external/cddl/osnet/dist/uts/common/fs/zfs/spa.c

   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 2008 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 /*
  28  * This file contains all the routines used when modifying on-disk SPA state.
  29  * This includes opening, importing, destroying, exporting a pool, and syncing a
  30  * pool.
  31  */
  32
  33 #include <sys/zfs_context.h>
  34 #include <sys/fm/fs/zfs.h>
  35 #include <sys/spa_impl.h>
  36 #include <sys/zio.h>
  37 #include <sys/zio_checksum.h>
  38 #include <sys/zio_compress.h>
  39 #include <sys/dmu.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/zap.h>
  42 #include <sys/zil.h>
  43 #include <sys/vdev_impl.h>
  44 #include <sys/metaslab.h>
  45 #include <sys/uberblock_impl.h>
  46 #include <sys/txg.h>
  47 #include <sys/avl.h>
  48 #include <sys/dmu_traverse.h>
  49 #include <sys/dmu_objset.h>
  50 #include <sys/unique.h>
  51 #include <sys/dsl_pool.h>
  52 #include <sys/dsl_dataset.h>
  53 #include <sys/dsl_dir.h>
  54 #include <sys/dsl_prop.h>
  55 #include <sys/dsl_synctask.h>
  56 #include <sys/fs/zfs.h>
  57 #include <sys/arc.h>
  58 #include <sys/callb.h>
  59 #include <sys/systeminfo.h>
  60 #include <sys/sunddi.h>
  61 #include <sys/spa_boot.h>
  62
  63 #include "zfs_prop.h"
  64 #include "zfs_comutil.h"
  65
  66 int zio_taskq_threads[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
  67         /*      ISSUE   INTR                                    */
  68         {       1,      1       },      /* ZIO_TYPE_NULL        */
  69         {       1,      8       },      /* ZIO_TYPE_READ        */
  70         {       8,      1       },      /* ZIO_TYPE_WRITE       */
  71         {       1,      1       },      /* ZIO_TYPE_FREE        */
  72         {       1,      1       },      /* ZIO_TYPE_CLAIM       */
  73         {       1,      1       },      /* ZIO_TYPE_IOCTL       */
  74 };
  75
  76 static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
  77 static boolean_t spa_has_active_shared_spare(spa_t *spa);
  78
  79 /*
  80  * ==========================================================================
  81  * SPA properties routines
  82  * ==========================================================================
  83  */
  84
  85 /*
  86  * Add a (source=src, propname=propval) list to an nvlist.
  87  */
  88 static void
  89 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
  90     uint64_t intval, zprop_source_t src)
  91 {
  92         const char *propname = zpool_prop_to_name(prop);
  93         nvlist_t *propval;
  94
  95         VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
  96         VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
  97
  98         if (strval != NULL)
  99                 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
 100         else
 101                 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
 102
 103         VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
 104         nvlist_free(propval);
 105 }
 106
 107 /*
 108  * Get property values from the spa configuration.
 109  */
 110 static void
 111 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
 112 {
 113         uint64_t size = spa_get_space(spa);
 114         uint64_t used = spa_get_alloc(spa);
 115         uint64_t cap, version;
 116         zprop_source_t src = ZPROP_SRC_NONE;
 117         spa_config_dirent_t *dp;
 118
 119         ASSERT(MUTEX_HELD(&spa->spa_props_lock));
 120
 121         /*
 122          * readonly properties
 123          */
 124         spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
 125         spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
 126         spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
 127         spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL, size - used, src);
 128
 129         cap = (size == 0) ? 0 : (used * 100 / size);
 130         spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
 131
 132         spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
 133         spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
 134             spa->spa_root_vdev->vdev_state, src);
 135
 136         /*
 137          * settable properties that are not stored in the pool property object.
 138          */
 139         version = spa_version(spa);
 140         if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
 141                 src = ZPROP_SRC_DEFAULT;
 142         else
 143                 src = ZPROP_SRC_LOCAL;
 144         spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
 145
 146         if (spa->spa_root != NULL)
 147                 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
 148                     0, ZPROP_SRC_LOCAL);
 149
 150         if ((dp = list_head(&spa->spa_config_list)) != NULL) {
 151                 if (dp->scd_path == NULL) {
 152                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 153                             "none", 0, ZPROP_SRC_LOCAL);
 154                 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
 155                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 156                             dp->scd_path, 0, ZPROP_SRC_LOCAL);
 157                 }
 158         }
 159 }
 160
 161 /*
 162  * Get zpool property values.
 163  */
 164 int
 165 spa_prop_get(spa_t *spa, nvlist_t **nvp)
 166 {
 167         zap_cursor_t zc;
 168         zap_attribute_t za;
 169         objset_t *mos = spa->spa_meta_objset;
 170         int err;
 171
 172         VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 173
 174         mutex_enter(&spa->spa_props_lock);
 175
 176         /*
 177          * Get properties from the spa config.
 178          */
 179         spa_prop_get_config(spa, nvp);
 180
 181         /* If no pool property object, no more prop to get. */
 182         if (spa->spa_pool_props_object == 0) {
 183                 mutex_exit(&spa->spa_props_lock);
 184                 return (0);
 185         }
 186
 187         /*
 188          * Get properties from the MOS pool property object.
 189          */
 190         for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
 191             (err = zap_cursor_retrieve(&zc, &za)) == 0;
 192             zap_cursor_advance(&zc)) {
 193                 uint64_t intval = 0;
 194                 char *strval = NULL;
 195                 zprop_source_t src = ZPROP_SRC_DEFAULT;
 196                 zpool_prop_t prop;
 197
 198                 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
 199                         continue;
 200
 201                 switch (za.za_integer_length) {
 202                 case 8:
 203                         /* integer property */
 204                         if (za.za_first_integer !=
 205                             zpool_prop_default_numeric(prop))
 206                                 src = ZPROP_SRC_LOCAL;
 207
 208                         if (prop == ZPOOL_PROP_BOOTFS) {
 209                                 dsl_pool_t *dp;
 210                                 dsl_dataset_t *ds = NULL;
 211
 212                                 dp = spa_get_dsl(spa);
 213                                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 214                                 if (err = dsl_dataset_hold_obj(dp,
 215                                     za.za_first_integer, FTAG, &ds)) {
 216                                         rw_exit(&dp->dp_config_rwlock);
 217                                         break;
 218                                 }
 219
 220                                 strval = kmem_alloc(
 221                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
 222                                     KM_SLEEP);
 223                                 dsl_dataset_name(ds, strval);
 224                                 dsl_dataset_rele(ds, FTAG);
 225                                 rw_exit(&dp->dp_config_rwlock);
 226                         } else {
 227                                 strval = NULL;
 228                                 intval = za.za_first_integer;
 229                         }
 230
 231                         spa_prop_add_list(*nvp, prop, strval, intval, src);
 232
 233                         if (strval != NULL)
 234                                 kmem_free(strval,
 235                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
 236
 237                         break;
 238
 239                 case 1:
 240                         /* string property */
 241                         strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
 242                         err = zap_lookup(mos, spa->spa_pool_props_object,
 243                             za.za_name, 1, za.za_num_integers, strval);
 244                         if (err) {
 245                                 kmem_free(strval, za.za_num_integers);
 246                                 break;
 247                         }
 248                         spa_prop_add_list(*nvp, prop, strval, 0, src);
 249                         kmem_free(strval, za.za_num_integers);
 250                         break;
 251
 252                 default:
 253                         break;
 254                 }
 255         }
 256         zap_cursor_fini(&zc);
 257         mutex_exit(&spa->spa_props_lock);
 258 out:
 259         if (err && err != ENOENT) {
 260                 nvlist_free(*nvp);
 261                 *nvp = NULL;
 262                 return (err);
 263         }
 264
 265         return (0);
 266 }
 267
 268 /*
 269  * Validate the given pool properties nvlist and modify the list
 270  * for the property values to be set.
 271  */
 272 static int
 273 spa_prop_validate(spa_t *spa, nvlist_t *props)
 274 {
 275         nvpair_t *elem;
 276         int error = 0, reset_bootfs = 0;
 277         uint64_t objnum;
 278
 279         elem = NULL;
 280         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 281                 zpool_prop_t prop;
 282                 char *propname, *strval;
 283                 uint64_t intval;
 284                 objset_t *os;
 285                 char *slash;
 286
 287                 propname = nvpair_name(elem);
 288
 289                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
 290                         return (EINVAL);
 291
 292                 switch (prop) {
 293                 case ZPOOL_PROP_VERSION:
 294                         error = nvpair_value_uint64(elem, &intval);
 295                         if (!error &&
 296                             (intval < spa_version(spa) || intval > SPA_VERSION))
 297                                 error = EINVAL;
 298                         break;
 299
 300                 case ZPOOL_PROP_DELEGATION:
 301                 case ZPOOL_PROP_AUTOREPLACE:
 302                 case ZPOOL_PROP_LISTSNAPS:
 303                         error = nvpair_value_uint64(elem, &intval);
 304                         if (!error && intval > 1)
 305                                 error = EINVAL;
 306                         break;
 307
 308                 case ZPOOL_PROP_BOOTFS:
 309                         if (spa_version(spa) < SPA_VERSION_BOOTFS) {
 310                                 error = ENOTSUP;
 311                                 break;
 312                         }
 313
 314                         /*
 315                          * Make sure the vdev config is bootable
 316                          */
 317                         if (!vdev_is_bootable(spa->spa_root_vdev)) {
 318                                 error = ENOTSUP;
 319                                 break;
 320                         }
 321
 322                         reset_bootfs = 1;
 323
 324                         error = nvpair_value_string(elem, &strval);
 325
 326                         if (!error) {
 327                                 uint64_t compress;
 328
 329                                 if (strval == NULL || strval[0] == '\0') {
 330                                         objnum = zpool_prop_default_numeric(
 331                                             ZPOOL_PROP_BOOTFS);
 332                                         break;
 333                                 }
 334
 335                                 if (error = dmu_objset_open(strval, DMU_OST_ZFS,
 336                                     DS_MODE_USER | DS_MODE_READONLY, &os))
 337                                         break;
 338
 339                                 /* We don't support gzip bootable datasets */
 340                                 if ((error = dsl_prop_get_integer(strval,
 341                                     zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 342                                     &compress, NULL)) == 0 &&
 343                                     !BOOTFS_COMPRESS_VALID(compress)) {
 344                                         error = ENOTSUP;
 345                                 } else {
 346                                         objnum = dmu_objset_id(os);
 347                                 }
 348                                 dmu_objset_close(os);
 349                         }
 350                         break;
 351
 352                 case ZPOOL_PROP_FAILUREMODE:
 353                         error = nvpair_value_uint64(elem, &intval);
 354                         if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
 355                             intval > ZIO_FAILURE_MODE_PANIC))
 356                                 error = EINVAL;
 357
 358                         /*
 359                          * This is a special case which only occurs when
 360                          * the pool has completely failed. This allows
 361                          * the user to change the in-core failmode property
 362                          * without syncing it out to disk (I/Os might
 363                          * currently be blocked). We do this by returning
 364                          * EIO to the caller (spa_prop_set) to trick it
 365                          * into thinking we encountered a property validation
 366                          * error.
 367                          */
 368                         if (!error && spa_suspended(spa)) {
 369                                 spa->spa_failmode = intval;
 370                                 error = EIO;
 371                         }
 372                         break;
 373
 374                 case ZPOOL_PROP_CACHEFILE:
 375                         if ((error = nvpair_value_string(elem, &strval)) != 0)
 376                                 break;
 377
 378                         if (strval[0] == '\0')
 379                                 break;
 380
 381                         if (strcmp(strval, "none") == 0)
 382                                 break;
 383
 384                         if (strval[0] != '/') {
 385                                 error = EINVAL;
 386                                 break;
 387                         }
 388
 389                         slash = strrchr(strval, '/');
 390                         ASSERT(slash != NULL);
 391
 392                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 393                             strcmp(slash, "/..") == 0)
 394                                 error = EINVAL;
 395                         break;
 396                 }
 397
 398                 if (error)
 399                         break;
 400         }
 401
 402         if (!error && reset_bootfs) {
 403                 error = nvlist_remove(props,
 404                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
 405
 406                 if (!error) {
 407                         error = nvlist_add_uint64(props,
 408                             zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
 409                 }
 410         }
 411
 412         return (error);
 413 }
 414
 415 int
 416 spa_prop_set(spa_t *spa, nvlist_t *nvp)
 417 {
 418         int error;
 419
 420         if ((error = spa_prop_validate(spa, nvp)) != 0)
 421                 return (error);
 422
 423         return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
 424             spa, nvp, 3));
 425 }
 426
 427 /*
 428  * If the bootfs property value is dsobj, clear it.
 429  */
 430 void
 431 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
 432 {
 433         if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
 434                 VERIFY(zap_remove(spa->spa_meta_objset,
 435                     spa->spa_pool_props_object,
 436                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
 437                 spa->spa_bootfs = 0;
 438         }
 439 }
 440
 441 /*
 442  * ==========================================================================
 443  * SPA state manipulation (open/create/destroy/import/export)
 444  * ==========================================================================
 445  */
 446
 447 static int
 448 spa_error_entry_compare(const void *a, const void *b)
 449 {
 450         spa_error_entry_t *sa = (spa_error_entry_t *)a;
 451         spa_error_entry_t *sb = (spa_error_entry_t *)b;
 452         int ret;
 453
 454         ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
 455             sizeof (zbookmark_t));
 456
 457         if (ret < 0)
 458                 return (-1);
 459         else if (ret > 0)
 460                 return (1);
 461         else
 462                 return (0);
 463 }
 464
 465 /*
 466  * Utility function which retrieves copies of the current logs and
 467  * re-initializes them in the process.
 468  */
 469 void
 470 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
 471 {
 472         ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
 473
 474         bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
 475         bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
 476
 477         avl_create(&spa->spa_errlist_scrub,
 478             spa_error_entry_compare, sizeof (spa_error_entry_t),
 479             offsetof(spa_error_entry_t, se_avl));
 480         avl_create(&spa->spa_errlist_last,
 481             spa_error_entry_compare, sizeof (spa_error_entry_t),
 482             offsetof(spa_error_entry_t, se_avl));
 483 }
 484
 485 /*
 486  * Activate an uninitialized pool.
