Mercurial > illumos > illumos-gate
annotate usr/src/uts/common/fs/zfs/spa.c @ 1732:9e3ae798af31
6280668 pluggable block allocation policy
6399301 initial read of space maps is super slow
6407365 large-sector disk support in ZFS
6407366 ADVANCE_NOLOCK gathers MOS
6407367 three-way deadlock between db_mtx, dbuf_hash[], and ms_lock
| author | bonwick |
|---|---|
| date | Sun, 02 Apr 2006 00:47:06 -0800 |
| parents | 0ab1193d47cb |
| children | a7c3bc84e012 |
| rev | line source |
|---|---|
| 789 | 1 /* |
| 2 * CDDL HEADER START | |
| 3 * | |
| 4 * The contents of this file are subject to the terms of the | |
| 1544 | 5 * Common Development and Distribution License (the "License"). |
| 6 * You may not use this file except in compliance with the License. | |
| 789 | 7 * |
| 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
| 9 * or http://www.opensolaris.org/os/licensing. | |
| 10 * See the License for the specific language governing permissions | |
| 11 * and limitations under the License. | |
| 12 * | |
| 13 * When distributing Covered Code, include this CDDL HEADER in each | |
| 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
| 15 * If applicable, add the following below this CDDL HEADER, with the | |
| 16 * fields enclosed by brackets "[]" replaced with your own identifying | |
| 17 * information: Portions Copyright [yyyy] [name of copyright owner] | |
| 18 * | |
| 19 * CDDL HEADER END | |
| 20 */ | |
| 21 /* | |
|
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. |
| 789 | 23 * Use is subject to license terms. |
| 24 */ | |
| 25 | |
| 26 #pragma ident "%Z%%M% %I% %E% SMI" | |
| 27 | |
| 28 /* | |
| 29 * This file contains all the routines used when modifying on-disk SPA state. | |
| 30 * This includes opening, importing, destroying, exporting a pool, and syncing a | |
| 31 * pool. | |
| 32 */ | |
| 33 | |
| 34 #include <sys/zfs_context.h> | |
| 1544 | 35 #include <sys/fm/fs/zfs.h> |
| 789 | 36 #include <sys/spa_impl.h> |
| 37 #include <sys/zio.h> | |
| 38 #include <sys/zio_checksum.h> | |
| 39 #include <sys/zio_compress.h> | |
| 40 #include <sys/dmu.h> | |
| 41 #include <sys/dmu_tx.h> | |
| 42 #include <sys/zap.h> | |
| 43 #include <sys/zil.h> | |
| 44 #include <sys/vdev_impl.h> | |
| 45 #include <sys/metaslab.h> | |
| 46 #include <sys/uberblock_impl.h> | |
| 47 #include <sys/txg.h> | |
| 48 #include <sys/avl.h> | |
| 49 #include <sys/dmu_traverse.h> | |
| 50 #include <sys/unique.h> | |
| 51 #include <sys/dsl_pool.h> | |
| 52 #include <sys/dsl_dir.h> | |
| 53 #include <sys/dsl_prop.h> | |
| 54 #include <sys/fs/zfs.h> | |
| 55 #include <sys/callb.h> | |
| 56 | |
| 57 /* | |
| 58 * ========================================================================== | |
| 59 * SPA state manipulation (open/create/destroy/import/export) | |
| 60 * ========================================================================== | |
| 61 */ | |
| 62 | |
| 1544 | 63 static int |
| 64 spa_error_entry_compare(const void *a, const void *b) | |
| 65 { | |
| 66 spa_error_entry_t *sa = (spa_error_entry_t *)a; | |
| 67 spa_error_entry_t *sb = (spa_error_entry_t *)b; | |
| 68 int ret; | |
| 69 | |
| 70 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, | |
| 71 sizeof (zbookmark_t)); | |
| 72 | |
| 73 if (ret < 0) | |
| 74 return (-1); | |
| 75 else if (ret > 0) | |
| 76 return (1); | |
| 77 else | |
| 78 return (0); | |
| 79 } | |
| 80 | |
| 81 /* | |
| 82 * Utility function which retrieves copies of the current logs and | |
| 83 * re-initializes them in the process. | |
| 84 */ | |
| 85 void | |
| 86 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) | |
| 87 { | |
| 88 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); | |
| 89 | |
| 90 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); | |
| 91 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); | |
| 92 | |
| 93 avl_create(&spa->spa_errlist_scrub, | |
| 94 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
| 95 offsetof(spa_error_entry_t, se_avl)); | |
| 96 avl_create(&spa->spa_errlist_last, | |
| 97 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
| 98 offsetof(spa_error_entry_t, se_avl)); | |
| 99 } | |
| 100 | |
| 789 | 101 /* |
| 102 * Activate an uninitialized pool. | |
| 103 */ | |
| 104 static void | |
| 105 spa_activate(spa_t *spa) | |
| 106 { | |
| 107 int t; | |
| 108 | |
| 109 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); | |
| 110 | |
| 111 spa->spa_state = POOL_STATE_ACTIVE; | |
| 112 | |
| 113 spa->spa_normal_class = metaslab_class_create(); | |
| 114 | |
| 115 for (t = 0; t < ZIO_TYPES; t++) { | |
| 116 spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue", | |
| 117 8, maxclsyspri, 50, INT_MAX, | |
| 118 TASKQ_PREPOPULATE); | |
| 119 spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr", | |
| 120 8, maxclsyspri, 50, INT_MAX, | |
| 121 TASKQ_PREPOPULATE); | |
| 122 } | |
| 123 | |
| 124 rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL); | |
| 125 | |
| 126 list_create(&spa->spa_dirty_list, sizeof (vdev_t), | |
| 127 offsetof(vdev_t, vdev_dirty_node)); | |
| 128 | |
| 129 txg_list_create(&spa->spa_vdev_txg_list, | |
| 130 offsetof(struct vdev, vdev_txg_node)); | |
| 1544 | 131 |
| 132 avl_create(&spa->spa_errlist_scrub, | |
| 133 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
| 134 offsetof(spa_error_entry_t, se_avl)); | |
| 135 avl_create(&spa->spa_errlist_last, | |
| 136 spa_error_entry_compare, sizeof (spa_error_entry_t), | |
| 137 offsetof(spa_error_entry_t, se_avl)); | |
| 789 | 138 } |
| 139 | |
| 140 /* | |
| 141 * Opposite of spa_activate(). | |
| 142 */ | |
| 143 static void | |
| 144 spa_deactivate(spa_t *spa) | |
| 145 { | |
| 146 int t; | |
| 147 | |
| 148 ASSERT(spa->spa_sync_on == B_FALSE); | |
| 149 ASSERT(spa->spa_dsl_pool == NULL); | |
| 150 ASSERT(spa->spa_root_vdev == NULL); | |
| 151 | |
| 152 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); | |
| 153 | |
| 154 txg_list_destroy(&spa->spa_vdev_txg_list); | |
| 155 | |
| 156 list_destroy(&spa->spa_dirty_list); | |
| 157 | |
| 158 rw_destroy(&spa->spa_traverse_lock); | |
| 159 | |
| 160 for (t = 0; t < ZIO_TYPES; t++) { | |
| 161 taskq_destroy(spa->spa_zio_issue_taskq[t]); | |
| 162 taskq_destroy(spa->spa_zio_intr_taskq[t]); | |
| 163 spa->spa_zio_issue_taskq[t] = NULL; | |
| 164 spa->spa_zio_intr_taskq[t] = NULL; | |
| 165 } | |
| 166 | |
| 167 metaslab_class_destroy(spa->spa_normal_class); | |
| 168 spa->spa_normal_class = NULL; | |
| 169 | |
| 1544 | 170 /* |
| 171 * If this was part of an import or the open otherwise failed, we may | |
| 172 * still have errors left in the queues. Empty them just in case. | |
| 173 */ | |
| 174 spa_errlog_drain(spa); | |
| 175 | |
| 176 avl_destroy(&spa->spa_errlist_scrub); | |
| 177 avl_destroy(&spa->spa_errlist_last); | |
| 178 | |
| 789 | 179 spa->spa_state = POOL_STATE_UNINITIALIZED; |
| 180 } | |
| 181 | |
| 182 /* | |
| 183 * Verify a pool configuration, and construct the vdev tree appropriately. This | |
| 184 * will create all the necessary vdevs in the appropriate layout, with each vdev | |
| 185 * in the CLOSED state. This will prep the pool before open/creation/import. | |
| 186 * All vdev validation is done by the vdev_alloc() routine. | |
| 187 */ | |
| 188 static vdev_t * | |
| 189 spa_config_parse(spa_t *spa, nvlist_t *nv, vdev_t *parent, uint_t id, int atype) | |
| 190 { | |
| 191 nvlist_t **child; | |
| 192 uint_t c, children; | |
| 193 vdev_t *vd; | |
| 194 | |
| 195 if ((vd = vdev_alloc(spa, nv, parent, id, atype)) == NULL) | |
| 196 return (NULL); | |
| 197 | |
| 198 if (vd->vdev_ops->vdev_op_leaf) | |
| 199 return (vd); | |
| 200 | |
| 201 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, | |
| 202 &child, &children) != 0) { | |
| 203 vdev_free(vd); | |
| 204 return (NULL); | |
| 205 } | |
| 206 | |
| 207 for (c = 0; c < children; c++) { | |
| 208 if (spa_config_parse(spa, child[c], vd, c, atype) == NULL) { | |
| 209 vdev_free(vd); | |
| 210 return (NULL); | |
| 211 } | |
| 212 } | |
| 213 | |
| 214 return (vd); | |
| 215 } | |
| 216 | |
| 217 /* | |
| 218 * Opposite of spa_load(). | |
| 219 */ | |
| 220 static void | |
| 221 spa_unload(spa_t *spa) | |
| 222 { | |
| 223 /* | |
| 1544 | 224 * Stop async tasks. |
| 225 */ | |
| 226 spa_async_suspend(spa); | |
| 227 | |
| 228 /* | |
| 789 | 229 * Stop syncing. |
| 230 */ | |
| 231 if (spa->spa_sync_on) { | |
| 232 txg_sync_stop(spa->spa_dsl_pool); | |
| 233 spa->spa_sync_on = B_FALSE; | |
| 234 } | |
| 235 | |
| 236 /* | |
| 237 * Wait for any outstanding prefetch I/O to complete. | |
| 238 */ | |
| 1544 | 239 spa_config_enter(spa, RW_WRITER, FTAG); |
| 240 spa_config_exit(spa, FTAG); | |
| 789 | 241 |
| 242 /* | |
| 243 * Close the dsl pool. | |
| 244 */ | |
| 245 if (spa->spa_dsl_pool) { | |
| 246 dsl_pool_close(spa->spa_dsl_pool); | |
| 247 spa->spa_dsl_pool = NULL; | |
| 248 } | |
| 249 | |
| 250 /* | |
| 251 * Close all vdevs. | |
| 252 */ | |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
253 if (spa->spa_root_vdev) |
| 789 | 254 vdev_free(spa->spa_root_vdev); |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
255 ASSERT(spa->spa_root_vdev == NULL); |
| 1544 | 256 |
| 257 spa->spa_async_suspended = 0; | |
| 789 | 258 } |
| 259 | |
| 260 /* | |
| 261 * Load an existing storage pool, using the pool's builtin spa_config as a | |
| 1544 | 262 * source of configuration information. |
| 789 | 263 */ |
| 264 static int | |
| 1544 | 265 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) |
| 789 | 266 { |
| 267 int error = 0; | |
| 268 nvlist_t *nvroot = NULL; | |
| 269 vdev_t *rvd; | |
| 270 uberblock_t *ub = &spa->spa_uberblock; | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
271 uint64_t config_cache_txg = spa->spa_config_txg; |
| 789 | 272 uint64_t pool_guid; |
| 273 zio_t *zio; | |
| 274 | |
| 1544 | 275 spa->spa_load_state = state; |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
276 |
| 789 | 277 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
278 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid) || |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
279 (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
280 &spa->spa_config_txg) && mosconfig)) { |
| 1544 | 281 error = EINVAL; |
| 282 goto out; | |
| 283 } | |
| 789 | 284 |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
285 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && |
| 1544 | 286 spa_guid_exists(pool_guid, 0)) { |
| 287 error = EEXIST; | |
| 288 goto out; | |
| 289 } | |
| 789 | 290 |
| 291 /* | |
| 292 * Parse the configuration into a vdev tree. | |
| 293 */ | |
| 1544 | 294 spa_config_enter(spa, RW_WRITER, FTAG); |
| 789 | 295 rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_LOAD); |
| 1544 | 296 spa_config_exit(spa, FTAG); |
| 789 | 297 |
| 1544 | 298 if (rvd == NULL) { |
| 299 error = EINVAL; | |
| 300 goto out; | |
| 301 } | |
| 789 | 302 |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
