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, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 * Copyright 2013 Saso Kiselkov. All rights reserved. 28 * Copyright (c) 2014 Integros [integros.com] 29 * Copyright 2016 Toomas Soome <tsoome@me.com> 30 * Copyright 2017 Joyent, Inc. 31 * Copyright (c) 2017 Datto Inc. 32 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. 33 */ 34 35 /* 36 * SPA: Storage Pool Allocator 37 * 38 * This file contains all the routines used when modifying on-disk SPA state. 39 * This includes opening, importing, destroying, exporting a pool, and syncing a 40 * pool. 41 */ 42 43 #include <sys/zfs_context.h> 44 #include <sys/fm/fs/zfs.h> 45 #include <sys/spa_impl.h> 46 #include <sys/zio.h> 47 #include <sys/zio_checksum.h> 48 #include <sys/dmu.h> 49 #include <sys/dmu_tx.h> 50 #include <sys/zap.h> 51 #include <sys/zil.h> 52 #include <sys/ddt.h> 53 #include <sys/vdev_impl.h> 54 #include <sys/vdev_removal.h> 55 #include <sys/vdev_indirect_mapping.h> 56 #include <sys/vdev_indirect_births.h> 57 #include <sys/metaslab.h> 58 #include <sys/metaslab_impl.h> 59 #include <sys/uberblock_impl.h> 60 #include <sys/txg.h> 61 #include <sys/avl.h> 62 #include <sys/bpobj.h> 63 #include <sys/dmu_traverse.h> 64 #include <sys/dmu_objset.h> 65 #include <sys/unique.h> 66 #include <sys/dsl_pool.h> 67 #include <sys/dsl_dataset.h> 68 #include <sys/dsl_dir.h> 69 #include <sys/dsl_prop.h> 70 #include <sys/dsl_synctask.h> 71 #include <sys/fs/zfs.h> 72 #include <sys/arc.h> 73 #include <sys/callb.h> 74 #include <sys/systeminfo.h> 75 #include <sys/spa_boot.h> 76 #include <sys/zfs_ioctl.h> 77 #include <sys/dsl_scan.h> 78 #include <sys/zfeature.h> 79 #include <sys/dsl_destroy.h> 80 #include <sys/abd.h> 81 82 #ifdef _KERNEL 83 #include <sys/bootprops.h> 84 #include <sys/callb.h> 85 #include <sys/cpupart.h> 86 #include <sys/pool.h> 87 #include <sys/sysdc.h> 88 #include <sys/zone.h> 89 #endif /* _KERNEL */ 90 91 #include "zfs_prop.h" 92 #include "zfs_comutil.h" 93 94 /* 95 * The interval, in seconds, at which failed configuration cache file writes 96 * should be retried. 97 */ 98 int zfs_ccw_retry_interval = 300; 99 100 typedef enum zti_modes { 101 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 102 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 103 ZTI_MODE_NULL, /* don't create a taskq */ 104 ZTI_NMODES 105 } zti_modes_t; 106 107 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 108 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 109 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 110 111 #define ZTI_N(n) ZTI_P(n, 1) 112 #define ZTI_ONE ZTI_N(1) 113 114 typedef struct zio_taskq_info { 115 zti_modes_t zti_mode; 116 uint_t zti_value; 117 uint_t zti_count; 118 } zio_taskq_info_t; 119 120 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 121 "issue", "issue_high", "intr", "intr_high" 122 }; 123 124 /* 125 * This table defines the taskq settings for each ZFS I/O type. When 126 * initializing a pool, we use this table to create an appropriately sized 127 * taskq. Some operations are low volume and therefore have a small, static 128 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 129 * macros. Other operations process a large amount of data; the ZTI_BATCH 130 * macro causes us to create a taskq oriented for throughput. Some operations 131 * are so high frequency and short-lived that the taskq itself can become a a 132 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 133 * additional degree of parallelism specified by the number of threads per- 134 * taskq and the number of taskqs; when dispatching an event in this case, the 135 * particular taskq is chosen at random. 136 * 137 * The different taskq priorities are to handle the different contexts (issue 138 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 139 * need to be handled with minimum delay. 140 */ 141 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 142 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 144 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 145 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 146 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 149 }; 150 151 static void spa_sync_version(void *arg, dmu_tx_t *tx); 152 static void spa_sync_props(void *arg, dmu_tx_t *tx); 153 static boolean_t spa_has_active_shared_spare(spa_t *spa); 154 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport); 155 static void spa_vdev_resilver_done(spa_t *spa); 156 157 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 158 id_t zio_taskq_psrset_bind = PS_NONE; 159 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 160 uint_t zio_taskq_basedc = 80; /* base duty cycle */ 161 162 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 163 extern int zfs_sync_pass_deferred_free; 164 165 /* 166 * Report any spa_load_verify errors found, but do not fail spa_load. 167 * This is used by zdb to analyze non-idle pools. 168 */ 169 boolean_t spa_load_verify_dryrun = B_FALSE; 170 171 /* 172 * This (illegal) pool name is used when temporarily importing a spa_t in order 173 * to get the vdev stats associated with the imported devices. 174 */ 175 #define TRYIMPORT_NAME "$import" 176 177 /* 178 * For debugging purposes: print out vdev tree during pool import. 179 */ 180 boolean_t spa_load_print_vdev_tree = B_FALSE; 181 182 /* 183 * A non-zero value for zfs_max_missing_tvds means that we allow importing 184 * pools with missing top-level vdevs. This is strictly intended for advanced 185 * pool recovery cases since missing data is almost inevitable. Pools with 186 * missing devices can only be imported read-only for safety reasons, and their 187 * fail-mode will be automatically set to "continue". 188 * 189 * With 1 missing vdev we should be able to import the pool and mount all 190 * datasets. User data that was not modified after the missing device has been 191 * added should be recoverable. This means that snapshots created prior to the 192 * addition of that device should be completely intact. 193 * 194 * With 2 missing vdevs, some datasets may fail to mount since there are 195 * dataset statistics that are stored as regular metadata. Some data might be 196 * recoverable if those vdevs were added recently. 197 * 198 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries 199 * may be missing entirely. Chances of data recovery are very low. Note that 200 * there are also risks of performing an inadvertent rewind as we might be 201 * missing all the vdevs with the latest uberblocks. 202 */ 203 uint64_t zfs_max_missing_tvds = 0; 204 205 /* 206 * The parameters below are similar to zfs_max_missing_tvds but are only 207 * intended for a preliminary open of the pool with an untrusted config which 208 * might be incomplete or out-dated. 209 * 210 * We are more tolerant for pools opened from a cachefile since we could have 211 * an out-dated cachefile where a device removal was not registered. 212 * We could have set the limit arbitrarily high but in the case where devices 213 * are really missing we would want to return the proper error codes; we chose 214 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available 215 * and we get a chance to retrieve the trusted config. 216 */ 217 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; 218 219 /* 220 * In the case where config was assembled by scanning device paths (/dev/dsks 221 * by default) we are less tolerant since all the existing devices should have 222 * been detected and we want spa_load to return the right error codes. 223 */ 224 uint64_t zfs_max_missing_tvds_scan = 0; 225 226 /* 227 * Debugging aid that pauses spa_sync() towards the end. 228 */ 229 boolean_t zfs_pause_spa_sync = B_FALSE; 230 231 /* 232 * ========================================================================== 233 * SPA properties routines 234 * ========================================================================== 235 */ 236 237 /* 238 * Add a (source=src, propname=propval) list to an nvlist. 239 */ 240 static void 241 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 242 uint64_t intval, zprop_source_t src) 243 { 244 const char *propname = zpool_prop_to_name(prop); 245 nvlist_t *propval; 246 247 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 248 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 249 250 if (strval != NULL) 251 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 252 else 253 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 254 255 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 256 nvlist_free(propval); 257 } 258 259 /* 260 * Get property values from the spa configuration. 261 */ 262 static void 263 spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 264 { 265 vdev_t *rvd = spa->spa_root_vdev; 266 dsl_pool_t *pool = spa->spa_dsl_pool; 267 uint64_t size, alloc, cap, version; 268 zprop_source_t src = ZPROP_SRC_NONE; 269 spa_config_dirent_t *dp; 270 metaslab_class_t *mc = spa_normal_class(spa); 271 272 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 273 274 if (rvd != NULL) { 275 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 276 size = metaslab_class_get_space(spa_normal_class(spa)); 277 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 278 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 279 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 280 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 281 size - alloc, src); 282 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, 283 spa->spa_checkpoint_info.sci_dspace, src); 284 285 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 286 metaslab_class_fragmentation(mc), src); 287 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 288 metaslab_class_expandable_space(mc), src); 289 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 290 (spa_mode(spa) == FREAD), src); 291 292 cap = (size == 0) ? 0 : (alloc * 100 / size); 293 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 294 295 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 296 ddt_get_pool_dedup_ratio(spa), src); 297 298 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 299 rvd->vdev_state, src); 300 301 version = spa_version(spa); 302 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 303 src = ZPROP_SRC_DEFAULT; 304 else 305 src = ZPROP_SRC_LOCAL; 306 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 307 } 308 309 if (pool != NULL) { 310 /* 311 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 312 * when opening pools before this version freedir will be NULL. 313 */ 314 if (pool->dp_free_dir != NULL) { 315 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 316 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 317 src); 318 } else { 319 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 320 NULL, 0, src); 321 } 322 323 if (pool->dp_leak_dir != NULL) { 324 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 325 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 326 src); 327 } else { 328 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 329 NULL, 0, src); 330 } 331 } 332 333 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 334 335 if (spa->spa_comment != NULL) { 336 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 337 0, ZPROP_SRC_LOCAL); 338 } 339 340 if (spa->spa_root != NULL) 341 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 342 0, ZPROP_SRC_LOCAL); 343 344 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 345 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 346 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 347 } else { 348 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 349 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 350 } 351 352 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 353 if (dp->scd_path == NULL) { 354 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 355 "none", 0, ZPROP_SRC_LOCAL); 356 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 357 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 358 dp->scd_path, 0, ZPROP_SRC_LOCAL); 359 } 360 } 361 } 362 363 /* 364 * Get zpool property values. 365 */ 366 int 367 spa_prop_get(spa_t *spa, nvlist_t **nvp) 368 { 369 objset_t *mos = spa->spa_meta_objset; 370 zap_cursor_t zc; 371 zap_attribute_t za; 372 int err; 373 374 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 375 376 mutex_enter(&spa->spa_props_lock); 377 378 /* 379 * Get properties from the spa config. 380 */ 381 spa_prop_get_config(spa, nvp); 382 383 /* If no pool property object, no more prop to get. */ 384 if (mos == NULL || spa->spa_pool_props_object == 0) { 385 mutex_exit(&spa->spa_props_lock); 386 return (0); 387 } 388 389 /* 390 * Get properties from the MOS pool property object. 391 */ 392 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 393 (err = zap_cursor_retrieve(&zc, &za)) == 0; 394 zap_cursor_advance(&zc)) { 395 uint64_t intval = 0; 396 char *strval = NULL; 397 zprop_source_t src = ZPROP_SRC_DEFAULT; 398 zpool_prop_t prop; 399 400 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) 401 continue; 402 403 switch (za.za_integer_length) { 404 case 8: 405 /* integer property */ 406 if (za.za_first_integer != 407 zpool_prop_default_numeric(prop)) 408 src = ZPROP_SRC_LOCAL; 409 410 if (prop == ZPOOL_PROP_BOOTFS) { 411 dsl_pool_t *dp; 412 dsl_dataset_t *ds = NULL; 413 414 dp = spa_get_dsl(spa); 415 dsl_pool_config_enter(dp, FTAG); 416 if (err = dsl_dataset_hold_obj(dp, 417 za.za_first_integer, FTAG, &ds)) { 418 dsl_pool_config_exit(dp, FTAG); 419 break; 420 } 421 422 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 423 KM_SLEEP); 424 dsl_dataset_name(ds, strval); 425 dsl_dataset_rele(ds, FTAG); 426 dsl_pool_config_exit(dp, FTAG); 427 } else { 428 strval = NULL; 429 intval = za.za_first_integer; 430 } 431 432 spa_prop_add_list(*nvp, prop, strval, intval, src); 433 434 if (strval != NULL) 435 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 436 437 break; 438 439 case 1: 440 /* string property */ 441 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 442 err = zap_lookup(mos, spa->spa_pool_props_object, 443 za.za_name, 1, za.za_num_integers, strval); 444 if (err) { 445 kmem_free(strval, za.za_num_integers); 446 break; 447 } 448 spa_prop_add_list(*nvp, prop, strval, 0, src); 449 kmem_free(strval, za.za_num_integers); 450 break; 451 452 default: 453 break; 454 } 455 } 456 zap_cursor_fini(&zc); 457 mutex_exit(&spa->spa_props_lock); 458 out: 459 if (err && err != ENOENT) { 460 nvlist_free(*nvp); 461 *nvp = NULL; 462 return (err); 463 } 464 465 return (0); 466 } 467 468 /* 469 * Validate the given pool properties nvlist and modify the list 470 * for the property values to be set. 471 */ 472 static int 473 spa_prop_validate(spa_t *spa, nvlist_t *props) 474 { 475 nvpair_t *elem; 476 int error = 0, reset_bootfs = 0; 477 uint64_t objnum = 0; 478 boolean_t has_feature = B_FALSE; 479 480 elem = NULL; 481 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 482 uint64_t intval; 483 char *strval, *slash, *check, *fname; 484 const char *propname = nvpair_name(elem); 485 zpool_prop_t prop = zpool_name_to_prop(propname); 486 487 switch (prop) { 488 case ZPOOL_PROP_INVAL: 489 if (!zpool_prop_feature(propname)) { 490 error = SET_ERROR(EINVAL); 491 break; 492 } 493 494 /* 495 * Sanitize the input. 496 */ 497 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 498 error = SET_ERROR(EINVAL); 499 break; 500 } 501 502 if (nvpair_value_uint64(elem, &intval) != 0) { 503 error = SET_ERROR(EINVAL); 504 break; 505 } 506 507 if (intval != 0) { 508 error = SET_ERROR(EINVAL); 509 break; 510 } 511 512 fname = strchr(propname, '@') + 1; 513 if (zfeature_lookup_name(fname, NULL) != 0) { 514 error = SET_ERROR(EINVAL); 515 break; 516 } 517 518 has_feature = B_TRUE; 519 break; 520 521 case ZPOOL_PROP_VERSION: 522 error = nvpair_value_uint64(elem, &intval); 523 if (!error && 524 (intval < spa_version(spa) || 525 intval > SPA_VERSION_BEFORE_FEATURES || 526 has_feature)) 527 error = SET_ERROR(EINVAL); 528 break; 529 530 case ZPOOL_PROP_DELEGATION: 531 case ZPOOL_PROP_AUTOREPLACE: 532 case ZPOOL_PROP_LISTSNAPS: 533 case ZPOOL_PROP_AUTOEXPAND: 534 error = nvpair_value_uint64(elem, &intval); 535 if (!error && intval > 1) 536 error = SET_ERROR(EINVAL); 537 break; 538 539 case ZPOOL_PROP_BOOTFS: 540 /* 541 * If the pool version is less than SPA_VERSION_BOOTFS, 542 * or the pool is still being created (version == 0), 543 * the bootfs property cannot be set. 544 */ 545 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 546 error = SET_ERROR(ENOTSUP); 547 break; 548 } 549 550 /* 551 * Make sure the vdev config is bootable 552 */ 553 if (!vdev_is_bootable(spa->spa_root_vdev)) { 554 error = SET_ERROR(ENOTSUP); 555 break; 556 } 557 558 reset_bootfs = 1; 559 560 error = nvpair_value_string(elem, &strval); 561 562 if (!error) { 563 objset_t *os; 564 uint64_t propval; 565 566 if (strval == NULL || strval[0] == '\0') { 567 objnum = zpool_prop_default_numeric( 568 ZPOOL_PROP_BOOTFS); 569 break; 570 } 571 572 if (error = dmu_objset_hold(strval, FTAG, &os)) 573 break; 574 575 /* 576 * Must be ZPL, and its property settings 577 * must be supported by GRUB (compression 578 * is not gzip, and large blocks are not used). 579 */ 580 581 if (dmu_objset_type(os) != DMU_OST_ZFS) { 582 error = SET_ERROR(ENOTSUP); 583 } else if ((error = 584 dsl_prop_get_int_ds(dmu_objset_ds(os), 585 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 586 &propval)) == 0 && 587 !BOOTFS_COMPRESS_VALID(propval)) { 588 error = SET_ERROR(ENOTSUP); 589 } else { 590 objnum = dmu_objset_id(os); 591 } 592 dmu_objset_rele(os, FTAG); 593 } 594 break; 595 596 case ZPOOL_PROP_FAILUREMODE: 597 error = nvpair_value_uint64(elem, &intval); 598 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 599 intval > ZIO_FAILURE_MODE_PANIC)) 600 error = SET_ERROR(EINVAL); 601 602 /* 603 * This is a special case which only occurs when 604 * the pool has completely failed. This allows 605 * the user to change the in-core failmode property 606 * without syncing it out to disk (I/Os might 607 * currently be blocked). We do this by returning 608 * EIO to the caller (spa_prop_set) to trick it 609 * into thinking we encountered a property validation 610 * error. 611 */ 612 if (!error && spa_suspended(spa)) { 613 spa->spa_failmode = intval; 614 error = SET_ERROR(EIO); 615 } 616 break; 617 618 case ZPOOL_PROP_CACHEFILE: 619 if ((error = nvpair_value_string(elem, &strval)) != 0) 620 break; 621 622 if (strval[0] == '\0') 623 break; 624 625 if (strcmp(strval, "none") == 0) 626 break; 627 628 if (strval[0] != '/') { 629 error = SET_ERROR(EINVAL); 630 break; 631 } 632 633 slash = strrchr(strval, '/'); 634 ASSERT(slash != NULL); 635 636 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 637 strcmp(slash, "/..") == 0) 638 error = SET_ERROR(EINVAL); 639 break; 640 641 case ZPOOL_PROP_COMMENT: 642 if ((error = nvpair_value_string(elem, &strval)) != 0) 643 break; 644 for (check = strval; *check != '\0'; check++) { 645 /* 646 * The kernel doesn't have an easy isprint() 647 * check. For this kernel check, we merely 648 * check ASCII apart from DEL. Fix this if 649 * there is an easy-to-use kernel isprint(). 650 */ 651 if (*check >= 0x7f) { 652 error = SET_ERROR(EINVAL); 653 break; 654 } 655 } 656 if (strlen(strval) > ZPROP_MAX_COMMENT) 657 error = E2BIG; 658 break; 659 660 case ZPOOL_PROP_DEDUPDITTO: 661 if (spa_version(spa) < SPA_VERSION_DEDUP) 662 error = SET_ERROR(ENOTSUP); 663 else 664 error = nvpair_value_uint64(elem, &intval); 665 if (error == 0 && 666 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 667 error = SET_ERROR(EINVAL); 668 break; 669 } 670 671 if (error) 672 break; 673 } 674 675 if (!error && reset_bootfs) { 676 error = nvlist_remove(props, 677 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 678 679 if (!error) { 680 error = nvlist_add_uint64(props, 681 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 682 } 683 } 684 685 return (error); 686 } 687 688 void 689 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 690 { 691 char *cachefile; 692 spa_config_dirent_t *dp; 693 694 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 695 &cachefile) != 0) 696 return; 697 698 dp = kmem_alloc(sizeof (spa_config_dirent_t), 699 KM_SLEEP); 700 701 if (cachefile[0] == '\0') 702 dp->scd_path = spa_strdup(spa_config_path); 703 else if (strcmp(cachefile, "none") == 0) 704 dp->scd_path = NULL; 705 else 706 dp->scd_path = spa_strdup(cachefile); 707 708 list_insert_head(&spa->spa_config_list, dp); 709 if (need_sync) 710 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 711 } 712 713 int 714 spa_prop_set(spa_t *spa, nvlist_t *nvp) 715 { 716 int error; 717 nvpair_t *elem = NULL; 718 boolean_t need_sync = B_FALSE; 719 720 if ((error = spa_prop_validate(spa, nvp)) != 0) 721 return (error); 722 723 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 724 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 725 726 if (prop == ZPOOL_PROP_CACHEFILE || 727 prop == ZPOOL_PROP_ALTROOT || 728 prop == ZPOOL_PROP_READONLY) 729 continue; 730 731 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { 732 uint64_t ver; 733 734 if (prop == ZPOOL_PROP_VERSION) { 735 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 736 } else { 737 ASSERT(zpool_prop_feature(nvpair_name(elem))); 738 ver = SPA_VERSION_FEATURES; 739 need_sync = B_TRUE; 740 } 741 742 /* Save time if the version is already set. */ 743 if (ver == spa_version(spa)) 744 continue; 745 746 /* 747 * In addition to the pool directory object, we might 748 * create the pool properties object, the features for 749 * read object, the features for write object, or the 750 * feature descriptions object. 751 */ 752 error = dsl_sync_task(spa->spa_name, NULL, 753 spa_sync_version, &ver, 754 6, ZFS_SPACE_CHECK_RESERVED); 755 if (error) 756 return (error); 757 continue; 758 } 759 760 need_sync = B_TRUE; 761 break; 762 } 763 764 if (need_sync) { 765 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 766 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 767 } 768 769 return (0); 770 } 771 772 /* 773 * If the bootfs property value is dsobj, clear it. 774 */ 775 void 776 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 777 { 778 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 779 VERIFY(zap_remove(spa->spa_meta_objset, 780 spa->spa_pool_props_object, 781 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 782 spa->spa_bootfs = 0; 783 } 784 } 785 786 /*ARGSUSED*/ 787 static int 788 spa_change_guid_check(void *arg, dmu_tx_t *tx) 789 { 790 uint64_t *newguid = arg; 791 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 792 vdev_t *rvd = spa->spa_root_vdev; 793 uint64_t vdev_state; 794 795 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 796 int error = (spa_has_checkpoint(spa)) ? 797 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 798 return (SET_ERROR(error)); 799 } 800 801 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 802 vdev_state = rvd->vdev_state; 803 spa_config_exit(spa, SCL_STATE, FTAG); 804 805 if (vdev_state != VDEV_STATE_HEALTHY) 806 return (SET_ERROR(ENXIO)); 807 808 ASSERT3U(spa_guid(spa), !=, *newguid); 809 810 return (0); 811 } 812 813 static void 814 spa_change_guid_sync(void *arg, dmu_tx_t *tx) 815 { 816 uint64_t *newguid = arg; 817 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 818 uint64_t oldguid; 819 vdev_t *rvd = spa->spa_root_vdev; 820 821 oldguid = spa_guid(spa); 822 823 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 824 rvd->vdev_guid = *newguid; 825 rvd->vdev_guid_sum += (*newguid - oldguid); 826 vdev_config_dirty(rvd); 827 spa_config_exit(spa, SCL_STATE, FTAG); 828 829 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 830 oldguid, *newguid); 831 } 832 833 /* 834 * Change the GUID for the pool. This is done so that we can later 835 * re-import a pool built from a clone of our own vdevs. We will modify 836 * the root vdev's guid, our own pool guid, and then mark all of our 837 * vdevs dirty. Note that we must make sure that all our vdevs are 838 * online when we do this, or else any vdevs that weren't present 839 * would be orphaned from our pool. We are also going to issue a 840 * sysevent to update any watchers. 841 */ 842 int 843 spa_change_guid(spa_t *spa) 844 { 845 int error; 846 uint64_t guid; 847 848 mutex_enter(&spa->spa_vdev_top_lock); 849 mutex_enter(&spa_namespace_lock); 850 guid = spa_generate_guid(NULL); 851 852 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 853 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 854 855 if (error == 0) { 856 spa_write_cachefile(spa, B_FALSE, B_TRUE); 857 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); 858 } 859 860 mutex_exit(&spa_namespace_lock); 861 mutex_exit(&spa->spa_vdev_top_lock); 862 863 return (error); 864 } 865 866 /* 867 * ========================================================================== 868 * SPA state manipulation (open/create/destroy/import/export) 869 * ========================================================================== 870 */ 871 872 static int 873 spa_error_entry_compare(const void *a, const void *b) 874 { 875 spa_error_entry_t *sa = (spa_error_entry_t *)a; 876 spa_error_entry_t *sb = (spa_error_entry_t *)b; 877 int ret; 878 879 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 880 sizeof (zbookmark_phys_t)); 881 882 if (ret < 0) 883 return (-1); 884 else if (ret > 0) 885 return (1); 886 else 887 return (0); 888 } 889 890 /* 891 * Utility function which retrieves copies of the current logs and 892 * re-initializes them in the process. 893 */ 894 void 895 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 896 { 897 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 898 899 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 900 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 901 902 avl_create(&spa->spa_errlist_scrub, 903 spa_error_entry_compare, sizeof (spa_error_entry_t), 904 offsetof(spa_error_entry_t, se_avl)); 905 avl_create(&spa->spa_errlist_last, 906 spa_error_entry_compare, sizeof (spa_error_entry_t), 907 offsetof(spa_error_entry_t, se_avl)); 908 } 909 910 static void 911 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 912 { 913 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 914 enum zti_modes mode = ztip->zti_mode; 915 uint_t value = ztip->zti_value; 916 uint_t count = ztip->zti_count; 917 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 918 char name[32]; 919 uint_t flags = 0; 920 boolean_t batch = B_FALSE; 921 922 if (mode == ZTI_MODE_NULL) { 923 tqs->stqs_count = 0; 924 tqs->stqs_taskq = NULL; 925 return; 926 } 927 928 ASSERT3U(count, >, 0); 929 930 tqs->stqs_count = count; 931 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 932 933 switch (mode) { 934 case ZTI_MODE_FIXED: 935 ASSERT3U(value, >=, 1); 936 value = MAX(value, 1); 937 break; 938 939 case ZTI_MODE_BATCH: 940 batch = B_TRUE; 941 flags |= TASKQ_THREADS_CPU_PCT; 942 value = zio_taskq_batch_pct; 943 break; 944 945 default: 946 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 947 "spa_activate()", 948 zio_type_name[t], zio_taskq_types[q], mode, value); 949 break; 950 } 951 952 for (uint_t i = 0; i < count; i++) { 953 taskq_t *tq; 954 955 if (count > 1) { 956 (void) snprintf(name, sizeof (name), "%s_%s_%u", 957 zio_type_name[t], zio_taskq_types[q], i); 958 } else { 959 (void) snprintf(name, sizeof (name), "%s_%s", 960 zio_type_name[t], zio_taskq_types[q]); 961 } 962 963 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 964 if (batch) 965 flags |= TASKQ_DC_BATCH; 966 967 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 968 spa->spa_proc, zio_taskq_basedc, flags); 969 } else { 970 pri_t pri = maxclsyspri; 971 /* 972 * The write issue taskq can be extremely CPU 973 * intensive. Run it at slightly lower priority 974 * than the other taskqs. 975 */ 976 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 977 pri--; 978 979 tq = taskq_create_proc(name, value, pri, 50, 980 INT_MAX, spa->spa_proc, flags); 981 } 982 983 tqs->stqs_taskq[i] = tq; 984 } 985 } 986 987 static void 988 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 989 { 990 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 991 992 if (tqs->stqs_taskq == NULL) { 993 ASSERT0(tqs->stqs_count); 994 return; 995 } 996 997 for (uint_t i = 0; i < tqs->stqs_count; i++) { 998 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 999 taskq_destroy(tqs->stqs_taskq[i]); 1000 } 1001 1002 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 1003 tqs->stqs_taskq = NULL; 1004 } 1005 1006 /* 1007 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 1008 * Note that a type may have multiple discrete taskqs to avoid lock contention 1009 * on the taskq itself. In that case we choose which taskq at random by using 1010 * the low bits of gethrtime(). 1011 */ 1012 void 1013 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 1014 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 1015 { 1016 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1017 taskq_t *tq; 1018 1019 ASSERT3P(tqs->stqs_taskq, !=, NULL); 1020 ASSERT3U(tqs->stqs_count, !