1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 */ 28 29 /* 30 * SPA: Storage Pool Allocator 31 * 32 * This file contains all the routines used when modifying on-disk SPA state. 33 * This includes opening, importing, destroying, exporting a pool, and syncing a 34 * pool. 35 */ 36 37 #include <sys/zfs_context.h> 38 #include <sys/fm/fs/zfs.h> 39 #include <sys/spa_impl.h> 40 #include <sys/zio.h> 41 #include <sys/zio_checksum.h> 42 #include <sys/dmu.h> 43 #include <sys/dmu_tx.h> 44 #include <sys/zap.h> 45 #include <sys/zil.h> 46 #include <sys/ddt.h> 47 #include <sys/vdev_impl.h> 48 #include <sys/metaslab.h> 49 #include <sys/metaslab_impl.h> 50 #include <sys/uberblock_impl.h> 51 #include <sys/txg.h> 52 #include <sys/avl.h> 53 #include <sys/dmu_traverse.h> 54 #include <sys/dmu_objset.h> 55 #include <sys/unique.h> 56 #include <sys/dsl_pool.h> 57 #include <sys/dsl_dataset.h> 58 #include <sys/dsl_dir.h> 59 #include <sys/dsl_prop.h> 60 #include <sys/dsl_synctask.h> 61 #include <sys/fs/zfs.h> 62 #include <sys/arc.h> 63 #include <sys/callb.h> 64 #include <sys/systeminfo.h> 65 #include <sys/spa_boot.h> 66 #include <sys/zfs_ioctl.h> 67 #include <sys/dsl_scan.h> 68 #include <sys/zfeature.h> 69 #include <sys/dsl_destroy.h> 70 71 #ifdef _KERNEL 72 #include <sys/bootprops.h> 73 #include <sys/callb.h> 74 #include <sys/cpupart.h> 75 #include <sys/pool.h> 76 #include <sys/sysdc.h> 77 #include <sys/zone.h> 78 #endif /* _KERNEL */ 79 80 #include "zfs_prop.h" 81 #include "zfs_comutil.h" 82 83 /* 84 * The interval, in seconds, at which failed configuration cache file writes 85 * should be retried. 86 */ 87 static int zfs_ccw_retry_interval = 300; 88 89 typedef enum zti_modes { 90 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 91 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 92 ZTI_MODE_NULL, /* don't create a taskq */ 93 ZTI_NMODES 94 } zti_modes_t; 95 96 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 97 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 98 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 99 100 #define ZTI_N(n) ZTI_P(n, 1) 101 #define ZTI_ONE ZTI_N(1) 102 103 typedef struct zio_taskq_info { 104 zti_modes_t zti_mode; 105 uint_t zti_value; 106 uint_t zti_count; 107 } zio_taskq_info_t; 108 109 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 110 "issue", "issue_high", "intr", "intr_high" 111 }; 112 113 /* 114 * This table defines the taskq settings for each ZFS I/O type. When 115 * initializing a pool, we use this table to create an appropriately sized 116 * taskq. Some operations are low volume and therefore have a small, static 117 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 118 * macros. Other operations process a large amount of data; the ZTI_BATCH 119 * macro causes us to create a taskq oriented for throughput. Some operations 120 * are so high frequency and short-lived that the taskq itself can become a a 121 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 122 * additional degree of parallelism specified by the number of threads per- 123 * taskq and the number of taskqs; when dispatching an event in this case, the 124 * particular taskq is chosen at random. 125 * 126 * The different taskq priorities are to handle the different contexts (issue 127 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 128 * need to be handled with minimum delay. 129 */ 130 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 131 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 132 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 133 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 134 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 135 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 136 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 137 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 138 }; 139 140 static void spa_sync_version(void *arg, dmu_tx_t *tx); 141 static void spa_sync_props(void *arg, dmu_tx_t *tx); 142 static boolean_t spa_has_active_shared_spare(spa_t *spa); 143 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 144 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 145 char **ereport); 146 static void spa_vdev_resilver_done(spa_t *spa); 147 148 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 149 id_t zio_taskq_psrset_bind = PS_NONE; 150 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 151 uint_t zio_taskq_basedc = 80; /* base duty cycle */ 152 153 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 154 extern int zfs_sync_pass_deferred_free; 155 156 /* 157 * This (illegal) pool name is used when temporarily importing a spa_t in order 158 * to get the vdev stats associated with the imported devices. 159 */ 160 #define TRYIMPORT_NAME "$import" 161 162 /* 163 * ========================================================================== 164 * SPA properties routines 165 * ========================================================================== 166 */ 167 168 /* 169 * Add a (source=src, propname=propval) list to an nvlist. 170 */ 171 static void 172 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 173 uint64_t intval, zprop_source_t src) 174 { 175 const char *propname = zpool_prop_to_name(prop); 176 nvlist_t *propval; 177 178 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 179 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 180 181 if (strval != NULL) 182 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 183 else 184 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 185 186 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 187 nvlist_free(propval); 188 } 189 190 /* 191 * Get property values from the spa configuration. 192 */ 193 static void 194 spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 195 { 196 vdev_t *rvd = spa->spa_root_vdev; 197 dsl_pool_t *pool = spa->spa_dsl_pool; 198 uint64_t size, alloc, cap, version; 199 zprop_source_t src = ZPROP_SRC_NONE; 200 spa_config_dirent_t *dp; 201 metaslab_class_t *mc = spa_normal_class(spa); 202 203 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 204 205 if (rvd != NULL) { 206 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 207 size = metaslab_class_get_space(spa_normal_class(spa)); 208 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 209 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 210 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 211 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 212 size - alloc, src); 213 214 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 215 metaslab_class_fragmentation(mc), src); 216 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 217 metaslab_class_expandable_space(mc), src); 218 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 219 (spa_mode(spa) == FREAD), src); 220 221 cap = (size == 0) ? 0 : (alloc * 100 / size); 222 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 223 224 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 225 ddt_get_pool_dedup_ratio(spa), src); 226 227 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 228 rvd->vdev_state, src); 229 230 version = spa_version(spa); 231 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 232 src = ZPROP_SRC_DEFAULT; 233 else 234 src = ZPROP_SRC_LOCAL; 235 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 236 } 237 238 if (pool != NULL) { 239 /* 240 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 241 * when opening pools before this version freedir will be NULL. 242 */ 243 if (pool->dp_free_dir != NULL) { 244 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 245 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 246 src); 247 } else { 248 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 249 NULL, 0, src); 250 } 251 252 if (pool->dp_leak_dir != NULL) { 253 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 254 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 255 src); 256 } else { 257 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 258 NULL, 0, src); 259 } 260 } 261 262 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 263 264 if (spa->spa_comment != NULL) { 265 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 266 0, ZPROP_SRC_LOCAL); 267 } 268 269 if (spa->spa_root != NULL) 270 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 271 0, ZPROP_SRC_LOCAL); 272 273 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 274 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 275 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 276 } else { 277 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 278 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 279 } 280 281 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 282 if (dp->scd_path == NULL) { 283 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 284 "none", 0, ZPROP_SRC_LOCAL); 285 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 286 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 287 dp->scd_path, 0, ZPROP_SRC_LOCAL); 288 } 289 } 290 } 291 292 /* 293 * Get zpool property values. 294 */ 295 int 296 spa_prop_get(spa_t *spa, nvlist_t **nvp) 297 { 298 objset_t *mos = spa->spa_meta_objset; 299 zap_cursor_t zc; 300 zap_attribute_t za; 301 int err; 302 303 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 304 305 mutex_enter(&spa->spa_props_lock); 306 307 /* 308 * Get properties from the spa config. 309 */ 310 spa_prop_get_config(spa, nvp); 311 312 /* If no pool property object, no more prop to get. */ 313 if (mos == NULL || spa->spa_pool_props_object == 0) { 314 mutex_exit(&spa->spa_props_lock); 315 return (0); 316 } 317 318 /* 319 * Get properties from the MOS pool property object. 320 */ 321 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 322 (err = zap_cursor_retrieve(&zc, &za)) == 0; 323 zap_cursor_advance(&zc)) { 324 uint64_t intval = 0; 325 char *strval = NULL; 326 zprop_source_t src = ZPROP_SRC_DEFAULT; 327 zpool_prop_t prop; 328 329 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 330 continue; 331 332 switch (za.za_integer_length) { 333 case 8: 334 /* integer property */ 335 if (za.za_first_integer != 336 zpool_prop_default_numeric(prop)) 337 src = ZPROP_SRC_LOCAL; 338 339 if (prop == ZPOOL_PROP_BOOTFS) { 340 dsl_pool_t *dp; 341 dsl_dataset_t *ds = NULL; 342 343 dp = spa_get_dsl(spa); 344 dsl_pool_config_enter(dp, FTAG); 345 if (err = dsl_dataset_hold_obj(dp, 346 za.za_first_integer, FTAG, &ds)) { 347 dsl_pool_config_exit(dp, FTAG); 348 break; 349 } 350 351 strval = kmem_alloc( 352 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 353 KM_SLEEP); 354 dsl_dataset_name(ds, strval); 355 dsl_dataset_rele(ds, FTAG); 356 dsl_pool_config_exit(dp, FTAG); 357 } else { 358 strval = NULL; 359 intval = za.za_first_integer; 360 } 361 362 spa_prop_add_list(*nvp, prop, strval, intval, src); 363 364 if (strval != NULL) 365 kmem_free(strval, 366 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 367 368 break; 369 370 case 1: 371 /* string property */ 372 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 373 err = zap_lookup(mos, spa->spa_pool_props_object, 374 za.za_name, 1, za.za_num_integers, strval); 375 if (err) { 376 kmem_free(strval, za.za_num_integers); 377 break; 378 } 379 spa_prop_add_list(*nvp, prop, strval, 0, src); 380 kmem_free(strval, za.za_num_integers); 381 break; 382 383 default: 384 break; 385 } 386 } 387 zap_cursor_fini(&zc); 388 mutex_exit(&spa->spa_props_lock); 389 out: 390 if (err && err != ENOENT) { 391 nvlist_free(*nvp); 392 *nvp = NULL; 393 return (err); 394 } 395 396 return (0); 397 } 398 399 /* 400 * Validate the given pool properties nvlist and modify the list 401 * for the property values to be set. 402 */ 403 static int 404 spa_prop_validate(spa_t *spa, nvlist_t *props) 405 { 406 nvpair_t *elem; 407 int error = 0, reset_bootfs = 0; 408 uint64_t objnum = 0; 409 boolean_t has_feature = B_FALSE; 410 411 elem = NULL; 412 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 413 uint64_t intval; 414 char *strval, *slash, *check, *fname; 415 const char *propname = nvpair_name(elem); 416 zpool_prop_t prop = zpool_name_to_prop(propname); 417 418 switch (prop) { 419 case ZPROP_INVAL: 420 if (!zpool_prop_feature(propname)) { 421 error = SET_ERROR(EINVAL); 422 break; 423 } 424 425 /* 426 * Sanitize the input. 427 */ 428 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 429 error = SET_ERROR(EINVAL); 430 break; 431 } 432 433 if (nvpair_value_uint64(elem, &intval) != 0) { 434 error = SET_ERROR(EINVAL); 435 break; 436 } 437 438 if (intval != 0) { 439 error = SET_ERROR(EINVAL); 440 break; 441 } 442 443 fname = strchr(propname, '@') + 1; 444 if (zfeature_lookup_name(fname, NULL) != 0) { 445 error = SET_ERROR(EINVAL); 446 break; 447 } 448 449 has_feature = B_TRUE; 450 break; 451 452 case ZPOOL_PROP_VERSION: 453 error = nvpair_value_uint64(elem, &intval); 454 if (!error && 455 (intval < spa_version(spa) || 456 intval > SPA_VERSION_BEFORE_FEATURES || 457 has_feature)) 458 error = SET_ERROR(EINVAL); 459 break; 460 461 case ZPOOL_PROP_DELEGATION: 462 case ZPOOL_PROP_AUTOREPLACE: 463 case ZPOOL_PROP_LISTSNAPS: 464 case ZPOOL_PROP_AUTOEXPAND: 465 error = nvpair_value_uint64(elem, &intval); 466 if (!error && intval > 1) 467 error = SET_ERROR(EINVAL); 468 break; 469 470 case ZPOOL_PROP_BOOTFS: 471 /* 472 * If the pool version is less than SPA_VERSION_BOOTFS, 473 * or the pool is still being created (version == 0), 474 * the bootfs property cannot be set. 475 */ 476 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 477 error = SET_ERROR(ENOTSUP); 478 break; 479 } 480 481 /* 482 * Make sure the vdev config is bootable 483 */ 484 if (!vdev_is_bootable(spa->spa_root_vdev)) { 485 error = SET_ERROR(ENOTSUP); 486 break; 487 } 488 489 reset_bootfs = 1; 490 491 error = nvpair_value_string(elem, &strval); 492 493 if (!error) { 494 objset_t *os; 495 uint64_t propval; 496 497 if (strval == NULL || strval[0] == '\0') { 498 objnum = zpool_prop_default_numeric( 499 ZPOOL_PROP_BOOTFS); 500 break; 501 } 502 503 if (error = dmu_objset_hold(strval, FTAG, &os)) 504 break; 505 506 /* 507 * Must be ZPL, and its property settings 508 * must be supported by GRUB (compression 509 * is not gzip, and large blocks are not used). 510 */ 511 512 if (dmu_objset_type(os) != DMU_OST_ZFS) { 513 error = SET_ERROR(ENOTSUP); 514 } else if ((error = 515 dsl_prop_get_int_ds(dmu_objset_ds(os), 516 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 517 &propval)) == 0 && 518 !BOOTFS_COMPRESS_VALID(propval)) { 519 error = SET_ERROR(ENOTSUP); 520 } else if ((error = 521 dsl_prop_get_int_ds(dmu_objset_ds(os), 522 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 523 &propval)) == 0 && 524 propval > SPA_OLD_MAXBLOCKSIZE) { 525 error = SET_ERROR(ENOTSUP); 526 } else { 527 objnum = dmu_objset_id(os); 528 } 529 dmu_objset_rele(os, FTAG); 530 } 531 break; 532 533 case ZPOOL_PROP_FAILUREMODE: 534 error = nvpair_value_uint64(elem, &intval); 535 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 536 intval > ZIO_FAILURE_MODE_PANIC)) 537 error = SET_ERROR(EINVAL); 538 539 /* 540 * This is a special case which only occurs when 541 * the pool has completely failed. This allows 542 * the user to change the in-core failmode property 543 * without syncing it out to disk (I/Os might 544 * currently be blocked). We do this by returning 545 * EIO to the caller (spa_prop_set) to trick it 546 * into thinking we encountered a property validation 547 * error. 548 */ 549 if (!error && spa_suspended(spa)) { 550 spa->spa_failmode = intval; 551 error = SET_ERROR(EIO); 552 } 553 break; 554 555 case ZPOOL_PROP_CACHEFILE: 556 if ((error = nvpair_value_string(elem, &strval)) != 0) 557 break; 558 559 if (strval[0] == '\0') 560 break; 561 562 if (strcmp(strval, "none") == 0) 563 break; 564 565 if (strval[0] != '/') { 566 error = SET_ERROR(EINVAL); 567 break; 568 } 569 570 slash = strrchr(strval, '/'); 571 ASSERT(slash != NULL); 572 573 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 574 strcmp(slash, "/..") == 0) 575 error = SET_ERROR(EINVAL); 576 break; 577 578 case ZPOOL_PROP_COMMENT: 579 if ((error = nvpair_value_string(elem, &strval)) != 0) 580 break; 581 for (check = strval; *check != '\0'; check++) { 582 /* 583 * The kernel doesn't have an easy isprint() 584 * check. For this kernel check, we merely 585 * check ASCII apart from DEL. Fix this if 586 * there is an easy-to-use kernel isprint(). 587 */ 588 if (*check >= 0x7f) { 589 error = SET_ERROR(EINVAL); 590 break; 591 } 592 check++; 593 } 594 if (strlen(strval) > ZPROP_MAX_COMMENT) 595 error = E2BIG; 596 break; 597 598 case ZPOOL_PROP_DEDUPDITTO: 599 if (spa_version(spa) < SPA_VERSION_DEDUP) 600 error = SET_ERROR(ENOTSUP); 601 else 602 error = nvpair_value_uint64(elem, &intval); 603 if (error == 0 && 604 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 605 error = SET_ERROR(EINVAL); 606 break; 607 } 608 609 if (error) 610 break; 611 } 612 613 if (!error && reset_bootfs) { 614 error = nvlist_remove(props, 615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 616 617 if (!error) { 618 error = nvlist_add_uint64(props, 619 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 620 } 621 } 622 623 return (error); 624 } 625 626 void 627 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 628 { 629 char *cachefile; 630 spa_config_dirent_t *dp; 631 632 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 633 &cachefile) != 0) 634 return; 635 636 dp = kmem_alloc(sizeof (spa_config_dirent_t), 637 KM_SLEEP); 638 639 if (cachefile[0] == '\0') 640 dp->scd_path = spa_strdup(spa_config_path); 641 else if (strcmp(cachefile, "none") == 0) 642 dp->scd_path = NULL; 643 else 644 dp->scd_path = spa_strdup(cachefile); 645 646 list_insert_head(&spa->spa_config_list, dp); 647 if (need_sync) 648 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 649 } 650 651 int 652 spa_prop_set(spa_t *spa, nvlist_t *nvp) 653 { 654 int error; 655 nvpair_t *elem = NULL; 656 boolean_t need_sync = B_FALSE; 657 658 if ((error = spa_prop_validate(spa, nvp)) != 0) 659 return (error); 660 661 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 662 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 663 664 if (prop == ZPOOL_PROP_CACHEFILE || 665 prop == ZPOOL_PROP_ALTROOT || 666 prop == ZPOOL_PROP_READONLY) 667 continue; 668 669 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 670 uint64_t ver; 671 672 if (prop == ZPOOL_PROP_VERSION) { 673 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 674 } else { 675 ASSERT(zpool_prop_feature(nvpair_name(elem))); 676 ver = SPA_VERSION_FEATURES; 677 need_sync = B_TRUE; 678 } 679 680 /* Save time if the version is already set. */ 681 if (ver == spa_version(spa)) 682 continue; 683 684 /* 685 * In addition to the pool directory object, we might 686 * create the pool properties object, the features for 687 * read object, the features for write object, or the 688 * feature descriptions object. 689 */ 690 error = dsl_sync_task(spa->spa_name, NULL, 691 spa_sync_version, &ver, 692 6, ZFS_SPACE_CHECK_RESERVED); 693 if (error) 694 return (error); 695 continue; 696 } 697 698 need_sync = B_TRUE; 699 break; 700 } 701 702 if (need_sync) { 703 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 704 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 705 } 706 707 return (0); 708 } 709 710 /* 711 * If the bootfs property value is dsobj, clear it. 712 */ 713 void 714 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 715 { 716 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 717 VERIFY(zap_remove(spa->spa_meta_objset, 718 spa->spa_pool_props_object, 719 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 720 spa->spa_bootfs = 0; 721 } 722 } 723 724 /*ARGSUSED*/ 725 static int 726 spa_change_guid_check(void *arg, dmu_tx_t *tx) 727 { 728 uint64_t *newguid = arg; 729 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 730 vdev_t *rvd = spa->spa_root_vdev; 731 uint64_t vdev_state; 732 733 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 734 vdev_state = rvd->vdev_state; 735 spa_config_exit(spa, SCL_STATE, FTAG); 736 737 if (vdev_state != VDEV_STATE_HEALTHY) 738 return (SET_ERROR(ENXIO)); 739 740 ASSERT3U(spa_guid(spa), !=, *newguid); 741 742 return (0); 743 } 744 745 static void 746 spa_change_guid_sync(void *arg, dmu_tx_t *tx) 747 { 748 uint64_t *newguid = arg; 749 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 750 uint64_t oldguid; 751 vdev_t *rvd = spa->spa_root_vdev; 752 753 oldguid = spa_guid(spa); 754 755 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 756 rvd->vdev_guid = *newguid; 757 rvd->vdev_guid_sum += (*newguid - oldguid); 758 vdev_config_dirty(rvd); 759 spa_config_exit(spa, SCL_STATE, FTAG); 760 761 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 762 oldguid, *newguid); 763 } 764 765 /* 766 * Change the GUID for the pool. This is done so that we can later 767 * re-import a pool built from a clone of our own vdevs. We will modify 768 * the root vdev's guid, our own pool guid, and then mark all of our 769 * vdevs dirty. Note that we must make sure that all our vdevs are 770 * online when we do this, or else any vdevs that weren't present 771 * would be orphaned from our pool. We are also going to issue a 772 * sysevent to update any watchers. 773 */ 774 int 775 spa_change_guid(spa_t *spa) 776 { 777 int error; 778 uint64_t guid; 779 780 mutex_enter(&spa->spa_vdev_top_lock); 781 mutex_enter(&spa_namespace_lock); 782 guid = spa_generate_guid(NULL); 783 784 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 785 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 786 787 if (error == 0) { 788 spa_config_sync(spa, B_FALSE, B_TRUE); 789 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 790 } 791 792 mutex_exit(&spa_namespace_lock); 793 mutex_exit(&spa->spa_vdev_top_lock); 794 795 return (error); 796 } 797 798 /* 799 * ========================================================================== 800 * SPA state manipulation (open/create/destroy/import/export) 801 * ========================================================================== 802 */ 803 804 static int 805 spa_error_entry_compare(const void *a, const void *b) 806 { 807 spa_error_entry_t *sa = (spa_error_entry_t *)a; 808 spa_error_entry_t *sb = (spa_error_entry_t *)b; 809 int ret; 810 811 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 812 sizeof (zbookmark_phys_t)); 813 814 if (ret < 0) 815 return (-1); 816 else if (ret > 0) 817 return (1); 818 else 819 return (0); 820 } 821 822 /* 823 * Utility function which retrieves copies of the current logs and 824 * re-initializes them in the process. 825 */ 826 void 827 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 828 { 829 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 830 831 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 832 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 833 834 avl_create(&spa->spa_errlist_scrub, 835 spa_error_entry_compare, sizeof (spa_error_entry_t), 836 offsetof(spa_error_entry_t, se_avl)); 837 avl_create(&spa->spa_errlist_last, 838 spa_error_entry_compare, sizeof (spa_error_entry_t), 839 offsetof(spa_error_entry_t, se_avl)); 840 } 841 842 static void 843 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 844 { 845 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 846 enum zti_modes mode = ztip->zti_mode; 847 uint_t value = ztip->zti_value; 848 uint_t count = ztip->zti_count; 849 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 850 char name[32]; 851 uint_t flags = 0; 852 boolean_t batch = B_FALSE; 853 854 if (mode == ZTI_MODE_NULL) { 855 tqs->stqs_count = 0; 856 tqs->stqs_taskq = NULL; 857 return; 858 } 859 860 ASSERT3U(count, >, 0); 861 862 tqs->stqs_count = count; 863 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 864 865 switch (mode) { 866 case ZTI_MODE_FIXED: 867 ASSERT3U(value, >=, 1); 868 value = MAX(value, 1); 869 break; 870 871 case ZTI_MODE_BATCH: 872 batch = B_TRUE; 873 flags |= TASKQ_THREADS_CPU_PCT; 874 value = zio_taskq_batch_pct; 875 break; 876 877 default: 878 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 879 "spa_activate()", 880 zio_type_name[t], zio_taskq_types[q], mode, value); 881 break; 882 } 883 884 for (uint_t i = 0; i < count; i++) { 885 taskq_t *tq; 886 887 if (count > 1) { 888 (void) snprintf(name, sizeof (name), "%s_%s_%u", 889 zio_type_name[t], zio_taskq_types[q], i); 890 } else { 891 (void) snprintf(name, sizeof (name), "%s_%s", 892 zio_type_name[t], zio_taskq_types[q]); 893 } 894 895 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 896 if (batch) 897 flags |= TASKQ_DC_BATCH; 898 899 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 900 spa->spa_proc, zio_taskq_basedc, flags); 901 } else { 902 pri_t pri = maxclsyspri; 903 /* 904 * The write issue taskq can be extremely CPU 905 * intensive. Run it at slightly lower priority 906 * than the other taskqs. 