303 ASSERT(spa->spa_root_vdev == rvd); |
| 789 | 304 ASSERT(spa_guid(spa) == pool_guid); |
| 305 | |
| 306 /* | |
| 307 * Try to open all vdevs, loading each label in the process. | |
| 308 */ | |
| 1544 | 309 if (vdev_open(rvd) != 0) { |
| 310 error = ENXIO; | |
| 311 goto out; | |
| 312 } | |
| 789 | 313 |
| 314 /* | |
| 315 * Find the best uberblock. | |
| 316 */ | |
| 317 bzero(ub, sizeof (uberblock_t)); | |
| 318 | |
| 319 zio = zio_root(spa, NULL, NULL, | |
| 320 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
| 321 vdev_uberblock_load(zio, rvd, ub); | |
| 322 error = zio_wait(zio); | |
| 323 | |
| 324 /* | |
| 325 * If we weren't able to find a single valid uberblock, return failure. | |
| 326 */ | |
| 327 if (ub->ub_txg == 0) { | |
| 1544 | 328 error = ENXIO; |
| 329 goto out; | |
| 330 } | |
| 331 | |
| 332 /* | |
| 333 * If the pool is newer than the code, we can't open it. | |
| 334 */ | |
| 335 if (ub->ub_version > UBERBLOCK_VERSION) { | |
| 336 error = ENOTSUP; | |
| 337 goto out; | |
| 789 | 338 } |
| 339 | |
| 340 /* | |
| 341 * If the vdev guid sum doesn't match the uberblock, we have an | |
| 342 * incomplete configuration. | |
| 343 */ | |
| 1732 | 344 if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { |
| 1544 | 345 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, |
| 346 VDEV_AUX_BAD_GUID_SUM); | |
| 347 error = ENXIO; | |
| 348 goto out; | |
| 789 | 349 } |
| 350 | |
| 351 /* | |
| 352 * Initialize internal SPA structures. | |
| 353 */ | |
| 354 spa->spa_state = POOL_STATE_ACTIVE; | |
| 355 spa->spa_ubsync = spa->spa_uberblock; | |
| 356 spa->spa_first_txg = spa_last_synced_txg(spa) + 1; | |
| 1544 | 357 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); |
| 358 if (error) { | |
| 359 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 360 VDEV_AUX_CORRUPT_DATA); | |
| 361 goto out; | |
| 362 } | |
| 789 | 363 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; |
| 364 | |
| 1544 | 365 if (zap_lookup(spa->spa_meta_objset, |
| 789 | 366 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
| 1544 | 367 sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { |
| 368 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 369 VDEV_AUX_CORRUPT_DATA); | |
| 370 error = EIO; | |
| 371 goto out; | |
| 372 } | |
| 789 | 373 |
| 374 if (!mosconfig) { | |
| 375 dmu_buf_t *db; | |
| 376 char *packed = NULL; | |
| 377 size_t nvsize = 0; | |
| 378 nvlist_t *newconfig = NULL; | |
| 379 | |
| 1544 | 380 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, |
| 381 spa->spa_config_object, FTAG, &db)); | |
| 789 | 382 nvsize = *(uint64_t *)db->db_data; |
| 1544 | 383 dmu_buf_rele(db, FTAG); |
| 789 | 384 |
| 385 packed = kmem_alloc(nvsize, KM_SLEEP); | |
| 1544 | 386 error = dmu_read(spa->spa_meta_objset, |
| 789 | 387 spa->spa_config_object, 0, nvsize, packed); |
| 388 if (error == 0) | |
| 389 error = nvlist_unpack(packed, nvsize, &newconfig, 0); | |
| 390 kmem_free(packed, nvsize); | |
| 391 | |
| 1544 | 392 if (error) { |
| 393 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 394 VDEV_AUX_CORRUPT_DATA); | |
| 395 error = EIO; | |
| 396 goto out; | |
| 397 } | |
| 789 | 398 |
| 399 spa_config_set(spa, newconfig); | |
| 400 | |
| 401 spa_unload(spa); | |
| 402 spa_deactivate(spa); | |
| 403 spa_activate(spa); | |
| 404 | |
| 1544 | 405 return (spa_load(spa, newconfig, state, B_TRUE)); |
| 406 } | |
| 407 | |
| 408 if (zap_lookup(spa->spa_meta_objset, | |
| 409 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, | |
| 410 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { | |
| 411 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 412 VDEV_AUX_CORRUPT_DATA); | |
| 413 error = EIO; | |
| 414 goto out; | |
| 789 | 415 } |
| 416 | |
| 1544 | 417 /* |
| 418 * Load the persistent error log. If we have an older pool, this will | |
| 419 * not be present. | |
| 420 */ | |
| 421 error = zap_lookup(spa->spa_meta_objset, | |
| 422 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, | |
| 423 sizeof (uint64_t), 1, &spa->spa_errlog_last); | |
| 424 if (error != 0 &&error != ENOENT) { | |
| 425 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 426 VDEV_AUX_CORRUPT_DATA); | |
| 427 error = EIO; | |
| 428 goto out; | |
| 429 } | |
| 430 | |
| 431 error = zap_lookup(spa->spa_meta_objset, | |
| 432 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, | |
| 433 sizeof (uint64_t), 1, &spa->spa_errlog_scrub); | |
| 434 if (error != 0 && error != ENOENT) { | |
| 435 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, | |
| 436 VDEV_AUX_CORRUPT_DATA); | |
| 437 error = EIO; | |
| 438 goto out; | |
| 439 } | |
| 789 | 440 |
| 441 /* | |
| 1544 | 442 * Load the vdev state for all top level vdevs. We need to grab the |
| 443 * config lock because all label I/O is done with the | |
| 444 * ZIO_FLAG_CONFIG_HELD flag. | |
| 789 | 445 */ |
| 1544 | 446 spa_config_enter(spa, RW_READER, FTAG); |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
447 error = vdev_load(rvd); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
448 spa_config_exit(spa, FTAG); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
449 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
450 if (error) |
| 1544 | 451 goto out; |
| 789 | 452 |
| 453 /* | |
| 454 * Propagate the leaf DTLs we just loaded all the way up the tree. | |
| 455 */ | |
| 1544 | 456 spa_config_enter(spa, RW_WRITER, FTAG); |
| 789 | 457 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); |
| 1544 | 458 spa_config_exit(spa, FTAG); |
| 789 | 459 |
| 460 /* | |
| 461 * Check the state of the root vdev. If it can't be opened, it | |
| 462 * indicates one or more toplevel vdevs are faulted. | |
| 463 */ | |
| 1544 | 464 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { |
| 465 error = ENXIO; | |
| 466 goto out; | |
| 467 } | |
| 789 | 468 |
| 1544 | 469 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
|
470 dmu_tx_t *tx; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
471 int need_update = B_FALSE; |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
472 int c; |
|
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
473 |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
474 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
475 * 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
|
476 * 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
|
477 */ |
|
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
478 tx = dmu_tx_create_assigned(spa_get_dsl(spa), |
| 789 | 479 spa_first_txg(spa)); |
| 480 dmu_objset_find(spa->spa_name, zil_claim, tx, 0); | |
| 481 dmu_tx_commit(tx); | |
| 482 | |
| 483 spa->spa_sync_on = B_TRUE; | |
| 484 txg_sync_start(spa->spa_dsl_pool); | |
| 485 | |
| 486 /* | |
| 487 * Wait for all claims to sync. | |
| 488 */ | |
| 489 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
|
490 |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
491 /* |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
492 * 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
|
493 * 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
|
494 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
495 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
|
496 state == SPA_LOAD_IMPORT) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
497 need_update = B_TRUE; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
498 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
499 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
|
500 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
|
501 need_update = B_TRUE; |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
502 |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
503 /* |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
504 * 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
|
505 * 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
|
506 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
507 if (need_update) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
508 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); |
| 789 | 509 } |
| 510 | |
| 1544 | 511 error = 0; |
| 512 out: | |
| 513 if (error) | |
| 514 zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0); | |
| 515 spa->spa_load_state = SPA_LOAD_NONE; | |
| 516 spa->spa_ena = 0; | |
| 517 | |
| 518 return (error); | |
| 789 | 519 } |
| 520 | |
| 521 /* | |
| 522 * Pool Open/Import | |
| 523 * | |
| 524 * The import case is identical to an open except that the configuration is sent | |
| 525 * down from userland, instead of grabbed from the configuration cache. For the | |
| 526 * case of an open, the pool configuration will exist in the | |
| 527 * POOL_STATE_UNITIALIZED state. | |
| 528 * | |
| 529 * The stats information (gen/count/ustats) is used to gather vdev statistics at | |
| 530 * the same time open the pool, without having to keep around the spa_t in some | |
| 531 * ambiguous state. | |
| 532 */ | |
| 533 static int | |
| 534 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) | |
| 535 { | |
| 536 spa_t *spa; | |
| 537 int error; | |
| 538 int loaded = B_FALSE; | |
| 539 int locked = B_FALSE; | |
| 540 | |
| 541 *spapp = NULL; | |
| 542 | |
| 543 /* | |
| 544 * As disgusting as this is, we need to support recursive calls to this | |
| 545 * function because dsl_dir_open() is called during spa_load(), and ends | |
| 546 * up calling spa_open() again. The real fix is to figure out how to | |
| 547 * avoid dsl_dir_open() calling this in the first place. | |
| 548 */ | |
| 549 if (mutex_owner(&spa_namespace_lock) != curthread) { | |
| 550 mutex_enter(&spa_namespace_lock); | |
| 551 locked = B_TRUE; | |
| 552 } | |
| 553 | |
| 554 if ((spa = spa_lookup(pool)) == NULL) { | |
| 555 if (locked) | |
| 556 mutex_exit(&spa_namespace_lock); | |
| 557 return (ENOENT); | |
| 558 } | |
| 559 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { | |
| 560 | |
| 561 spa_activate(spa); | |
| 562 | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
563 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); |
| 789 | 564 |
| 565 if (error == EBADF) { | |
| 566 /* | |
| 567 * If vdev_load() returns EBADF, it indicates that one | |
| 568 * of the vdevs indicates that the pool has been | |
| 569 * exported or destroyed. If this is the case, the | |
| 570 * config cache is out of sync and we should remove the | |
| 571 * pool from the namespace. | |
| 572 */ | |
| 573 spa_unload(spa); | |
| 574 spa_deactivate(spa); | |
| 575 spa_remove(spa); | |
| 576 spa_config_sync(); | |
| 577 if (locked) | |
| 578 mutex_exit(&spa_namespace_lock); | |
| 579 return (ENOENT); | |
| 1544 | 580 } |
| 581 | |
| 582 if (error) { | |
| 789 | 583 /* |
| 584 * We can't open the pool, but we still have useful | |
| 585 * information: the state of each vdev after the | |
| 586 * attempted vdev_open(). Return this to the user. | |
| 587 */ | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
588 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