=, 0); 1021 1022 if (tqs->stqs_count == 1) { 1023 tq = tqs->stqs_taskq[0]; 1024 } else { 1025 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 1026 } 1027 1028 taskq_dispatch_ent(tq, func, arg, flags, ent); 1029 } 1030 1031 static void 1032 spa_create_zio_taskqs(spa_t *spa) 1033 { 1034 for (int t = 0; t < ZIO_TYPES; t++) { 1035 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1036 spa_taskqs_init(spa, t, q); 1037 } 1038 } 1039 } 1040 1041 #ifdef _KERNEL 1042 static void 1043 spa_thread(void *arg) 1044 { 1045 callb_cpr_t cprinfo; 1046 1047 spa_t *spa = arg; 1048 user_t *pu = PTOU(curproc); 1049 1050 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1051 spa->spa_name); 1052 1053 ASSERT(curproc != &p0); 1054 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1055 "zpool-%s", spa->spa_name); 1056 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1057 1058 /* bind this thread to the requested psrset */ 1059 if (zio_taskq_psrset_bind != PS_NONE) { 1060 pool_lock(); 1061 mutex_enter(&cpu_lock); 1062 mutex_enter(&pidlock); 1063 mutex_enter(&curproc->p_lock); 1064 1065 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1066 0, NULL, NULL) == 0) { 1067 curthread->t_bind_pset = zio_taskq_psrset_bind; 1068 } else { 1069 cmn_err(CE_WARN, 1070 "Couldn't bind process for zfs pool \"%s\" to " 1071 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1072 } 1073 1074 mutex_exit(&curproc->p_lock); 1075 mutex_exit(&pidlock); 1076 mutex_exit(&cpu_lock); 1077 pool_unlock(); 1078 } 1079 1080 if (zio_taskq_sysdc) { 1081 sysdc_thread_enter(curthread, 100, 0); 1082 } 1083 1084 spa->spa_proc = curproc; 1085 spa->spa_did = curthread->t_did; 1086 1087 spa_create_zio_taskqs(spa); 1088 1089 mutex_enter(&spa->spa_proc_lock); 1090 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1091 1092 spa->spa_proc_state = SPA_PROC_ACTIVE; 1093 cv_broadcast(&spa->spa_proc_cv); 1094 1095 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1096 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1097 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1098 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1099 1100 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1101 spa->spa_proc_state = SPA_PROC_GONE; 1102 spa->spa_proc = &p0; 1103 cv_broadcast(&spa->spa_proc_cv); 1104 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1105 1106 mutex_enter(&curproc->p_lock); 1107 lwp_exit(); 1108 } 1109 #endif 1110 1111 /* 1112 * Activate an uninitialized pool. 1113 */ 1114 static void 1115 spa_activate(spa_t *spa, int mode) 1116 { 1117 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1118 1119 spa->spa_state = POOL_STATE_ACTIVE; 1120 spa->spa_mode = mode; 1121 1122 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1123 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1124 1125 /* Try to create a covering process */ 1126 mutex_enter(&spa->spa_proc_lock); 1127 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1128 ASSERT(spa->spa_proc == &p0); 1129 spa->spa_did = 0; 1130 1131 /* Only create a process if we're going to be around a while. */ 1132 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1133 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1134 NULL, 0) == 0) { 1135 spa->spa_proc_state = SPA_PROC_CREATED; 1136 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1137 cv_wait(&spa->spa_proc_cv, 1138 &spa->spa_proc_lock); 1139 } 1140 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1141 ASSERT(spa->spa_proc != &p0); 1142 ASSERT(spa->spa_did != 0); 1143 } else { 1144 #ifdef _KERNEL 1145 cmn_err(CE_WARN, 1146 "Couldn't create process for zfs pool \"%s\"\n", 1147 spa->spa_name); 1148 #endif 1149 } 1150 } 1151 mutex_exit(&spa->spa_proc_lock); 1152 1153 /* If we didn't create a process, we need to create our taskqs. */ 1154 if (spa->spa_proc == &p0) { 1155 spa_create_zio_taskqs(spa); 1156 } 1157 1158 for (size_t i = 0; i < TXG_SIZE; i++) 1159 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0); 1160 1161 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1162 offsetof(vdev_t, vdev_config_dirty_node)); 1163 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1164 offsetof(objset_t, os_evicting_node)); 1165 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1166 offsetof(vdev_t, vdev_state_dirty_node)); 1167 1168 txg_list_create(&spa->spa_vdev_txg_list, spa, 1169 offsetof(struct vdev, vdev_txg_node)); 1170 1171 avl_create(&spa->spa_errlist_scrub, 1172 spa_error_entry_compare, sizeof (spa_error_entry_t), 1173 offsetof(spa_error_entry_t, se_avl)); 1174 avl_create(&spa->spa_errlist_last, 1175 spa_error_entry_compare, sizeof (spa_error_entry_t), 1176 offsetof(spa_error_entry_t, se_avl)); 1177 } 1178 1179 /* 1180 * Opposite of spa_activate(). 1181 */ 1182 static void 1183 spa_deactivate(spa_t *spa) 1184 { 1185 ASSERT(spa->spa_sync_on == B_FALSE); 1186 ASSERT(spa->spa_dsl_pool == NULL); 1187 ASSERT(spa->spa_root_vdev == NULL); 1188 ASSERT(spa->spa_async_zio_root == NULL); 1189 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1190 1191 spa_evicting_os_wait(spa); 1192 1193 txg_list_destroy(&spa->spa_vdev_txg_list); 1194 1195 list_destroy(&spa->spa_config_dirty_list); 1196 list_destroy(&spa->spa_evicting_os_list); 1197 list_destroy(&spa->spa_state_dirty_list); 1198 1199 for (int t = 0; t < ZIO_TYPES; t++) { 1200 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1201 spa_taskqs_fini(spa, t, q); 1202 } 1203 } 1204 1205 for (size_t i = 0; i < TXG_SIZE; i++) { 1206 ASSERT3P(spa->spa_txg_zio[i], !=, NULL); 1207 VERIFY0(zio_wait(spa->spa_txg_zio[i])); 1208 spa->spa_txg_zio[i] = NULL; 1209 } 1210 1211 metaslab_class_destroy(spa->spa_normal_class); 1212 spa->spa_normal_class = NULL; 1213 1214 metaslab_class_destroy(spa->spa_log_class); 1215 spa->spa_log_class = NULL; 1216 1217 /* 1218 * If this was part of an import or the open otherwise failed, we may 1219 * still have errors left in the queues. Empty them just in case. 1220 */ 1221 spa_errlog_drain(spa); 1222 1223 avl_destroy(&spa->spa_errlist_scrub); 1224 avl_destroy(&spa->spa_errlist_last); 1225 1226 spa->spa_state = POOL_STATE_UNINITIALIZED; 1227 1228 mutex_enter(&spa->spa_proc_lock); 1229 if (spa->spa_proc_state != SPA_PROC_NONE) { 1230 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1231 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1232 cv_broadcast(&spa->spa_proc_cv); 1233 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1234 ASSERT(spa->spa_proc != &p0); 1235 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1236 } 1237 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1238 spa->spa_proc_state = SPA_PROC_NONE; 1239 } 1240 ASSERT(spa->spa_proc == &p0); 1241 mutex_exit(&spa->spa_proc_lock); 1242 1243 /* 1244 * We want to make sure spa_thread() has actually exited the ZFS 1245 * module, so that the module can't be unloaded out from underneath 1246 * it. 1247 */ 1248 if (spa->spa_did != 0) { 1249 thread_join(spa->spa_did); 1250 spa->spa_did = 0; 1251 } 1252 } 1253 1254 /* 1255 * Verify a pool configuration, and construct the vdev tree appropriately. This 1256 * will create all the necessary vdevs in the appropriate layout, with each vdev 1257 * in the CLOSED state. This will prep the pool before open/creation/import. 1258 * All vdev validation is done by the vdev_alloc() routine. 1259 */ 1260 static int 1261 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1262 uint_t id, int atype) 1263 { 1264 nvlist_t **child; 1265 uint_t children; 1266 int error; 1267 1268 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1269 return (error); 1270 1271 if ((*vdp)->vdev_ops->vdev_op_leaf) 1272 return (0); 1273 1274 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1275 &child, &children); 1276 1277 if (error == ENOENT) 1278 return (0); 1279 1280 if (error) { 1281 vdev_free(*vdp); 1282 *vdp = NULL; 1283 return (SET_ERROR(EINVAL)); 1284 } 1285 1286 for (int c = 0; c < children; c++) { 1287 vdev_t *vd; 1288 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1289 atype)) != 0) { 1290 vdev_free(*vdp); 1291 *vdp = NULL; 1292 return (error); 1293 } 1294 } 1295 1296 ASSERT(*vdp != NULL); 1297 1298 return (0); 1299 } 1300 1301 /* 1302 * Opposite of spa_load(). 1303 */ 1304 static void 1305 spa_unload(spa_t *spa) 1306 { 1307 int i; 1308 1309 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1310 1311 spa_load_note(spa, "UNLOADING"); 1312 1313 /* 1314 * Stop async tasks. 1315 */ 1316 spa_async_suspend(spa); 1317 1318 /* 1319 * Stop syncing. 1320 */ 1321 if (spa->spa_sync_on) { 1322 txg_sync_stop(spa->spa_dsl_pool); 1323 spa->spa_sync_on = B_FALSE; 1324 } 1325 1326 /* 1327 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1328 * to call it earlier, before we wait for async i/o to complete. 1329 * This ensures that there is no async metaslab prefetching, by 1330 * calling taskq_wait(mg_taskq). 1331 */ 1332 if (spa->spa_root_vdev != NULL) { 1333 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1334 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1335 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1336 spa_config_exit(spa, SCL_ALL, FTAG); 1337 } 1338 1339 /* 1340 * Wait for any outstanding async I/O to complete. 1341 */ 1342 if (spa->spa_async_zio_root != NULL) { 1343 for (int i = 0; i < max_ncpus; i++) 1344 (void) zio_wait(spa->spa_async_zio_root[i]); 1345 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1346 spa->spa_async_zio_root = NULL; 1347 } 1348 1349 if (spa->spa_vdev_removal != NULL) { 1350 spa_vdev_removal_destroy(spa->spa_vdev_removal); 1351 spa->spa_vdev_removal = NULL; 1352 } 1353 1354 if (spa->spa_condense_zthr != NULL) { 1355 ASSERT(!zthr_isrunning(spa->spa_condense_zthr)); 1356 zthr_destroy(spa->spa_condense_zthr); 1357 spa->spa_condense_zthr = NULL; 1358 } 1359 1360 if (spa->spa_checkpoint_discard_zthr != NULL) { 1361 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr)); 1362 zthr_destroy(spa->spa_checkpoint_discard_zthr); 1363 spa->spa_checkpoint_discard_zthr = NULL; 1364 } 1365 1366 spa_condense_fini(spa); 1367 1368 bpobj_close(&spa->spa_deferred_bpobj); 1369 1370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1371 1372 /* 1373 * Close all vdevs. 1374 */ 1375 if (spa->spa_root_vdev) 1376 vdev_free(spa->spa_root_vdev); 1377 ASSERT(spa->spa_root_vdev == NULL); 1378 1379 /* 1380 * Close the dsl pool. 1381 */ 1382 if (spa->spa_dsl_pool) { 1383 dsl_pool_close(spa->spa_dsl_pool); 1384 spa->spa_dsl_pool = NULL; 1385 spa->spa_meta_objset = NULL; 1386 } 1387 1388 ddt_unload(spa); 1389 1390 /* 1391 * Drop and purge level 2 cache 1392 */ 1393 spa_l2cache_drop(spa); 1394 1395 for (i = 0; i < spa->spa_spares.sav_count; i++) 1396 vdev_free(spa->spa_spares.sav_vdevs[i]); 1397 if (spa->spa_spares.sav_vdevs) { 1398 kmem_free(spa->spa_spares.sav_vdevs, 1399 spa->spa_spares.sav_count * sizeof (void *)); 1400 spa->spa_spares.sav_vdevs = NULL; 1401 } 1402 if (spa->spa_spares.sav_config) { 1403 nvlist_free(spa->spa_spares.sav_config); 1404 spa->spa_spares.sav_config = NULL; 1405 } 1406 spa->spa_spares.sav_count = 0; 1407 1408 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1409 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1410 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1411 } 1412 if (spa->spa_l2cache.sav_vdevs) { 1413 kmem_free(spa->spa_l2cache.sav_vdevs, 1414 spa->spa_l2cache.sav_count * sizeof (void *)); 1415 spa->spa_l2cache.sav_vdevs = NULL; 1416 } 1417 if (spa->spa_l2cache.sav_config) { 1418 nvlist_free(spa->spa_l2cache.sav_config); 1419 spa->spa_l2cache.sav_config = NULL; 1420 } 1421 spa->spa_l2cache.sav_count = 0; 1422 1423 spa->spa_async_suspended = 0; 1424 1425 spa->spa_indirect_vdevs_loaded = B_FALSE; 1426 1427 if (spa->spa_comment != NULL) { 1428 spa_strfree(spa->spa_comment); 1429 spa->spa_comment = NULL; 1430 } 1431 1432 spa_config_exit(spa, SCL_ALL, FTAG); 1433 } 1434 1435 /* 1436 * Load (or re-load) the current list of vdevs describing the active spares for 1437 * this pool. When this is called, we have some form of basic information in 1438 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1439 * then re-generate a more complete list including status information. 1440 */ 1441 void 1442 spa_load_spares(spa_t *spa) 1443 { 1444 nvlist_t **spares; 1445 uint_t nspares; 1446 int i; 1447 vdev_t *vd, *tvd; 1448 1449 #ifndef _KERNEL 1450 /* 1451 * zdb opens both the current state of the pool and the 1452 * checkpointed state (if present), with a different spa_t. 1453 * 1454 * As spare vdevs are shared among open pools, we skip loading 1455 * them when we load the checkpointed state of the pool. 1456 */ 1457 if (!spa_writeable(spa)) 1458 return; 1459 #endif 1460 1461 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1462 1463 /* 1464 * First, close and free any existing spare vdevs. 1465 */ 1466 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1467 vd = spa->spa_spares.sav_vdevs[i]; 1468 1469 /* Undo the call to spa_activate() below */ 1470 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1471 B_FALSE)) != NULL && tvd->vdev_isspare) 1472 spa_spare_remove(tvd); 1473 vdev_close(vd); 1474 vdev_free(vd); 1475 } 1476 1477 if (spa->spa_spares.sav_vdevs) 1478 kmem_free(spa->spa_spares.sav_vdevs, 1479 spa->spa_spares.sav_count * sizeof (void *)); 1480 1481 if (spa->spa_spares.sav_config == NULL) 1482 nspares = 0; 1483 else 1484 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1485 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1486 1487 spa->spa_spares.sav_count = (int)nspares; 1488 spa->spa_spares.sav_vdevs = NULL; 1489 1490 if (nspares == 0) 1491 return; 1492 1493 /* 1494 * Construct the array of vdevs, opening them to get status in the 1495 * process. For each spare, there is potentially two different vdev_t 1496 * structures associated with it: one in the list of spares (used only 1497 * for basic validation purposes) and one in the active vdev 1498 * configuration (if it's spared in). During this phase we open and 1499 * validate each vdev on the spare list. If the vdev also exists in the 1500 * active configuration, then we also mark this vdev as an active spare. 1501 */ 1502 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1503 KM_SLEEP); 1504 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1505 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1506 VDEV_ALLOC_SPARE) == 0); 1507 ASSERT(vd != NULL); 1508 1509 spa->spa_spares.sav_vdevs[i] = vd; 1510 1511 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1512 B_FALSE)) != NULL) { 1513 if (!tvd->vdev_isspare) 1514 spa_spare_add(tvd); 1515 1516 /* 1517 * We only mark the spare active if we were successfully 1518 * able to load the vdev. Otherwise, importing a pool 1519 * with a bad active spare would result in strange 1520 * behavior, because multiple pool would think the spare 1521 * is actively in use. 1522 * 1523 * There is a vulnerability here to an equally bizarre 1524 * circumstance, where a dead active spare is later 1525 * brought back to life (onlined or otherwise). Given 1526 * the rarity of this scenario, and the extra complexity 1527 * it adds, we ignore the possibility. 1528 */ 1529 if (!vdev_is_dead(tvd)) 1530 spa_spare_activate(tvd); 1531 } 1532 1533 vd->vdev_top = vd; 1534 vd->vdev_aux = &spa->spa_spares; 1535 1536 if (vdev_open(vd) != 0) 1537 continue; 1538 1539 if (vdev_validate_aux(vd) == 0) 1540 spa_spare_add(vd); 1541 } 1542 1543 /* 1544 * Recompute the stashed list of spares, with status information 1545 * this time. 1546 */ 1547 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1548 DATA_TYPE_NVLIST_ARRAY) == 0); 1549 1550 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1551 KM_SLEEP); 1552 for (i = 0; i < spa->spa_spares.sav_count; i++) 1553 spares[i] = vdev_config_generate(spa, 1554 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1555 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1556 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1557 for (i = 0; i < spa->spa_spares.sav_count; i++) 1558 nvlist_free(spares[i]); 1559 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1560 } 1561 1562 /* 1563 * Load (or re-load) the current list of vdevs describing the active l2cache for 1564 * this pool. When this is called, we have some form of basic information in 1565 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1566 * then re-generate a more complete list including status information. 1567 * Devices which are already active have their details maintained, and are 1568 * not re-opened. 1569 */ 1570 void 1571 spa_load_l2cache(spa_t *spa) 1572 { 1573 nvlist_t **l2cache; 1574 uint_t nl2cache; 1575 int i, j, oldnvdevs; 1576 uint64_t guid; 1577 vdev_t *vd, **oldvdevs, **newvdevs; 1578 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1579 1580 #ifndef _KERNEL 1581 /* 1582 * zdb opens both the current state of the pool and the 1583 * checkpointed state (if present), with a different spa_t. 1584 * 1585 * As L2 caches are part of the ARC which is shared among open 1586 * pools, we skip loading them when we load the checkpointed 1587 * state of the pool. 1588 */ 1589 if (!spa_writeable(spa)) 1590 return; 1591 #endif 1592 1593 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1594 1595 if (sav->sav_config != NULL) { 1596 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1597 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1598 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1599 } else { 1600 nl2cache = 0; 1601 newvdevs = NULL; 1602 } 1603 1604 oldvdevs = sav->sav_vdevs; 1605 oldnvdevs = sav->sav_count; 1606 sav->sav_vdevs = NULL; 1607 sav->sav_count = 0; 1608 1609 /* 1610 * Process new nvlist of vdevs. 1611 */ 1612 for (i = 0; i < nl2cache; i++) { 1613 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1614 &guid) == 0); 1615 1616 newvdevs[i] = NULL; 1617 for (j = 0; j < oldnvdevs; j++) { 1618 vd = oldvdevs[j]; 1619 if (vd != NULL && guid == vd->vdev_guid) { 1620 /* 1621 * Retain previous vdev for add/remove ops. 1622 */ 1623 newvdevs[i] = vd; 1624 oldvdevs[j] = NULL; 1625 break; 1626 } 1627 } 1628 1629 if (newvdevs[i] == NULL) { 1630 /* 1631 * Create new vdev 1632 */ 1633 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1634 VDEV_ALLOC_L2CACHE) == 0); 1635 ASSERT(vd != NULL); 1636 newvdevs[i] = vd; 1637 1638 /* 1639 * Commit this vdev as an l2cache device, 1640 * even if it fails to open. 1641 */ 1642 spa_l2cache_add(vd); 1643 1644 vd->vdev_top = vd; 1645 vd->vdev_aux = sav; 1646 1647 spa_l2cache_activate(vd); 1648 1649 if (vdev_open(vd) != 0) 1650 continue; 1651 1652 (void) vdev_validate_aux(vd); 1653 1654 if (!vdev_is_dead(vd)) 1655 l2arc_add_vdev(spa, vd); 1656 } 1657 } 1658 1659 /* 1660 * Purge vdevs that were dropped 1661 */ 1662 for (i = 0; i < oldnvdevs; i++) { 1663 uint64_t pool; 1664 1665 vd = oldvdevs[i]; 1666 if (vd != NULL) { 1667 ASSERT(vd->vdev_isl2cache); 1668 1669 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1670 pool != 0ULL && l2arc_vdev_present(vd)) 1671 l2arc_remove_vdev(vd); 1672 vdev_clear_stats(vd); 1673 vdev_free(vd); 1674 } 1675 } 1676 1677 if (oldvdevs) 1678 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1679 1680 if (sav->sav_config == NULL) 1681 goto out; 1682 1683 sav->sav_vdevs = newvdevs; 1684 sav->sav_count = (int)nl2cache; 1685 1686 /* 1687 * Recompute the stashed list of l2cache devices, with status 1688 * information this time. 1689 */ 1690 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1691 DATA_TYPE_NVLIST_ARRAY) == 0); 1692 1693 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1694 for (i = 0; i < sav->sav_count; i++) 1695 l2cache[i] = vdev_config_generate(spa, 1696 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1697 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1698 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1699 out: 1700 for (i = 0; i < sav->sav_count; i++) 1701 nvlist_free(l2cache[i]); 1702 if (sav->sav_count) 1703 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1704 } 1705 1706 static int 1707 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1708 { 1709 dmu_buf_t *db; 1710 char *packed = NULL; 1711 size_t nvsize = 0; 1712 int error; 1713 *value = NULL; 1714 1715 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1716 if (error != 0) 1717 return (error); 1718 1719 nvsize = *(uint64_t *)db->db_data; 1720 dmu_buf_rele(db, FTAG); 1721 1722 packed = kmem_alloc(nvsize, KM_SLEEP); 1723 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1724 DMU_READ_PREFETCH); 1725 if (error == 0) 1726 error = nvlist_unpack(packed, nvsize, value, 0); 1727 kmem_free(packed, nvsize); 1728 1729 return (error); 1730 } 1731 1732 /* 1733 * Concrete top-level vdevs that are not missing and are not logs. At every 1734 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. 1735 */ 1736 static uint64_t 1737 spa_healthy_core_tvds(spa_t *spa) 1738 { 1739 vdev_t *rvd = spa->spa_root_vdev; 1740 uint64_t tvds = 0; 1741 1742 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 1743 vdev_t *vd = rvd->vdev_child[i]; 1744 if (vd->vdev_islog) 1745 continue; 1746 if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) 1747 tvds++; 1748 } 1749 1750 return (tvds); 1751 } 1752 1753 /* 1754 * Checks to see if the given vdev could not be opened, in which case we post a 1755 * sysevent to notify the autoreplace code that the device has been removed. 1756 */ 1757 static void 1758 spa_check_removed(vdev_t *vd) 1759 { 1760 for (uint64_t c = 0; c < vd->vdev_children; c++) 1761 spa_check_removed(vd->vdev_child[c]); 1762 1763 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1764 vdev_is_concrete(vd)) { 1765 zfs_post_autoreplace(vd->vdev_spa, vd); 1766 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); 1767 } 1768 } 1769 1770 static int 1771 spa_check_for_missing_logs(spa_t *spa) 1772 { 1773 vdev_t *rvd = spa->spa_root_vdev; 1774 1775 /* 1776 * If we're doing a normal import, then build up any additional 1777 * diagnostic information about missing log devices. 1778 * We'll pass this up to the user for further processing. 1779 */ 1780 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1781 nvlist_t **child, *nv; 1782 uint64_t idx = 0; 1783 1784 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1785 KM_SLEEP); 1786 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1787 1788 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1789 vdev_t *tvd = rvd->vdev_child[c]; 1790 1791 /* 1792 * We consider a device as missing only if it failed 1793 * to open (i.e. offline or faulted is not considered 1794 * as missing). 1795 */ 1796 if (tvd->vdev_islog && 1797 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1798 child[idx++] = vdev_config_generate(spa, tvd, 1799 B_FALSE, VDEV_CONFIG_MISSING); 1800 } 1801 } 1802 1803 if (idx > 0) { 1804 fnvlist_add_nvlist_array(nv, 1805 ZPOOL_CONFIG_CHILDREN, child, idx); 1806 fnvlist_add_nvlist(spa->spa_load_info, 1807 ZPOOL_CONFIG_MISSING_DEVICES, nv); 1808 1809 for (uint64_t i = 0; i < idx; i++) 1810 nvlist_free(child[i]); 1811 } 1812 nvlist_free(nv); 1813 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1814 1815 if (idx > 0) { 1816 spa_load_failed(spa, "some log devices are missing"); 1817 return (SET_ERROR(ENXIO)); 1818 } 1819 } else { 1820 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1821 vdev_t *tvd = rvd->vdev_child[c]; 1822 1823 if (tvd->vdev_islog && 1824 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1825 spa_set_log_state(spa, SPA_LOG_CLEAR); 1826 spa_load_note(spa, "some log devices are " 1827 "missing, ZIL is dropped."); 1828 break; 1829 } 1830 } 1831 } 1832 1833 return (0); 1834 } 1835 1836 /* 1837 * Check for missing log devices 1838 */ 1839 static boolean_t 1840 spa_check_logs(spa_t *spa) 1841 { 1842 boolean_t rv = B_FALSE; 1843 dsl_pool_t *dp = spa_get_dsl(spa); 1844 1845 switch (spa->spa_log_state) { 1846 case SPA_LOG_MISSING: 1847 /* need to recheck in case slog has been restored */ 1848 case SPA_LOG_UNKNOWN: 1849 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1850 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1851 if (rv) 1852 spa_set_log_state(spa, SPA_LOG_MISSING); 1853 break; 1854 } 1855 return (rv); 1856 } 1857 1858 static boolean_t 1859 spa_passivate_log(spa_t *spa) 1860 { 1861 vdev_t *rvd = spa->spa_root_vdev; 1862 boolean_t slog_found = B_FALSE; 1863 1864 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1865 1866 if (!spa_has_slogs(spa)) 1867 return (B_FALSE); 1868 1869 for (int c = 0; c < rvd->vdev_children; c++) { 1870 vdev_t *tvd = rvd->vdev_child[c]; 1871 metaslab_group_t *mg = tvd->vdev_mg; 1872 1873 if (tvd->vdev_islog) { 1874 metaslab_group_passivate(mg); 1875 slog_found = B_TRUE; 1876 } 1877 } 1878 1879 return (slog_found); 1880 } 1881 1882 static void 1883 spa_activate_log(spa_t *spa) 1884 { 1885 vdev_t *rvd = spa->spa_root_vdev; 1886 1887 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1888 1889 for (int c = 0; c < rvd->vdev_children; c++) { 1890 vdev_t *tvd = rvd->vdev_child[c]; 1891 metaslab_group_t *mg = tvd->vdev_mg; 1892 1893 if (tvd->vdev_islog) 1894 metaslab_group_activate(mg); 1895 } 1896 } 1897 1898 int 1899 spa_reset_logs(spa_t *spa) 1900 { 1901 int error; 1902 1903 error = dmu_objset_find(spa_name(spa), zil_reset, 1904 NULL, DS_FIND_CHILDREN); 1905 if (error == 0) { 1906 /* 1907 * We successfully offlined the log device, sync out the 1908 * current txg so that the "stubby" block can be removed 1909 * by zil_sync(). 1910 */ 1911 txg_wait_synced(spa->spa_dsl_pool, 0); 1912 } 1913 return (error); 1914 } 1915 1916 static void 1917 spa_aux_check_removed(spa_aux_vdev_t *sav) 1918 { 1919 for (int i = 0; i < sav->sav_count; i++) 1920 spa_check_removed(sav->sav_vdevs[i]); 1921 } 1922 1923 void 1924 spa_claim_notify(zio_t *zio) 1925 { 1926 spa_t *spa = zio->io_spa; 1927 1928 if (zio->io_error) 1929 return; 1930 1931 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1932 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1933 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1934 mutex_exit(&spa->spa_props_lock); 1935 } 1936 1937 typedef struct spa_load_error { 1938 uint64_t sle_meta_count; 1939 uint64_t sle_data_count; 1940 } spa_load_error_t; 1941 1942 static void 1943 spa_load_verify_done(zio_t *zio) 1944 { 1945 blkptr_t *bp = zio->io_bp; 1946 spa_load_error_t *sle = zio->io_private; 1947 dmu_object_type_t type = BP_GET_TYPE(bp); 1948 int error = zio->io_error; 1949 spa_t *spa = zio->io_spa; 1950 1951 abd_free(zio->io_abd); 1952 if (error) { 1953 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1954 type != DMU_OT_INTENT_LOG) 1955 atomic_inc_64(&sle->sle_meta_count); 1956 else 1957 atomic_inc_64(&sle->sle_data_count); 1958 } 1959 1960 mutex_enter(&spa->spa_scrub_lock); 1961 spa->spa_scrub_inflight--; 1962 cv_broadcast(&spa->spa_scrub_io_cv); 1963 mutex_exit(&spa->spa_scrub_lock); 1964 } 1965 1966 /* 1967 * Maximum number of concurrent scrub i/os to create while verifying 1968 * a pool while importing it. 1969 */ 1970 int spa_load_verify_maxinflight = 10000; 1971 boolean_t spa_load_verify_metadata = B_TRUE; 1972 boolean_t spa_load_verify_data = B_TRUE; 1973 1974 /*ARGSUSED*/ 1975 static int 1976 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1977 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1978 { 1979 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1980 return (0); 1981 /* 1982 * Note: normally this routine will not be called if 1983 * spa_load_verify_metadata is not set. However, it may be useful 1984 * to manually set the flag after the traversal has begun. 1985 */ 1986 if (!spa_load_verify_metadata) 1987 return (0); 1988 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 1989 return (0); 1990 1991 zio_t *rio = arg; 1992 size_t size = BP_GET_PSIZE(bp); 1993 1994 mutex_enter(&spa->spa_scrub_lock); 1995 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1996 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1997 spa->spa_scrub_inflight++; 1998 mutex_exit(&spa->spa_scrub_lock); 1999 2000 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2001 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2002 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2003 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2004 return (0); 2005 } 2006 2007 /* ARGSUSED */ 2008 int 2009 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2010 { 2011 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2012 return (SET_ERROR(ENAMETOOLONG)); 2013 2014 return (0); 2015 } 2016 2017 static int 2018 spa_load_verify(spa_t *spa) 2019 { 2020 zio_t *rio; 2021 spa_load_error_t sle = { 0 }; 2022 zpool_rewind_policy_t policy; 2023 boolean_t verify_ok = B_FALSE; 2024 int error = 0; 2025 2026 zpool_get_rewind_policy(spa->spa_config, &policy); 2027 2028 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 2029 return (0); 2030 2031 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2032 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2033 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2034 DS_FIND_CHILDREN); 2035 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2036 if (error != 0) 2037 return (error); 2038 2039 rio = zio_root(spa, NULL, &sle, 2040 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2041 2042 if (spa_load_verify_metadata) { 2043 if (spa->spa_extreme_rewind) { 2044 spa_load_note(spa, "performing a complete scan of the " 2045 "pool since extreme rewind is on. This may take " 2046 "a very long time.\n (spa_load_verify_data=%u, " 2047 "spa_load_verify_metadata=%u)", 2048 spa_load_verify_data, spa_load_verify_metadata); 2049 } 2050 error = traverse_pool(spa, spa->spa_verify_min_txg, 2051 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2052 spa_load_verify_cb, rio); 2053 } 2054 2055 (void) zio_wait(rio); 2056 2057 spa->spa_load_meta_errors = sle.sle_meta_count; 2058 spa->spa_load_data_errors = sle.sle_data_count; 2059 2060 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { 2061 spa_load_note(spa, "spa_load_verify found %llu metadata errors " 2062 "and %llu data errors", (u_longlong_t)sle.sle_meta_count, 2063 (u_longlong_t)sle.sle_data_count); 2064 } 2065 2066 if (spa_load_verify_dryrun || 2067 (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2068 sle.sle_data_count <= policy.zrp_maxdata)) { 2069 int64_t loss = 0; 2070 2071 verify_ok = B_TRUE; 2072 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2073 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2074 2075 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2076 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2077 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2078 VERIFY(nvlist_add_int64(spa->spa_load_info, 2079 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2080 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2081 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2082 } else { 2083 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2084 } 2085 2086 if (spa_load_verify_dryrun) 2087 return (0); 2088 2089 if (error) { 2090 if (error != ENXIO && error != EIO) 2091 error = SET_ERROR(EIO); 2092 return (error); 2093 } 2094 2095 return (verify_ok ? 0 : EIO); 2096 } 2097 2098 /* 2099 * Find a value in the pool props object. 