | |
1908 } | |
1909 } | |
1910 } | |
1911 | |
1912 /* | |
1913 * We only support removing a hot spare, and only if it's not currently | |
1914 * in use in this pool. | |
1915 */ | |
1916 if (nv == NULL && vd == NULL) { | |
1917 ret = ENOENT; | |
1918 goto out; | |
1919 } | |
1920 | |
1921 if (nv == NULL && vd != NULL) { | |
1922 ret = ENOTSUP; | |
1923 goto out; | |
1924 } | |
1925 | |
1926 if (!unspare && nv != NULL && vd != NULL) { | |
1927 ret = EBUSY; | |
1928 goto out; | |
1929 } | |
1930 | |
1931 if (nspares == 1) { | |
1932 newspares = NULL; | |
1933 } else { | |
1934 newspares = kmem_alloc((nspares - 1) * sizeof (void *), | |
1935 KM_SLEEP); | |
1936 for (i = 0, j = 0; i < nspares; i++) { | |
1937 if (spares[i] != nv) | |
1938 VERIFY(nvlist_dup(spares[i], | |
1939 &newspares[j++], KM_SLEEP) == 0); | |
1940 } | |
1941 } | |
1942 | |
1943 VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1944 DATA_TYPE_NVLIST_ARRAY) == 0); | |
1945 VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES, | |
1946 newspares, nspares - 1) == 0); | |
1947 for (i = 0; i < nspares - 1; i++) | |
1948 nvlist_free(newspares[i]); | |
1949 kmem_free(newspares, (nspares - 1) * sizeof (void *)); | |
1950 spa_load_spares(spa); | |
1951 spa->spa_sync_spares = B_TRUE; | |
1952 | |
1953 out: | |
1954 spa_config_exit(spa, FTAG); | |
1955 | |
1956 return (ret); | |
789 | 1957 } |
1958 | |
1959 /* | |
1544 | 1960 * Find any device that's done replacing, so we can detach it. |
789 | 1961 */ |
1544 | 1962 static vdev_t * |
1963 spa_vdev_replace_done_hunt(vdev_t *vd) | |
789 | 1964 { |
1544 | 1965 vdev_t *newvd, *oldvd; |
789 | 1966 int c; |
1967 | |
1544 | 1968 for (c = 0; c < vd->vdev_children; c++) { |
1969 oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]); | |
1970 if (oldvd != NULL) | |
1971 return (oldvd); | |
1972 } | |
789 | 1973 |
1974 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { | |
1544 | 1975 oldvd = vd->vdev_child[0]; |
1976 newvd = vd->vdev_child[1]; | |
789 | 1977 |
1544 | 1978 mutex_enter(&newvd->vdev_dtl_lock); |
1979 if (newvd->vdev_dtl_map.sm_space == 0 && | |
1980 newvd->vdev_dtl_scrub.sm_space == 0) { | |
1981 mutex_exit(&newvd->vdev_dtl_lock); | |
1982 return (oldvd); | |
1983 } | |
1984 mutex_exit(&newvd->vdev_dtl_lock); | |
1985 } | |
789 | 1986 |
1544 | 1987 return (NULL); |
789 | 1988 } |
1989 | |
1544 | 1990 static void |
789 | 1991 spa_vdev_replace_done(spa_t *spa) |
1992 { | |
1544 | 1993 vdev_t *vd; |
2082 | 1994 vdev_t *pvd; |
1544 | 1995 uint64_t guid; |
2082 | 1996 uint64_t pguid = 0; |
789 | 1997 |
1544 | 1998 spa_config_enter(spa, RW_READER, FTAG); |
789 | 1999 |
1544 | 2000 while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) { |
2001 guid = vd->vdev_guid; | |
2082 | 2002 /* |
2003 * If we have just finished replacing a hot spared device, then | |
2004 * we need to detach the parent's first child (the original hot | |
2005 * spare) as well. | |
2006 */ | |
2007 pvd = vd->vdev_parent; | |
2008 if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops && | |
2009 pvd->vdev_id == 0) { | |
2010 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); | |
2011 ASSERT(pvd->vdev_parent->vdev_children == 2); | |
2012 pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid; | |
2013 } | |
1544 | 2014 spa_config_exit(spa, FTAG); |
2015 if (spa_vdev_detach(spa, guid, B_TRUE) != 0) | |
2016 return; | |
2082 | 2017 if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0) |
2018 return; | |
1544 | 2019 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2020 } |
2021 | |
1544 | 2022 spa_config_exit(spa, FTAG); |
789 | 2023 } |
2024 | |
2025 /* | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2026 * 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
|
2027 * on spa_vdev_enter/exit() to synchronize the labels and cache. |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2028 */ |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2029 int |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2030 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
|
2031 { |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2032 vdev_t *rvd, *vd; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2033 uint64_t txg; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2034 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2035 rvd = spa->spa_root_vdev; |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2036 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2037 txg = spa_vdev_enter(spa); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2038 |
2082 | 2039 if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) { |
2040 /* | |
2041 * Determine if this is a reference to a hot spare. In that | |
2042 * case, update the path as stored in the spare list. | |
2043 */ | |
2044 nvlist_t **spares; | |
2045 uint_t i, nspares; | |
2046 if (spa->spa_sparelist != NULL) { | |
2047 VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist, | |
2048 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); | |
2049 for (i = 0; i < nspares; i++) { | |
2050 uint64_t theguid; | |
2051 VERIFY(nvlist_lookup_uint64(spares[i], | |
2052 ZPOOL_CONFIG_GUID, &theguid) == 0); | |
2053 if (theguid == guid) | |
2054 break; | |