|
589 spa_config_enter(spa, RW_READER, FTAG); |
| 789 | 590 *config = spa_config_generate(spa, NULL, -1ULL, |
| 591 B_TRUE); | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
592 spa_config_exit(spa, FTAG); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
593 } |
| 789 | 594 spa_unload(spa); |
| 595 spa_deactivate(spa); | |
| 1544 | 596 spa->spa_last_open_failed = B_TRUE; |
| 789 | 597 if (locked) |
| 598 mutex_exit(&spa_namespace_lock); | |
| 599 *spapp = NULL; | |
| 600 return (error); | |
| 1544 | 601 } else { |
| 602 zfs_post_ok(spa, NULL); | |
| 603 spa->spa_last_open_failed = B_FALSE; | |
| 789 | 604 } |
| 605 | |
| 606 loaded = B_TRUE; | |
| 607 } | |
| 608 | |
| 609 spa_open_ref(spa, tag); | |
| 610 if (locked) | |
| 611 mutex_exit(&spa_namespace_lock); | |
| 612 | |
| 613 *spapp = spa; | |
| 614 | |
| 615 if (config != NULL) { | |
| 1544 | 616 spa_config_enter(spa, RW_READER, FTAG); |
| 789 | 617 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); |
| 1544 | 618 spa_config_exit(spa, FTAG); |
| 789 | 619 } |
| 620 | |
| 621 /* | |
| 622 * If we just loaded the pool, resilver anything that's out of date. | |
| 623 */ | |
| 624 if (loaded && (spa_mode & FWRITE)) | |
| 625 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
| 626 | |
| 627 return (0); | |
| 628 } | |
| 629 | |
| 630 int | |
| 631 spa_open(const char *name, spa_t **spapp, void *tag) | |
| 632 { | |
| 633 return (spa_open_common(name, spapp, tag, NULL)); | |
| 634 } | |
| 635 | |
| 1544 | 636 /* |
| 637 * Lookup the given spa_t, incrementing the inject count in the process, | |
| 638 * preventing it from being exported or destroyed. | |
| 639 */ | |
| 640 spa_t * | |
| 641 spa_inject_addref(char *name) | |
| 642 { | |
| 643 spa_t *spa; | |
| 644 | |
| 645 mutex_enter(&spa_namespace_lock); | |
| 646 if ((spa = spa_lookup(name)) == NULL) { | |
| 647 mutex_exit(&spa_namespace_lock); | |
| 648 return (NULL); | |
| 649 } | |
| 650 spa->spa_inject_ref++; | |
| 651 mutex_exit(&spa_namespace_lock); | |
| 652 | |
| 653 return (spa); | |
| 654 } | |
| 655 | |
| 656 void | |
| 657 spa_inject_delref(spa_t *spa) | |
| 658 { | |
| 659 mutex_enter(&spa_namespace_lock); | |
| 660 spa->spa_inject_ref--; | |
| 661 mutex_exit(&spa_namespace_lock); | |
| 662 } | |
| 663 | |
| 789 | 664 int |
| 1544 | 665 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) |
| 789 | 666 { |
| 667 int error; | |
| 668 spa_t *spa; | |
| 669 | |
| 670 *config = NULL; | |
| 671 error = spa_open_common(name, &spa, FTAG, config); | |
| 672 | |
| 1544 | 673 if (spa && *config != NULL) |
| 674 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, | |
| 675 spa_get_errlog_size(spa)) == 0); | |
| 676 | |
| 677 /* | |
| 678 * We want to get the alternate root even for faulted pools, so we cheat | |
| 679 * and call spa_lookup() directly. | |
| 680 */ | |
| 681 if (altroot) { | |
| 682 if (spa == NULL) { | |
| 683 mutex_enter(&spa_namespace_lock); | |
| 684 spa = spa_lookup(name); | |
| 685 if (spa) | |
| 686 spa_altroot(spa, altroot, buflen); | |
| 687 else | |
| 688 altroot[0] = '\0'; | |
| 689 spa = NULL; | |
| 690 mutex_exit(&spa_namespace_lock); | |
| 691 } else { | |
| 692 spa_altroot(spa, altroot, buflen); | |
| 693 } | |
| 694 } | |
| 695 | |
| 789 | 696 if (spa != NULL) |
| 697 spa_close(spa, FTAG); | |
| 698 | |
| 699 return (error); | |
| 700 } | |
| 701 | |
| 702 /* | |
| 703 * Pool Creation | |
| 704 */ | |
| 705 int | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
706 spa_create(const char *pool, nvlist_t *nvroot, const char *altroot) |
| 789 | 707 { |
| 708 spa_t *spa; | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
709 vdev_t *rvd; |
| 789 | 710 dsl_pool_t *dp; |
| 711 dmu_tx_t *tx; | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
712 int c, error; |
| 789 | 713 uint64_t txg = TXG_INITIAL; |
| 714 | |
| 715 /* | |
| 716 * If this pool already exists, return failure. | |
| 717 */ | |
| 718 mutex_enter(&spa_namespace_lock); | |
| 719 if (spa_lookup(pool) != NULL) { | |
| 720 mutex_exit(&spa_namespace_lock); | |
| 721 return (EEXIST); | |
| 722 } | |
| 723 | |
| 724 /* | |
| 725 * Allocate a new spa_t structure. | |
| 726 */ | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
727 spa = spa_add(pool, altroot); |
| 789 | 728 spa_activate(spa); |
| 729 | |
| 730 spa->spa_uberblock.ub_txg = txg - 1; | |
| 731 spa->spa_ubsync = spa->spa_uberblock; | |
| 732 | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
733 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
734 * Create the root vdev. |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
735 */ |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
736 spa_config_enter(spa, RW_WRITER, FTAG); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
737 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
738 rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
739 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
740 ASSERT(spa->spa_root_vdev == rvd); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
741 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
742 if (rvd == NULL) { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
743 error = EINVAL; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
744 } else { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
745 if ((error = vdev_create(rvd, txg)) == 0) { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
746 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
|
747 vdev_init(rvd->vdev_child[c], txg); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
748 vdev_config_dirty(rvd); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
749 } |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
750 } |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
751 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
752 spa_config_exit(spa, FTAG); |
| 789 | 753 |
| 754 if (error) { | |
| 755 spa_unload(spa); | |
| 756 spa_deactivate(spa); | |
| 757 spa_remove(spa); | |
| 758 mutex_exit(&spa_namespace_lock); | |
| 759 return (error); | |
| 760 } | |
| 761 | |
| 762 spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg); | |
| 763 spa->spa_meta_objset = dp->dp_meta_objset; | |
| 764 | |
| 765 tx = dmu_tx_create_assigned(dp, txg); | |
| 766 | |
| 767 /* | |
| 768 * Create the pool config object. | |
| 769 */ | |
| 770 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, | |
| 771 DMU_OT_PACKED_NVLIST, 1 << 14, | |
| 772 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); | |
| 773 | |
| 1544 | 774 if (zap_add(spa->spa_meta_objset, |
| 789 | 775 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, |
| 1544 | 776 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { |
| 777 cmn_err(CE_PANIC, "failed to add pool config"); | |
| 778 } | |
| 789 | 779 |
| 780 /* | |
| 781 * Create the deferred-free bplist object. Turn off compression | |
| 782 * because sync-to-convergence takes longer if the blocksize | |
| 783 * keeps changing. | |
| 784 */ | |
| 785 spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, | |
| 786 1 << 14, tx); | |
| 787 dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, | |
| 788 ZIO_COMPRESS_OFF, tx); | |
| 789 | |
| 1544 | 790 if (zap_add(spa->spa_meta_objset, |
| 789 | 791 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, |
| 1544 | 792 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { |
| 793 cmn_err(CE_PANIC, "failed to add bplist"); | |
| 794 } | |
| 789 | 795 |
| 796 dmu_tx_commit(tx); | |
| 797 | |
| 798 spa->spa_sync_on = B_TRUE; | |
| 799 txg_sync_start(spa->spa_dsl_pool); | |
| 800 | |
| 801 /* | |
| 802 * We explicitly wait for the first transaction to complete so that our | |
| 803 * bean counters are appropriately updated. | |
| 804 */ | |
| 805 txg_wait_synced(spa->spa_dsl_pool, txg); | |
| 806 | |
| 807 spa_config_sync(); | |
| 808 | |
| 809 mutex_exit(&spa_namespace_lock); | |
| 810 | |
| 811 return (0); | |
| 812 } | |
| 813 | |
| 814 /* | |
| 815 * Import the given pool into the system. We set up the necessary spa_t and | |
| 816 * then call spa_load() to do the dirty work. | |
| 817 */ | |
| 818 int | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
819 spa_import(const char *pool, nvlist_t *config, const char *altroot) |
| 789 | 820 { |
| 821 spa_t *spa; | |
| 822 int error; | |
| 823 | |
| 824 if (!(spa_mode & FWRITE)) | |
| 825 return (EROFS); | |
| 826 | |
| 827 /* | |
| 828 * If a pool with this name exists, return failure. | |
| 829 */ | |
| 830 mutex_enter(&spa_namespace_lock); | |
| 831 if (spa_lookup(pool) != NULL) { | |
| 832 mutex_exit(&spa_namespace_lock); | |
| 833 return (EEXIST); | |
| 834 } | |
| 835 | |
| 836 /* | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
837 * Create and initialize the spa structure. |
| 789 | 838 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
839 spa = spa_add(pool, altroot); |
| 789 | 840 spa_activate(spa); |
| 841 | |
| 842 /* | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
843 * Pass off the heavy lifting to spa_load(). |
| 1732 | 844 * Pass TRUE for mosconfig because the user-supplied config |
| 845 * 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
|
846 */ |
| 1732 | 847 error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE); |
| 789 | 848 |
| 849 if (error) { | |
| 850 spa_unload(spa); | |
| 851 spa_deactivate(spa); | |
| 852 spa_remove(spa); | |
| 853 mutex_exit(&spa_namespace_lock); | |
| 854 return (error); | |
| 855 } | |
| 856 | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
857 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
858 * 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
|
859 */ |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
860 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
|
861 |
| 789 | 862 mutex_exit(&spa_namespace_lock); |
| 863 | |
| 864 /* | |
| 865 * Resilver anything that's out of date. | |
| 866 */ | |
| 867 if (spa_mode & FWRITE) | |
| 868 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
| 869 | |
| 870 return (0); | |
| 871 } | |
| 872 | |
| 873 /* | |
| 874 * This (illegal) pool name is used when temporarily importing a spa_t in order | |
| 875 * to get the vdev stats associated with the imported devices. | |
| 876 */ | |
| 877 #define TRYIMPORT_NAME "$import" | |
| 878 | |
| 879 nvlist_t * | |
| 880 spa_tryimport(nvlist_t *tryconfig) | |
| 881 { | |
| 882 nvlist_t *config = NULL; | |
| 883 char *poolname; | |
| 884 spa_t *spa; | |
| 885 uint64_t state; | |
| 886 | |
| 887 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) | |
| 888 return (NULL); | |
| 889 | |
| 890 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) | |
| 891 return (NULL); | |
| 892 | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
893 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
894 * Create and initialize the spa structure. |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
895 */ |