2100 */ 2101 static void 2102 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2103 { 2104 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2105 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2106 } 2107 2108 /* 2109 * Find a value in the pool directory object. 2110 */ 2111 static int 2112 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) 2113 { 2114 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2115 name, sizeof (uint64_t), 1, val); 2116 2117 if (error != 0 && (error != ENOENT || log_enoent)) { 2118 spa_load_failed(spa, "couldn't get '%s' value in MOS directory " 2119 "[error=%d]", name, error); 2120 } 2121 2122 return (error); 2123 } 2124 2125 static int 2126 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2127 { 2128 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2129 return (SET_ERROR(err)); 2130 } 2131 2132 static void 2133 spa_spawn_aux_threads(spa_t *spa) 2134 { 2135 ASSERT(spa_writeable(spa)); 2136 2137 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2138 2139 spa_start_indirect_condensing_thread(spa); 2140 2141 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); 2142 spa->spa_checkpoint_discard_zthr = 2143 zthr_create(spa_checkpoint_discard_thread_check, 2144 spa_checkpoint_discard_thread, spa); 2145 } 2146 2147 /* 2148 * Fix up config after a partly-completed split. This is done with the 2149 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2150 * pool have that entry in their config, but only the splitting one contains 2151 * a list of all the guids of the vdevs that are being split off. 2152 * 2153 * This function determines what to do with that list: either rejoin 2154 * all the disks to the pool, or complete the splitting process. To attempt 2155 * the rejoin, each disk that is offlined is marked online again, and 2156 * we do a reopen() call. If the vdev label for every disk that was 2157 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2158 * then we call vdev_split() on each disk, and complete the split. 2159 * 2160 * Otherwise we leave the config alone, with all the vdevs in place in 2161 * the original pool. 2162 */ 2163 static void 2164 spa_try_repair(spa_t *spa, nvlist_t *config) 2165 { 2166 uint_t extracted; 2167 uint64_t *glist; 2168 uint_t i, gcount; 2169 nvlist_t *nvl; 2170 vdev_t **vd; 2171 boolean_t attempt_reopen; 2172 2173 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2174 return; 2175 2176 /* check that the config is complete */ 2177 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2178 &glist, &gcount) != 0) 2179 return; 2180 2181 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2182 2183 /* attempt to online all the vdevs & validate */ 2184 attempt_reopen = B_TRUE; 2185 for (i = 0; i < gcount; i++) { 2186 if (glist[i] == 0) /* vdev is hole */ 2187 continue; 2188 2189 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2190 if (vd[i] == NULL) { 2191 /* 2192 * Don't bother attempting to reopen the disks; 2193 * just do the split. 2194 */ 2195 attempt_reopen = B_FALSE; 2196 } else { 2197 /* attempt to re-online it */ 2198 vd[i]->vdev_offline = B_FALSE; 2199 } 2200 } 2201 2202 if (attempt_reopen) { 2203 vdev_reopen(spa->spa_root_vdev); 2204 2205 /* check each device to see what state it's in */ 2206 for (extracted = 0, i = 0; i < gcount; i++) { 2207 if (vd[i] != NULL && 2208 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2209 break; 2210 ++extracted; 2211 } 2212 } 2213 2214 /* 2215 * If every disk has been moved to the new pool, or if we never 2216 * even attempted to look at them, then we split them off for 2217 * good. 2218 */ 2219 if (!attempt_reopen || gcount == extracted) { 2220 for (i = 0; i < gcount; i++) 2221 if (vd[i] != NULL) 2222 vdev_split(vd[i]); 2223 vdev_reopen(spa->spa_root_vdev); 2224 } 2225 2226 kmem_free(vd, gcount * sizeof (vdev_t *)); 2227 } 2228 2229 static int 2230 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) 2231 { 2232 char *ereport = FM_EREPORT_ZFS_POOL; 2233 int error; 2234 2235 spa->spa_load_state = state; 2236 2237 gethrestime(&spa->spa_loaded_ts); 2238 error = spa_load_impl(spa, type, &ereport); 2239 2240 /* 2241 * Don't count references from objsets that are already closed 2242 * and are making their way through the eviction process. 2243 */ 2244 spa_evicting_os_wait(spa); 2245 spa->spa_minref = refcount_count(&spa->spa_refcount); 2246 if (error) { 2247 if (error != EEXIST) { 2248 spa->spa_loaded_ts.tv_sec = 0; 2249 spa->spa_loaded_ts.tv_nsec = 0; 2250 } 2251 if (error != EBADF) { 2252 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2253 } 2254 } 2255 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2256 spa->spa_ena = 0; 2257 2258 return (error); 2259 } 2260 2261 /* 2262 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2263 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2264 * spa's per-vdev ZAP list. 2265 */ 2266 static uint64_t 2267 vdev_count_verify_zaps(vdev_t *vd) 2268 { 2269 spa_t *spa = vd->vdev_spa; 2270 uint64_t total = 0; 2271 if (vd->vdev_top_zap != 0) { 2272 total++; 2273 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2274 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2275 } 2276 if (vd->vdev_leaf_zap != 0) { 2277 total++; 2278 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2279 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2280 } 2281 2282 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2283 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2284 } 2285 2286 return (total); 2287 } 2288 2289 static int 2290 spa_verify_host(spa_t *spa, nvlist_t *mos_config) 2291 { 2292 uint64_t hostid; 2293 char *hostname; 2294 uint64_t myhostid = 0; 2295 2296 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, 2297 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2298 hostname = fnvlist_lookup_string(mos_config, 2299 ZPOOL_CONFIG_HOSTNAME); 2300 2301 myhostid = zone_get_hostid(NULL); 2302 2303 if (hostid != 0 && myhostid != 0 && hostid != myhostid) { 2304 cmn_err(CE_WARN, "pool '%s' could not be " 2305 "loaded as it was last accessed by " 2306 "another system (host: %s hostid: 0x%llx). " 2307 "See: http://illumos.org/msg/ZFS-8000-EY", 2308 spa_name(spa), hostname, (u_longlong_t)hostid); 2309 spa_load_failed(spa, "hostid verification failed: pool " 2310 "last accessed by host: %s (hostid: 0x%llx)", 2311 hostname, (u_longlong_t)hostid); 2312 return (SET_ERROR(EBADF)); 2313 } 2314 } 2315 2316 return (0); 2317 } 2318 2319 static int 2320 spa_ld_parse_config(spa_t *spa, spa_import_type_t type) 2321 { 2322 int error = 0; 2323 nvlist_t *nvtree, *nvl, *config = spa->spa_config; 2324 int parse; 2325 vdev_t *rvd; 2326 uint64_t pool_guid; 2327 char *comment; 2328 2329 /* 2330 * Versioning wasn't explicitly added to the label until later, so if 2331 * it's not present treat it as the initial version. 2332 */ 2333 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2334 &spa->spa_ubsync.ub_version) != 0) 2335 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2336 2337 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 2338 spa_load_failed(spa, "invalid config provided: '%s' missing", 2339 ZPOOL_CONFIG_POOL_GUID); 2340 return (SET_ERROR(EINVAL)); 2341 } 2342 2343 /* 2344 * If we are doing an import, ensure that the pool is not already 2345 * imported by checking if its pool guid already exists in the 2346 * spa namespace. 2347 * 2348 * The only case that we allow an already imported pool to be 2349 * imported again, is when the pool is checkpointed and we want to 2350 * look at its checkpointed state from userland tools like zdb. 2351 */ 2352 #ifdef _KERNEL 2353 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2354 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2355 spa_guid_exists(pool_guid, 0)) { 2356 #else 2357 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2358 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2359 spa_guid_exists(pool_guid, 0) && 2360 !spa_importing_readonly_checkpoint(spa)) { 2361 #endif 2362 spa_load_failed(spa, "a pool with guid %llu is already open", 2363 (u_longlong_t)pool_guid); 2364 return (SET_ERROR(EEXIST)); 2365 } 2366 2367 spa->spa_config_guid = pool_guid; 2368 2369 nvlist_free(spa->spa_load_info); 2370 spa->spa_load_info = fnvlist_alloc(); 2371 2372 ASSERT(spa->spa_comment == NULL); 2373 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2374 spa->spa_comment = spa_strdup(comment); 2375 2376 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2377 &spa->spa_config_txg); 2378 2379 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) 2380 spa->spa_config_splitting = fnvlist_dup(nvl); 2381 2382 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { 2383 spa_load_failed(spa, "invalid config provided: '%s' missing", 2384 ZPOOL_CONFIG_VDEV_TREE); 2385 return (SET_ERROR(EINVAL)); 2386 } 2387 2388 /* 2389 * Create "The Godfather" zio to hold all async IOs 2390 */ 2391 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2392 KM_SLEEP); 2393 for (int i = 0; i < max_ncpus; i++) { 2394 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2395 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2396 ZIO_FLAG_GODFATHER); 2397 } 2398 2399 /* 2400 * Parse the configuration into a vdev tree. We explicitly set the 2401 * value that will be returned by spa_version() since parsing the 2402 * configuration requires knowing the version number. 2403 */ 2404 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2405 parse = (type == SPA_IMPORT_EXISTING ? 2406 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2407 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); 2408 spa_config_exit(spa, SCL_ALL, FTAG); 2409 2410 if (error != 0) { 2411 spa_load_failed(spa, "unable to parse config [error=%d]", 2412 error); 2413 return (error); 2414 } 2415 2416 ASSERT(spa->spa_root_vdev == rvd); 2417 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2418 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2419 2420 if (type != SPA_IMPORT_ASSEMBLE) { 2421 ASSERT(spa_guid(spa) == pool_guid); 2422 } 2423 2424 return (0); 2425 } 2426 2427 /* 2428 * Recursively open all vdevs in the vdev tree. This function is called twice: 2429 * first with the untrusted config, then with the trusted config. 2430 */ 2431 static int 2432 spa_ld_open_vdevs(spa_t *spa) 2433 { 2434 int error = 0; 2435 2436 /* 2437 * spa_missing_tvds_allowed defines how many top-level vdevs can be 2438 * missing/unopenable for the root vdev to be still considered openable. 2439 */ 2440 if (spa->spa_trust_config) { 2441 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; 2442 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { 2443 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; 2444 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { 2445 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; 2446 } else { 2447 spa->spa_missing_tvds_allowed = 0; 2448 } 2449 2450 spa->spa_missing_tvds_allowed = 2451 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); 2452 2453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2454 error = vdev_open(spa->spa_root_vdev); 2455 spa_config_exit(spa, SCL_ALL, FTAG); 2456 2457 if (spa->spa_missing_tvds != 0) { 2458 spa_load_note(spa, "vdev tree has %lld missing top-level " 2459 "vdevs.", (u_longlong_t)spa->spa_missing_tvds); 2460 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) { 2461 /* 2462 * Although theoretically we could allow users to open 2463 * incomplete pools in RW mode, we'd need to add a lot 2464 * of extra logic (e.g. adjust pool space to account 2465 * for missing vdevs). 2466 * This limitation also prevents users from accidentally 2467 * opening the pool in RW mode during data recovery and 2468 * damaging it further. 2469 */ 2470 spa_load_note(spa, "pools with missing top-level " 2471 "vdevs can only be opened in read-only mode."); 2472 error = SET_ERROR(ENXIO); 2473 } else { 2474 spa_load_note(spa, "current settings allow for maximum " 2475 "%lld missing top-level vdevs at this stage.", 2476 (u_longlong_t)spa->spa_missing_tvds_allowed); 2477 } 2478 } 2479 if (error != 0) { 2480 spa_load_failed(spa, "unable to open vdev tree [error=%d]", 2481 error); 2482 } 2483 if (spa->spa_missing_tvds != 0 || error != 0) 2484 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); 2485 2486 return (error); 2487 } 2488 2489 /* 2490 * We need to validate the vdev labels against the configuration that 2491 * we have in hand. This function is called twice: first with an untrusted 2492 * config, then with a trusted config. The validation is more strict when the 2493 * config is trusted. 2494 */ 2495 static int 2496 spa_ld_validate_vdevs(spa_t *spa) 2497 { 2498 int error = 0; 2499 vdev_t *rvd = spa->spa_root_vdev; 2500 2501 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2502 error = vdev_validate(rvd); 2503 spa_config_exit(spa, SCL_ALL, FTAG); 2504 2505 if (error != 0) { 2506 spa_load_failed(spa, "vdev_validate failed [error=%d]", error); 2507 return (error); 2508 } 2509 2510 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 2511 spa_load_failed(spa, "cannot open vdev tree after invalidating " 2512 "some vdevs"); 2513 vdev_dbgmsg_print_tree(rvd, 2); 2514 return (SET_ERROR(ENXIO)); 2515 } 2516 2517 return (0); 2518 } 2519 2520 static void 2521 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) 2522 { 2523 spa->spa_state = POOL_STATE_ACTIVE; 2524 spa->spa_ubsync = spa->spa_uberblock; 2525 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2526 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2527 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2528 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2529 spa->spa_claim_max_txg = spa->spa_first_txg; 2530 spa->spa_prev_software_version = ub->ub_software_version; 2531 } 2532 2533 static int 2534 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) 2535 { 2536 vdev_t *rvd = spa->spa_root_vdev; 2537 nvlist_t *label; 2538 uberblock_t *ub = &spa->spa_uberblock; 2539 2540 /* 2541 * If we are opening the checkpointed state of the pool by 2542 * rewinding to it, at this point we will have written the 2543 * checkpointed uberblock to the vdev labels, so searching 2544 * the labels will find the right uberblock. However, if 2545 * we are opening the checkpointed state read-only, we have 2546 * not modified the labels. Therefore, we must ignore the 2547 * labels and continue using the spa_uberblock that was set 2548 * by spa_ld_checkpoint_rewind. 2549 * 2550 * Note that it would be fine to ignore the labels when 2551 * rewinding (opening writeable) as well. However, if we 2552 * crash just after writing the labels, we will end up 2553 * searching the labels. Doing so in the common case means 2554 * that this code path gets exercised normally, rather than 2555 * just in the edge case. 2556 */ 2557 if (ub->ub_checkpoint_txg != 0 && 2558 spa_importing_readonly_checkpoint(spa)) { 2559 spa_ld_select_uberblock_done(spa, ub); 2560 return (0); 2561 } 2562 2563 /* 2564 * Find the best uberblock. 2565 */ 2566 vdev_uberblock_load(rvd, ub, &label); 2567 2568 /* 2569 * If we weren't able to find a single valid uberblock, return failure. 2570 */ 2571 if (ub->ub_txg == 0) { 2572 nvlist_free(label); 2573 spa_load_failed(spa, "no valid uberblock found"); 2574 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2575 } 2576 2577 spa_load_note(spa, "using uberblock with txg=%llu", 2578 (u_longlong_t)ub->ub_txg); 2579 2580 /* 2581 * If the pool has an unsupported version we can't open it. 2582 */ 2583 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2584 nvlist_free(label); 2585 spa_load_failed(spa, "version %llu is not supported", 2586 (u_longlong_t)ub->ub_version); 2587 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2588 } 2589 2590 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2591 nvlist_t *features; 2592 2593 /* 2594 * If we weren't able to find what's necessary for reading the 2595 * MOS in the label, return failure. 2596 */ 2597 if (label == NULL) { 2598 spa_load_failed(spa, "label config unavailable"); 2599 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2600 ENXIO)); 2601 } 2602 2603 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, 2604 &features) != 0) { 2605 nvlist_free(label); 2606 spa_load_failed(spa, "invalid label: '%s' missing", 2607 ZPOOL_CONFIG_FEATURES_FOR_READ); 2608 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2609 ENXIO)); 2610 } 2611 2612 /* 2613 * Update our in-core representation with the definitive values 2614 * from the label. 2615 */ 2616 nvlist_free(spa->spa_label_features); 2617 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2618 } 2619 2620 nvlist_free(label); 2621 2622 /* 2623 * Look through entries in the label nvlist's features_for_read. If 2624 * there is a feature listed there which we don't understand then we 2625 * cannot open a pool. 2626 */ 2627 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2628 nvlist_t *unsup_feat; 2629 2630 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2631 0); 2632 2633 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2634 NULL); nvp != NULL; 2635 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2636 if (!zfeature_is_supported(nvpair_name(nvp))) { 2637 VERIFY(nvlist_add_string(unsup_feat, 2638 nvpair_name(nvp), "") == 0); 2639 } 2640 } 2641 2642 if (!nvlist_empty(unsup_feat)) { 2643 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2644 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2645 nvlist_free(unsup_feat); 2646 spa_load_failed(spa, "some features are unsupported"); 2647 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2648 ENOTSUP)); 2649 } 2650 2651 nvlist_free(unsup_feat); 2652 } 2653 2654 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2655 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2656 spa_try_repair(spa, spa->spa_config); 2657 spa_config_exit(spa, SCL_ALL, FTAG); 2658 nvlist_free(spa->spa_config_splitting); 2659 spa->spa_config_splitting = NULL; 2660 } 2661 2662 /* 2663 * Initialize internal SPA structures. 2664 */ 2665 spa_ld_select_uberblock_done(spa, ub); 2666 2667 return (0); 2668 } 2669 2670 static int 2671 spa_ld_open_rootbp(spa_t *spa) 2672 { 2673 int error = 0; 2674 vdev_t *rvd = spa->spa_root_vdev; 2675 2676 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2677 if (error != 0) { 2678 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " 2679 "[error=%d]", error); 2680 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2681 } 2682 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2683 2684 return (0); 2685 } 2686 2687 static int 2688 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, 2689 boolean_t reloading) 2690 { 2691 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 2692 nvlist_t *nv, *mos_config, *policy; 2693 int error = 0, copy_error; 2694 uint64_t healthy_tvds, healthy_tvds_mos; 2695 uint64_t mos_config_txg; 2696 2697 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) 2698 != 0) 2699 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2700 2701 /* 2702 * If we're assembling a pool from a split, the config provided is 2703 * already trusted so there is nothing to do. 2704 */ 2705 if (type == SPA_IMPORT_ASSEMBLE) 2706 return (0); 2707 2708 healthy_tvds = spa_healthy_core_tvds(spa); 2709 2710 if (load_nvlist(spa, spa->spa_config_object, &mos_config) 2711 != 0) { 2712 spa_load_failed(spa, "unable to retrieve MOS config"); 2713 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2714 } 2715 2716 /* 2717 * If we are doing an open, pool owner wasn't verified yet, thus do 2718 * the verification here. 2719 */ 2720 if (spa->spa_load_state == SPA_LOAD_OPEN) { 2721 error = spa_verify_host(spa, mos_config); 2722 if (error != 0) { 2723 nvlist_free(mos_config); 2724 return (error); 2725 } 2726 } 2727 2728 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); 2729 2730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2731 2732 /* 2733 * Build a new vdev tree from the trusted config 2734 */ 2735 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 2736 2737 /* 2738 * Vdev paths in the MOS may be obsolete. If the untrusted config was 2739 * obtained by scanning /dev/dsk, then it will have the right vdev 2740 * paths. We update the trusted MOS config with this information. 2741 * We first try to copy the paths with vdev_copy_path_strict, which 2742 * succeeds only when both configs have exactly the same vdev tree. 2743 * If that fails, we fall back to a more flexible method that has a 2744 * best effort policy. 2745 */ 2746 copy_error = vdev_copy_path_strict(rvd, mrvd); 2747 if (copy_error != 0 || spa_load_print_vdev_tree) { 2748 spa_load_note(spa, "provided vdev tree:"); 2749 vdev_dbgmsg_print_tree(rvd, 2); 2750 spa_load_note(spa, "MOS vdev tree:"); 2751 vdev_dbgmsg_print_tree(mrvd, 2); 2752 } 2753 if (copy_error != 0) { 2754 spa_load_note(spa, "vdev_copy_path_strict failed, falling " 2755 "back to vdev_copy_path_relaxed"); 2756 vdev_copy_path_relaxed(rvd, mrvd); 2757 } 2758 2759 vdev_close(rvd); 2760 vdev_free(rvd); 2761 spa->spa_root_vdev = mrvd; 2762 rvd = mrvd; 2763 spa_config_exit(spa, SCL_ALL, FTAG); 2764 2765 /* 2766 * We will use spa_config if we decide to reload the spa or if spa_load 2767 * fails and we rewind. We must thus regenerate the config using the 2768 * MOS information with the updated paths. Rewind policy is an import 2769 * setting and is not in the MOS. We copy it over to our new, trusted 2770 * config. 2771 */ 2772 mos_config_txg = fnvlist_lookup_uint64(mos_config, 2773 ZPOOL_CONFIG_POOL_TXG); 2774 nvlist_free(mos_config); 2775 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); 2776 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_REWIND_POLICY, 2777 &policy) == 0) 2778 fnvlist_add_nvlist(mos_config, ZPOOL_REWIND_POLICY, policy); 2779 spa_config_set(spa, mos_config); 2780 spa->spa_config_source = SPA_CONFIG_SRC_MOS; 2781 2782 /* 2783 * Now that we got the config from the MOS, we should be more strict 2784 * in checking blkptrs and can make assumptions about the consistency 2785 * of the vdev tree. spa_trust_config must be set to true before opening 2786 * vdevs in order for them to be writeable. 2787 */ 2788 spa->spa_trust_config = B_TRUE; 2789 2790 /* 2791 * Open and validate the new vdev tree 2792 */ 2793 error = spa_ld_open_vdevs(spa); 2794 if (error != 0) 2795 return (error); 2796 2797 error = spa_ld_validate_vdevs(spa); 2798 if (error != 0) 2799 return (error); 2800 2801 if (copy_error != 0 || spa_load_print_vdev_tree) { 2802 spa_load_note(spa, "final vdev tree:"); 2803 vdev_dbgmsg_print_tree(rvd, 2); 2804 } 2805 2806 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && 2807 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { 2808 /* 2809 * Sanity check to make sure that we are indeed loading the 2810 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds 2811 * in the config provided and they happened to be the only ones 2812 * to have the latest uberblock, we could involuntarily perform 2813 * an extreme rewind. 2814 */ 2815 healthy_tvds_mos = spa_healthy_core_tvds(spa); 2816 if (healthy_tvds_mos - healthy_tvds >= 2817 SPA_SYNC_MIN_VDEVS) { 2818 spa_load_note(spa, "config provided misses too many " 2819 "top-level vdevs compared to MOS (%lld vs %lld). ", 2820 (u_longlong_t)healthy_tvds, 2821 (u_longlong_t)healthy_tvds_mos); 2822 spa_load_note(spa, "vdev tree:"); 2823 vdev_dbgmsg_print_tree(rvd, 2); 2824 if (reloading) { 2825 spa_load_failed(spa, "config was already " 2826 "provided from MOS. Aborting."); 2827 return (spa_vdev_err(rvd, 2828 VDEV_AUX_CORRUPT_DATA, EIO)); 2829 } 2830 spa_load_note(spa, "spa must be reloaded using MOS " 2831 "config"); 2832 return (SET_ERROR(EAGAIN)); 2833 } 2834 } 2835 2836 error = spa_check_for_missing_logs(spa); 2837 if (error != 0) 2838 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2839 2840 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { 2841 spa_load_failed(spa, "uberblock guid sum doesn't match MOS " 2842 "guid sum (%llu != %llu)", 2843 (u_longlong_t)spa->spa_uberblock.ub_guid_sum, 2844 (u_longlong_t)rvd->vdev_guid_sum); 2845 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2846 ENXIO)); 2847 } 2848 2849 return (0); 2850 } 2851 2852 static int 2853 spa_ld_open_indirect_vdev_metadata(spa_t *spa) 2854 { 2855 int error = 0; 2856 vdev_t *rvd = spa->spa_root_vdev; 2857 2858 /* 2859 * Everything that we read before spa_remove_init() must be stored 2860 * on concreted vdevs. Therefore we do this as early as possible. 2861 */ 2862 error = spa_remove_init(spa); 2863 if (error != 0) { 2864 spa_load_failed(spa, "spa_remove_init failed [error=%d]", 2865 error); 2866 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2867 } 2868 2869 /* 2870 * Retrieve information needed to condense indirect vdev mappings. 2871 */ 2872 error = spa_condense_init(spa); 2873 if (error != 0) { 2874 spa_load_failed(spa, "spa_condense_init failed [error=%d]", 2875 error); 2876 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 2877 } 2878 2879 return (0); 2880 } 2881 2882 static int 2883 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) 2884 { 2885 int error = 0; 2886 vdev_t *rvd = spa->spa_root_vdev; 2887 2888 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2889 boolean_t missing_feat_read = B_FALSE; 2890 nvlist_t *unsup_feat, *enabled_feat; 2891 2892 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2893 &spa->spa_feat_for_read_obj, B_TRUE) != 0) { 2894 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2895 } 2896 2897 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2898 &spa->spa_feat_for_write_obj, B_TRUE) != 0) { 2899 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2900 } 2901 2902 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2903 &spa->spa_feat_desc_obj, B_TRUE) != 0) { 2904 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2905 } 2906 2907 enabled_feat = fnvlist_alloc(); 2908 unsup_feat = fnvlist_alloc(); 2909 2910 if (!spa_features_check(spa, B_FALSE, 2911 unsup_feat, enabled_feat)) 2912 missing_feat_read = B_TRUE; 2913 2914 if (spa_writeable(spa) || 2915 spa->spa_load_state == SPA_LOAD_TRYIMPORT) { 2916 if (!spa_features_check(spa, B_TRUE, 2917 unsup_feat, enabled_feat)) { 2918 *missing_feat_writep = B_TRUE; 2919 } 2920 } 2921 2922 fnvlist_add_nvlist(spa->spa_load_info, 2923 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2924 2925 if (!nvlist_empty(unsup_feat)) { 2926 fnvlist_add_nvlist(spa->spa_load_info, 2927 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2928 } 2929 2930 fnvlist_free(enabled_feat); 2931 fnvlist_free(unsup_feat); 2932 2933 if (!missing_feat_read) { 2934 fnvlist_add_boolean(spa->spa_load_info, 2935 ZPOOL_CONFIG_CAN_RDONLY); 2936 } 2937 2938 /* 2939 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2940 * twofold: to determine whether the pool is available for 2941 * import in read-write mode and (if it is not) whether the 2942 * pool is available for import in read-only mode. If the pool 2943 * is available for import in read-write mode, it is displayed 2944 * as available in userland; if it is not available for import 2945 * in read-only mode, it is displayed as unavailable in 2946 * userland. If the pool is available for import in read-only 2947 * mode but not read-write mode, it is displayed as unavailable 2948 * in userland with a special note that the pool is actually 2949 * available for open in read-only mode. 2950 * 2951 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2952 * missing a feature for write, we must first determine whether 2953 * the pool can be opened read-only before returning to 2954 * userland in order to know whether to display the 2955 * abovementioned note. 2956 */ 2957 if (missing_feat_read || (*missing_feat_writep && 2958 spa_writeable(spa))) { 2959 spa_load_failed(spa, "pool uses unsupported features"); 2960 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2961 ENOTSUP)); 2962 } 2963 2964 /* 2965 * Load refcounts for ZFS features from disk into an in-memory 2966 * cache during SPA initialization. 2967 */ 2968 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2969 uint64_t refcount; 2970 2971 error = feature_get_refcount_from_disk(spa, 2972 &spa_feature_table[i], &refcount); 2973 if (error == 0) { 2974 spa->spa_feat_refcount_cache[i] = refcount; 2975 } else if (error == ENOTSUP) { 2976 spa->spa_feat_refcount_cache[i] = 2977 SPA_FEATURE_DISABLED; 2978 } else { 2979 spa_load_failed(spa, "error getting refcount " 2980 "for feature %s [error=%d]", 2981 spa_feature_table[i].fi_guid, error); 2982 return (spa_vdev_err(rvd, 2983 VDEV_AUX_CORRUPT_DATA, EIO)); 2984 } 2985 } 2986 } 2987 2988 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2989 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2990 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) 2991 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2992 } 2993 2994 return (0); 2995 } 2996 2997 static int 2998 spa_ld_load_special_directories(spa_t *spa) 2999 { 3000 int error = 0; 3001 vdev_t *rvd = spa->spa_root_vdev; 3002 3003 spa->spa_is_initializing = B_TRUE; 3004 error = dsl_pool_open(spa->spa_dsl_pool); 3005 spa->spa_is_initializing = B_FALSE; 3006 if (error != 0) { 3007 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); 3008 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3009 } 3010 3011 return (0); 3012 } 3013 3014 static int 3015 spa_ld_get_props(spa_t *spa) 3016 { 3017 int error = 0; 3018 uint64_t obj; 3019 vdev_t *rvd = spa->spa_root_vdev; 3020 3021 /* Grab the secret checksum salt from the MOS. */ 3022 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3023 DMU_POOL_CHECKSUM_SALT, 1, 3024 sizeof (spa->spa_cksum_salt.zcs_bytes), 3025 spa->spa_cksum_salt.zcs_bytes); 3026 if (error == ENOENT) { 3027 /* Generate a new salt for subsequent use */ 3028 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3029 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3030 } else if (error != 0) { 3031 spa_load_failed(spa, "unable to retrieve checksum salt from " 3032 "MOS [error=%d]", error); 3033 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3034 } 3035 3036 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) 3037 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3038 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 3039 if (error != 0) { 3040 spa_load_failed(spa, "error opening deferred-frees bpobj " 3041 "[error=%d]", error); 3042 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3043 } 3044 3045 /* 3046 * Load the bit that tells us to use the new accounting function 3047 * (raid-z deflation). If we have an older pool, this will not 3048 * be present. 