907 */ 908 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 909 pri--; 910 911 tq = taskq_create_proc(name, value, pri, 50, 912 INT_MAX, spa->spa_proc, flags); 913 } 914 915 tqs->stqs_taskq[i] = tq; 916 } 917 } 918 919 static void 920 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 921 { 922 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 923 924 if (tqs->stqs_taskq == NULL) { 925 ASSERT0(tqs->stqs_count); 926 return; 927 } 928 929 for (uint_t i = 0; i < tqs->stqs_count; i++) { 930 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 931 taskq_destroy(tqs->stqs_taskq[i]); 932 } 933 934 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 935 tqs->stqs_taskq = NULL; 936 } 937 938 /* 939 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 940 * Note that a type may have multiple discrete taskqs to avoid lock contention 941 * on the taskq itself. In that case we choose which taskq at random by using 942 * the low bits of gethrtime(). 943 */ 944 void 945 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 946 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 947 { 948 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 949 taskq_t *tq; 950 951 ASSERT3P(tqs->stqs_taskq, !=, NULL); 952 ASSERT3U(tqs->stqs_count, !=, 0); 953 954 if (tqs->stqs_count == 1) { 955 tq = tqs->stqs_taskq[0]; 956 } else { 957 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 958 } 959 960 taskq_dispatch_ent(tq, func, arg, flags, ent); 961 } 962 963 static void 964 spa_create_zio_taskqs(spa_t *spa) 965 { 966 for (int t = 0; t < ZIO_TYPES; t++) { 967 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 968 spa_taskqs_init(spa, t, q); 969 } 970 } 971 } 972 973 #ifdef _KERNEL 974 static void 975 spa_thread(void *arg) 976 { 977 callb_cpr_t cprinfo; 978 979 spa_t *spa = arg; 980 user_t *pu = PTOU(curproc); 981 982 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 983 spa->spa_name); 984 985 ASSERT(curproc != &p0); 986 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 987 "zpool-%s", spa->spa_name); 988 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 989 990 /* bind this thread to the requested psrset */ 991 if (zio_taskq_psrset_bind != PS_NONE) { 992 pool_lock(); 993 mutex_enter(&cpu_lock); 994 mutex_enter(&pidlock); 995 mutex_enter(&curproc->p_lock); 996 997 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 998 0, NULL, NULL) == 0) { 999 curthread->t_bind_pset = zio_taskq_psrset_bind; 1000 } else { 1001 cmn_err(CE_WARN, 1002 "Couldn't bind process for zfs pool \"%s\" to " 1003 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1004 } 1005 1006 mutex_exit(&curproc->p_lock); 1007 mutex_exit(&pidlock); 1008 mutex_exit(&cpu_lock); 1009 pool_unlock(); 1010 } 1011 1012 if (zio_taskq_sysdc) { 1013 sysdc_thread_enter(curthread, 100, 0); 1014 } 1015 1016 spa->spa_proc = curproc; 1017 spa->spa_did = curthread->t_did; 1018 1019 spa_create_zio_taskqs(spa); 1020 1021 mutex_enter(&spa->spa_proc_lock); 1022 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1023 1024 spa->spa_proc_state = SPA_PROC_ACTIVE; 1025 cv_broadcast(&spa->spa_proc_cv); 1026 1027 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1028 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1029 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1030 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1031 1032 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1033 spa->spa_proc_state = SPA_PROC_GONE; 1034 spa->spa_proc = &p0; 1035 cv_broadcast(&spa->spa_proc_cv); 1036 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1037 1038 mutex_enter(&curproc->p_lock); 1039 lwp_exit(); 1040 } 1041 #endif 1042 1043 /* 1044 * Activate an uninitialized pool. 1045 */ 1046 static void 1047 spa_activate(spa_t *spa, int mode) 1048 { 1049 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1050 1051 spa->spa_state = POOL_STATE_ACTIVE; 1052 spa->spa_mode = mode; 1053 1054 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1055 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1056 1057 /* Try to create a covering process */ 1058 mutex_enter(&spa->spa_proc_lock); 1059 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1060 ASSERT(spa->spa_proc == &p0); 1061 spa->spa_did = 0; 1062 1063 /* Only create a process if we're going to be around a while. */ 1064 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1065 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1066 NULL, 0) == 0) { 1067 spa->spa_proc_state = SPA_PROC_CREATED; 1068 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1069 cv_wait(&spa->spa_proc_cv, 1070 &spa->spa_proc_lock); 1071 } 1072 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1073 ASSERT(spa->spa_proc != &p0); 1074 ASSERT(spa->spa_did != 0); 1075 } else { 1076 #ifdef _KERNEL 1077 cmn_err(CE_WARN, 1078 "Couldn't create process for zfs pool \"%s\"\n", 1079 spa->spa_name); 1080 #endif 1081 } 1082 } 1083 mutex_exit(&spa->spa_proc_lock); 1084 1085 /* If we didn't create a process, we need to create our taskqs. */ 1086 if (spa->spa_proc == &p0) { 1087 spa_create_zio_taskqs(spa); 1088 } 1089 1090 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1091 offsetof(vdev_t, vdev_config_dirty_node)); 1092 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1093 offsetof(objset_t, os_evicting_node)); 1094 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1095 offsetof(vdev_t, vdev_state_dirty_node)); 1096 1097 txg_list_create(&spa->spa_vdev_txg_list, 1098 offsetof(struct vdev, vdev_txg_node)); 1099 1100 avl_create(&spa->spa_errlist_scrub, 1101 spa_error_entry_compare, sizeof (spa_error_entry_t), 1102 offsetof(spa_error_entry_t, se_avl)); 1103 avl_create(&spa->spa_errlist_last, 1104 spa_error_entry_compare, sizeof (spa_error_entry_t), 1105 offsetof(spa_error_entry_t, se_avl)); 1106 } 1107 1108 /* 1109 * Opposite of spa_activate(). 1110 */ 1111 static void 1112 spa_deactivate(spa_t *spa) 1113 { 1114 ASSERT(spa->spa_sync_on == B_FALSE); 1115 ASSERT(spa->spa_dsl_pool == NULL); 1116 ASSERT(spa->spa_root_vdev == NULL); 1117 ASSERT(spa->spa_async_zio_root == NULL); 1118 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1119 1120 spa_evicting_os_wait(spa); 1121 1122 txg_list_destroy(&spa->spa_vdev_txg_list); 1123 1124 list_destroy(&spa->spa_config_dirty_list); 1125 list_destroy(&spa->spa_evicting_os_list); 1126 list_destroy(&spa->spa_state_dirty_list); 1127 1128 for (int t = 0; t < ZIO_TYPES; t++) { 1129 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1130 spa_taskqs_fini(spa, t, q); 1131 } 1132 } 1133 1134 metaslab_class_destroy(spa->spa_normal_class); 1135 spa->spa_normal_class = NULL; 1136 1137 metaslab_class_destroy(spa->spa_log_class); 1138 spa->spa_log_class = NULL; 1139 1140 /* 1141 * If this was part of an import or the open otherwise failed, we may 1142 * still have errors left in the queues. Empty them just in case. 1143 */ 1144 spa_errlog_drain(spa); 1145 1146 avl_destroy(&spa->spa_errlist_scrub); 1147 avl_destroy(&spa->spa_errlist_last); 1148 1149 spa->spa_state = POOL_STATE_UNINITIALIZED; 1150 1151 mutex_enter(&spa->spa_proc_lock); 1152 if (spa->spa_proc_state != SPA_PROC_NONE) { 1153 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1154 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1155 cv_broadcast(&spa->spa_proc_cv); 1156 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1157 ASSERT(spa->spa_proc != &p0); 1158 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1159 } 1160 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1161 spa->spa_proc_state = SPA_PROC_NONE; 1162 } 1163 ASSERT(spa->spa_proc == &p0); 1164 mutex_exit(&spa->spa_proc_lock); 1165 1166 /* 1167 * We want to make sure spa_thread() has actually exited the ZFS 1168 * module, so that the module can't be unloaded out from underneath 1169 * it. 1170 */ 1171 if (spa->spa_did != 0) { 1172 thread_join(spa->spa_did); 1173 spa->spa_did = 0; 1174 } 1175 } 1176 1177 /* 1178 * Verify a pool configuration, and construct the vdev tree appropriately. This 1179 * will create all the necessary vdevs in the appropriate layout, with each vdev 1180 * in the CLOSED state. This will prep the pool before open/creation/import. 1181 * All vdev validation is done by the vdev_alloc() routine. 1182 */ 1183 static int 1184 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1185 uint_t id, int atype) 1186 { 1187 nvlist_t **child; 1188 uint_t children; 1189 int error; 1190 1191 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1192 return (error); 1193 1194 if ((*vdp)->vdev_ops->vdev_op_leaf) 1195 return (0); 1196 1197 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1198 &child, &children); 1199 1200 if (error == ENOENT) 1201 return (0); 1202 1203 if (error) { 1204 vdev_free(*vdp); 1205 *vdp = NULL; 1206 return (SET_ERROR(EINVAL)); 1207 } 1208 1209 for (int c = 0; c < children; c++) { 1210 vdev_t *vd; 1211 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1212 atype)) != 0) { 1213 vdev_free(*vdp); 1214 *vdp = NULL; 1215 return (error); 1216 } 1217 } 1218 1219 ASSERT(*vdp != NULL); 1220 1221 return (0); 1222 } 1223 1224 /* 1225 * Opposite of spa_load(). 1226 */ 1227 static void 1228 spa_unload(spa_t *spa) 1229 { 1230 int i; 1231 1232 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1233 1234 /* 1235 * Stop async tasks. 1236 */ 1237 spa_async_suspend(spa); 1238 1239 /* 1240 * Stop syncing. 1241 */ 1242 if (spa->spa_sync_on) { 1243 txg_sync_stop(spa->spa_dsl_pool); 1244 spa->spa_sync_on = B_FALSE; 1245 } 1246 1247 /* 1248 * Wait for any outstanding async I/O to complete. 1249 */ 1250 if (spa->spa_async_zio_root != NULL) { 1251 for (int i = 0; i < max_ncpus; i++) 1252 (void) zio_wait(spa->spa_async_zio_root[i]); 1253 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1254 spa->spa_async_zio_root = NULL; 1255 } 1256 1257 bpobj_close(&spa->spa_deferred_bpobj); 1258 1259 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1260 1261 /* 1262 * Close all vdevs. 1263 */ 1264 if (spa->spa_root_vdev) 1265 vdev_free(spa->spa_root_vdev); 1266 ASSERT(spa->spa_root_vdev == NULL); 1267 1268 /* 1269 * Close the dsl pool. 1270 */ 1271 if (spa->spa_dsl_pool) { 1272 dsl_pool_close(spa->spa_dsl_pool); 1273 spa->spa_dsl_pool = NULL; 1274 spa->spa_meta_objset = NULL; 1275 } 1276 1277 ddt_unload(spa); 1278 1279 1280 /* 1281 * Drop and purge level 2 cache 1282 */ 1283 spa_l2cache_drop(spa); 1284 1285 for (i = 0; i < spa->spa_spares.sav_count; i++) 1286 vdev_free(spa->spa_spares.sav_vdevs[i]); 1287 if (spa->spa_spares.sav_vdevs) { 1288 kmem_free(spa->spa_spares.sav_vdevs, 1289 spa->spa_spares.sav_count * sizeof (void *)); 1290 spa->spa_spares.sav_vdevs = NULL; 1291 } 1292 if (spa->spa_spares.sav_config) { 1293 nvlist_free(spa->spa_spares.sav_config); 1294 spa->spa_spares.sav_config = NULL; 1295 } 1296 spa->spa_spares.sav_count = 0; 1297 1298 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1299 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1300 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1301 } 1302 if (spa->spa_l2cache.sav_vdevs) { 1303 kmem_free(spa->spa_l2cache.sav_vdevs, 1304 spa->spa_l2cache.sav_count * sizeof (void *)); 1305 spa->spa_l2cache.sav_vdevs = NULL; 1306 } 1307 if (spa->spa_l2cache.sav_config) { 1308 nvlist_free(spa->spa_l2cache.sav_config); 1309 spa->spa_l2cache.sav_config = NULL; 1310 } 1311 spa->spa_l2cache.sav_count = 0; 1312 1313 spa->spa_async_suspended = 0; 1314 1315 if (spa->spa_comment != NULL) { 1316 spa_strfree(spa->spa_comment); 1317 spa->spa_comment = NULL; 1318 } 1319 1320 spa_config_exit(spa, SCL_ALL, FTAG); 1321 } 1322 1323 /* 1324 * Load (or re-load) the current list of vdevs describing the active spares for 1325 * this pool. When this is called, we have some form of basic information in 1326 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1327 * then re-generate a more complete list including status information. 1328 */ 1329 static void 1330 spa_load_spares(spa_t *spa) 1331 { 1332 nvlist_t **spares; 1333 uint_t nspares; 1334 int i; 1335 vdev_t *vd, *tvd; 1336 1337 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1338 1339 /* 1340 * First, close and free any existing spare vdevs. 1341 */ 1342 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1343 vd = spa->spa_spares.sav_vdevs[i]; 1344 1345 /* Undo the call to spa_activate() below */ 1346 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1347 B_FALSE)) != NULL && tvd->vdev_isspare) 1348 spa_spare_remove(tvd); 1349 vdev_close(vd); 1350 vdev_free(vd); 1351 } 1352 1353 if (spa->spa_spares.sav_vdevs) 1354 kmem_free(spa->spa_spares.sav_vdevs, 1355 spa->spa_spares.sav_count * sizeof (void *)); 1356 1357 if (spa->spa_spares.sav_config == NULL) 1358 nspares = 0; 1359 else 1360 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1361 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1362 1363 spa->spa_spares.sav_count = (int)nspares; 1364 spa->spa_spares.sav_vdevs = NULL; 1365 1366 if (nspares == 0) 1367 return; 1368 1369 /* 1370 * Construct the array of vdevs, opening them to get status in the 1371 * process. For each spare, there is potentially two different vdev_t 1372 * structures associated with it: one in the list of spares (used only 1373 * for basic validation purposes) and one in the active vdev 1374 * configuration (if it's spared in). During this phase we open and 1375 * validate each vdev on the spare list. If the vdev also exists in the 1376 * active configuration, then we also mark this vdev as an active spare. 1377 */ 1378 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1379 KM_SLEEP); 1380 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1381 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1382 VDEV_ALLOC_SPARE) == 0); 1383 ASSERT(vd != NULL); 1384 1385 spa->spa_spares.sav_vdevs[i] = vd; 1386 1387 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1388 B_FALSE)) != NULL) { 1389 if (!tvd->vdev_isspare) 1390 spa_spare_add(tvd); 1391 1392 /* 1393 * We only mark the spare active if we were successfully 1394 * able to load the vdev. Otherwise, importing a pool 1395 * with a bad active spare would result in strange 1396 * behavior, because multiple pool would think the spare 1397 * is actively in use. 1398 * 1399 * There is a vulnerability here to an equally bizarre 1400 * circumstance, where a dead active spare is later 1401 * brought back to life (onlined or otherwise). Given 1402 * the rarity of this scenario, and the extra complexity 1403 * it adds, we ignore the possibility. 1404 */ 1405 if (!vdev_is_dead(tvd)) 1406 spa_spare_activate(tvd); 1407 } 1408 1409 vd->vdev_top = vd; 1410 vd->vdev_aux = &spa->spa_spares; 1411 1412 if (vdev_open(vd) != 0) 1413 continue; 1414 1415 if (vdev_validate_aux(vd) == 0) 1416 spa_spare_add(vd); 1417 } 1418 1419 /* 1420 * Recompute the stashed list of spares, with status information 1421 * this time. 1422 */ 1423 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1424 DATA_TYPE_NVLIST_ARRAY) == 0); 1425 1426 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1427 KM_SLEEP); 1428 for (i = 0; i < spa->spa_spares.sav_count; i++) 1429 spares[i] = vdev_config_generate(spa, 1430 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1431 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1432 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1433 for (i = 0; i < spa->spa_spares.sav_count; i++) 1434 nvlist_free(spares[i]); 1435 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1436 } 1437 1438 /* 1439 * Load (or re-load) the current list of vdevs describing the active l2cache for 1440 * this pool. When this is called, we have some form of basic information in 1441 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1442 * then re-generate a more complete list including status information. 1443 * Devices which are already active have their details maintained, and are 1444 * not re-opened. 1445 */ 1446 static void 1447 spa_load_l2cache(spa_t *spa) 1448 { 1449 nvlist_t **l2cache; 1450 uint_t nl2cache; 1451 int i, j, oldnvdevs; 1452 uint64_t guid; 1453 vdev_t *vd, **oldvdevs, **newvdevs; 1454 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1455 1456 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1457 1458 if (sav->sav_config != NULL) { 1459 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1460 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1461 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1462 } else { 1463 nl2cache = 0; 1464 newvdevs = NULL; 1465 } 1466 1467 oldvdevs = sav->sav_vdevs; 1468 oldnvdevs = sav->sav_count; 1469 sav->sav_vdevs = NULL; 1470 sav->sav_count = 0; 1471 1472 /* 1473 * Process new nvlist of vdevs. 1474 */ 1475 for (i = 0; i < nl2cache; i++) { 1476 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1477 &guid) == 0); 1478 1479 newvdevs[i] = NULL; 1480 for (j = 0; j < oldnvdevs; j++) { 1481 vd = oldvdevs[j]; 1482 if (vd != NULL && guid == vd->vdev_guid) { 1483 /* 1484 * Retain previous vdev for add/remove ops. 1485 */ 1486 newvdevs[i] = vd; 1487 oldvdevs[j] = NULL; 1488 break; 1489 } 1490 } 1491 1492 if (newvdevs[i] == NULL) { 1493 /* 1494 * Create new vdev 1495 */ 1496 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1497 VDEV_ALLOC_L2CACHE) == 0); 1498 ASSERT(vd != NULL); 1499 newvdevs[i] = vd; 1500 1501 /* 1502 * Commit this vdev as an l2cache device, 1503 * even if it fails to open. 1504 */ 1505 spa_l2cache_add(vd); 1506 1507 vd->vdev_top = vd; 1508 vd->vdev_aux = sav; 1509 1510 spa_l2cache_activate(vd); 1511 1512 if (vdev_open(vd) != 0) 1513 continue; 1514 1515 (void) vdev_validate_aux(vd); 1516 1517 if (!vdev_is_dead(vd)) { 1518 boolean_t do_rebuild = B_FALSE; 1519 1520 (void) nvlist_lookup_boolean_value(l2cache[i], 1521 ZPOOL_CONFIG_L2CACHE_PERSISTENT, 1522 &do_rebuild); 1523 l2arc_add_vdev(spa, vd, do_rebuild); 1524 } 1525 } 1526 } 1527 1528 /* 1529 * Purge vdevs that were dropped 1530 */ 1531 for (i = 0; i < oldnvdevs; i++) { 1532 uint64_t pool; 1533 1534 vd = oldvdevs[i]; 1535 if (vd != NULL) { 1536 ASSERT(vd->vdev_isl2cache); 1537 1538 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1539 pool != 0ULL && l2arc_vdev_present(vd)) 1540 l2arc_remove_vdev(vd); 1541 vdev_clear_stats(vd); 1542 vdev_free(vd); 1543 } 1544 } 1545 1546 if (oldvdevs) 1547 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1548 1549 if (sav->sav_config == NULL) 1550 goto out; 1551 1552 sav->sav_vdevs = newvdevs; 1553 sav->sav_count = (int)nl2cache; 1554 1555 /* 1556 * Recompute the stashed list of l2cache devices, with status 1557 * information this time. 1558 */ 1559 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1560 DATA_TYPE_NVLIST_ARRAY) == 0); 1561 1562 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1563 for (i = 0; i < sav->sav_count; i++) 1564 l2cache[i] = vdev_config_generate(spa, 1565 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1566 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1567 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1568 out: 1569 for (i = 0; i < sav->sav_count; i++) 1570 nvlist_free(l2cache[i]); 1571 if (sav->sav_count) 1572 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1573 } 1574 1575 static int 1576 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1577 { 1578 dmu_buf_t *db; 1579 char *packed = NULL; 1580 size_t nvsize = 0; 1581 int error; 1582 *value = NULL; 1583 1584 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1585 if (error != 0) 1586 return (error); 1587 1588 nvsize = *(uint64_t *)db->db_data; 1589 dmu_buf_rele(db, FTAG); 1590 1591 packed = kmem_alloc(nvsize, KM_SLEEP); 1592 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1593 DMU_READ_PREFETCH); 1594 if (error == 0) 1595 error = nvlist_unpack(packed, nvsize, value, 0); 1596 kmem_free(packed, nvsize); 1597 1598 return (error); 1599 } 1600 1601 /* 1602 * Checks to see if the given vdev could not be opened, in which case we post a 1603 * sysevent to notify the autoreplace code that the device has been removed. 1604 */ 1605 static void 1606 spa_check_removed(vdev_t *vd) 1607 { 1608 for (int c = 0; c < vd->vdev_children; c++) 1609 spa_check_removed(vd->vdev_child[c]); 1610 1611 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1612 !vd->vdev_ishole) { 1613 zfs_post_autoreplace(vd->vdev_spa, vd); 1614 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1615 } 1616 } 1617 1618 /* 1619 * Validate the current config against the MOS config 1620 */ 1621 static boolean_t 1622 spa_config_valid(spa_t *spa, nvlist_t *config) 1623 { 1624 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1625 nvlist_t *nv; 1626 1627 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1628 1629 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1630 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1631 1632 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1633 1634 /* 1635 * If we're doing a normal import, then build up any additional 1636 * diagnostic information about missing devices in this config. 1637 * We'll pass this up to the user for further processing. 1638 */ 1639 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1640 nvlist_t **child, *nv; 1641 uint64_t idx = 0; 1642 1643 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1644 KM_SLEEP); 1645 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1646 1647 for (int c = 0; c < rvd->vdev_children; c++) { 1648 vdev_t *tvd = rvd->vdev_child[c]; 1649 vdev_t *mtvd = mrvd->vdev_child[c]; 1650 1651 if (tvd->vdev_ops == &vdev_missing_ops && 1652 mtvd->vdev_ops != &vdev_missing_ops && 1653 mtvd->vdev_islog) 1654 child[idx++] = vdev_config_generate(spa, mtvd, 1655 B_FALSE, 0); 1656 } 1657 1658 if (idx) { 1659 VERIFY(nvlist_add_nvlist_array(nv, 1660 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1661 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1662 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1663 1664 for (int i = 0; i < idx; i++) 1665 nvlist_free(child[i]); 1666 } 1667 nvlist_free(nv); 1668 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1669 } 1670 1671 /* 1672 * Compare the root vdev tree with the information we have 1673 * from the MOS config (mrvd). Check each top-level vdev 1674 * with the corresponding MOS config top-level (mtvd). 1675 */ 1676 for (int c = 0; c < rvd->vdev_children; c++) { 1677 vdev_t *tvd = rvd->vdev_child[c]; 1678 vdev_t *mtvd = mrvd->vdev_child[c]; 1679 1680 /* 1681 * Resolve any "missing" vdevs in the current configuration. 1682 * If we find that the MOS config has more accurate information 1683 * about the top-level vdev then use that vdev instead. 1684 */ 1685 if (tvd->vdev_ops == &vdev_missing_ops && 1686 mtvd->vdev_ops != &vdev_missing_ops) { 1687 1688 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1689 continue; 1690 1691 /* 1692 * Device specific actions. 1693 */ 1694 if (mtvd->vdev_islog) { 1695 spa_set_log_state(spa, SPA_LOG_CLEAR); 1696 } else { 1697 /* 1698 * XXX - once we have 'readonly' pool 1699 * support we should be able to handle 1700 * missing data devices by transitioning 1701 * the pool to readonly. 1702 */ 1703 continue; 1704 } 1705 1706 /* 1707 * Swap the missing vdev with the data we were 1708 * able to obtain from the MOS config. 1709 */ 1710 vdev_remove_child(rvd, tvd); 1711 vdev_remove_child(mrvd, mtvd); 1712 1713 vdev_add_child(rvd, mtvd); 1714 vdev_add_child(mrvd, tvd); 1715 1716 spa_config_exit(spa, SCL_ALL, FTAG); 1717 vdev_load(mtvd); 1718 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1719 1720 vdev_reopen(rvd); 1721 } else if (mtvd->vdev_islog) { 1722 /* 1723 * Load the slog device's state from the MOS config 1724 * since it's possible that the label does not 1725 * contain the most up-to-date information. 1726 */ 1727 vdev_load_log_state(tvd, mtvd); 1728 vdev_reopen(tvd); 1729 } 1730 } 1731 vdev_free(mrvd); 1732 spa_config_exit(spa, SCL_ALL, FTAG); 1733 1734 /* 1735 * Ensure we were able to validate the config. 