 487  */
 488 static void
 489 spa_activate(spa_t *spa)
 490 {
 491         ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
 492
 493         spa->spa_state = POOL_STATE_ACTIVE;
 494
 495         spa->spa_normal_class = metaslab_class_create();
 496         spa->spa_log_class = metaslab_class_create();
 497
 498         for (int t = 0; t < ZIO_TYPES; t++) {
 499                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
 500                         spa->spa_zio_taskq[t][q] = taskq_create("spa_zio",
 501                             zio_taskq_threads[t][q], maxclsyspri, 50,
 502                             INT_MAX, TASKQ_PREPOPULATE);
 503                 }
 504         }
 505
 506         list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
 507             offsetof(vdev_t, vdev_config_dirty_node));
 508         list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
 509             offsetof(vdev_t, vdev_state_dirty_node));
 510
 511         txg_list_create(&spa->spa_vdev_txg_list,
 512             offsetof(struct vdev, vdev_txg_node));
 513
 514         avl_create(&spa->spa_errlist_scrub,
 515             spa_error_entry_compare, sizeof (spa_error_entry_t),
 516             offsetof(spa_error_entry_t, se_avl));
 517         avl_create(&spa->spa_errlist_last,
 518             spa_error_entry_compare, sizeof (spa_error_entry_t),
 519             offsetof(spa_error_entry_t, se_avl));
 520 }
 521
 522 /*
 523  * Opposite of spa_activate().
 524  */
 525 static void
 526 spa_deactivate(spa_t *spa)
 527 {
 528         ASSERT(spa->spa_sync_on == B_FALSE);
 529         ASSERT(spa->spa_dsl_pool == NULL);
 530         ASSERT(spa->spa_root_vdev == NULL);
 531
 532         ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
 533
 534         txg_list_destroy(&spa->spa_vdev_txg_list);
 535
 536         list_destroy(&spa->spa_config_dirty_list);
 537         list_destroy(&spa->spa_state_dirty_list);
 538
 539         for (int t = 0; t < ZIO_TYPES; t++) {
 540                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
 541                         taskq_destroy(spa->spa_zio_taskq[t][q]);
 542                         spa->spa_zio_taskq[t][q] = NULL;
 543                 }
 544         }
 545
 546         metaslab_class_destroy(spa->spa_normal_class);
 547         spa->spa_normal_class = NULL;
 548
 549         metaslab_class_destroy(spa->spa_log_class);
 550         spa->spa_log_class = NULL;
 551
 552         /*
 553          * If this was part of an import or the open otherwise failed, we may
 554          * still have errors left in the queues.  Empty them just in case.
 555          */
 556         spa_errlog_drain(spa);
 557
 558         avl_destroy(&spa->spa_errlist_scrub);
 559         avl_destroy(&spa->spa_errlist_last);
 560
 561         spa->spa_state = POOL_STATE_UNINITIALIZED;
 562 }
 563
 564 /*
 565  * Verify a pool configuration, and construct the vdev tree appropriately.  This
 566  * will create all the necessary vdevs in the appropriate layout, with each vdev
 567  * in the CLOSED state.  This will prep the pool before open/creation/import.
 568  * All vdev validation is done by the vdev_alloc() routine.
 569  */
 570 static int
 571 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
 572     uint_t id, int atype)
 573 {
 574         nvlist_t **child;
 575         uint_t c, children;
 576         int error;
 577
 578         if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
 579                 return (error);
 580
 581         if ((*vdp)->vdev_ops->vdev_op_leaf)
 582                 return (0);
 583
 584         error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 585             &child, &children);
 586
 587         if (error == ENOENT)
 588                 return (0);
 589
 590         if (error) {
 591                 vdev_free(*vdp);
 592                 *vdp = NULL;
 593                 return (EINVAL);
 594         }
 595
 596         for (c = 0; c < children; c++) {
 597                 vdev_t *vd;
 598                 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
 599                     atype)) != 0) {
 600                         vdev_free(*vdp);
 601                         *vdp = NULL;
 602                         return (error);
 603                 }
 604         }
 605
 606         ASSERT(*vdp != NULL);
 607
 608         return (0);
 609 }
 610
 611 /*
 612  * Opposite of spa_load().
 613  */
 614 static void
 615 spa_unload(spa_t *spa)
 616 {
 617         int i;
 618
 619         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 620
 621         /*
 622          * Stop async tasks.
 623          */
 624         spa_async_suspend(spa);
 625
 626         /*
 627          * Stop syncing.
 628          */
 629         if (spa->spa_sync_on) {
 630                 txg_sync_stop(spa->spa_dsl_pool);
 631                 spa->spa_sync_on = B_FALSE;
 632         }
 633
 634         /*
 635          * Wait for any outstanding async I/O to complete.
 636          */
 637         mutex_enter(&spa->spa_async_root_lock);
 638         while (spa->spa_async_root_count != 0)
 639                 cv_wait(&spa->spa_async_root_cv, &spa->spa_async_root_lock);
 640         mutex_exit(&spa->spa_async_root_lock);
 641
 642         /*
 643          * Drop and purge level 2 cache
 644          */
 645         spa_l2cache_drop(spa);
 646
 647         /*
 648          * Close the dsl pool.
 649          */
 650         if (spa->spa_dsl_pool) {
 651                 dsl_pool_close(spa->spa_dsl_pool);
 652                 spa->spa_dsl_pool = NULL;
 653         }
 654
 655         /*
 656          * Close all vdevs.
 657          */
 658         if (spa->spa_root_vdev)
 659                 vdev_free(spa->spa_root_vdev);
 660         ASSERT(spa->spa_root_vdev == NULL);
 661
 662         for (i = 0; i < spa->spa_spares.sav_count; i++)
 663                 vdev_free(spa->spa_spares.sav_vdevs[i]);
 664         if (spa->spa_spares.sav_vdevs) {
 665                 kmem_free(spa->spa_spares.sav_vdevs,
 666                     spa->spa_spares.sav_count * sizeof (void *));
 667                 spa->spa_spares.sav_vdevs = NULL;
 668         }
 669         if (spa->spa_spares.sav_config) {
 670                 nvlist_free(spa->spa_spares.sav_config);
 671                 spa->spa_spares.sav_config = NULL;
 672         }
 673         spa->spa_spares.sav_count = 0;
 674
 675         for (i = 0; i < spa->spa_l2cache.sav_count; i++)
 676                 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
 677         if (spa->spa_l2cache.sav_vdevs) {
 678                 kmem_free(spa->spa_l2cache.sav_vdevs,
 679                     spa->spa_l2cache.sav_count * sizeof (void *));
 680                 spa->spa_l2cache.sav_vdevs = NULL;
 681         }
 682         if (spa->spa_l2cache.sav_config) {
 683                 nvlist_free(spa->spa_l2cache.sav_config);
 684                 spa->spa_l2cache.sav_config = NULL;
 685         }
 686         spa->spa_l2cache.sav_count = 0;
 687
 688         spa->spa_async_suspended = 0;
 689 }
 690
 691 /*
 692  * Load (or re-load) the current list of vdevs describing the active spares for
 693  * this pool.  When this is called, we have some form of basic information in
 694  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
 695  * then re-generate a more complete list including status information.
 696  */
 697 static void
 698 spa_load_spares(spa_t *spa)
 699 {
 700         nvlist_t **spares;
 701         uint_t nspares;
 702         int i;
 703         vdev_t *vd, *tvd;
 704
 705         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 706
 707         /*
 708          * First, close and free any existing spare vdevs.
 709          */
 710         for (i = 0; i < spa->spa_spares.sav_count; i++) {
 711                 vd = spa->spa_spares.sav_vdevs[i];
 712
 713                 /* Undo the call to spa_activate() below */
 714                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
 715                     B_FALSE)) != NULL && tvd->vdev_isspare)
 716                         spa_spare_remove(tvd);
 717                 vdev_close(vd);
 718                 vdev_free(vd);
 719         }
 720
 721         if (spa->spa_spares.sav_vdevs)
 722                 kmem_free(spa->spa_spares.sav_vdevs,
 723                     spa->spa_spares.sav_count * sizeof (void *));
 724
 725         if (spa->spa_spares.sav_config == NULL)
 726                 nspares = 0;
 727         else
 728                 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
 729                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
 730
 731         spa->spa_spares.sav_count = (int)nspares;
 732         spa->spa_spares.sav_vdevs = NULL;
 733
 734         if (nspares == 0)
 735                 return;
 736
 737         /*
 738          * Construct the array of vdevs, opening them to get status in the
 739          * process.   For each spare, there is potentially two different vdev_t
 740          * structures associated with it: one in the list of spares (used only
 741          * for basic validation purposes) and one in the active vdev
 742          * configuration (if it's spared in).  During this phase we open and
 743          * validate each vdev on the spare list.  If the vdev also exists in the
 744          * active configuration, then we also mark this vdev as an active spare.
 745          */
 746         spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
 747             KM_SLEEP);
 748         for (i = 0; i < spa->spa_spares.sav_count; i++) {
 749                 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
 750                     VDEV_ALLOC_SPARE) == 0);
 751                 ASSERT(vd != NULL);
 752
 753                 spa->spa_spares.sav_vdevs[i] = vd;
 754
 755                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
 756                     B_FALSE)) != NULL) {
 757                         if (!tvd->vdev_isspare)
 758                                 spa_spare_add(tvd);
 759
 760                         /*
 761                          * We only mark the spare active if we were successfully
 762                          * able to load the vdev.  Otherwise, importing a pool
 763                          * with a bad active spare would result in strange
 764                          * behavior, because multiple pool would think the spare
 765                          * is actively in use.
 766                          *
 767                          * There is a vulnerability here to an equally bizarre
 768                          * circumstance, where a dead active spare is later
 769                          * brought back to life (onlined or otherwise).  Given
 770                          * the rarity of this scenario, and the extra complexity
 771                          * it adds, we ignore the possibility.
 772                          */
 773                         if (!vdev_is_dead(tvd))
 774                                 spa_spare_activate(tvd);
 775                 }
 776
 777                 vd->vdev_top = vd;
 778
 779                 if (vdev_open(vd) != 0)
 780                         continue;
 781
 782                 if (vdev_validate_aux(vd) == 0)
 783                         spa_spare_add(vd);
 784         }
 785
 786         /*
 787          * Recompute the stashed list of spares, with status information
 788          * this time.
 789          */
 790         VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
 791             DATA_TYPE_NVLIST_ARRAY) == 0);
 792
 793         spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
 794             KM_SLEEP);
 795         for (i = 0; i < spa->spa_spares.sav_count; i++)
 796                 spares[i] = vdev_config_generate(spa,
 797                     spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
 798         VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
 799             ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
 800         for (i = 0; i < spa->spa_spares.sav_count; i++)
 801                 nvlist_free(spares[i]);
 802         kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
 803 }
 804
 805 /*
 806  * Load (or re-load) the current list of vdevs describing the active l2cache for
 807  * this pool.  When this is called, we have some form of basic information in
 808  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
 809  * then re-generate a more complete list including status information.
 810  * Devices which are already active have their details maintained, and are
 811  * not re-opened.
 812  */
 813 static void
 814 spa_load_l2cache(spa_t *spa)
 815 {
 816         nvlist_t **l2cache;
 817         uint_t nl2cache;
 818         int i, j, oldnvdevs;
 819         uint64_t guid, size;
 820         vdev_t *vd, **oldvdevs, **newvdevs;
 821         spa_aux_vdev_t *sav = &spa->spa_l2cache;
 822
 823         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 824
 825         if (sav->sav_config != NULL) {
 826                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
 827                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
 828                 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
 829         } else {
 830                 nl2cache = 0;
 831         }
 832
 833         oldvdevs = sav->sav_vdevs;
 834         oldnvdevs = sav->sav_count;
 835         sav->sav_vdevs = NULL;
 836         sav->sav_count = 0;
 837
 838         /*
 839          * Process new nvlist of vdevs.
 840          */
 841         for (i = 0; i < nl2cache; i++) {
 842                 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
 843                     &guid) == 0);
 844
 845                 newvdevs[i] = NULL;
 846                 for (j = 0; j < oldnvdevs; j++) {
 847                         vd = oldvdevs[j];
 848                         if (vd != NULL && guid == vd->vdev_guid) {
 849                                 /*
 850                                  * Retain previous vdev for add/remove ops.
 851                                  */
 852                                 newvdevs[i] = vd;
 853                                 oldvdevs[j] = NULL;
 854                                 break;
 855                         }
 856                 }
 857
 858                 if (newvdevs[i] == NULL) {
 859                         /*
 860                          * Create new vdev
 861                          */
 862                         VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
 863                             VDEV_ALLOC_L2CACHE) == 0);
 864                         ASSERT(vd != NULL);
 865                         newvdevs[i] = vd;
 866
 867                         /*
 868                          * Commit this vdev as an l2cache device,
 869                          * even if it fails to open.
 870                          */
 871                         spa_l2cache_add(vd);
 872
 873                         vd->vdev_top = vd;
 874                         vd->vdev_aux = sav;
 875
 876                         spa_l2cache_activate(vd);
 877
 878                         if (vdev_open(vd) != 0)
 879                                 continue;
 880
 881                         (void) vdev_validate_aux(vd);
 882
 883                         if (!vdev_is_dead(vd)) {
 884                                 size = vdev_get_rsize(vd);
 885                                 l2arc_add_vdev(spa, vd,
 886                                     VDEV_LABEL_START_SIZE,
 887                                     size - VDEV_LABEL_START_SIZE);
 888                         }
 889                 }
 890         }
 891
 892         /*
 893          * Purge vdevs that were dropped
 894          */
 895         for (i = 0; i < oldnvdevs; i++) {
 896                 uint64_t pool;
 897
 898                 vd = oldvdevs[i];
 899                 if (vd != NULL) {
 900                         if ((spa_mode & FWRITE) &&
 901                             spa_l2cache_exists(vd->vdev_guid, &pool) &&
 902                             pool != 0ULL &&
 903                             l2arc_vdev_present(vd)) {
 904                                 l2arc_remove_vdev(vd);
 905                         }
 906                         (void) vdev_close(vd);
 907                         spa_l2cache_remove(vd);
 908                 }
 909         }
 910
 911         if (oldvdevs)
 912                 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
 913
 914         if (sav->sav_config == NULL)
 915                 goto out;
 916
 917         sav->sav_vdevs = newvdevs;
 918         sav->sav_count = (int)nl2cache;
 919
 920         /*
 921          * Recompute the stashed list of l2cache devices, with status
 922          * information this time.