| 789 | 896 mutex_enter(&spa_namespace_lock); |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
897 spa = spa_add(TRYIMPORT_NAME, NULL); |
| 789 | 898 spa_activate(spa); |
| 899 | |
| 900 /* | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
901 * Pass off the heavy lifting to spa_load(). |
| 1732 | 902 * Pass TRUE for mosconfig because the user-supplied config |
| 903 * is actually the one to trust when doing an import. | |
| 789 | 904 */ |
| 1732 | 905 (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); |
| 789 | 906 |
| 907 /* | |
| 908 * If 'tryconfig' was at least parsable, return the current config. | |
| 909 */ | |
| 910 if (spa->spa_root_vdev != NULL) { | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
911 spa_config_enter(spa, RW_READER, FTAG); |
| 789 | 912 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
|
913 spa_config_exit(spa, FTAG); |
| 789 | 914 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, |
| 915 poolname) == 0); | |
| 916 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, | |
| 917 state) == 0); | |
| 918 } | |
| 919 | |
| 920 spa_unload(spa); | |
| 921 spa_deactivate(spa); | |
| 922 spa_remove(spa); | |
| 923 mutex_exit(&spa_namespace_lock); | |
| 924 | |
| 925 return (config); | |
| 926 } | |
| 927 | |
| 928 /* | |
| 929 * Pool export/destroy | |
| 930 * | |
| 931 * The act of destroying or exporting a pool is very simple. We make sure there | |
| 932 * is no more pending I/O and any references to the pool are gone. Then, we | |
| 933 * update the pool state and sync all the labels to disk, removing the | |
| 934 * configuration from the cache afterwards. | |
| 935 */ | |
| 936 static int | |
| 937 spa_export_common(char *pool, int new_state) | |
| 938 { | |
| 939 spa_t *spa; | |
| 940 | |
| 941 if (!(spa_mode & FWRITE)) | |
| 942 return (EROFS); | |
| 943 | |
| 944 mutex_enter(&spa_namespace_lock); | |
| 945 if ((spa = spa_lookup(pool)) == NULL) { | |
| 946 mutex_exit(&spa_namespace_lock); | |
| 947 return (ENOENT); | |
| 948 } | |
| 949 | |
| 950 /* | |
| 1544 | 951 * Put a hold on the pool, drop the namespace lock, stop async tasks, |
| 952 * reacquire the namespace lock, and see if we can export. | |
| 953 */ | |
| 954 spa_open_ref(spa, FTAG); | |
| 955 mutex_exit(&spa_namespace_lock); | |
| 956 spa_async_suspend(spa); | |
| 957 mutex_enter(&spa_namespace_lock); | |
| 958 spa_close(spa, FTAG); | |
| 959 | |
| 960 /* | |
| 789 | 961 * The pool will be in core if it's openable, |
| 962 * in which case we can modify its state. | |
| 963 */ | |
| 964 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { | |
| 965 /* | |
| 966 * Objsets may be open only because they're dirty, so we | |
| 967 * have to force it to sync before checking spa_refcnt. | |
| 968 */ | |
| 969 spa_scrub_suspend(spa); | |
| 970 txg_wait_synced(spa->spa_dsl_pool, 0); | |
| 971 | |
| 1544 | 972 /* |
| 973 * A pool cannot be exported or destroyed if there are active | |
| 974 * references. If we are resetting a pool, allow references by | |
| 975 * fault injection handlers. | |
| 976 */ | |
| 977 if (!spa_refcount_zero(spa) || | |
| 978 (spa->spa_inject_ref != 0 && | |
| 979 new_state != POOL_STATE_UNINITIALIZED)) { | |
| 789 | 980 spa_scrub_resume(spa); |
| 1544 | 981 spa_async_resume(spa); |
| 789 | 982 mutex_exit(&spa_namespace_lock); |
| 983 return (EBUSY); | |
| 984 } | |
| 985 | |
| 986 spa_scrub_resume(spa); | |
| 987 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); | |
| 988 | |
| 989 /* | |
| 990 * We want this to be reflected on every label, | |
| 991 * so mark them all dirty. spa_unload() will do the | |
| 992 * final sync that pushes these changes out. | |
| 993 */ | |
| 1544 | 994 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
|
995 spa_config_enter(spa, RW_WRITER, FTAG); |
| 1544 | 996 spa->spa_state = new_state; |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
997 spa->spa_final_txg = spa_last_synced_txg(spa) + 1; |
| 1544 | 998 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
|
999 spa_config_exit(spa, FTAG); |
| 1544 | 1000 } |
| 789 | 1001 } |
| 1002 | |
| 1003 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
| 1004 spa_unload(spa); | |
| 1005 spa_deactivate(spa); | |
| 1006 } | |
| 1007 | |
| 1544 | 1008 if (new_state != POOL_STATE_UNINITIALIZED) { |
| 1009 spa_remove(spa); | |
| 1010 spa_config_sync(); | |
| 1011 } | |
| 789 | 1012 mutex_exit(&spa_namespace_lock); |
| 1013 | |
| 1014 return (0); | |
| 1015 } | |
| 1016 | |
| 1017 /* | |
| 1018 * Destroy a storage pool. | |
| 1019 */ | |
| 1020 int | |
| 1021 spa_destroy(char *pool) | |
| 1022 { | |
| 1023 return (spa_export_common(pool, POOL_STATE_DESTROYED)); | |
| 1024 } | |
| 1025 | |
| 1026 /* | |
| 1027 * Export a storage pool. | |
| 1028 */ | |
| 1029 int | |
| 1030 spa_export(char *pool) | |
| 1031 { | |
| 1032 return (spa_export_common(pool, POOL_STATE_EXPORTED)); | |
| 1033 } | |
| 1034 | |
| 1035 /* | |
| 1544 | 1036 * Similar to spa_export(), this unloads the spa_t without actually removing it |
| 1037 * from the namespace in any way. | |
| 1038 */ | |
| 1039 int | |
| 1040 spa_reset(char *pool) | |
| 1041 { | |
| 1042 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED)); | |
| 1043 } | |
| 1044 | |
| 1045 | |
| 1046 /* | |
| 789 | 1047 * ========================================================================== |
| 1048 * Device manipulation | |
| 1049 * ========================================================================== | |
| 1050 */ | |
| 1051 | |
| 1052 /* | |
| 1053 * Add capacity to a storage pool. | |
| 1054 */ | |
| 1055 int | |
| 1056 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) | |
| 1057 { | |
| 1058 uint64_t txg; | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1059 int c, error; |
| 789 | 1060 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
|
1061 vdev_t *vd, *tvd; |
| 789 | 1062 |
| 1063 txg = spa_vdev_enter(spa); | |
| 1064 | |
| 1065 vd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
| 1066 | |
| 1067 if (vd == NULL) | |
| 1068 return (spa_vdev_exit(spa, vd, txg, EINVAL)); | |
| 1069 | |
| 1070 if ((error = vdev_create(vd, txg)) != 0) | |
| 1071 return (spa_vdev_exit(spa, vd, txg, error)); | |
| 1072 | |
| 1073 /* | |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1074 * Transfer each new top-level vdev from vd to rvd. |
| 789 | 1075 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1076 for (c = 0; c < vd->vdev_children; c++) { |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1077 tvd = vd->vdev_child[c]; |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1078 vdev_remove_child(vd, tvd); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1079 tvd->vdev_id = rvd->vdev_children; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1080 vdev_add_child(rvd, tvd); |
| 789 | 1081 vdev_config_dirty(tvd); |
| 1082 } | |
| 1083 | |
| 1084 /* | |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1085 * 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
|
1086 * 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
|
1087 * 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
|
1088 * 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
|
1089 * 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
|
1090 * 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
|
1091 * 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
|
1092 * |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1093 * 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
|
1094 * 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
|
1095 * steps will be completed the next time we load the pool. |
| 789 | 1096 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1097 (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
|
1098 |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1099 mutex_enter(&spa_namespace_lock); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1100 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
|
1101 mutex_exit(&spa_namespace_lock); |
| 789 | 1102 |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1103 return (0); |
| 789 | 1104 } |
| 1105 | |
| 1106 /* | |
| 1107 * Attach a device to a mirror. The arguments are the path to any device | |
| 1108 * in the mirror, and the nvroot for the new device. If the path specifies | |
| 1109 * a device that is not mirrored, we automatically insert the mirror vdev. | |
| 1110 * | |
| 1111 * If 'replacing' is specified, the new device is intended to replace the | |
| 1112 * existing device; in this case the two devices are made into their own | |
| 1113 * mirror using the 'replacing' vdev, which is functionally idendical to | |
| 1114 * the mirror vdev (it actually reuses all the same ops) but has a few | |
| 1115 * extra rules: you can't attach to it after it's been created, and upon | |
| 1116 * completion of resilvering, the first disk (the one being replaced) | |
| 1117 * is automatically detached. | |
| 1118 */ | |
| 1119 int | |
| 1544 | 1120 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) |
| 789 | 1121 { |
| 1122 uint64_t txg, open_txg; | |
| 1123 int error; | |
| 1124 vdev_t *rvd = spa->spa_root_vdev; | |
| 1125 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; | |
| 1126 vdev_ops_t *pvops = replacing ? &vdev_replacing_ops : &vdev_mirror_ops; | |
| 1127 | |
| 1128 txg = spa_vdev_enter(spa); | |
| 1129 | |
| 1544 | 1130 oldvd = vdev_lookup_by_guid(rvd, guid); |
| 789 | 1131 |
| 1132 if (oldvd == NULL) | |
| 1133 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
| 1134 | |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1135 if (!oldvd->vdev_ops->vdev_op_leaf) |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1136 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
|
1137 |
| 789 | 1138 pvd = oldvd->vdev_parent; |
| 1139 | |
| 1140 /* | |
| 1141 * The parent must be a mirror or the root, unless we're replacing; | |
| 1142 * in that case, the parent can be anything but another replacing vdev. | |
| 1143 */ | |
| 1144 if (pvd->vdev_ops != &vdev_mirror_ops && | |
| 1145 pvd->vdev_ops != &vdev_root_ops && | |
| 1146 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) | |
| 1147 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
| 1148 | |
| 1149 newrootvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD); | |
| 1150 | |
| 1151 if (newrootvd == NULL || newrootvd->vdev_children != 1) | |
| 1152 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
| 1153 | |
| 1154 newvd = newrootvd->vdev_child[0]; | |
| 1155 | |
| 1156 if (!newvd->vdev_ops->vdev_op_leaf) | |
| 1157 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); | |
| 1158 | |
| 1159 if ((error = vdev_create(newrootvd, txg)) != 0) | |
| 1160 return (spa_vdev_exit(spa, newrootvd, txg, error)); | |
| 1161 | |
|
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
|
1162 /* |
|
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
|
1163 * 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
|
1164 * 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