1736 */ 1737 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1738 } 1739 1740 /* 1741 * Check for missing log devices 1742 */ 1743 static boolean_t 1744 spa_check_logs(spa_t *spa) 1745 { 1746 boolean_t rv = B_FALSE; 1747 dsl_pool_t *dp = spa_get_dsl(spa); 1748 1749 switch (spa->spa_log_state) { 1750 case SPA_LOG_MISSING: 1751 /* need to recheck in case slog has been restored */ 1752 case SPA_LOG_UNKNOWN: 1753 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1754 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1755 if (rv) 1756 spa_set_log_state(spa, SPA_LOG_MISSING); 1757 break; 1758 } 1759 return (rv); 1760 } 1761 1762 static boolean_t 1763 spa_passivate_log(spa_t *spa) 1764 { 1765 vdev_t *rvd = spa->spa_root_vdev; 1766 boolean_t slog_found = B_FALSE; 1767 1768 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1769 1770 if (!spa_has_slogs(spa)) 1771 return (B_FALSE); 1772 1773 for (int c = 0; c < rvd->vdev_children; c++) { 1774 vdev_t *tvd = rvd->vdev_child[c]; 1775 metaslab_group_t *mg = tvd->vdev_mg; 1776 1777 if (tvd->vdev_islog) { 1778 metaslab_group_passivate(mg); 1779 slog_found = B_TRUE; 1780 } 1781 } 1782 1783 return (slog_found); 1784 } 1785 1786 static void 1787 spa_activate_log(spa_t *spa) 1788 { 1789 vdev_t *rvd = spa->spa_root_vdev; 1790 1791 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1792 1793 for (int c = 0; c < rvd->vdev_children; c++) { 1794 vdev_t *tvd = rvd->vdev_child[c]; 1795 metaslab_group_t *mg = tvd->vdev_mg; 1796 1797 if (tvd->vdev_islog) 1798 metaslab_group_activate(mg); 1799 } 1800 } 1801 1802 int 1803 spa_offline_log(spa_t *spa) 1804 { 1805 int error; 1806 1807 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1808 NULL, DS_FIND_CHILDREN); 1809 if (error == 0) { 1810 /* 1811 * We successfully offlined the log device, sync out the 1812 * current txg so that the "stubby" block can be removed 1813 * by zil_sync(). 1814 */ 1815 txg_wait_synced(spa->spa_dsl_pool, 0); 1816 } 1817 return (error); 1818 } 1819 1820 static void 1821 spa_aux_check_removed(spa_aux_vdev_t *sav) 1822 { 1823 for (int i = 0; i < sav->sav_count; i++) 1824 spa_check_removed(sav->sav_vdevs[i]); 1825 } 1826 1827 void 1828 spa_claim_notify(zio_t *zio) 1829 { 1830 spa_t *spa = zio->io_spa; 1831 1832 if (zio->io_error) 1833 return; 1834 1835 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1836 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1837 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1838 mutex_exit(&spa->spa_props_lock); 1839 } 1840 1841 typedef struct spa_load_error { 1842 uint64_t sle_meta_count; 1843 uint64_t sle_data_count; 1844 } spa_load_error_t; 1845 1846 static void 1847 spa_load_verify_done(zio_t *zio) 1848 { 1849 blkptr_t *bp = zio->io_bp; 1850 spa_load_error_t *sle = zio->io_private; 1851 dmu_object_type_t type = BP_GET_TYPE(bp); 1852 int error = zio->io_error; 1853 spa_t *spa = zio->io_spa; 1854 1855 if (error) { 1856 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1857 type != DMU_OT_INTENT_LOG) 1858 atomic_inc_64(&sle->sle_meta_count); 1859 else 1860 atomic_inc_64(&sle->sle_data_count); 1861 } 1862 zio_data_buf_free(zio->io_data, zio->io_size); 1863 1864 mutex_enter(&spa->spa_scrub_lock); 1865 spa->spa_scrub_inflight--; 1866 cv_broadcast(&spa->spa_scrub_io_cv); 1867 mutex_exit(&spa->spa_scrub_lock); 1868 } 1869 1870 /* 1871 * Maximum number of concurrent scrub i/os to create while verifying 1872 * a pool while importing it. 1873 */ 1874 int spa_load_verify_maxinflight = 10000; 1875 boolean_t spa_load_verify_metadata = B_TRUE; 1876 boolean_t spa_load_verify_data = B_TRUE; 1877 1878 /*ARGSUSED*/ 1879 static int 1880 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1881 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1882 { 1883 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1884 return (0); 1885 /* 1886 * Note: normally this routine will not be called if 1887 * spa_load_verify_metadata is not set. However, it may be useful 1888 * to manually set the flag after the traversal has begun. 1889 */ 1890 if (!spa_load_verify_metadata) 1891 return (0); 1892 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1893 return (0); 1894 1895 zio_t *rio = arg; 1896 size_t size = BP_GET_PSIZE(bp); 1897 void *data = zio_data_buf_alloc(size); 1898 1899 mutex_enter(&spa->spa_scrub_lock); 1900 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1901 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1902 spa->spa_scrub_inflight++; 1903 mutex_exit(&spa->spa_scrub_lock); 1904 1905 zio_nowait(zio_read(rio, spa, bp, data, size, 1906 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1907 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1908 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1909 return (0); 1910 } 1911 1912 static int 1913 spa_load_verify(spa_t *spa) 1914 { 1915 zio_t *rio; 1916 spa_load_error_t sle = { 0 }; 1917 zpool_rewind_policy_t policy; 1918 boolean_t verify_ok = B_FALSE; 1919 int error = 0; 1920 1921 zpool_get_rewind_policy(spa->spa_config, &policy); 1922 1923 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1924 return (0); 1925 1926 rio = zio_root(spa, NULL, &sle, 1927 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1928 1929 if (spa_load_verify_metadata) { 1930 error = traverse_pool(spa, spa->spa_verify_min_txg, 1931 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1932 spa_load_verify_cb, rio); 1933 } 1934 1935 (void) zio_wait(rio); 1936 1937 spa->spa_load_meta_errors = sle.sle_meta_count; 1938 spa->spa_load_data_errors = sle.sle_data_count; 1939 1940 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1941 sle.sle_data_count <= policy.zrp_maxdata) { 1942 int64_t loss = 0; 1943 1944 verify_ok = B_TRUE; 1945 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1946 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1947 1948 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1949 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1950 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1951 VERIFY(nvlist_add_int64(spa->spa_load_info, 1952 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1953 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1954 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1955 } else { 1956 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1957 } 1958 1959 if (error) { 1960 if (error != ENXIO && error != EIO) 1961 error = SET_ERROR(EIO); 1962 return (error); 1963 } 1964 1965 return (verify_ok ? 0 : EIO); 1966 } 1967 1968 /* 1969 * Find a value in the pool props object. 1970 */ 1971 static void 1972 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1973 { 1974 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1975 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1976 } 1977 1978 /* 1979 * Find a value in the pool directory object. 1980 */ 1981 static int 1982 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1983 { 1984 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1985 name, sizeof (uint64_t), 1, val)); 1986 } 1987 1988 static int 1989 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1990 { 1991 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1992 return (err); 1993 } 1994 1995 /* 1996 * Fix up config after a partly-completed split. This is done with the 1997 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1998 * pool have that entry in their config, but only the splitting one contains 1999 * a list of all the guids of the vdevs that are being split off. 2000 * 2001 * This function determines what to do with that list: either rejoin 2002 * all the disks to the pool, or complete the splitting process. To attempt 2003 * the rejoin, each disk that is offlined is marked online again, and 2004 * we do a reopen() call. If the vdev label for every disk that was 2005 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2006 * then we call vdev_split() on each disk, and complete the split. 2007 * 2008 * Otherwise we leave the config alone, with all the vdevs in place in 2009 * the original pool. 2010 */ 2011 static void 2012 spa_try_repair(spa_t *spa, nvlist_t *config) 2013 { 2014 uint_t extracted; 2015 uint64_t *glist; 2016 uint_t i, gcount; 2017 nvlist_t *nvl; 2018 vdev_t **vd; 2019 boolean_t attempt_reopen; 2020 2021 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2022 return; 2023 2024 /* check that the config is complete */ 2025 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2026 &glist, &gcount) != 0) 2027 return; 2028 2029 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2030 2031 /* attempt to online all the vdevs & validate */ 2032 attempt_reopen = B_TRUE; 2033 for (i = 0; i < gcount; i++) { 2034 if (glist[i] == 0) /* vdev is hole */ 2035 continue; 2036 2037 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2038 if (vd[i] == NULL) { 2039 /* 2040 * Don't bother attempting to reopen the disks; 2041 * just do the split. 2042 */ 2043 attempt_reopen = B_FALSE; 2044 } else { 2045 /* attempt to re-online it */ 2046 vd[i]->vdev_offline = B_FALSE; 2047 } 2048 } 2049 2050 if (attempt_reopen) { 2051 vdev_reopen(spa->spa_root_vdev); 2052 2053 /* check each device to see what state it's in */ 2054 for (extracted = 0, i = 0; i < gcount; i++) { 2055 if (vd[i] != NULL && 2056 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2057 break; 2058 ++extracted; 2059 } 2060 } 2061 2062 /* 2063 * If every disk has been moved to the new pool, or if we never 2064 * even attempted to look at them, then we split them off for 2065 * good. 2066 */ 2067 if (!attempt_reopen || gcount == extracted) { 2068 for (i = 0; i < gcount; i++) 2069 if (vd[i] != NULL) 2070 vdev_split(vd[i]); 2071 vdev_reopen(spa->spa_root_vdev); 2072 } 2073 2074 kmem_free(vd, gcount * sizeof (vdev_t *)); 2075 } 2076 2077 static int 2078 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2079 boolean_t mosconfig) 2080 { 2081 nvlist_t *config = spa->spa_config; 2082 char *ereport = FM_EREPORT_ZFS_POOL; 2083 char *comment; 2084 int error; 2085 uint64_t pool_guid; 2086 nvlist_t *nvl; 2087 2088 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2089 return (SET_ERROR(EINVAL)); 2090 2091 ASSERT(spa->spa_comment == NULL); 2092 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2093 spa->spa_comment = spa_strdup(comment); 2094 2095 /* 2096 * Versioning wasn't explicitly added to the label until later, so if 2097 * it's not present treat it as the initial version. 2098 */ 2099 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2100 &spa->spa_ubsync.ub_version) != 0) 2101 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2102 2103 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2104 &spa->spa_config_txg); 2105 2106 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2107 spa_guid_exists(pool_guid, 0)) { 2108 error = SET_ERROR(EEXIST); 2109 } else { 2110 spa->spa_config_guid = pool_guid; 2111 2112 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2113 &nvl) == 0) { 2114 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2115 KM_SLEEP) == 0); 2116 } 2117 2118 nvlist_free(spa->spa_load_info); 2119 spa->spa_load_info = fnvlist_alloc(); 2120 2121 gethrestime(&spa->spa_loaded_ts); 2122 error = spa_load_impl(spa, pool_guid, config, state, type, 2123 mosconfig, &ereport); 2124 } 2125 2126 /* 2127 * Don't count references from objsets that are already closed 2128 * and are making their way through the eviction process. 2129 */ 2130 spa_evicting_os_wait(spa); 2131 spa->spa_minref = refcount_count(&spa->spa_refcount); 2132 if (error) { 2133 if (error != EEXIST) { 2134 spa->spa_loaded_ts.tv_sec = 0; 2135 spa->spa_loaded_ts.tv_nsec = 0; 2136 } 2137 if (error != EBADF) { 2138 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2139 } 2140 } 2141 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2142 spa->spa_ena = 0; 2143 2144 return (error); 2145 } 2146 2147 /* 2148 * Load an existing storage pool, using the pool's builtin spa_config as a 2149 * source of configuration information. 2150 */ 2151 static int 2152 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2153 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2154 char **ereport) 2155 { 2156 int error = 0; 2157 nvlist_t *nvroot = NULL; 2158 nvlist_t *label; 2159 vdev_t *rvd; 2160 uberblock_t *ub = &spa->spa_uberblock; 2161 uint64_t children, config_cache_txg = spa->spa_config_txg; 2162 int orig_mode = spa->spa_mode; 2163 int parse; 2164 uint64_t obj; 2165 boolean_t missing_feat_write = B_FALSE; 2166 2167 /* 2168 * If this is an untrusted config, access the pool in read-only mode. 2169 * This prevents things like resilvering recently removed devices. 2170 */ 2171 if (!mosconfig) 2172 spa->spa_mode = FREAD; 2173 2174 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2175 2176 spa->spa_load_state = state; 2177 2178 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2179 return (SET_ERROR(EINVAL)); 2180 2181 parse = (type == SPA_IMPORT_EXISTING ? 2182 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2183 2184 /* 2185 * Create "The Godfather" zio to hold all async IOs 2186 */ 2187 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2188 KM_SLEEP); 2189 for (int i = 0; i < max_ncpus; i++) { 2190 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2191 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2192 ZIO_FLAG_GODFATHER); 2193 } 2194 2195 /* 2196 * Parse the configuration into a vdev tree. We explicitly set the 2197 * value that will be returned by spa_version() since parsing the 2198 * configuration requires knowing the version number. 2199 */ 2200 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2201 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2202 spa_config_exit(spa, SCL_ALL, FTAG); 2203 2204 if (error != 0) 2205 return (error); 2206 2207 ASSERT(spa->spa_root_vdev == rvd); 2208 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2209 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2210 2211 if (type != SPA_IMPORT_ASSEMBLE) { 2212 ASSERT(spa_guid(spa) == pool_guid); 2213 } 2214 2215 /* 2216 * Try to open all vdevs, loading each label in the process. 2217 */ 2218 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2219 error = vdev_open(rvd); 2220 spa_config_exit(spa, SCL_ALL, FTAG); 2221 if (error != 0) 2222 return (error); 2223 2224 /* 2225 * We need to validate the vdev labels against the configuration that 2226 * we have in hand, which is dependent on the setting of mosconfig. If 2227 * mosconfig is true then we're validating the vdev labels based on 2228 * that config. Otherwise, we're validating against the cached config 2229 * (zpool.cache) that was read when we loaded the zfs module, and then 2230 * later we will recursively call spa_load() and validate against 2231 * the vdev config. 2232 * 2233 * If we're assembling a new pool that's been split off from an 2234 * existing pool, the labels haven't yet been updated so we skip 2235 * validation for now. 2236 */ 2237 if (type != SPA_IMPORT_ASSEMBLE) { 2238 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2239 error = vdev_validate(rvd, mosconfig); 2240 spa_config_exit(spa, SCL_ALL, FTAG); 2241 2242 if (error != 0) 2243 return (error); 2244 2245 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2246 return (SET_ERROR(ENXIO)); 2247 } 2248 2249 /* 2250 * Find the best uberblock. 2251 */ 2252 vdev_uberblock_load(rvd, ub, &label); 2253 2254 /* 2255 * If we weren't able to find a single valid uberblock, return failure. 2256 */ 2257 if (ub->ub_txg == 0) { 2258 nvlist_free(label); 2259 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2260 } 2261 2262 /* 2263 * If the pool has an unsupported version we can't open it. 2264 */ 2265 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2266 nvlist_free(label); 2267 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2268 } 2269 2270 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2271 nvlist_t *features; 2272 2273 /* 2274 * If we weren't able to find what's necessary for reading the 2275 * MOS in the label, return failure. 2276 */ 2277 if (label == NULL || nvlist_lookup_nvlist(label, 2278 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2279 nvlist_free(label); 2280 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2281 ENXIO)); 2282 } 2283 2284 /* 2285 * Update our in-core representation with the definitive values 2286 * from the label. 2287 */ 2288 nvlist_free(spa->spa_label_features); 2289 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2290 } 2291 2292 nvlist_free(label); 2293 2294 /* 2295 * Look through entries in the label nvlist's features_for_read. If 2296 * there is a feature listed there which we don't understand then we 2297 * cannot open a pool. 2298 */ 2299 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2300 nvlist_t *unsup_feat; 2301 2302 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2303 0); 2304 2305 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2306 NULL); nvp != NULL; 2307 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2308 if (!zfeature_is_supported(nvpair_name(nvp))) { 2309 VERIFY(nvlist_add_string(unsup_feat, 2310 nvpair_name(nvp), "") == 0); 2311 } 2312 } 2313 2314 if (!nvlist_empty(unsup_feat)) { 2315 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2316 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2317 nvlist_free(unsup_feat); 2318 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2319 ENOTSUP)); 2320 } 2321 2322 nvlist_free(unsup_feat); 2323 } 2324 2325 /* 2326 * If the vdev guid sum doesn't match the uberblock, we have an 2327 * incomplete configuration. We first check to see if the pool 2328 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2329 * If it is, defer the vdev_guid_sum check till later so we 2330 * can handle missing vdevs. 2331 */ 2332 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2333 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2334 rvd->vdev_guid_sum != ub->ub_guid_sum) 2335 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2336 2337 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2338 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2339 spa_try_repair(spa, config); 2340 spa_config_exit(spa, SCL_ALL, FTAG); 2341 nvlist_free(spa->spa_config_splitting); 2342 spa->spa_config_splitting = NULL; 2343 } 2344 2345 /* 2346 * Initialize internal SPA structures. 2347 */ 2348 spa->spa_state = POOL_STATE_ACTIVE; 2349 spa->spa_ubsync = spa->spa_uberblock; 2350 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2351 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2352 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2353 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2354 spa->spa_claim_max_txg = spa->spa_first_txg; 2355 spa->spa_prev_software_version = ub->ub_software_version; 2356 2357 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2358 if (error) 2359 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2360 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2361 2362 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2363 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2364 2365 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2366 boolean_t missing_feat_read = B_FALSE; 2367 nvlist_t *unsup_feat, *enabled_feat; 2368 2369 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2370 &spa->spa_feat_for_read_obj) != 0) { 2371 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2372 } 2373 2374 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2375 &spa->spa_feat_for_write_obj) != 0) { 2376 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2377 } 2378 2379 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2380 &spa->spa_feat_desc_obj) != 0) { 2381 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2382 } 2383 2384 enabled_feat = fnvlist_alloc(); 2385 unsup_feat = fnvlist_alloc(); 2386 2387 if (!spa_features_check(spa, B_FALSE, 2388 unsup_feat, enabled_feat)) 2389 missing_feat_read = B_TRUE; 2390 2391 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2392 if (!spa_features_check(spa, B_TRUE, 2393 unsup_feat, enabled_feat)) { 2394 missing_feat_write = B_TRUE; 2395 } 2396 } 2397 2398 fnvlist_add_nvlist(spa->spa_load_info, 2399 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2400 2401 if (!nvlist_empty(unsup_feat)) { 2402 fnvlist_add_nvlist(spa->spa_load_info, 2403 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2404 } 2405 2406 fnvlist_free(enabled_feat); 2407 fnvlist_free(unsup_feat); 2408 2409 if (!missing_feat_read) { 2410 fnvlist_add_boolean(spa->spa_load_info, 2411 ZPOOL_CONFIG_CAN_RDONLY); 2412 } 2413 2414 /* 2415 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2416 * twofold: to determine whether the pool is available for 2417 * import in read-write mode and (if it is not) whether the 2418 * pool is available for import in read-only mode. If the pool 2419 * is available for import in read-write mode, it is displayed 2420 * as available in userland; if it is not available for import 2421 * in read-only mode, it is displayed as unavailable in 2422 * userland. If the pool is available for import in read-only 2423 * mode but not read-write mode, it is displayed as unavailable 2424 * in userland with a special note that the pool is actually 2425 * available for open in read-only mode. 2426 * 2427 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2428 * missing a feature for write, we must first determine whether 2429 * the pool can be opened read-only before returning to 2430 * userland in order to know whether to display the 2431 * abovementioned note. 2432 */ 2433 if (missing_feat_read || (missing_feat_write && 2434 spa_writeable(spa))) { 2435 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2436 ENOTSUP)); 2437 } 2438 2439 /* 2440 * Load refcounts for ZFS features from disk into an in-memory 2441 * cache during SPA initialization. 2442 */ 2443 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2444 uint64_t refcount; 2445 2446 error = feature_get_refcount_from_disk(spa, 2447 &spa_feature_table[i], &refcount); 2448 if (error == 0) { 2449 spa->spa_feat_refcount_cache[i] = refcount; 2450 } else if (error == ENOTSUP) { 2451 spa->spa_feat_refcount_cache[i] = 2452 SPA_FEATURE_DISABLED; 2453 } else { 2454 return (spa_vdev_err(rvd, 2455 VDEV_AUX_CORRUPT_DATA, EIO)); 2456 } 2457 } 2458 } 2459 2460 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2461 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2462 &spa->spa_feat_enabled_txg_obj) != 0) 2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2464 } 2465 2466 spa->spa_is_initializing = B_TRUE; 2467 error = dsl_pool_open(spa->spa_dsl_pool); 2468 spa->spa_is_initializing = B_FALSE; 2469 if (error != 0) 2470 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2471 2472 if (!mosconfig) { 2473 uint64_t hostid; 2474 nvlist_t *policy = NULL, *nvconfig; 2475 2476 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2477 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2478 2479 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2480 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2481 char *hostname; 2482 unsigned long myhostid = 0; 2483 2484 VERIFY(nvlist_lookup_string(nvconfig, 2485 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2486 2487 #ifdef _KERNEL 2488 myhostid = zone_get_hostid(NULL); 2489 #else /* _KERNEL */ 2490 /* 2491 * We're emulating the system's hostid in userland, so 2492 * we can't use zone_get_hostid(). 2493 */ 2494 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2495 #endif /* _KERNEL */ 2496 if (hostid != 0 && myhostid != 0 && 2497 hostid != myhostid) { 2498 nvlist_free(nvconfig); 2499 cmn_err(CE_WARN, "pool '%s' could not be " 2500 "loaded as it was last accessed by " 2501 "another system (host: %s hostid: 0x%lx). " 2502 "See: http://illumos.org/msg/ZFS-8000-EY", 2503 spa_name(spa), hostname, 2504 (unsigned long)hostid); 2505 return (SET_ERROR(EBADF)); 2506 } 2507 } 2508 if (nvlist_lookup_nvlist(spa->spa_config, 2509 ZPOOL_REWIND_POLICY, &policy) == 0) 2510 VERIFY(nvlist_add_nvlist(nvconfig, 2511 ZPOOL_REWIND_POLICY, policy) == 0); 2512 2513 spa_config_set(spa, nvconfig); 2514 spa_unload(spa); 2515 spa_deactivate(spa); 2516 spa_activate(spa, orig_mode); 2517 2518 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2519 } 2520 2521 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2522 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2523 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2524 if (error != 0) 2525 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2526 2527 /* 2528 * Load the bit that tells us to use the new accounting function 2529 * (raid-z deflation). If we have an older pool, this will not 2530 * be present. 2531 */ 2532 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2533 if (error != 0 && error != ENOENT) 2534 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2535 2536 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2537 &spa->spa_creation_version); 2538 if (error != 0 && error != ENOENT) 2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2540 2541 /* 2542 * Load the persistent error log. If we have an older pool, this will 2543 * not be present. 2544 */ 2545 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2546 if (error != 0 && error != ENOENT) 2547 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2548 2549 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2550 &spa->spa_errlog_scrub); 2551 if (error != 0 && error != ENOENT) 2552 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2553 2554 /* 2555 * Load the history object. If we have an older pool, this 2556 * will not be present. 