 923          */
 924         VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
 925             DATA_TYPE_NVLIST_ARRAY) == 0);
 926
 927         l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
 928         for (i = 0; i < sav->sav_count; i++)
 929                 l2cache[i] = vdev_config_generate(spa,
 930                     sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
 931         VERIFY(nvlist_add_nvlist_array(sav->sav_config,
 932             ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
 933 out:
 934         for (i = 0; i < sav->sav_count; i++)
 935                 nvlist_free(l2cache[i]);
 936         if (sav->sav_count)
 937                 kmem_free(l2cache, sav->sav_count * sizeof (void *));
 938 }
 939
 940 static int
 941 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
 942 {
 943         dmu_buf_t *db;
 944         char *packed = NULL;
 945         size_t nvsize = 0;
 946         int error;
 947         *value = NULL;
 948
 949         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
 950         nvsize = *(uint64_t *)db->db_data;
 951         dmu_buf_rele(db, FTAG);
 952
 953         packed = kmem_alloc(nvsize, KM_SLEEP);
 954         error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
 955         if (error == 0)
 956                 error = nvlist_unpack(packed, nvsize, value, 0);
 957         kmem_free(packed, nvsize);
 958
 959         return (error);
 960 }
 961
 962 /*
 963  * Checks to see if the given vdev could not be opened, in which case we post a
 964  * sysevent to notify the autoreplace code that the device has been removed.
 965  */
 966 static void
 967 spa_check_removed(vdev_t *vd)
 968 {
 969         int c;
 970
 971         for (c = 0; c < vd->vdev_children; c++)
 972                 spa_check_removed(vd->vdev_child[c]);
 973
 974         if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
 975                 zfs_post_autoreplace(vd->vdev_spa, vd);
 976                 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
 977         }
 978 }
 979
 980 /*
 981  * Check for missing log devices
 982  */
 983 int
 984 spa_check_logs(spa_t *spa)
 985 {
 986         switch (spa->spa_log_state) {
 987         case SPA_LOG_MISSING:
 988                 /* need to recheck in case slog has been restored */
 989         case SPA_LOG_UNKNOWN:
 990                 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
 991                     DS_FIND_CHILDREN)) {
 992                         spa->spa_log_state = SPA_LOG_MISSING;
 993                         return (1);
 994                 }
 995                 break;
 996
 997         case SPA_LOG_CLEAR:
 998                 (void) dmu_objset_find(spa->spa_name, zil_clear_log_chain, NULL,
 999                     DS_FIND_CHILDREN);
1000                 break;
1001         }
1002         spa->spa_log_state = SPA_LOG_GOOD;
1003         return (0);
1004 }
1005
1006 /*
1007  * Load an existing storage pool, using the pool's builtin spa_config as a
1008  * source of configuration information.
1009  */
1010 static int
1011 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1012 {
1013         int error = 0;
1014         nvlist_t *nvroot = NULL;
1015         vdev_t *rvd;
1016         uberblock_t *ub = &spa->spa_uberblock;
1017         uint64_t config_cache_txg = spa->spa_config_txg;
1018         uint64_t pool_guid;
1019         uint64_t version;
1020         uint64_t autoreplace = 0;
1021         char *ereport = FM_EREPORT_ZFS_POOL;
1022
1023         ASSERT(MUTEX_HELD(&spa_namespace_lock));
1024
1025         spa->spa_load_state = state;
1026
1027         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1028             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1029                 error = EINVAL;
1030                 goto out;
1031         }
1032
1033         /*
1034          * Versioning wasn't explicitly added to the label until later, so if
1035          * it's not present treat it as the initial version.
1036          */
1037         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1038                 version = SPA_VERSION_INITIAL;
1039
1040         (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1041             &spa->spa_config_txg);
1042
1043         if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1044             spa_guid_exists(pool_guid, 0)) {
1045                 error = EEXIST;
1046                 goto out;
1047         }
1048
1049         spa->spa_load_guid = pool_guid;
1050
1051         /*
1052          * Parse the configuration into a vdev tree.  We explicitly set the
1053          * value that will be returned by spa_version() since parsing the
1054          * configuration requires knowing the version number.
1055          */
1056         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1057         spa->spa_ubsync.ub_version = version;
1058         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1059         spa_config_exit(spa, SCL_ALL, FTAG);
1060
1061         if (error != 0)
1062                 goto out;
1063
1064         ASSERT(spa->spa_root_vdev == rvd);
1065         ASSERT(spa_guid(spa) == pool_guid);
1066
1067         /*
1068          * Try to open all vdevs, loading each label in the process.
1069          */
1070         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1071         error = vdev_open(rvd);
1072         spa_config_exit(spa, SCL_ALL, FTAG);
1073         if (error != 0)
1074                 goto out;
1075
1076         /*
1077          * Validate the labels for all leaf vdevs.  We need to grab the config
1078          * lock because all label I/O is done with ZIO_FLAG_CONFIG_WRITER.
1079          */
1080         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1081         error = vdev_validate(rvd);
1082         spa_config_exit(spa, SCL_ALL, FTAG);
1083
1084         if (error != 0)
1085                 goto out;
1086
1087         if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1088                 error = ENXIO;
1089                 goto out;
1090         }
1091
1092         /*
1093          * Find the best uberblock.
1094          */
1095         vdev_uberblock_load(NULL, rvd, ub);
1096
1097         /*
1098          * If we weren't able to find a single valid uberblock, return failure.
1099          */
1100         if (ub->ub_txg == 0) {
1101                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1102                     VDEV_AUX_CORRUPT_DATA);
1103                 error = ENXIO;
1104                 goto out;
1105         }
1106
1107         /*
1108          * If the pool is newer than the code, we can't open it.
1109          */
1110         if (ub->ub_version > SPA_VERSION) {
1111                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1112                     VDEV_AUX_VERSION_NEWER);
1113                 error = ENOTSUP;
1114                 goto out;
1115         }
1116
1117         /*
1118          * If the vdev guid sum doesn't match the uberblock, we have an
1119          * incomplete configuration.
1120          */
1121         if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1122                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1123                     VDEV_AUX_BAD_GUID_SUM);
1124                 error = ENXIO;
1125                 goto out;
1126         }
1127
1128         /*
1129          * Initialize internal SPA structures.
1130          */
1131         spa->spa_state = POOL_STATE_ACTIVE;
1132         spa->spa_ubsync = spa->spa_uberblock;
1133         spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1134         error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1135         if (error) {
1136                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1137                     VDEV_AUX_CORRUPT_DATA);
1138                 goto out;
1139         }
1140         spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1141
1142         if (zap_lookup(spa->spa_meta_objset,
1143             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1144             sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1145                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1146                     VDEV_AUX_CORRUPT_DATA);
1147                 error = EIO;
1148                 goto out;
1149         }
1150
1151         if (!mosconfig) {
1152                 nvlist_t *newconfig;
1153                 uint64_t hostid;
1154
1155                 if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1156                         vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1157                             VDEV_AUX_CORRUPT_DATA);
1158                         error = EIO;
1159                         goto out;
1160                 }
1161
1162                 if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1163                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1164                         char *hostname;
1165                         unsigned long myhostid = 0;
1166
1167                         VERIFY(nvlist_lookup_string(newconfig,
1168                             ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1169
1170                         (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1171                         if (hostid != 0 && myhostid != 0 &&
1172                             (unsigned long)hostid != myhostid) {
1173                                 cmn_err(CE_WARN, "pool '%s' could not be "
1174                                     "loaded as it was last accessed by "
1175                                     "another system (host: %s hostid: 0x%lx). "
1176                                     "See: http://www.sun.com/msg/ZFS-8000-EY",
1177                                     spa_name(spa), hostname,
1178                                     (unsigned long)hostid);
1179                                 error = EBADF;
1180                                 goto out;
1181                         }
1182                 }
1183
1184                 spa_config_set(spa, newconfig);
1185                 spa_unload(spa);
1186                 spa_deactivate(spa);
1187                 spa_activate(spa);
1188
1189                 return (spa_load(spa, newconfig, state, B_TRUE));
1190         }
1191
1192         if (zap_lookup(spa->spa_meta_objset,
1193             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1194             sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1195                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1196                     VDEV_AUX_CORRUPT_DATA);
1197                 error = EIO;
1198                 goto out;
1199         }
1200
1201         /*
1202          * Load the bit that tells us to use the new accounting function
1203          * (raid-z deflation).  If we have an older pool, this will not
1204          * be present.
1205          */
1206         error = zap_lookup(spa->spa_meta_objset,
1207             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1208             sizeof (uint64_t), 1, &spa->spa_deflate);
1209         if (error != 0 && error != ENOENT) {
1210                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1211                     VDEV_AUX_CORRUPT_DATA);
1212                 error = EIO;
1213                 goto out;
1214         }
1215
1216         /*
1217          * Load the persistent error log.  If we have an older pool, this will
1218          * not be present.
1219          */
1220         error = zap_lookup(spa->spa_meta_objset,
1221             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1222             sizeof (uint64_t), 1, &spa->spa_errlog_last);
1223         if (error != 0 && error != ENOENT) {
1224                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1225                     VDEV_AUX_CORRUPT_DATA);
1226                 error = EIO;
1227                 goto out;
1228         }
1229
1230         error = zap_lookup(spa->spa_meta_objset,
1231             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1232             sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1233         if (error != 0 && error != ENOENT) {
1234                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1235                     VDEV_AUX_CORRUPT_DATA);
1236                 error = EIO;
1237                 goto out;
1238         }
1239
1240         /*
1241          * Load the history object.  If we have an older pool, this
1242          * will not be present.
1243          */
1244         error = zap_lookup(spa->spa_meta_objset,
1245             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1246             sizeof (uint64_t), 1, &spa->spa_history);
1247         if (error != 0 && error != ENOENT) {
1248                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1249                     VDEV_AUX_CORRUPT_DATA);
1250                 error = EIO;
1251                 goto out;
1252         }
1253
1254         /*
1255          * Load any hot spares for this pool.
1256          */
1257         error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1258             DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1259         if (error != 0 && error != ENOENT) {
1260                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1261                     VDEV_AUX_CORRUPT_DATA);
1262                 error = EIO;
1263                 goto out;
1264         }
1265         if (error == 0) {
1266                 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1267                 if (load_nvlist(spa, spa->spa_spares.sav_object,
1268                     &spa->spa_spares.sav_config) != 0) {
1269                         vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1270                             VDEV_AUX_CORRUPT_DATA);
1271                         error = EIO;
1272                         goto out;
1273                 }
1274
1275                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1276                 spa_load_spares(spa);
1277                 spa_config_exit(spa, SCL_ALL, FTAG);
1278         }
1279
1280         /*
1281          * Load any level 2 ARC devices for this pool.
1282          */
1283         error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1284             DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1285             &spa->spa_l2cache.sav_object);
1286         if (error != 0 && error != ENOENT) {
1287                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1288                     VDEV_AUX_CORRUPT_DATA);
1289                 error = EIO;
1290                 goto out;
1291         }
1292         if (error == 0) {
1293                 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1294                 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1295                     &spa->spa_l2cache.sav_config) != 0) {
1296                         vdev_set_state(rvd, B_TRUE,
1297                             VDEV_STATE_CANT_OPEN,
1298                             VDEV_AUX_CORRUPT_DATA);
1299                         error = EIO;
1300                         goto out;
1301                 }
1302
1303                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1304                 spa_load_l2cache(spa);
1305                 spa_config_exit(spa, SCL_ALL, FTAG);
1306         }
1307
1308         if (spa_check_logs(spa)) {
1309                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1310                     VDEV_AUX_BAD_LOG);
1311                 error = ENXIO;
1312                 ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1313                 goto out;
1314         }
1315
1316
1317         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1318
1319         error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1320             DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1321
1322         if (error && error != ENOENT) {
1323                 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1324                     VDEV_AUX_CORRUPT_DATA);
1325                 error = EIO;
1326                 goto out;
1327         }
1328
1329         if (error == 0) {
1330                 (void) zap_lookup(spa->spa_meta_objset,
1331                     spa->spa_pool_props_object,
1332                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1333                     sizeof (uint64_t), 1, &spa->spa_bootfs);
1334                 (void) zap_lookup(spa->spa_meta_objset,
1335                     spa->spa_pool_props_object,
1336                     zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1337                     sizeof (uint64_t), 1, &autoreplace);
1338                 (void) zap_lookup(spa->spa_meta_objset,
1339                     spa->spa_pool_props_object,
1340                     zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1341                     sizeof (uint64_t), 1, &spa->spa_delegation);
1342                 (void) zap_lookup(spa->spa_meta_objset,
1343                     spa->spa_pool_props_object,
1344                     zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1345                     sizeof (uint64_t), 1, &spa->spa_failmode);
1346         }
1347
1348         /*
1349          * If the 'autoreplace' property is set, then post a resource notifying
1350          * the ZFS DE that it should not issue any faults for unopenable
1351          * devices.  We also iterate over the vdevs, and post a sysevent for any
1352          * unopenable vdevs so that the normal autoreplace handler can take
1353          * over.
1354          */
1355         if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1356                 spa_check_removed(spa->spa_root_vdev);
1357
1358         /*
1359          * Load the vdev state for all toplevel vdevs.
1360          */
1361         vdev_load(rvd);
1362
1363         /*
1364          * Propagate the leaf DTLs we just loaded all the way up the tree.
1365          */
1366         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1367         vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1368         spa_config_exit(spa, SCL_ALL, FTAG);
1369
1370         /*
1371          * Check the state of the root vdev.  If it can't be opened, it
1372          * indicates one or more toplevel vdevs are faulted.
1373          */
1374         if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1375                 error = ENXIO;
1376                 goto out;
1377         }
1378
1379         if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
1380                 dmu_tx_t *tx;
1381                 int need_update = B_FALSE;
1382                 int c;
1383
1384                 /*
1385                  * Claim log blocks that haven't been committed yet.
1386                  * This must all happen in a single txg.
1387                  */
1388                 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1389                     spa_first_txg(spa));
1390                 (void) dmu_objset_find(spa_name(spa),
1391                     zil_claim, tx, DS_FIND_CHILDREN);
1392                 dmu_tx_commit(tx);
1393
1394                 spa->spa_sync_on = B_TRUE;
1395                 txg_sync_start(spa->spa_dsl_pool);
1396
1397                 /*
1398                  * Wait for all claims to sync.
1399                  */
1400                 txg_wait_synced(spa->spa_dsl_pool, 0);
1401
1402                 /*
1403                  * If the config cache is stale, or we have uninitialized
1404                  * metaslabs (see spa_vdev_add()), then update the config.