2055 } | |
2056 | |
2057 if (i == nspares) | |
2058 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); | |
2059 | |
2060 VERIFY(nvlist_add_string(spares[i], | |
2061 ZPOOL_CONFIG_PATH, newpath) == 0); | |
2062 spa_load_spares(spa); | |
2063 spa->spa_sync_spares = B_TRUE; | |
2064 return (spa_vdev_exit(spa, NULL, txg, 0)); | |
2065 } else { | |
2066 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); | |
2067 } | |
2068 } | |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2069 |
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
2070 if (!vd->vdev_ops->vdev_op_leaf) |
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
2071 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
|
2072 |
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2073 spa_strfree(vd->vdev_path); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2074 vd->vdev_path = spa_strdup(newpath); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2075 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2076 vdev_config_dirty(vd->vdev_top); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2077 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2078 return (spa_vdev_exit(spa, NULL, txg, 0)); |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2079 } |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2080 |
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
2081 /* |
789 | 2082 * ========================================================================== |
2083 * SPA Scrubbing | |
2084 * ========================================================================== | |
2085 */ | |
2086 | |
1544 | 2087 void |
2088 spa_scrub_throttle(spa_t *spa, int direction) | |
2089 { | |
2090 mutex_enter(&spa->spa_scrub_lock); | |
2091 spa->spa_scrub_throttled += direction; | |
2092 ASSERT(spa->spa_scrub_throttled >= 0); | |
2093 if (spa->spa_scrub_throttled == 0) | |
2094 cv_broadcast(&spa->spa_scrub_io_cv); | |
2095 mutex_exit(&spa->spa_scrub_lock); | |
2096 } | |
789 | 2097 |
2098 static void | |
2099 spa_scrub_io_done(zio_t *zio) | |
2100 { | |
2101 spa_t *spa = zio->io_spa; | |
2102 | |
2103 zio_buf_free(zio->io_data, zio->io_size); | |
2104 | |
2105 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2106 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
|
2107 vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev; |
789 | 2108 spa->spa_scrub_errors++; |
2109 mutex_enter(&vd->vdev_stat_lock); | |
2110 vd->vdev_stat.vs_scrub_errors++; | |
2111 mutex_exit(&vd->vdev_stat_lock); | |
2112 } | |
1544 | 2113 if (--spa->spa_scrub_inflight == 0) { |
2114 cv_broadcast(&spa->spa_scrub_io_cv); | |
2115 ASSERT(spa->spa_scrub_throttled == 0); | |
2116 } | |
2117 mutex_exit(&spa->spa_scrub_lock); | |
789 | 2118 } |
2119 | |
2120 static void | |
1544 | 2121 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags, |
2122 zbookmark_t *zb) | |
789 | 2123 { |
2124 size_t size = BP_GET_LSIZE(bp); | |
2125 void *data = zio_buf_alloc(size); | |
2126 | |
2127 mutex_enter(&spa->spa_scrub_lock); | |
2128 spa->spa_scrub_inflight++; | |
2129 mutex_exit(&spa->spa_scrub_lock); | |
2130 | |
1544 | 2131 if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET) |
2132 flags |= ZIO_FLAG_SPECULATIVE; /* intent log block */ | |
2133 | |
1807
35c8b566d7af
6410711 intent log blocks don't get invited to pool parties
bonwick
parents:
1775
diff
changeset
|
2134 flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL; |
1544 | 2135 |
789 | 2136 zio_nowait(zio_read(NULL, spa, bp, data, size, |
1544 | 2137 spa_scrub_io_done, NULL, priority, flags, zb)); |
789 | 2138 } |
2139 | |
2140 /* ARGSUSED */ | |
2141 static int | |
2142 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a) | |
2143 { | |
2144 blkptr_t *bp = &bc->bc_blkptr; | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2145 vdev_t *vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2146 dva_t *dva = bp->blk_dva; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2147 int needs_resilver = B_FALSE; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2148 int d; |
789 | 2149 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2150 if (bc->bc_errno) { |
789 | 2151 /* |
2152 * We can't scrub this block, but we can continue to scrub | |
2153 * the rest of the pool. Note the error and move along. | |
2154 */ | |
2155 mutex_enter(&spa->spa_scrub_lock); | |
2156 spa->spa_scrub_errors++; | |
2157 mutex_exit(&spa->spa_scrub_lock); | |
2158 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2159 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2160 vd->vdev_stat.vs_scrub_errors++; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2161 mutex_exit(&vd->vdev_stat_lock); |
789 | 2162 |
2163 return (ERESTART); | |
2164 } | |
2165 | |
2166 ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg); | |
2167 | |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2168 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
|
2169 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
|
2170 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2171 ASSERT(vd != NULL); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2172 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2173 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2174 * 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