2557 */ 2558 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2559 if (error != 0 && error != ENOENT) 2560 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2561 2562 /* 2563 * If we're assembling the pool from the split-off vdevs of 2564 * an existing pool, we don't want to attach the spares & cache 2565 * devices. 2566 */ 2567 2568 /* 2569 * Load any hot spares for this pool. 2570 */ 2571 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2572 if (error != 0 && error != ENOENT) 2573 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2574 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2575 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2576 if (load_nvlist(spa, spa->spa_spares.sav_object, 2577 &spa->spa_spares.sav_config) != 0) 2578 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2579 2580 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2581 spa_load_spares(spa); 2582 spa_config_exit(spa, SCL_ALL, FTAG); 2583 } else if (error == 0) { 2584 spa->spa_spares.sav_sync = B_TRUE; 2585 } 2586 2587 /* 2588 * Load any level 2 ARC devices for this pool. 2589 */ 2590 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2591 &spa->spa_l2cache.sav_object); 2592 if (error != 0 && error != ENOENT) 2593 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2594 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2595 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2596 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2597 &spa->spa_l2cache.sav_config) != 0) 2598 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2599 2600 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2601 spa_load_l2cache(spa); 2602 spa_config_exit(spa, SCL_ALL, FTAG); 2603 } else if (error == 0) { 2604 spa->spa_l2cache.sav_sync = B_TRUE; 2605 } 2606 2607 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2608 2609 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2610 if (error && error != ENOENT) 2611 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2612 2613 if (error == 0) { 2614 uint64_t autoreplace; 2615 2616 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2617 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2618 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2619 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2620 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2621 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2622 &spa->spa_dedup_ditto); 2623 2624 spa->spa_autoreplace = (autoreplace != 0); 2625 } 2626 2627 /* 2628 * If the 'autoreplace' property is set, then post a resource notifying 2629 * the ZFS DE that it should not issue any faults for unopenable 2630 * devices. We also iterate over the vdevs, and post a sysevent for any 2631 * unopenable vdevs so that the normal autoreplace handler can take 2632 * over. 2633 */ 2634 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2635 spa_check_removed(spa->spa_root_vdev); 2636 /* 2637 * For the import case, this is done in spa_import(), because 2638 * at this point we're using the spare definitions from 2639 * the MOS config, not necessarily from the userland config. 2640 */ 2641 if (state != SPA_LOAD_IMPORT) { 2642 spa_aux_check_removed(&spa->spa_spares); 2643 spa_aux_check_removed(&spa->spa_l2cache); 2644 } 2645 } 2646 2647 /* 2648 * Load the vdev state for all toplevel vdevs. 2649 */ 2650 vdev_load(rvd); 2651 2652 /* 2653 * Propagate the leaf DTLs we just loaded all the way up the tree. 2654 */ 2655 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2656 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2657 spa_config_exit(spa, SCL_ALL, FTAG); 2658 2659 /* 2660 * Load the DDTs (dedup tables). 2661 */ 2662 error = ddt_load(spa); 2663 if (error != 0) 2664 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2665 2666 spa_update_dspace(spa); 2667 2668 /* 2669 * Validate the config, using the MOS config to fill in any 2670 * information which might be missing. If we fail to validate 2671 * the config then declare the pool unfit for use. If we're 2672 * assembling a pool from a split, the log is not transferred 2673 * over. 2674 */ 2675 if (type != SPA_IMPORT_ASSEMBLE) { 2676 nvlist_t *nvconfig; 2677 2678 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2679 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2680 2681 if (!spa_config_valid(spa, nvconfig)) { 2682 nvlist_free(nvconfig); 2683 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2684 ENXIO)); 2685 } 2686 nvlist_free(nvconfig); 2687 2688 /* 2689 * Now that we've validated the config, check the state of the 2690 * root vdev. If it can't be opened, it indicates one or 2691 * more toplevel vdevs are faulted. 2692 */ 2693 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2694 return (SET_ERROR(ENXIO)); 2695 2696 if (spa_writeable(spa) && spa_check_logs(spa)) { 2697 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2698 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2699 } 2700 } 2701 2702 if (missing_feat_write) { 2703 ASSERT(state == SPA_LOAD_TRYIMPORT); 2704 2705 /* 2706 * At this point, we know that we can open the pool in 2707 * read-only mode but not read-write mode. We now have enough 2708 * information and can return to userland. 2709 */ 2710 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2711 } 2712 2713 /* 2714 * We've successfully opened the pool, verify that we're ready 2715 * to start pushing transactions. 2716 */ 2717 if (state != SPA_LOAD_TRYIMPORT) { 2718 if (error = spa_load_verify(spa)) 2719 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2720 error)); 2721 } 2722 2723 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2724 spa->spa_load_max_txg == UINT64_MAX)) { 2725 dmu_tx_t *tx; 2726 int need_update = B_FALSE; 2727 dsl_pool_t *dp = spa_get_dsl(spa); 2728 2729 ASSERT(state != SPA_LOAD_TRYIMPORT); 2730 2731 /* 2732 * Claim log blocks that haven't been committed yet. 2733 * This must all happen in a single txg. 2734 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2735 * invoked from zil_claim_log_block()'s i/o done callback. 2736 * Price of rollback is that we abandon the log. 2737 */ 2738 spa->spa_claiming = B_TRUE; 2739 2740 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2741 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2742 zil_claim, tx, DS_FIND_CHILDREN); 2743 dmu_tx_commit(tx); 2744 2745 spa->spa_claiming = B_FALSE; 2746 2747 spa_set_log_state(spa, SPA_LOG_GOOD); 2748 spa->spa_sync_on = B_TRUE; 2749 txg_sync_start(spa->spa_dsl_pool); 2750 2751 /* 2752 * Wait for all claims to sync. We sync up to the highest 2753 * claimed log block birth time so that claimed log blocks 2754 * don't appear to be from the future. spa_claim_max_txg 2755 * will have been set for us by either zil_check_log_chain() 2756 * (invoked from spa_check_logs()) or zil_claim() above. 2757 */ 2758 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2759 2760 /* 2761 * If the config cache is stale, or we have uninitialized 2762 * metaslabs (see spa_vdev_add()), then update the config. 2763 * 2764 * If this is a verbatim import, trust the current 2765 * in-core spa_config and update the disk labels. 2766 */ 2767 if (config_cache_txg != spa->spa_config_txg || 2768 state == SPA_LOAD_IMPORT || 2769 state == SPA_LOAD_RECOVER || 2770 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2771 need_update = B_TRUE; 2772 2773 for (int c = 0; c < rvd->vdev_children; c++) 2774 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2775 need_update = B_TRUE; 2776 2777 /* 2778 * Update the config cache asychronously in case we're the 2779 * root pool, in which case the config cache isn't writable yet. 2780 */ 2781 if (need_update) 2782 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2783 2784 /* 2785 * Check all DTLs to see if anything needs resilvering. 2786 */ 2787 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2788 vdev_resilver_needed(rvd, NULL, NULL)) 2789 spa_async_request(spa, SPA_ASYNC_RESILVER); 2790 2791 /* 2792 * Log the fact that we booted up (so that we can detect if 2793 * we rebooted in the middle of an operation). 2794 */ 2795 spa_history_log_version(spa, "open"); 2796 2797 /* 2798 * Delete any inconsistent datasets. 2799 */ 2800 (void) dmu_objset_find(spa_name(spa), 2801 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2802 2803 /* 2804 * Clean up any stale temporary dataset userrefs. 2805 */ 2806 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2807 } 2808 2809 spa_async_request(spa, SPA_ASYNC_L2CACHE_REBUILD); 2810 2811 return (0); 2812 } 2813 2814 static int 2815 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2816 { 2817 int mode = spa->spa_mode; 2818 2819 spa_unload(spa); 2820 spa_deactivate(spa); 2821 2822 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2823 2824 spa_activate(spa, mode); 2825 spa_async_suspend(spa); 2826 2827 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2828 } 2829 2830 /* 2831 * If spa_load() fails this function will try loading prior txg's. If 2832 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2833 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2834 * function will not rewind the pool and will return the same error as 2835 * spa_load(). 2836 */ 2837 static int 2838 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2839 uint64_t max_request, int rewind_flags) 2840 { 2841 nvlist_t *loadinfo = NULL; 2842 nvlist_t *config = NULL; 2843 int load_error, rewind_error; 2844 uint64_t safe_rewind_txg; 2845 uint64_t min_txg; 2846 2847 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2848 spa->spa_load_max_txg = spa->spa_load_txg; 2849 spa_set_log_state(spa, SPA_LOG_CLEAR); 2850 } else { 2851 spa->spa_load_max_txg = max_request; 2852 if (max_request != UINT64_MAX) 2853 spa->spa_extreme_rewind = B_TRUE; 2854 } 2855 2856 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2857 mosconfig); 2858 if (load_error == 0) 2859 return (0); 2860 2861 if (spa->spa_root_vdev != NULL) 2862 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2863 2864 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2865 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2866 2867 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2868 nvlist_free(config); 2869 return (load_error); 2870 } 2871 2872 if (state == SPA_LOAD_RECOVER) { 2873 /* Price of rolling back is discarding txgs, including log */ 2874 spa_set_log_state(spa, SPA_LOG_CLEAR); 2875 } else { 2876 /* 2877 * If we aren't rolling back save the load info from our first 2878 * import attempt so that we can restore it after attempting 2879 * to rewind. 2880 */ 2881 loadinfo = spa->spa_load_info; 2882 spa->spa_load_info = fnvlist_alloc(); 2883 } 2884 2885 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2886 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2887 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2888 TXG_INITIAL : safe_rewind_txg; 2889 2890 /* 2891 * Continue as long as we're finding errors, we're still within 2892 * the acceptable rewind range, and we're still finding uberblocks 2893 */ 2894 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2895 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2896 if (spa->spa_load_max_txg < safe_rewind_txg) 2897 spa->spa_extreme_rewind = B_TRUE; 2898 rewind_error = spa_load_retry(spa, state, mosconfig); 2899 } 2900 2901 spa->spa_extreme_rewind = B_FALSE; 2902 spa->spa_load_max_txg = UINT64_MAX; 2903 2904 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2905 spa_config_set(spa, config); 2906 2907 if (state == SPA_LOAD_RECOVER) { 2908 ASSERT3P(loadinfo, ==, NULL); 2909 return (rewind_error); 2910 } else { 2911 /* Store the rewind info as part of the initial load info */ 2912 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2913 spa->spa_load_info); 2914 2915 /* Restore the initial load info */ 2916 fnvlist_free(spa->spa_load_info); 2917 spa->spa_load_info = loadinfo; 2918 2919 return (load_error); 2920 } 2921 } 2922 2923 /* 2924 * Pool Open/Import 2925 * 2926 * The import case is identical to an open except that the configuration is sent 2927 * down from userland, instead of grabbed from the configuration cache. For the 2928 * case of an open, the pool configuration will exist in the 2929 * POOL_STATE_UNINITIALIZED state. 2930 * 2931 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2932 * the same time open the pool, without having to keep around the spa_t in some 2933 * ambiguous state. 2934 */ 2935 static int 2936 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2937 nvlist_t **config) 2938 { 2939 spa_t *spa; 2940 spa_load_state_t state = SPA_LOAD_OPEN; 2941 int error; 2942 int locked = B_FALSE; 2943 2944 *spapp = NULL; 2945 2946 /* 2947 * As disgusting as this is, we need to support recursive calls to this 2948 * function because dsl_dir_open() is called during spa_load(), and ends 2949 * up calling spa_open() again. The real fix is to figure out how to 2950 * avoid dsl_dir_open() calling this in the first place. 2951 */ 2952 if (mutex_owner(&spa_namespace_lock) != curthread) { 2953 mutex_enter(&spa_namespace_lock); 2954 locked = B_TRUE; 2955 } 2956 2957 if ((spa = spa_lookup(pool)) == NULL) { 2958 if (locked) 2959 mutex_exit(&spa_namespace_lock); 2960 return (SET_ERROR(ENOENT)); 2961 } 2962 2963 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2964 zpool_rewind_policy_t policy; 2965 2966 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2967 &policy); 2968 if (policy.zrp_request & ZPOOL_DO_REWIND) 2969 state = SPA_LOAD_RECOVER; 2970 2971 spa_activate(spa, spa_mode_global); 2972 2973 if (state != SPA_LOAD_RECOVER) 2974 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2975 2976 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2977 policy.zrp_request); 2978 2979 if (error == EBADF) { 2980 /* 2981 * If vdev_validate() returns failure (indicated by 2982 * EBADF), it indicates that one of the vdevs indicates 2983 * that the pool has been exported or destroyed. If 2984 * this is the case, the config cache is out of sync and 2985 * we should remove the pool from the namespace. 2986 */ 2987 spa_unload(spa); 2988 spa_deactivate(spa); 2989 spa_config_sync(spa, B_TRUE, B_TRUE); 2990 spa_remove(spa); 2991 if (locked) 2992 mutex_exit(&spa_namespace_lock); 2993 return (SET_ERROR(ENOENT)); 2994 } 2995 2996 if (error) { 2997 /* 2998 * We can't open the pool, but we still have useful 2999 * information: the state of each vdev after the 3000 * attempted vdev_open(). Return this to the user. 3001 */ 3002 if (config != NULL && spa->spa_config) { 3003 VERIFY(nvlist_dup(spa->spa_config, config, 3004 KM_SLEEP) == 0); 3005 VERIFY(nvlist_add_nvlist(*config, 3006 ZPOOL_CONFIG_LOAD_INFO, 3007 spa->spa_load_info) == 0); 3008 } 3009 spa_unload(spa); 3010 spa_deactivate(spa); 3011 spa->spa_last_open_failed = error; 3012 if (locked) 3013 mutex_exit(&spa_namespace_lock); 3014 *spapp = NULL; 3015 return (error); 3016 } 3017 } 3018 3019 spa_open_ref(spa, tag); 3020 3021 if (config != NULL) 3022 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3023 3024 /* 3025 * If we've recovered the pool, pass back any information we 3026 * gathered while doing the load. 3027 */ 3028 if (state == SPA_LOAD_RECOVER) { 3029 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3030 spa->spa_load_info) == 0); 3031 } 3032 3033 if (locked) { 3034 spa->spa_last_open_failed = 0; 3035 spa->spa_last_ubsync_txg = 0; 3036 spa->spa_load_txg = 0; 3037 mutex_exit(&spa_namespace_lock); 3038 } 3039 3040 *spapp = spa; 3041 3042 return (0); 3043 } 3044 3045 int 3046 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3047 nvlist_t **config) 3048 { 3049 return (spa_open_common(name, spapp, tag, policy, config)); 3050 } 3051 3052 int 3053 spa_open(const char *name, spa_t **spapp, void *tag) 3054 { 3055 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3056 } 3057 3058 /* 3059 * Lookup the given spa_t, incrementing the inject count in the process, 3060 * preventing it from being exported or destroyed. 3061 */ 3062 spa_t * 3063 spa_inject_addref(char *name) 3064 { 3065 spa_t *spa; 3066 3067 mutex_enter(&spa_namespace_lock); 3068 if ((spa = spa_lookup(name)) == NULL) { 3069 mutex_exit(&spa_namespace_lock); 3070 return (NULL); 3071 } 3072 spa->spa_inject_ref++; 3073 mutex_exit(&spa_namespace_lock); 3074 3075 return (spa); 3076 } 3077 3078 void 3079 spa_inject_delref(spa_t *spa) 3080 { 3081 mutex_enter(&spa_namespace_lock); 3082 spa->spa_inject_ref--; 3083 mutex_exit(&spa_namespace_lock); 3084 } 3085 3086 /* 3087 * Add spares device information to the nvlist. 3088 */ 3089 static void 3090 spa_add_spares(spa_t *spa, nvlist_t *config) 3091 { 3092 nvlist_t **spares; 3093 uint_t i, nspares; 3094 nvlist_t *nvroot; 3095 uint64_t guid; 3096 vdev_stat_t *vs; 3097 uint_t vsc; 3098 uint64_t pool; 3099 3100 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3101 3102 if (spa->spa_spares.sav_count == 0) 3103 return; 3104 3105 VERIFY(nvlist_lookup_nvlist(config, 3106 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3107 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3108 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3109 if (nspares != 0) { 3110 VERIFY(nvlist_add_nvlist_array(nvroot, 3111 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3112 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3113 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3114 3115 /* 3116 * Go through and find any spares which have since been 3117 * repurposed as an active spare. If this is the case, update 3118 * their status appropriately. 3119 */ 3120 for (i = 0; i < nspares; i++) { 3121 VERIFY(nvlist_lookup_uint64(spares[i], 3122 ZPOOL_CONFIG_GUID, &guid) == 0); 3123 if (spa_spare_exists(guid, &pool, NULL) && 3124 pool != 0ULL) { 3125 VERIFY(nvlist_lookup_uint64_array( 3126 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3127 (uint64_t **)&vs, &vsc) == 0); 3128 vs->vs_state = VDEV_STATE_CANT_OPEN; 3129 vs->vs_aux = VDEV_AUX_SPARED; 3130 } 3131 } 3132 } 3133 } 3134 3135 /* 3136 * Add l2cache device information to the nvlist, including vdev stats. 3137 */ 3138 static void 3139 spa_add_l2cache(spa_t *spa, nvlist_t *config) 3140 { 3141 nvlist_t **l2cache; 3142 uint_t i, j, nl2cache; 3143 nvlist_t *nvroot; 3144 uint64_t guid; 3145 vdev_t *vd; 3146 vdev_stat_t *vs; 3147 uint_t vsc; 3148 3149 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3150 3151 if (spa->spa_l2cache.sav_count == 0) 3152 return; 3153 3154 VERIFY(nvlist_lookup_nvlist(config, 3155 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3156 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3157 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3158 if (nl2cache != 0) { 3159 VERIFY(nvlist_add_nvlist_array(nvroot, 3160 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3161 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3162 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3163 3164 /* 3165 * Update level 2 cache device stats. 3166 */ 3167 3168 for (i = 0; i < nl2cache; i++) { 3169 VERIFY(nvlist_lookup_uint64(l2cache[i], 3170 ZPOOL_CONFIG_GUID, &guid) == 0); 3171 3172 vd = NULL; 3173 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3174 if (guid == 3175 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3176 vd = spa->spa_l2cache.sav_vdevs[j]; 3177 break; 3178 } 3179 } 3180 ASSERT(vd != NULL); 3181 3182 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3183 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3184 == 0); 3185 vdev_get_stats(vd, vs); 3186 } 3187 } 3188 } 3189 3190 static void 3191 spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3192 { 3193 nvlist_t *features; 3194 zap_cursor_t zc; 3195 zap_attribute_t za; 3196 3197 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3198 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3199 3200 if (spa->spa_feat_for_read_obj != 0) { 3201 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3202 spa->spa_feat_for_read_obj); 3203 zap_cursor_retrieve(&zc, &za) == 0; 3204 zap_cursor_advance(&zc)) { 3205 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3206 za.za_num_integers == 1); 3207 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3208 za.za_first_integer)); 3209 } 3210 zap_cursor_fini(&zc); 3211 } 3212 3213 if (spa->spa_feat_for_write_obj != 0) { 3214 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3215 spa->spa_feat_for_write_obj); 3216 zap_cursor_retrieve(&zc, &za) == 0; 3217 zap_cursor_advance(&zc)) { 3218 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3219 za.za_num_integers == 1); 3220 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3221 za.za_first_integer)); 3222 } 3223 zap_cursor_fini(&zc); 3224 } 3225 3226 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3227 features) == 0); 3228 nvlist_free(features); 3229 } 3230 3231 int 3232 spa_get_stats(const char *name, nvlist_t **config, 3233 char *altroot, size_t buflen) 3234 { 3235 int error; 3236 spa_t *spa; 3237 3238 *config = NULL; 3239 error = spa_open_common(name, &spa, FTAG, NULL, config); 3240 3241 if (spa != NULL) { 3242 /* 3243 * This still leaves a window of inconsistency where the spares 3244 * or l2cache devices could change and the config would be 3245 * self-inconsistent. 3246 */ 3247 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3248 3249 if (*config != NULL) { 3250 uint64_t loadtimes[2]; 3251 3252 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3253 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3254 VERIFY(nvlist_add_uint64_array(*config, 3255 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3256 3257 VERIFY(nvlist_add_uint64(*config, 3258 ZPOOL_CONFIG_ERRCOUNT, 3259 spa_get_errlog_size(spa)) == 0); 3260 3261 if (spa_suspended(spa)) 3262 VERIFY(nvlist_add_uint64(*config, 3263 ZPOOL_CONFIG_SUSPENDED, 3264 spa->spa_failmode) == 0); 3265 3266 spa_add_spares(spa, *config); 3267 spa_add_l2cache(spa, *config); 3268 spa_add_feature_stats(spa, *config); 3269 } 3270 } 3271 3272 /* 3273 * We want to get the alternate root even for faulted pools, so we cheat 3274 * and call spa_lookup() directly. 3275 */ 3276 if (altroot) { 3277 if (spa == NULL) { 3278 mutex_enter(&spa_namespace_lock); 3279 spa = spa_lookup(name); 3280 if (spa) 3281 spa_altroot(spa, altroot, buflen); 3282 else 3283 altroot[0] = '\0'; 3284 spa = NULL; 3285 mutex_exit(&spa_namespace_lock); 3286 } else { 3287 spa_altroot(spa, altroot, buflen); 3288 } 3289 } 3290 3291 if (spa != NULL) { 3292 spa_config_exit(spa, SCL_CONFIG, FTAG); 3293 spa_close(spa, FTAG); 3294 } 3295 3296 return (error); 3297 } 3298 3299 /* 3300 * Validate that the auxiliary device array is well formed. We must have an 3301 * array of nvlists, each which describes a valid leaf vdev. If this is an 3302 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3303 * specified, as long as they are well-formed. 3304 */ 3305 static int 3306 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3307 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3308 vdev_labeltype_t label) 3309 { 3310 nvlist_t **dev; 3311 uint_t i, ndev; 3312 vdev_t *vd; 3313 int error; 3314 3315 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3316 3317 /* 3318 * It's acceptable to have no devs specified. 3319 */ 3320 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3321 return (0); 3322 3323 if (ndev == 0) 3324 return (SET_ERROR(EINVAL)); 3325 3326 /* 3327 * Make sure the pool is formatted with a version that supports this 3328 * device type. 3329 */ 3330 if (spa_version(spa) < version) 3331 return (SET_ERROR(ENOTSUP)); 3332 3333 /* 3334 * Set the pending device list so we correctly handle device in-use 3335 * checking. 