|
1165 */ |
|
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
|
1166 if (newvd->vdev_psize < vdev_get_rsize(oldvd)) |
| 789 | 1167 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); |
| 1168 | |
| 1732 | 1169 /* |
| 1170 * The new device cannot have a higher alignment requirement | |
| 1171 * than the top-level vdev. | |
| 1172 */ | |
| 1173 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) | |
| 789 | 1174 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); |
| 1175 | |
| 1176 /* | |
| 1177 * If this is an in-place replacement, update oldvd's path and devid | |
| 1178 * to make it distinguishable from newvd, and unopenable from now on. | |
| 1179 */ | |
| 1180 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { | |
| 1181 spa_strfree(oldvd->vdev_path); | |
| 1182 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, | |
| 1183 KM_SLEEP); | |
| 1184 (void) sprintf(oldvd->vdev_path, "%s/%s", | |
| 1185 newvd->vdev_path, "old"); | |
| 1186 if (oldvd->vdev_devid != NULL) { | |
| 1187 spa_strfree(oldvd->vdev_devid); | |
| 1188 oldvd->vdev_devid = NULL; | |
| 1189 } | |
| 1190 } | |
| 1191 | |
| 1192 /* | |
| 1193 * If the parent is not a mirror, or if we're replacing, | |
| 1194 * insert the new mirror/replacing vdev above oldvd. | |
| 1195 */ | |
| 1196 if (pvd->vdev_ops != pvops) | |
| 1197 pvd = vdev_add_parent(oldvd, pvops); | |
| 1198 | |
| 1199 ASSERT(pvd->vdev_top->vdev_parent == rvd); | |
| 1200 ASSERT(pvd->vdev_ops == pvops); | |
| 1201 ASSERT(oldvd->vdev_parent == pvd); | |
| 1202 | |
| 1203 /* | |
| 1204 * Extract the new device from its root and add it to pvd. | |
| 1205 */ | |
| 1206 vdev_remove_child(newrootvd, newvd); | |
| 1207 newvd->vdev_id = pvd->vdev_children; | |
| 1208 vdev_add_child(pvd, newvd); | |
| 1209 | |
| 1544 | 1210 /* |
| 1211 * If newvd is smaller than oldvd, but larger than its rsize, | |
| 1212 * the addition of newvd may have decreased our parent's asize. | |
| 1213 */ | |
| 1214 pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); | |
| 1215 | |
| 789 | 1216 tvd = newvd->vdev_top; |
| 1217 ASSERT(pvd->vdev_top == tvd); | |
| 1218 ASSERT(tvd->vdev_parent == rvd); | |
| 1219 | |
| 1220 vdev_config_dirty(tvd); | |
| 1221 | |
| 1222 /* | |
| 1223 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate | |
| 1224 * upward when spa_vdev_exit() calls vdev_dtl_reassess(). | |
| 1225 */ | |
| 1226 open_txg = txg + TXG_CONCURRENT_STATES - 1; | |
| 1227 | |
| 1228 mutex_enter(&newvd->vdev_dtl_lock); | |
| 1229 space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, | |
| 1230 open_txg - TXG_INITIAL + 1); | |
| 1231 mutex_exit(&newvd->vdev_dtl_lock); | |
| 1232 | |
| 1544 | 1233 dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg); |
| 1234 | |
| 789 | 1235 /* |
| 1236 * Mark newvd's DTL dirty in this txg. | |
| 1237 */ | |
| 1732 | 1238 vdev_dirty(tvd, VDD_DTL, newvd, txg); |
| 789 | 1239 |
| 1240 (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); | |
| 1241 | |
| 1242 /* | |
| 1243 * Kick off a resilver to update newvd. | |
| 1244 */ | |
| 1245 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
| 1246 | |
| 1247 return (0); | |
| 1248 } | |
| 1249 | |
| 1250 /* | |
| 1251 * Detach a device from a mirror or replacing vdev. | |
| 1252 * If 'replace_done' is specified, only detach if the parent | |
| 1253 * is a replacing vdev. | |
| 1254 */ | |
| 1255 int | |
| 1544 | 1256 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) |
| 789 | 1257 { |
| 1258 uint64_t txg; | |
| 1259 int c, t, error; | |
| 1260 vdev_t *rvd = spa->spa_root_vdev; | |
| 1261 vdev_t *vd, *pvd, *cvd, *tvd; | |
| 1262 | |
| 1263 txg = spa_vdev_enter(spa); | |
| 1264 | |
| 1544 | 1265 vd = vdev_lookup_by_guid(rvd, guid); |
| 789 | 1266 |
| 1267 if (vd == NULL) | |
| 1268 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); | |
| 1269 | |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1270 if (!vd->vdev_ops->vdev_op_leaf) |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1271 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
|
1272 |
| 789 | 1273 pvd = vd->vdev_parent; |
| 1274 | |
| 1275 /* | |
| 1276 * If replace_done is specified, only remove this device if it's | |
| 1277 * the first child of a replacing vdev. | |
| 1278 */ | |
| 1279 if (replace_done && | |
| 1280 (vd->vdev_id != 0 || pvd->vdev_ops != &vdev_replacing_ops)) | |
| 1281 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
| 1282 | |
| 1283 /* | |
| 1284 * Only mirror and replacing vdevs support detach. | |
| 1285 */ | |
| 1286 if (pvd->vdev_ops != &vdev_replacing_ops && | |
| 1287 pvd->vdev_ops != &vdev_mirror_ops) | |
| 1288 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); | |
| 1289 | |
| 1290 /* | |
| 1291 * If there's only one replica, you can't detach it. | |
| 1292 */ | |
| 1293 if (pvd->vdev_children <= 1) | |
| 1294 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
| 1295 | |
| 1296 /* | |
| 1297 * If all siblings have non-empty DTLs, this device may have the only | |
| 1298 * valid copy of the data, which means we cannot safely detach it. | |
| 1299 * | |
| 1300 * XXX -- as in the vdev_offline() case, we really want a more | |
| 1301 * precise DTL check. | |
| 1302 */ | |
| 1303 for (c = 0; c < pvd->vdev_children; c++) { | |
| 1304 uint64_t dirty; | |
| 1305 | |
| 1306 cvd = pvd->vdev_child[c]; | |
| 1307 if (cvd == vd) | |
| 1308 continue; | |
| 1309 if (vdev_is_dead(cvd)) | |
| 1310 continue; | |
| 1311 mutex_enter(&cvd->vdev_dtl_lock); | |
| 1312 dirty = cvd->vdev_dtl_map.sm_space | | |
| 1313 cvd->vdev_dtl_scrub.sm_space; | |
| 1314 mutex_exit(&cvd->vdev_dtl_lock); | |
| 1315 if (!dirty) | |
| 1316 break; | |
| 1317 } | |
| 1318 if (c == pvd->vdev_children) | |
| 1319 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); | |
| 1320 | |
| 1321 /* | |
| 1322 * Erase the disk labels so the disk can be used for other things. | |
| 1323 * This must be done after all other error cases are handled, | |
| 1324 * but before we disembowel vd (so we can still do I/O to it). | |
| 1325 * But if we can't do it, don't treat the error as fatal -- | |
| 1326 * it may be that the unwritability of the disk is the reason | |
| 1327 * it's being detached! | |
| 1328 */ | |
| 1329 error = vdev_label_init(vd, 0); | |
| 1330 if (error) | |
| 1331 dprintf("unable to erase labels on %s\n", vdev_description(vd)); | |
| 1332 | |
| 1333 /* | |
| 1334 * Remove vd from its parent and compact the parent's children. | |
| 1335 */ | |
| 1336 vdev_remove_child(pvd, vd); | |
| 1337 vdev_compact_children(pvd); | |
| 1338 | |
| 1339 /* | |
| 1340 * Remember one of the remaining children so we can get tvd below. | |
| 1341 */ | |
| 1342 cvd = pvd->vdev_child[0]; | |
| 1343 | |
| 1344 /* | |
| 1345 * If the parent mirror/replacing vdev only has one child, | |
| 1346 * the parent is no longer needed. Remove it from the tree. | |
| 1347 */ | |
| 1348 if (pvd->vdev_children == 1) | |
| 1349 vdev_remove_parent(cvd); | |
| 1350 | |
| 1351 /* | |
| 1352 * We don't set tvd until now because the parent we just removed | |
| 1353 * may have been the previous top-level vdev. | |
| 1354 */ | |
| 1355 tvd = cvd->vdev_top; | |
| 1356 ASSERT(tvd->vdev_parent == rvd); | |
| 1357 | |
| 1358 /* | |
| 1359 * Reopen this top-level vdev to reassess health after detach. | |
| 1360 */ | |
| 1544 | 1361 vdev_reopen(tvd); |
| 789 | 1362 |
| 1363 /* | |
| 1364 * If the device we just detached was smaller than the others, | |
| 1732 | 1365 * it may be possible to add metaslabs (i.e. grow the pool). |
| 1366 * vdev_metaslab_init() can't fail because the existing metaslabs | |
| 1367 * are already in core, so there's nothing to read from disk. | |
| 789 | 1368 */ |
| 1732 | 1369 VERIFY(vdev_metaslab_init(tvd, txg) == 0); |
| 789 | 1370 |
| 1371 vdev_config_dirty(tvd); | |
| 1372 | |
| 1373 /* | |
| 1374 * Mark vd's DTL as dirty in this txg. | |
| 1375 * vdev_dtl_sync() will see that vd->vdev_detached is set | |
| 1376 * and free vd's DTL object in syncing context. | |
| 1377 * But first make sure we're not on any *other* txg's DTL list, | |
| 1378 * to prevent vd from being accessed after it's freed. | |
| 1379 */ | |
| 1380 for (t = 0; t < TXG_SIZE; t++) | |
| 1381 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); | |
| 1732 | 1382 vd->vdev_detached = B_TRUE; |
| 1383 vdev_dirty(tvd, VDD_DTL, vd, txg); | |
| 789 | 1384 |
| 1544 | 1385 dprintf("detached %s in txg %llu\n", vd->vdev_path, txg); |
| 789 | 1386 |
| 1387 return (spa_vdev_exit(spa, vd, txg, 0)); | |
| 1388 } | |
| 1389 | |
| 1390 /* | |
| 1544 | 1391 * Find any device that's done replacing, so we can detach it. |
| 789 | 1392 */ |
| 1544 | 1393 static vdev_t * |
| 1394 spa_vdev_replace_done_hunt(vdev_t *vd) | |
| 789 | 1395 { |
| 1544 | 1396 vdev_t *newvd, *oldvd; |
| 789 | 1397 int c; |
| 1398 | |
| 1544 | 1399 for (c = 0; c < vd->vdev_children; c++) { |
| 1400 oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]); | |
| 1401 if (oldvd != NULL) | |
| 1402 return (oldvd); | |
| 1403 } | |
| 789 | 1404 |
| 1405 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { | |
| 1544 | 1406 oldvd = vd->vdev_child[0]; |
| 1407 newvd = vd->vdev_child[1]; | |
| 789 | 1408 |
| 1544 | 1409 mutex_enter(&newvd->vdev_dtl_lock); |
| 1410 if (newvd->vdev_dtl_map.sm_space == 0 && | |
| 1411 newvd->vdev_dtl_scrub.sm_space == 0) { | |
| 1412 mutex_exit(&newvd->vdev_dtl_lock); | |
| 1413 return (oldvd); | |
| 1414 } | |
| 1415 mutex_exit(&newvd->vdev_dtl_lock); | |
| 1416 } | |
| 789 | 1417 |
| 1544 | 1418 return (NULL); |
| 789 | 1419 } |
| 1420 | |
| 1544 | 1421 static void |
| 789 | 1422 spa_vdev_replace_done(spa_t *spa) |
| 1423 { | |
| 1544 | 1424 vdev_t *vd; |
| 1425 uint64_t guid; | |
| 789 | 1426 |
| 1544 | 1427 spa_config_enter(spa, RW_READER, FTAG); |
| 789 | 1428 |
| 1544 | 1429 while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) { |
| 1430 guid = vd->vdev_guid; | |
| 1431 spa_config_exit(spa, FTAG); | |
| 1432 if (spa_vdev_detach(spa, guid, B_TRUE) != 0) | |
| 1433 return; | |
| 1434 spa_config_enter(spa, RW_READER, FTAG); | |
| 789 | 1435 } |
| 1436 | |
| 1544 | 1437 spa_config_exit(spa, FTAG); |
| 789 | 1438 } |
| 1439 | |
| 1440 /* | |
|
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1441 * 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
|
1442 * on spa_vdev_enter/exit() to synchronize the labels and cache. |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1443 */ |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1444 int |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1445 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
|