|
2175 * 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
|
2176 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2177 mutex_enter(&vd->vdev_stat_lock); |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2178 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
|
2179 mutex_exit(&vd->vdev_stat_lock); |
789 | 2180 |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2181 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
|
2182 if (DVA_GET_GANG(&dva[d])) { |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2183 /* |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2184 * Gang members may be spread across multiple |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2185 * 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
|
2186 * pool-wide DTL. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2187 * 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
|
2188 * 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
|
2189 * happen. |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2190 */ |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2191 vd = spa->spa_root_vdev; |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2192 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2193 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
|
2194 bp->blk_birth, 1)) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2195 needs_resilver = B_TRUE; |
789 | 2196 } |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2197 } |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2198 |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2199 if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING) |
789 | 2200 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB, |
1544 | 2201 ZIO_FLAG_SCRUB, &bc->bc_bookmark); |
1775
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2202 else if (needs_resilver) |
e51e26b432c0
6410698 ZFS metadata needs to be more highly replicated (ditto blocks)
billm
parents:
1760
diff
changeset
|
2203 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
|
2204 ZIO_FLAG_RESILVER, &bc->bc_bookmark); |
789 | 2205 |
2206 return (0); | |
2207 } | |
2208 | |
2209 static void | |
2210 spa_scrub_thread(spa_t *spa) | |
2211 { | |
2212 callb_cpr_t cprinfo; | |
2213 traverse_handle_t *th = spa->spa_scrub_th; | |
2214 vdev_t *rvd = spa->spa_root_vdev; | |
2215 pool_scrub_type_t scrub_type = spa->spa_scrub_type; | |
2216 int error = 0; | |
2217 boolean_t complete; | |
2218 | |
2219 CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG); | |
2220 | |
797
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2221 /* |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2222 * 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
|
2223 * wait for that to complete. |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2224 */ |
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
2225 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
|
2226 |
1544 | 2227 dprintf("start %s mintxg=%llu maxtxg=%llu\n", |
2228 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
2229 spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg); | |
2230 | |
2231 spa_config_enter(spa, RW_WRITER, FTAG); | |
2232 vdev_reopen(rvd); /* purge all vdev caches */ | |
789 | 2233 vdev_config_dirty(rvd); /* rewrite all disk labels */ |
2234 vdev_scrub_stat_update(rvd, scrub_type, B_FALSE); | |
1544 | 2235 spa_config_exit(spa, FTAG); |
789 | 2236 |
2237 mutex_enter(&spa->spa_scrub_lock); | |
2238 spa->spa_scrub_errors = 0; | |
2239 spa->spa_scrub_active = 1; | |
1544 | 2240 ASSERT(spa->spa_scrub_inflight == 0); |
2241 ASSERT(spa->spa_scrub_throttled == 0); | |
789 | 2242 |
2243 while (!spa->spa_scrub_stop) { | |
2244 CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
1544 | 2245 while (spa->spa_scrub_suspended) { |
789 | 2246 spa->spa_scrub_active = 0; |
2247 cv_broadcast(&spa->spa_scrub_cv); | |
2248 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2249 spa->spa_scrub_active = 1; | |
2250 } | |
2251 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock); | |
2252 | |
2253 if (spa->spa_scrub_restart_txg != 0) | |
2254 break; | |
2255 | |
2256 mutex_exit(&spa->spa_scrub_lock); | |
2257 error = traverse_more(th); | |
2258 mutex_enter(&spa->spa_scrub_lock); | |
2259 if (error != EAGAIN) | |
2260 break; | |
1544 | 2261 |
2262 while (spa->spa_scrub_throttled > 0) | |
2263 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
789 | 2264 } |
2265 | |
2266 while (spa->spa_scrub_inflight) | |
2267 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2268 | |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2269 spa->spa_scrub_active = 0; |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2270 cv_broadcast(&spa->spa_scrub_cv); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2271 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2272 mutex_exit(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2273 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2274 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
|
2275 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2276 mutex_enter(&spa->spa_scrub_lock); |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2277 |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2278 /* |