1405                  */
1406                 if (config_cache_txg != spa->spa_config_txg ||
1407                     state == SPA_LOAD_IMPORT)
1408                         need_update = B_TRUE;
1409
1410                 for (c = 0; c < rvd->vdev_children; c++)
1411                         if (rvd->vdev_child[c]->vdev_ms_array == 0)
1412                                 need_update = B_TRUE;
1413
1414                 /*
1415                  * Update the config cache asychronously in case we're the
1416                  * root pool, in which case the config cache isn't writable yet.
1417                  */
1418                 if (need_update)
1419                         spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1420         }
1421
1422         error = 0;
1423 out:
1424         spa->spa_minref = refcount_count(&spa->spa_refcount);
1425         if (error && error != EBADF)
1426                 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1427         spa->spa_load_state = SPA_LOAD_NONE;
1428         spa->spa_ena = 0;
1429
1430         return (error);
1431 }
1432
1433 /*
1434  * Pool Open/Import
1435  *
1436  * The import case is identical to an open except that the configuration is sent
1437  * down from userland, instead of grabbed from the configuration cache.  For the
1438  * case of an open, the pool configuration will exist in the
1439  * POOL_STATE_UNINITIALIZED state.
1440  *
1441  * The stats information (gen/count/ustats) is used to gather vdev statistics at
1442  * the same time open the pool, without having to keep around the spa_t in some
1443  * ambiguous state.
1444  */
1445 static int
1446 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1447 {
1448         spa_t *spa;
1449         int error;
1450         int locked = B_FALSE;
1451
1452         *spapp = NULL;
1453
1454         /*
1455          * As disgusting as this is, we need to support recursive calls to this
1456          * function because dsl_dir_open() is called during spa_load(), and ends
1457          * up calling spa_open() again.  The real fix is to figure out how to
1458          * avoid dsl_dir_open() calling this in the first place.
1459          */
1460         if (mutex_owner(&spa_namespace_lock) != curthread) {
1461                 mutex_enter(&spa_namespace_lock);
1462                 locked = B_TRUE;
1463         }
1464
1465         if ((spa = spa_lookup(pool)) == NULL) {
1466                 if (locked)
1467                         mutex_exit(&spa_namespace_lock);
1468                 return (ENOENT);
1469         }
1470         if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1471
1472                 spa_activate(spa);
1473
1474                 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1475
1476                 if (error == EBADF) {
1477                         /*
1478                          * If vdev_validate() returns failure (indicated by
1479                          * EBADF), it indicates that one of the vdevs indicates
1480                          * that the pool has been exported or destroyed.  If
1481                          * this is the case, the config cache is out of sync and
1482                          * we should remove the pool from the namespace.
1483                          */
1484                         spa_unload(spa);
1485                         spa_deactivate(spa);
1486                         spa_config_sync(spa, B_TRUE, B_TRUE);
1487                         spa_remove(spa);
1488                         if (locked)
1489                                 mutex_exit(&spa_namespace_lock);
1490                         return (ENOENT);
1491                 }
1492
1493                 if (error) {
1494                         /*
1495                          * We can't open the pool, but we still have useful
1496                          * information: the state of each vdev after the
1497                          * attempted vdev_open().  Return this to the user.
1498                          */
1499                         if (config != NULL && spa->spa_root_vdev != NULL)
1500                                 *config = spa_config_generate(spa, NULL, -1ULL,
1501                                     B_TRUE);
1502                         spa_unload(spa);
1503                         spa_deactivate(spa);
1504                         spa->spa_last_open_failed = B_TRUE;
1505                         if (locked)
1506                                 mutex_exit(&spa_namespace_lock);
1507                         *spapp = NULL;
1508                         return (error);
1509                 } else {
1510                         spa->spa_last_open_failed = B_FALSE;
1511                 }
1512         }
1513
1514         spa_open_ref(spa, tag);
1515
1516         if (locked)
1517                 mutex_exit(&spa_namespace_lock);
1518
1519         *spapp = spa;
1520
1521         if (config != NULL)
1522                 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1523
1524         return (0);
1525 }
1526
1527 int
1528 spa_open(const char *name, spa_t **spapp, void *tag)
1529 {
1530         return (spa_open_common(name, spapp, tag, NULL));
1531 }
1532
1533 /*
1534  * Lookup the given spa_t, incrementing the inject count in the process,
1535  * preventing it from being exported or destroyed.
1536  */
1537 spa_t *
1538 spa_inject_addref(char *name)
1539 {
1540         spa_t *spa;
1541
1542         mutex_enter(&spa_namespace_lock);
1543         if ((spa = spa_lookup(name)) == NULL) {
1544                 mutex_exit(&spa_namespace_lock);
1545                 return (NULL);
1546         }
1547         spa->spa_inject_ref++;
1548         mutex_exit(&spa_namespace_lock);
1549
1550         return (spa);
1551 }
1552
1553 void
1554 spa_inject_delref(spa_t *spa)
1555 {
1556         mutex_enter(&spa_namespace_lock);
1557         spa->spa_inject_ref--;
1558         mutex_exit(&spa_namespace_lock);
1559 }
1560
1561 /*
1562  * Add spares device information to the nvlist.
1563  */
1564 static void
1565 spa_add_spares(spa_t *spa, nvlist_t *config)
1566 {
1567         nvlist_t **spares;
1568         uint_t i, nspares;
1569         nvlist_t *nvroot;
1570         uint64_t guid;
1571         vdev_stat_t *vs;
1572         uint_t vsc;
1573         uint64_t pool;
1574
1575         if (spa->spa_spares.sav_count == 0)
1576                 return;
1577
1578         VERIFY(nvlist_lookup_nvlist(config,
1579             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1580         VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1581             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1582         if (nspares != 0) {
1583                 VERIFY(nvlist_add_nvlist_array(nvroot,
1584                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1585                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
1586                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1587
1588                 /*
1589                  * Go through and find any spares which have since been
1590                  * repurposed as an active spare.  If this is the case, update
1591                  * their status appropriately.
1592                  */
1593                 for (i = 0; i < nspares; i++) {
1594                         VERIFY(nvlist_lookup_uint64(spares[i],
1595                             ZPOOL_CONFIG_GUID, &guid) == 0);
1596                         if (spa_spare_exists(guid, &pool, NULL) &&
1597                             pool != 0ULL) {
1598                                 VERIFY(nvlist_lookup_uint64_array(
1599                                     spares[i], ZPOOL_CONFIG_STATS,
1600                                     (uint64_t **)&vs, &vsc) == 0);
1601                                 vs->vs_state = VDEV_STATE_CANT_OPEN;
1602                                 vs->vs_aux = VDEV_AUX_SPARED;
1603                         }
1604                 }
1605         }
1606 }
1607
1608 /*
1609  * Add l2cache device information to the nvlist, including vdev stats.
1610  */
1611 static void
1612 spa_add_l2cache(spa_t *spa, nvlist_t *config)
1613 {
1614         nvlist_t **l2cache;
1615         uint_t i, j, nl2cache;
1616         nvlist_t *nvroot;
1617         uint64_t guid;
1618         vdev_t *vd;
1619         vdev_stat_t *vs;
1620         uint_t vsc;
1621
1622         if (spa->spa_l2cache.sav_count == 0)
1623                 return;
1624
1625         spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1626
1627         VERIFY(nvlist_lookup_nvlist(config,
1628             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1629         VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1630             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1631         if (nl2cache != 0) {
1632                 VERIFY(nvlist_add_nvlist_array(nvroot,
1633                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1634                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
1635                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1636
1637                 /*
1638                  * Update level 2 cache device stats.
1639                  */
1640
1641                 for (i = 0; i < nl2cache; i++) {
1642                         VERIFY(nvlist_lookup_uint64(l2cache[i],
1643                             ZPOOL_CONFIG_GUID, &guid) == 0);
1644
1645                         vd = NULL;
1646                         for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1647                                 if (guid ==
1648                                     spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1649                                         vd = spa->spa_l2cache.sav_vdevs[j];
1650                                         break;
1651                                 }
1652                         }
1653                         ASSERT(vd != NULL);
1654
1655                         VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1656                             ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1657                         vdev_get_stats(vd, vs);
1658                 }
1659         }
1660
1661         spa_config_exit(spa, SCL_CONFIG, FTAG);
1662 }
1663
1664 int
1665 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1666 {
1667         int error;
1668         spa_t *spa;
1669
1670         *config = NULL;
1671         error = spa_open_common(name, &spa, FTAG, config);
1672
1673         if (spa && *config != NULL) {
1674                 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1675                     spa_get_errlog_size(spa)) == 0);
1676
1677                 if (spa_suspended(spa))
1678                         VERIFY(nvlist_add_uint64(*config,
1679                             ZPOOL_CONFIG_SUSPENDED, spa->spa_failmode) == 0);
1680
1681                 spa_add_spares(spa, *config);
1682                 spa_add_l2cache(spa, *config);
1683         }
1684
1685         /*
1686          * We want to get the alternate root even for faulted pools, so we cheat
1687          * and call spa_lookup() directly.
1688          */
1689         if (altroot) {
1690                 if (spa == NULL) {
1691                         mutex_enter(&spa_namespace_lock);
1692                         spa = spa_lookup(name);
1693                         if (spa)
1694                                 spa_altroot(spa, altroot, buflen);
1695                         else
1696                                 altroot[0] = '\0';
1697                         spa = NULL;
1698                         mutex_exit(&spa_namespace_lock);
1699                 } else {
1700                         spa_altroot(spa, altroot, buflen);
1701                 }
1702         }
1703
1704         if (spa != NULL)
1705                 spa_close(spa, FTAG);
1706
1707         return (error);
1708 }
1709
1710 /*
1711  * Validate that the auxiliary device array is well formed.  We must have an
1712  * array of nvlists, each which describes a valid leaf vdev.  If this is an
1713  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1714  * specified, as long as they are well-formed.
1715  */
1716 static int
1717 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1718     spa_aux_vdev_t *sav, const char *config, uint64_t version,
1719     vdev_labeltype_t label)
1720 {
1721         nvlist_t **dev;
1722         uint_t i, ndev;
1723         vdev_t *vd;
1724         int error;
1725
1726         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1727
1728         /*
1729          * It's acceptable to have no devs specified.
1730          */
1731         if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1732                 return (0);
1733
1734         if (ndev == 0)
1735                 return (EINVAL);
1736
1737         /*
1738          * Make sure the pool is formatted with a version that supports this
1739          * device type.
1740          */
1741         if (spa_version(spa) < version)
1742                 return (ENOTSUP);
1743
1744         /*
1745          * Set the pending device list so we correctly handle device in-use
1746          * checking.
1747          */
1748         sav->sav_pending = dev;
1749         sav->sav_npending = ndev;
1750
1751         for (i = 0; i < ndev; i++) {
1752                 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1753                     mode)) != 0)
1754                         goto out;
1755
1756                 if (!vd->vdev_ops->vdev_op_leaf) {
1757                         vdev_free(vd);
1758                         error = EINVAL;
1759                         goto out;
1760                 }
1761
1762                 /*
1763                  * The L2ARC currently only supports disk devices in
1764                  * kernel context.  For user-level testing, we allow it.
1765                  */
1766 #ifdef _KERNEL
1767                 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1768                     strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1769                         error = ENOTBLK;
1770                         goto out;
1771                 }
1772 #endif
1773                 vd->vdev_top = vd;
1774
1775                 if ((error = vdev_open(vd)) == 0 &&
1776                     (error = vdev_label_init(vd, crtxg, label)) == 0) {
1777                         VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1778                             vd->vdev_guid) == 0);
1779                 }
1780
1781                 vdev_free(vd);
1782
1783                 if (error &&
1784                     (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1785                         goto out;
1786                 else
1787                         error = 0;
1788         }
1789
1790 out:
1791         sav->sav_pending = NULL;
1792         sav->sav_npending = 0;
1793         return (error);
1794 }
1795
1796 static int
1797 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1798 {
1799         int error;
1800
1801         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1802
1803         if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1804             &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1805             VDEV_LABEL_SPARE)) != 0) {
1806                 return (error);
1807         }
1808
1809         return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1810             &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
1811             VDEV_LABEL_L2CACHE));
1812 }
1813
1814 static void
1815 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
1816     const char *config)
1817 {
1818         int i;
1819
1820         if (sav->sav_config != NULL) {
1821                 nvlist_t **olddevs;
1822                 uint_t oldndevs;
1823                 nvlist_t **newdevs;
1824
1825                 /*
1826                  * Generate new dev list by concatentating with the
1827                  * current dev list.
1828                  */
1829                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
1830                     &olddevs, &oldndevs) == 0);
1831
1832                 newdevs = kmem_alloc(sizeof (void *) *
1833                     (ndevs + oldndevs), KM_SLEEP);
1834                 for (i = 0; i < oldndevs; i++)
1835                         VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
1836                             KM_SLEEP) == 0);
1837                 for (i = 0; i < ndevs; i++)
1838                         VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
1839                             KM_SLEEP) == 0);
1840
1841                 VERIFY(nvlist_remove(sav->sav_config, config,
1842                     DATA_TYPE_NVLIST_ARRAY) == 0);
1843
1844                 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1845                     config, newdevs, ndevs + oldndevs) == 0);
1846                 for (i = 0; i < oldndevs + ndevs; i++)
1847                         nvlist_free(newdevs[i]);
1848                 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
1849         } else {
1850                 /*
1851                  * Generate a new dev list.
1852                  */
1853                 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
1854                     KM_SLEEP) == 0);
1855                 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
1856                     devs, ndevs) == 0);
1857         }
1858 }
1859
1860 /*
1861  * Stop and drop level 2 ARC devices
1862  */
1863 void
1864 spa_l2cache_drop(spa_t *spa)
1865 {
1866         vdev_t *vd;
1867         int i;
1868         spa_aux_vdev_t *sav = &spa->spa_l2cache;
1869
1870         for (i = 0; i < sav->sav_count; i++) {
1871                 uint64_t pool;
1872
1873                 vd = sav->sav_vdevs[i];
1874                 ASSERT(vd != NULL);
1875
1876                 if ((spa_mode & FWRITE) &&
1877                     spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL &&
1878                     l2arc_vdev_present(vd)) {
1879                         l2arc_remove_vdev(vd);
1880                 }
1881                 if (vd->vdev_isl2cache)
1882                         spa_l2cache_remove(vd);
1883                 vdev_clear_stats(vd);
1884                 (void) vdev_close(vd);
1885         }
1886 }
1887
1888 /*
1889  * Pool Creation
1890  */
1891 int
1892 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
1893     const char *history_str, nvlist_t *zplprops)
1894 {
1895         spa_t *spa;
1896         char *altroot = NULL;
1897         vdev_t *rvd;
1898         dsl_pool_t *dp;
1899         dmu_tx_t *tx;
1900         int c, error = 0;
1901         uint64_t txg = TXG_INITIAL;
1902         nvlist_t **spares, **l2cache;
1903         uint_t nspares, nl2cache;
1904         uint64_t version;
1905
1906         /*
1907          * If this pool already exists, return failure.