1446 { |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1447 vdev_t *rvd, *vd; |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1448 uint64_t txg; |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1449 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1450 rvd = spa->spa_root_vdev; |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1451 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1452 txg = spa_vdev_enter(spa); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1453 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1454 if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1455 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1456 |
|
1585
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1457 if (!vd->vdev_ops->vdev_op_leaf) |
|
4ad213e858a9
6395480 ztest ASSERT: rbt.bt_objset == wbt.bt_objset, line 2041
bonwick
parents:
1544
diff
changeset
|
1458 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
|
1459 |
|
1354
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1460 spa_strfree(vd->vdev_path); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1461 vd->vdev_path = spa_strdup(newpath); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1462 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1463 vdev_config_dirty(vd->vdev_top); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1464 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1465 return (spa_vdev_exit(spa, NULL, txg, 0)); |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1466 } |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1467 |
|
81359ee1ee63
6362672 import gets confused about overlapping slices
eschrock
parents:
1175
diff
changeset
|
1468 /* |
| 789 | 1469 * ========================================================================== |
| 1470 * SPA Scrubbing | |
| 1471 * ========================================================================== | |
| 1472 */ | |
| 1473 | |
| 1544 | 1474 void |
| 1475 spa_scrub_throttle(spa_t *spa, int direction) | |
| 1476 { | |
| 1477 mutex_enter(&spa->spa_scrub_lock); | |
| 1478 spa->spa_scrub_throttled += direction; | |
| 1479 ASSERT(spa->spa_scrub_throttled >= 0); | |
| 1480 if (spa->spa_scrub_throttled == 0) | |
| 1481 cv_broadcast(&spa->spa_scrub_io_cv); | |
| 1482 mutex_exit(&spa->spa_scrub_lock); | |
| 1483 } | |
| 789 | 1484 |
| 1485 static void | |
| 1486 spa_scrub_io_done(zio_t *zio) | |
| 1487 { | |
| 1488 spa_t *spa = zio->io_spa; | |
| 1489 | |
| 1490 zio_buf_free(zio->io_data, zio->io_size); | |
| 1491 | |
| 1492 mutex_enter(&spa->spa_scrub_lock); | |
| 1544 | 1493 if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { |
| 1494 vdev_t *vd = zio->io_vd; | |
| 789 | 1495 spa->spa_scrub_errors++; |
| 1496 mutex_enter(&vd->vdev_stat_lock); | |
| 1497 vd->vdev_stat.vs_scrub_errors++; | |
| 1498 mutex_exit(&vd->vdev_stat_lock); | |
| 1499 } | |
| 1544 | 1500 if (--spa->spa_scrub_inflight == 0) { |
| 1501 cv_broadcast(&spa->spa_scrub_io_cv); | |
| 1502 ASSERT(spa->spa_scrub_throttled == 0); | |
| 1503 } | |
| 1504 mutex_exit(&spa->spa_scrub_lock); | |
| 789 | 1505 } |
| 1506 | |
| 1507 static void | |
| 1544 | 1508 spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags, |
| 1509 zbookmark_t *zb) | |
| 789 | 1510 { |
| 1511 size_t size = BP_GET_LSIZE(bp); | |
| 1512 void *data = zio_buf_alloc(size); | |
| 1513 | |
| 1514 mutex_enter(&spa->spa_scrub_lock); | |
| 1515 spa->spa_scrub_inflight++; | |
| 1516 mutex_exit(&spa->spa_scrub_lock); | |
| 1517 | |
| 1544 | 1518 if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET) |
| 1519 flags |= ZIO_FLAG_SPECULATIVE; /* intent log block */ | |
| 1520 | |
| 1521 flags |= ZIO_FLAG_CANFAIL; | |
| 1522 | |
| 789 | 1523 zio_nowait(zio_read(NULL, spa, bp, data, size, |
| 1544 | 1524 spa_scrub_io_done, NULL, priority, flags, zb)); |
| 789 | 1525 } |
| 1526 | |
| 1527 /* ARGSUSED */ | |
| 1528 static int | |
| 1529 spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a) | |
| 1530 { | |
| 1531 blkptr_t *bp = &bc->bc_blkptr; | |
| 1532 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[0])); | |
| 1533 | |
| 1534 if (bc->bc_errno || vd == NULL) { | |
| 1535 /* | |
| 1536 * We can't scrub this block, but we can continue to scrub | |
| 1537 * the rest of the pool. Note the error and move along. | |
| 1538 */ | |
| 1539 mutex_enter(&spa->spa_scrub_lock); | |
| 1540 spa->spa_scrub_errors++; | |
| 1541 mutex_exit(&spa->spa_scrub_lock); | |
| 1542 | |
| 1543 if (vd != NULL) { | |
| 1544 mutex_enter(&vd->vdev_stat_lock); | |
| 1545 vd->vdev_stat.vs_scrub_errors++; | |
| 1546 mutex_exit(&vd->vdev_stat_lock); | |
| 1547 } | |
| 1548 | |
| 1549 return (ERESTART); | |
| 1550 } | |
| 1551 | |
| 1552 ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg); | |
| 1553 | |
| 1554 /* | |
| 1555 * Keep track of how much data we've examined so that | |
| 1556 * zpool(1M) status can make useful progress reports. | |
| 1557 */ | |
| 1558 mutex_enter(&vd->vdev_stat_lock); | |
| 1559 vd->vdev_stat.vs_scrub_examined += BP_GET_ASIZE(bp); | |
| 1560 mutex_exit(&vd->vdev_stat_lock); | |
| 1561 | |
| 1562 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) { | |
| 1563 if (DVA_GET_GANG(&bp->blk_dva[0])) { | |
| 1564 /* | |
| 1565 * Gang members may be spread across multiple vdevs, | |
| 1566 * so the best we can do is look at the pool-wide DTL. | |
| 1567 * XXX -- it would be better to change our allocation | |
| 1568 * policy to ensure that this can't happen. | |
| 1569 */ | |
| 1570 vd = spa->spa_root_vdev; | |
| 1571 } | |
| 1572 if (vdev_dtl_contains(&vd->vdev_dtl_map, bp->blk_birth, 1)) { | |
| 1573 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER, | |
| 1544 | 1574 ZIO_FLAG_RESILVER, &bc->bc_bookmark); |
| 789 | 1575 } |
| 1576 } else { | |
| 1577 spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB, | |
| 1544 | 1578 ZIO_FLAG_SCRUB, &bc->bc_bookmark); |
| 789 | 1579 } |
| 1580 | |
| 1581 return (0); | |
| 1582 } | |
| 1583 | |
| 1584 static void | |
| 1585 spa_scrub_thread(spa_t *spa) | |
| 1586 { | |
| 1587 callb_cpr_t cprinfo; | |
| 1588 traverse_handle_t *th = spa->spa_scrub_th; | |
| 1589 vdev_t *rvd = spa->spa_root_vdev; | |
| 1590 pool_scrub_type_t scrub_type = spa->spa_scrub_type; | |
| 1591 int error = 0; | |
| 1592 boolean_t complete; | |
| 1593 | |
| 1594 CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG); | |
| 1595 | |
|
797
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1596 /* |
|
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1597 * 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
|
1598 * wait for that to complete. |
|
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1599 */ |
|
af56ba8b7e41
6344108 snapshot create/delete interlock with scrub/resilver must sync txg
bonwick
parents:
789
diff
changeset
|
1600 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
|
1601 |
| 1544 | 1602 dprintf("start %s mintxg=%llu maxtxg=%llu\n", |
| 1603 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
| 1604 spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg); | |
| 1605 | |
| 1606 spa_config_enter(spa, RW_WRITER, FTAG); | |
| 1607 vdev_reopen(rvd); /* purge all vdev caches */ | |
| 789 | 1608 vdev_config_dirty(rvd); /* rewrite all disk labels */ |
| 1609 vdev_scrub_stat_update(rvd, scrub_type, B_FALSE); | |
| 1544 | 1610 spa_config_exit(spa, FTAG); |
| 789 | 1611 |
| 1612 mutex_enter(&spa->spa_scrub_lock); | |
| 1613 spa->spa_scrub_errors = 0; | |
| 1614 spa->spa_scrub_active = 1; | |
| 1544 | 1615 ASSERT(spa->spa_scrub_inflight == 0); |
| 1616 ASSERT(spa->spa_scrub_throttled == 0); | |
| 789 | 1617 |
| 1618 while (!spa->spa_scrub_stop) { | |
| 1619 CALLB_CPR_SAFE_BEGIN(&cprinfo); | |
| 1544 | 1620 while (spa->spa_scrub_suspended) { |
| 789 | 1621 spa->spa_scrub_active = 0; |
| 1622 cv_broadcast(&spa->spa_scrub_cv); | |
| 1623 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
| 1624 spa->spa_scrub_active = 1; | |
| 1625 } | |
| 1626 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock); | |
| 1627 | |
| 1628 if (spa->spa_scrub_restart_txg != 0) | |
| 1629 break; | |
| 1630 | |
| 1631 mutex_exit(&spa->spa_scrub_lock); | |
| 1632 error = traverse_more(th); | |
| 1633 mutex_enter(&spa->spa_scrub_lock); | |
| 1634 if (error != EAGAIN) | |
| 1635 break; | |
| 1544 | 1636 |
| 1637 while (spa->spa_scrub_throttled > 0) | |
| 1638 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
| 789 | 1639 } |
| 1640 | |
| 1641 while (spa->spa_scrub_inflight) | |
| 1642 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
| 1643 | |
|
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1644 spa->spa_scrub_active = 0; |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1645 cv_broadcast(&spa->spa_scrub_cv); |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1646 |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1647 mutex_exit(&spa->spa_scrub_lock); |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1648 |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1649 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
|
1650 |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1651 mutex_enter(&spa->spa_scrub_lock); |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1652 |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1653 /* |
|
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1654 * 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
|
1655 * 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
|
1656 * 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
|
1657 */ |
| 789 | 1658 if (spa->spa_scrub_restart_txg != 0) |
| 1659 error = ERESTART; | |
| 1660 | |
| 1544 | 1661 if (spa->spa_scrub_stop) |
| 1662 error = EINTR; | |
| 1663 | |
| 789 | 1664 /* |
| 1544 | 1665 * Even if there were uncorrectable errors, we consider the scrub |
| 1666 * completed. The downside is that if there is a transient error during | |
| 1667 * a resilver, we won't resilver the data properly to the target. But | |
| 1668 * if the damage is permanent (more likely) we will resilver forever, | |
| 1669 * which isn't really acceptable. Since there is enough information for | |
| 1670 * the user to know what has failed and why, this seems like a more | |
| 1671 * tractable approach. | |
| 789 | 1672 */ |
| 1544 | 1673 complete = (error == 0); |
| 789 | 1674 |
| 1544 | 1675 dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n", |
| 1676 scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", | |
| 789 | 1677 spa->spa_scrub_maxtxg, complete ? "done" : "FAILED", |
| 1678 error, spa->spa_scrub_errors, spa->spa_scrub_stop); | |
| 1679 | |
| 1680 mutex_exit(&spa->spa_scrub_lock); | |
| 1681 | |
| 1682 /* | |
| 1683 * If the scrub/resilver completed, update all DTLs to reflect this. | |
| 1684 * Whether it succeeded or not, vacate all temporary scrub DTLs. | |
| 1685 */ | |
| 1686 vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1, | |
| 1687 complete ? spa->spa_scrub_maxtxg : 0, B_TRUE); | |
| 1688 vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete); | |
| 1544 | 1689 spa_errlog_rotate(spa); |
|
1601
438b928f80c7
6397197 ADVANCE_ZIL should only visit claimed-but-not-yet-replayed logs
bonwick
parents:
1585
diff
changeset
|
1690 |
| 1544 | 1691 spa_config_exit(spa, FTAG); |
| 789 | 1692 |
| 1693 mutex_enter(&spa->spa_scrub_lock); | |