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2279 * 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
|
2280 * 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
|
2281 * 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
|
2282 */ |
789 | 2283 if (spa->spa_scrub_restart_txg != 0) |
2284 error = ERESTART; | |
2285 | |
1544 | 2286 if (spa->spa_scrub_stop) |
2287 error = EINTR; | |
2288 | |
789 | 2289 /* |
1544 | 2290 * Even if there were uncorrectable errors, we consider the scrub |
2291 * completed. The downside is that if there is a transient error during | |
2292 * a resilver, we won't resilver the data properly to the target. But | |
2293 * if the damage is permanent (more likely) we will resilver forever, | |
2294 * which isn't really acceptable. Since there is enough information for | |
2295 * the user to know what has failed and why, this seems like a more | |
2296 * tractable approach. | |
789 | 2297 */ |
1544 | 2298 complete = (error == 0); |
789 | 2299 |
1544 | 2300 dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n", |
2301 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
789 | 2302 spa->spa_scrub_maxtxg, complete ? "done" : "FAILED", |
2303 error, spa->spa_scrub_errors, spa->spa_scrub_stop); | |
2304 | |
2305 mutex_exit(&spa->spa_scrub_lock); | |
2306 | |
2307 /* | |
2308 * If the scrub/resilver completed, update all DTLs to reflect this. | |
2309 * Whether it succeeded or not, vacate all temporary scrub DTLs. | |
2310 */ | |
2311 vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1, | |
2312 complete ? spa->spa_scrub_maxtxg : 0, B_TRUE); | |
2313 vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete); | |
1544 | 2314 spa_errlog_rotate(spa); |
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
2315 |
1544 | 2316 spa_config_exit(spa, FTAG); |
789 | 2317 |
2318 mutex_enter(&spa->spa_scrub_lock); | |
2319 | |
1544 | 2320 /* |
2321 * We may have finished replacing a device. | |
2322 * Let the async thread assess this and handle the detach. | |
2323 */ | |
2324 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); | |
789 | 2325 |
2326 /* | |
2327 * If we were told to restart, our final act is to start a new scrub. | |
2328 */ | |
2329 if (error == ERESTART) | |
1544 | 2330 spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ? |
2331 SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB); | |
789 | 2332 |
1544 | 2333 spa->spa_scrub_type = POOL_SCRUB_NONE; |
2334 spa->spa_scrub_active = 0; | |
2335 spa->spa_scrub_thread = NULL; | |
2336 cv_broadcast(&spa->spa_scrub_cv); | |
789 | 2337 CALLB_CPR_EXIT(&cprinfo); /* drops &spa->spa_scrub_lock */ |
2338 thread_exit(); | |
2339 } | |
2340 | |
2341 void | |
2342 spa_scrub_suspend(spa_t *spa) | |
2343 { | |
2344 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2345 spa->spa_scrub_suspended++; |
789 | 2346 while (spa->spa_scrub_active) { |
2347 cv_broadcast(&spa->spa_scrub_cv); | |
2348 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2349 } | |
2350 while (spa->spa_scrub_inflight) | |
2351 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
2352 mutex_exit(&spa->spa_scrub_lock); | |
2353 } | |
2354 | |
2355 void | |
2356 spa_scrub_resume(spa_t *spa) | |
2357 { | |
2358 mutex_enter(&spa->spa_scrub_lock); | |
1544 | 2359 ASSERT(spa->spa_scrub_suspended != 0); |
2360 if (--spa->spa_scrub_suspended == 0) | |
789 | 2361 cv_broadcast(&spa->spa_scrub_cv); |
2362 mutex_exit(&spa->spa_scrub_lock); | |
2363 } | |
2364 | |
2365 void | |
2366 spa_scrub_restart(spa_t *spa, uint64_t txg) | |
2367 { | |
2368 /* | |
2369 * Something happened (e.g. snapshot create/delete) that means | |
2370 * we must restart any in-progress scrubs. The itinerary will | |
2371 * fix this properly. | |
2372 */ | |
2373 mutex_enter(&spa->spa_scrub_lock); | |
2374 spa->spa_scrub_restart_txg = txg; | |
2375 mutex_exit(&spa->spa_scrub_lock); | |
2376 } | |
2377 | |
1544 | 2378 int |
2379 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force) | |
789 | 2380 { |
2381 space_seg_t *ss; | |
2382 uint64_t mintxg, maxtxg; | |
2383 vdev_t *rvd = spa->spa_root_vdev; | |
2384 | |
2385 if ((uint_t)type >= POOL_SCRUB_TYPES) | |
2386 return (ENOTSUP); | |
2387 | |
1544 | 2388 mutex_enter(&spa->spa_scrub_lock); |
2389 | |
789 | 2390 /* |
2391 * If there's a scrub or resilver already in progress, stop it. | |
2392 */ | |
2393 while (spa->spa_scrub_thread != NULL) { | |
2394 /* | |
2395 * Don't stop a resilver unless forced. | |
2396 */ | |
1544 | 2397 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) { |
2398 mutex_exit(&spa->spa_scrub_lock); | |
789 | 2399 return (EBUSY); |
1544 | 2400 } |
789 | 2401 spa->spa_scrub_stop = 1; |
2402 cv_broadcast(&spa->spa_scrub_cv); | |
2403 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
2404 } | |
2405 | |
2406 /* | |
2407 * Terminate the previous traverse. | |
2408 */ | |
2409 if (spa->spa_scrub_th != NULL) { | |
2410 traverse_fini(spa->spa_scrub_th); | |
2411 spa->spa_scrub_th = NULL; | |
2412 } | |
2413 | |
1544 | 2414 if (rvd == NULL) { |
2415 ASSERT(spa->spa_scrub_stop == 0); | |
2416 ASSERT(spa->spa_scrub_type == type); | |