1908          */
1909         mutex_enter(&spa_namespace_lock);
1910         if (spa_lookup(pool) != NULL) {
1911                 mutex_exit(&spa_namespace_lock);
1912                 return (EEXIST);
1913         }
1914
1915         /*
1916          * Allocate a new spa_t structure.
1917          */
1918         (void) nvlist_lookup_string(props,
1919             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
1920         spa = spa_add(pool, altroot);
1921         spa_activate(spa);
1922
1923         spa->spa_uberblock.ub_txg = txg - 1;
1924
1925         if (props && (error = spa_prop_validate(spa, props))) {
1926                 spa_unload(spa);
1927                 spa_deactivate(spa);
1928                 spa_remove(spa);
1929                 mutex_exit(&spa_namespace_lock);
1930                 return (error);
1931         }
1932
1933         if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
1934             &version) != 0)
1935                 version = SPA_VERSION;
1936         ASSERT(version <= SPA_VERSION);
1937         spa->spa_uberblock.ub_version = version;
1938         spa->spa_ubsync = spa->spa_uberblock;
1939
1940         /*
1941          * Create the root vdev.
1942          */
1943         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1944
1945         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1946
1947         ASSERT(error != 0 || rvd != NULL);
1948         ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1949
1950         if (error == 0 && !zfs_allocatable_devs(nvroot))
1951                 error = EINVAL;
1952
1953         if (error == 0 &&
1954             (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1955             (error = spa_validate_aux(spa, nvroot, txg,
1956             VDEV_ALLOC_ADD)) == 0) {
1957                 for (c = 0; c < rvd->vdev_children; c++)
1958                         vdev_init(rvd->vdev_child[c], txg);
1959                 vdev_config_dirty(rvd);
1960         }
1961
1962         spa_config_exit(spa, SCL_ALL, FTAG);
1963
1964         if (error != 0) {
1965                 spa_unload(spa);
1966                 spa_deactivate(spa);
1967                 spa_remove(spa);
1968                 mutex_exit(&spa_namespace_lock);
1969                 return (error);
1970         }
1971
1972         /*
1973          * Get the list of spares, if specified.
1974          */
1975         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1976             &spares, &nspares) == 0) {
1977                 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
1978                     KM_SLEEP) == 0);
1979                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1980                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1981                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1982                 spa_load_spares(spa);
1983                 spa_config_exit(spa, SCL_ALL, FTAG);
1984                 spa->spa_spares.sav_sync = B_TRUE;
1985         }
1986
1987         /*
1988          * Get the list of level 2 cache devices, if specified.
1989          */
1990         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1991             &l2cache, &nl2cache) == 0) {
1992                 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
1993                     NV_UNIQUE_NAME, KM_SLEEP) == 0);
1994                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
1995                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1996                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1997                 spa_load_l2cache(spa);
1998                 spa_config_exit(spa, SCL_ALL, FTAG);
1999                 spa->spa_l2cache.sav_sync = B_TRUE;
2000         }
2001
2002         spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2003         spa->spa_meta_objset = dp->dp_meta_objset;
2004
2005         tx = dmu_tx_create_assigned(dp, txg);
2006
2007         /*
2008          * Create the pool config object.
2009          */
2010         spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2011             DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2012             DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2013
2014         if (zap_add(spa->spa_meta_objset,
2015             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2016             sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2017                 cmn_err(CE_PANIC, "failed to add pool config");
2018         }
2019
2020         /* Newly created pools with the right version are always deflated. */
2021         if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2022                 spa->spa_deflate = TRUE;
2023                 if (zap_add(spa->spa_meta_objset,
2024                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2025                     sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2026                         cmn_err(CE_PANIC, "failed to add deflate");
2027                 }
2028         }
2029
2030         /*
2031          * Create the deferred-free bplist object.  Turn off compression
2032          * because sync-to-convergence takes longer if the blocksize
2033          * keeps changing.
2034          */
2035         spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2036             1 << 14, tx);
2037         dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2038             ZIO_COMPRESS_OFF, tx);
2039
2040         if (zap_add(spa->spa_meta_objset,
2041             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2042             sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2043                 cmn_err(CE_PANIC, "failed to add bplist");
2044         }
2045
2046         /*
2047          * Create the pool's history object.
2048          */
2049         if (version >= SPA_VERSION_ZPOOL_HISTORY)
2050                 spa_history_create_obj(spa, tx);
2051
2052         /*
2053          * Set pool properties.
2054          */
2055         spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2056         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2057         spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2058         if (props)
2059                 spa_sync_props(spa, props, CRED(), tx);
2060
2061         dmu_tx_commit(tx);
2062
2063         spa->spa_sync_on = B_TRUE;
2064         txg_sync_start(spa->spa_dsl_pool);
2065
2066         /*
2067          * We explicitly wait for the first transaction to complete so that our
2068          * bean counters are appropriately updated.
2069          */
2070         txg_wait_synced(spa->spa_dsl_pool, txg);
2071
2072         spa_config_sync(spa, B_FALSE, B_TRUE);
2073
2074         if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2075                 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2076
2077         mutex_exit(&spa_namespace_lock);
2078
2079         spa->spa_minref = refcount_count(&spa->spa_refcount);
2080
2081         return (0);
2082 }
2083
2084 /*
2085  * Import the given pool into the system.  We set up the necessary spa_t and
2086  * then call spa_load() to do the dirty work.
2087  */
2088 static int
2089 spa_import_common(const char *pool, nvlist_t *config, nvlist_t *props,
2090     boolean_t isroot, boolean_t allowfaulted)
2091 {
2092         spa_t *spa;
2093         char *altroot = NULL;
2094         int error, loaderr;
2095         nvlist_t *nvroot;
2096         nvlist_t **spares, **l2cache;
2097         uint_t nspares, nl2cache;
2098
2099         /*
2100          * If a pool with this name exists, return failure.
2101          */
2102         mutex_enter(&spa_namespace_lock);
2103         if ((spa = spa_lookup(pool)) != NULL) {
2104                 if (isroot) {
2105                         /*
2106                          * Remove the existing root pool from the
2107                          * namespace so that we can replace it with
2108                          * the correct config we just read in.
2109                          */
2110                         ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
2111                         spa_remove(spa);
2112                 } else {
2113                         mutex_exit(&spa_namespace_lock);
2114                         return (EEXIST);
2115                 }
2116         }
2117
2118         /*
2119          * Create and initialize the spa structure.
2120          */
2121         (void) nvlist_lookup_string(props,
2122             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2123         spa = spa_add(pool, altroot);
2124         spa_activate(spa);
2125
2126         if (allowfaulted)
2127                 spa->spa_import_faulted = B_TRUE;
2128         spa->spa_is_root = isroot;
2129
2130         /*
2131          * Pass off the heavy lifting to spa_load().
2132          * Pass TRUE for mosconfig (unless this is a root pool) because
2133          * the user-supplied config is actually the one to trust when
2134          * doing an import.
2135          */
2136         loaderr = error = spa_load(spa, config, SPA_LOAD_IMPORT, !isroot);
2137
2138         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2139         /*
2140          * Toss any existing sparelist, as it doesn't have any validity anymore,
2141          * and conflicts with spa_has_spare().
2142          */
2143         if (!isroot && spa->spa_spares.sav_config) {
2144                 nvlist_free(spa->spa_spares.sav_config);
2145                 spa->spa_spares.sav_config = NULL;
2146                 spa_load_spares(spa);
2147         }
2148         if (!isroot && spa->spa_l2cache.sav_config) {
2149                 nvlist_free(spa->spa_l2cache.sav_config);
2150                 spa->spa_l2cache.sav_config = NULL;
2151                 spa_load_l2cache(spa);
2152         }
2153
2154         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2155             &nvroot) == 0);
2156         if (error == 0)
2157                 error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE);
2158         if (error == 0)
2159                 error = spa_validate_aux(spa, nvroot, -1ULL,
2160                     VDEV_ALLOC_L2CACHE);
2161         spa_config_exit(spa, SCL_ALL, FTAG);
2162
2163         if (error != 0 || (props && (error = spa_prop_set(spa, props)))) {
2164                 if (loaderr != 0 && loaderr != EINVAL && allowfaulted) {
2165                         /*
2166                          * If we failed to load the pool, but 'allowfaulted' is
2167                          * set, then manually set the config as if the config
2168                          * passed in was specified in the cache file.
2169                          */
2170                         error = 0;
2171                         spa->spa_import_faulted = B_FALSE;
2172                         if (spa->spa_config == NULL)
2173                                 spa->spa_config = spa_config_generate(spa,
2174                                     NULL, -1ULL, B_TRUE);
2175                         spa_unload(spa);
2176                         spa_deactivate(spa);
2177                         spa_config_sync(spa, B_FALSE, B_TRUE);
2178                 } else {
2179                         spa_unload(spa);
2180                         spa_deactivate(spa);
2181                         spa_remove(spa);
2182                 }
2183                 mutex_exit(&spa_namespace_lock);
2184                 return (error);
2185         }
2186
2187         /*
2188          * Override any spares and level 2 cache devices as specified by
2189          * the user, as these may have correct device names/devids, etc.
2190          */
2191         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2192             &spares, &nspares) == 0) {
2193                 if (spa->spa_spares.sav_config)
2194                         VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2195                             ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2196                 else
2197                         VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2198                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
2199                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2200                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2201                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2202                 spa_load_spares(spa);
2203                 spa_config_exit(spa, SCL_ALL, FTAG);
2204                 spa->spa_spares.sav_sync = B_TRUE;
2205         }
2206         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2207             &l2cache, &nl2cache) == 0) {
2208                 if (spa->spa_l2cache.sav_config)
2209                         VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2210                             ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2211                 else
2212                         VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2213                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
2214                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2215                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2216                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2217                 spa_load_l2cache(spa);
2218                 spa_config_exit(spa, SCL_ALL, FTAG);
2219                 spa->spa_l2cache.sav_sync = B_TRUE;
2220         }
2221
2222         if (spa_mode & FWRITE) {
2223                 /*
2224                  * Update the config cache to include the newly-imported pool.
2225                  */
2226                 spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, isroot);
2227         }
2228
2229         spa->spa_import_faulted = B_FALSE;
2230         mutex_exit(&spa_namespace_lock);
2231
2232         return (0);
2233 }
2234
2235 #ifdef _KERNEL
2236 /*
2237  * Build a "root" vdev for a top level vdev read in from a rootpool
2238  * device label.
2239  */
2240 static void
2241 spa_build_rootpool_config(nvlist_t *config)
2242 {
2243         nvlist_t *nvtop, *nvroot;
2244         uint64_t pgid;
2245
2246         /*
2247          * Add this top-level vdev to the child array.
2248          */
2249         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2250             == 0);
2251         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2252             == 0);
2253
2254         /*
2255          * Put this pool's top-level vdevs into a root vdev.
2256          */
2257         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2258         VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2259             == 0);
2260         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2261         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2262         VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2263             &nvtop, 1) == 0);
2264
2265         /*
2266          * Replace the existing vdev_tree with the new root vdev in
2267          * this pool's configuration (remove the old, add the new).
2268          */
2269         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2270         nvlist_free(nvroot);
2271 }
2272
2273 /*
2274  * Get the root pool information from the root disk, then import the root pool
2275  * during the system boot up time.
2276  */
2277 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2278
2279 int
2280 spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf,
2281     uint64_t *besttxg)
2282 {
2283         nvlist_t *config;
2284         uint64_t txg;
2285         int error;
2286
2287         if (error = vdev_disk_read_rootlabel(devpath, devid, &config))
2288                 return (error);
2289
2290         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2291
2292         if (bestconf != NULL)
2293                 *bestconf = config;
2294         else
2295                 nvlist_free(config);
2296         *besttxg = txg;
2297         return (0);
2298 }
2299
2300 boolean_t
2301 spa_rootdev_validate(nvlist_t *nv)
2302 {
2303         uint64_t ival;
2304
2305         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2306             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2307             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2308                 return (B_FALSE);
2309
2310         return (B_TRUE);
2311 }
2312
2313
2314 /*
2315  * Given the boot device's physical path or devid, check if the device
2316  * is in a valid state.  If so, return the configuration from the vdev
2317  * label.
2318  */
2319 int
2320 spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf)
2321 {
2322         nvlist_t *conf = NULL;
2323         uint64_t txg = 0;
2324         nvlist_t *nvtop, **child;
2325         char *type;
2326         char *bootpath = NULL;
2327         uint_t children, c;
2328         char *tmp;
2329         int error;
2330
2331         if (devpath && ((tmp = strchr(devpath, ' ')) != NULL))
2332                 *tmp = '\0';
2333         if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) {
2334                 cmn_err(CE_NOTE, "error reading device label");
2335                 return (error);
2336         }
2337         if (txg == 0) {
2338                 cmn_err(CE_NOTE, "this device is detached");
2339                 nvlist_free(conf);
2340                 return (EINVAL);
2341         }
2342
2343         VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE,
2344             &nvtop) == 0);
2345         VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0);
2346
2347         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2348                 if (spa_rootdev_validate(nvtop)) {
2349                         goto out;
2350                 } else {
2351                         nvlist_free(conf);
2352                         return (EINVAL);
2353                 }
2354         }
2355
2356         ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0);
2357
2358         VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN,
2359             &child, &children) == 0);
2360
2361         /*
2362          * Go thru vdevs in the mirror to see if the given device
2363          * has the most recent txg. Only the device with the most
2364          * recent txg has valid information and should be booted.
2365          */
2366         for (c = 0; c < children; c++) {
2367                 char *cdevid, *cpath;
2368                 uint64_t tmptxg;
2369
2370                 if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH,
2371                     &cpath) != 0)
2372                         return (EINVAL);
2373                 if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_DEVID,
2374                     &cdevid) != 0)
2375                         return (EINVAL);
2376                 if ((spa_check_rootconf(cpath, cdevid, NULL,
2377                     &tmptxg) == 0) && (tmptxg > txg)) {
2378                         txg = tmptxg;
2379                         VERIFY(nvlist_lookup_string(child[c],
2380                             ZPOOL_CONFIG_PATH, &bootpath) == 0);
2381                 }
2382         }
2383
2384         /* Does the best device match the one we've booted from? */
2385         if (bootpath) {
2386                 cmn_err(CE_NOTE, "try booting from '%s'", bootpath);
2387                 return (EINVAL);
2388         }
2389 out:
2390         *bestconf = conf;
2391         return (0);
2392 }
2393
2394 /*
2395  * Import a root pool.