3049 */ 3050 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); 3051 if (error != 0 && error != ENOENT) 3052 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3053 3054 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 3055 &spa->spa_creation_version, B_FALSE); 3056 if (error != 0 && error != ENOENT) 3057 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3058 3059 /* 3060 * Load the persistent error log. If we have an older pool, this will 3061 * not be present. 3062 */ 3063 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, 3064 B_FALSE); 3065 if (error != 0 && error != ENOENT) 3066 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3067 3068 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 3069 &spa->spa_errlog_scrub, B_FALSE); 3070 if (error != 0 && error != ENOENT) 3071 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3072 3073 /* 3074 * Load the history object. If we have an older pool, this 3075 * will not be present. 3076 */ 3077 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); 3078 if (error != 0 && error != ENOENT) 3079 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3080 3081 /* 3082 * Load the per-vdev ZAP map. If we have an older pool, this will not 3083 * be present; in this case, defer its creation to a later time to 3084 * avoid dirtying the MOS this early / out of sync context. See 3085 * spa_sync_config_object. 3086 */ 3087 3088 /* The sentinel is only available in the MOS config. */ 3089 nvlist_t *mos_config; 3090 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { 3091 spa_load_failed(spa, "unable to retrieve MOS config"); 3092 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3093 } 3094 3095 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 3096 &spa->spa_all_vdev_zaps, B_FALSE); 3097 3098 if (error == ENOENT) { 3099 VERIFY(!nvlist_exists(mos_config, 3100 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 3101 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 3102 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3103 } else if (error != 0) { 3104 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3105 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 3106 /* 3107 * An older version of ZFS overwrote the sentinel value, so 3108 * we have orphaned per-vdev ZAPs in the MOS. Defer their 3109 * destruction to later; see spa_sync_config_object. 3110 */ 3111 spa->spa_avz_action = AVZ_ACTION_DESTROY; 3112 /* 3113 * We're assuming that no vdevs have had their ZAPs created 3114 * before this. Better be sure of it. 3115 */ 3116 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3117 } 3118 nvlist_free(mos_config); 3119 3120 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3121 3122 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, 3123 B_FALSE); 3124 if (error && error != ENOENT) 3125 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3126 3127 if (error == 0) { 3128 uint64_t autoreplace; 3129 3130 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 3131 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 3132 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 3133 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 3134 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 3135 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 3136 &spa->spa_dedup_ditto); 3137 3138 spa->spa_autoreplace = (autoreplace != 0); 3139 } 3140 3141 /* 3142 * If we are importing a pool with missing top-level vdevs, 3143 * we enforce that the pool doesn't panic or get suspended on 3144 * error since the likelihood of missing data is extremely high. 3145 */ 3146 if (spa->spa_missing_tvds > 0 && 3147 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && 3148 spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3149 spa_load_note(spa, "forcing failmode to 'continue' " 3150 "as some top level vdevs are missing"); 3151 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; 3152 } 3153 3154 return (0); 3155 } 3156 3157 static int 3158 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) 3159 { 3160 int error = 0; 3161 vdev_t *rvd = spa->spa_root_vdev; 3162 3163 /* 3164 * If we're assembling the pool from the split-off vdevs of 3165 * an existing pool, we don't want to attach the spares & cache 3166 * devices. 3167 */ 3168 3169 /* 3170 * Load any hot spares for this pool. 3171 */ 3172 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, 3173 B_FALSE); 3174 if (error != 0 && error != ENOENT) 3175 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3176 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3177 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 3178 if (load_nvlist(spa, spa->spa_spares.sav_object, 3179 &spa->spa_spares.sav_config) != 0) { 3180 spa_load_failed(spa, "error loading spares nvlist"); 3181 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3182 } 3183 3184 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3185 spa_load_spares(spa); 3186 spa_config_exit(spa, SCL_ALL, FTAG); 3187 } else if (error == 0) { 3188 spa->spa_spares.sav_sync = B_TRUE; 3189 } 3190 3191 /* 3192 * Load any level 2 ARC devices for this pool. 3193 */ 3194 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 3195 &spa->spa_l2cache.sav_object, B_FALSE); 3196 if (error != 0 && error != ENOENT) 3197 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3198 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3199 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 3200 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 3201 &spa->spa_l2cache.sav_config) != 0) { 3202 spa_load_failed(spa, "error loading l2cache nvlist"); 3203 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3204 } 3205 3206 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3207 spa_load_l2cache(spa); 3208 spa_config_exit(spa, SCL_ALL, FTAG); 3209 } else if (error == 0) { 3210 spa->spa_l2cache.sav_sync = B_TRUE; 3211 } 3212 3213 return (0); 3214 } 3215 3216 static int 3217 spa_ld_load_vdev_metadata(spa_t *spa) 3218 { 3219 int error = 0; 3220 vdev_t *rvd = spa->spa_root_vdev; 3221 3222 /* 3223 * If the 'autoreplace' property is set, then post a resource notifying 3224 * the ZFS DE that it should not issue any faults for unopenable 3225 * devices. We also iterate over the vdevs, and post a sysevent for any 3226 * unopenable vdevs so that the normal autoreplace handler can take 3227 * over. 3228 */ 3229 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3230 spa_check_removed(spa->spa_root_vdev); 3231 /* 3232 * For the import case, this is done in spa_import(), because 3233 * at this point we're using the spare definitions from 3234 * the MOS config, not necessarily from the userland config. 3235 */ 3236 if (spa->spa_load_state != SPA_LOAD_IMPORT) { 3237 spa_aux_check_removed(&spa->spa_spares); 3238 spa_aux_check_removed(&spa->spa_l2cache); 3239 } 3240 } 3241 3242 /* 3243 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. 3244 */ 3245 error = vdev_load(rvd); 3246 if (error != 0) { 3247 spa_load_failed(spa, "vdev_load failed [error=%d]", error); 3248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 3249 } 3250 3251 /* 3252 * Propagate the leaf DTLs we just loaded all the way up the vdev tree. 3253 */ 3254 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3255 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 3256 spa_config_exit(spa, SCL_ALL, FTAG); 3257 3258 return (0); 3259 } 3260 3261 static int 3262 spa_ld_load_dedup_tables(spa_t *spa) 3263 { 3264 int error = 0; 3265 vdev_t *rvd = spa->spa_root_vdev; 3266 3267 error = ddt_load(spa); 3268 if (error != 0) { 3269 spa_load_failed(spa, "ddt_load failed [error=%d]", error); 3270 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3271 } 3272 3273 return (0); 3274 } 3275 3276 static int 3277 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport) 3278 { 3279 vdev_t *rvd = spa->spa_root_vdev; 3280 3281 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { 3282 boolean_t missing = spa_check_logs(spa); 3283 if (missing) { 3284 if (spa->spa_missing_tvds != 0) { 3285 spa_load_note(spa, "spa_check_logs failed " 3286 "so dropping the logs"); 3287 } else { 3288 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 3289 spa_load_failed(spa, "spa_check_logs failed"); 3290 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, 3291 ENXIO)); 3292 } 3293 } 3294 } 3295 3296 return (0); 3297 } 3298 3299 static int 3300 spa_ld_verify_pool_data(spa_t *spa) 3301 { 3302 int error = 0; 3303 vdev_t *rvd = spa->spa_root_vdev; 3304 3305 /* 3306 * We've successfully opened the pool, verify that we're ready 3307 * to start pushing transactions. 3308 */ 3309 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3310 error = spa_load_verify(spa); 3311 if (error != 0) { 3312 spa_load_failed(spa, "spa_load_verify failed " 3313 "[error=%d]", error); 3314 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 3315 error)); 3316 } 3317 } 3318 3319 return (0); 3320 } 3321 3322 static void 3323 spa_ld_claim_log_blocks(spa_t *spa) 3324 { 3325 dmu_tx_t *tx; 3326 dsl_pool_t *dp = spa_get_dsl(spa); 3327 3328 /* 3329 * Claim log blocks that haven't been committed yet. 3330 * This must all happen in a single txg. 3331 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 3332 * invoked from zil_claim_log_block()'s i/o done callback. 3333 * Price of rollback is that we abandon the log. 3334 */ 3335 spa->spa_claiming = B_TRUE; 3336 3337 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 3338 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 3339 zil_claim, tx, DS_FIND_CHILDREN); 3340 dmu_tx_commit(tx); 3341 3342 spa->spa_claiming = B_FALSE; 3343 3344 spa_set_log_state(spa, SPA_LOG_GOOD); 3345 } 3346 3347 static void 3348 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, 3349 boolean_t update_config_cache) 3350 { 3351 vdev_t *rvd = spa->spa_root_vdev; 3352 int need_update = B_FALSE; 3353 3354 /* 3355 * If the config cache is stale, or we have uninitialized 3356 * metaslabs (see spa_vdev_add()), then update the config. 3357 * 3358 * If this is a verbatim import, trust the current 3359 * in-core spa_config and update the disk labels. 3360 */ 3361 if (update_config_cache || config_cache_txg != spa->spa_config_txg || 3362 spa->spa_load_state == SPA_LOAD_IMPORT || 3363 spa->spa_load_state == SPA_LOAD_RECOVER || 3364 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 3365 need_update = B_TRUE; 3366 3367 for (int c = 0; c < rvd->vdev_children; c++) 3368 if (rvd->vdev_child[c]->vdev_ms_array == 0) 3369 need_update = B_TRUE; 3370 3371 /* 3372 * Update the config cache asychronously in case we're the 3373 * root pool, in which case the config cache isn't writable yet. 3374 */ 3375 if (need_update) 3376 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3377 } 3378 3379 static void 3380 spa_ld_prepare_for_reload(spa_t *spa) 3381 { 3382 int mode = spa->spa_mode; 3383 int async_suspended = spa->spa_async_suspended; 3384 3385 spa_unload(spa); 3386 spa_deactivate(spa); 3387 spa_activate(spa, mode); 3388 3389 /* 3390 * We save the value of spa_async_suspended as it gets reset to 0 by 3391 * spa_unload(). We want to restore it back to the original value before 3392 * returning as we might be calling spa_async_resume() later. 3393 */ 3394 spa->spa_async_suspended = async_suspended; 3395 } 3396 3397 static int 3398 spa_ld_read_checkpoint_txg(spa_t *spa) 3399 { 3400 uberblock_t checkpoint; 3401 int error = 0; 3402 3403 ASSERT0(spa->spa_checkpoint_txg); 3404 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3405 3406 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3407 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3408 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3409 3410 if (error == ENOENT) 3411 return (0); 3412 3413 if (error != 0) 3414 return (error); 3415 3416 ASSERT3U(checkpoint.ub_txg, !=, 0); 3417 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); 3418 ASSERT3U(checkpoint.ub_timestamp, !=, 0); 3419 spa->spa_checkpoint_txg = checkpoint.ub_txg; 3420 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; 3421 3422 return (0); 3423 } 3424 3425 static int 3426 spa_ld_mos_init(spa_t *spa, spa_import_type_t type) 3427 { 3428 int error = 0; 3429 3430 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3431 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3432 3433 /* 3434 * Never trust the config that is provided unless we are assembling 3435 * a pool following a split. 3436 * This means don't trust blkptrs and the vdev tree in general. This 3437 * also effectively puts the spa in read-only mode since 3438 * spa_writeable() checks for spa_trust_config to be true. 3439 * We will later load a trusted config from the MOS. 3440 */ 3441 if (type != SPA_IMPORT_ASSEMBLE) 3442 spa->spa_trust_config = B_FALSE; 3443 3444 /* 3445 * Parse the config provided to create a vdev tree. 3446 */ 3447 error = spa_ld_parse_config(spa, type); 3448 if (error != 0) 3449 return (error); 3450 3451 /* 3452 * Now that we have the vdev tree, try to open each vdev. This involves 3453 * opening the underlying physical device, retrieving its geometry and 3454 * probing the vdev with a dummy I/O. The state of each vdev will be set 3455 * based on the success of those operations. After this we'll be ready 3456 * to read from the vdevs. 3457 */ 3458 error = spa_ld_open_vdevs(spa); 3459 if (error != 0) 3460 return (error); 3461 3462 /* 3463 * Read the label of each vdev and make sure that the GUIDs stored 3464 * there match the GUIDs in the config provided. 3465 * If we're assembling a new pool that's been split off from an 3466 * existing pool, the labels haven't yet been updated so we skip 3467 * validation for now. 3468 */ 3469 if (type != SPA_IMPORT_ASSEMBLE) { 3470 error = spa_ld_validate_vdevs(spa); 3471 if (error != 0) 3472 return (error); 3473 } 3474 3475 /* 3476 * Read all vdev labels to find the best uberblock (i.e. latest, 3477 * unless spa_load_max_txg is set) and store it in spa_uberblock. We 3478 * get the list of features required to read blkptrs in the MOS from 3479 * the vdev label with the best uberblock and verify that our version 3480 * of zfs supports them all. 3481 */ 3482 error = spa_ld_select_uberblock(spa, type); 3483 if (error != 0) 3484 return (error); 3485 3486 /* 3487 * Pass that uberblock to the dsl_pool layer which will open the root 3488 * blkptr. This blkptr points to the latest version of the MOS and will 3489 * allow us to read its contents. 3490 */ 3491 error = spa_ld_open_rootbp(spa); 3492 if (error != 0) 3493 return (error); 3494 3495 return (0); 3496 } 3497 3498 static int 3499 spa_ld_checkpoint_rewind(spa_t *spa) 3500 { 3501 uberblock_t checkpoint; 3502 int error = 0; 3503 3504 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3505 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3506 3507 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3508 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3509 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3510 3511 if (error != 0) { 3512 spa_load_failed(spa, "unable to retrieve checkpointed " 3513 "uberblock from the MOS config [error=%d]", error); 3514 3515 if (error == ENOENT) 3516 error = ZFS_ERR_NO_CHECKPOINT; 3517 3518 return (error); 3519 } 3520 3521 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); 3522 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); 3523 3524 /* 3525 * We need to update the txg and timestamp of the checkpointed 3526 * uberblock to be higher than the latest one. This ensures that 3527 * the checkpointed uberblock is selected if we were to close and 3528 * reopen the pool right after we've written it in the vdev labels. 3529 * (also see block comment in vdev_uberblock_compare) 3530 */ 3531 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; 3532 checkpoint.ub_timestamp = gethrestime_sec(); 3533 3534 /* 3535 * Set current uberblock to be the checkpointed uberblock. 3536 */ 3537 spa->spa_uberblock = checkpoint; 3538 3539 /* 3540 * If we are doing a normal rewind, then the pool is open for 3541 * writing and we sync the "updated" checkpointed uberblock to 3542 * disk. Once this is done, we've basically rewound the whole 3543 * pool and there is no way back. 3544 * 3545 * There are cases when we don't want to attempt and sync the 3546 * checkpointed uberblock to disk because we are opening a 3547 * pool as read-only. Specifically, verifying the checkpointed 3548 * state with zdb, and importing the checkpointed state to get 3549 * a "preview" of its content. 3550 */ 3551 if (spa_writeable(spa)) { 3552 vdev_t *rvd = spa->spa_root_vdev; 3553 3554 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3555 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 3556 int svdcount = 0; 3557 int children = rvd->vdev_children; 3558 int c0 = spa_get_random(children); 3559 3560 for (int c = 0; c < children; c++) { 3561 vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; 3562 3563 /* Stop when revisiting the first vdev */ 3564 if (c > 0 && svd[0] == vd) 3565 break; 3566 3567 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 3568 !vdev_is_concrete(vd)) 3569 continue; 3570 3571 svd[svdcount++] = vd; 3572 if (svdcount == SPA_SYNC_MIN_VDEVS) 3573 break; 3574 } 3575 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); 3576 if (error == 0) 3577 spa->spa_last_synced_guid = rvd->vdev_guid; 3578 spa_config_exit(spa, SCL_ALL, FTAG); 3579 3580 if (error != 0) { 3581 spa_load_failed(spa, "failed to write checkpointed " 3582 "uberblock to the vdev labels [error=%d]", error); 3583 return (error); 3584 } 3585 } 3586 3587 return (0); 3588 } 3589 3590 static int 3591 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, 3592 boolean_t *update_config_cache) 3593 { 3594 int error; 3595 3596 /* 3597 * Parse the config for pool, open and validate vdevs, 3598 * select an uberblock, and use that uberblock to open 3599 * the MOS. 3600 */ 3601 error = spa_ld_mos_init(spa, type); 3602 if (error != 0) 3603 return (error); 3604 3605 /* 3606 * Retrieve the trusted config stored in the MOS and use it to create 3607 * a new, exact version of the vdev tree, then reopen all vdevs. 3608 */ 3609 error = spa_ld_trusted_config(spa, type, B_FALSE); 3610 if (error == EAGAIN) { 3611 if (update_config_cache != NULL) 3612 *update_config_cache = B_TRUE; 3613 3614 /* 3615 * Redo the loading process with the trusted config if it is 3616 * too different from the untrusted config. 3617 */ 3618 spa_ld_prepare_for_reload(spa); 3619 spa_load_note(spa, "RELOADING"); 3620 error = spa_ld_mos_init(spa, type); 3621 if (error != 0) 3622 return (error); 3623 3624 error = spa_ld_trusted_config(spa, type, B_TRUE); 3625 if (error != 0) 3626 return (error); 3627 3628 } else if (error != 0) { 3629 return (error); 3630 } 3631 3632 return (0); 3633 } 3634 3635 /* 3636 * Load an existing storage pool, using the config provided. This config 3637 * describes which vdevs are part of the pool and is later validated against 3638 * partial configs present in each vdev's label and an entire copy of the 3639 * config stored in the MOS. 3640 */ 3641 static int 3642 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport) 3643 { 3644 int error = 0; 3645 boolean_t missing_feat_write = B_FALSE; 3646 boolean_t checkpoint_rewind = 3647 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3648 boolean_t update_config_cache = B_FALSE; 3649 3650 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3651 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3652 3653 spa_load_note(spa, "LOADING"); 3654 3655 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); 3656 if (error != 0) 3657 return (error); 3658 3659 /* 3660 * If we are rewinding to the checkpoint then we need to repeat 3661 * everything we've done so far in this function but this time 3662 * selecting the checkpointed uberblock and using that to open 3663 * the MOS. 3664 */ 3665 if (checkpoint_rewind) { 3666 /* 3667 * If we are rewinding to the checkpoint update config cache 3668 * anyway. 3669 */ 3670 update_config_cache = B_TRUE; 3671 3672 /* 3673 * Extract the checkpointed uberblock from the current MOS 3674 * and use this as the pool's uberblock from now on. If the 3675 * pool is imported as writeable we also write the checkpoint 3676 * uberblock to the labels, making the rewind permanent. 3677 */ 3678 error = spa_ld_checkpoint_rewind(spa); 3679 if (error != 0) 3680 return (error); 3681 3682 /* 3683 * Redo the loading process process again with the 3684 * checkpointed uberblock. 3685 */ 3686 spa_ld_prepare_for_reload(spa); 3687 spa_load_note(spa, "LOADING checkpointed uberblock"); 3688 error = spa_ld_mos_with_trusted_config(spa, type, NULL); 3689 if (error != 0) 3690 return (error); 3691 } 3692 3693 /* 3694 * Retrieve the checkpoint txg if the pool has a checkpoint. 3695 */ 3696 error = spa_ld_read_checkpoint_txg(spa); 3697 if (error != 0) 3698 return (error); 3699 3700 /* 3701 * Retrieve the mapping of indirect vdevs. Those vdevs were removed 3702 * from the pool and their contents were re-mapped to other vdevs. Note 3703 * that everything that we read before this step must have been 3704 * rewritten on concrete vdevs after the last device removal was 3705 * initiated. Otherwise we could be reading from indirect vdevs before 3706 * we have loaded their mappings. 3707 */ 3708 error = spa_ld_open_indirect_vdev_metadata(spa); 3709 if (error != 0) 3710 return (error); 3711 3712 /* 3713 * Retrieve the full list of active features from the MOS and check if 3714 * they are all supported. 3715 */ 3716 error = spa_ld_check_features(spa, &missing_feat_write); 3717 if (error != 0) 3718 return (error); 3719 3720 /* 3721 * Load several special directories from the MOS needed by the dsl_pool 3722 * layer. 3723 */ 3724 error = spa_ld_load_special_directories(spa); 3725 if (error != 0) 3726 return (error); 3727 3728 /* 3729 * Retrieve pool properties from the MOS. 3730 */ 3731 error = spa_ld_get_props(spa); 3732 if (error != 0) 3733 return (error); 3734 3735 /* 3736 * Retrieve the list of auxiliary devices - cache devices and spares - 3737 * and open them. 3738 */ 3739 error = spa_ld_open_aux_vdevs(spa, type); 3740 if (error != 0) 3741 return (error); 3742 3743 /* 3744 * Load the metadata for all vdevs. Also check if unopenable devices 3745 * should be autoreplaced. 3746 */ 3747 error = spa_ld_load_vdev_metadata(spa); 3748 if (error != 0) 3749 return (error); 3750 3751 error = spa_ld_load_dedup_tables(spa); 3752 if (error != 0) 3753 return (error); 3754 3755 /* 3756 * Verify the logs now to make sure we don't have any unexpected errors 3757 * when we claim log blocks later. 3758 */ 3759 error = spa_ld_verify_logs(spa, type, ereport); 3760 if (error != 0) 3761 return (error); 3762 3763 if (missing_feat_write) { 3764 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); 3765 3766 /* 3767 * At this point, we know that we can open the pool in 3768 * read-only mode but not read-write mode. We now have enough 3769 * information and can return to userland. 3770 */ 3771 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, 3772 ENOTSUP)); 3773 } 3774 3775 /* 3776 * Traverse the last txgs to make sure the pool was left off in a safe 3777 * state. When performing an extreme rewind, we verify the whole pool, 3778 * which can take a very long time. 3779 */ 3780 error = spa_ld_verify_pool_data(spa); 3781 if (error != 0) 3782 return (error); 3783 3784 /* 3785 * Calculate the deflated space for the pool. This must be done before 3786 * we write anything to the pool because we'd need to update the space 3787 * accounting using the deflated sizes. 3788 */ 3789 spa_update_dspace(spa); 3790 3791 /* 3792 * We have now retrieved all the information we needed to open the 3793 * pool. If we are importing the pool in read-write mode, a few 3794 * additional steps must be performed to finish the import. 3795 */ 3796 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || 3797 spa->spa_load_max_txg == UINT64_MAX)) { 3798 uint64_t config_cache_txg = spa->spa_config_txg; 3799 3800 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); 3801 3802 /* 3803 * In case of a checkpoint rewind, log the original txg 3804 * of the checkpointed uberblock. 3805 */ 3806 if (checkpoint_rewind) { 3807 spa_history_log_internal(spa, "checkpoint rewind", 3808 NULL, "rewound state to txg=%llu", 3809 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); 3810 } 3811 3812 /* 3813 * Traverse the ZIL and claim all blocks. 3814 */ 3815 spa_ld_claim_log_blocks(spa); 3816 3817 /* 3818 * Kick-off the syncing thread. 3819 */ 3820 spa->spa_sync_on = B_TRUE; 3821 txg_sync_start(spa->spa_dsl_pool); 3822 3823 /* 3824 * Wait for all claims to sync. We sync up to the highest 3825 * claimed log block birth time so that claimed log blocks 3826 * don't appear to be from the future. spa_claim_max_txg 3827 * will have been set for us by ZIL traversal operations 3828 * performed above. 3829 */ 3830 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 3831 3832 /* 3833 * Check if we need to request an update of the config. On the 3834 * next sync, we would update the config stored in vdev labels 3835 * and the cachefile (by default /etc/zfs/zpool.cache). 3836 */ 3837 spa_ld_check_for_config_update(spa, config_cache_txg, 3838 update_config_cache); 3839 3840 /* 3841 * Check all DTLs to see if anything needs resilvering. 3842 */ 3843 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 3844 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) 3845 spa_async_request(spa, SPA_ASYNC_RESILVER); 3846 3847 /* 3848 * Log the fact that we booted up (so that we can detect if 3849 * we rebooted in the middle of an operation). 3850 */ 3851 spa_history_log_version(spa, "open"); 3852 3853 /* 3854 * Delete any inconsistent datasets. 3855 */ 3856 (void) dmu_objset_find(spa_name(spa), 3857 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3858 3859 /* 3860 * Clean up any stale temporary dataset userrefs. 3861 */ 3862 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3863 3864 spa_restart_removal(spa); 3865 3866 spa_spawn_aux_threads(spa); 3867 } 3868 3869 spa_load_note(spa, "LOADED"); 3870 3871 return (0); 3872 } 3873 3874 static int 3875 spa_load_retry(spa_t *spa, spa_load_state_t state) 3876 { 3877 int mode = spa->spa_mode; 3878 3879 spa_unload(spa); 3880 spa_deactivate(spa); 3881 3882 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3883 3884 spa_activate(spa, mode); 3885 spa_async_suspend(spa); 3886 3887 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", 3888 (u_longlong_t)spa->spa_load_max_txg); 3889 3890 return (spa_load(spa, state, SPA_IMPORT_EXISTING)); 3891 } 3892 3893 /* 3894 * If spa_load() fails this function will try loading prior txg's. If 3895 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3896 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3897 * function will not rewind the pool and will return the same error as 3898 * spa_load(). 3899 */ 3900 static int 3901 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, 3902 int rewind_flags) 3903 { 3904 nvlist_t *loadinfo = NULL; 3905 nvlist_t *config = NULL; 3906 int load_error, rewind_error; 3907 uint64_t safe_rewind_txg; 3908 uint64_t min_txg; 3909 3910 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 3911 spa->spa_load_max_txg = spa->spa_load_txg; 3912 spa_set_log_state(spa, SPA_LOG_CLEAR); 3913 } else { 3914 spa->spa_load_max_txg = max_request; 3915 if (max_request != UINT64_MAX) 3916 spa->spa_extreme_rewind = B_TRUE; 3917 } 3918 3919 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); 3920 if (load_error == 0) 3921 return (0); 3922 if (load_error == ZFS_ERR_NO_CHECKPOINT) { 3923 /* 3924 * When attempting checkpoint-rewind on a pool with no 3925 * checkpoint, we should not attempt to load uberblocks 3926 * from previous txgs when spa_load fails. 3927 */ 3928 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3929 return (load_error); 3930 } 3931 3932 if (spa->spa_root_vdev != NULL) 3933 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3934 3935 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 3936 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 3937 3938 if (rewind_flags & ZPOOL_NEVER_REWIND) { 3939 nvlist_free(config); 3940 return (load_error); 3941 } 3942 3943 if (state == SPA_LOAD_RECOVER) { 3944 /* Price of rolling back is discarding txgs, including log */ 3945 spa_set_log_state(spa, SPA_LOG_CLEAR); 3946 } else { 3947 /* 3948 * If we aren't rolling back save the load info from our first 3949 * import attempt so that we can restore it after attempting 3950 * to rewind. 3951 */ 3952 loadinfo = spa->spa_load_info; 3953 spa->spa_load_info = fnvlist_alloc(); 3954 } 3955 3956 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 3957 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 3958 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 3959 TXG_INITIAL : safe_rewind_txg; 3960 3961 /* 3962 * Continue as long as we're finding errors, we're still within 3963 * the acceptable rewind range, and we're still finding uberblocks 3964 */ 3965 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 3966 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 3967 if (spa->spa_load_max_txg < safe_rewind_txg) 3968 spa->spa_extreme_rewind = B_TRUE; 3969 rewind_error = spa_load_retry(spa, state); 3970 } 3971 3972 spa->spa_extreme_rewind = B_FALSE; 3973 spa->spa_load_max_txg = UINT64_MAX; 3974 3975 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3976 spa_config_set(spa, config); 3977 else 3978 nvlist_free(config); 3979 3980 if (state == SPA_LOAD_RECOVER) { 3981 ASSERT3P(loadinfo, ==, NULL); 3982 return (rewind_error); 3983 } else { 3984 /* Store the rewind info as part of the initial load info */ 3985 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3986 spa->spa_load_info); 3987 3988 /* Restore the initial load info */ 3989 fnvlist_free(spa->spa_load_info); 3990 spa->spa_load_info = loadinfo; 3991 3992 return (load_error); 3993 } 3994 } 3995 3996 /* 3997 * Pool Open/Import 3998 * 3999 * The import case is identical to an open except that the configuration is sent 4000 * down from userland, instead of grabbed from the configuration cache. For the 4001 * case of an open, the pool configuration will exist in the 4002 * POOL_STATE_UNINITIALIZED state. 