3336 */ 3337 sav->sav_pending = dev; 3338 sav->sav_npending = ndev; 3339 3340 for (i = 0; i < ndev; i++) { 3341 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3342 mode)) != 0) 3343 goto out; 3344 3345 if (!vd->vdev_ops->vdev_op_leaf) { 3346 vdev_free(vd); 3347 error = SET_ERROR(EINVAL); 3348 goto out; 3349 } 3350 3351 /* 3352 * The L2ARC currently only supports disk devices in 3353 * kernel context. For user-level testing, we allow it. 3354 */ 3355 #ifdef _KERNEL 3356 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3357 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3358 error = SET_ERROR(ENOTBLK); 3359 vdev_free(vd); 3360 goto out; 3361 } 3362 #endif 3363 vd->vdev_top = vd; 3364 3365 if ((error = vdev_open(vd)) == 0 && 3366 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3367 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3368 vd->vdev_guid) == 0); 3369 } 3370 3371 vdev_free(vd); 3372 3373 if (error && 3374 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3375 goto out; 3376 else 3377 error = 0; 3378 } 3379 3380 out: 3381 sav->sav_pending = NULL; 3382 sav->sav_npending = 0; 3383 return (error); 3384 } 3385 3386 static int 3387 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3388 { 3389 int error; 3390 3391 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3392 3393 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3394 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3395 VDEV_LABEL_SPARE)) != 0) { 3396 return (error); 3397 } 3398 3399 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3400 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3401 VDEV_LABEL_L2CACHE)); 3402 } 3403 3404 static void 3405 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3406 const char *config) 3407 { 3408 int i; 3409 3410 if (sav->sav_config != NULL) { 3411 nvlist_t **olddevs; 3412 uint_t oldndevs; 3413 nvlist_t **newdevs; 3414 3415 /* 3416 * Generate new dev list by concatentating with the 3417 * current dev list. 3418 */ 3419 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3420 &olddevs, &oldndevs) == 0); 3421 3422 newdevs = kmem_alloc(sizeof (void *) * 3423 (ndevs + oldndevs), KM_SLEEP); 3424 for (i = 0; i < oldndevs; i++) 3425 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3426 KM_SLEEP) == 0); 3427 for (i = 0; i < ndevs; i++) 3428 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3429 KM_SLEEP) == 0); 3430 3431 VERIFY(nvlist_remove(sav->sav_config, config, 3432 DATA_TYPE_NVLIST_ARRAY) == 0); 3433 3434 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3435 config, newdevs, ndevs + oldndevs) == 0); 3436 for (i = 0; i < oldndevs + ndevs; i++) 3437 nvlist_free(newdevs[i]); 3438 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3439 } else { 3440 /* 3441 * Generate a new dev list. 3442 */ 3443 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3444 KM_SLEEP) == 0); 3445 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3446 devs, ndevs) == 0); 3447 } 3448 } 3449 3450 /* 3451 * Stop and drop level 2 ARC devices 3452 */ 3453 void 3454 spa_l2cache_drop(spa_t *spa) 3455 { 3456 vdev_t *vd; 3457 int i; 3458 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3459 3460 for (i = 0; i < sav->sav_count; i++) { 3461 uint64_t pool; 3462 3463 vd = sav->sav_vdevs[i]; 3464 ASSERT(vd != NULL); 3465 3466 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3467 pool != 0ULL && l2arc_vdev_present(vd)) 3468 l2arc_remove_vdev(vd); 3469 } 3470 } 3471 3472 /* 3473 * Pool Creation 3474 */ 3475 int 3476 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3477 nvlist_t *zplprops) 3478 { 3479 spa_t *spa; 3480 char *altroot = NULL; 3481 vdev_t *rvd; 3482 dsl_pool_t *dp; 3483 dmu_tx_t *tx; 3484 int error = 0; 3485 uint64_t txg = TXG_INITIAL; 3486 nvlist_t **spares, **l2cache; 3487 uint_t nspares, nl2cache; 3488 uint64_t version, obj; 3489 boolean_t has_features; 3490 3491 /* 3492 * If this pool already exists, return failure. 3493 */ 3494 mutex_enter(&spa_namespace_lock); 3495 if (spa_lookup(pool) != NULL) { 3496 mutex_exit(&spa_namespace_lock); 3497 return (SET_ERROR(EEXIST)); 3498 } 3499 3500 /* 3501 * Allocate a new spa_t structure. 3502 */ 3503 (void) nvlist_lookup_string(props, 3504 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3505 spa = spa_add(pool, NULL, altroot); 3506 spa_activate(spa, spa_mode_global); 3507 3508 if (props && (error = spa_prop_validate(spa, props))) { 3509 spa_deactivate(spa); 3510 spa_remove(spa); 3511 mutex_exit(&spa_namespace_lock); 3512 return (error); 3513 } 3514 3515 has_features = B_FALSE; 3516 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3517 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3518 if (zpool_prop_feature(nvpair_name(elem))) 3519 has_features = B_TRUE; 3520 } 3521 3522 if (has_features || nvlist_lookup_uint64(props, 3523 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3524 version = SPA_VERSION; 3525 } 3526 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3527 3528 spa->spa_first_txg = txg; 3529 spa->spa_uberblock.ub_txg = txg - 1; 3530 spa->spa_uberblock.ub_version = version; 3531 spa->spa_ubsync = spa->spa_uberblock; 3532 3533 /* 3534 * Create "The Godfather" zio to hold all async IOs 3535 */ 3536 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3537 KM_SLEEP); 3538 for (int i = 0; i < max_ncpus; i++) { 3539 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3540 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3541 ZIO_FLAG_GODFATHER); 3542 } 3543 3544 /* 3545 * Create the root vdev. 3546 */ 3547 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3548 3549 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3550 3551 ASSERT(error != 0 || rvd != NULL); 3552 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3553 3554 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3555 error = SET_ERROR(EINVAL); 3556 3557 if (error == 0 && 3558 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3559 (error = spa_validate_aux(spa, nvroot, txg, 3560 VDEV_ALLOC_ADD)) == 0) { 3561 for (int c = 0; c < rvd->vdev_children; c++) { 3562 vdev_metaslab_set_size(rvd->vdev_child[c]); 3563 vdev_expand(rvd->vdev_child[c], txg); 3564 } 3565 } 3566 3567 spa_config_exit(spa, SCL_ALL, FTAG); 3568 3569 if (error != 0) { 3570 spa_unload(spa); 3571 spa_deactivate(spa); 3572 spa_remove(spa); 3573 mutex_exit(&spa_namespace_lock); 3574 return (error); 3575 } 3576 3577 /* 3578 * Get the list of spares, if specified. 3579 */ 3580 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3581 &spares, &nspares) == 0) { 3582 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3583 KM_SLEEP) == 0); 3584 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3585 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3586 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3587 spa_load_spares(spa); 3588 spa_config_exit(spa, SCL_ALL, FTAG); 3589 spa->spa_spares.sav_sync = B_TRUE; 3590 } 3591 3592 /* 3593 * Get the list of level 2 cache devices, if specified. 3594 */ 3595 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3596 &l2cache, &nl2cache) == 0) { 3597 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3598 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3599 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3600 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3601 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3602 spa_load_l2cache(spa); 3603 spa_config_exit(spa, SCL_ALL, FTAG); 3604 spa->spa_l2cache.sav_sync = B_TRUE; 3605 } 3606 3607 spa->spa_is_initializing = B_TRUE; 3608 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3609 spa->spa_meta_objset = dp->dp_meta_objset; 3610 spa->spa_is_initializing = B_FALSE; 3611 3612 /* 3613 * Create DDTs (dedup tables). 3614 */ 3615 ddt_create(spa); 3616 3617 spa_update_dspace(spa); 3618 3619 tx = dmu_tx_create_assigned(dp, txg); 3620 3621 /* 3622 * Create the pool config object. 3623 */ 3624 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3625 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3626 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3627 3628 if (zap_add(spa->spa_meta_objset, 3629 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3630 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3631 cmn_err(CE_PANIC, "failed to add pool config"); 3632 } 3633 3634 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3635 spa_feature_create_zap_objects(spa, tx); 3636 3637 if (zap_add(spa->spa_meta_objset, 3638 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3639 sizeof (uint64_t), 1, &version, tx) != 0) { 3640 cmn_err(CE_PANIC, "failed to add pool version"); 3641 } 3642 3643 /* Newly created pools with the right version are always deflated. */ 3644 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3645 spa->spa_deflate = TRUE; 3646 if (zap_add(spa->spa_meta_objset, 3647 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3648 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3649 cmn_err(CE_PANIC, "failed to add deflate"); 3650 } 3651 } 3652 3653 /* 3654 * Create the deferred-free bpobj. Turn off compression 3655 * because sync-to-convergence takes longer if the blocksize 3656 * keeps changing. 3657 */ 3658 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3659 dmu_object_set_compress(spa->spa_meta_objset, obj, 3660 ZIO_COMPRESS_OFF, tx); 3661 if (zap_add(spa->spa_meta_objset, 3662 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3663 sizeof (uint64_t), 1, &obj, tx) != 0) { 3664 cmn_err(CE_PANIC, "failed to add bpobj"); 3665 } 3666 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3667 spa->spa_meta_objset, obj)); 3668 3669 /* 3670 * Create the pool's history object. 3671 */ 3672 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3673 spa_history_create_obj(spa, tx); 3674 3675 /* 3676 * Set pool properties. 3677 */ 3678 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3679 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3680 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3681 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3682 3683 if (props != NULL) { 3684 spa_configfile_set(spa, props, B_FALSE); 3685 spa_sync_props(props, tx); 3686 } 3687 3688 dmu_tx_commit(tx); 3689 3690 spa->spa_sync_on = B_TRUE; 3691 txg_sync_start(spa->spa_dsl_pool); 3692 3693 /* 3694 * We explicitly wait for the first transaction to complete so that our 3695 * bean counters are appropriately updated. 3696 */ 3697 txg_wait_synced(spa->spa_dsl_pool, txg); 3698 3699 spa_config_sync(spa, B_FALSE, B_TRUE); 3700 3701 spa_history_log_version(spa, "create"); 3702 3703 /* 3704 * Don't count references from objsets that are already closed 3705 * and are making their way through the eviction process. 3706 */ 3707 spa_evicting_os_wait(spa); 3708 spa->spa_minref = refcount_count(&spa->spa_refcount); 3709 3710 mutex_exit(&spa_namespace_lock); 3711 3712 return (0); 3713 } 3714 3715 #ifdef _KERNEL 3716 /* 3717 * Get the root pool information from the root disk, then import the root pool 3718 * during the system boot up time. 3719 */ 3720 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3721 3722 static nvlist_t * 3723 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3724 { 3725 nvlist_t *config; 3726 nvlist_t *nvtop, *nvroot; 3727 uint64_t pgid; 3728 3729 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3730 return (NULL); 3731 3732 /* 3733 * Add this top-level vdev to the child array. 3734 */ 3735 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3736 &nvtop) == 0); 3737 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3738 &pgid) == 0); 3739 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3740 3741 /* 3742 * Put this pool's top-level vdevs into a root vdev. 3743 */ 3744 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3745 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3746 VDEV_TYPE_ROOT) == 0); 3747 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3748 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3749 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3750 &nvtop, 1) == 0); 3751 3752 /* 3753 * Replace the existing vdev_tree with the new root vdev in 3754 * this pool's configuration (remove the old, add the new). 3755 */ 3756 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3757 nvlist_free(nvroot); 3758 return (config); 3759 } 3760 3761 /* 3762 * Walk the vdev tree and see if we can find a device with "better" 3763 * configuration. A configuration is "better" if the label on that 3764 * device has a more recent txg. 3765 */ 3766 static void 3767 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3768 { 3769 for (int c = 0; c < vd->vdev_children; c++) 3770 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3771 3772 if (vd->vdev_ops->vdev_op_leaf) { 3773 nvlist_t *label; 3774 uint64_t label_txg; 3775 3776 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3777 &label) != 0) 3778 return; 3779 3780 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3781 &label_txg) == 0); 3782 3783 /* 3784 * Do we have a better boot device? 3785 */ 3786 if (label_txg > *txg) { 3787 *txg = label_txg; 3788 *avd = vd; 3789 } 3790 nvlist_free(label); 3791 } 3792 } 3793 3794 /* 3795 * Import a root pool. 3796 * 3797 * For x86. devpath_list will consist of devid and/or physpath name of 3798 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3799 * The GRUB "findroot" command will return the vdev we should boot. 3800 * 3801 * For Sparc, devpath_list consists the physpath name of the booting device 3802 * no matter the rootpool is a single device pool or a mirrored pool. 3803 * e.g. 3804 * "/pci@1f,0/ide@d/disk@0,0:a" 3805 */ 3806 int 3807 spa_import_rootpool(char *devpath, char *devid) 3808 { 3809 spa_t *spa; 3810 vdev_t *rvd, *bvd, *avd = NULL; 3811 nvlist_t *config, *nvtop; 3812 uint64_t guid, txg; 3813 char *pname; 3814 int error; 3815 3816 /* 3817 * Read the label from the boot device and generate a configuration. 3818 */ 3819 config = spa_generate_rootconf(devpath, devid, &guid); 3820 #if defined(_OBP) && defined(_KERNEL) 3821 if (config == NULL) { 3822 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3823 /* iscsi boot */ 3824 get_iscsi_bootpath_phy(devpath); 3825 config = spa_generate_rootconf(devpath, devid, &guid); 3826 } 3827 } 3828 #endif 3829 if (config == NULL) { 3830 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3831 devpath); 3832 return (SET_ERROR(EIO)); 3833 } 3834 3835 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3836 &pname) == 0); 3837 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3838 3839 mutex_enter(&spa_namespace_lock); 3840 if ((spa = spa_lookup(pname)) != NULL) { 3841 /* 3842 * Remove the existing root pool from the namespace so that we 3843 * can replace it with the correct config we just read in. 3844 */ 3845 spa_remove(spa); 3846 } 3847 3848 spa = spa_add(pname, config, NULL); 3849 spa->spa_is_root = B_TRUE; 3850 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3851 3852 /* 3853 * Build up a vdev tree based on the boot device's label config. 3854 */ 3855 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3856 &nvtop) == 0); 3857 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3858 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3859 VDEV_ALLOC_ROOTPOOL); 3860 spa_config_exit(spa, SCL_ALL, FTAG); 3861 if (error) { 3862 mutex_exit(&spa_namespace_lock); 3863 nvlist_free(config); 3864 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3865 pname); 3866 return (error); 3867 } 3868 3869 /* 3870 * Get the boot vdev. 3871 */ 3872 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3873 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3874 (u_longlong_t)guid); 3875 error = SET_ERROR(ENOENT); 3876 goto out; 3877 } 3878 3879 /* 3880 * Determine if there is a better boot device. 3881 */ 3882 avd = bvd; 3883 spa_alt_rootvdev(rvd, &avd, &txg); 3884 if (avd != bvd) { 3885 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3886 "try booting from '%s'", avd->vdev_path); 3887 error = SET_ERROR(EINVAL); 3888 goto out; 3889 } 3890 3891 /* 3892 * If the boot device is part of a spare vdev then ensure that 3893 * we're booting off the active spare. 3894 */ 3895 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3896 !bvd->vdev_isspare) { 3897 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3898 "try booting from '%s'", 3899 bvd->vdev_parent-> 3900 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3901 error = SET_ERROR(EINVAL); 3902 goto out; 3903 } 3904 3905 error = 0; 3906 out: 3907 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3908 vdev_free(rvd); 3909 spa_config_exit(spa, SCL_ALL, FTAG); 3910 mutex_exit(&spa_namespace_lock); 3911 3912 nvlist_free(config); 3913 return (error); 3914 } 3915 3916 #endif 3917 3918 /* 3919 * Import a non-root pool into the system. 3920 */ 3921 int 3922 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3923 { 3924 spa_t *spa; 3925 char *altroot = NULL; 3926 spa_load_state_t state = SPA_LOAD_IMPORT; 3927 zpool_rewind_policy_t policy; 3928 uint64_t mode = spa_mode_global; 3929 uint64_t readonly = B_FALSE; 3930 int error; 3931 nvlist_t *nvroot; 3932 nvlist_t **spares, **l2cache; 3933 uint_t nspares, nl2cache; 3934 3935 /* 3936 * If a pool with this name exists, return failure. 3937 */ 3938 mutex_enter(&spa_namespace_lock); 3939 if (spa_lookup(pool) != NULL) { 3940 mutex_exit(&spa_namespace_lock); 3941 return (SET_ERROR(EEXIST)); 3942 } 3943 3944 /* 3945 * Create and initialize the spa structure. 3946 */ 3947 (void) nvlist_lookup_string(props, 3948 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3949 (void) nvlist_lookup_uint64(props, 3950 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 3951 if (readonly) 3952 mode = FREAD; 3953 spa = spa_add(pool, config, altroot); 3954 spa->spa_import_flags = flags; 3955 3956 /* 3957 * Verbatim import - Take a pool and insert it into the namespace 3958 * as if it had been loaded at boot. 3959 */ 3960 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 3961 if (props != NULL) 3962 spa_configfile_set(spa, props, B_FALSE); 3963 3964 spa_config_sync(spa, B_FALSE, B_TRUE); 3965 3966 mutex_exit(&spa_namespace_lock); 3967 return (0); 3968 } 3969 3970 spa_activate(spa, mode); 3971 3972 /* 3973 * Don't start async tasks until we know everything is healthy. 3974 */ 3975 spa_async_suspend(spa); 3976 3977 zpool_get_rewind_policy(config, &policy); 3978 if (policy.zrp_request & ZPOOL_DO_REWIND) 3979 state = SPA_LOAD_RECOVER; 3980 3981 /* 3982 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 3983 * because the user-supplied config is actually the one to trust when 3984 * doing an import. 3985 */ 3986 if (state != SPA_LOAD_RECOVER) 3987 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3988 3989 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 3990 policy.zrp_request); 3991 3992 /* 3993 * Propagate anything learned while loading the pool and pass it 3994 * back to caller (i.e. rewind info, missing devices, etc). 3995 */ 3996 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 3997 spa->spa_load_info) == 0); 3998 3999 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4000 /* 4001 * Toss any existing sparelist, as it doesn't have any validity 4002 * anymore, and conflicts with spa_has_spare(). 4003 */ 4004 if (spa->spa_spares.sav_config) { 4005 nvlist_free(spa->spa_spares.sav_config); 4006 spa->spa_spares.sav_config = NULL; 4007 spa_load_spares(spa); 4008 } 4009 if (spa->spa_l2cache.sav_config) { 4010 nvlist_free(spa->spa_l2cache.sav_config); 4011 spa->spa_l2cache.sav_config = NULL; 4012 spa_load_l2cache(spa); 4013 } 4014 4015 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4016 &nvroot) == 0); 4017 if (error == 0) 4018 error = spa_validate_aux(spa, nvroot, -1ULL, 4019 VDEV_ALLOC_SPARE); 4020 if (error == 0) 4021 error = spa_validate_aux(spa, nvroot, -1ULL, 4022 VDEV_ALLOC_L2CACHE); 4023 spa_config_exit(spa, SCL_ALL, FTAG); 4024 4025 if (props != NULL) 4026 spa_configfile_set(spa, props, B_FALSE); 4027 4028 if (error != 0 || (props && spa_writeable(spa) && 4029 (error = spa_prop_set(spa, props)))) { 4030 spa_unload(spa); 4031 spa_deactivate(spa); 4032 spa_remove(spa); 4033 mutex_exit(&spa_namespace_lock); 4034 return (error); 4035 } 4036 4037 spa_async_resume(spa); 4038 4039 /* 4040 * Override any spares and level 2 cache devices as specified by 4041 * the user, as these may have correct device names/devids, etc. 4042 */ 4043 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4044 &spares, &nspares) == 0) { 4045 if (spa->spa_spares.sav_config) 4046 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4047 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4048 else 4049 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4050 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4051 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4052 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4053 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4054 spa_load_spares(spa); 4055 spa_config_exit(spa, SCL_ALL, FTAG); 4056 spa->spa_spares.sav_sync = B_TRUE; 4057 } 4058 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4059 &l2cache, &nl2cache) == 0) { 4060 if (spa->spa_l2cache.sav_config) 4061 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4062 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4063 else 4064 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4065 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4066 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4067 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4068 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4069 spa_load_l2cache(spa); 4070 spa_config_exit(spa, SCL_ALL, FTAG); 4071 spa->spa_l2cache.sav_sync = B_TRUE; 4072 } 4073 4074 /* 4075 * Check for any removed devices. 4076 */ 4077 if (spa->spa_autoreplace) { 4078 spa_aux_check_removed(&spa->spa_spares); 4079 spa_aux_check_removed(&spa->spa_l2cache); 4080 } 4081 4082 if (spa_writeable(spa)) { 4083 /* 4084 * Update the config cache to include the newly-imported pool. 4085 */ 4086 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4087 } 4088 4089 /* 4090 * It's possible that the pool was expanded while it was exported. 4091 * We kick off an async task to handle this for us. 4092 */ 4093 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4094 4095 mutex_exit(&spa_namespace_lock); 4096 spa_history_log_version(spa, "import"); 4097 4098 return (0); 4099 } 4100 4101 nvlist_t * 4102 spa_tryimport(nvlist_t *tryconfig) 4103 { 4104 nvlist_t *config = NULL; 4105 char *poolname; 4106 spa_t *spa; 4107 uint64_t state; 4108 int error; 4109 4110 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4111 return (NULL); 4112 4113 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4114 return (NULL); 4115 4116 /* 4117 * Create and initialize the spa structure. 4118 */ 4119 mutex_enter(&spa_namespace_lock); 4120 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4121 spa_activate(spa, FREAD); 4122 4123 /* 4124 * Pass off the heavy lifting to spa_load(). 4125 * Pass TRUE for mosconfig because the user-supplied config 4126 * is actually the one to trust when doing an import. 4127 */ 4128 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4129 4130 /* 4131 * If 'tryconfig' was at least parsable, return the current config. 4132 */ 4133 if (spa->spa_root_vdev != NULL) { 4134 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4135 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4136 poolname) == 0); 4137 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4138 state) == 0); 4139 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4140 spa->spa_uberblock.ub_timestamp) == 0); 4141 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4142 spa->spa_load_info) == 0); 4143 4144 /* 4145 * If the bootfs property exists on this pool then we 4146 * copy it out so that external consumers can tell which 4147 * pools are bootable. 4148 */ 4149 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4150 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4151 4152 /* 4153 * We have to play games with the name since the 4154 * pool was opened as TRYIMPORT_NAME. 4155 */ 4156 if (dsl_dsobj_to_dsname(spa_name(spa), 4157 spa->spa_bootfs, tmpname) == 0) { 4158 char *cp; 4159 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4160 4161 cp = strchr(tmpname, '/'); 4162 if (cp == NULL) { 4163 (void) strlcpy(dsname, tmpname, 4164 MAXPATHLEN); 4165 } else { 4166 (void) snprintf(dsname, MAXPATHLEN, 4167 "%s/%s", poolname, ++cp); 4168 } 4169 VERIFY(nvlist_add_string(config, 4170 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4171 kmem_free(dsname, MAXPATHLEN); 4172 } 4173 kmem_free(tmpname, MAXPATHLEN); 4174 } 4175 4176 /* 4177 * Add the list of hot spares and level 2 cache devices. 4178 */ 4179 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4180 spa_add_spares(spa, config); 4181 spa_add_l2cache(spa, config); 4182 spa_config_exit(spa, SCL_CONFIG, FTAG); 4183 } 4184 4185 spa_unload(spa); 4186 spa_deactivate(spa); 4187 spa_remove(spa); 4188 mutex_exit(&spa_namespace_lock); 4189 4190 return (config); 4191 } 4192 4193 /* 4194 * Pool export/destroy 4195 * 4196 * The act of destroying or exporting a pool is very simple. We make sure there 4197 * is no more pending I/O and any references to the pool are gone. Then, we 4198 * update the pool state and sync all the labels to disk, removing the 4199 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4200 * we don't sync the labels or remove the configuration cache. 4201 */ 4202 static int 4203 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4204 boolean_t force, boolean_t hardforce) 4205 { 4206 spa_t *spa; 4207 4208 if (oldconfig) 4209 *oldconfig = NULL; 4210 4211 if (!(spa_mode_global & FWRITE)) 4212 return (SET_ERROR(EROFS)); 4213 4214 mutex_enter(&spa_namespace_lock); 4215 if ((spa = spa_lookup(pool)) == NULL) { 4216 mutex_exit(&spa_namespace_lock); 4217 return (SET_ERROR(ENOENT)); 4218 } 4219 4220 /* 4221 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4222 * reacquire the namespace lock, and see if we can export. 