| 1694 | |
| 1544 | 1695 /* |
| 1696 * We may have finished replacing a device. | |
| 1697 * Let the async thread assess this and handle the detach. | |
| 1698 */ | |
| 1699 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); | |
| 789 | 1700 |
| 1701 /* | |
| 1702 * If we were told to restart, our final act is to start a new scrub. | |
| 1703 */ | |
| 1704 if (error == ERESTART) | |
| 1544 | 1705 spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ? |
| 1706 SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB); | |
| 789 | 1707 |
| 1544 | 1708 spa->spa_scrub_type = POOL_SCRUB_NONE; |
| 1709 spa->spa_scrub_active = 0; | |
| 1710 spa->spa_scrub_thread = NULL; | |
| 1711 cv_broadcast(&spa->spa_scrub_cv); | |
| 789 | 1712 CALLB_CPR_EXIT(&cprinfo); /* drops &spa->spa_scrub_lock */ |
| 1713 thread_exit(); | |
| 1714 } | |
| 1715 | |
| 1716 void | |
| 1717 spa_scrub_suspend(spa_t *spa) | |
| 1718 { | |
| 1719 mutex_enter(&spa->spa_scrub_lock); | |
| 1544 | 1720 spa->spa_scrub_suspended++; |
| 789 | 1721 while (spa->spa_scrub_active) { |
| 1722 cv_broadcast(&spa->spa_scrub_cv); | |
| 1723 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
| 1724 } | |
| 1725 while (spa->spa_scrub_inflight) | |
| 1726 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); | |
| 1727 mutex_exit(&spa->spa_scrub_lock); | |
| 1728 } | |
| 1729 | |
| 1730 void | |
| 1731 spa_scrub_resume(spa_t *spa) | |
| 1732 { | |
| 1733 mutex_enter(&spa->spa_scrub_lock); | |
| 1544 | 1734 ASSERT(spa->spa_scrub_suspended != 0); |
| 1735 if (--spa->spa_scrub_suspended == 0) | |
| 789 | 1736 cv_broadcast(&spa->spa_scrub_cv); |
| 1737 mutex_exit(&spa->spa_scrub_lock); | |
| 1738 } | |
| 1739 | |
| 1740 void | |
| 1741 spa_scrub_restart(spa_t *spa, uint64_t txg) | |
| 1742 { | |
| 1743 /* | |
| 1744 * Something happened (e.g. snapshot create/delete) that means | |
| 1745 * we must restart any in-progress scrubs. The itinerary will | |
| 1746 * fix this properly. | |
| 1747 */ | |
| 1748 mutex_enter(&spa->spa_scrub_lock); | |
| 1749 spa->spa_scrub_restart_txg = txg; | |
| 1750 mutex_exit(&spa->spa_scrub_lock); | |
| 1751 } | |
| 1752 | |
| 1544 | 1753 int |
| 1754 spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force) | |
| 789 | 1755 { |
| 1756 space_seg_t *ss; | |
| 1757 uint64_t mintxg, maxtxg; | |
| 1758 vdev_t *rvd = spa->spa_root_vdev; | |
| 1759 | |
| 1760 if ((uint_t)type >= POOL_SCRUB_TYPES) | |
| 1761 return (ENOTSUP); | |
| 1762 | |
| 1544 | 1763 mutex_enter(&spa->spa_scrub_lock); |
| 1764 | |
| 789 | 1765 /* |
| 1766 * If there's a scrub or resilver already in progress, stop it. | |
| 1767 */ | |
| 1768 while (spa->spa_scrub_thread != NULL) { | |
| 1769 /* | |
| 1770 * Don't stop a resilver unless forced. | |
| 1771 */ | |
| 1544 | 1772 if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) { |
| 1773 mutex_exit(&spa->spa_scrub_lock); | |
| 789 | 1774 return (EBUSY); |
| 1544 | 1775 } |
| 789 | 1776 spa->spa_scrub_stop = 1; |
| 1777 cv_broadcast(&spa->spa_scrub_cv); | |
| 1778 cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); | |
| 1779 } | |
| 1780 | |
| 1781 /* | |
| 1782 * Terminate the previous traverse. | |
| 1783 */ | |
| 1784 if (spa->spa_scrub_th != NULL) { | |
| 1785 traverse_fini(spa->spa_scrub_th); | |
| 1786 spa->spa_scrub_th = NULL; | |
| 1787 } | |
| 1788 | |
| 1544 | 1789 if (rvd == NULL) { |
| 1790 ASSERT(spa->spa_scrub_stop == 0); | |
| 1791 ASSERT(spa->spa_scrub_type == type); | |
| 1792 ASSERT(spa->spa_scrub_restart_txg == 0); | |
| 1793 mutex_exit(&spa->spa_scrub_lock); | |
| 1794 return (0); | |
| 1795 } | |
| 789 | 1796 |
| 1797 mintxg = TXG_INITIAL - 1; | |
| 1798 maxtxg = spa_last_synced_txg(spa) + 1; | |
| 1799 | |
| 1544 | 1800 mutex_enter(&rvd->vdev_dtl_lock); |
| 789 | 1801 |
| 1544 | 1802 if (rvd->vdev_dtl_map.sm_space == 0) { |
| 1803 /* | |
| 1804 * The pool-wide DTL is empty. | |
| 1732 | 1805 * If this is a resilver, there's nothing to do except |
| 1806 * check whether any in-progress replacements have completed. | |
| 1544 | 1807 */ |
| 1732 | 1808 if (type == POOL_SCRUB_RESILVER) { |
| 1544 | 1809 type = POOL_SCRUB_NONE; |
| 1732 | 1810 spa_async_request(spa, SPA_ASYNC_REPLACE_DONE); |
| 1811 } | |
| 1544 | 1812 } else { |
| 1813 /* | |
| 1814 * The pool-wide DTL is non-empty. | |
| 1815 * If this is a normal scrub, upgrade to a resilver instead. | |
| 1816 */ | |
| 1817 if (type == POOL_SCRUB_EVERYTHING) | |
| 1818 type = POOL_SCRUB_RESILVER; | |
| 1819 } | |
| 789 | 1820 |
| 1544 | 1821 if (type == POOL_SCRUB_RESILVER) { |
| 789 | 1822 /* |
| 1823 * Determine the resilvering boundaries. | |
| 1824 * | |
| 1825 * Note: (mintxg, maxtxg) is an open interval, | |
| 1826 * i.e. mintxg and maxtxg themselves are not included. | |
| 1827 * | |
| 1828 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1 | |
| 1829 * so we don't claim to resilver a txg that's still changing. | |
| 1830 */ | |
| 1831 ss = avl_first(&rvd->vdev_dtl_map.sm_root); | |
| 1544 | 1832 mintxg = ss->ss_start - 1; |
| 789 | 1833 ss = avl_last(&rvd->vdev_dtl_map.sm_root); |
| 1544 | 1834 maxtxg = MIN(ss->ss_end, maxtxg); |
| 789 | 1835 } |
| 1836 | |
| 1544 | 1837 mutex_exit(&rvd->vdev_dtl_lock); |
| 1838 | |
| 1839 spa->spa_scrub_stop = 0; | |
| 1840 spa->spa_scrub_type = type; | |
| 1841 spa->spa_scrub_restart_txg = 0; | |
| 1842 | |
| 1843 if (type != POOL_SCRUB_NONE) { | |
| 1844 spa->spa_scrub_mintxg = mintxg; | |
| 789 | 1845 spa->spa_scrub_maxtxg = maxtxg; |
| 1846 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
|
1847 ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL, |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1848 ZIO_FLAG_CANFAIL); |
| 789 | 1849 traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg); |
| 1850 spa->spa_scrub_thread = thread_create(NULL, 0, | |
| 1851 spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri); | |
| 1852 } | |
| 1853 | |
| 1544 | 1854 mutex_exit(&spa->spa_scrub_lock); |
| 1855 | |
| 789 | 1856 return (0); |
| 1857 } | |
| 1858 | |
| 1544 | 1859 /* |
| 1860 * ========================================================================== | |
| 1861 * SPA async task processing | |
| 1862 * ========================================================================== | |
| 1863 */ | |
| 1864 | |
| 1865 static void | |
| 1866 spa_async_reopen(spa_t *spa) | |
| 789 | 1867 { |
| 1544 | 1868 vdev_t *rvd = spa->spa_root_vdev; |
| 1869 vdev_t *tvd; | |
| 1870 int c; | |
| 1871 | |
| 1872 spa_config_enter(spa, RW_WRITER, FTAG); | |
| 1873 | |
| 1874 for (c = 0; c < rvd->vdev_children; c++) { | |
| 1875 tvd = rvd->vdev_child[c]; | |
| 1876 if (tvd->vdev_reopen_wanted) { | |
| 1877 tvd->vdev_reopen_wanted = 0; | |
| 1878 vdev_reopen(tvd); | |
| 1879 } | |
| 1880 } | |
| 789 | 1881 |
| 1544 | 1882 spa_config_exit(spa, FTAG); |
| 1883 } | |
| 1884 | |
| 1885 static void | |
| 1886 spa_async_thread(spa_t *spa) | |
| 1887 { | |
| 1888 int tasks; | |
| 1889 | |
| 1890 ASSERT(spa->spa_sync_on); | |
| 789 | 1891 |
| 1544 | 1892 mutex_enter(&spa->spa_async_lock); |
| 1893 tasks = spa->spa_async_tasks; | |
| 1894 spa->spa_async_tasks = 0; | |
| 1895 mutex_exit(&spa->spa_async_lock); | |
| 1896 | |
| 1897 /* | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1898 * 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
|
1899 */ |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1900 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1901 mutex_enter(&spa_namespace_lock); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1902 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
|
1903 mutex_exit(&spa_namespace_lock); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1904 } |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1905 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1906 /* |
| 1544 | 1907 * See if any devices need to be reopened. |
| 1908 */ | |
| 1909 if (tasks & SPA_ASYNC_REOPEN) | |
| 1910 spa_async_reopen(spa); | |
| 1911 | |
| 1912 /* | |
| 1913 * If any devices are done replacing, detach them. | |
| 1914 */ | |
| 1915 if (tasks & SPA_ASYNC_REPLACE_DONE) | |
| 789 | 1916 spa_vdev_replace_done(spa); |
| 1917 | |
| 1544 | 1918 /* |
| 1919 * Kick off a scrub. | |
| 1920 */ | |
| 1921 if (tasks & SPA_ASYNC_SCRUB) | |
| 1922 VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0); | |
| 1923 | |
| 1924 /* | |
| 1925 * Kick off a resilver. | |
| 1926 */ | |
| 1927 if (tasks & SPA_ASYNC_RESILVER) | |
| 1928 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); | |
| 1929 | |
| 1930 /* | |
| 1931 * Let the world know that we're done. | |
| 1932 */ | |
| 1933 mutex_enter(&spa->spa_async_lock); | |
| 1934 spa->spa_async_thread = NULL; | |
| 1935 cv_broadcast(&spa->spa_async_cv); | |
| 1936 mutex_exit(&spa->spa_async_lock); | |
| 1937 thread_exit(); | |
| 1938 } | |
| 1939 | |
| 1940 void | |
| 1941 spa_async_suspend(spa_t *spa) | |
| 1942 { | |
| 1943 mutex_enter(&spa->spa_async_lock); | |
| 1944 spa->spa_async_suspended++; | |
| 1945 while (spa->spa_async_thread != NULL) | |
| 1946 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); | |
| 1947 mutex_exit(&spa->spa_async_lock); | |
| 1948 } | |
| 1949 | |
| 1950 void | |
| 1951 spa_async_resume(spa_t *spa) | |
| 1952 { | |
| 1953 mutex_enter(&spa->spa_async_lock); | |
| 1954 ASSERT(spa->spa_async_suspended != 0); | |
| 1955 spa->spa_async_suspended--; | |
| 1956 mutex_exit(&spa->spa_async_lock); | |
| 1957 } | |
| 1958 | |
| 1959 static void | |
| 1960 spa_async_dispatch(spa_t *spa) | |
| 1961 { | |
| 1962 mutex_enter(&spa->spa_async_lock); | |
| 1963 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
|
1964 spa->spa_async_thread == NULL && |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
1965 rootdir != NULL && !vn_is_readonly(rootdir)) |
| 1544 | 1966 spa->spa_async_thread = thread_create(NULL, 0, |
| 1967 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); | |
| 1968 mutex_exit(&spa->spa_async_lock); | |
| 1969 } | |
| 1970 | |
| 1971 void | |
| 1972 spa_async_request(spa_t *spa, int task) | |
| 1973 { | |
| 1974 mutex_enter(&spa->spa_async_lock); | |
| 1975 spa->spa_async_tasks |= task; | |
| 1976 mutex_exit(&spa->spa_async_lock); | |
| 789 | 1977 } |
| 1978 | |
| 1979 /* | |
| 1980 * ========================================================================== | |
| 1981 * SPA syncing routines | |
| 1982 * ========================================================================== | |
| 1983 */ | |
| 1984 | |
| 1985 static void | |
| 1986 spa_sync_deferred_frees(spa_t *spa, uint64_t txg) | |
| 1987 { | |
| 1988 bplist_t *bpl = &spa->spa_sync_bplist; | |
| 1989 dmu_tx_t *tx; | |
| 1990 blkptr_t blk; | |
| 1991 uint64_t itor = 0; | |
| 1992 zio_t *zio; | |
| 1993 int error; | |
| 1994 uint8_t c = 1; | |
| 1995 | |
| 1996 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD); | |
| 1997 | |
| 1998 while (bplist_iterate(bpl, &itor, &blk) == 0) | |
| 1999 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL)); | |