2417 ASSERT(spa->spa_scrub_restart_txg == 0); | |
2418 mutex_exit(&spa->spa_scrub_lock); | |
2419 return (0); | |
2420 } | |
789 | 2421 |
2422 mintxg = TXG_INITIAL - 1; | |
2423 maxtxg = spa_last_synced_txg(spa) + 1; | |
2424 | |
1544 | 2425 mutex_enter(&rvd->vdev_dtl_lock); |
789 | 2426 |
1544 | 2427 if (rvd->vdev_dtl_map.sm_space == 0) { |
2428 /* | |
2429 * The pool-wide DTL is empty. | |
1732 | 2430 * If this is a resilver, there's nothing to do except |
2431 * check whether any in-progress replacements have completed. | |
1544 | 2432 */ |
1732 | 2433 if (type == POOL_SCRUB_RESILVER) { |
1544 | 2434 type = POOL_SCRUB_NONE; |
1732 | 2435 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); |
2436 } | |
1544 | 2437 } else { |
2438 /* | |
2439 * The pool-wide DTL is non-empty. | |
2440 * If this is a normal scrub, upgrade to a resilver instead. | |
2441 */ | |
2442 if (type == POOL_SCRUB_EVERYTHING) | |
2443 type = POOL_SCRUB_RESILVER; | |
2444 } | |
789 | 2445 |
1544 | 2446 if (type == POOL_SCRUB_RESILVER) { |
789 | 2447 /* |
2448 * Determine the resilvering boundaries. | |
2449 * | |
2450 * Note: (mintxg, maxtxg) is an open interval, | |
2451 * i.e. mintxg and maxtxg themselves are not included. | |
2452 * | |
2453 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1 | |
2454 * so we don't claim to resilver a txg that's still changing. | |
2455 */ | |
2456 ss = avl_first(&rvd->vdev_dtl_map.sm_root); | |
1544 | 2457 mintxg = ss->ss_start - 1; |
789 | 2458 ss = avl_last(&rvd->vdev_dtl_map.sm_root); |
1544 | 2459 maxtxg = MIN(ss->ss_end, maxtxg); |
789 | 2460 } |
2461 | |
1544 | 2462 mutex_exit(&rvd->vdev_dtl_lock); |
2463 | |
2464 spa->spa_scrub_stop = 0; | |
2465 spa->spa_scrub_type = type; | |
2466 spa->spa_scrub_restart_txg = 0; | |
2467 | |
2468 if (type != POOL_SCRUB_NONE) { | |
2469 spa->spa_scrub_mintxg = mintxg; | |
789 | 2470 spa->spa_scrub_maxtxg = maxtxg; |
2471 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
|
2472 ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL, |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2473 ZIO_FLAG_CANFAIL); |
789 | 2474 traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg); |
2475 spa->spa_scrub_thread = thread_create(NULL, 0, | |
2476 spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri); | |
2477 } | |
2478 | |
1544 | 2479 mutex_exit(&spa->spa_scrub_lock); |
2480 | |
789 | 2481 return (0); |
2482 } | |
2483 | |
1544 | 2484 /* |
2485 * ========================================================================== | |
2486 * SPA async task processing | |
2487 * ========================================================================== | |
2488 */ | |
2489 | |
2490 static void | |
2491 spa_async_reopen(spa_t *spa) | |
789 | 2492 { |
1544 | 2493 vdev_t *rvd = spa->spa_root_vdev; |
2494 vdev_t *tvd; | |
2495 int c; | |
2496 | |
2497 spa_config_enter(spa, RW_WRITER, FTAG); | |
2498 | |
2499 for (c = 0; c < rvd->vdev_children; c++) { | |
2500 tvd = rvd->vdev_child[c]; | |
2501 if (tvd->vdev_reopen_wanted) { | |
2502 tvd->vdev_reopen_wanted = 0; | |
2503 vdev_reopen(tvd); | |
2504 } | |
2505 } | |
789 | 2506 |
1544 | 2507 spa_config_exit(spa, FTAG); |
2508 } | |
2509 | |
2510 static void | |
2511 spa_async_thread(spa_t *spa) | |
2512 { | |
2513 int tasks; | |
2514 | |
2515 ASSERT(spa->spa_sync_on); | |
789 | 2516 |
1544 | 2517 mutex_enter(&spa->spa_async_lock); |
2518 tasks = spa->spa_async_tasks; | |
2519 spa->spa_async_tasks = 0; | |
2520 mutex_exit(&spa->spa_async_lock); | |
2521 | |
2522 /* | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2523 * 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
|
2524 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2525 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2526 mutex_enter(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2527 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
|
2528 mutex_exit(&spa_namespace_lock); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2529 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2530 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2531 /* |
1544 | 2532 * See if any devices need to be reopened. |
2533 */ | |
2534 if (tasks & SPA_ASYNC_REOPEN) | |
2535 spa_async_reopen(spa); | |
2536 | |
2537 /* | |
2538 * If any devices are done replacing, detach them. | |
2539 */ | |
2540 if (tasks & SPA_ASYNC_REPLACE_DONE) | |
789 | 2541 spa_vdev_replace_done(spa); |
2542 | |
1544 | 2543 /* |
2544 * Kick off a scrub. | |
2545 */ | |
2546 if (tasks & SPA_ASYNC_SCRUB) | |
2547 VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0); | |
2548 | |
2549 /* | |
2550 * Kick off a resilver. | |
2551 */ | |
2552 if (tasks & SPA_ASYNC_RESILVER) | |
2553 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
2554 | |
2555 /* | |
2556 * Let the world know that we're done. | |
2557 */ | |
2558 mutex_enter(&spa->spa_async_lock); | |
2559 spa->spa_async_thread = NULL; | |
2560 cv_broadcast(&spa->spa_async_cv); | |
2561 mutex_exit(&spa->spa_async_lock); | |
2562 thread_exit(); | |
2563 } | |
2564 | |
2565 void | |
2566 spa_async_suspend(spa_t *spa) | |
2567 { | |
2568 mutex_enter(&spa->spa_async_lock); | |
2569 spa->spa_async_suspended++; | |