4223 */ 4224 spa_open_ref(spa, FTAG); 4225 mutex_exit(&spa_namespace_lock); 4226 spa_async_suspend(spa); 4227 mutex_enter(&spa_namespace_lock); 4228 spa_close(spa, FTAG); 4229 4230 /* 4231 * The pool will be in core if it's openable, 4232 * in which case we can modify its state. 4233 */ 4234 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4235 /* 4236 * Objsets may be open only because they're dirty, so we 4237 * have to force it to sync before checking spa_refcnt. 4238 */ 4239 txg_wait_synced(spa->spa_dsl_pool, 0); 4240 spa_evicting_os_wait(spa); 4241 4242 /* 4243 * A pool cannot be exported or destroyed if there are active 4244 * references. If we are resetting a pool, allow references by 4245 * fault injection handlers. 4246 */ 4247 if (!spa_refcount_zero(spa) || 4248 (spa->spa_inject_ref != 0 && 4249 new_state != POOL_STATE_UNINITIALIZED)) { 4250 spa_async_resume(spa); 4251 mutex_exit(&spa_namespace_lock); 4252 return (SET_ERROR(EBUSY)); 4253 } 4254 4255 /* 4256 * A pool cannot be exported if it has an active shared spare. 4257 * This is to prevent other pools stealing the active spare 4258 * from an exported pool. At user's own will, such pool can 4259 * be forcedly exported. 4260 */ 4261 if (!force && new_state == POOL_STATE_EXPORTED && 4262 spa_has_active_shared_spare(spa)) { 4263 spa_async_resume(spa); 4264 mutex_exit(&spa_namespace_lock); 4265 return (SET_ERROR(EXDEV)); 4266 } 4267 4268 /* 4269 * We want this to be reflected on every label, 4270 * so mark them all dirty. spa_unload() will do the 4271 * final sync that pushes these changes out. 4272 */ 4273 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4274 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4275 spa->spa_state = new_state; 4276 spa->spa_final_txg = spa_last_synced_txg(spa) + 4277 TXG_DEFER_SIZE + 1; 4278 vdev_config_dirty(spa->spa_root_vdev); 4279 spa_config_exit(spa, SCL_ALL, FTAG); 4280 } 4281 } 4282 4283 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4284 4285 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4286 spa_unload(spa); 4287 spa_deactivate(spa); 4288 } 4289 4290 if (oldconfig && spa->spa_config) 4291 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4292 4293 if (new_state != POOL_STATE_UNINITIALIZED) { 4294 if (!hardforce) 4295 spa_config_sync(spa, B_TRUE, B_TRUE); 4296 spa_remove(spa); 4297 } 4298 mutex_exit(&spa_namespace_lock); 4299 4300 return (0); 4301 } 4302 4303 /* 4304 * Destroy a storage pool. 4305 */ 4306 int 4307 spa_destroy(char *pool) 4308 { 4309 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4310 B_FALSE, B_FALSE)); 4311 } 4312 4313 /* 4314 * Export a storage pool. 4315 */ 4316 int 4317 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4318 boolean_t hardforce) 4319 { 4320 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4321 force, hardforce)); 4322 } 4323 4324 /* 4325 * Similar to spa_export(), this unloads the spa_t without actually removing it 4326 * from the namespace in any way. 4327 */ 4328 int 4329 spa_reset(char *pool) 4330 { 4331 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4332 B_FALSE, B_FALSE)); 4333 } 4334 4335 /* 4336 * ========================================================================== 4337 * Device manipulation 4338 * ========================================================================== 4339 */ 4340 4341 /* 4342 * Add a device to a storage pool. 4343 */ 4344 int 4345 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4346 { 4347 uint64_t txg, id; 4348 int error; 4349 vdev_t *rvd = spa->spa_root_vdev; 4350 vdev_t *vd, *tvd; 4351 nvlist_t **spares, **l2cache; 4352 uint_t nspares, nl2cache; 4353 4354 ASSERT(spa_writeable(spa)); 4355 4356 txg = spa_vdev_enter(spa); 4357 4358 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4359 VDEV_ALLOC_ADD)) != 0) 4360 return (spa_vdev_exit(spa, NULL, txg, error)); 4361 4362 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4363 4364 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4365 &nspares) != 0) 4366 nspares = 0; 4367 4368 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4369 &nl2cache) != 0) 4370 nl2cache = 0; 4371 4372 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4373 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4374 4375 if (vd->vdev_children != 0 && 4376 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4377 return (spa_vdev_exit(spa, vd, txg, error)); 4378 4379 /* 4380 * We must validate the spares and l2cache devices after checking the 4381 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4382 */ 4383 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4384 return (spa_vdev_exit(spa, vd, txg, error)); 4385 4386 /* 4387 * Transfer each new top-level vdev from vd to rvd. 4388 */ 4389 for (int c = 0; c < vd->vdev_children; c++) { 4390 4391 /* 4392 * Set the vdev id to the first hole, if one exists. 4393 */ 4394 for (id = 0; id < rvd->vdev_children; id++) { 4395 if (rvd->vdev_child[id]->vdev_ishole) { 4396 vdev_free(rvd->vdev_child[id]); 4397 break; 4398 } 4399 } 4400 tvd = vd->vdev_child[c]; 4401 vdev_remove_child(vd, tvd); 4402 tvd->vdev_id = id; 4403 vdev_add_child(rvd, tvd); 4404 vdev_config_dirty(tvd); 4405 } 4406 4407 if (nspares != 0) { 4408 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4409 ZPOOL_CONFIG_SPARES); 4410 spa_load_spares(spa); 4411 spa->spa_spares.sav_sync = B_TRUE; 4412 } 4413 4414 if (nl2cache != 0) { 4415 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4416 ZPOOL_CONFIG_L2CACHE); 4417 spa_load_l2cache(spa); 4418 spa->spa_l2cache.sav_sync = B_TRUE; 4419 } 4420 4421 /* 4422 * We have to be careful when adding new vdevs to an existing pool. 4423 * If other threads start allocating from these vdevs before we 4424 * sync the config cache, and we lose power, then upon reboot we may 4425 * fail to open the pool because there are DVAs that the config cache 4426 * can't translate. Therefore, we first add the vdevs without 4427 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4428 * and then let spa_config_update() initialize the new metaslabs. 4429 * 4430 * spa_load() checks for added-but-not-initialized vdevs, so that 4431 * if we lose power at any point in this sequence, the remaining 4432 * steps will be completed the next time we load the pool. 4433 */ 4434 (void) spa_vdev_exit(spa, vd, txg, 0); 4435 4436 mutex_enter(&spa_namespace_lock); 4437 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4438 mutex_exit(&spa_namespace_lock); 4439 4440 return (0); 4441 } 4442 4443 /* 4444 * Attach a device to a mirror. The arguments are the path to any device 4445 * in the mirror, and the nvroot for the new device. If the path specifies 4446 * a device that is not mirrored, we automatically insert the mirror vdev. 4447 * 4448 * If 'replacing' is specified, the new device is intended to replace the 4449 * existing device; in this case the two devices are made into their own 4450 * mirror using the 'replacing' vdev, which is functionally identical to 4451 * the mirror vdev (it actually reuses all the same ops) but has a few 4452 * extra rules: you can't attach to it after it's been created, and upon 4453 * completion of resilvering, the first disk (the one being replaced) 4454 * is automatically detached. 4455 */ 4456 int 4457 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4458 { 4459 uint64_t txg, dtl_max_txg; 4460 vdev_t *rvd = spa->spa_root_vdev; 4461 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4462 vdev_ops_t *pvops; 4463 char *oldvdpath, *newvdpath; 4464 int newvd_isspare; 4465 int error; 4466 4467 ASSERT(spa_writeable(spa)); 4468 4469 txg = spa_vdev_enter(spa); 4470 4471 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4472 4473 if (oldvd == NULL) 4474 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4475 4476 if (!oldvd->vdev_ops->vdev_op_leaf) 4477 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4478 4479 pvd = oldvd->vdev_parent; 4480 4481 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4482 VDEV_ALLOC_ATTACH)) != 0) 4483 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4484 4485 if (newrootvd->vdev_children != 1) 4486 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4487 4488 newvd = newrootvd->vdev_child[0]; 4489 4490 if (!newvd->vdev_ops->vdev_op_leaf) 4491 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4492 4493 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4494 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4495 4496 /* 4497 * Spares can't replace logs 4498 */ 4499 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4500 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4501 4502 if (!replacing) { 4503 /* 4504 * For attach, the only allowable parent is a mirror or the root 4505 * vdev. 4506 */ 4507 if (pvd->vdev_ops != &vdev_mirror_ops && 4508 pvd->vdev_ops != &vdev_root_ops) 4509 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4510 4511 pvops = &vdev_mirror_ops; 4512 } else { 4513 /* 4514 * Active hot spares can only be replaced by inactive hot 4515 * spares. 4516 */ 4517 if (pvd->vdev_ops == &vdev_spare_ops && 4518 oldvd->vdev_isspare && 4519 !spa_has_spare(spa, newvd->vdev_guid)) 4520 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4521 4522 /* 4523 * If the source is a hot spare, and the parent isn't already a 4524 * spare, then we want to create a new hot spare. Otherwise, we 4525 * want to create a replacing vdev. The user is not allowed to 4526 * attach to a spared vdev child unless the 'isspare' state is 4527 * the same (spare replaces spare, non-spare replaces 4528 * non-spare). 4529 */ 4530 if (pvd->vdev_ops == &vdev_replacing_ops && 4531 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4532 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4533 } else if (pvd->vdev_ops == &vdev_spare_ops && 4534 newvd->vdev_isspare != oldvd->vdev_isspare) { 4535 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4536 } 4537 4538 if (newvd->vdev_isspare) 4539 pvops = &vdev_spare_ops; 4540 else 4541 pvops = &vdev_replacing_ops; 4542 } 4543 4544 /* 4545 * Make sure the new device is big enough. 4546 */ 4547 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4548 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4549 4550 /* 4551 * The new device cannot have a higher alignment requirement 4552 * than the top-level vdev. 4553 */ 4554 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4555 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4556 4557 /* 4558 * If this is an in-place replacement, update oldvd's path and devid 4559 * to make it distinguishable from newvd, and unopenable from now on. 4560 */ 4561 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4562 spa_strfree(oldvd->vdev_path); 4563 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4564 KM_SLEEP); 4565 (void) sprintf(oldvd->vdev_path, "%s/%s", 4566 newvd->vdev_path, "old"); 4567 if (oldvd->vdev_devid != NULL) { 4568 spa_strfree(oldvd->vdev_devid); 4569 oldvd->vdev_devid = NULL; 4570 } 4571 } 4572 4573 /* mark the device being resilvered */ 4574 newvd->vdev_resilver_txg = txg; 4575 4576 /* 4577 * If the parent is not a mirror, or if we're replacing, insert the new 4578 * mirror/replacing/spare vdev above oldvd. 4579 */ 4580 if (pvd->vdev_ops != pvops) 4581 pvd = vdev_add_parent(oldvd, pvops); 4582 4583 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4584 ASSERT(pvd->vdev_ops == pvops); 4585 ASSERT(oldvd->vdev_parent == pvd); 4586 4587 /* 4588 * Extract the new device from its root and add it to pvd. 4589 */ 4590 vdev_remove_child(newrootvd, newvd); 4591 newvd->vdev_id = pvd->vdev_children; 4592 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4593 vdev_add_child(pvd, newvd); 4594 4595 tvd = newvd->vdev_top; 4596 ASSERT(pvd->vdev_top == tvd); 4597 ASSERT(tvd->vdev_parent == rvd); 4598 4599 vdev_config_dirty(tvd); 4600 4601 /* 4602 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4603 * for any dmu_sync-ed blocks. It will propagate upward when 4604 * spa_vdev_exit() calls vdev_dtl_reassess(). 4605 */ 4606 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4607 4608 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4609 dtl_max_txg - TXG_INITIAL); 4610 4611 if (newvd->vdev_isspare) { 4612 spa_spare_activate(newvd); 4613 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4614 } 4615 4616 oldvdpath = spa_strdup(oldvd->vdev_path); 4617 newvdpath = spa_strdup(newvd->vdev_path); 4618 newvd_isspare = newvd->vdev_isspare; 4619 4620 /* 4621 * Mark newvd's DTL dirty in this txg. 4622 */ 4623 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4624 4625 /* 4626 * Schedule the resilver to restart in the future. We do this to 4627 * ensure that dmu_sync-ed blocks have been stitched into the 4628 * respective datasets. 4629 */ 4630 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4631 4632 /* 4633 * Commit the config 4634 */ 4635 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4636 4637 spa_history_log_internal(spa, "vdev attach", NULL, 4638 "%s vdev=%s %s vdev=%s", 4639 replacing && newvd_isspare ? "spare in" : 4640 replacing ? "replace" : "attach", newvdpath, 4641 replacing ? "for" : "to", oldvdpath); 4642 4643 spa_strfree(oldvdpath); 4644 spa_strfree(newvdpath); 4645 4646 if (spa->spa_bootfs) 4647 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4648 4649 return (0); 4650 } 4651 4652 /* 4653 * Detach a device from a mirror or replacing vdev. 4654 * 4655 * If 'replace_done' is specified, only detach if the parent 4656 * is a replacing vdev. 4657 */ 4658 int 4659 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4660 { 4661 uint64_t txg; 4662 int error; 4663 vdev_t *rvd = spa->spa_root_vdev; 4664 vdev_t *vd, *pvd, *cvd, *tvd; 4665 boolean_t unspare = B_FALSE; 4666 uint64_t unspare_guid = 0; 4667 char *vdpath; 4668 4669 ASSERT(spa_writeable(spa)); 4670 4671 txg = spa_vdev_enter(spa); 4672 4673 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4674 4675 if (vd == NULL) 4676 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4677 4678 if (!vd->vdev_ops->vdev_op_leaf) 4679 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4680 4681 pvd = vd->vdev_parent; 4682 4683 /* 4684 * If the parent/child relationship is not as expected, don't do it. 4685 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4686 * vdev that's replacing B with C. The user's intent in replacing 4687 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4688 * the replace by detaching C, the expected behavior is to end up 4689 * M(A,B). But suppose that right after deciding to detach C, 4690 * the replacement of B completes. We would have M(A,C), and then 4691 * ask to detach C, which would leave us with just A -- not what 4692 * the user wanted. To prevent this, we make sure that the 4693 * parent/child relationship hasn't changed -- in this example, 4694 * that C's parent is still the replacing vdev R. 4695 */ 4696 if (pvd->vdev_guid != pguid && pguid != 0) 4697 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4698 4699 /* 4700 * Only 'replacing' or 'spare' vdevs can be replaced. 4701 */ 4702 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4703 pvd->vdev_ops != &vdev_spare_ops) 4704 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4705 4706 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4707 spa_version(spa) >= SPA_VERSION_SPARES); 4708 4709 /* 4710 * Only mirror, replacing, and spare vdevs support detach. 4711 */ 4712 if (pvd->vdev_ops != &vdev_replacing_ops && 4713 pvd->vdev_ops != &vdev_mirror_ops && 4714 pvd->vdev_ops != &vdev_spare_ops) 4715 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4716 4717 /* 4718 * If this device has the only valid copy of some data, 4719 * we cannot safely detach it. 4720 */ 4721 if (vdev_dtl_required(vd)) 4722 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4723 4724 ASSERT(pvd->vdev_children >= 2); 4725 4726 /* 4727 * If we are detaching the second disk from a replacing vdev, then 4728 * check to see if we changed the original vdev's path to have "/old" 4729 * at the end in spa_vdev_attach(). If so, undo that change now. 4730 */ 4731 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4732 vd->vdev_path != NULL) { 4733 size_t len = strlen(vd->vdev_path); 4734 4735 for (int c = 0; c < pvd->vdev_children; c++) { 4736 cvd = pvd->vdev_child[c]; 4737 4738 if (cvd == vd || cvd->vdev_path == NULL) 4739 continue; 4740 4741 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4742 strcmp(cvd->vdev_path + len, "/old") == 0) { 4743 spa_strfree(cvd->vdev_path); 4744 cvd->vdev_path = spa_strdup(vd->vdev_path); 4745 break; 4746 } 4747 } 4748 } 4749 4750 /* 4751 * If we are detaching the original disk from a spare, then it implies 4752 * that the spare should become a real disk, and be removed from the 4753 * active spare list for the pool. 4754 */ 4755 if (pvd->vdev_ops == &vdev_spare_ops && 4756 vd->vdev_id == 0 && 4757 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4758 unspare = B_TRUE; 4759 4760 /* 4761 * Erase the disk labels so the disk can be used for other things. 4762 * This must be done after all other error cases are handled, 4763 * but before we disembowel vd (so we can still do I/O to it). 4764 * But if we can't do it, don't treat the error as fatal -- 4765 * it may be that the unwritability of the disk is the reason 4766 * it's being detached! 4767 */ 4768 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4769 4770 /* 4771 * Remove vd from its parent and compact the parent's children. 4772 */ 4773 vdev_remove_child(pvd, vd); 4774 vdev_compact_children(pvd); 4775 4776 /* 4777 * Remember one of the remaining children so we can get tvd below. 4778 */ 4779 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4780 4781 /* 4782 * If we need to remove the remaining child from the list of hot spares, 4783 * do it now, marking the vdev as no longer a spare in the process. 4784 * We must do this before vdev_remove_parent(), because that can 4785 * change the GUID if it creates a new toplevel GUID. For a similar 4786 * reason, we must remove the spare now, in the same txg as the detach; 4787 * otherwise someone could attach a new sibling, change the GUID, and 4788 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4789 */ 4790 if (unspare) { 4791 ASSERT(cvd->vdev_isspare); 4792 spa_spare_remove(cvd); 4793 unspare_guid = cvd->vdev_guid; 4794 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4795 cvd->vdev_unspare = B_TRUE; 4796 } 4797 4798 /* 4799 * If the parent mirror/replacing vdev only has one child, 4800 * the parent is no longer needed. Remove it from the tree. 4801 */ 4802 if (pvd->vdev_children == 1) { 4803 if (pvd->vdev_ops == &vdev_spare_ops) 4804 cvd->vdev_unspare = B_FALSE; 4805 vdev_remove_parent(cvd); 4806 } 4807 4808 4809 /* 4810 * We don't set tvd until now because the parent we just removed 4811 * may have been the previous top-level vdev. 4812 */ 4813 tvd = cvd->vdev_top; 4814 ASSERT(tvd->vdev_parent == rvd); 4815 4816 /* 4817 * Reevaluate the parent vdev state. 4818 */ 4819 vdev_propagate_state(cvd); 4820 4821 /* 4822 * If the 'autoexpand' property is set on the pool then automatically 4823 * try to expand the size of the pool. For example if the device we 4824 * just detached was smaller than the others, it may be possible to 4825 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4826 * first so that we can obtain the updated sizes of the leaf vdevs. 4827 */ 4828 if (spa->spa_autoexpand) { 4829 vdev_reopen(tvd); 4830 vdev_expand(tvd, txg); 4831 } 4832 4833 vdev_config_dirty(tvd); 4834 4835 /* 4836 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4837 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4838 * But first make sure we're not on any *other* txg's DTL list, to 4839 * prevent vd from being accessed after it's freed. 4840 */ 4841 vdpath = spa_strdup(vd->vdev_path); 4842 for (int t = 0; t < TXG_SIZE; t++) 4843 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4844 vd->vdev_detached = B_TRUE; 4845 vdev_dirty(tvd, VDD_DTL, vd, txg); 4846 4847 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4848 4849 /* hang on to the spa before we release the lock */ 4850 spa_open_ref(spa, FTAG); 4851 4852 error = spa_vdev_exit(spa, vd, txg, 0); 4853 4854 spa_history_log_internal(spa, "detach", NULL, 4855 "vdev=%s", vdpath); 4856 spa_strfree(vdpath); 4857 4858 /* 4859 * If this was the removal of the original device in a hot spare vdev, 4860 * then we want to go through and remove the device from the hot spare 4861 * list of every other pool. 4862 */ 4863 if (unspare) { 4864 spa_t *altspa = NULL; 4865 4866 mutex_enter(&spa_namespace_lock); 4867 while ((altspa = spa_next(altspa)) != NULL) { 4868 if (altspa->spa_state != POOL_STATE_ACTIVE || 4869 altspa == spa) 4870 continue; 4871 4872 spa_open_ref(altspa, FTAG); 4873 mutex_exit(&spa_namespace_lock); 4874 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 4875 mutex_enter(&spa_namespace_lock); 4876 spa_close(altspa, FTAG); 4877 } 4878 mutex_exit(&spa_namespace_lock); 4879 4880 /* search the rest of the vdevs for spares to remove */ 4881 spa_vdev_resilver_done(spa); 4882 } 4883 4884 /* all done with the spa; OK to release */ 4885 mutex_enter(&spa_namespace_lock); 4886 spa_close(spa, FTAG); 4887 mutex_exit(&spa_namespace_lock); 4888 4889 return (error); 4890 } 4891 4892 /* 4893 * Split a set of devices from their mirrors, and create a new pool from them. 4894 */ 4895 int 4896 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 4897 nvlist_t *props, boolean_t exp) 4898 { 4899 int error = 0; 4900 uint64_t txg, *glist; 4901 spa_t *newspa; 4902 uint_t c, children, lastlog; 4903 nvlist_t **child, *nvl, *tmp; 4904 dmu_tx_t *tx; 4905 char *altroot = NULL; 4906 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 4907 boolean_t activate_slog; 4908 4909 ASSERT(spa_writeable(spa)); 4910 4911 txg = spa_vdev_enter(spa); 4912 4913 /* clear the log and flush everything up to now */ 4914 activate_slog = spa_passivate_log(spa); 4915 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4916 error = spa_offline_log(spa); 4917 txg = spa_vdev_config_enter(spa); 4918 4919 if (activate_slog) 4920 spa_activate_log(spa); 4921 4922 if (error != 0) 4923 return (spa_vdev_exit(spa, NULL, txg, error)); 4924 4925 /* check new spa name before going any further */ 4926 if (spa_lookup(newname) != NULL) 4927 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 4928 4929 /* 4930 * scan through all the children to ensure they're all mirrors 4931 */ 4932 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 4933 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 4934 &children) != 0) 4935 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4936 4937 /* first, check to ensure we've got the right child count */ 4938 rvd = spa->spa_root_vdev; 4939 lastlog = 0; 4940 for (c = 0; c < rvd->vdev_children; c++) { 4941 vdev_t *vd = rvd->vdev_child[c]; 4942 4943 /* don't count the holes & logs as children */ 4944 if (vd->vdev_islog || vd->vdev_ishole) { 4945 if (lastlog == 0) 4946 lastlog = c; 4947 continue; 4948 } 4949 4950 lastlog = 0; 4951 } 4952 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 4953 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4954 4955 /* next, ensure no spare or cache devices are part of the split */ 4956 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 4957 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 4958 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4959 4960 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 4961 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 4962 4963 /* then, loop over each vdev and validate it */ 4964 for (c = 0; c < children; c++) { 4965 uint64_t is_hole = 0; 4966 4967 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 4968 &is_hole); 4969 4970 if (is_hole != 0) { 4971 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 4972 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 4973 continue; 4974 } else { 4975 error = SET_ERROR(EINVAL); 4976 break; 4977 } 4978 } 4979 4980 /* which disk is going to be split? */ 4981 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 4982 &glist[c]) != 0) { 4983 error = SET_ERROR(EINVAL); 4984 break; 4985 } 4986 4987 /* look it up in the spa */ 4988 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 4989 if (vml[c] == NULL) { 4990 error = SET_ERROR(ENODEV); 4991 break; 4992 } 4993 4994 /* make sure there's nothing stopping the split */ 4995 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 4996 vml[c]->vdev_islog || 4997 vml[c]->vdev_ishole || 4998 vml[c]->vdev_isspare || 4999 vml[c]->vdev_isl2cache || 5000 !vdev_writeable(vml[c]) || 5001 vml[c]->vdev_children != 0 || 5002 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5003 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5004 error = SET_ERROR(EINVAL); 5005 break; 5006 } 5007 5008 if (vdev_dtl_required(vml[c])) { 5009 error = SET_ERROR(EBUSY); 5010 break; 5011 } 5012 5013 /* we need certain info from the top level */ 5014 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5015 vml[c]->vdev_top->vdev_ms_array) == 0); 5016 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5017 vml[c]->vdev_top->vdev_ms_shift) == 0); 5018 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5019 vml[c]->vdev_top->vdev_asize) == 0); 5020 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5021 vml[c]->vdev_top->vdev_ashift) == 0); 5022 } 5023 5024 if (error != 0) { 5025 kmem_free(vml, children * sizeof (vdev_t *)); 5026 kmem_free(glist, children * sizeof (uint64_t)); 5027 return (spa_vdev_exit(spa, NULL, txg, error)); 5028 } 5029 5030 /* stop writers from using the disks */ 5031 for (c = 0; c < children; c++) { 5032 if (vml[c] != NULL) 5033 vml[c]->vdev_offline = B_TRUE; 5034 } 5035 vdev_reopen(spa->spa_root_vdev); 5036 5037 /* 5038 * Temporarily record the splitting vdevs in the spa config. This 5039 * will disappear once the config is regenerated. 