4003 * 4004 * The stats information (gen/count/ustats) is used to gather vdev statistics at 4005 * the same time open the pool, without having to keep around the spa_t in some 4006 * ambiguous state. 4007 */ 4008 static int 4009 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 4010 nvlist_t **config) 4011 { 4012 spa_t *spa; 4013 spa_load_state_t state = SPA_LOAD_OPEN; 4014 int error; 4015 int locked = B_FALSE; 4016 4017 *spapp = NULL; 4018 4019 /* 4020 * As disgusting as this is, we need to support recursive calls to this 4021 * function because dsl_dir_open() is called during spa_load(), and ends 4022 * up calling spa_open() again. The real fix is to figure out how to 4023 * avoid dsl_dir_open() calling this in the first place. 4024 */ 4025 if (mutex_owner(&spa_namespace_lock) != curthread) { 4026 mutex_enter(&spa_namespace_lock); 4027 locked = B_TRUE; 4028 } 4029 4030 if ((spa = spa_lookup(pool)) == NULL) { 4031 if (locked) 4032 mutex_exit(&spa_namespace_lock); 4033 return (SET_ERROR(ENOENT)); 4034 } 4035 4036 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 4037 zpool_rewind_policy_t policy; 4038 4039 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 4040 &policy); 4041 if (policy.zrp_request & ZPOOL_DO_REWIND) 4042 state = SPA_LOAD_RECOVER; 4043 4044 spa_activate(spa, spa_mode_global); 4045 4046 if (state != SPA_LOAD_RECOVER) 4047 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4048 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 4049 4050 zfs_dbgmsg("spa_open_common: opening %s", pool); 4051 error = spa_load_best(spa, state, policy.zrp_txg, 4052 policy.zrp_request); 4053 4054 if (error == EBADF) { 4055 /* 4056 * If vdev_validate() returns failure (indicated by 4057 * EBADF), it indicates that one of the vdevs indicates 4058 * that the pool has been exported or destroyed. If 4059 * this is the case, the config cache is out of sync and 4060 * we should remove the pool from the namespace. 4061 */ 4062 spa_unload(spa); 4063 spa_deactivate(spa); 4064 spa_write_cachefile(spa, B_TRUE, B_TRUE); 4065 spa_remove(spa); 4066 if (locked) 4067 mutex_exit(&spa_namespace_lock); 4068 return (SET_ERROR(ENOENT)); 4069 } 4070 4071 if (error) { 4072 /* 4073 * We can't open the pool, but we still have useful 4074 * information: the state of each vdev after the 4075 * attempted vdev_open(). Return this to the user. 4076 */ 4077 if (config != NULL && spa->spa_config) { 4078 VERIFY(nvlist_dup(spa->spa_config, config, 4079 KM_SLEEP) == 0); 4080 VERIFY(nvlist_add_nvlist(*config, 4081 ZPOOL_CONFIG_LOAD_INFO, 4082 spa->spa_load_info) == 0); 4083 } 4084 spa_unload(spa); 4085 spa_deactivate(spa); 4086 spa->spa_last_open_failed = error; 4087 if (locked) 4088 mutex_exit(&spa_namespace_lock); 4089 *spapp = NULL; 4090 return (error); 4091 } 4092 } 4093 4094 spa_open_ref(spa, tag); 4095 4096 if (config != NULL) 4097 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4098 4099 /* 4100 * If we've recovered the pool, pass back any information we 4101 * gathered while doing the load. 4102 */ 4103 if (state == SPA_LOAD_RECOVER) { 4104 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 4105 spa->spa_load_info) == 0); 4106 } 4107 4108 if (locked) { 4109 spa->spa_last_open_failed = 0; 4110 spa->spa_last_ubsync_txg = 0; 4111 spa->spa_load_txg = 0; 4112 mutex_exit(&spa_namespace_lock); 4113 } 4114 4115 *spapp = spa; 4116 4117 return (0); 4118 } 4119 4120 int 4121 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 4122 nvlist_t **config) 4123 { 4124 return (spa_open_common(name, spapp, tag, policy, config)); 4125 } 4126 4127 int 4128 spa_open(const char *name, spa_t **spapp, void *tag) 4129 { 4130 return (spa_open_common(name, spapp, tag, NULL, NULL)); 4131 } 4132 4133 /* 4134 * Lookup the given spa_t, incrementing the inject count in the process, 4135 * preventing it from being exported or destroyed. 4136 */ 4137 spa_t * 4138 spa_inject_addref(char *name) 4139 { 4140 spa_t *spa; 4141 4142 mutex_enter(&spa_namespace_lock); 4143 if ((spa = spa_lookup(name)) == NULL) { 4144 mutex_exit(&spa_namespace_lock); 4145 return (NULL); 4146 } 4147 spa->spa_inject_ref++; 4148 mutex_exit(&spa_namespace_lock); 4149 4150 return (spa); 4151 } 4152 4153 void 4154 spa_inject_delref(spa_t *spa) 4155 { 4156 mutex_enter(&spa_namespace_lock); 4157 spa->spa_inject_ref--; 4158 mutex_exit(&spa_namespace_lock); 4159 } 4160 4161 /* 4162 * Add spares device information to the nvlist. 4163 */ 4164 static void 4165 spa_add_spares(spa_t *spa, nvlist_t *config) 4166 { 4167 nvlist_t **spares; 4168 uint_t i, nspares; 4169 nvlist_t *nvroot; 4170 uint64_t guid; 4171 vdev_stat_t *vs; 4172 uint_t vsc; 4173 uint64_t pool; 4174 4175 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4176 4177 if (spa->spa_spares.sav_count == 0) 4178 return; 4179 4180 VERIFY(nvlist_lookup_nvlist(config, 4181 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4182 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 4183 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4184 if (nspares != 0) { 4185 VERIFY(nvlist_add_nvlist_array(nvroot, 4186 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4187 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4188 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4189 4190 /* 4191 * Go through and find any spares which have since been 4192 * repurposed as an active spare. If this is the case, update 4193 * their status appropriately. 4194 */ 4195 for (i = 0; i < nspares; i++) { 4196 VERIFY(nvlist_lookup_uint64(spares[i], 4197 ZPOOL_CONFIG_GUID, &guid) == 0); 4198 if (spa_spare_exists(guid, &pool, NULL) && 4199 pool != 0ULL) { 4200 VERIFY(nvlist_lookup_uint64_array( 4201 spares[i], ZPOOL_CONFIG_VDEV_STATS, 4202 (uint64_t **)&vs, &vsc) == 0); 4203 vs->vs_state = VDEV_STATE_CANT_OPEN; 4204 vs->vs_aux = VDEV_AUX_SPARED; 4205 } 4206 } 4207 } 4208 } 4209 4210 /* 4211 * Add l2cache device information to the nvlist, including vdev stats. 4212 */ 4213 static void 4214 spa_add_l2cache(spa_t *spa, nvlist_t *config) 4215 { 4216 nvlist_t **l2cache; 4217 uint_t i, j, nl2cache; 4218 nvlist_t *nvroot; 4219 uint64_t guid; 4220 vdev_t *vd; 4221 vdev_stat_t *vs; 4222 uint_t vsc; 4223 4224 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4225 4226 if (spa->spa_l2cache.sav_count == 0) 4227 return; 4228 4229 VERIFY(nvlist_lookup_nvlist(config, 4230 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4231 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 4232 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4233 if (nl2cache != 0) { 4234 VERIFY(nvlist_add_nvlist_array(nvroot, 4235 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4236 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4237 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4238 4239 /* 4240 * Update level 2 cache device stats. 4241 */ 4242 4243 for (i = 0; i < nl2cache; i++) { 4244 VERIFY(nvlist_lookup_uint64(l2cache[i], 4245 ZPOOL_CONFIG_GUID, &guid) == 0); 4246 4247 vd = NULL; 4248 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 4249 if (guid == 4250 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 4251 vd = spa->spa_l2cache.sav_vdevs[j]; 4252 break; 4253 } 4254 } 4255 ASSERT(vd != NULL); 4256 4257 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 4258 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 4259 == 0); 4260 vdev_get_stats(vd, vs); 4261 } 4262 } 4263 } 4264 4265 static void 4266 spa_add_feature_stats(spa_t *spa, nvlist_t *config) 4267 { 4268 nvlist_t *features; 4269 zap_cursor_t zc; 4270 zap_attribute_t za; 4271 4272 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4273 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4274 4275 if (spa->spa_feat_for_read_obj != 0) { 4276 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4277 spa->spa_feat_for_read_obj); 4278 zap_cursor_retrieve(&zc, &za) == 0; 4279 zap_cursor_advance(&zc)) { 4280 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4281 za.za_num_integers == 1); 4282 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4283 za.za_first_integer)); 4284 } 4285 zap_cursor_fini(&zc); 4286 } 4287 4288 if (spa->spa_feat_for_write_obj != 0) { 4289 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4290 spa->spa_feat_for_write_obj); 4291 zap_cursor_retrieve(&zc, &za) == 0; 4292 zap_cursor_advance(&zc)) { 4293 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4294 za.za_num_integers == 1); 4295 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4296 za.za_first_integer)); 4297 } 4298 zap_cursor_fini(&zc); 4299 } 4300 4301 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 4302 features) == 0); 4303 nvlist_free(features); 4304 } 4305 4306 int 4307 spa_get_stats(const char *name, nvlist_t **config, 4308 char *altroot, size_t buflen) 4309 { 4310 int error; 4311 spa_t *spa; 4312 4313 *config = NULL; 4314 error = spa_open_common(name, &spa, FTAG, NULL, config); 4315 4316 if (spa != NULL) { 4317 /* 4318 * This still leaves a window of inconsistency where the spares 4319 * or l2cache devices could change and the config would be 4320 * self-inconsistent. 4321 */ 4322 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4323 4324 if (*config != NULL) { 4325 uint64_t loadtimes[2]; 4326 4327 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 4328 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 4329 VERIFY(nvlist_add_uint64_array(*config, 4330 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 4331 4332 VERIFY(nvlist_add_uint64(*config, 4333 ZPOOL_CONFIG_ERRCOUNT, 4334 spa_get_errlog_size(spa)) == 0); 4335 4336 if (spa_suspended(spa)) 4337 VERIFY(nvlist_add_uint64(*config, 4338 ZPOOL_CONFIG_SUSPENDED, 4339 spa->spa_failmode) == 0); 4340 4341 spa_add_spares(spa, *config); 4342 spa_add_l2cache(spa, *config); 4343 spa_add_feature_stats(spa, *config); 4344 } 4345 } 4346 4347 /* 4348 * We want to get the alternate root even for faulted pools, so we cheat 4349 * and call spa_lookup() directly. 4350 */ 4351 if (altroot) { 4352 if (spa == NULL) { 4353 mutex_enter(&spa_namespace_lock); 4354 spa = spa_lookup(name); 4355 if (spa) 4356 spa_altroot(spa, altroot, buflen); 4357 else 4358 altroot[0] = '\0'; 4359 spa = NULL; 4360 mutex_exit(&spa_namespace_lock); 4361 } else { 4362 spa_altroot(spa, altroot, buflen); 4363 } 4364 } 4365 4366 if (spa != NULL) { 4367 spa_config_exit(spa, SCL_CONFIG, FTAG); 4368 spa_close(spa, FTAG); 4369 } 4370 4371 return (error); 4372 } 4373 4374 /* 4375 * Validate that the auxiliary device array is well formed. We must have an 4376 * array of nvlists, each which describes a valid leaf vdev. If this is an 4377 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 4378 * specified, as long as they are well-formed. 4379 */ 4380 static int 4381 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 4382 spa_aux_vdev_t *sav, const char *config, uint64_t version, 4383 vdev_labeltype_t label) 4384 { 4385 nvlist_t **dev; 4386 uint_t i, ndev; 4387 vdev_t *vd; 4388 int error; 4389 4390 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4391 4392 /* 4393 * It's acceptable to have no devs specified. 4394 */ 4395 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 4396 return (0); 4397 4398 if (ndev == 0) 4399 return (SET_ERROR(EINVAL)); 4400 4401 /* 4402 * Make sure the pool is formatted with a version that supports this 4403 * device type. 4404 */ 4405 if (spa_version(spa) < version) 4406 return (SET_ERROR(ENOTSUP)); 4407 4408 /* 4409 * Set the pending device list so we correctly handle device in-use 4410 * checking. 4411 */ 4412 sav->sav_pending = dev; 4413 sav->sav_npending = ndev; 4414 4415 for (i = 0; i < ndev; i++) { 4416 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 4417 mode)) != 0) 4418 goto out; 4419 4420 if (!vd->vdev_ops->vdev_op_leaf) { 4421 vdev_free(vd); 4422 error = SET_ERROR(EINVAL); 4423 goto out; 4424 } 4425 4426 /* 4427 * The L2ARC currently only supports disk devices in 4428 * kernel context. For user-level testing, we allow it. 4429 */ 4430 #ifdef _KERNEL 4431 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 4432 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 4433 error = SET_ERROR(ENOTBLK); 4434 vdev_free(vd); 4435 goto out; 4436 } 4437 #endif 4438 vd->vdev_top = vd; 4439 4440 if ((error = vdev_open(vd)) == 0 && 4441 (error = vdev_label_init(vd, crtxg, label)) == 0) { 4442 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 4443 vd->vdev_guid) == 0); 4444 } 4445 4446 vdev_free(vd); 4447 4448 if (error && 4449 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 4450 goto out; 4451 else 4452 error = 0; 4453 } 4454 4455 out: 4456 sav->sav_pending = NULL; 4457 sav->sav_npending = 0; 4458 return (error); 4459 } 4460 4461 static int 4462 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 4463 { 4464 int error; 4465 4466 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4467 4468 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4469 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 4470 VDEV_LABEL_SPARE)) != 0) { 4471 return (error); 4472 } 4473 4474 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4475 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 4476 VDEV_LABEL_L2CACHE)); 4477 } 4478 4479 static void 4480 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 4481 const char *config) 4482 { 4483 int i; 4484 4485 if (sav->sav_config != NULL) { 4486 nvlist_t **olddevs; 4487 uint_t oldndevs; 4488 nvlist_t **newdevs; 4489 4490 /* 4491 * Generate new dev list by concatentating with the 4492 * current dev list. 4493 */ 4494 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 4495 &olddevs, &oldndevs) == 0); 4496 4497 newdevs = kmem_alloc(sizeof (void *) * 4498 (ndevs + oldndevs), KM_SLEEP); 4499 for (i = 0; i < oldndevs; i++) 4500 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 4501 KM_SLEEP) == 0); 4502 for (i = 0; i < ndevs; i++) 4503 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 4504 KM_SLEEP) == 0); 4505 4506 VERIFY(nvlist_remove(sav->sav_config, config, 4507 DATA_TYPE_NVLIST_ARRAY) == 0); 4508 4509 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 4510 config, newdevs, ndevs + oldndevs) == 0); 4511 for (i = 0; i < oldndevs + ndevs; i++) 4512 nvlist_free(newdevs[i]); 4513 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 4514 } else { 4515 /* 4516 * Generate a new dev list. 4517 */ 4518 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 4519 KM_SLEEP) == 0); 4520 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 4521 devs, ndevs) == 0); 4522 } 4523 } 4524 4525 /* 4526 * Stop and drop level 2 ARC devices 4527 */ 4528 void 4529 spa_l2cache_drop(spa_t *spa) 4530 { 4531 vdev_t *vd; 4532 int i; 4533 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4534 4535 for (i = 0; i < sav->sav_count; i++) { 4536 uint64_t pool; 4537 4538 vd = sav->sav_vdevs[i]; 4539 ASSERT(vd != NULL); 4540 4541 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 4542 pool != 0ULL && l2arc_vdev_present(vd)) 4543 l2arc_remove_vdev(vd); 4544 } 4545 } 4546 4547 /* 4548 * Pool Creation 4549 */ 4550 int 4551 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 4552 nvlist_t *zplprops) 4553 { 4554 spa_t *spa; 4555 char *altroot = NULL; 4556 vdev_t *rvd; 4557 dsl_pool_t *dp; 4558 dmu_tx_t *tx; 4559 int error = 0; 4560 uint64_t txg = TXG_INITIAL; 4561 nvlist_t **spares, **l2cache; 4562 uint_t nspares, nl2cache; 4563 uint64_t version, obj; 4564 boolean_t has_features; 4565 4566 /* 4567 * If this pool already exists, return failure. 4568 */ 4569 mutex_enter(&spa_namespace_lock); 4570 if (spa_lookup(pool) != NULL) { 4571 mutex_exit(&spa_namespace_lock); 4572 return (SET_ERROR(EEXIST)); 4573 } 4574 4575 /* 4576 * Allocate a new spa_t structure. 4577 */ 4578 (void) nvlist_lookup_string(props, 4579 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4580 spa = spa_add(pool, NULL, altroot); 4581 spa_activate(spa, spa_mode_global); 4582 4583 if (props && (error = spa_prop_validate(spa, props))) { 4584 spa_deactivate(spa); 4585 spa_remove(spa); 4586 mutex_exit(&spa_namespace_lock); 4587 return (error); 4588 } 4589 4590 has_features = B_FALSE; 4591 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 4592 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 4593 if (zpool_prop_feature(nvpair_name(elem))) 4594 has_features = B_TRUE; 4595 } 4596 4597 if (has_features || nvlist_lookup_uint64(props, 4598 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 4599 version = SPA_VERSION; 4600 } 4601 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 4602 4603 spa->spa_first_txg = txg; 4604 spa->spa_uberblock.ub_txg = txg - 1; 4605 spa->spa_uberblock.ub_version = version; 4606 spa->spa_ubsync = spa->spa_uberblock; 4607 spa->spa_load_state = SPA_LOAD_CREATE; 4608 spa->spa_removing_phys.sr_state = DSS_NONE; 4609 spa->spa_removing_phys.sr_removing_vdev = -1; 4610 spa->spa_removing_phys.sr_prev_indirect_vdev = -1; 4611 4612 /* 4613 * Create "The Godfather" zio to hold all async IOs 4614 */ 4615 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 4616 KM_SLEEP); 4617 for (int i = 0; i < max_ncpus; i++) { 4618 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 4619 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 4620 ZIO_FLAG_GODFATHER); 4621 } 4622 4623 /* 4624 * Create the root vdev. 4625 */ 4626 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4627 4628 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 4629 4630 ASSERT(error != 0 || rvd != NULL); 4631 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 4632 4633 if (error == 0 && !zfs_allocatable_devs(nvroot)) 4634 error = SET_ERROR(EINVAL); 4635 4636 if (error == 0 && 4637 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 4638 (error = spa_validate_aux(spa, nvroot, txg, 4639 VDEV_ALLOC_ADD)) == 0) { 4640 for (int c = 0; c < rvd->vdev_children; c++) { 4641 vdev_metaslab_set_size(rvd->vdev_child[c]); 4642 vdev_expand(rvd->vdev_child[c], txg); 4643 } 4644 } 4645 4646 spa_config_exit(spa, SCL_ALL, FTAG); 4647 4648 if (error != 0) { 4649 spa_unload(spa); 4650 spa_deactivate(spa); 4651 spa_remove(spa); 4652 mutex_exit(&spa_namespace_lock); 4653 return (error); 4654 } 4655 4656 /* 4657 * Get the list of spares, if specified. 4658 */ 4659 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4660 &spares, &nspares) == 0) { 4661 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 4662 KM_SLEEP) == 0); 4663 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4664 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4665 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4666 spa_load_spares(spa); 4667 spa_config_exit(spa, SCL_ALL, FTAG); 4668 spa->spa_spares.sav_sync = B_TRUE; 4669 } 4670 4671 /* 4672 * Get the list of level 2 cache devices, if specified. 4673 */ 4674 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4675 &l2cache, &nl2cache) == 0) { 4676 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4677 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4678 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4679 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4681 spa_load_l2cache(spa); 4682 spa_config_exit(spa, SCL_ALL, FTAG); 4683 spa->spa_l2cache.sav_sync = B_TRUE; 4684 } 4685 4686 spa->spa_is_initializing = B_TRUE; 4687 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 4688 spa->spa_meta_objset = dp->dp_meta_objset; 4689 spa->spa_is_initializing = B_FALSE; 4690 4691 /* 4692 * Create DDTs (dedup tables). 4693 */ 4694 ddt_create(spa); 4695 4696 spa_update_dspace(spa); 4697 4698 tx = dmu_tx_create_assigned(dp, txg); 4699 4700 /* 4701 * Create the pool config object. 4702 */ 4703 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 4704 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 4705 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 4706 4707 if (zap_add(spa->spa_meta_objset, 4708 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 4709 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 4710 cmn_err(CE_PANIC, "failed to add pool config"); 4711 } 4712 4713 if (spa_version(spa) >= SPA_VERSION_FEATURES) 4714 spa_feature_create_zap_objects(spa, tx); 4715 4716 if (zap_add(spa->spa_meta_objset, 4717 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 4718 sizeof (uint64_t), 1, &version, tx) != 0) { 4719 cmn_err(CE_PANIC, "failed to add pool version"); 4720 } 4721 4722 /* Newly created pools with the right version are always deflated. */ 4723 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 4724 spa->spa_deflate = TRUE; 4725 if (zap_add(spa->spa_meta_objset, 4726 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 4727 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 4728 cmn_err(CE_PANIC, "failed to add deflate"); 4729 } 4730 } 4731 4732 /* 4733 * Create the deferred-free bpobj. Turn off compression 4734 * because sync-to-convergence takes longer if the blocksize 4735 * keeps changing. 4736 */ 4737 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 4738 dmu_object_set_compress(spa->spa_meta_objset, obj, 4739 ZIO_COMPRESS_OFF, tx); 4740 if (zap_add(spa->spa_meta_objset, 4741 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 4742 sizeof (uint64_t), 1, &obj, tx) != 0) { 4743 cmn_err(CE_PANIC, "failed to add bpobj"); 4744 } 4745 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 4746 spa->spa_meta_objset, obj)); 4747 4748 /* 4749 * Create the pool's history object. 4750 */ 4751 if (version >= SPA_VERSION_ZPOOL_HISTORY) 4752 spa_history_create_obj(spa, tx); 4753 4754 /* 4755 * Generate some random noise for salted checksums to operate on. 4756 */ 4757 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 4758 sizeof (spa->spa_cksum_salt.zcs_bytes)); 4759 4760 /* 4761 * Set pool properties. 4762 */ 4763 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 4764 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 4765 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 4766 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 4767 4768 if (props != NULL) { 4769 spa_configfile_set(spa, props, B_FALSE); 4770 spa_sync_props(props, tx); 4771 } 4772 4773 dmu_tx_commit(tx); 4774 4775 spa->spa_sync_on = B_TRUE; 4776 txg_sync_start(spa->spa_dsl_pool); 4777 4778 /* 4779 * We explicitly wait for the first transaction to complete so that our 4780 * bean counters are appropriately updated. 4781 */ 4782 txg_wait_synced(spa->spa_dsl_pool, txg); 4783 4784 spa_spawn_aux_threads(spa); 4785 4786 spa_write_cachefile(spa, B_FALSE, B_TRUE); 4787 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); 4788 4789 spa_history_log_version(spa, "create"); 4790 4791 /* 4792 * Don't count references from objsets that are already closed 4793 * and are making their way through the eviction process. 4794 */ 4795 spa_evicting_os_wait(spa); 4796 spa->spa_minref = refcount_count(&spa->spa_refcount); 4797 spa->spa_load_state = SPA_LOAD_NONE; 4798 4799 mutex_exit(&spa_namespace_lock); 4800 4801 return (0); 4802 } 4803 4804 #ifdef _KERNEL 4805 /* 4806 * Get the root pool information from the root disk, then import the root pool 4807 * during the system boot up time. 4808 */ 4809 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 4810 4811 static nvlist_t * 4812 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 4813 { 4814 nvlist_t *config; 4815 nvlist_t *nvtop, *nvroot; 4816 uint64_t pgid; 4817 4818 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 4819 return (NULL); 4820 4821 /* 4822 * Add this top-level vdev to the child array. 4823 */ 4824 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4825 &nvtop) == 0); 4826 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4827 &pgid) == 0); 4828 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 4829 4830 /* 4831 * Put this pool's top-level vdevs into a root vdev. 4832 */ 4833 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4834 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4835 VDEV_TYPE_ROOT) == 0); 4836 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4837 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4838 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4839 &nvtop, 1) == 0); 4840 4841 /* 4842 * Replace the existing vdev_tree with the new root vdev in 4843 * this pool's configuration (remove the old, add the new). 4844 */ 4845 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4846 nvlist_free(nvroot); 4847 return (config); 4848 } 4849 4850 /* 4851 * Walk the vdev tree and see if we can find a device with "better" 4852 * configuration. A configuration is "better" if the label on that 4853 * device has a more recent txg. 4854 */ 4855 static void 4856 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 4857 { 4858 for (int c = 0; c < vd->vdev_children; c++) 4859 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 4860 4861 if (vd->vdev_ops->vdev_op_leaf) { 4862 nvlist_t *label; 4863 uint64_t label_txg; 4864 4865 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 4866 &label) != 0) 4867 return; 4868 4869 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 4870 &label_txg) == 0); 4871 4872 /* 4873 * Do we have a better boot device? 4874 */ 4875 if (label_txg > *txg) { 4876 *txg = label_txg; 4877 *avd = vd; 4878 } 4879 nvlist_free(label); 4880 } 4881 } 4882 4883 /* 4884 * Import a root pool. 4885 * 4886 * For x86. devpath_list will consist of devid and/or physpath name of 4887 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 4888 * The GRUB "findroot" command will return the vdev we should boot. 4889 * 4890 * For Sparc, devpath_list consists the physpath name of the booting device 4891 * no matter the rootpool is a single device pool or a mirrored pool. 4892 * e.g. 4893 * "/pci@1f,0/ide@d/disk@0,0:a" 4894 */ 4895 int 4896 spa_import_rootpool(char *devpath, char *devid) 4897 { 4898 spa_t *spa; 4899 vdev_t *rvd, *bvd, *avd = NULL; 4900 nvlist_t *config, *nvtop; 4901 uint64_t guid, txg; 4902 char *pname; 4903 int error; 4904 4905 /* 4906 * Read the label from the boot device and generate a configuration. 4907 */ 4908 config = spa_generate_rootconf(devpath, devid, &guid); 4909 #if defined(_OBP) && defined(_KERNEL) 4910 if (config == NULL) { 4911 if (strstr(devpath, "/iscsi/ssd") != NULL) { 4912 /* iscsi boot */ 4913 get_iscsi_bootpath_phy(devpath); 4914 config = spa_generate_rootconf(devpath, devid, &guid); 4915 } 4916 } 4917 #endif 4918 if (config == NULL) { 4919 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 4920 devpath); 4921 return (SET_ERROR(EIO)); 4922 } 4923 4924 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4925 &pname) == 0); 4926 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 4927 4928 mutex_enter(&spa_namespace_lock); 4929 if ((spa = spa_lookup(pname)) != NULL) { 4930 /* 4931 * Remove the existing root pool from the namespace so that we 4932 * can replace it with the correct config we just read in. 4933 */ 4934 spa_remove(spa); 4935 } 4936 4937 spa = spa_add(pname, config, NULL); 4938 spa->spa_is_root = B_TRUE; 4939 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4940 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4941 &spa->spa_ubsync.ub_version) != 0) 4942 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4943 4944 /* 4945 * Build up a vdev tree based on the boot device's label config. 4946 */ 4947 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4948 &nvtop) == 0); 4949 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4950 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4951 VDEV_ALLOC_ROOTPOOL); 4952 spa_config_exit(spa, SCL_ALL, FTAG); 4953 if (error) { 4954 mutex_exit(&spa_namespace_lock); 4955 nvlist_free(config); 4956 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4957 pname); 4958 return (error); 4959 } 4960 4961 /* 4962 * Get the boot vdev. 4963 */ 4964 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 4965 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 4966 (u_longlong_t)guid); 4967 error = SET_ERROR(ENOENT); 4968 goto out; 4969 } 4970 4971 /* 4972 * Determine if there is a better boot device. 4973 */ 4974 avd = bvd; 4975 spa_alt_rootvdev(rvd, &avd, &txg); 4976 if (avd != bvd) { 4977 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4978 "try booting from '%s'", avd->vdev_path); 4979 error = SET_ERROR(EINVAL); 4980 goto out; 4981 } 4982 4983 /* 4984 * If the boot device is part of a spare vdev then ensure that 4985 * we're booting off the active spare. 4986 */ 4987 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4988 !bvd->vdev_isspare) { 4989 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4990 "try booting from '%s'", 4991 bvd->vdev_parent-> 4992 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4993 error = SET_ERROR(EINVAL); 4994 goto out; 4995 } 4996 4997 error = 0; 4998 out: 4999 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5000 vdev_free(rvd); 5001 spa_config_exit(spa, SCL_ALL, FTAG); 5002 mutex_exit(&spa_namespace_lock); 5003 5004 nvlist_free(config); 5005 return (error); 5006 } 5007 5008 #endif 5009 5010 /* 5011 * Import a non-root pool into the system. 5012 */ 5013 int 5014 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 5015 { 5016 spa_t *spa; 5017 char *altroot = NULL; 5018 spa_load_state_t state = SPA_LOAD_IMPORT; 5019 zpool_rewind_policy_t policy; 5020 uint64_t mode = spa_mode_global; 5021 uint64_t readonly = B_FALSE; 5022 int error; 5023 nvlist_t *nvroot; 5024 nvlist_t **spares, **l2cache; 5025 uint_t nspares, nl2cache; 5026 5027 /* 5028 * If a pool with this name exists, return failure. 5029 */ 5030 mutex_enter(&spa_namespace_lock); 5031 if (spa_lookup(pool) != NULL) { 5032 mutex_exit(&spa_namespace_lock); 5033 return (SET_ERROR(EEXIST)); 5034 } 5035 5036 /* 5037 * Create and initialize the spa structure. 