2570 while (spa->spa_async_thread != NULL) | |
2571 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
2572 mutex_exit(&spa->spa_async_lock); | |
2573 } | |
2574 | |
2575 void | |
2576 spa_async_resume(spa_t *spa) | |
2577 { | |
2578 mutex_enter(&spa->spa_async_lock); | |
2579 ASSERT(spa->spa_async_suspended != 0); | |
2580 spa->spa_async_suspended--; | |
2581 mutex_exit(&spa->spa_async_lock); | |
2582 } | |
2583 | |
2584 static void | |
2585 spa_async_dispatch(spa_t *spa) | |
2586 { | |
2587 mutex_enter(&spa->spa_async_lock); | |
2588 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
|
2589 spa->spa_async_thread == NULL && |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2590 rootdir != NULL && !vn_is_readonly(rootdir)) |
1544 | 2591 spa->spa_async_thread = thread_create(NULL, 0, |
2592 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
2593 mutex_exit(&spa->spa_async_lock); | |
2594 } | |
2595 | |
2596 void | |
2597 spa_async_request(spa_t *spa, int task) | |
2598 { | |
2599 mutex_enter(&spa->spa_async_lock); | |
2600 spa->spa_async_tasks |= task; | |
2601 mutex_exit(&spa->spa_async_lock); | |
789 | 2602 } |
2603 | |
2604 /* | |
2605 * ========================================================================== | |
2606 * SPA syncing routines | |
2607 * ========================================================================== | |
2608 */ | |
2609 | |
2610 static void | |
2611 spa_sync_deferred_frees(spa_t *spa, uint64_t txg) | |
2612 { | |
2613 bplist_t *bpl = &spa->spa_sync_bplist; | |
2614 dmu_tx_t *tx; | |
2615 blkptr_t blk; | |
2616 uint64_t itor = 0; | |
2617 zio_t *zio; | |
2618 int error; | |
2619 uint8_t c = 1; | |
2620 | |
2621 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD); | |
2622 | |
2623 while (bplist_iterate(bpl, &itor, &blk) == 0) | |
2624 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL)); | |
2625 | |
2626 error = zio_wait(zio); | |
2627 ASSERT3U(error, ==, 0); | |
2628 | |
2629 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
2630 bplist_vacate(bpl, tx); | |
2631 | |
2632 /* | |
2633 * Pre-dirty the first block so we sync to convergence faster. | |
2634 * (Usually only the first block is needed.) | |
2635 */ | |
2636 dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); | |
2637 dmu_tx_commit(tx); | |
2638 } | |
2639 | |
2640 static void | |
2082 | 2641 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) |
2642 { | |
2643 char *packed = NULL; | |
2644 size_t nvsize = 0; | |
2645 dmu_buf_t *db; | |
2646 | |
2647 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); | |
2648 | |
2649 packed = kmem_alloc(nvsize, KM_SLEEP); | |
2650 | |
2651 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, | |
2652 KM_SLEEP) == 0); | |
2653 | |
2654 dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx); | |
2655 | |
2656 kmem_free(packed, nvsize); | |
2657 | |
2658 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); | |
2659 dmu_buf_will_dirty(db, tx); | |
2660 *(uint64_t *)db->db_data = nvsize; | |
2661 dmu_buf_rele(db, FTAG); | |
2662 } | |
2663 | |
2664 static void | |
2665 spa_sync_spares(spa_t *spa, dmu_tx_t *tx) | |
2666 { | |
2667 nvlist_t *nvroot; | |
2668 nvlist_t **spares; | |
2669 int i; | |
2670 | |
2671 if (!spa->spa_sync_spares) | |
2672 return; | |
2673 | |
2674 /* | |
2675 * Update the MOS nvlist describing the list of available spares. | |
2676 * spa_validate_spares() will have already made sure this nvlist is | |
2677 * valid and the vdevs are labelled appropriately. | |
2678 */ | |
2679 if (spa->spa_spares_object == 0) { | |
2680 spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset, | |
2681 DMU_OT_PACKED_NVLIST, 1 << 14, | |
2682 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
2683 VERIFY(zap_update(spa->spa_meta_objset, | |
2684 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES, | |
2685 sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0); | |
2686 } | |
2687 | |
2688 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); | |
2689 if (spa->spa_nspares == 0) { | |
2690 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
2691 NULL, 0) == 0); | |
2692 } else { | |
2693 spares = kmem_alloc(spa->spa_nspares * sizeof (void *), | |
2694 KM_SLEEP); | |
2695 for (i = 0; i < spa->spa_nspares; i++) | |
2696 spares[i] = vdev_config_generate(spa, | |
2697 spa->spa_spares[i], B_FALSE, B_TRUE); | |
2698 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, | |
2699 spares, spa->spa_nspares) == 0); | |
2700 for (i = 0; i < spa->spa_nspares; i++) | |
2701 nvlist_free(spares[i]); | |
2702 kmem_free(spares, spa->spa_nspares * sizeof (void *)); | |
2703 } | |
2704 | |
2705 spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx); | |
2706 | |
2707 spa->spa_sync_spares = B_FALSE; | |
2708 } | |
2709 | |
2710 static void | |
789 | 2711 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) |
2712 { | |
2713 nvlist_t *config; | |
2714 | |
2715 if (list_is_empty(&spa->spa_dirty_list)) | |
2716 return; | |
2717 | |
2718 config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE); | |
2719 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2720 if (spa->spa_config_syncing) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2721 nvlist_free(spa->spa_config_syncing); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2722 spa->spa_config_syncing = config; |