5040 */ 5041 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5042 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5043 glist, children) == 0); 5044 kmem_free(glist, children * sizeof (uint64_t)); 5045 5046 mutex_enter(&spa->spa_props_lock); 5047 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5048 nvl) == 0); 5049 mutex_exit(&spa->spa_props_lock); 5050 spa->spa_config_splitting = nvl; 5051 vdev_config_dirty(spa->spa_root_vdev); 5052 5053 /* configure and create the new pool */ 5054 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5055 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5056 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5057 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5058 spa_version(spa)) == 0); 5059 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5060 spa->spa_config_txg) == 0); 5061 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5062 spa_generate_guid(NULL)) == 0); 5063 (void) nvlist_lookup_string(props, 5064 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5065 5066 /* add the new pool to the namespace */ 5067 newspa = spa_add(newname, config, altroot); 5068 newspa->spa_config_txg = spa->spa_config_txg; 5069 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5070 5071 /* release the spa config lock, retaining the namespace lock */ 5072 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5073 5074 if (zio_injection_enabled) 5075 zio_handle_panic_injection(spa, FTAG, 1); 5076 5077 spa_activate(newspa, spa_mode_global); 5078 spa_async_suspend(newspa); 5079 5080 /* create the new pool from the disks of the original pool */ 5081 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5082 if (error) 5083 goto out; 5084 5085 /* if that worked, generate a real config for the new pool */ 5086 if (newspa->spa_root_vdev != NULL) { 5087 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5088 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5089 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5090 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5091 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5092 B_TRUE)); 5093 } 5094 5095 /* set the props */ 5096 if (props != NULL) { 5097 spa_configfile_set(newspa, props, B_FALSE); 5098 error = spa_prop_set(newspa, props); 5099 if (error) 5100 goto out; 5101 } 5102 5103 /* flush everything */ 5104 txg = spa_vdev_config_enter(newspa); 5105 vdev_config_dirty(newspa->spa_root_vdev); 5106 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5107 5108 if (zio_injection_enabled) 5109 zio_handle_panic_injection(spa, FTAG, 2); 5110 5111 spa_async_resume(newspa); 5112 5113 /* finally, update the original pool's config */ 5114 txg = spa_vdev_config_enter(spa); 5115 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5116 error = dmu_tx_assign(tx, TXG_WAIT); 5117 if (error != 0) 5118 dmu_tx_abort(tx); 5119 for (c = 0; c < children; c++) { 5120 if (vml[c] != NULL) { 5121 vdev_split(vml[c]); 5122 if (error == 0) 5123 spa_history_log_internal(spa, "detach", tx, 5124 "vdev=%s", vml[c]->vdev_path); 5125 vdev_free(vml[c]); 5126 } 5127 } 5128 vdev_config_dirty(spa->spa_root_vdev); 5129 spa->spa_config_splitting = NULL; 5130 nvlist_free(nvl); 5131 if (error == 0) 5132 dmu_tx_commit(tx); 5133 (void) spa_vdev_exit(spa, NULL, txg, 0); 5134 5135 if (zio_injection_enabled) 5136 zio_handle_panic_injection(spa, FTAG, 3); 5137 5138 /* split is complete; log a history record */ 5139 spa_history_log_internal(newspa, "split", NULL, 5140 "from pool %s", spa_name(spa)); 5141 5142 kmem_free(vml, children * sizeof (vdev_t *)); 5143 5144 /* if we're not going to mount the filesystems in userland, export */ 5145 if (exp) 5146 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5147 B_FALSE, B_FALSE); 5148 5149 return (error); 5150 5151 out: 5152 spa_unload(newspa); 5153 spa_deactivate(newspa); 5154 spa_remove(newspa); 5155 5156 txg = spa_vdev_config_enter(spa); 5157 5158 /* re-online all offlined disks */ 5159 for (c = 0; c < children; c++) { 5160 if (vml[c] != NULL) 5161 vml[c]->vdev_offline = B_FALSE; 5162 } 5163 vdev_reopen(spa->spa_root_vdev); 5164 5165 nvlist_free(spa->spa_config_splitting); 5166 spa->spa_config_splitting = NULL; 5167 (void) spa_vdev_exit(spa, NULL, txg, error); 5168 5169 kmem_free(vml, children * sizeof (vdev_t *)); 5170 return (error); 5171 } 5172 5173 static nvlist_t * 5174 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5175 { 5176 for (int i = 0; i < count; i++) { 5177 uint64_t guid; 5178 5179 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5180 &guid) == 0); 5181 5182 if (guid == target_guid) 5183 return (nvpp[i]); 5184 } 5185 5186 return (NULL); 5187 } 5188 5189 static void 5190 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5191 nvlist_t *dev_to_remove) 5192 { 5193 nvlist_t **newdev = NULL; 5194 5195 if (count > 1) 5196 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5197 5198 for (int i = 0, j = 0; i < count; i++) { 5199 if (dev[i] == dev_to_remove) 5200 continue; 5201 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5202 } 5203 5204 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5205 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5206 5207 for (int i = 0; i < count - 1; i++) 5208 nvlist_free(newdev[i]); 5209 5210 if (count > 1) 5211 kmem_free(newdev, (count - 1) * sizeof (void *)); 5212 } 5213 5214 /* 5215 * Evacuate the device. 5216 */ 5217 static int 5218 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5219 { 5220 uint64_t txg; 5221 int error = 0; 5222 5223 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5224 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5225 ASSERT(vd == vd->vdev_top); 5226 5227 /* 5228 * Evacuate the device. We don't hold the config lock as writer 5229 * since we need to do I/O but we do keep the 5230 * spa_namespace_lock held. Once this completes the device 5231 * should no longer have any blocks allocated on it. 5232 */ 5233 if (vd->vdev_islog) { 5234 if (vd->vdev_stat.vs_alloc != 0) 5235 error = spa_offline_log(spa); 5236 } else { 5237 error = SET_ERROR(ENOTSUP); 5238 } 5239 5240 if (error) 5241 return (error); 5242 5243 /* 5244 * The evacuation succeeded. Remove any remaining MOS metadata 5245 * associated with this vdev, and wait for these changes to sync. 5246 */ 5247 ASSERT0(vd->vdev_stat.vs_alloc); 5248 txg = spa_vdev_config_enter(spa); 5249 vd->vdev_removing = B_TRUE; 5250 vdev_dirty_leaves(vd, VDD_DTL, txg); 5251 vdev_config_dirty(vd); 5252 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5253 5254 return (0); 5255 } 5256 5257 /* 5258 * Complete the removal by cleaning up the namespace. 5259 */ 5260 static void 5261 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5262 { 5263 vdev_t *rvd = spa->spa_root_vdev; 5264 uint64_t id = vd->vdev_id; 5265 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5266 5267 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5268 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5269 ASSERT(vd == vd->vdev_top); 5270 5271 /* 5272 * Only remove any devices which are empty. 5273 */ 5274 if (vd->vdev_stat.vs_alloc != 0) 5275 return; 5276 5277 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5278 5279 if (list_link_active(&vd->vdev_state_dirty_node)) 5280 vdev_state_clean(vd); 5281 if (list_link_active(&vd->vdev_config_dirty_node)) 5282 vdev_config_clean(vd); 5283 5284 vdev_free(vd); 5285 5286 if (last_vdev) { 5287 vdev_compact_children(rvd); 5288 } else { 5289 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5290 vdev_add_child(rvd, vd); 5291 } 5292 vdev_config_dirty(rvd); 5293 5294 /* 5295 * Reassess the health of our root vdev. 5296 */ 5297 vdev_reopen(rvd); 5298 } 5299 5300 /* 5301 * Remove a device from the pool - 5302 * 5303 * Removing a device from the vdev namespace requires several steps 5304 * and can take a significant amount of time. As a result we use 5305 * the spa_vdev_config_[enter/exit] functions which allow us to 5306 * grab and release the spa_config_lock while still holding the namespace 5307 * lock. During each step the configuration is synced out. 5308 * 5309 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5310 * devices. 5311 */ 5312 int 5313 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5314 { 5315 vdev_t *vd; 5316 metaslab_group_t *mg; 5317 nvlist_t **spares, **l2cache, *nv; 5318 uint64_t txg = 0; 5319 uint_t nspares, nl2cache; 5320 int error = 0; 5321 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5322 5323 ASSERT(spa_writeable(spa)); 5324 5325 if (!locked) 5326 txg = spa_vdev_enter(spa); 5327 5328 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5329 5330 if (spa->spa_spares.sav_vdevs != NULL && 5331 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5332 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5333 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5334 /* 5335 * Only remove the hot spare if it's not currently in use 5336 * in this pool. 5337 */ 5338 if (vd == NULL || unspare) { 5339 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5340 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5341 spa_load_spares(spa); 5342 spa->spa_spares.sav_sync = B_TRUE; 5343 } else { 5344 error = SET_ERROR(EBUSY); 5345 } 5346 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5347 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5348 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5349 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5350 /* 5351 * Cache devices can always be removed. 5352 */ 5353 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5354 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5355 spa_load_l2cache(spa); 5356 spa->spa_l2cache.sav_sync = B_TRUE; 5357 } else if (vd != NULL && vd->vdev_islog) { 5358 ASSERT(!locked); 5359 ASSERT(vd == vd->vdev_top); 5360 5361 mg = vd->vdev_mg; 5362 5363 /* 5364 * Stop allocating from this vdev. 5365 */ 5366 metaslab_group_passivate(mg); 5367 5368 /* 5369 * Wait for the youngest allocations and frees to sync, 5370 * and then wait for the deferral of those frees to finish. 5371 */ 5372 spa_vdev_config_exit(spa, NULL, 5373 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5374 5375 /* 5376 * Attempt to evacuate the vdev. 5377 */ 5378 error = spa_vdev_remove_evacuate(spa, vd); 5379 5380 txg = spa_vdev_config_enter(spa); 5381 5382 /* 5383 * If we couldn't evacuate the vdev, unwind. 5384 */ 5385 if (error) { 5386 metaslab_group_activate(mg); 5387 return (spa_vdev_exit(spa, NULL, txg, error)); 5388 } 5389 5390 /* 5391 * Clean up the vdev namespace. 5392 */ 5393 spa_vdev_remove_from_namespace(spa, vd); 5394 5395 } else if (vd != NULL) { 5396 /* 5397 * Normal vdevs cannot be removed (yet). 5398 */ 5399 error = SET_ERROR(ENOTSUP); 5400 } else { 5401 /* 5402 * There is no vdev of any kind with the specified guid. 5403 */ 5404 error = SET_ERROR(ENOENT); 5405 } 5406 5407 if (!locked) 5408 return (spa_vdev_exit(spa, NULL, txg, error)); 5409 5410 return (error); 5411 } 5412 5413 /* 5414 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5415 * currently spared, so we can detach it. 5416 */ 5417 static vdev_t * 5418 spa_vdev_resilver_done_hunt(vdev_t *vd) 5419 { 5420 vdev_t *newvd, *oldvd; 5421 5422 for (int c = 0; c < vd->vdev_children; c++) { 5423 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5424 if (oldvd != NULL) 5425 return (oldvd); 5426 } 5427 5428 /* 5429 * Check for a completed replacement. We always consider the first 5430 * vdev in the list to be the oldest vdev, and the last one to be 5431 * the newest (see spa_vdev_attach() for how that works). In 5432 * the case where the newest vdev is faulted, we will not automatically 5433 * remove it after a resilver completes. This is OK as it will require 5434 * user intervention to determine which disk the admin wishes to keep. 5435 */ 5436 if (vd->vdev_ops == &vdev_replacing_ops) { 5437 ASSERT(vd->vdev_children > 1); 5438 5439 newvd = vd->vdev_child[vd->vdev_children - 1]; 5440 oldvd = vd->vdev_child[0]; 5441 5442 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5443 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5444 !vdev_dtl_required(oldvd)) 5445 return (oldvd); 5446 } 5447 5448 /* 5449 * Check for a completed resilver with the 'unspare' flag set. 5450 */ 5451 if (vd->vdev_ops == &vdev_spare_ops) { 5452 vdev_t *first = vd->vdev_child[0]; 5453 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5454 5455 if (last->vdev_unspare) { 5456 oldvd = first; 5457 newvd = last; 5458 } else if (first->vdev_unspare) { 5459 oldvd = last; 5460 newvd = first; 5461 } else { 5462 oldvd = NULL; 5463 } 5464 5465 if (oldvd != NULL && 5466 vdev_dtl_empty(newvd, DTL_MISSING) && 5467 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5468 !vdev_dtl_required(oldvd)) 5469 return (oldvd); 5470 5471 /* 5472 * If there are more than two spares attached to a disk, 5473 * and those spares are not required, then we want to 5474 * attempt to free them up now so that they can be used 5475 * by other pools. Once we're back down to a single 5476 * disk+spare, we stop removing them. 5477 */ 5478 if (vd->vdev_children > 2) { 5479 newvd = vd->vdev_child[1]; 5480 5481 if (newvd->vdev_isspare && last->vdev_isspare && 5482 vdev_dtl_empty(last, DTL_MISSING) && 5483 vdev_dtl_empty(last, DTL_OUTAGE) && 5484 !vdev_dtl_required(newvd)) 5485 return (newvd); 5486 } 5487 } 5488 5489 return (NULL); 5490 } 5491 5492 static void 5493 spa_vdev_resilver_done(spa_t *spa) 5494 { 5495 vdev_t *vd, *pvd, *ppvd; 5496 uint64_t guid, sguid, pguid, ppguid; 5497 5498 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5499 5500 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5501 pvd = vd->vdev_parent; 5502 ppvd = pvd->vdev_parent; 5503 guid = vd->vdev_guid; 5504 pguid = pvd->vdev_guid; 5505 ppguid = ppvd->vdev_guid; 5506 sguid = 0; 5507 /* 5508 * If we have just finished replacing a hot spared device, then 5509 * we need to detach the parent's first child (the original hot 5510 * spare) as well. 5511 */ 5512 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5513 ppvd->vdev_children == 2) { 5514 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5515 sguid = ppvd->vdev_child[1]->vdev_guid; 5516 } 5517 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5518 5519 spa_config_exit(spa, SCL_ALL, FTAG); 5520 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5521 return; 5522 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5523 return; 5524 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5525 } 5526 5527 spa_config_exit(spa, SCL_ALL, FTAG); 5528 } 5529 5530 /* 5531 * Update the stored path or FRU for this vdev. 5532 */ 5533 int 5534 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5535 boolean_t ispath) 5536 { 5537 vdev_t *vd; 5538 boolean_t sync = B_FALSE; 5539 5540 ASSERT(spa_writeable(spa)); 5541 5542 spa_vdev_state_enter(spa, SCL_ALL); 5543 5544 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5545 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5546 5547 if (!vd->vdev_ops->vdev_op_leaf) 5548 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5549 5550 if (ispath) { 5551 if (strcmp(value, vd->vdev_path) != 0) { 5552 spa_strfree(vd->vdev_path); 5553 vd->vdev_path = spa_strdup(value); 5554 sync = B_TRUE; 5555 } 5556 } else { 5557 if (vd->vdev_fru == NULL) { 5558 vd->vdev_fru = spa_strdup(value); 5559 sync = B_TRUE; 5560 } else if (strcmp(value, vd->vdev_fru) != 0) { 5561 spa_strfree(vd->vdev_fru); 5562 vd->vdev_fru = spa_strdup(value); 5563 sync = B_TRUE; 5564 } 5565 } 5566 5567 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5568 } 5569 5570 int 5571 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5572 { 5573 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5574 } 5575 5576 int 5577 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5578 { 5579 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5580 } 5581 5582 /* 5583 * ========================================================================== 5584 * SPA Scanning 5585 * ========================================================================== 5586 */ 5587 5588 int 5589 spa_scan_stop(spa_t *spa) 5590 { 5591 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5592 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5593 return (SET_ERROR(EBUSY)); 5594 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5595 } 5596 5597 int 5598 spa_scan(spa_t *spa, pool_scan_func_t func) 5599 { 5600 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5601 5602 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5603 return (SET_ERROR(ENOTSUP)); 5604 5605 /* 5606 * If a resilver was requested, but there is no DTL on a 5607 * writeable leaf device, we have nothing to do. 5608 */ 5609 if (func == POOL_SCAN_RESILVER && 5610 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5611 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5612 return (0); 5613 } 5614 5615 return (dsl_scan(spa->spa_dsl_pool, func)); 5616 } 5617 5618 /* 5619 * ========================================================================== 5620 * SPA async task processing 5621 * ========================================================================== 5622 */ 5623 5624 static void 5625 spa_async_remove(spa_t *spa, vdev_t *vd) 5626 { 5627 if (vd->vdev_remove_wanted) { 5628 vd->vdev_remove_wanted = B_FALSE; 5629 vd->vdev_delayed_close = B_FALSE; 5630 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5631 5632 /* 5633 * We want to clear the stats, but we don't want to do a full 5634 * vdev_clear() as that will cause us to throw away 5635 * degraded/faulted state as well as attempt to reopen the 5636 * device, all of which is a waste. 5637 */ 5638 vd->vdev_stat.vs_read_errors = 0; 5639 vd->vdev_stat.vs_write_errors = 0; 5640 vd->vdev_stat.vs_checksum_errors = 0; 5641 5642 vdev_state_dirty(vd->vdev_top); 5643 } 5644 5645 for (int c = 0; c < vd->vdev_children; c++) 5646 spa_async_remove(spa, vd->vdev_child[c]); 5647 } 5648 5649 static void 5650 spa_async_probe(spa_t *spa, vdev_t *vd) 5651 { 5652 if (vd->vdev_probe_wanted) { 5653 vd->vdev_probe_wanted = B_FALSE; 5654 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5655 } 5656 5657 for (int c = 0; c < vd->vdev_children; c++) 5658 spa_async_probe(spa, vd->vdev_child[c]); 5659 } 5660 5661 static void 5662 spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5663 { 5664 sysevent_id_t eid; 5665 nvlist_t *attr; 5666 char *physpath; 5667 5668 if (!spa->spa_autoexpand) 5669 return; 5670 5671 for (int c = 0; c < vd->vdev_children; c++) { 5672 vdev_t *cvd = vd->vdev_child[c]; 5673 spa_async_autoexpand(spa, cvd); 5674 } 5675 5676 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5677 return; 5678 5679 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5680 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5681 5682 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5683 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5684 5685 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5686 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 5687 5688 nvlist_free(attr); 5689 kmem_free(physpath, MAXPATHLEN); 5690 } 5691 5692 static void 5693 spa_async_thread(spa_t *spa) 5694 { 5695 int tasks; 5696 5697 ASSERT(spa->spa_sync_on); 5698 5699 mutex_enter(&spa->spa_async_lock); 5700 tasks = spa->spa_async_tasks; 5701 spa->spa_async_tasks = 0; 5702 mutex_exit(&spa->spa_async_lock); 5703 5704 /* 5705 * See if the config needs to be updated. 5706 */ 5707 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5708 uint64_t old_space, new_space; 5709 5710 mutex_enter(&spa_namespace_lock); 5711 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5712 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5713 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5714 mutex_exit(&spa_namespace_lock); 5715 5716 /* 5717 * If the pool grew as a result of the config update, 5718 * then log an internal history event. 5719 */ 5720 if (new_space != old_space) { 5721 spa_history_log_internal(spa, "vdev online", NULL, 5722 "pool '%s' size: %llu(+%llu)", 5723 spa_name(spa), new_space, new_space - old_space); 5724 } 5725 } 5726 5727 /* 5728 * See if any devices need to be marked REMOVED. 5729 */ 5730 if (tasks & SPA_ASYNC_REMOVE) { 5731 spa_vdev_state_enter(spa, SCL_NONE); 5732 spa_async_remove(spa, spa->spa_root_vdev); 5733 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5734 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5735 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5736 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5737 (void) spa_vdev_state_exit(spa, NULL, 0); 5738 } 5739 5740 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5741 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5742 spa_async_autoexpand(spa, spa->spa_root_vdev); 5743 spa_config_exit(spa, SCL_CONFIG, FTAG); 5744 } 5745 5746 /* 5747 * See if any devices need to be probed. 5748 */ 5749 if (tasks & SPA_ASYNC_PROBE) { 5750 spa_vdev_state_enter(spa, SCL_NONE); 5751 spa_async_probe(spa, spa->spa_root_vdev); 5752 (void) spa_vdev_state_exit(spa, NULL, 0); 5753 } 5754 5755 /* 5756 * If any devices are done replacing, detach them. 5757 */ 5758 if (tasks & SPA_ASYNC_RESILVER_DONE) 5759 spa_vdev_resilver_done(spa); 5760 5761 /* 5762 * Kick off a resilver. 5763 */ 5764 if (tasks & SPA_ASYNC_RESILVER) 5765 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5766 5767 /* 5768 * Kick off L2 cache rebuilding. 5769 */ 5770 if (tasks & SPA_ASYNC_L2CACHE_REBUILD) 5771 l2arc_spa_rebuild_start(spa); 5772 5773 /* 5774 * Let the world know that we're done. 5775 */ 5776 mutex_enter(&spa->spa_async_lock); 5777 spa->spa_async_thread = NULL; 5778 cv_broadcast(&spa->spa_async_cv); 5779 mutex_exit(&spa->spa_async_lock); 5780 thread_exit(); 5781 } 5782 5783 void 5784 spa_async_suspend(spa_t *spa) 5785 { 5786 mutex_enter(&spa->spa_async_lock); 5787 spa->spa_async_suspended++; 5788 while (spa->spa_async_thread != NULL) 5789 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5790 mutex_exit(&spa->spa_async_lock); 5791 } 5792 5793 void 5794 spa_async_resume(spa_t *spa) 5795 { 5796 mutex_enter(&spa->spa_async_lock); 5797 ASSERT(spa->spa_async_suspended != 0); 5798 spa->spa_async_suspended--; 5799 mutex_exit(&spa->spa_async_lock); 5800 } 5801 5802 static boolean_t 5803 spa_async_tasks_pending(spa_t *spa) 5804 { 5805 uint_t non_config_tasks; 5806 uint_t config_task; 5807 boolean_t config_task_suspended; 5808 5809 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; 5810 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 5811 if (spa->spa_ccw_fail_time == 0) { 5812 config_task_suspended = B_FALSE; 5813 } else { 5814 config_task_suspended = 5815 (gethrtime() - spa->spa_ccw_fail_time) < 5816 (zfs_ccw_retry_interval * NANOSEC); 5817 } 5818 5819 return (non_config_tasks || (config_task && !config_task_suspended)); 5820 } 5821 5822 static void 5823 spa_async_dispatch(spa_t *spa) 5824 { 5825 mutex_enter(&spa->spa_async_lock); 5826 if (spa_async_tasks_pending(spa) && 5827 !spa->spa_async_suspended && 5828 spa->spa_async_thread == NULL && 5829 rootdir != NULL) 5830 spa->spa_async_thread = thread_create(NULL, 0, 5831 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5832 mutex_exit(&spa->spa_async_lock); 5833 } 5834 5835 void 5836 spa_async_request(spa_t *spa, int task) 5837 { 5838 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5839 mutex_enter(&spa->spa_async_lock); 5840 spa->spa_async_tasks |= task; 5841 mutex_exit(&spa->spa_async_lock); 5842 } 5843 5844 /* 5845 * ========================================================================== 5846 * SPA syncing routines 5847 * ========================================================================== 5848 */ 5849 5850 static int 5851 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5852 { 5853 bpobj_t *bpo = arg; 5854 bpobj_enqueue(bpo, bp, tx); 5855 return (0); 5856 } 5857 5858 static int 5859 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5860 { 5861 zio_t *zio = arg; 5862 5863 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5864 zio->io_flags)); 5865 return (0); 5866 } 5867 5868 /* 5869 * Note: this simple function is not inlined to make it easier to dtrace the 5870 * amount of time spent syncing frees. 