| 2000 | |
| 2001 error = zio_wait(zio); | |
| 2002 ASSERT3U(error, ==, 0); | |
| 2003 | |
| 2004 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
| 2005 bplist_vacate(bpl, tx); | |
| 2006 | |
| 2007 /* | |
| 2008 * Pre-dirty the first block so we sync to convergence faster. | |
| 2009 * (Usually only the first block is needed.) | |
| 2010 */ | |
| 2011 dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); | |
| 2012 dmu_tx_commit(tx); | |
| 2013 } | |
| 2014 | |
| 2015 static void | |
| 2016 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) | |
| 2017 { | |
| 2018 nvlist_t *config; | |
| 2019 char *packed = NULL; | |
| 2020 size_t nvsize = 0; | |
| 2021 dmu_buf_t *db; | |
| 2022 | |
| 2023 if (list_is_empty(&spa->spa_dirty_list)) | |
| 2024 return; | |
| 2025 | |
| 2026 config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE); | |
| 2027 | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2028 if (spa->spa_config_syncing) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2029 nvlist_free(spa->spa_config_syncing); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2030 spa->spa_config_syncing = config; |
| 789 | 2031 |
| 2032 VERIFY(nvlist_size(config, &nvsize, NV_ENCODE_XDR) == 0); | |
| 2033 | |
| 2034 packed = kmem_alloc(nvsize, KM_SLEEP); | |
| 2035 | |
| 1544 | 2036 VERIFY(nvlist_pack(config, &packed, &nvsize, NV_ENCODE_XDR, |
| 2037 KM_SLEEP) == 0); | |
| 789 | 2038 |
| 2039 dmu_write(spa->spa_meta_objset, spa->spa_config_object, 0, nvsize, | |
| 2040 packed, tx); | |
| 2041 | |
| 2042 kmem_free(packed, nvsize); | |
| 2043 | |
| 1544 | 2044 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, |
| 2045 spa->spa_config_object, FTAG, &db)); | |
| 789 | 2046 dmu_buf_will_dirty(db, tx); |
| 2047 *(uint64_t *)db->db_data = nvsize; | |
| 1544 | 2048 dmu_buf_rele(db, FTAG); |
| 789 | 2049 } |
| 2050 | |
| 2051 /* | |
| 2052 * Sync the specified transaction group. New blocks may be dirtied as | |
| 2053 * part of the process, so we iterate until it converges. | |
| 2054 */ | |
| 2055 void | |
| 2056 spa_sync(spa_t *spa, uint64_t txg) | |
| 2057 { | |
| 2058 dsl_pool_t *dp = spa->spa_dsl_pool; | |
| 2059 objset_t *mos = spa->spa_meta_objset; | |
| 2060 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
|
2061 vdev_t *rvd = spa->spa_root_vdev; |
| 789 | 2062 vdev_t *vd; |
| 2063 dmu_tx_t *tx; | |
| 2064 int dirty_vdevs; | |
| 2065 | |
| 2066 /* | |
| 2067 * Lock out configuration changes. | |
| 2068 */ | |
| 1544 | 2069 spa_config_enter(spa, RW_READER, FTAG); |
| 789 | 2070 |
| 2071 spa->spa_syncing_txg = txg; | |
| 2072 spa->spa_sync_pass = 0; | |
| 2073 | |
| 1544 | 2074 VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); |
| 789 | 2075 |
| 2076 /* | |
| 2077 * If anything has changed in this txg, push the deferred frees | |
| 2078 * from the previous txg. If not, leave them alone so that we | |
| 2079 * don't generate work on an otherwise idle system. | |
| 2080 */ | |
| 2081 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || | |
| 2082 !txg_list_empty(&dp->dp_dirty_dirs, txg)) | |
| 2083 spa_sync_deferred_frees(spa, txg); | |
| 2084 | |
| 2085 /* | |
| 2086 * Iterate to convergence. | |
| 2087 */ | |
| 2088 do { | |
| 2089 spa->spa_sync_pass++; | |
| 2090 | |
| 2091 tx = dmu_tx_create_assigned(dp, txg); | |
| 2092 spa_sync_config_object(spa, tx); | |
| 2093 dmu_tx_commit(tx); | |
| 2094 | |
| 1544 | 2095 spa_errlog_sync(spa, txg); |
| 2096 | |
| 789 | 2097 dsl_pool_sync(dp, txg); |
| 2098 | |
| 2099 dirty_vdevs = 0; | |
| 2100 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { | |
| 2101 vdev_sync(vd, txg); | |
| 2102 dirty_vdevs++; | |
| 2103 } | |
| 2104 | |
| 2105 tx = dmu_tx_create_assigned(dp, txg); | |
| 2106 bplist_sync(bpl, tx); | |
| 2107 dmu_tx_commit(tx); | |
| 2108 | |
| 2109 } while (dirty_vdevs); | |
| 2110 | |
| 2111 bplist_close(bpl); | |
| 2112 | |
| 2113 dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); | |
| 2114 | |
| 2115 /* | |
| 2116 * Rewrite the vdev configuration (which includes the uberblock) | |
| 2117 * to commit the transaction group. | |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2118 * |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2119 * 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
|
2120 * 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
|
2121 * 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
|
2122 * If the write fails, try the next vdev until we're tried them all. |
| 789 | 2123 */ |
|
1635
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2124 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
|
2125 VERIFY(vdev_config_sync(rvd, txg) == 0); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2126 } else { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2127 int children = rvd->vdev_children; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2128 int c0 = spa_get_random(children); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2129 int c; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2130 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2131 for (c = 0; c < children; c++) { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2132 vd = rvd->vdev_child[(c0 + c) % children]; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2133 if (vd->vdev_ms_array == 0) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2134 continue; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2135 if (vdev_config_sync(vd, txg) == 0) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2136 break; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2137 } |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2138 if (c == children) |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2139 VERIFY(vdev_config_sync(rvd, txg) == 0); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2140 } |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2141 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2142 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2143 * Clear the dirty config list. |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2144 */ |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2145 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
|
2146 vdev_config_clean(vd); |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2147 |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2148 /* |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2149 * 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
|
2150 * 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
|
2151 */ |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2152 if (spa->spa_config_syncing != NULL) { |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2153 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
|
2154 spa->spa_config_txg = txg; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2155 spa->spa_config_syncing = NULL; |
|
0ab1193d47cb
6398664 zpool detach: missing argument to error message causes core dump
bonwick
parents:
1601
diff
changeset
|
2156 } |
| 789 | 2157 |
| 2158 /* | |
| 2159 * Make a stable copy of the fully synced uberblock. | |
| 2160 * We use this as the root for pool traversals. | |
| 2161 */ | |
| 2162 spa->spa_traverse_wanted = 1; /* tells traverse_more() to stop */ | |
| 2163 | |
| 2164 spa_scrub_suspend(spa); /* stop scrubbing and finish I/Os */ | |
| 2165 | |
| 2166 rw_enter(&spa->spa_traverse_lock, RW_WRITER); | |
| 2167 spa->spa_traverse_wanted = 0; | |
| 2168 spa->spa_ubsync = spa->spa_uberblock; | |
| 2169 rw_exit(&spa->spa_traverse_lock); | |
| 2170 | |
| 2171 spa_scrub_resume(spa); /* resume scrub with new ubsync */ | |
| 2172 | |
| 2173 /* | |
| 2174 * Clean up the ZIL records for the synced txg. | |
| 2175 */ | |
| 2176 dsl_pool_zil_clean(dp); | |
| 2177 | |
| 2178 /* | |
| 2179 * Update usable space statistics. | |
| 2180 */ | |
| 2181 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) | |
| 2182 vdev_sync_done(vd, txg); | |
| 2183 | |
| 2184 /* | |
| 2185 * It had better be the case that we didn't dirty anything | |
| 2186 * since spa_sync_labels(). | |
| 2187 */ | |
| 2188 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); | |
| 2189 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); | |
| 2190 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); | |
| 2191 ASSERT(bpl->bpl_queue == NULL); | |
| 2192 | |
| 1544 | 2193 spa_config_exit(spa, FTAG); |
| 2194 | |
| 2195 /* | |
| 2196 * If any async tasks have been requested, kick them off. | |
| 2197 */ | |
| 2198 spa_async_dispatch(spa); | |
| 789 | 2199 } |
| 2200 | |
| 2201 /* | |
| 2202 * Sync all pools. We don't want to hold the namespace lock across these | |
| 2203 * operations, so we take a reference on the spa_t and drop the lock during the | |
| 2204 * sync. | |
| 2205 */ | |
| 2206 void | |
| 2207 spa_sync_allpools(void) | |
| 2208 { | |
| 2209 spa_t *spa = NULL; | |
| 2210 mutex_enter(&spa_namespace_lock); | |
| 2211 while ((spa = spa_next(spa)) != NULL) { | |
| 2212 if (spa_state(spa) != POOL_STATE_ACTIVE) | |
| 2213 continue; | |
| 2214 spa_open_ref(spa, FTAG); | |
| 2215 mutex_exit(&spa_namespace_lock); | |
| 2216 txg_wait_synced(spa_get_dsl(spa), 0); | |
| 2217 mutex_enter(&spa_namespace_lock); | |
| 2218 spa_close(spa, FTAG); | |
| 2219 } | |
| 2220 mutex_exit(&spa_namespace_lock); | |
| 2221 } | |
| 2222 | |
| 2223 /* | |
| 2224 * ========================================================================== | |
| 2225 * Miscellaneous routines | |
| 2226 * ========================================================================== | |
| 2227 */ | |
| 2228 | |
| 2229 /* | |
| 2230 * Remove all pools in the system. | |
| 2231 */ | |
| 2232 void | |
| 2233 spa_evict_all(void) | |
| 2234 { | |
| 2235 spa_t *spa; | |
| 2236 | |
| 2237 /* | |
| 2238 * Remove all cached state. All pools should be closed now, | |
| 2239 * so every spa in the AVL tree should be unreferenced. | |
| 2240 */ | |
| 2241 mutex_enter(&spa_namespace_lock); | |
| 2242 while ((spa = spa_next(NULL)) != NULL) { | |
| 2243 /* | |
| 1544 | 2244 * Stop async tasks. The async thread may need to detach |
| 2245 * a device that's been replaced, which requires grabbing | |
| 2246 * spa_namespace_lock, so we must drop it here. | |
| 789 | 2247 */ |
| 2248 spa_open_ref(spa, FTAG); | |
| 2249 mutex_exit(&spa_namespace_lock); | |
| 1544 | 2250 spa_async_suspend(spa); |
| 789 | 2251 VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); |
| 2252 mutex_enter(&spa_namespace_lock); | |
| 2253 spa_close(spa, FTAG); | |
| 2254 | |
| 2255 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { | |
| 2256 spa_unload(spa); | |
| 2257 spa_deactivate(spa); | |
| 2258 } | |
| 2259 spa_remove(spa); | |
| 2260 } | |
| 2261 mutex_exit(&spa_namespace_lock); | |
| 2262 } | |
| 1544 | 2263 |
| 2264 vdev_t * | |
| 2265 spa_lookup_by_guid(spa_t *spa, uint64_t guid) | |
| 2266 { | |
| 2267 return (vdev_lookup_by_guid(spa->spa_root_vdev, guid)); | |
| 2268 } |