2396  *
2397  * For x86. devpath_list will consist of devid and/or physpath name of
2398  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2399  * The GRUB "findroot" command will return the vdev we should boot.
2400  *
2401  * For Sparc, devpath_list consists the physpath name of the booting device
2402  * no matter the rootpool is a single device pool or a mirrored pool.
2403  * e.g.
2404  *      "/pci@1f,0/ide@d/disk@0,0:a"
2405  */
2406 int
2407 spa_import_rootpool(char *devpath, char *devid)
2408 {
2409         nvlist_t *conf = NULL;
2410         char *pname;
2411         int error;
2412
2413         /*
2414          * Get the vdev pathname and configuation from the most
2415          * recently updated vdev (highest txg).
2416          */
2417         if (error = spa_get_rootconf(devpath, devid, &conf))
2418                 goto msg_out;
2419
2420         /*
2421          * Add type "root" vdev to the config.
2422          */
2423         spa_build_rootpool_config(conf);
2424
2425         VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2426
2427         /*
2428          * We specify 'allowfaulted' for this to be treated like spa_open()
2429          * instead of spa_import().  This prevents us from marking vdevs as
2430          * persistently unavailable, and generates FMA ereports as if it were a
2431          * pool open, not import.
2432          */
2433         error = spa_import_common(pname, conf, NULL, B_TRUE, B_TRUE);
2434         ASSERT(error != EEXIST);
2435
2436         nvlist_free(conf);
2437         return (error);
2438
2439 msg_out:
2440         cmn_err(CE_NOTE, "\n"
2441             "  ***************************************************  \n"
2442             "  *  This device is not bootable!                   *  \n"
2443             "  *  It is either offlined or detached or faulted.  *  \n"
2444             "  *  Please try to boot from a different device.    *  \n"
2445             "  ***************************************************  ");
2446
2447         return (error);
2448 }
2449 #endif
2450
2451 /*
2452  * Import a non-root pool into the system.
2453  */
2454 int
2455 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2456 {
2457         return (spa_import_common(pool, config, props, B_FALSE, B_FALSE));
2458 }
2459
2460 int
2461 spa_import_faulted(const char *pool, nvlist_t *config, nvlist_t *props)
2462 {
2463         return (spa_import_common(pool, config, props, B_FALSE, B_TRUE));
2464 }
2465
2466
2467 /*
2468  * This (illegal) pool name is used when temporarily importing a spa_t in order
2469  * to get the vdev stats associated with the imported devices.
2470  */
2471 #define TRYIMPORT_NAME  "$import"
2472
2473 nvlist_t *
2474 spa_tryimport(nvlist_t *tryconfig)
2475 {
2476         nvlist_t *config = NULL;
2477         char *poolname;
2478         spa_t *spa;
2479         uint64_t state;
2480
2481         if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2482                 return (NULL);
2483
2484         if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2485                 return (NULL);
2486
2487         /*
2488          * Create and initialize the spa structure.
2489          */
2490         mutex_enter(&spa_namespace_lock);
2491         spa = spa_add(TRYIMPORT_NAME, NULL);
2492         spa_activate(spa);
2493
2494         /*
2495          * Pass off the heavy lifting to spa_load().
2496          * Pass TRUE for mosconfig because the user-supplied config
2497          * is actually the one to trust when doing an import.
2498          */
2499         (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2500
2501         /*
2502          * If 'tryconfig' was at least parsable, return the current config.
2503          */
2504         if (spa->spa_root_vdev != NULL) {
2505                 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2506                 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2507                     poolname) == 0);
2508                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2509                     state) == 0);
2510                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2511                     spa->spa_uberblock.ub_timestamp) == 0);
2512
2513                 /*
2514                  * If the bootfs property exists on this pool then we
2515                  * copy it out so that external consumers can tell which
2516                  * pools are bootable.
2517                  */
2518                 if (spa->spa_bootfs) {
2519                         char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2520
2521                         /*
2522                          * We have to play games with the name since the
2523                          * pool was opened as TRYIMPORT_NAME.
2524                          */
2525                         if (dsl_dsobj_to_dsname(spa_name(spa),
2526                             spa->spa_bootfs, tmpname) == 0) {
2527                                 char *cp;
2528                                 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2529
2530                                 cp = strchr(tmpname, '/');
2531                                 if (cp == NULL) {
2532                                         (void) strlcpy(dsname, tmpname,
2533                                             MAXPATHLEN);
2534                                 } else {
2535                                         (void) snprintf(dsname, MAXPATHLEN,
2536                                             "%s/%s", poolname, ++cp);
2537                                 }
2538                                 VERIFY(nvlist_add_string(config,
2539                                     ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2540                                 kmem_free(dsname, MAXPATHLEN);
2541                         }
2542                         kmem_free(tmpname, MAXPATHLEN);
2543                 }
2544
2545                 /*
2546                  * Add the list of hot spares and level 2 cache devices.
2547                  */
2548                 spa_add_spares(spa, config);
2549                 spa_add_l2cache(spa, config);
2550         }
2551
2552         spa_unload(spa);
2553         spa_deactivate(spa);
2554         spa_remove(spa);
2555         mutex_exit(&spa_namespace_lock);
2556
2557         return (config);
2558 }
2559
2560 /*
2561  * Pool export/destroy
2562  *
2563  * The act of destroying or exporting a pool is very simple.  We make sure there
2564  * is no more pending I/O and any references to the pool are gone.  Then, we
2565  * update the pool state and sync all the labels to disk, removing the
2566  * configuration from the cache afterwards.
2567  */
2568 static int
2569 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2570     boolean_t force)
2571 {
2572         spa_t *spa;
2573
2574         if (oldconfig)
2575                 *oldconfig = NULL;
2576
2577         if (!(spa_mode & FWRITE))
2578                 return (EROFS);
2579
2580         mutex_enter(&spa_namespace_lock);
2581         if ((spa = spa_lookup(pool)) == NULL) {
2582                 mutex_exit(&spa_namespace_lock);
2583                 return (ENOENT);
2584         }
2585
2586         /*
2587          * Put a hold on the pool, drop the namespace lock, stop async tasks,
2588          * reacquire the namespace lock, and see if we can export.
2589          */
2590         spa_open_ref(spa, FTAG);
2591         mutex_exit(&spa_namespace_lock);
2592         spa_async_suspend(spa);
2593         mutex_enter(&spa_namespace_lock);
2594         spa_close(spa, FTAG);
2595
2596         /*
2597          * The pool will be in core if it's openable,
2598          * in which case we can modify its state.
2599          */
2600         if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2601                 /*
2602                  * Objsets may be open only because they're dirty, so we
2603                  * have to force it to sync before checking spa_refcnt.
2604                  */
2605                 txg_wait_synced(spa->spa_dsl_pool, 0);
2606
2607                 /*
2608                  * A pool cannot be exported or destroyed if there are active
2609                  * references.  If we are resetting a pool, allow references by
2610                  * fault injection handlers.
2611                  */
2612                 if (!spa_refcount_zero(spa) ||
2613                     (spa->spa_inject_ref != 0 &&
2614                     new_state != POOL_STATE_UNINITIALIZED)) {
2615                         spa_async_resume(spa);
2616                         mutex_exit(&spa_namespace_lock);
2617                         return (EBUSY);
2618                 }
2619
2620                 /*
2621                  * A pool cannot be exported if it has an active shared spare.
2622                  * This is to prevent other pools stealing the active spare
2623                  * from an exported pool. At user's own will, such pool can
2624                  * be forcedly exported.
2625                  */
2626                 if (!force && new_state == POOL_STATE_EXPORTED &&
2627                     spa_has_active_shared_spare(spa)) {
2628                         spa_async_resume(spa);
2629                         mutex_exit(&spa_namespace_lock);
2630                         return (EXDEV);
2631                 }
2632
2633                 /*
2634                  * We want this to be reflected on every label,
2635                  * so mark them all dirty.  spa_unload() will do the
2636                  * final sync that pushes these changes out.
2637                  */
2638                 if (new_state != POOL_STATE_UNINITIALIZED) {
2639                         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2640                         spa->spa_state = new_state;
2641                         spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2642                         vdev_config_dirty(spa->spa_root_vdev);
2643                         spa_config_exit(spa, SCL_ALL, FTAG);
2644                 }
2645         }
2646
2647         spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2648
2649         if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2650                 spa_unload(spa);
2651                 spa_deactivate(spa);
2652         }
2653
2654         if (oldconfig && spa->spa_config)
2655                 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2656
2657         if (new_state != POOL_STATE_UNINITIALIZED) {
2658                 spa_config_sync(spa, B_TRUE, B_TRUE);
2659                 spa_remove(spa);
2660         }
2661         mutex_exit(&spa_namespace_lock);
2662
2663         return (0);
2664 }
2665
2666 /*
2667  * Destroy a storage pool.
2668  */
2669 int
2670 spa_destroy(char *pool)
2671 {
2672         return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, B_FALSE));
2673 }
2674
2675 /*
2676  * Export a storage pool.
2677  */
2678 int
2679 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force)
2680 {
2681         return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, force));
2682 }
2683
2684 /*
2685  * Similar to spa_export(), this unloads the spa_t without actually removing it
2686  * from the namespace in any way.
2687  */
2688 int
2689 spa_reset(char *pool)
2690 {
2691         return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2692             B_FALSE));
2693 }
2694
2695 /*
2696  * ==========================================================================
2697  * Device manipulation
2698  * ==========================================================================
2699  */
2700
2701 /*
2702  * Add a device to a storage pool.
2703  */
2704 int
2705 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2706 {
2707         uint64_t txg;
2708         int c, error;
2709         vdev_t *rvd = spa->spa_root_vdev;
2710         vdev_t *vd, *tvd;
2711         nvlist_t **spares, **l2cache;
2712         uint_t nspares, nl2cache;
2713
2714         txg = spa_vdev_enter(spa);
2715
2716         if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2717             VDEV_ALLOC_ADD)) != 0)
2718                 return (spa_vdev_exit(spa, NULL, txg, error));
2719
2720         spa->spa_pending_vdev = vd;     /* spa_vdev_exit() will clear this */
2721
2722         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2723             &nspares) != 0)
2724                 nspares = 0;
2725
2726         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2727             &nl2cache) != 0)
2728                 nl2cache = 0;
2729
2730         if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2731                 return (spa_vdev_exit(spa, vd, txg, EINVAL));
2732
2733         if (vd->vdev_children != 0 &&
2734             (error = vdev_create(vd, txg, B_FALSE)) != 0)
2735                 return (spa_vdev_exit(spa, vd, txg, error));
2736
2737         /*
2738          * We must validate the spares and l2cache devices after checking the
2739          * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2740          */
2741         if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2742                 return (spa_vdev_exit(spa, vd, txg, error));
2743
2744         /*
2745          * Transfer each new top-level vdev from vd to rvd.
2746          */
2747         for (c = 0; c < vd->vdev_children; c++) {
2748                 tvd = vd->vdev_child[c];
2749                 vdev_remove_child(vd, tvd);
2750                 tvd->vdev_id = rvd->vdev_children;
2751                 vdev_add_child(rvd, tvd);
2752                 vdev_config_dirty(tvd);
2753         }
2754
2755         if (nspares != 0) {
2756                 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2757                     ZPOOL_CONFIG_SPARES);
2758                 spa_load_spares(spa);
2759                 spa->spa_spares.sav_sync = B_TRUE;
2760         }
2761
2762         if (nl2cache != 0) {
2763                 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2764                     ZPOOL_CONFIG_L2CACHE);
2765                 spa_load_l2cache(spa);
2766                 spa->spa_l2cache.sav_sync = B_TRUE;
2767         }
2768
2769         /*
2770          * We have to be careful when adding new vdevs to an existing pool.
2771          * If other threads start allocating from these vdevs before we
2772          * sync the config cache, and we lose power, then upon reboot we may
2773          * fail to open the pool because there are DVAs that the config cache
2774          * can't translate.  Therefore, we first add the vdevs without
2775          * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2776          * and then let spa_config_update() initialize the new metaslabs.
2777          *
2778          * spa_load() checks for added-but-not-initialized vdevs, so that
2779          * if we lose power at any point in this sequence, the remaining
2780          * steps will be completed the next time we load the pool.
2781          */
2782         (void) spa_vdev_exit(spa, vd, txg, 0);
2783
2784         mutex_enter(&spa_namespace_lock);
2785         spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2786         mutex_exit(&spa_namespace_lock);
2787
2788         return (0);
2789 }
2790
2791 /*
2792  * Attach a device to a mirror.  The arguments are the path to any device
2793  * in the mirror, and the nvroot for the new device.  If the path specifies
2794  * a device that is not mirrored, we automatically insert the mirror vdev.
2795  *
2796  * If 'replacing' is specified, the new device is intended to replace the
2797  * existing device; in this case the two devices are made into their own
2798  * mirror using the 'replacing' vdev, which is functionally identical to
2799  * the mirror vdev (it actually reuses all the same ops) but has a few
2800  * extra rules: you can't attach to it after it's been created, and upon
2801  * completion of resilvering, the first disk (the one being replaced)
2802  * is automatically detached.
2803  */
2804 int
2805 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2806 {
2807         uint64_t txg, open_txg;
2808         vdev_t *rvd = spa->spa_root_vdev;
2809         vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2810         vdev_ops_t *pvops;
2811         dmu_tx_t *tx;
2812         char *oldvdpath, *newvdpath;
2813         int newvd_isspare;
2814         int error;
2815
2816         txg = spa_vdev_enter(spa);
2817
2818         oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
2819
2820         if (oldvd == NULL)
2821                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2822
2823         if (!oldvd->vdev_ops->vdev_op_leaf)
2824                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2825
2826         pvd = oldvd->vdev_parent;
2827
2828         if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
2829             VDEV_ALLOC_ADD)) != 0)
2830                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
2831
2832         if (newrootvd->vdev_children != 1)
2833                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2834
2835         newvd = newrootvd->vdev_child[0];
2836
2837         if (!newvd->vdev_ops->vdev_op_leaf)
2838                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2839
2840         if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
2841                 return (spa_vdev_exit(spa, newrootvd, txg, error));
2842
2843         /*
2844          * Spares can't replace logs
2845          */
2846         if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
2847                 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2848
2849         if (!replacing) {
2850                 /*
2851                  * For attach, the only allowable parent is a mirror or the root
2852                  * vdev.