5871 */ 5872 static void 5873 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 5874 { 5875 zio_t *zio = zio_root(spa, NULL, NULL, 0); 5876 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 5877 VERIFY(zio_wait(zio) == 0); 5878 } 5879 5880 /* 5881 * Note: this simple function is not inlined to make it easier to dtrace the 5882 * amount of time spent syncing deferred frees. 5883 */ 5884 static void 5885 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 5886 { 5887 zio_t *zio = zio_root(spa, NULL, NULL, 0); 5888 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 5889 spa_free_sync_cb, zio, tx), ==, 0); 5890 VERIFY0(zio_wait(zio)); 5891 } 5892 5893 5894 static void 5895 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5896 { 5897 char *packed = NULL; 5898 size_t bufsize; 5899 size_t nvsize = 0; 5900 dmu_buf_t *db; 5901 5902 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5903 5904 /* 5905 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5906 * information. This avoids the dmu_buf_will_dirty() path and 5907 * saves us a pre-read to get data we don't actually care about. 5908 */ 5909 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5910 packed = kmem_alloc(bufsize, KM_SLEEP); 5911 5912 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5913 KM_SLEEP) == 0); 5914 bzero(packed + nvsize, bufsize - nvsize); 5915 5916 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5917 5918 kmem_free(packed, bufsize); 5919 5920 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5921 dmu_buf_will_dirty(db, tx); 5922 *(uint64_t *)db->db_data = nvsize; 5923 dmu_buf_rele(db, FTAG); 5924 } 5925 5926 static void 5927 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5928 const char *config, const char *entry) 5929 { 5930 nvlist_t *nvroot; 5931 nvlist_t **list; 5932 int i; 5933 5934 if (!sav->sav_sync) 5935 return; 5936 5937 /* 5938 * Update the MOS nvlist describing the list of available devices. 5939 * spa_validate_aux() will have already made sure this nvlist is 5940 * valid and the vdevs are labeled appropriately. 5941 */ 5942 if (sav->sav_object == 0) { 5943 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5944 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5945 sizeof (uint64_t), tx); 5946 VERIFY(zap_update(spa->spa_meta_objset, 5947 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5948 &sav->sav_object, tx) == 0); 5949 } 5950 5951 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5952 if (sav->sav_count == 0) { 5953 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 5954 } else { 5955 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5956 for (i = 0; i < sav->sav_count; i++) 5957 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 5958 B_FALSE, VDEV_CONFIG_L2CACHE); 5959 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5960 sav->sav_count) == 0); 5961 for (i = 0; i < sav->sav_count; i++) 5962 nvlist_free(list[i]); 5963 kmem_free(list, sav->sav_count * sizeof (void *)); 5964 } 5965 5966 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 5967 nvlist_free(nvroot); 5968 5969 sav->sav_sync = B_FALSE; 5970 } 5971 5972 static void 5973 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 5974 { 5975 nvlist_t *config; 5976 5977 if (list_is_empty(&spa->spa_config_dirty_list)) 5978 return; 5979 5980 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5981 5982 config = spa_config_generate(spa, spa->spa_root_vdev, 5983 dmu_tx_get_txg(tx), B_FALSE); 5984 5985 /* 5986 * If we're upgrading the spa version then make sure that 5987 * the config object gets updated with the correct version. 5988 */ 5989 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 5990 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5991 spa->spa_uberblock.ub_version); 5992 5993 spa_config_exit(spa, SCL_STATE, FTAG); 5994 5995 if (spa->spa_config_syncing) 5996 nvlist_free(spa->spa_config_syncing); 5997 spa->spa_config_syncing = config; 5998 5999 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6000 } 6001 6002 static void 6003 spa_sync_version(void *arg, dmu_tx_t *tx) 6004 { 6005 uint64_t *versionp = arg; 6006 uint64_t version = *versionp; 6007 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6008 6009 /* 6010 * Setting the version is special cased when first creating the pool. 6011 */ 6012 ASSERT(tx->tx_txg != TXG_INITIAL); 6013 6014 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6015 ASSERT(version >= spa_version(spa)); 6016 6017 spa->spa_uberblock.ub_version = version; 6018 vdev_config_dirty(spa->spa_root_vdev); 6019 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6020 } 6021 6022 /* 6023 * Set zpool properties. 6024 */ 6025 static void 6026 spa_sync_props(void *arg, dmu_tx_t *tx) 6027 { 6028 nvlist_t *nvp = arg; 6029 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6030 objset_t *mos = spa->spa_meta_objset; 6031 nvpair_t *elem = NULL; 6032 6033 mutex_enter(&spa->spa_props_lock); 6034 6035 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6036 uint64_t intval; 6037 char *strval, *fname; 6038 zpool_prop_t prop; 6039 const char *propname; 6040 zprop_type_t proptype; 6041 spa_feature_t fid; 6042 6043 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6044 case ZPROP_INVAL: 6045 /* 6046 * We checked this earlier in spa_prop_validate(). 6047 */ 6048 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6049 6050 fname = strchr(nvpair_name(elem), '@') + 1; 6051 VERIFY0(zfeature_lookup_name(fname, &fid)); 6052 6053 spa_feature_enable(spa, fid, tx); 6054 spa_history_log_internal(spa, "set", tx, 6055 "%s=enabled", nvpair_name(elem)); 6056 break; 6057 6058 case ZPOOL_PROP_VERSION: 6059 intval = fnvpair_value_uint64(elem); 6060 /* 6061 * The version is synced seperatly before other 6062 * properties and should be correct by now. 6063 */ 6064 ASSERT3U(spa_version(spa), >=, intval); 6065 break; 6066 6067 case ZPOOL_PROP_ALTROOT: 6068 /* 6069 * 'altroot' is a non-persistent property. It should 6070 * have been set temporarily at creation or import time. 6071 */ 6072 ASSERT(spa->spa_root != NULL); 6073 break; 6074 6075 case ZPOOL_PROP_READONLY: 6076 case ZPOOL_PROP_CACHEFILE: 6077 /* 6078 * 'readonly' and 'cachefile' are also non-persisitent 6079 * properties. 6080 */ 6081 break; 6082 case ZPOOL_PROP_COMMENT: 6083 strval = fnvpair_value_string(elem); 6084 if (spa->spa_comment != NULL) 6085 spa_strfree(spa->spa_comment); 6086 spa->spa_comment = spa_strdup(strval); 6087 /* 6088 * We need to dirty the configuration on all the vdevs 6089 * so that their labels get updated. It's unnecessary 6090 * to do this for pool creation since the vdev's 6091 * configuratoin has already been dirtied. 6092 */ 6093 if (tx->tx_txg != TXG_INITIAL) 6094 vdev_config_dirty(spa->spa_root_vdev); 6095 spa_history_log_internal(spa, "set", tx, 6096 "%s=%s", nvpair_name(elem), strval); 6097 break; 6098 default: 6099 /* 6100 * Set pool property values in the poolprops mos object. 6101 */ 6102 if (spa->spa_pool_props_object == 0) { 6103 spa->spa_pool_props_object = 6104 zap_create_link(mos, DMU_OT_POOL_PROPS, 6105 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6106 tx); 6107 } 6108 6109 /* normalize the property name */ 6110 propname = zpool_prop_to_name(prop); 6111 proptype = zpool_prop_get_type(prop); 6112 6113 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6114 ASSERT(proptype == PROP_TYPE_STRING); 6115 strval = fnvpair_value_string(elem); 6116 VERIFY0(zap_update(mos, 6117 spa->spa_pool_props_object, propname, 6118 1, strlen(strval) + 1, strval, tx)); 6119 spa_history_log_internal(spa, "set", tx, 6120 "%s=%s", nvpair_name(elem), strval); 6121 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6122 intval = fnvpair_value_uint64(elem); 6123 6124 if (proptype == PROP_TYPE_INDEX) { 6125 const char *unused; 6126 VERIFY0(zpool_prop_index_to_string( 6127 prop, intval, &unused)); 6128 } 6129 VERIFY0(zap_update(mos, 6130 spa->spa_pool_props_object, propname, 6131 8, 1, &intval, tx)); 6132 spa_history_log_internal(spa, "set", tx, 6133 "%s=%lld", nvpair_name(elem), intval); 6134 } else { 6135 ASSERT(0); /* not allowed */ 6136 } 6137 6138 switch (prop) { 6139 case ZPOOL_PROP_DELEGATION: 6140 spa->spa_delegation = intval; 6141 break; 6142 case ZPOOL_PROP_BOOTFS: 6143 spa->spa_bootfs = intval; 6144 break; 6145 case ZPOOL_PROP_FAILUREMODE: 6146 spa->spa_failmode = intval; 6147 break; 6148 case ZPOOL_PROP_AUTOEXPAND: 6149 spa->spa_autoexpand = intval; 6150 if (tx->tx_txg != TXG_INITIAL) 6151 spa_async_request(spa, 6152 SPA_ASYNC_AUTOEXPAND); 6153 break; 6154 case ZPOOL_PROP_DEDUPDITTO: 6155 spa->spa_dedup_ditto = intval; 6156 break; 6157 default: 6158 break; 6159 } 6160 } 6161 6162 } 6163 6164 mutex_exit(&spa->spa_props_lock); 6165 } 6166 6167 /* 6168 * Perform one-time upgrade on-disk changes. spa_version() does not 6169 * reflect the new version this txg, so there must be no changes this 6170 * txg to anything that the upgrade code depends on after it executes. 6171 * Therefore this must be called after dsl_pool_sync() does the sync 6172 * tasks. 6173 */ 6174 static void 6175 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6176 { 6177 dsl_pool_t *dp = spa->spa_dsl_pool; 6178 6179 ASSERT(spa->spa_sync_pass == 1); 6180 6181 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6182 6183 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6184 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6185 dsl_pool_create_origin(dp, tx); 6186 6187 /* Keeping the origin open increases spa_minref */ 6188 spa->spa_minref += 3; 6189 } 6190 6191 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6192 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6193 dsl_pool_upgrade_clones(dp, tx); 6194 } 6195 6196 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6197 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6198 dsl_pool_upgrade_dir_clones(dp, tx); 6199 6200 /* Keeping the freedir open increases spa_minref */ 6201 spa->spa_minref += 3; 6202 } 6203 6204 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6205 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6206 spa_feature_create_zap_objects(spa, tx); 6207 } 6208 6209 /* 6210 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6211 * when possibility to use lz4 compression for metadata was added 6212 * Old pools that have this feature enabled must be upgraded to have 6213 * this feature active 6214 */ 6215 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6216 boolean_t lz4_en = spa_feature_is_enabled(spa, 6217 SPA_FEATURE_LZ4_COMPRESS); 6218 boolean_t lz4_ac = spa_feature_is_active(spa, 6219 SPA_FEATURE_LZ4_COMPRESS); 6220 6221 if (lz4_en && !lz4_ac) 6222 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6223 } 6224 rrw_exit(&dp->dp_config_rwlock, FTAG); 6225 } 6226 6227 /* 6228 * Sync the specified transaction group. New blocks may be dirtied as 6229 * part of the process, so we iterate until it converges. 6230 */ 6231 void 6232 spa_sync(spa_t *spa, uint64_t txg) 6233 { 6234 dsl_pool_t *dp = spa->spa_dsl_pool; 6235 objset_t *mos = spa->spa_meta_objset; 6236 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6237 vdev_t *rvd = spa->spa_root_vdev; 6238 vdev_t *vd; 6239 dmu_tx_t *tx; 6240 int error; 6241 6242 VERIFY(spa_writeable(spa)); 6243 6244 /* 6245 * Lock out configuration changes. 6246 */ 6247 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6248 6249 spa->spa_syncing_txg = txg; 6250 spa->spa_sync_pass = 0; 6251 6252 /* 6253 * If there are any pending vdev state changes, convert them 6254 * into config changes that go out with this transaction group. 6255 */ 6256 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6257 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6258 /* 6259 * We need the write lock here because, for aux vdevs, 6260 * calling vdev_config_dirty() modifies sav_config. 6261 * This is ugly and will become unnecessary when we 6262 * eliminate the aux vdev wart by integrating all vdevs 6263 * into the root vdev tree. 6264 */ 6265 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6266 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6267 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6268 vdev_state_clean(vd); 6269 vdev_config_dirty(vd); 6270 } 6271 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6272 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6273 } 6274 spa_config_exit(spa, SCL_STATE, FTAG); 6275 6276 tx = dmu_tx_create_assigned(dp, txg); 6277 6278 spa->spa_sync_starttime = gethrtime(); 6279 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6280 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6281 6282 /* 6283 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6284 * set spa_deflate if we have no raid-z vdevs. 6285 */ 6286 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6287 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6288 int i; 6289 6290 for (i = 0; i < rvd->vdev_children; i++) { 6291 vd = rvd->vdev_child[i]; 6292 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6293 break; 6294 } 6295 if (i == rvd->vdev_children) { 6296 spa->spa_deflate = TRUE; 6297 VERIFY(0 == zap_add(spa->spa_meta_objset, 6298 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6299 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6300 } 6301 } 6302 6303 /* 6304 * Iterate to convergence. 6305 */ 6306 do { 6307 int pass = ++spa->spa_sync_pass; 6308 6309 spa_sync_config_object(spa, tx); 6310 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6311 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6312 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6313 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6314 spa_errlog_sync(spa, txg); 6315 dsl_pool_sync(dp, txg); 6316 6317 if (pass < zfs_sync_pass_deferred_free) { 6318 spa_sync_frees(spa, free_bpl, tx); 6319 } else { 6320 /* 6321 * We can not defer frees in pass 1, because 6322 * we sync the deferred frees later in pass 1. 6323 */ 6324 ASSERT3U(pass, >, 1); 6325 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6326 &spa->spa_deferred_bpobj, tx); 6327 } 6328 6329 ddt_sync(spa, txg); 6330 dsl_scan_sync(dp, tx); 6331 6332 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6333 vdev_sync(vd, txg); 6334 6335 if (pass == 1) { 6336 spa_sync_upgrades(spa, tx); 6337 ASSERT3U(txg, >=, 6338 spa->spa_uberblock.ub_rootbp.blk_birth); 6339 /* 6340 * Note: We need to check if the MOS is dirty 6341 * because we could have marked the MOS dirty 6342 * without updating the uberblock (e.g. if we 6343 * have sync tasks but no dirty user data). We 6344 * need to check the uberblock's rootbp because 6345 * it is updated if we have synced out dirty 6346 * data (though in this case the MOS will most 6347 * likely also be dirty due to second order 6348 * effects, we don't want to rely on that here). 6349 */ 6350 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6351 !dmu_objset_is_dirty(mos, txg)) { 6352 /* 6353 * Nothing changed on the first pass, 6354 * therefore this TXG is a no-op. Avoid 6355 * syncing deferred frees, so that we 6356 * can keep this TXG as a no-op. 6357 */ 6358 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6359 txg)); 6360 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6361 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6362 break; 6363 } 6364 spa_sync_deferred_frees(spa, tx); 6365 } 6366 6367 } while (dmu_objset_is_dirty(mos, txg)); 6368 6369 /* 6370 * Rewrite the vdev configuration (which includes the uberblock) 6371 * to commit the transaction group. 6372 * 6373 * If there are no dirty vdevs, we sync the uberblock to a few 6374 * random top-level vdevs that are known to be visible in the 6375 * config cache (see spa_vdev_add() for a complete description). 6376 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6377 */ 6378 for (;;) { 6379 /* 6380 * We hold SCL_STATE to prevent vdev open/close/etc. 6381 * while we're attempting to write the vdev labels. 6382 */ 6383 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6384 6385 if (list_is_empty(&spa->spa_config_dirty_list)) { 6386 vdev_t *svd[SPA_DVAS_PER_BP]; 6387 int svdcount = 0; 6388 int children = rvd->vdev_children; 6389 int c0 = spa_get_random(children); 6390 6391 for (int c = 0; c < children; c++) { 6392 vd = rvd->vdev_child[(c0 + c) % children]; 6393 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6394 continue; 6395 svd[svdcount++] = vd; 6396 if (svdcount == SPA_DVAS_PER_BP) 6397 break; 6398 } 6399 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6400 if (error != 0) 6401 error = vdev_config_sync(svd, svdcount, txg, 6402 B_TRUE); 6403 } else { 6404 error = vdev_config_sync(rvd->vdev_child, 6405 rvd->vdev_children, txg, B_FALSE); 6406 if (error != 0) 6407 error = vdev_config_sync(rvd->vdev_child, 6408 rvd->vdev_children, txg, B_TRUE); 6409 } 6410 6411 if (error == 0) 6412 spa->spa_last_synced_guid = rvd->vdev_guid; 6413 6414 spa_config_exit(spa, SCL_STATE, FTAG); 6415 6416 if (error == 0) 6417 break; 6418 zio_suspend(spa, NULL); 6419 zio_resume_wait(spa); 6420 } 6421 dmu_tx_commit(tx); 6422 6423 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6424 6425 /* 6426 * Clear the dirty config list. 6427 */ 6428 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6429 vdev_config_clean(vd); 6430 6431 /* 6432 * Now that the new config has synced transactionally, 6433 * let it become visible to the config cache. 6434 */ 6435 if (spa->spa_config_syncing != NULL) { 6436 spa_config_set(spa, spa->spa_config_syncing); 6437 spa->spa_config_txg = txg; 6438 spa->spa_config_syncing = NULL; 6439 } 6440 6441 spa->spa_ubsync = spa->spa_uberblock; 6442 6443 dsl_pool_sync_done(dp, txg); 6444 6445 /* 6446 * Update usable space statistics. 6447 */ 6448 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6449 vdev_sync_done(vd, txg); 6450 6451 spa_update_dspace(spa); 6452 6453 /* 6454 * It had better be the case that we didn't dirty anything 6455 * since vdev_config_sync(). 6456 */ 6457 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6458 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6459 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6460 6461 spa->spa_sync_pass = 0; 6462 6463 spa_config_exit(spa, SCL_CONFIG, FTAG); 6464 6465 spa_handle_ignored_writes(spa); 6466 6467 /* 6468 * If any async tasks have been requested, kick them off. 6469 */ 6470 spa_async_dispatch(spa); 6471 } 6472 6473 /* 6474 * Sync all pools. We don't want to hold the namespace lock across these 6475 * operations, so we take a reference on the spa_t and drop the lock during the 6476 * sync. 6477 */ 6478 void 6479 spa_sync_allpools(void) 6480 { 6481 spa_t *spa = NULL; 6482 mutex_enter(&spa_namespace_lock); 6483 while ((spa = spa_next(spa)) != NULL) { 6484 if (spa_state(spa) != POOL_STATE_ACTIVE || 6485 !spa_writeable(spa) || spa_suspended(spa)) 6486 continue; 6487 spa_open_ref(spa, FTAG); 6488 mutex_exit(&spa_namespace_lock); 6489 txg_wait_synced(spa_get_dsl(spa), 0); 6490 mutex_enter(&spa_namespace_lock); 6491 spa_close(spa, FTAG); 6492 } 6493 mutex_exit(&spa_namespace_lock); 6494 } 6495 6496 /* 6497 * ========================================================================== 6498 * Miscellaneous routines 6499 * ========================================================================== 6500 */ 6501 6502 /* 6503 * Remove all pools in the system. 6504 */ 6505 void 6506 spa_evict_all(void) 6507 { 6508 spa_t *spa; 6509 6510 /* 6511 * Remove all cached state. All pools should be closed now, 6512 * so every spa in the AVL tree should be unreferenced. 6513 */ 6514 mutex_enter(&spa_namespace_lock); 6515 while ((spa = spa_next(NULL)) != NULL) { 6516 /* 6517 * Stop async tasks. The async thread may need to detach 6518 * a device that's been replaced, which requires grabbing 6519 * spa_namespace_lock, so we must drop it here. 6520 */ 6521 spa_open_ref(spa, FTAG); 6522 mutex_exit(&spa_namespace_lock); 6523 spa_async_suspend(spa); 6524 mutex_enter(&spa_namespace_lock); 6525 spa_close(spa, FTAG); 6526 6527 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6528 spa_unload(spa); 6529 spa_deactivate(spa); 6530 } 6531 spa_remove(spa); 6532 } 6533 mutex_exit(&spa_namespace_lock); 6534 } 6535 6536 vdev_t * 6537 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6538 { 6539 vdev_t *vd; 6540 int i; 6541 6542 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6543 return (vd); 6544 6545 if (aux) { 6546 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6547 vd = spa->spa_l2cache.sav_vdevs[i]; 6548 if (vd->vdev_guid == guid) 6549 return (vd); 6550 } 6551 6552 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6553 vd = spa->spa_spares.sav_vdevs[i]; 6554 if (vd->vdev_guid == guid) 6555 return (vd); 6556 } 6557 } 6558 6559 return (NULL); 6560 } 6561 6562 void 6563 spa_upgrade(spa_t *spa, uint64_t version) 6564 { 6565 ASSERT(spa_writeable(spa)); 6566 6567 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6568 6569 /* 6570 * This should only be called for a non-faulted pool, and since a 6571 * future version would result in an unopenable pool, this shouldn't be 6572 * possible. 6573 */ 6574 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6575 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6576 6577 spa->spa_uberblock.ub_version = version; 6578 vdev_config_dirty(spa->spa_root_vdev); 6579 6580 spa_config_exit(spa, SCL_ALL, FTAG); 6581 6582 txg_wait_synced(spa_get_dsl(spa), 0); 6583 } 6584 6585 boolean_t 6586 spa_has_spare(spa_t *spa, uint64_t guid) 6587 { 6588 int i; 6589 uint64_t spareguid; 6590 spa_aux_vdev_t *sav = &spa->spa_spares; 6591 6592 for (i = 0; i < sav->sav_count; i++) 6593 if (sav->sav_vdevs[i]->vdev_guid == guid) 6594 return (B_TRUE); 6595 6596 for (i = 0; i < sav->sav_npending; i++) { 6597 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6598 &spareguid) == 0 && spareguid == guid) 6599 return (B_TRUE); 6600 } 6601 6602 return (B_FALSE); 6603 } 6604 6605 /* 6606 * Check if a pool has an active shared spare device. 6607 * Note: reference count of an active spare is 2, as a spare and as a replace 6608 */ 6609 static boolean_t 6610 spa_has_active_shared_spare(spa_t *spa) 6611 { 6612 int i, refcnt; 6613 uint64_t pool; 6614 spa_aux_vdev_t *sav = &spa->spa_spares; 6615 6616 for (i = 0; i < sav->sav_count; i++) { 6617 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6618 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6619 refcnt > 2) 6620 return (B_TRUE); 6621 } 6622 6623 return (B_FALSE); 6624 } 6625 6626 /* 6627 * Post a sysevent corresponding to the given event. The 'name' must be one of 6628 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6629 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6630 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6631 * or zdb as real changes. 6632 */ 6633 void 6634 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6635 { 6636 #ifdef _KERNEL 6637 sysevent_t *ev; 6638 sysevent_attr_list_t *attr = NULL; 6639 sysevent_value_t value; 6640 sysevent_id_t eid; 6641 6642 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6643 SE_SLEEP); 6644 6645 value.value_type = SE_DATA_TYPE_STRING; 6646 value.value.sv_string = spa_name(spa); 6647 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6648 goto done; 6649 6650 value.value_type = SE_DATA_TYPE_UINT64; 6651 value.value.sv_uint64 = spa_guid(spa); 6652 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6653 goto done; 6654 6655 if (vd) { 6656 value.value_type = SE_DATA_TYPE_UINT64; 6657 value.value.sv_uint64 = vd->vdev_guid; 6658 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6659 SE_SLEEP) != 0) 6660 goto done; 6661 6662 if (vd->vdev_path) { 6663 value.value_type = SE_DATA_TYPE_STRING; 6664 value.value.sv_string = vd->vdev_path; 6665 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6666 &value, SE_SLEEP) != 0) 6667 goto done; 6668 } 6669 } 6670 6671 if (sysevent_attach_attributes(ev, attr) != 0) 6672 goto done; 6673 attr = NULL; 6674 6675 (void) log_sysevent(ev, SE_SLEEP, &eid); 6676 6677 done: 6678 if (attr) 6679 sysevent_free_attr(attr); 6680 sysevent_free(ev); 6681 #endif 6682 } 6683