5038 */ 5039 (void) nvlist_lookup_string(props, 5040 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5041 (void) nvlist_lookup_uint64(props, 5042 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 5043 if (readonly) 5044 mode = FREAD; 5045 spa = spa_add(pool, config, altroot); 5046 spa->spa_import_flags = flags; 5047 5048 /* 5049 * Verbatim import - Take a pool and insert it into the namespace 5050 * as if it had been loaded at boot. 5051 */ 5052 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 5053 if (props != NULL) 5054 spa_configfile_set(spa, props, B_FALSE); 5055 5056 spa_write_cachefile(spa, B_FALSE, B_TRUE); 5057 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5058 zfs_dbgmsg("spa_import: verbatim import of %s", pool); 5059 mutex_exit(&spa_namespace_lock); 5060 return (0); 5061 } 5062 5063 spa_activate(spa, mode); 5064 5065 /* 5066 * Don't start async tasks until we know everything is healthy. 5067 */ 5068 spa_async_suspend(spa); 5069 5070 zpool_get_rewind_policy(config, &policy); 5071 if (policy.zrp_request & ZPOOL_DO_REWIND) 5072 state = SPA_LOAD_RECOVER; 5073 5074 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; 5075 5076 if (state != SPA_LOAD_RECOVER) { 5077 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 5078 zfs_dbgmsg("spa_import: importing %s", pool); 5079 } else { 5080 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " 5081 "(RECOVERY MODE)", pool, (longlong_t)policy.zrp_txg); 5082 } 5083 error = spa_load_best(spa, state, policy.zrp_txg, policy.zrp_request); 5084 5085 /* 5086 * Propagate anything learned while loading the pool and pass it 5087 * back to caller (i.e. rewind info, missing devices, etc). 5088 */ 5089 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5090 spa->spa_load_info) == 0); 5091 5092 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5093 /* 5094 * Toss any existing sparelist, as it doesn't have any validity 5095 * anymore, and conflicts with spa_has_spare(). 5096 */ 5097 if (spa->spa_spares.sav_config) { 5098 nvlist_free(spa->spa_spares.sav_config); 5099 spa->spa_spares.sav_config = NULL; 5100 spa_load_spares(spa); 5101 } 5102 if (spa->spa_l2cache.sav_config) { 5103 nvlist_free(spa->spa_l2cache.sav_config); 5104 spa->spa_l2cache.sav_config = NULL; 5105 spa_load_l2cache(spa); 5106 } 5107 5108 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5109 &nvroot) == 0); 5110 if (error == 0) 5111 error = spa_validate_aux(spa, nvroot, -1ULL, 5112 VDEV_ALLOC_SPARE); 5113 if (error == 0) 5114 error = spa_validate_aux(spa, nvroot, -1ULL, 5115 VDEV_ALLOC_L2CACHE); 5116 spa_config_exit(spa, SCL_ALL, FTAG); 5117 5118 if (props != NULL) 5119 spa_configfile_set(spa, props, B_FALSE); 5120 5121 if (error != 0 || (props && spa_writeable(spa) && 5122 (error = spa_prop_set(spa, props)))) { 5123 spa_unload(spa); 5124 spa_deactivate(spa); 5125 spa_remove(spa); 5126 mutex_exit(&spa_namespace_lock); 5127 return (error); 5128 } 5129 5130 spa_async_resume(spa); 5131 5132 /* 5133 * Override any spares and level 2 cache devices as specified by 5134 * the user, as these may have correct device names/devids, etc. 5135 */ 5136 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 5137 &spares, &nspares) == 0) { 5138 if (spa->spa_spares.sav_config) 5139 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 5140 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 5141 else 5142 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 5143 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5144 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 5145 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 5146 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5147 spa_load_spares(spa); 5148 spa_config_exit(spa, SCL_ALL, FTAG); 5149 spa->spa_spares.sav_sync = B_TRUE; 5150 } 5151 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 5152 &l2cache, &nl2cache) == 0) { 5153 if (spa->spa_l2cache.sav_config) 5154 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 5155 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 5156 else 5157 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 5158 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5159 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 5160 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 5161 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5162 spa_load_l2cache(spa); 5163 spa_config_exit(spa, SCL_ALL, FTAG); 5164 spa->spa_l2cache.sav_sync = B_TRUE; 5165 } 5166 5167 /* 5168 * Check for any removed devices. 5169 */ 5170 if (spa->spa_autoreplace) { 5171 spa_aux_check_removed(&spa->spa_spares); 5172 spa_aux_check_removed(&spa->spa_l2cache); 5173 } 5174 5175 if (spa_writeable(spa)) { 5176 /* 5177 * Update the config cache to include the newly-imported pool. 5178 */ 5179 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5180 } 5181 5182 /* 5183 * It's possible that the pool was expanded while it was exported. 5184 * We kick off an async task to handle this for us. 5185 */ 5186 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 5187 5188 spa_history_log_version(spa, "import"); 5189 5190 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5191 5192 mutex_exit(&spa_namespace_lock); 5193 5194 return (0); 5195 } 5196 5197 nvlist_t * 5198 spa_tryimport(nvlist_t *tryconfig) 5199 { 5200 nvlist_t *config = NULL; 5201 char *poolname, *cachefile; 5202 spa_t *spa; 5203 uint64_t state; 5204 int error; 5205 zpool_rewind_policy_t policy; 5206 5207 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 5208 return (NULL); 5209 5210 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 5211 return (NULL); 5212 5213 /* 5214 * Create and initialize the spa structure. 5215 */ 5216 mutex_enter(&spa_namespace_lock); 5217 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 5218 spa_activate(spa, FREAD); 5219 5220 /* 5221 * Rewind pool if a max txg was provided. Note that even though we 5222 * retrieve the complete rewind policy, only the rewind txg is relevant 5223 * for tryimport. 5224 */ 5225 zpool_get_rewind_policy(spa->spa_config, &policy); 5226 if (policy.zrp_txg != UINT64_MAX) { 5227 spa->spa_load_max_txg = policy.zrp_txg; 5228 spa->spa_extreme_rewind = B_TRUE; 5229 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", 5230 poolname, (longlong_t)policy.zrp_txg); 5231 } else { 5232 zfs_dbgmsg("spa_tryimport: importing %s", poolname); 5233 } 5234 5235 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) 5236 == 0) { 5237 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); 5238 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 5239 } else { 5240 spa->spa_config_source = SPA_CONFIG_SRC_SCAN; 5241 } 5242 5243 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); 5244 5245 /* 5246 * If 'tryconfig' was at least parsable, return the current config. 5247 */ 5248 if (spa->spa_root_vdev != NULL) { 5249 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 5250 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 5251 poolname) == 0); 5252 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5253 state) == 0); 5254 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 5255 spa->spa_uberblock.ub_timestamp) == 0); 5256 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5257 spa->spa_load_info) == 0); 5258 5259 /* 5260 * If the bootfs property exists on this pool then we 5261 * copy it out so that external consumers can tell which 5262 * pools are bootable. 5263 */ 5264 if ((!error || error == EEXIST) && spa->spa_bootfs) { 5265 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5266 5267 /* 5268 * We have to play games with the name since the 5269 * pool was opened as TRYIMPORT_NAME. 5270 */ 5271 if (dsl_dsobj_to_dsname(spa_name(spa), 5272 spa->spa_bootfs, tmpname) == 0) { 5273 char *cp; 5274 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5275 5276 cp = strchr(tmpname, '/'); 5277 if (cp == NULL) { 5278 (void) strlcpy(dsname, tmpname, 5279 MAXPATHLEN); 5280 } else { 5281 (void) snprintf(dsname, MAXPATHLEN, 5282 "%s/%s", poolname, ++cp); 5283 } 5284 VERIFY(nvlist_add_string(config, 5285 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 5286 kmem_free(dsname, MAXPATHLEN); 5287 } 5288 kmem_free(tmpname, MAXPATHLEN); 5289 } 5290 5291 /* 5292 * Add the list of hot spares and level 2 cache devices. 5293 */ 5294 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5295 spa_add_spares(spa, config); 5296 spa_add_l2cache(spa, config); 5297 spa_config_exit(spa, SCL_CONFIG, FTAG); 5298 } 5299 5300 spa_unload(spa); 5301 spa_deactivate(spa); 5302 spa_remove(spa); 5303 mutex_exit(&spa_namespace_lock); 5304 5305 return (config); 5306 } 5307 5308 /* 5309 * Pool export/destroy 5310 * 5311 * The act of destroying or exporting a pool is very simple. We make sure there 5312 * is no more pending I/O and any references to the pool are gone. Then, we 5313 * update the pool state and sync all the labels to disk, removing the 5314 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 5315 * we don't sync the labels or remove the configuration cache. 5316 */ 5317 static int 5318 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 5319 boolean_t force, boolean_t hardforce) 5320 { 5321 spa_t *spa; 5322 5323 if (oldconfig) 5324 *oldconfig = NULL; 5325 5326 if (!(spa_mode_global & FWRITE)) 5327 return (SET_ERROR(EROFS)); 5328 5329 mutex_enter(&spa_namespace_lock); 5330 if ((spa = spa_lookup(pool)) == NULL) { 5331 mutex_exit(&spa_namespace_lock); 5332 return (SET_ERROR(ENOENT)); 5333 } 5334 5335 /* 5336 * Put a hold on the pool, drop the namespace lock, stop async tasks, 5337 * reacquire the namespace lock, and see if we can export. 5338 */ 5339 spa_open_ref(spa, FTAG); 5340 mutex_exit(&spa_namespace_lock); 5341 spa_async_suspend(spa); 5342 mutex_enter(&spa_namespace_lock); 5343 spa_close(spa, FTAG); 5344 5345 /* 5346 * The pool will be in core if it's openable, 5347 * in which case we can modify its state. 5348 */ 5349 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 5350 /* 5351 * Objsets may be open only because they're dirty, so we 5352 * have to force it to sync before checking spa_refcnt. 5353 */ 5354 txg_wait_synced(spa->spa_dsl_pool, 0); 5355 spa_evicting_os_wait(spa); 5356 5357 /* 5358 * A pool cannot be exported or destroyed if there are active 5359 * references. If we are resetting a pool, allow references by 5360 * fault injection handlers. 5361 */ 5362 if (!spa_refcount_zero(spa) || 5363 (spa->spa_inject_ref != 0 && 5364 new_state != POOL_STATE_UNINITIALIZED)) { 5365 spa_async_resume(spa); 5366 mutex_exit(&spa_namespace_lock); 5367 return (SET_ERROR(EBUSY)); 5368 } 5369 5370 /* 5371 * A pool cannot be exported if it has an active shared spare. 5372 * This is to prevent other pools stealing the active spare 5373 * from an exported pool. At user's own will, such pool can 5374 * be forcedly exported. 5375 */ 5376 if (!force && new_state == POOL_STATE_EXPORTED && 5377 spa_has_active_shared_spare(spa)) { 5378 spa_async_resume(spa); 5379 mutex_exit(&spa_namespace_lock); 5380 return (SET_ERROR(EXDEV)); 5381 } 5382 5383 /* 5384 * We want this to be reflected on every label, 5385 * so mark them all dirty. spa_unload() will do the 5386 * final sync that pushes these changes out. 5387 */ 5388 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 5389 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5390 spa->spa_state = new_state; 5391 spa->spa_final_txg = spa_last_synced_txg(spa) + 5392 TXG_DEFER_SIZE + 1; 5393 vdev_config_dirty(spa->spa_root_vdev); 5394 spa_config_exit(spa, SCL_ALL, FTAG); 5395 } 5396 } 5397 5398 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); 5399 5400 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 5401 spa_unload(spa); 5402 spa_deactivate(spa); 5403 } 5404 5405 if (oldconfig && spa->spa_config) 5406 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 5407 5408 if (new_state != POOL_STATE_UNINITIALIZED) { 5409 if (!hardforce) 5410 spa_write_cachefile(spa, B_TRUE, B_TRUE); 5411 spa_remove(spa); 5412 } 5413 mutex_exit(&spa_namespace_lock); 5414 5415 return (0); 5416 } 5417 5418 /* 5419 * Destroy a storage pool. 5420 */ 5421 int 5422 spa_destroy(char *pool) 5423 { 5424 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 5425 B_FALSE, B_FALSE)); 5426 } 5427 5428 /* 5429 * Export a storage pool. 5430 */ 5431 int 5432 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 5433 boolean_t hardforce) 5434 { 5435 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 5436 force, hardforce)); 5437 } 5438 5439 /* 5440 * Similar to spa_export(), this unloads the spa_t without actually removing it 5441 * from the namespace in any way. 5442 */ 5443 int 5444 spa_reset(char *pool) 5445 { 5446 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 5447 B_FALSE, B_FALSE)); 5448 } 5449 5450 /* 5451 * ========================================================================== 5452 * Device manipulation 5453 * ========================================================================== 5454 */ 5455 5456 /* 5457 * Add a device to a storage pool. 5458 */ 5459 int 5460 spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 5461 { 5462 uint64_t txg, id; 5463 int error; 5464 vdev_t *rvd = spa->spa_root_vdev; 5465 vdev_t *vd, *tvd; 5466 nvlist_t **spares, **l2cache; 5467 uint_t nspares, nl2cache; 5468 5469 ASSERT(spa_writeable(spa)); 5470 5471 txg = spa_vdev_enter(spa); 5472 5473 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 5474 VDEV_ALLOC_ADD)) != 0) 5475 return (spa_vdev_exit(spa, NULL, txg, error)); 5476 5477 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 5478 5479 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 5480 &nspares) != 0) 5481 nspares = 0; 5482 5483 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 5484 &nl2cache) != 0) 5485 nl2cache = 0; 5486 5487 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 5488 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5489 5490 if (vd->vdev_children != 0 && 5491 (error = vdev_create(vd, txg, B_FALSE)) != 0) 5492 return (spa_vdev_exit(spa, vd, txg, error)); 5493 5494 /* 5495 * We must validate the spares and l2cache devices after checking the 5496 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 5497 */ 5498 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 5499 return (spa_vdev_exit(spa, vd, txg, error)); 5500 5501 /* 5502 * If we are in the middle of a device removal, we can only add 5503 * devices which match the existing devices in the pool. 5504 * If we are in the middle of a removal, or have some indirect 5505 * vdevs, we can not add raidz toplevels. 5506 */ 5507 if (spa->spa_vdev_removal != NULL || 5508 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5509 for (int c = 0; c < vd->vdev_children; c++) { 5510 tvd = vd->vdev_child[c]; 5511 if (spa->spa_vdev_removal != NULL && 5512 tvd->vdev_ashift != 5513 spa->spa_vdev_removal->svr_vdev->vdev_ashift) { 5514 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5515 } 5516 /* Fail if top level vdev is raidz */ 5517 if (tvd->vdev_ops == &vdev_raidz_ops) { 5518 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5519 } 5520 /* 5521 * Need the top level mirror to be 5522 * a mirror of leaf vdevs only 5523 */ 5524 if (tvd->vdev_ops == &vdev_mirror_ops) { 5525 for (uint64_t cid = 0; 5526 cid < tvd->vdev_children; cid++) { 5527 vdev_t *cvd = tvd->vdev_child[cid]; 5528 if (!cvd->vdev_ops->vdev_op_leaf) { 5529 return (spa_vdev_exit(spa, vd, 5530 txg, EINVAL)); 5531 } 5532 } 5533 } 5534 } 5535 } 5536 5537 for (int c = 0; c < vd->vdev_children; c++) { 5538 5539 /* 5540 * Set the vdev id to the first hole, if one exists. 5541 */ 5542 for (id = 0; id < rvd->vdev_children; id++) { 5543 if (rvd->vdev_child[id]->vdev_ishole) { 5544 vdev_free(rvd->vdev_child[id]); 5545 break; 5546 } 5547 } 5548 tvd = vd->vdev_child[c]; 5549 vdev_remove_child(vd, tvd); 5550 tvd->vdev_id = id; 5551 vdev_add_child(rvd, tvd); 5552 vdev_config_dirty(tvd); 5553 } 5554 5555 if (nspares != 0) { 5556 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 5557 ZPOOL_CONFIG_SPARES); 5558 spa_load_spares(spa); 5559 spa->spa_spares.sav_sync = B_TRUE; 5560 } 5561 5562 if (nl2cache != 0) { 5563 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 5564 ZPOOL_CONFIG_L2CACHE); 5565 spa_load_l2cache(spa); 5566 spa->spa_l2cache.sav_sync = B_TRUE; 5567 } 5568 5569 /* 5570 * We have to be careful when adding new vdevs to an existing pool. 5571 * If other threads start allocating from these vdevs before we 5572 * sync the config cache, and we lose power, then upon reboot we may 5573 * fail to open the pool because there are DVAs that the config cache 5574 * can't translate. Therefore, we first add the vdevs without 5575 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 5576 * and then let spa_config_update() initialize the new metaslabs. 5577 * 5578 * spa_load() checks for added-but-not-initialized vdevs, so that 5579 * if we lose power at any point in this sequence, the remaining 5580 * steps will be completed the next time we load the pool. 5581 */ 5582 (void) spa_vdev_exit(spa, vd, txg, 0); 5583 5584 mutex_enter(&spa_namespace_lock); 5585 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5586 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); 5587 mutex_exit(&spa_namespace_lock); 5588 5589 return (0); 5590 } 5591 5592 /* 5593 * Attach a device to a mirror. The arguments are the path to any device 5594 * in the mirror, and the nvroot for the new device. If the path specifies 5595 * a device that is not mirrored, we automatically insert the mirror vdev. 5596 * 5597 * If 'replacing' is specified, the new device is intended to replace the 5598 * existing device; in this case the two devices are made into their own 5599 * mirror using the 'replacing' vdev, which is functionally identical to 5600 * the mirror vdev (it actually reuses all the same ops) but has a few 5601 * extra rules: you can't attach to it after it's been created, and upon 5602 * completion of resilvering, the first disk (the one being replaced) 5603 * is automatically detached. 5604 */ 5605 int 5606 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 5607 { 5608 uint64_t txg, dtl_max_txg; 5609 vdev_t *rvd = spa->spa_root_vdev; 5610 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 5611 vdev_ops_t *pvops; 5612 char *oldvdpath, *newvdpath; 5613 int newvd_isspare; 5614 int error; 5615 5616 ASSERT(spa_writeable(spa)); 5617 5618 txg = spa_vdev_enter(spa); 5619 5620 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 5621 5622 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5623 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 5624 error = (spa_has_checkpoint(spa)) ? 5625 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 5626 return (spa_vdev_exit(spa, NULL, txg, error)); 5627 } 5628 5629 if (spa->spa_vdev_removal != NULL || 5630 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5631 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5632 } 5633 5634 if (oldvd == NULL) 5635 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5636 5637 if (!oldvd->vdev_ops->vdev_op_leaf) 5638 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5639 5640 pvd = oldvd->vdev_parent; 5641 5642 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 5643 VDEV_ALLOC_ATTACH)) != 0) 5644 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5645 5646 if (newrootvd->vdev_children != 1) 5647 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5648 5649 newvd = newrootvd->vdev_child[0]; 5650 5651 if (!newvd->vdev_ops->vdev_op_leaf) 5652 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5653 5654 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 5655 return (spa_vdev_exit(spa, newrootvd, txg, error)); 5656 5657 /* 5658 * Spares can't replace logs 5659 */ 5660 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 5661 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5662 5663 if (!replacing) { 5664 /* 5665 * For attach, the only allowable parent is a mirror or the root 5666 * vdev. 5667 */ 5668 if (pvd->vdev_ops != &vdev_mirror_ops && 5669 pvd->vdev_ops != &vdev_root_ops) 5670 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5671 5672 pvops = &vdev_mirror_ops; 5673 } else { 5674 /* 5675 * Active hot spares can only be replaced by inactive hot 5676 * spares. 5677 */ 5678 if (pvd->vdev_ops == &vdev_spare_ops && 5679 oldvd->vdev_isspare && 5680 !spa_has_spare(spa, newvd->vdev_guid)) 5681 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5682 5683 /* 5684 * If the source is a hot spare, and the parent isn't already a 5685 * spare, then we want to create a new hot spare. Otherwise, we 5686 * want to create a replacing vdev. The user is not allowed to 5687 * attach to a spared vdev child unless the 'isspare' state is 5688 * the same (spare replaces spare, non-spare replaces 5689 * non-spare). 5690 */ 5691 if (pvd->vdev_ops == &vdev_replacing_ops && 5692 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 5693 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5694 } else if (pvd->vdev_ops == &vdev_spare_ops && 5695 newvd->vdev_isspare != oldvd->vdev_isspare) { 5696 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5697 } 5698 5699 if (newvd->vdev_isspare) 5700 pvops = &vdev_spare_ops; 5701 else 5702 pvops = &vdev_replacing_ops; 5703 } 5704 5705 /* 5706 * Make sure the new device is big enough. 5707 */ 5708 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 5709 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 5710 5711 /* 5712 * The new device cannot have a higher alignment requirement 5713 * than the top-level vdev. 5714 */ 5715 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 5716 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 5717 5718 /* 5719 * If this is an in-place replacement, update oldvd's path and devid 5720 * to make it distinguishable from newvd, and unopenable from now on. 5721 */ 5722 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 5723 spa_strfree(oldvd->vdev_path); 5724 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 5725 KM_SLEEP); 5726 (void) sprintf(oldvd->vdev_path, "%s/%s", 5727 newvd->vdev_path, "old"); 5728 if (oldvd->vdev_devid != NULL) { 5729 spa_strfree(oldvd->vdev_devid); 5730 oldvd->vdev_devid = NULL; 5731 } 5732 } 5733 5734 /* mark the device being resilvered */ 5735 newvd->vdev_resilver_txg = txg; 5736 5737 /* 5738 * If the parent is not a mirror, or if we're replacing, insert the new 5739 * mirror/replacing/spare vdev above oldvd. 5740 */ 5741 if (pvd->vdev_ops != pvops) 5742 pvd = vdev_add_parent(oldvd, pvops); 5743 5744 ASSERT(pvd->vdev_top->vdev_parent == rvd); 5745 ASSERT(pvd->vdev_ops == pvops); 5746 ASSERT(oldvd->vdev_parent == pvd); 5747 5748 /* 5749 * Extract the new device from its root and add it to pvd. 5750 */ 5751 vdev_remove_child(newrootvd, newvd); 5752 newvd->vdev_id = pvd->vdev_children; 5753 newvd->vdev_crtxg = oldvd->vdev_crtxg; 5754 vdev_add_child(pvd, newvd); 5755 5756 tvd = newvd->vdev_top; 5757 ASSERT(pvd->vdev_top == tvd); 5758 ASSERT(tvd->vdev_parent == rvd); 5759 5760 vdev_config_dirty(tvd); 5761 5762 /* 5763 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 5764 * for any dmu_sync-ed blocks. It will propagate upward when 5765 * spa_vdev_exit() calls vdev_dtl_reassess(). 5766 */ 5767 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 5768 5769 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 5770 dtl_max_txg - TXG_INITIAL); 5771 5772 if (newvd->vdev_isspare) { 5773 spa_spare_activate(newvd); 5774 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); 5775 } 5776 5777 oldvdpath = spa_strdup(oldvd->vdev_path); 5778 newvdpath = spa_strdup(newvd->vdev_path); 5779 newvd_isspare = newvd->vdev_isspare; 5780 5781 /* 5782 * Mark newvd's DTL dirty in this txg. 5783 */ 5784 vdev_dirty(tvd, VDD_DTL, newvd, txg); 5785 5786 /* 5787 * Schedule the resilver to restart in the future. We do this to 5788 * ensure that dmu_sync-ed blocks have been stitched into the 5789 * respective datasets. 5790 */ 5791 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 5792 5793 if (spa->spa_bootfs) 5794 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); 5795 5796 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); 5797 5798 /* 5799 * Commit the config 5800 */ 5801 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 5802 5803 spa_history_log_internal(spa, "vdev attach", NULL, 5804 "%s vdev=%s %s vdev=%s", 5805 replacing && newvd_isspare ? "spare in" : 5806 replacing ? "replace" : "attach", newvdpath, 5807 replacing ? "for" : "to", oldvdpath); 5808 5809 spa_strfree(oldvdpath); 5810 spa_strfree(newvdpath); 5811 5812 return (0); 5813 } 5814 5815 /* 5816 * Detach a device from a mirror or replacing vdev. 5817 * 5818 * If 'replace_done' is specified, only detach if the parent 5819 * is a replacing vdev. 5820 */ 5821 int 5822 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 5823 { 5824 uint64_t txg; 5825 int error; 5826 vdev_t *rvd = spa->spa_root_vdev; 5827 vdev_t *vd, *pvd, *cvd, *tvd; 5828 boolean_t unspare = B_FALSE; 5829 uint64_t unspare_guid = 0; 5830 char *vdpath; 5831 5832 ASSERT(spa_writeable(spa)); 5833 5834 txg = spa_vdev_enter(spa); 5835 5836 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5837 5838 /* 5839 * Besides being called directly from the userland through the 5840 * ioctl interface, spa_vdev_detach() can be potentially called 5841 * at the end of spa_vdev_resilver_done(). 5842 * 5843 * In the regular case, when we have a checkpoint this shouldn't 5844 * happen as we never empty the DTLs of a vdev during the scrub 5845 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() 5846 * should never get here when we have a checkpoint. 5847 * 5848 * That said, even in a case when we checkpoint the pool exactly 5849 * as spa_vdev_resilver_done() calls this function everything 5850 * should be fine as the resilver will return right away. 5851 */ 5852 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5853 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 5854 error = (spa_has_checkpoint(spa)) ? 5855 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 5856 return (spa_vdev_exit(spa, NULL, txg, error)); 5857 } 5858 5859 if (vd == NULL) 5860 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5861 5862 if (!vd->vdev_ops->vdev_op_leaf) 5863 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5864 5865 pvd = vd->vdev_parent; 5866 5867 /* 5868 * If the parent/child relationship is not as expected, don't do it. 5869 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5870 * vdev that's replacing B with C. The user's intent in replacing 5871 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5872 * the replace by detaching C, the expected behavior is to end up 5873 * M(A,B). But suppose that right after deciding to detach C, 5874 * the replacement of B completes. We would have M(A,C), and then 5875 * ask to detach C, which would leave us with just A -- not what 5876 * the user wanted. To prevent this, we make sure that the 5877 * parent/child relationship hasn't changed -- in this example, 5878 * that C's parent is still the replacing vdev R. 5879 */ 5880 if (pvd->vdev_guid != pguid && pguid != 0) 5881 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5882 5883 /* 5884 * Only 'replacing' or 'spare' vdevs can be replaced. 5885 */ 5886 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5887 pvd->vdev_ops != &vdev_spare_ops) 5888 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5889 5890 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5891 spa_version(spa) >= SPA_VERSION_SPARES); 5892 5893 /* 5894 * Only mirror, replacing, and spare vdevs support detach. 5895 */ 5896 if (pvd->vdev_ops != &vdev_replacing_ops && 5897 pvd->vdev_ops != &vdev_mirror_ops && 5898 pvd->vdev_ops != &vdev_spare_ops) 5899 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5900 5901 /* 5902 * If this device has the only valid copy of some data, 5903 * we cannot safely detach it. 5904 */ 5905 if (vdev_dtl_required(vd)) 5906 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5907 5908 ASSERT(pvd->vdev_children >= 2); 5909 5910 /* 5911 * If we are detaching the second disk from a replacing vdev, then 5912 * check to see if we changed the original vdev's path to have "/old" 5913 * at the end in spa_vdev_attach(). If so, undo that change now. 5914 */ 5915 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5916 vd->vdev_path != NULL) { 5917 size_t len = strlen(vd->vdev_path); 5918 5919 for (int c = 0; c < pvd->vdev_children; c++) { 5920 cvd = pvd->vdev_child[c]; 5921 5922 if (cvd == vd || cvd->vdev_path == NULL) 5923 continue; 5924 5925 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5926 strcmp(cvd->vdev_path + len, "/old") == 0) { 5927 spa_strfree(cvd->vdev_path); 5928 cvd->vdev_path = spa_strdup(vd->vdev_path); 5929 break; 5930 } 5931 } 5932 } 5933 5934 /* 5935 * If we are detaching the original disk from a spare, then it implies 5936 * that the spare should become a real disk, and be removed from the 5937 * active spare list for the pool. 5938 */ 5939 if (pvd->vdev_ops == &vdev_spare_ops && 5940 vd->vdev_id == 0 && 5941 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5942 unspare = B_TRUE; 5943 5944 /* 5945 * Erase the disk labels so the disk can be used for other things. 5946 * This must be done after all other error cases are handled, 5947 * but before we disembowel vd (so we can still do I/O to it). 5948 * But if we can't do it, don't treat the error as fatal -- 5949 * it may be that the unwritability of the disk is the reason 5950 * it's being detached! 5951 */ 5952 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5953 5954 /* 5955 * Remove vd from its parent and compact the parent's children. 5956 */ 5957 vdev_remove_child(pvd, vd); 5958 vdev_compact_children(pvd); 5959 5960 /* 5961 * Remember one of the remaining children so we can get tvd below. 5962 */ 5963 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5964 5965 /* 5966 * If we need to remove the remaining child from the list of hot spares, 5967 * do it now, marking the vdev as no longer a spare in the process. 