789 | 2723 |
2082 | 2724 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); |
789 | 2725 } |
2726 | |
2727 /* | |
2728 * Sync the specified transaction group. New blocks may be dirtied as | |
2729 * part of the process, so we iterate until it converges. | |
2730 */ | |
2731 void | |
2732 spa_sync(spa_t *spa, uint64_t txg) | |
2733 { | |
2734 dsl_pool_t *dp = spa->spa_dsl_pool; | |
2735 objset_t *mos = spa->spa_meta_objset; | |
2736 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
|
2737 vdev_t *rvd = spa->spa_root_vdev; |
789 | 2738 vdev_t *vd; |
2739 dmu_tx_t *tx; | |
2740 int dirty_vdevs; | |
2741 | |
2742 /* | |
2743 * Lock out configuration changes. | |
2744 */ | |
1544 | 2745 spa_config_enter(spa, RW_READER, FTAG); |
789 | 2746 |
2747 spa->spa_syncing_txg = txg; | |
2748 spa->spa_sync_pass = 0; | |
2749 | |
1544 | 2750 VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); |
789 | 2751 |
2082 | 2752 tx = dmu_tx_create_assigned(dp, txg); |
2753 | |
2754 /* | |
2755 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg, | |
2756 * set spa_deflate if we have no raid-z vdevs. | |
2757 */ | |
2758 if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE && | |
2759 spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) { | |
2760 int i; | |
2761 | |
2762 for (i = 0; i < rvd->vdev_children; i++) { | |
2763 vd = rvd->vdev_child[i]; | |
2764 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) | |
2765 break; | |
2766 } | |
2767 if (i == rvd->vdev_children) { | |
2768 spa->spa_deflate = TRUE; | |
2769 VERIFY(0 == zap_add(spa->spa_meta_objset, | |
2770 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, | |
2771 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); | |
2772 } | |
2773 } | |
2774 | |
789 | 2775 /* |
2776 * If anything has changed in this txg, push the deferred frees | |
2777 * from the previous txg. If not, leave them alone so that we | |
2778 * don't generate work on an otherwise idle system. | |
2779 */ | |
2780 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || | |
2781 !txg_list_empty(&dp->dp_dirty_dirs, txg)) | |
2782 spa_sync_deferred_frees(spa, txg); | |
2783 | |
2784 /* | |
2785 * Iterate to convergence. | |
2786 */ | |
2787 do { | |
2788 spa->spa_sync_pass++; | |
2789 | |
2790 spa_sync_config_object(spa, tx); | |
2082 | 2791 spa_sync_spares(spa, tx); |
1544 | 2792 spa_errlog_sync(spa, txg); |
789 | 2793 dsl_pool_sync(dp, txg); |
2794 | |
2795 dirty_vdevs = 0; | |
2796 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { | |
2797 vdev_sync(vd, txg); | |
2798 dirty_vdevs++; | |
2799 } | |
2800 | |
2801 bplist_sync(bpl, tx); | |
2802 } while (dirty_vdevs); | |
2803 | |
2804 bplist_close(bpl); | |
2805 | |
2806 dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); | |
2807 | |
2808 /* | |
2809 * Rewrite the vdev configuration (which includes the uberblock) | |
2810 * to commit the transaction group. | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2811 * |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2812 * 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
|
2813 * 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
|
2814 * 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
|
2815 * If the write fails, try the next vdev until we're tried them all. |
789 | 2816 */ |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2817 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
|
2818 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2819 } else { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2820 int children = rvd->vdev_children; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2821 int c0 = spa_get_random(children); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2822 int c; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2823 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2824 for (c = 0; c < children; c++) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2825 vd = rvd->vdev_child[(c0 + c) % children]; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2826 if (vd->vdev_ms_array == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2827 continue; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2828 if (vdev_config_sync(vd, txg) == 0) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2829 break; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2830 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2831 if (c == children) |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2832 VERIFY(vdev_config_sync(rvd, txg) == 0); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2833 } |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2834 |
2082 | 2835 dmu_tx_commit(tx); |
2836 | |
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2837 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2838 * Clear the dirty config list. |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2839 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2840 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
|