2853                  */
2854                 if (pvd->vdev_ops != &vdev_mirror_ops &&
2855                     pvd->vdev_ops != &vdev_root_ops)
2856                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2857
2858                 pvops = &vdev_mirror_ops;
2859         } else {
2860                 /*
2861                  * Active hot spares can only be replaced by inactive hot
2862                  * spares.
2863                  */
2864                 if (pvd->vdev_ops == &vdev_spare_ops &&
2865                     pvd->vdev_child[1] == oldvd &&
2866                     !spa_has_spare(spa, newvd->vdev_guid))
2867                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2868
2869                 /*
2870                  * If the source is a hot spare, and the parent isn't already a
2871                  * spare, then we want to create a new hot spare.  Otherwise, we
2872                  * want to create a replacing vdev.  The user is not allowed to
2873                  * attach to a spared vdev child unless the 'isspare' state is
2874                  * the same (spare replaces spare, non-spare replaces
2875                  * non-spare).
2876                  */
2877                 if (pvd->vdev_ops == &vdev_replacing_ops)
2878                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2879                 else if (pvd->vdev_ops == &vdev_spare_ops &&
2880                     newvd->vdev_isspare != oldvd->vdev_isspare)
2881                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2882                 else if (pvd->vdev_ops != &vdev_spare_ops &&
2883                     newvd->vdev_isspare)
2884                         pvops = &vdev_spare_ops;
2885                 else
2886                         pvops = &vdev_replacing_ops;
2887         }
2888
2889         /*
2890          * Compare the new device size with the replaceable/attachable
2891          * device size.
2892          */
2893         if (newvd->vdev_psize < vdev_get_rsize(oldvd))
2894                 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
2895
2896         /*
2897          * The new device cannot have a higher alignment requirement
2898          * than the top-level vdev.
2899          */
2900         if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
2901                 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
2902
2903         /*
2904          * If this is an in-place replacement, update oldvd's path and devid
2905          * to make it distinguishable from newvd, and unopenable from now on.
2906          */
2907         if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
2908                 spa_strfree(oldvd->vdev_path);
2909                 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
2910                     KM_SLEEP);
2911                 (void) sprintf(oldvd->vdev_path, "%s/%s",
2912                     newvd->vdev_path, "old");
2913                 if (oldvd->vdev_devid != NULL) {
2914                         spa_strfree(oldvd->vdev_devid);
2915                         oldvd->vdev_devid = NULL;
2916                 }
2917         }
2918
2919         /*
2920          * If the parent is not a mirror, or if we're replacing, insert the new
2921          * mirror/replacing/spare vdev above oldvd.
2922          */
2923         if (pvd->vdev_ops != pvops)
2924                 pvd = vdev_add_parent(oldvd, pvops);
2925
2926         ASSERT(pvd->vdev_top->vdev_parent == rvd);
2927         ASSERT(pvd->vdev_ops == pvops);
2928         ASSERT(oldvd->vdev_parent == pvd);
2929
2930         /*
2931          * Extract the new device from its root and add it to pvd.
2932          */
2933         vdev_remove_child(newrootvd, newvd);
2934         newvd->vdev_id = pvd->vdev_children;
2935         vdev_add_child(pvd, newvd);
2936
2937         /*
2938          * If newvd is smaller than oldvd, but larger than its rsize,
2939          * the addition of newvd may have decreased our parent's asize.
2940          */
2941         pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
2942
2943         tvd = newvd->vdev_top;
2944         ASSERT(pvd->vdev_top == tvd);
2945         ASSERT(tvd->vdev_parent == rvd);
2946
2947         vdev_config_dirty(tvd);
2948
2949         /*
2950          * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
2951          * upward when spa_vdev_exit() calls vdev_dtl_reassess().
2952          */
2953         open_txg = txg + TXG_CONCURRENT_STATES - 1;
2954
2955         mutex_enter(&newvd->vdev_dtl_lock);
2956         space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
2957             open_txg - TXG_INITIAL + 1);
2958         mutex_exit(&newvd->vdev_dtl_lock);
2959
2960         if (newvd->vdev_isspare)
2961                 spa_spare_activate(newvd);
2962         oldvdpath = spa_strdup(oldvd->vdev_path);
2963         newvdpath = spa_strdup(newvd->vdev_path);
2964         newvd_isspare = newvd->vdev_isspare;
2965
2966         /*
2967          * Mark newvd's DTL dirty in this txg.
2968          */
2969         vdev_dirty(tvd, VDD_DTL, newvd, txg);
2970
2971         (void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
2972
2973         tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
2974         if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
2975                 spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
2976                     CRED(),  "%s vdev=%s %s vdev=%s",
2977                     replacing && newvd_isspare ? "spare in" :
2978                     replacing ? "replace" : "attach", newvdpath,
2979                     replacing ? "for" : "to", oldvdpath);
2980                 dmu_tx_commit(tx);
2981         } else {
2982                 dmu_tx_abort(tx);
2983         }
2984
2985         spa_strfree(oldvdpath);
2986         spa_strfree(newvdpath);
2987
2988         /*
2989          * Kick off a resilver to update newvd.
2990          */
2991         VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
2992
2993         return (0);
2994 }
2995
2996 /*
2997  * Detach a device from a mirror or replacing vdev.
2998  * If 'replace_done' is specified, only detach if the parent
2999  * is a replacing vdev.
3000  */
3001 int
3002 spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
3003 {
3004         uint64_t txg;
3005         int c, t, error;
3006         vdev_t *rvd = spa->spa_root_vdev;
3007         vdev_t *vd, *pvd, *cvd, *tvd;
3008         boolean_t unspare = B_FALSE;
3009         uint64_t unspare_guid;
3010         size_t len;
3011
3012         txg = spa_vdev_enter(spa);
3013
3014         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3015
3016         if (vd == NULL)
3017                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3018
3019         if (!vd->vdev_ops->vdev_op_leaf)
3020                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3021
3022         pvd = vd->vdev_parent;
3023
3024         /*
3025          * If replace_done is specified, only remove this device if it's
3026          * the first child of a replacing vdev.  For the 'spare' vdev, either
3027          * disk can be removed.
3028          */
3029         if (replace_done) {
3030                 if (pvd->vdev_ops == &vdev_replacing_ops) {
3031                         if (vd->vdev_id != 0)
3032                                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3033                 } else if (pvd->vdev_ops != &vdev_spare_ops) {
3034                         return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3035                 }
3036         }
3037
3038         ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3039             spa_version(spa) >= SPA_VERSION_SPARES);
3040
3041         /*
3042          * Only mirror, replacing, and spare vdevs support detach.
3043          */
3044         if (pvd->vdev_ops != &vdev_replacing_ops &&
3045             pvd->vdev_ops != &vdev_mirror_ops &&
3046             pvd->vdev_ops != &vdev_spare_ops)
3047                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3048
3049         /*
3050          * If there's only one replica, you can't detach it.
3051          */
3052         if (pvd->vdev_children <= 1)
3053                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3054
3055         /*
3056          * If all siblings have non-empty DTLs, this device may have the only
3057          * valid copy of the data, which means we cannot safely detach it.
3058          *
3059          * XXX -- as in the vdev_offline() case, we really want a more
3060          * precise DTL check.
3061          */
3062         for (c = 0; c < pvd->vdev_children; c++) {
3063                 uint64_t dirty;
3064
3065                 cvd = pvd->vdev_child[c];
3066                 if (cvd == vd)
3067                         continue;
3068                 if (vdev_is_dead(cvd))
3069                         continue;
3070                 mutex_enter(&cvd->vdev_dtl_lock);
3071                 dirty = cvd->vdev_dtl_map.sm_space |
3072                     cvd->vdev_dtl_scrub.sm_space;
3073                 mutex_exit(&cvd->vdev_dtl_lock);
3074                 if (!dirty)
3075                         break;
3076         }
3077
3078         if (c == pvd->vdev_children)
3079                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3080
3081         /*
3082          * If we are detaching the second disk from a replacing vdev, then
3083          * check to see if we changed the original vdev's path to have "/old"
3084          * at the end in spa_vdev_attach().  If so, undo that change now.
3085          */
3086         if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3087             pvd->vdev_child[0]->vdev_path != NULL &&
3088             pvd->vdev_child[1]->vdev_path != NULL) {
3089                 ASSERT(pvd->vdev_child[1] == vd);
3090                 cvd = pvd->vdev_child[0];
3091                 len = strlen(vd->vdev_path);
3092                 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3093                     strcmp(cvd->vdev_path + len, "/old") == 0) {
3094                         spa_strfree(cvd->vdev_path);
3095                         cvd->vdev_path = spa_strdup(vd->vdev_path);
3096                 }
3097         }
3098
3099         /*
3100          * If we are detaching the original disk from a spare, then it implies
3101          * that the spare should become a real disk, and be removed from the
3102          * active spare list for the pool.
3103          */
3104         if (pvd->vdev_ops == &vdev_spare_ops &&
3105             vd->vdev_id == 0)
3106                 unspare = B_TRUE;
3107
3108         /*
3109          * Erase the disk labels so the disk can be used for other things.
3110          * This must be done after all other error cases are handled,
3111          * but before we disembowel vd (so we can still do I/O to it).
3112          * But if we can't do it, don't treat the error as fatal --
3113          * it may be that the unwritability of the disk is the reason
3114          * it's being detached!
3115          */
3116         error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3117
3118         /*
3119          * Remove vd from its parent and compact the parent's children.
3120          */
3121         vdev_remove_child(pvd, vd);
3122         vdev_compact_children(pvd);
3123
3124         /*
3125          * Remember one of the remaining children so we can get tvd below.
3126          */
3127         cvd = pvd->vdev_child[0];
3128
3129         /*
3130          * If we need to remove the remaining child from the list of hot spares,
3131          * do it now, marking the vdev as no longer a spare in the process.  We
3132          * must do this before vdev_remove_parent(), because that can change the
3133          * GUID if it creates a new toplevel GUID.
3134          */
3135         if (unspare) {
3136                 ASSERT(cvd->vdev_isspare);
3137                 spa_spare_remove(cvd);
3138                 unspare_guid = cvd->vdev_guid;
3139         }
3140
3141         /*
3142          * If the parent mirror/replacing vdev only has one child,
3143          * the parent is no longer needed.  Remove it from the tree.
3144          */
3145         if (pvd->vdev_children == 1)
3146                 vdev_remove_parent(cvd);
3147
3148         /*
3149          * We don't set tvd until now because the parent we just removed
3150          * may have been the previous top-level vdev.
3151          */
3152         tvd = cvd->vdev_top;
3153         ASSERT(tvd->vdev_parent == rvd);
3154
3155         /*
3156          * Reevaluate the parent vdev state.
3157          */
3158         vdev_propagate_state(cvd);
3159
3160         /*
3161          * If the device we just detached was smaller than the others, it may be
3162          * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3163          * can't fail because the existing metaslabs are already in core, so
3164          * there's nothing to read from disk.
3165          */
3166         VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3167
3168         vdev_config_dirty(tvd);
3169
3170         /*
3171          * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3172          * vd->vdev_detached is set and free vd's DTL object in syncing context.
3173          * But first make sure we're not on any *other* txg's DTL list, to
3174          * prevent vd from being accessed after it's freed.
3175          */
3176         for (t = 0; t < TXG_SIZE; t++)
3177                 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3178         vd->vdev_detached = B_TRUE;
3179         vdev_dirty(tvd, VDD_DTL, vd, txg);
3180
3181         spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3182
3183         error = spa_vdev_exit(spa, vd, txg, 0);
3184
3185         /*
3186          * If this was the removal of the original device in a hot spare vdev,
3187          * then we want to go through and remove the device from the hot spare
3188          * list of every other pool.
3189          */
3190         if (unspare) {
3191                 spa = NULL;
3192                 mutex_enter(&spa_namespace_lock);
3193                 while ((spa = spa_next(spa)) != NULL) {
3194                         if (spa->spa_state != POOL_STATE_ACTIVE)
3195                                 continue;
3196                         spa_open_ref(spa, FTAG);
3197                         mutex_exit(&spa_namespace_lock);
3198                         (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3199                         mutex_enter(&spa_namespace_lock);
3200                         spa_close(spa, FTAG);
3201                 }
3202                 mutex_exit(&spa_namespace_lock);
3203         }
3204
3205         return (error);
3206 }
3207
3208 static nvlist_t *
3209 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3210 {
3211         for (int i = 0; i < count; i++) {
3212                 uint64_t guid;
3213
3214                 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3215                     &guid) == 0);
3216
3217                 if (guid == target_guid)
3218                         return (nvpp[i]);
3219         }
3220
3221         return (NULL);
3222 }
3223
3224 static void
3225 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3226         nvlist_t *dev_to_remove)
3227 {
3228         nvlist_t **newdev = NULL;
3229
3230         if (count > 1)
3231                 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3232
3233         for (int i = 0, j = 0; i < count; i++) {
3234                 if (dev[i] == dev_to_remove)
3235                         continue;
3236                 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3237         }
3238
3239         VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3240         VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3241
3242         for (int i = 0; i < count - 1; i++)
3243                 nvlist_free(newdev[i]);
3244
3245         if (count > 1)
3246                 kmem_free(newdev, (count - 1) * sizeof (void *));
3247 }
3248
3249 /*
3250  * Remove a device from the pool.  Currently, this supports removing only hot
3251  * spares and level 2 ARC devices.
3252  */
3253 int
3254 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3255 {
3256         vdev_t *vd;
3257         nvlist_t **spares, **l2cache, *nv;
3258         uint_t nspares, nl2cache;
3259         uint64_t txg;
3260         int error = 0;
3261
3262         txg = spa_vdev_enter(spa);
3263
3264         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3265
3266         if (spa->spa_spares.sav_vdevs != NULL &&
3267             nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3268             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3269             (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3270                 /*
3271                  * Only remove the hot spare if it's not currently in use
3272                  * in this pool.
3273                  */
3274                 if (vd == NULL || unspare) {
3275                         spa_vdev_remove_aux(spa->spa_spares.sav_config,
3276                             ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3277                         spa_load_spares(spa);
3278                         spa->spa_spares.sav_sync = B_TRUE;
3279                 } else {
3280                         error = EBUSY;
3281                 }
3282         } else if (spa->spa_l2cache.sav_vdevs != NULL &&
3283             nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3284             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3285             (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3286                 /*
3287                  * Cache devices can always be removed.