5968 * We must do this before vdev_remove_parent(), because that can 5969 * change the GUID if it creates a new toplevel GUID. For a similar 5970 * reason, we must remove the spare now, in the same txg as the detach; 5971 * otherwise someone could attach a new sibling, change the GUID, and 5972 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5973 */ 5974 if (unspare) { 5975 ASSERT(cvd->vdev_isspare); 5976 spa_spare_remove(cvd); 5977 unspare_guid = cvd->vdev_guid; 5978 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5979 cvd->vdev_unspare = B_TRUE; 5980 } 5981 5982 /* 5983 * If the parent mirror/replacing vdev only has one child, 5984 * the parent is no longer needed. Remove it from the tree. 5985 */ 5986 if (pvd->vdev_children == 1) { 5987 if (pvd->vdev_ops == &vdev_spare_ops) 5988 cvd->vdev_unspare = B_FALSE; 5989 vdev_remove_parent(cvd); 5990 } 5991 5992 5993 /* 5994 * We don't set tvd until now because the parent we just removed 5995 * may have been the previous top-level vdev. 5996 */ 5997 tvd = cvd->vdev_top; 5998 ASSERT(tvd->vdev_parent == rvd); 5999 6000 /* 6001 * Reevaluate the parent vdev state. 6002 */ 6003 vdev_propagate_state(cvd); 6004 6005 /* 6006 * If the 'autoexpand' property is set on the pool then automatically 6007 * try to expand the size of the pool. For example if the device we 6008 * just detached was smaller than the others, it may be possible to 6009 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 6010 * first so that we can obtain the updated sizes of the leaf vdevs. 6011 */ 6012 if (spa->spa_autoexpand) { 6013 vdev_reopen(tvd); 6014 vdev_expand(tvd, txg); 6015 } 6016 6017 vdev_config_dirty(tvd); 6018 6019 /* 6020 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 6021 * vd->vdev_detached is set and free vd's DTL object in syncing context. 6022 * But first make sure we're not on any *other* txg's DTL list, to 6023 * prevent vd from being accessed after it's freed. 6024 */ 6025 vdpath = spa_strdup(vd->vdev_path); 6026 for (int t = 0; t < TXG_SIZE; t++) 6027 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 6028 vd->vdev_detached = B_TRUE; 6029 vdev_dirty(tvd, VDD_DTL, vd, txg); 6030 6031 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); 6032 6033 /* hang on to the spa before we release the lock */ 6034 spa_open_ref(spa, FTAG); 6035 6036 error = spa_vdev_exit(spa, vd, txg, 0); 6037 6038 spa_history_log_internal(spa, "detach", NULL, 6039 "vdev=%s", vdpath); 6040 spa_strfree(vdpath); 6041 6042 /* 6043 * If this was the removal of the original device in a hot spare vdev, 6044 * then we want to go through and remove the device from the hot spare 6045 * list of every other pool. 6046 */ 6047 if (unspare) { 6048 spa_t *altspa = NULL; 6049 6050 mutex_enter(&spa_namespace_lock); 6051 while ((altspa = spa_next(altspa)) != NULL) { 6052 if (altspa->spa_state != POOL_STATE_ACTIVE || 6053 altspa == spa) 6054 continue; 6055 6056 spa_open_ref(altspa, FTAG); 6057 mutex_exit(&spa_namespace_lock); 6058 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 6059 mutex_enter(&spa_namespace_lock); 6060 spa_close(altspa, FTAG); 6061 } 6062 mutex_exit(&spa_namespace_lock); 6063 6064 /* search the rest of the vdevs for spares to remove */ 6065 spa_vdev_resilver_done(spa); 6066 } 6067 6068 /* all done with the spa; OK to release */ 6069 mutex_enter(&spa_namespace_lock); 6070 spa_close(spa, FTAG); 6071 mutex_exit(&spa_namespace_lock); 6072 6073 return (error); 6074 } 6075 6076 /* 6077 * Split a set of devices from their mirrors, and create a new pool from them. 6078 */ 6079 int 6080 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 6081 nvlist_t *props, boolean_t exp) 6082 { 6083 int error = 0; 6084 uint64_t txg, *glist; 6085 spa_t *newspa; 6086 uint_t c, children, lastlog; 6087 nvlist_t **child, *nvl, *tmp; 6088 dmu_tx_t *tx; 6089 char *altroot = NULL; 6090 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 6091 boolean_t activate_slog; 6092 6093 ASSERT(spa_writeable(spa)); 6094 6095 txg = spa_vdev_enter(spa); 6096 6097 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6098 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6099 error = (spa_has_checkpoint(spa)) ? 6100 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6101 return (spa_vdev_exit(spa, NULL, txg, error)); 6102 } 6103 6104 /* clear the log and flush everything up to now */ 6105 activate_slog = spa_passivate_log(spa); 6106 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6107 error = spa_reset_logs(spa); 6108 txg = spa_vdev_config_enter(spa); 6109 6110 if (activate_slog) 6111 spa_activate_log(spa); 6112 6113 if (error != 0) 6114 return (spa_vdev_exit(spa, NULL, txg, error)); 6115 6116 /* check new spa name before going any further */ 6117 if (spa_lookup(newname) != NULL) 6118 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 6119 6120 /* 6121 * scan through all the children to ensure they're all mirrors 6122 */ 6123 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 6124 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 6125 &children) != 0) 6126 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6127 6128 /* first, check to ensure we've got the right child count */ 6129 rvd = spa->spa_root_vdev; 6130 lastlog = 0; 6131 for (c = 0; c < rvd->vdev_children; c++) { 6132 vdev_t *vd = rvd->vdev_child[c]; 6133 6134 /* don't count the holes & logs as children */ 6135 if (vd->vdev_islog || !vdev_is_concrete(vd)) { 6136 if (lastlog == 0) 6137 lastlog = c; 6138 continue; 6139 } 6140 6141 lastlog = 0; 6142 } 6143 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 6144 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6145 6146 /* next, ensure no spare or cache devices are part of the split */ 6147 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 6148 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 6149 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6150 6151 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 6152 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 6153 6154 /* then, loop over each vdev and validate it */ 6155 for (c = 0; c < children; c++) { 6156 uint64_t is_hole = 0; 6157 6158 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 6159 &is_hole); 6160 6161 if (is_hole != 0) { 6162 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 6163 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 6164 continue; 6165 } else { 6166 error = SET_ERROR(EINVAL); 6167 break; 6168 } 6169 } 6170 6171 /* which disk is going to be split? */ 6172 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 6173 &glist[c]) != 0) { 6174 error = SET_ERROR(EINVAL); 6175 break; 6176 } 6177 6178 /* look it up in the spa */ 6179 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 6180 if (vml[c] == NULL) { 6181 error = SET_ERROR(ENODEV); 6182 break; 6183 } 6184 6185 /* make sure there's nothing stopping the split */ 6186 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 6187 vml[c]->vdev_islog || 6188 !vdev_is_concrete(vml[c]) || 6189 vml[c]->vdev_isspare || 6190 vml[c]->vdev_isl2cache || 6191 !vdev_writeable(vml[c]) || 6192 vml[c]->vdev_children != 0 || 6193 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 6194 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 6195 error = SET_ERROR(EINVAL); 6196 break; 6197 } 6198 6199 if (vdev_dtl_required(vml[c])) { 6200 error = SET_ERROR(EBUSY); 6201 break; 6202 } 6203 6204 /* we need certain info from the top level */ 6205 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 6206 vml[c]->vdev_top->vdev_ms_array) == 0); 6207 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 6208 vml[c]->vdev_top->vdev_ms_shift) == 0); 6209 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 6210 vml[c]->vdev_top->vdev_asize) == 0); 6211 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 6212 vml[c]->vdev_top->vdev_ashift) == 0); 6213 6214 /* transfer per-vdev ZAPs */ 6215 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 6216 VERIFY0(nvlist_add_uint64(child[c], 6217 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 6218 6219 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 6220 VERIFY0(nvlist_add_uint64(child[c], 6221 ZPOOL_CONFIG_VDEV_TOP_ZAP, 6222 vml[c]->vdev_parent->vdev_top_zap)); 6223 } 6224 6225 if (error != 0) { 6226 kmem_free(vml, children * sizeof (vdev_t *)); 6227 kmem_free(glist, children * sizeof (uint64_t)); 6228 return (spa_vdev_exit(spa, NULL, txg, error)); 6229 } 6230 6231 /* stop writers from using the disks */ 6232 for (c = 0; c < children; c++) { 6233 if (vml[c] != NULL) 6234 vml[c]->vdev_offline = B_TRUE; 6235 } 6236 vdev_reopen(spa->spa_root_vdev); 6237 6238 /* 6239 * Temporarily record the splitting vdevs in the spa config. This 6240 * will disappear once the config is regenerated. 6241 */ 6242 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6243 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 6244 glist, children) == 0); 6245 kmem_free(glist, children * sizeof (uint64_t)); 6246 6247 mutex_enter(&spa->spa_props_lock); 6248 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 6249 nvl) == 0); 6250 mutex_exit(&spa->spa_props_lock); 6251 spa->spa_config_splitting = nvl; 6252 vdev_config_dirty(spa->spa_root_vdev); 6253 6254 /* configure and create the new pool */ 6255 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 6256 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 6257 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 6258 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6259 spa_version(spa)) == 0); 6260 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 6261 spa->spa_config_txg) == 0); 6262 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 6263 spa_generate_guid(NULL)) == 0); 6264 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 6265 (void) nvlist_lookup_string(props, 6266 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 6267 6268 /* add the new pool to the namespace */ 6269 newspa = spa_add(newname, config, altroot); 6270 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 6271 newspa->spa_config_txg = spa->spa_config_txg; 6272 spa_set_log_state(newspa, SPA_LOG_CLEAR); 6273 6274 /* release the spa config lock, retaining the namespace lock */ 6275 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6276 6277 if (zio_injection_enabled) 6278 zio_handle_panic_injection(spa, FTAG, 1); 6279 6280 spa_activate(newspa, spa_mode_global); 6281 spa_async_suspend(newspa); 6282 6283 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; 6284 6285 /* create the new pool from the disks of the original pool */ 6286 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); 6287 if (error) 6288 goto out; 6289 6290 /* if that worked, generate a real config for the new pool */ 6291 if (newspa->spa_root_vdev != NULL) { 6292 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 6293 NV_UNIQUE_NAME, KM_SLEEP) == 0); 6294 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 6295 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 6296 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 6297 B_TRUE)); 6298 } 6299 6300 /* set the props */ 6301 if (props != NULL) { 6302 spa_configfile_set(newspa, props, B_FALSE); 6303 error = spa_prop_set(newspa, props); 6304 if (error) 6305 goto out; 6306 } 6307 6308 /* flush everything */ 6309 txg = spa_vdev_config_enter(newspa); 6310 vdev_config_dirty(newspa->spa_root_vdev); 6311 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 6312 6313 if (zio_injection_enabled) 6314 zio_handle_panic_injection(spa, FTAG, 2); 6315 6316 spa_async_resume(newspa); 6317 6318 /* finally, update the original pool's config */ 6319 txg = spa_vdev_config_enter(spa); 6320 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 6321 error = dmu_tx_assign(tx, TXG_WAIT); 6322 if (error != 0) 6323 dmu_tx_abort(tx); 6324 for (c = 0; c < children; c++) { 6325 if (vml[c] != NULL) { 6326 vdev_split(vml[c]); 6327 if (error == 0) 6328 spa_history_log_internal(spa, "detach", tx, 6329 "vdev=%s", vml[c]->vdev_path); 6330 6331 vdev_free(vml[c]); 6332 } 6333 } 6334 spa->spa_avz_action = AVZ_ACTION_REBUILD; 6335 vdev_config_dirty(spa->spa_root_vdev); 6336 spa->spa_config_splitting = NULL; 6337 nvlist_free(nvl); 6338 if (error == 0) 6339 dmu_tx_commit(tx); 6340 (void) spa_vdev_exit(spa, NULL, txg, 0); 6341 6342 if (zio_injection_enabled) 6343 zio_handle_panic_injection(spa, FTAG, 3); 6344 6345 /* split is complete; log a history record */ 6346 spa_history_log_internal(newspa, "split", NULL, 6347 "from pool %s", spa_name(spa)); 6348 6349 kmem_free(vml, children * sizeof (vdev_t *)); 6350 6351 /* if we're not going to mount the filesystems in userland, export */ 6352 if (exp) 6353 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 6354 B_FALSE, B_FALSE); 6355 6356 return (error); 6357 6358 out: 6359 spa_unload(newspa); 6360 spa_deactivate(newspa); 6361 spa_remove(newspa); 6362 6363 txg = spa_vdev_config_enter(spa); 6364 6365 /* re-online all offlined disks */ 6366 for (c = 0; c < children; c++) { 6367 if (vml[c] != NULL) 6368 vml[c]->vdev_offline = B_FALSE; 6369 } 6370 vdev_reopen(spa->spa_root_vdev); 6371 6372 nvlist_free(spa->spa_config_splitting); 6373 spa->spa_config_splitting = NULL; 6374 (void) spa_vdev_exit(spa, NULL, txg, error); 6375 6376 kmem_free(vml, children * sizeof (vdev_t *)); 6377 return (error); 6378 } 6379 6380 /* 6381 * Find any device that's done replacing, or a vdev marked 'unspare' that's 6382 * currently spared, so we can detach it. 6383 */ 6384 static vdev_t * 6385 spa_vdev_resilver_done_hunt(vdev_t *vd) 6386 { 6387 vdev_t *newvd, *oldvd; 6388 6389 for (int c = 0; c < vd->vdev_children; c++) { 6390 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 6391 if (oldvd != NULL) 6392 return (oldvd); 6393 } 6394 6395 /* 6396 * Check for a completed replacement. We always consider the first 6397 * vdev in the list to be the oldest vdev, and the last one to be 6398 * the newest (see spa_vdev_attach() for how that works). In 6399 * the case where the newest vdev is faulted, we will not automatically 6400 * remove it after a resilver completes. This is OK as it will require 6401 * user intervention to determine which disk the admin wishes to keep. 6402 */ 6403 if (vd->vdev_ops == &vdev_replacing_ops) { 6404 ASSERT(vd->vdev_children > 1); 6405 6406 newvd = vd->vdev_child[vd->vdev_children - 1]; 6407 oldvd = vd->vdev_child[0]; 6408 6409 if (vdev_dtl_empty(newvd, DTL_MISSING) && 6410 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6411 !vdev_dtl_required(oldvd)) 6412 return (oldvd); 6413 } 6414 6415 /* 6416 * Check for a completed resilver with the 'unspare' flag set. 6417 */ 6418 if (vd->vdev_ops == &vdev_spare_ops) { 6419 vdev_t *first = vd->vdev_child[0]; 6420 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 6421 6422 if (last->vdev_unspare) { 6423 oldvd = first; 6424 newvd = last; 6425 } else if (first->vdev_unspare) { 6426 oldvd = last; 6427 newvd = first; 6428 } else { 6429 oldvd = NULL; 6430 } 6431 6432 if (oldvd != NULL && 6433 vdev_dtl_empty(newvd, DTL_MISSING) && 6434 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6435 !vdev_dtl_required(oldvd)) 6436 return (oldvd); 6437 6438 /* 6439 * If there are more than two spares attached to a disk, 6440 * and those spares are not required, then we want to 6441 * attempt to free them up now so that they can be used 6442 * by other pools. Once we're back down to a single 6443 * disk+spare, we stop removing them. 6444 */ 6445 if (vd->vdev_children > 2) { 6446 newvd = vd->vdev_child[1]; 6447 6448 if (newvd->vdev_isspare && last->vdev_isspare && 6449 vdev_dtl_empty(last, DTL_MISSING) && 6450 vdev_dtl_empty(last, DTL_OUTAGE) && 6451 !vdev_dtl_required(newvd)) 6452 return (newvd); 6453 } 6454 } 6455 6456 return (NULL); 6457 } 6458 6459 static void 6460 spa_vdev_resilver_done(spa_t *spa) 6461 { 6462 vdev_t *vd, *pvd, *ppvd; 6463 uint64_t guid, sguid, pguid, ppguid; 6464 6465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6466 6467 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 6468 pvd = vd->vdev_parent; 6469 ppvd = pvd->vdev_parent; 6470 guid = vd->vdev_guid; 6471 pguid = pvd->vdev_guid; 6472 ppguid = ppvd->vdev_guid; 6473 sguid = 0; 6474 /* 6475 * If we have just finished replacing a hot spared device, then 6476 * we need to detach the parent's first child (the original hot 6477 * spare) as well. 6478 */ 6479 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 6480 ppvd->vdev_children == 2) { 6481 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 6482 sguid = ppvd->vdev_child[1]->vdev_guid; 6483 } 6484 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 6485 6486 spa_config_exit(spa, SCL_ALL, FTAG); 6487 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 6488 return; 6489 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 6490 return; 6491 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6492 } 6493 6494 spa_config_exit(spa, SCL_ALL, FTAG); 6495 } 6496 6497 /* 6498 * Update the stored path or FRU for this vdev. 6499 */ 6500 int 6501 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 6502 boolean_t ispath) 6503 { 6504 vdev_t *vd; 6505 boolean_t sync = B_FALSE; 6506 6507 ASSERT(spa_writeable(spa)); 6508 6509 spa_vdev_state_enter(spa, SCL_ALL); 6510 6511 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 6512 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 6513 6514 if (!vd->vdev_ops->vdev_op_leaf) 6515 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 6516 6517 if (ispath) { 6518 if (strcmp(value, vd->vdev_path) != 0) { 6519 spa_strfree(vd->vdev_path); 6520 vd->vdev_path = spa_strdup(value); 6521 sync = B_TRUE; 6522 } 6523 } else { 6524 if (vd->vdev_fru == NULL) { 6525 vd->vdev_fru = spa_strdup(value); 6526 sync = B_TRUE; 6527 } else if (strcmp(value, vd->vdev_fru) != 0) { 6528 spa_strfree(vd->vdev_fru); 6529 vd->vdev_fru = spa_strdup(value); 6530 sync = B_TRUE; 6531 } 6532 } 6533 6534 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 6535 } 6536 6537 int 6538 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 6539 { 6540 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 6541 } 6542 6543 int 6544 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 6545 { 6546 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 6547 } 6548 6549 /* 6550 * ========================================================================== 6551 * SPA Scanning 6552 * ========================================================================== 6553 */ 6554 int 6555 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 6556 { 6557 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6558 6559 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6560 return (SET_ERROR(EBUSY)); 6561 6562 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 6563 } 6564 6565 int 6566 spa_scan_stop(spa_t *spa) 6567 { 6568 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6569 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6570 return (SET_ERROR(EBUSY)); 6571 return (dsl_scan_cancel(spa->spa_dsl_pool)); 6572 } 6573 6574 int 6575 spa_scan(spa_t *spa, pool_scan_func_t func) 6576 { 6577 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6578 6579 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 6580 return (SET_ERROR(ENOTSUP)); 6581 6582 /* 6583 * If a resilver was requested, but there is no DTL on a 6584 * writeable leaf device, we have nothing to do. 6585 */ 6586 if (func == POOL_SCAN_RESILVER && 6587 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 6588 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 6589 return (0); 6590 } 6591 6592 return (dsl_scan(spa->spa_dsl_pool, func)); 6593 } 6594 6595 /* 6596 * ========================================================================== 6597 * SPA async task processing 6598 * ========================================================================== 6599 */ 6600 6601 static void 6602 spa_async_remove(spa_t *spa, vdev_t *vd) 6603 { 6604 if (vd->vdev_remove_wanted) { 6605 vd->vdev_remove_wanted = B_FALSE; 6606 vd->vdev_delayed_close = B_FALSE; 6607 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6608 6609 /* 6610 * We want to clear the stats, but we don't want to do a full 6611 * vdev_clear() as that will cause us to throw away 6612 * degraded/faulted state as well as attempt to reopen the 6613 * device, all of which is a waste. 6614 */ 6615 vd->vdev_stat.vs_read_errors = 0; 6616 vd->vdev_stat.vs_write_errors = 0; 6617 vd->vdev_stat.vs_checksum_errors = 0; 6618 6619 vdev_state_dirty(vd->vdev_top); 6620 } 6621 6622 for (int c = 0; c < vd->vdev_children; c++) 6623 spa_async_remove(spa, vd->vdev_child[c]); 6624 } 6625 6626 static void 6627 spa_async_probe(spa_t *spa, vdev_t *vd) 6628 { 6629 if (vd->vdev_probe_wanted) { 6630 vd->vdev_probe_wanted = B_FALSE; 6631 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6632 } 6633 6634 for (int c = 0; c < vd->vdev_children; c++) 6635 spa_async_probe(spa, vd->vdev_child[c]); 6636 } 6637 6638 static void 6639 spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6640 { 6641 sysevent_id_t eid; 6642 nvlist_t *attr; 6643 char *physpath; 6644 6645 if (!spa->spa_autoexpand) 6646 return; 6647 6648 for (int c = 0; c < vd->vdev_children; c++) { 6649 vdev_t *cvd = vd->vdev_child[c]; 6650 spa_async_autoexpand(spa, cvd); 6651 } 6652 6653 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6654 return; 6655 6656 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6657 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6658 6659 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6660 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6661 6662 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6663 ESC_DEV_DLE, attr, &eid, DDI_SLEEP); 6664 6665 nvlist_free(attr); 6666 kmem_free(physpath, MAXPATHLEN); 6667 } 6668 6669 static void 6670 spa_async_thread(void *arg) 6671 { 6672 spa_t *spa = (spa_t *)arg; 6673 int tasks; 6674 6675 ASSERT(spa->spa_sync_on); 6676 6677 mutex_enter(&spa->spa_async_lock); 6678 tasks = spa->spa_async_tasks; 6679 spa->spa_async_tasks = 0; 6680 mutex_exit(&spa->spa_async_lock); 6681 6682 /* 6683 * See if the config needs to be updated. 6684 */ 6685 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6686 uint64_t old_space, new_space; 6687 6688 mutex_enter(&spa_namespace_lock); 6689 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6690 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6691 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6692 mutex_exit(&spa_namespace_lock); 6693 6694 /* 6695 * If the pool grew as a result of the config update, 6696 * then log an internal history event. 6697 */ 6698 if (new_space != old_space) { 6699 spa_history_log_internal(spa, "vdev online", NULL, 6700 "pool '%s' size: %llu(+%llu)", 6701 spa_name(spa), new_space, new_space - old_space); 6702 } 6703 } 6704 6705 /* 6706 * See if any devices need to be marked REMOVED. 6707 */ 6708 if (tasks & SPA_ASYNC_REMOVE) { 6709 spa_vdev_state_enter(spa, SCL_NONE); 6710 spa_async_remove(spa, spa->spa_root_vdev); 6711 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6712 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6713 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6714 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6715 (void) spa_vdev_state_exit(spa, NULL, 0); 6716 } 6717 6718 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6719 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6720 spa_async_autoexpand(spa, spa->spa_root_vdev); 6721 spa_config_exit(spa, SCL_CONFIG, FTAG); 6722 } 6723 6724 /* 6725 * See if any devices need to be probed. 6726 */ 6727 if (tasks & SPA_ASYNC_PROBE) { 6728 spa_vdev_state_enter(spa, SCL_NONE); 6729 spa_async_probe(spa, spa->spa_root_vdev); 6730 (void) spa_vdev_state_exit(spa, NULL, 0); 6731 } 6732 6733 /* 6734 * If any devices are done replacing, detach them. 6735 */ 6736 if (tasks & SPA_ASYNC_RESILVER_DONE) 6737 spa_vdev_resilver_done(spa); 6738 6739 /* 6740 * Kick off a resilver. 6741 */ 6742 if (tasks & SPA_ASYNC_RESILVER) 6743 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6744 6745 /* 6746 * Let the world know that we're done. 6747 */ 6748 mutex_enter(&spa->spa_async_lock); 6749 spa->spa_async_thread = NULL; 6750 cv_broadcast(&spa->spa_async_cv); 6751 mutex_exit(&spa->spa_async_lock); 6752 thread_exit(); 6753 } 6754 6755 void 6756 spa_async_suspend(spa_t *spa) 6757 { 6758 mutex_enter(&spa->spa_async_lock); 6759 spa->spa_async_suspended++; 6760 while (spa->spa_async_thread != NULL) 6761 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6762 mutex_exit(&spa->spa_async_lock); 6763 6764 spa_vdev_remove_suspend(spa); 6765 6766 zthr_t *condense_thread = spa->spa_condense_zthr; 6767 if (condense_thread != NULL && zthr_isrunning(condense_thread)) 6768 VERIFY0(zthr_cancel(condense_thread)); 6769 6770 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 6771 if (discard_thread != NULL && zthr_isrunning(discard_thread)) 6772 VERIFY0(zthr_cancel(discard_thread)); 6773 } 6774 6775 void 6776 spa_async_resume(spa_t *spa) 6777 { 6778 mutex_enter(&spa->spa_async_lock); 6779 ASSERT(spa->spa_async_suspended != 0); 6780 spa->spa_async_suspended--; 6781 mutex_exit(&spa->spa_async_lock); 6782 spa_restart_removal(spa); 6783 6784 zthr_t *condense_thread = spa->spa_condense_zthr; 6785 if (condense_thread != NULL && !zthr_isrunning(condense_thread)) 6786 zthr_resume(condense_thread); 6787 6788 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 6789 if (discard_thread != NULL && !zthr_isrunning(discard_thread)) 6790 zthr_resume(discard_thread); 6791 } 6792 6793 static boolean_t 6794 spa_async_tasks_pending(spa_t *spa) 6795 { 6796 uint_t non_config_tasks; 6797 uint_t config_task; 6798 boolean_t config_task_suspended; 6799 6800 non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE; 6801 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6802 if (spa->spa_ccw_fail_time == 0) { 6803 config_task_suspended = B_FALSE; 6804 } else { 6805 config_task_suspended = 6806 (gethrtime() - spa->spa_ccw_fail_time) < 6807 (zfs_ccw_retry_interval * NANOSEC); 6808 } 6809 6810 return (non_config_tasks || (config_task && !config_task_suspended)); 6811 } 6812 6813 static void 6814 spa_async_dispatch(spa_t *spa) 6815 { 6816 mutex_enter(&spa->spa_async_lock); 6817 if (spa_async_tasks_pending(spa) && 6818 !spa->spa_async_suspended && 6819 spa->spa_async_thread == NULL && 6820 rootdir != NULL) 6821 spa->spa_async_thread = thread_create(NULL, 0, 6822 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6823 mutex_exit(&spa->spa_async_lock); 6824 } 6825 6826 void 6827 spa_async_request(spa_t *spa, int task) 6828 { 6829 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6830 mutex_enter(&spa->spa_async_lock); 6831 spa->spa_async_tasks |= task; 6832 mutex_exit(&spa->spa_async_lock); 6833 } 6834 6835 /* 6836 * ========================================================================== 6837 * SPA syncing routines 6838 * ========================================================================== 6839 */ 6840 6841 static int 6842 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6843 { 6844 bpobj_t *bpo = arg; 6845 bpobj_enqueue(bpo, bp, tx); 6846 return (0); 6847 } 6848 6849 static int 6850 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6851 { 6852 zio_t *zio = arg; 6853 6854 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6855 zio->io_flags)); 6856 return (0); 6857 } 6858 6859 /* 6860 * Note: this simple function is not inlined to make it easier to dtrace the 6861 * amount of time spent syncing frees. 6862 */ 6863 static void 6864 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6865 { 6866 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6867 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6868 VERIFY(zio_wait(zio) == 0); 6869 } 6870 6871 /* 6872 * Note: this simple function is not inlined to make it easier to dtrace the 6873 * amount of time spent syncing deferred frees. 6874 */ 6875 static void 6876 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6877 { 6878 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6879 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6880 spa_free_sync_cb, zio, tx), ==, 0); 6881 VERIFY0(zio_wait(zio)); 6882 } 6883 6884 6885 static void 6886 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6887 { 6888 char *packed = NULL; 6889 size_t bufsize; 6890 size_t nvsize = 0; 6891 dmu_buf_t *db; 6892 6893 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6894 6895 /* 6896 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6897 * information. This avoids the dmu_buf_will_dirty() path and 6898 * saves us a pre-read to get data we don't actually care about. 6899 */ 6900 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6901 packed = kmem_alloc(bufsize, KM_SLEEP); 6902 6903 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6904 KM_SLEEP) == 0); 6905 bzero(packed + nvsize, bufsize - nvsize); 6906 6907 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6908 6909 kmem_free(packed, bufsize); 6910 6911 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6912 dmu_buf_will_dirty(db, tx); 6913 *(uint64_t *)db->db_data = nvsize; 6914 dmu_buf_rele(db, FTAG); 6915 } 6916 6917 static void 6918 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6919 const char *config, const char *entry) 6920 { 6921 nvlist_t *nvroot; 6922 nvlist_t **list; 6923 int i; 6924 6925 if (!sav->sav_sync) 6926 return; 6927 6928 /* 6929 * Update the MOS nvlist describing the list of available devices. 6930 * spa_validate_aux() will have already made sure this nvlist is 6931 * valid and the vdevs are labeled appropriately. 6932 */ 6933 if (sav->sav_object == 0) { 6934 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6935 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6936 sizeof (uint64_t), tx); 6937 VERIFY(zap_update(spa->spa_meta_objset, 6938 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6939 &sav->sav_object, tx) == 0); 6940 } 6941 6942 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6943 if (sav->sav_count == 0) { 6944 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6945 } else { 6946 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6947 for (i = 0; i < sav->sav_count; i++) 6948 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6949 B_FALSE, VDEV_CONFIG_L2CACHE); 6950 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6951 sav->sav_count) == 0); 6952 for (i = 0; i < sav->sav_count; i++) 6953 nvlist_free(list[i]); 6954 kmem_free(list, sav->sav_count * sizeof (void *)); 6955 } 6956 6957 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6958 nvlist_free(nvroot); 6959 6960 sav->sav_sync = B_FALSE; 6961 } 6962 6963 /* 6964 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 6965 * The all-vdev ZAP must be empty. 