2841 vdev_config_clean(vd); |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2842 |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2843 /* |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2844 * 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
|
2845 * 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
|
2846 */ |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2847 if (spa->spa_config_syncing != NULL) { |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2848 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
|
2849 spa->spa_config_txg = txg; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2850 spa->spa_config_syncing = NULL; |
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2851 } |
789 | 2852 |
2853 /* | |
2854 * Make a stable copy of the fully synced uberblock. | |
2855 * We use this as the root for pool traversals. | |
2856 */ | |
2857 spa->spa_traverse_wanted = 1; /* tells traverse_more() to stop */ | |
2858 | |
2859 spa_scrub_suspend(spa); /* stop scrubbing and finish I/Os */ | |
2860 | |
2861 rw_enter(&spa->spa_traverse_lock, RW_WRITER); | |
2862 spa->spa_traverse_wanted = 0; | |
2863 spa->spa_ubsync = spa->spa_uberblock; | |
2864 rw_exit(&spa->spa_traverse_lock); | |
2865 | |
2866 spa_scrub_resume(spa); /* resume scrub with new ubsync */ | |
2867 | |
2868 /* | |
2869 * Clean up the ZIL records for the synced txg. | |
2870 */ | |
2871 dsl_pool_zil_clean(dp); | |
2872 | |
2873 /* | |
2874 * Update usable space statistics. | |
2875 */ | |
2876 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) | |
2877 vdev_sync_done(vd, txg); | |
2878 | |
2879 /* | |
2880 * It had better be the case that we didn't dirty anything | |
2082 | 2881 * since vdev_config_sync(). |
789 | 2882 */ |
2883 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
2884 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
2885 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
2886 ASSERT(bpl->bpl_queue == NULL); | |
2887 | |
1544 | 2888 spa_config_exit(spa, FTAG); |
2889 | |
2890 /* | |
2891 * If any async tasks have been requested, kick them off. | |
2892 */ | |
2893 spa_async_dispatch(spa); | |
789 | 2894 } |
2895 | |
2896 /* | |
2897 * Sync all pools. We don't want to hold the namespace lock across these | |
2898 * operations, so we take a reference on the spa_t and drop the lock during the | |
2899 * sync. | |
2900 */ | |
2901 void | |
2902 spa_sync_allpools(void) | |
2903 { | |
2904 spa_t *spa = NULL; | |
2905 mutex_enter(&spa_namespace_lock); | |
2906 while ((spa = spa_next(spa)) != NULL) { | |
2907 if (spa_state(spa) != POOL_STATE_ACTIVE) | |
2908 continue; | |
2909 spa_open_ref(spa, FTAG); | |
2910 mutex_exit(&spa_namespace_lock); | |
2911 txg_wait_synced(spa_get_dsl(spa), 0); | |
2912 mutex_enter(&spa_namespace_lock); | |
2913 spa_close(spa, FTAG); | |
2914 } | |
2915 mutex_exit(&spa_namespace_lock); | |
2916 } | |
2917 | |
2918 /* | |
2919 * ========================================================================== | |
2920 * Miscellaneous routines | |
2921 * ========================================================================== | |
2922 */ | |
2923 | |
2924 /* | |
2925 * Remove all pools in the system. | |
2926 */ | |
2927 void | |
2928 spa_evict_all(void) | |
2929 { | |
2930 spa_t *spa; | |
2931 | |
2932 /* | |
2933 * Remove all cached state. All pools should be closed now, | |
2934 * so every spa in the AVL tree should be unreferenced. | |
2935 */ | |
2936 mutex_enter(&spa_namespace_lock); | |
2937 while ((spa = spa_next(NULL)) != NULL) { | |
2938 /* | |
1544 | 2939 * Stop async tasks. The async thread may need to detach |
2940 * a device that's been replaced, which requires grabbing | |
2941 * spa_namespace_lock, so we must drop it here. | |
789 | 2942 */ |
2943 spa_open_ref(spa, FTAG); | |
2944 mutex_exit(&spa_namespace_lock); | |
1544 | 2945 spa_async_suspend(spa); |
789 | 2946 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); |
2947 mutex_enter(&spa_namespace_lock); | |
2948 spa_close(spa, FTAG); | |
2949 | |
2950 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
2951 spa_unload(spa); | |
2952 spa_deactivate(spa); | |
2953 } | |
2954 spa_remove(spa); | |
2955 } | |
2956 mutex_exit(&spa_namespace_lock); | |
2957 } | |
1544 | 2958 |
2959 vdev_t * | |
2960 spa_lookup_by_guid(spa_t *spa, uint64_t guid) | |
2961 { | |
2962 return (vdev_lookup_by_guid(spa->spa_root_vdev, guid)); | |
2963 } | |
1760 | 2964 |
2965 void | |
2966 spa_upgrade(spa_t *spa) | |
2967 { | |
2968 spa_config_enter(spa, RW_WRITER, FTAG); | |
2969 | |
2970 /* | |
2971 * This should only be called for a non-faulted pool, and since a | |
2972 * future version would result in an unopenable pool, this shouldn't be | |
2973 * possible. | |
2974 */ | |
2975 ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION); | |
2976 | |
2977 spa->spa_uberblock.ub_version = ZFS_VERSION; | |
2978 vdev_config_dirty(spa->spa_root_vdev); | |
2979 | |
2980 spa_config_exit(spa, FTAG); | |
2082 | 2981 |
2982 txg_wait_synced(spa_get_dsl(spa), 0); | |
1760 | 2983 } |
2082 | 2984 |
2985 boolean_t | |
2986 spa_has_spare(spa_t *spa, uint64_t guid) | |
2987 { | |
2988 int i; | |
2989 | |
2990 for (i = 0; i < spa->spa_nspares; i++) | |
2991 if (spa->spa_spares[i]->vdev_guid == guid) | |
2992 return (B_TRUE); | |
2993 | |
2994 return (B_FALSE); | |
2995 } |