6966 */ 6967 static void 6968 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 6969 { 6970 spa_t *spa = vd->vdev_spa; 6971 if (vd->vdev_top_zap != 0) { 6972 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6973 vd->vdev_top_zap, tx)); 6974 } 6975 if (vd->vdev_leaf_zap != 0) { 6976 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6977 vd->vdev_leaf_zap, tx)); 6978 } 6979 for (uint64_t i = 0; i < vd->vdev_children; i++) { 6980 spa_avz_build(vd->vdev_child[i], avz, tx); 6981 } 6982 } 6983 6984 static void 6985 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6986 { 6987 nvlist_t *config; 6988 6989 /* 6990 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 6991 * its config may not be dirty but we still need to build per-vdev ZAPs. 6992 * Similarly, if the pool is being assembled (e.g. after a split), we 6993 * need to rebuild the AVZ although the config may not be dirty. 6994 */ 6995 if (list_is_empty(&spa->spa_config_dirty_list) && 6996 spa->spa_avz_action == AVZ_ACTION_NONE) 6997 return; 6998 6999 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7000 7001 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 7002 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 7003 spa->spa_all_vdev_zaps != 0); 7004 7005 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 7006 /* Make and build the new AVZ */ 7007 uint64_t new_avz = zap_create(spa->spa_meta_objset, 7008 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 7009 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 7010 7011 /* Diff old AVZ with new one */ 7012 zap_cursor_t zc; 7013 zap_attribute_t za; 7014 7015 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7016 spa->spa_all_vdev_zaps); 7017 zap_cursor_retrieve(&zc, &za) == 0; 7018 zap_cursor_advance(&zc)) { 7019 uint64_t vdzap = za.za_first_integer; 7020 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 7021 vdzap) == ENOENT) { 7022 /* 7023 * ZAP is listed in old AVZ but not in new one; 7024 * destroy it 7025 */ 7026 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 7027 tx)); 7028 } 7029 } 7030 7031 zap_cursor_fini(&zc); 7032 7033 /* Destroy the old AVZ */ 7034 VERIFY0(zap_destroy(spa->spa_meta_objset, 7035 spa->spa_all_vdev_zaps, tx)); 7036 7037 /* Replace the old AVZ in the dir obj with the new one */ 7038 VERIFY0(zap_update(spa->spa_meta_objset, 7039 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 7040 sizeof (new_avz), 1, &new_avz, tx)); 7041 7042 spa->spa_all_vdev_zaps = new_avz; 7043 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 7044 zap_cursor_t zc; 7045 zap_attribute_t za; 7046 7047 /* Walk through the AVZ and destroy all listed ZAPs */ 7048 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7049 spa->spa_all_vdev_zaps); 7050 zap_cursor_retrieve(&zc, &za) == 0; 7051 zap_cursor_advance(&zc)) { 7052 uint64_t zap = za.za_first_integer; 7053 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 7054 } 7055 7056 zap_cursor_fini(&zc); 7057 7058 /* Destroy and unlink the AVZ itself */ 7059 VERIFY0(zap_destroy(spa->spa_meta_objset, 7060 spa->spa_all_vdev_zaps, tx)); 7061 VERIFY0(zap_remove(spa->spa_meta_objset, 7062 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 7063 spa->spa_all_vdev_zaps = 0; 7064 } 7065 7066 if (spa->spa_all_vdev_zaps == 0) { 7067 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 7068 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 7069 DMU_POOL_VDEV_ZAP_MAP, tx); 7070 } 7071 spa->spa_avz_action = AVZ_ACTION_NONE; 7072 7073 /* Create ZAPs for vdevs that don't have them. */ 7074 vdev_construct_zaps(spa->spa_root_vdev, tx); 7075 7076 config = spa_config_generate(spa, spa->spa_root_vdev, 7077 dmu_tx_get_txg(tx), B_FALSE); 7078 7079 /* 7080 * If we're upgrading the spa version then make sure that 7081 * the config object gets updated with the correct version. 7082 */ 7083 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 7084 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 7085 spa->spa_uberblock.ub_version); 7086 7087 spa_config_exit(spa, SCL_STATE, FTAG); 7088 7089 nvlist_free(spa->spa_config_syncing); 7090 spa->spa_config_syncing = config; 7091 7092 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 7093 } 7094 7095 static void 7096 spa_sync_version(void *arg, dmu_tx_t *tx) 7097 { 7098 uint64_t *versionp = arg; 7099 uint64_t version = *versionp; 7100 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7101 7102 /* 7103 * Setting the version is special cased when first creating the pool. 7104 */ 7105 ASSERT(tx->tx_txg != TXG_INITIAL); 7106 7107 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 7108 ASSERT(version >= spa_version(spa)); 7109 7110 spa->spa_uberblock.ub_version = version; 7111 vdev_config_dirty(spa->spa_root_vdev); 7112 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 7113 } 7114 7115 /* 7116 * Set zpool properties. 7117 */ 7118 static void 7119 spa_sync_props(void *arg, dmu_tx_t *tx) 7120 { 7121 nvlist_t *nvp = arg; 7122 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7123 objset_t *mos = spa->spa_meta_objset; 7124 nvpair_t *elem = NULL; 7125 7126 mutex_enter(&spa->spa_props_lock); 7127 7128 while ((elem = nvlist_next_nvpair(nvp, elem))) { 7129 uint64_t intval; 7130 char *strval, *fname; 7131 zpool_prop_t prop; 7132 const char *propname; 7133 zprop_type_t proptype; 7134 spa_feature_t fid; 7135 7136 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 7137 case ZPOOL_PROP_INVAL: 7138 /* 7139 * We checked this earlier in spa_prop_validate(). 7140 */ 7141 ASSERT(zpool_prop_feature(nvpair_name(elem))); 7142 7143 fname = strchr(nvpair_name(elem), '@') + 1; 7144 VERIFY0(zfeature_lookup_name(fname, &fid)); 7145 7146 spa_feature_enable(spa, fid, tx); 7147 spa_history_log_internal(spa, "set", tx, 7148 "%s=enabled", nvpair_name(elem)); 7149 break; 7150 7151 case ZPOOL_PROP_VERSION: 7152 intval = fnvpair_value_uint64(elem); 7153 /* 7154 * The version is synced seperatly before other 7155 * properties and should be correct by now. 7156 */ 7157 ASSERT3U(spa_version(spa), >=, intval); 7158 break; 7159 7160 case ZPOOL_PROP_ALTROOT: 7161 /* 7162 * 'altroot' is a non-persistent property. It should 7163 * have been set temporarily at creation or import time. 7164 */ 7165 ASSERT(spa->spa_root != NULL); 7166 break; 7167 7168 case ZPOOL_PROP_READONLY: 7169 case ZPOOL_PROP_CACHEFILE: 7170 /* 7171 * 'readonly' and 'cachefile' are also non-persisitent 7172 * properties. 7173 */ 7174 break; 7175 case ZPOOL_PROP_COMMENT: 7176 strval = fnvpair_value_string(elem); 7177 if (spa->spa_comment != NULL) 7178 spa_strfree(spa->spa_comment); 7179 spa->spa_comment = spa_strdup(strval); 7180 /* 7181 * We need to dirty the configuration on all the vdevs 7182 * so that their labels get updated. It's unnecessary 7183 * to do this for pool creation since the vdev's 7184 * configuratoin has already been dirtied. 7185 */ 7186 if (tx->tx_txg != TXG_INITIAL) 7187 vdev_config_dirty(spa->spa_root_vdev); 7188 spa_history_log_internal(spa, "set", tx, 7189 "%s=%s", nvpair_name(elem), strval); 7190 break; 7191 default: 7192 /* 7193 * Set pool property values in the poolprops mos object. 7194 */ 7195 if (spa->spa_pool_props_object == 0) { 7196 spa->spa_pool_props_object = 7197 zap_create_link(mos, DMU_OT_POOL_PROPS, 7198 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 7199 tx); 7200 } 7201 7202 /* normalize the property name */ 7203 propname = zpool_prop_to_name(prop); 7204 proptype = zpool_prop_get_type(prop); 7205 7206 if (nvpair_type(elem) == DATA_TYPE_STRING) { 7207 ASSERT(proptype == PROP_TYPE_STRING); 7208 strval = fnvpair_value_string(elem); 7209 VERIFY0(zap_update(mos, 7210 spa->spa_pool_props_object, propname, 7211 1, strlen(strval) + 1, strval, tx)); 7212 spa_history_log_internal(spa, "set", tx, 7213 "%s=%s", nvpair_name(elem), strval); 7214 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 7215 intval = fnvpair_value_uint64(elem); 7216 7217 if (proptype == PROP_TYPE_INDEX) { 7218 const char *unused; 7219 VERIFY0(zpool_prop_index_to_string( 7220 prop, intval, &unused)); 7221 } 7222 VERIFY0(zap_update(mos, 7223 spa->spa_pool_props_object, propname, 7224 8, 1, &intval, tx)); 7225 spa_history_log_internal(spa, "set", tx, 7226 "%s=%lld", nvpair_name(elem), intval); 7227 } else { 7228 ASSERT(0); /* not allowed */ 7229 } 7230 7231 switch (prop) { 7232 case ZPOOL_PROP_DELEGATION: 7233 spa->spa_delegation = intval; 7234 break; 7235 case ZPOOL_PROP_BOOTFS: 7236 spa->spa_bootfs = intval; 7237 break; 7238 case ZPOOL_PROP_FAILUREMODE: 7239 spa->spa_failmode = intval; 7240 break; 7241 case ZPOOL_PROP_AUTOEXPAND: 7242 spa->spa_autoexpand = intval; 7243 if (tx->tx_txg != TXG_INITIAL) 7244 spa_async_request(spa, 7245 SPA_ASYNC_AUTOEXPAND); 7246 break; 7247 case ZPOOL_PROP_DEDUPDITTO: 7248 spa->spa_dedup_ditto = intval; 7249 break; 7250 default: 7251 break; 7252 } 7253 } 7254 7255 } 7256 7257 mutex_exit(&spa->spa_props_lock); 7258 } 7259 7260 /* 7261 * Perform one-time upgrade on-disk changes. spa_version() does not 7262 * reflect the new version this txg, so there must be no changes this 7263 * txg to anything that the upgrade code depends on after it executes. 7264 * Therefore this must be called after dsl_pool_sync() does the sync 7265 * tasks. 7266 */ 7267 static void 7268 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 7269 { 7270 dsl_pool_t *dp = spa->spa_dsl_pool; 7271 7272 ASSERT(spa->spa_sync_pass == 1); 7273 7274 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 7275 7276 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 7277 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 7278 dsl_pool_create_origin(dp, tx); 7279 7280 /* Keeping the origin open increases spa_minref */ 7281 spa->spa_minref += 3; 7282 } 7283 7284 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 7285 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 7286 dsl_pool_upgrade_clones(dp, tx); 7287 } 7288 7289 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 7290 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 7291 dsl_pool_upgrade_dir_clones(dp, tx); 7292 7293 /* Keeping the freedir open increases spa_minref */ 7294 spa->spa_minref += 3; 7295 } 7296 7297 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 7298 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7299 spa_feature_create_zap_objects(spa, tx); 7300 } 7301 7302 /* 7303 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 7304 * when possibility to use lz4 compression for metadata was added 7305 * Old pools that have this feature enabled must be upgraded to have 7306 * this feature active 7307 */ 7308 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7309 boolean_t lz4_en = spa_feature_is_enabled(spa, 7310 SPA_FEATURE_LZ4_COMPRESS); 7311 boolean_t lz4_ac = spa_feature_is_active(spa, 7312 SPA_FEATURE_LZ4_COMPRESS); 7313 7314 if (lz4_en && !lz4_ac) 7315 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 7316 } 7317 7318 /* 7319 * If we haven't written the salt, do so now. Note that the 7320 * feature may not be activated yet, but that's fine since 7321 * the presence of this ZAP entry is backwards compatible. 7322 */ 7323 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 7324 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 7325 VERIFY0(zap_add(spa->spa_meta_objset, 7326 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 7327 sizeof (spa->spa_cksum_salt.zcs_bytes), 7328 spa->spa_cksum_salt.zcs_bytes, tx)); 7329 } 7330 7331 rrw_exit(&dp->dp_config_rwlock, FTAG); 7332 } 7333 7334 static void 7335 vdev_indirect_state_sync_verify(vdev_t *vd) 7336 { 7337 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 7338 vdev_indirect_births_t *vib = vd->vdev_indirect_births; 7339 7340 if (vd->vdev_ops == &vdev_indirect_ops) { 7341 ASSERT(vim != NULL); 7342 ASSERT(vib != NULL); 7343 } 7344 7345 if (vdev_obsolete_sm_object(vd) != 0) { 7346 ASSERT(vd->vdev_obsolete_sm != NULL); 7347 ASSERT(vd->vdev_removing || 7348 vd->vdev_ops == &vdev_indirect_ops); 7349 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); 7350 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); 7351 7352 ASSERT3U(vdev_obsolete_sm_object(vd), ==, 7353 space_map_object(vd->vdev_obsolete_sm)); 7354 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, 7355 space_map_allocated(vd->vdev_obsolete_sm)); 7356 } 7357 ASSERT(vd->vdev_obsolete_segments != NULL); 7358 7359 /* 7360 * Since frees / remaps to an indirect vdev can only 7361 * happen in syncing context, the obsolete segments 7362 * tree must be empty when we start syncing. 7363 */ 7364 ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); 7365 } 7366 7367 /* 7368 * Sync the specified transaction group. New blocks may be dirtied as 7369 * part of the process, so we iterate until it converges. 7370 */ 7371 void 7372 spa_sync(spa_t *spa, uint64_t txg) 7373 { 7374 dsl_pool_t *dp = spa->spa_dsl_pool; 7375 objset_t *mos = spa->spa_meta_objset; 7376 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 7377 vdev_t *rvd = spa->spa_root_vdev; 7378 vdev_t *vd; 7379 dmu_tx_t *tx; 7380 int error; 7381 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 7382 zfs_vdev_queue_depth_pct / 100; 7383 7384 VERIFY(spa_writeable(spa)); 7385 7386 /* 7387 * Wait for i/os issued in open context that need to complete 7388 * before this txg syncs. 7389 */ 7390 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK])); 7391 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0); 7392 7393 /* 7394 * Lock out configuration changes. 7395 */ 7396 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7397 7398 spa->spa_syncing_txg = txg; 7399 spa->spa_sync_pass = 0; 7400 7401 mutex_enter(&spa->spa_alloc_lock); 7402 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7403 mutex_exit(&spa->spa_alloc_lock); 7404 7405 /* 7406 * If there are any pending vdev state changes, convert them 7407 * into config changes that go out with this transaction group. 7408 */ 7409 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7410 while (list_head(&spa->spa_state_dirty_list) != NULL) { 7411 /* 7412 * We need the write lock here because, for aux vdevs, 7413 * calling vdev_config_dirty() modifies sav_config. 7414 * This is ugly and will become unnecessary when we 7415 * eliminate the aux vdev wart by integrating all vdevs 7416 * into the root vdev tree. 7417 */ 7418 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7419 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 7420 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 7421 vdev_state_clean(vd); 7422 vdev_config_dirty(vd); 7423 } 7424 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7425 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 7426 } 7427 spa_config_exit(spa, SCL_STATE, FTAG); 7428 7429 tx = dmu_tx_create_assigned(dp, txg); 7430 7431 spa->spa_sync_starttime = gethrtime(); 7432 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 7433 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 7434 7435 /* 7436 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 7437 * set spa_deflate if we have no raid-z vdevs. 7438 */ 7439 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 7440 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 7441 int i; 7442 7443 for (i = 0; i < rvd->vdev_children; i++) { 7444 vd = rvd->vdev_child[i]; 7445 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 7446 break; 7447 } 7448 if (i == rvd->vdev_children) { 7449 spa->spa_deflate = TRUE; 7450 VERIFY(0 == zap_add(spa->spa_meta_objset, 7451 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 7452 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 7453 } 7454 } 7455 7456 /* 7457 * Set the top-level vdev's max queue depth. Evaluate each 7458 * top-level's async write queue depth in case it changed. 7459 * The max queue depth will not change in the middle of syncing 7460 * out this txg. 7461 */ 7462 uint64_t queue_depth_total = 0; 7463 for (int c = 0; c < rvd->vdev_children; c++) { 7464 vdev_t *tvd = rvd->vdev_child[c]; 7465 metaslab_group_t *mg = tvd->vdev_mg; 7466 7467 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 7468 !metaslab_group_initialized(mg)) 7469 continue; 7470 7471 /* 7472 * It is safe to do a lock-free check here because only async 7473 * allocations look at mg_max_alloc_queue_depth, and async 7474 * allocations all happen from spa_sync(). 7475 */ 7476 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); 7477 mg->mg_max_alloc_queue_depth = max_queue_depth; 7478 queue_depth_total += mg->mg_max_alloc_queue_depth; 7479 } 7480 metaslab_class_t *mc = spa_normal_class(spa); 7481 ASSERT0(refcount_count(&mc->mc_alloc_slots)); 7482 mc->mc_alloc_max_slots = queue_depth_total; 7483 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 7484 7485 ASSERT3U(mc->mc_alloc_max_slots, <=, 7486 max_queue_depth * rvd->vdev_children); 7487 7488 for (int c = 0; c < rvd->vdev_children; c++) { 7489 vdev_t *vd = rvd->vdev_child[c]; 7490 vdev_indirect_state_sync_verify(vd); 7491 7492 if (vdev_indirect_should_condense(vd)) { 7493 spa_condense_indirect_start_sync(vd, tx); 7494 break; 7495 } 7496 } 7497 7498 /* 7499 * Iterate to convergence. 7500 */ 7501 do { 7502 int pass = ++spa->spa_sync_pass; 7503 7504 spa_sync_config_object(spa, tx); 7505 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 7506 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 7507 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 7508 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 7509 spa_errlog_sync(spa, txg); 7510 dsl_pool_sync(dp, txg); 7511 7512 if (pass < zfs_sync_pass_deferred_free) { 7513 spa_sync_frees(spa, free_bpl, tx); 7514 } else { 7515 /* 7516 * We can not defer frees in pass 1, because 7517 * we sync the deferred frees later in pass 1. 7518 */ 7519 ASSERT3U(pass, >, 1); 7520 bplist_iterate(free_bpl, bpobj_enqueue_cb, 7521 &spa->spa_deferred_bpobj, tx); 7522 } 7523 7524 ddt_sync(spa, txg); 7525 dsl_scan_sync(dp, tx); 7526 7527 if (spa->spa_vdev_removal != NULL) 7528 svr_sync(spa, tx); 7529 7530 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 7531 != NULL) 7532 vdev_sync(vd, txg); 7533 7534 if (pass == 1) { 7535 spa_sync_upgrades(spa, tx); 7536 ASSERT3U(txg, >=, 7537 spa->spa_uberblock.ub_rootbp.blk_birth); 7538 /* 7539 * Note: We need to check if the MOS is dirty 7540 * because we could have marked the MOS dirty 7541 * without updating the uberblock (e.g. if we 7542 * have sync tasks but no dirty user data). We 7543 * need to check the uberblock's rootbp because 7544 * it is updated if we have synced out dirty 7545 * data (though in this case the MOS will most 7546 * likely also be dirty due to second order 7547 * effects, we don't want to rely on that here). 7548 */ 7549 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 7550 !dmu_objset_is_dirty(mos, txg)) { 7551 /* 7552 * Nothing changed on the first pass, 7553 * therefore this TXG is a no-op. Avoid 7554 * syncing deferred frees, so that we 7555 * can keep this TXG as a no-op. 7556 */ 7557 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 7558 txg)); 7559 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7560 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 7561 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, 7562 txg)); 7563 break; 7564 } 7565 spa_sync_deferred_frees(spa, tx); 7566 } 7567 7568 } while (dmu_objset_is_dirty(mos, txg)); 7569 7570 if (!list_is_empty(&spa->spa_config_dirty_list)) { 7571 /* 7572 * Make sure that the number of ZAPs for all the vdevs matches 7573 * the number of ZAPs in the per-vdev ZAP list. This only gets 7574 * called if the config is dirty; otherwise there may be 7575 * outstanding AVZ operations that weren't completed in 7576 * spa_sync_config_object. 7577 */ 7578 uint64_t all_vdev_zap_entry_count; 7579 ASSERT0(zap_count(spa->spa_meta_objset, 7580 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 7581 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 7582 all_vdev_zap_entry_count); 7583 } 7584 7585 if (spa->spa_vdev_removal != NULL) { 7586 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); 7587 } 7588 7589 /* 7590 * Rewrite the vdev configuration (which includes the uberblock) 7591 * to commit the transaction group. 7592 * 7593 * If there are no dirty vdevs, we sync the uberblock to a few 7594 * random top-level vdevs that are known to be visible in the 7595 * config cache (see spa_vdev_add() for a complete description). 7596 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 7597 */ 7598 for (;;) { 7599 /* 7600 * We hold SCL_STATE to prevent vdev open/close/etc. 7601 * while we're attempting to write the vdev labels. 7602 */ 7603 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7604 7605 if (list_is_empty(&spa->spa_config_dirty_list)) { 7606 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 7607 int svdcount = 0; 7608 int children = rvd->vdev_children; 7609 int c0 = spa_get_random(children); 7610 7611 for (int c = 0; c < children; c++) { 7612 vd = rvd->vdev_child[(c0 + c) % children]; 7613 7614 /* Stop when revisiting the first vdev */ 7615 if (c > 0 && svd[0] == vd) 7616 break; 7617 7618 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 7619 !vdev_is_concrete(vd)) 7620 continue; 7621 7622 svd[svdcount++] = vd; 7623 if (svdcount == SPA_SYNC_MIN_VDEVS) 7624 break; 7625 } 7626 error = vdev_config_sync(svd, svdcount, txg); 7627 } else { 7628 error = vdev_config_sync(rvd->vdev_child, 7629 rvd->vdev_children, txg); 7630 } 7631 7632 if (error == 0) 7633 spa->spa_last_synced_guid = rvd->vdev_guid; 7634 7635 spa_config_exit(spa, SCL_STATE, FTAG); 7636 7637 if (error == 0) 7638 break; 7639 zio_suspend(spa, NULL); 7640 zio_resume_wait(spa); 7641 } 7642 dmu_tx_commit(tx); 7643 7644 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 7645 7646 /* 7647 * Clear the dirty config list. 7648 */ 7649 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 7650 vdev_config_clean(vd); 7651 7652 /* 7653 * Now that the new config has synced transactionally, 7654 * let it become visible to the config cache. 7655 */ 7656 if (spa->spa_config_syncing != NULL) { 7657 spa_config_set(spa, spa->spa_config_syncing); 7658 spa->spa_config_txg = txg; 7659 spa->spa_config_syncing = NULL; 7660 } 7661 7662 dsl_pool_sync_done(dp, txg); 7663 7664 mutex_enter(&spa->spa_alloc_lock); 7665 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7666 mutex_exit(&spa->spa_alloc_lock); 7667 7668 /* 7669 * Update usable space statistics. 7670 */ 7671 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 7672 vdev_sync_done(vd, txg); 7673 7674 spa_update_dspace(spa); 7675 7676 /* 7677 * It had better be the case that we didn't dirty anything 7678 * since vdev_config_sync(). 7679 */ 7680 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 7681 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7682 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 7683 7684 while (zfs_pause_spa_sync) 7685 delay(1); 7686 7687 spa->spa_sync_pass = 0; 7688 7689 /* 7690 * Update the last synced uberblock here. We want to do this at 7691 * the end of spa_sync() so that consumers of spa_last_synced_txg() 7692 * will be guaranteed that all the processing associated with 7693 * that txg has been completed. 7694 */ 7695 spa->spa_ubsync = spa->spa_uberblock; 7696 spa_config_exit(spa, SCL_CONFIG, FTAG); 7697 7698 spa_handle_ignored_writes(spa); 7699 7700 /* 7701 * If any async tasks have been requested, kick them off. 7702 */ 7703 spa_async_dispatch(spa); 7704 } 7705 7706 /* 7707 * Sync all pools. We don't want to hold the namespace lock across these 7708 * operations, so we take a reference on the spa_t and drop the lock during the 7709 * sync. 7710 */ 7711 void 7712 spa_sync_allpools(void) 7713 { 7714 spa_t *spa = NULL; 7715 mutex_enter(&spa_namespace_lock); 7716 while ((spa = spa_next(spa)) != NULL) { 7717 if (spa_state(spa) != POOL_STATE_ACTIVE || 7718 !spa_writeable(spa) || spa_suspended(spa)) 7719 continue; 7720 spa_open_ref(spa, FTAG); 7721 mutex_exit(&spa_namespace_lock); 7722 txg_wait_synced(spa_get_dsl(spa), 0); 7723 mutex_enter(&spa_namespace_lock); 7724 spa_close(spa, FTAG); 7725 } 7726 mutex_exit(&spa_namespace_lock); 7727 } 7728 7729 /* 7730 * ========================================================================== 7731 * Miscellaneous routines 7732 * ========================================================================== 7733 */ 7734 7735 /* 7736 * Remove all pools in the system. 7737 */ 7738 void 7739 spa_evict_all(void) 7740 { 7741 spa_t *spa; 7742 7743 /* 7744 * Remove all cached state. All pools should be closed now, 7745 * so every spa in the AVL tree should be unreferenced. 7746 */ 7747 mutex_enter(&spa_namespace_lock); 7748 while ((spa = spa_next(NULL)) != NULL) { 7749 /* 7750 * Stop async tasks. The async thread may need to detach 7751 * a device that's been replaced, which requires grabbing 7752 * spa_namespace_lock, so we must drop it here. 7753 */ 7754 spa_open_ref(spa, FTAG); 7755 mutex_exit(&spa_namespace_lock); 7756 spa_async_suspend(spa); 7757 mutex_enter(&spa_namespace_lock); 7758 spa_close(spa, FTAG); 7759 7760 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 7761 spa_unload(spa); 7762 spa_deactivate(spa); 7763 } 7764 spa_remove(spa); 7765 } 7766 mutex_exit(&spa_namespace_lock); 7767 } 7768 7769 vdev_t * 7770 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 7771 { 7772 vdev_t *vd; 7773 int i; 7774 7775 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 7776 return (vd); 7777 7778 if (aux) { 7779 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 7780 vd = spa->spa_l2cache.sav_vdevs[i]; 7781 if (vd->vdev_guid == guid) 7782 return (vd); 7783 } 7784 7785 for (i = 0; i < spa->spa_spares.sav_count; i++) { 7786 vd = spa->spa_spares.sav_vdevs[i]; 7787 if (vd->vdev_guid == guid) 7788 return (vd); 7789 } 7790 } 7791 7792 return (NULL); 7793 } 7794 7795 void 7796 spa_upgrade(spa_t *spa, uint64_t version) 7797 { 7798 ASSERT(spa_writeable(spa)); 7799 7800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 7801 7802 /* 7803 * This should only be called for a non-faulted pool, and since a 7804 * future version would result in an unopenable pool, this shouldn't be 7805 * possible. 7806 */ 7807 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 7808 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 7809 7810 spa->spa_uberblock.ub_version = version; 7811 vdev_config_dirty(spa->spa_root_vdev); 7812 7813 spa_config_exit(spa, SCL_ALL, FTAG); 7814 7815 txg_wait_synced(spa_get_dsl(spa), 0); 7816 } 7817 7818 boolean_t 7819 spa_has_spare(spa_t *spa, uint64_t guid) 7820 { 7821 int i; 7822 uint64_t spareguid; 7823 spa_aux_vdev_t *sav = &spa->spa_spares; 7824 7825 for (i = 0; i < sav->sav_count; i++) 7826 if (sav->sav_vdevs[i]->vdev_guid == guid) 7827 return (B_TRUE); 7828 7829 for (i = 0; i < sav->sav_npending; i++) { 7830 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 7831 &spareguid) == 0 && spareguid == guid) 7832 return (B_TRUE); 7833 } 7834 7835 return (B_FALSE); 7836 } 7837 7838 /* 7839 * Check if a pool has an active shared spare device. 7840 * Note: reference count of an active spare is 2, as a spare and as a replace 7841 */ 7842 static boolean_t 7843 spa_has_active_shared_spare(spa_t *spa) 7844 { 7845 int i, refcnt; 7846 uint64_t pool; 7847 spa_aux_vdev_t *sav = &spa->spa_spares; 7848 7849 for (i = 0; i < sav->sav_count; i++) { 7850 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7851 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7852 refcnt > 2) 7853 return (B_TRUE); 7854 } 7855 7856 return (B_FALSE); 7857 } 7858 7859 sysevent_t * 7860 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7861 { 7862 sysevent_t *ev = NULL; 7863 #ifdef _KERNEL 7864 sysevent_attr_list_t *attr = NULL; 7865 sysevent_value_t value; 7866 7867 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7868 SE_SLEEP); 7869 ASSERT(ev != NULL); 7870 7871 value.value_type = SE_DATA_TYPE_STRING; 7872 value.value.sv_string = spa_name(spa); 7873 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7874 goto done; 7875 7876 value.value_type = SE_DATA_TYPE_UINT64; 7877 value.value.sv_uint64 = spa_guid(spa); 7878 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7879 goto done; 7880 7881 if (vd) { 7882 value.value_type = SE_DATA_TYPE_UINT64; 7883 value.value.sv_uint64 = vd->vdev_guid; 7884 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7885 SE_SLEEP) != 0) 7886 goto done; 7887 7888 if (vd->vdev_path) { 7889 value.value_type = SE_DATA_TYPE_STRING; 7890 value.value.sv_string = vd->vdev_path; 7891 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7892 &value, SE_SLEEP) != 0) 7893 goto done; 7894 } 7895 } 7896 7897 if (hist_nvl != NULL) { 7898 fnvlist_merge((nvlist_t *)attr, hist_nvl); 7899 } 7900 7901 if (sysevent_attach_attributes(ev, attr) != 0) 7902 goto done; 7903 attr = NULL; 7904 7905 done: 7906 if (attr) 7907 sysevent_free_attr(attr); 7908 7909 #endif 7910 return (ev); 7911 } 7912 7913 void 7914 spa_event_post(sysevent_t *ev) 7915 { 7916 #ifdef _KERNEL 7917 sysevent_id_t eid; 7918 7919 (void) log_sysevent(ev, SE_SLEEP, &eid); 7920 sysevent_free(ev); 7921 #endif 7922 } 7923 7924 void 7925 spa_event_discard(sysevent_t *ev) 7926 { 7927 #ifdef _KERNEL 7928 sysevent_free(ev); 7929 #endif 7930 } 7931 7932 /* 7933 * Post a sysevent corresponding to the given event. The 'name' must be one of 7934 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7935 * filled in from the spa and (optionally) the vdev and history nvl. This 7936 * doesn't do anything in the userland libzpool, as we don't want consumers to 7937 * misinterpret ztest or zdb as real changes. 7938 */ 7939 void 7940 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7941 { 7942 spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); 7943 } 7944