3288                  */
3289                 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3290                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3291                 spa_load_l2cache(spa);
3292                 spa->spa_l2cache.sav_sync = B_TRUE;
3293         } else if (vd != NULL) {
3294                 /*
3295                  * Normal vdevs cannot be removed (yet).
3296                  */
3297                 error = ENOTSUP;
3298         } else {
3299                 /*
3300                  * There is no vdev of any kind with the specified guid.
3301                  */
3302                 error = ENOENT;
3303         }
3304
3305         return (spa_vdev_exit(spa, NULL, txg, error));
3306 }
3307
3308 /*
3309  * Find any device that's done replacing, or a vdev marked 'unspare' that's
3310  * current spared, so we can detach it.
3311  */
3312 static vdev_t *
3313 spa_vdev_resilver_done_hunt(vdev_t *vd)
3314 {
3315         vdev_t *newvd, *oldvd;
3316         int c;
3317
3318         for (c = 0; c < vd->vdev_children; c++) {
3319                 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3320                 if (oldvd != NULL)
3321                         return (oldvd);
3322         }
3323
3324         /*
3325          * Check for a completed replacement.
3326          */
3327         if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3328                 oldvd = vd->vdev_child[0];
3329                 newvd = vd->vdev_child[1];
3330
3331                 mutex_enter(&newvd->vdev_dtl_lock);
3332                 if (newvd->vdev_dtl_map.sm_space == 0 &&
3333                     newvd->vdev_dtl_scrub.sm_space == 0) {
3334                         mutex_exit(&newvd->vdev_dtl_lock);
3335                         return (oldvd);
3336                 }
3337                 mutex_exit(&newvd->vdev_dtl_lock);
3338         }
3339
3340         /*
3341          * Check for a completed resilver with the 'unspare' flag set.
3342          */
3343         if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3344                 newvd = vd->vdev_child[0];
3345                 oldvd = vd->vdev_child[1];
3346
3347                 mutex_enter(&newvd->vdev_dtl_lock);
3348                 if (newvd->vdev_unspare &&
3349                     newvd->vdev_dtl_map.sm_space == 0 &&
3350                     newvd->vdev_dtl_scrub.sm_space == 0) {
3351                         newvd->vdev_unspare = 0;
3352                         mutex_exit(&newvd->vdev_dtl_lock);
3353                         return (oldvd);
3354                 }
3355                 mutex_exit(&newvd->vdev_dtl_lock);
3356         }
3357
3358         return (NULL);
3359 }
3360
3361 static void
3362 spa_vdev_resilver_done(spa_t *spa)
3363 {
3364         vdev_t *vd;
3365         vdev_t *pvd;
3366         uint64_t guid;
3367         uint64_t pguid = 0;
3368
3369         spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3370
3371         while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3372                 guid = vd->vdev_guid;
3373                 /*
3374                  * If we have just finished replacing a hot spared device, then
3375                  * we need to detach the parent's first child (the original hot
3376                  * spare) as well.
3377                  */
3378                 pvd = vd->vdev_parent;
3379                 if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3380                     pvd->vdev_id == 0) {
3381                         ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3382                         ASSERT(pvd->vdev_parent->vdev_children == 2);
3383                         pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
3384                 }
3385                 spa_config_exit(spa, SCL_CONFIG, FTAG);
3386                 if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
3387                         return;
3388                 if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
3389                         return;
3390                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3391         }
3392
3393         spa_config_exit(spa, SCL_CONFIG, FTAG);
3394 }
3395
3396 /*
3397  * Update the stored path for this vdev.  Dirty the vdev configuration, relying
3398  * on spa_vdev_enter/exit() to synchronize the labels and cache.
3399  */
3400 int
3401 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3402 {
3403         vdev_t *vd;
3404         uint64_t txg;
3405
3406         txg = spa_vdev_enter(spa);
3407
3408         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) {
3409                 /*
3410                  * Determine if this is a reference to a hot spare device.  If
3411                  * it is, update the path manually as there is no associated
3412                  * vdev_t that can be synced to disk.
3413                  */
3414                 nvlist_t **spares;
3415                 uint_t i, nspares;
3416
3417                 if (spa->spa_spares.sav_config != NULL) {
3418                         VERIFY(nvlist_lookup_nvlist_array(
3419                             spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
3420                             &spares, &nspares) == 0);
3421                         for (i = 0; i < nspares; i++) {
3422                                 uint64_t theguid;
3423                                 VERIFY(nvlist_lookup_uint64(spares[i],
3424                                     ZPOOL_CONFIG_GUID, &theguid) == 0);
3425                                 if (theguid == guid) {
3426                                         VERIFY(nvlist_add_string(spares[i],
3427                                             ZPOOL_CONFIG_PATH, newpath) == 0);
3428                                         spa_load_spares(spa);
3429                                         spa->spa_spares.sav_sync = B_TRUE;
3430                                         return (spa_vdev_exit(spa, NULL, txg,
3431                                             0));
3432                                 }
3433                         }
3434                 }
3435
3436                 return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3437         }
3438
3439         if (!vd->vdev_ops->vdev_op_leaf)
3440                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3441
3442         spa_strfree(vd->vdev_path);
3443         vd->vdev_path = spa_strdup(newpath);
3444
3445         vdev_config_dirty(vd->vdev_top);
3446
3447         return (spa_vdev_exit(spa, NULL, txg, 0));
3448 }
3449
3450 /*
3451  * ==========================================================================
3452  * SPA Scrubbing
3453  * ==========================================================================
3454  */
3455
3456 int
3457 spa_scrub(spa_t *spa, pool_scrub_type_t type)
3458 {
3459         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3460
3461         if ((uint_t)type >= POOL_SCRUB_TYPES)
3462                 return (ENOTSUP);
3463
3464         /*
3465          * If a resilver was requested, but there is no DTL on a
3466          * writeable leaf device, we have nothing to do.
3467          */
3468         if (type == POOL_SCRUB_RESILVER &&
3469             !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3470                 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3471                 return (0);
3472         }
3473
3474         if (type == POOL_SCRUB_EVERYTHING &&
3475             spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3476             spa->spa_dsl_pool->dp_scrub_isresilver)
3477                 return (EBUSY);
3478
3479         if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3480                 return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3481         } else if (type == POOL_SCRUB_NONE) {
3482                 return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3483         } else {
3484                 return (EINVAL);
3485         }
3486 }
3487
3488 /*
3489  * ==========================================================================
3490  * SPA async task processing
3491  * ==========================================================================
3492  */
3493
3494 static void
3495 spa_async_remove(spa_t *spa, vdev_t *vd)
3496 {
3497         if (vd->vdev_remove_wanted) {
3498                 vd->vdev_remove_wanted = 0;
3499                 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3500                 vdev_clear(spa, vd);
3501                 vdev_state_dirty(vd->vdev_top);
3502         }
3503
3504         for (int c = 0; c < vd->vdev_children; c++)
3505                 spa_async_remove(spa, vd->vdev_child[c]);
3506 }
3507
3508 static void
3509 spa_async_probe(spa_t *spa, vdev_t *vd)
3510 {
3511         if (vd->vdev_probe_wanted) {
3512                 vd->vdev_probe_wanted = 0;
3513                 vdev_reopen(vd);        /* vdev_open() does the actual probe */
3514         }
3515
3516         for (int c = 0; c < vd->vdev_children; c++)
3517                 spa_async_probe(spa, vd->vdev_child[c]);
3518 }
3519
3520 static void
3521 spa_async_thread(spa_t *spa)
3522 {
3523         int tasks;
3524
3525         ASSERT(spa->spa_sync_on);
3526
3527         mutex_enter(&spa->spa_async_lock);
3528         tasks = spa->spa_async_tasks;
3529         spa->spa_async_tasks = 0;
3530         mutex_exit(&spa->spa_async_lock);
3531
3532         /*
3533          * See if the config needs to be updated.
3534          */
3535         if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3536                 mutex_enter(&spa_namespace_lock);
3537                 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3538                 mutex_exit(&spa_namespace_lock);
3539         }
3540
3541         /*
3542          * See if any devices need to be marked REMOVED.
3543          */
3544         if (tasks & SPA_ASYNC_REMOVE) {
3545                 spa_vdev_state_enter(spa);
3546                 spa_async_remove(spa, spa->spa_root_vdev);
3547                 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3548                         spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3549                 for (int i = 0; i < spa->spa_spares.sav_count; i++)
3550                         spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3551                 (void) spa_vdev_state_exit(spa, NULL, 0);
3552         }
3553
3554         /*
3555          * See if any devices need to be probed.
3556          */
3557         if (tasks & SPA_ASYNC_PROBE) {
3558                 spa_vdev_state_enter(spa);
3559                 spa_async_probe(spa, spa->spa_root_vdev);
3560                 (void) spa_vdev_state_exit(spa, NULL, 0);
3561         }
3562
3563         /*
3564          * If any devices are done replacing, detach them.
3565          */
3566         if (tasks & SPA_ASYNC_RESILVER_DONE)
3567                 spa_vdev_resilver_done(spa);
3568
3569         /*
3570          * Kick off a resilver.
3571          */
3572         if (tasks & SPA_ASYNC_RESILVER)
3573                 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3574
3575         /*
3576          * Let the world know that we're done.
3577          */
3578         mutex_enter(&spa->spa_async_lock);
3579         spa->spa_async_thread = NULL;
3580         cv_broadcast(&spa->spa_async_cv);
3581         mutex_exit(&spa->spa_async_lock);
3582         thread_exit();
3583 }
3584
3585 void
3586 spa_async_suspend(spa_t *spa)
3587 {
3588         mutex_enter(&spa->spa_async_lock);
3589         spa->spa_async_suspended++;
3590         while (spa->spa_async_thread != NULL)
3591                 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3592         mutex_exit(&spa->spa_async_lock);
3593 }
3594
3595 void
3596 spa_async_resume(spa_t *spa)
3597 {
3598         mutex_enter(&spa->spa_async_lock);
3599         ASSERT(spa->spa_async_suspended != 0);
3600         spa->spa_async_suspended--;
3601         mutex_exit(&spa->spa_async_lock);
3602 }
3603
3604 static void
3605 spa_async_dispatch(spa_t *spa)
3606 {
3607         mutex_enter(&spa->spa_async_lock);
3608         if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3609             spa->spa_async_thread == NULL &&
3610             rootdir != NULL && !vn_is_readonly(rootdir))
3611                 spa->spa_async_thread = thread_create(NULL, 0,
3612                     spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3613         mutex_exit(&spa->spa_async_lock);
3614 }
3615
3616 void
3617 spa_async_request(spa_t *spa, int task)
3618 {
3619         mutex_enter(&spa->spa_async_lock);
3620         spa->spa_async_tasks |= task;
3621         mutex_exit(&spa->spa_async_lock);
3622 }
3623
3624 /*
3625  * ==========================================================================
3626  * SPA syncing routines
3627  * ==========================================================================
3628  */
3629
3630 static void
3631 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3632 {
3633         bplist_t *bpl = &spa->spa_sync_bplist;
3634         dmu_tx_t *tx;
3635         blkptr_t blk;
3636         uint64_t itor = 0;
3637         zio_t *zio;
3638         int error;
3639         uint8_t c = 1;
3640
3641         zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3642
3643         while (bplist_iterate(bpl, &itor, &blk) == 0) {
3644                 ASSERT(blk.blk_birth < txg);
3645                 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3646                     ZIO_FLAG_MUSTSUCCEED));
3647         }
3648
3649         error = zio_wait(zio);
3650         ASSERT3U(error, ==, 0);
3651
3652         tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3653         bplist_vacate(bpl, tx);
3654
3655         /*
3656          * Pre-dirty the first block so we sync to convergence faster.
3657          * (Usually only the first block is needed.)
3658          */
3659         dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3660         dmu_tx_commit(tx);
3661 }
3662
3663 static void
3664 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3665 {
3666         char *packed = NULL;
3667         size_t bufsize;
3668         size_t nvsize = 0;
3669         dmu_buf_t *db;
3670
3671         VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3672
3673         /*
3674          * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3675          * information.  This avoids the dbuf_will_dirty() path and
3676          * saves us a pre-read to get data we don't actually care about.
3677          */
3678         bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3679         packed = kmem_alloc(bufsize, KM_SLEEP);
3680
3681         VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3682             KM_SLEEP) == 0);
3683         bzero(packed + nvsize, bufsize - nvsize);
3684
3685         dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3686
3687         kmem_free(packed, bufsize);
3688
3689         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3690         dmu_buf_will_dirty(db, tx);
3691         *(uint64_t *)db->db_data = nvsize;
3692         dmu_buf_rele(db, FTAG);
3693 }
3694
3695 static void
3696 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3697     const char *config, const char *entry)
3698 {
3699         nvlist_t *nvroot;
3700         nvlist_t **list;
3701         int i;
3702
3703         if (!sav->sav_sync)
3704                 return;
3705
3706         /*
3707          * Update the MOS nvlist describing the list of available devices.
3708          * spa_validate_aux() will have already made sure this nvlist is
3709          * valid and the vdevs are labeled appropriately.
3710          */
3711         if (sav->sav_object == 0) {
3712                 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3713                     DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3714                     sizeof (uint64_t), tx);
3715                 VERIFY(zap_update(spa->spa_meta_objset,
3716                     DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3717                     &sav->sav_object, tx) == 0);
3718         }
3719
3720         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3721         if (sav->sav_count == 0) {
3722                 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3723         } else {
3724                 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3725                 for (i = 0; i < sav->sav_count; i++)
3726                         list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3727                             B_FALSE, B_FALSE, B_TRUE);
3728                 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3729                     sav->sav_count) == 0);
3730                 for (i = 0; i < sav->sav_count; i++)
3731                         nvlist_free(list[i]);
3732                 kmem_free(list, sav->sav_count * sizeof (void *));
3733         }
3734
3735         spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3736         nvlist_free(nvroot);
3737
3738         sav->sav_sync = B_FALSE;
3739 }
3740
3741 static void
3742 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3743 {
3744         nvlist_t *config;
3745
3746         if (list_is_empty(&spa->spa_config_dirty_list))
3747                 return;
3748
3749         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3750
3751         config = spa_config_generate(spa, spa->spa_root_vdev,
3752             dmu_tx_get_txg(tx), B_FALSE);
3753
3754         spa_config_exit(spa, SCL_STATE, FTAG);
3755
3756         if (spa->spa_config_syncing)
3757                 nvlist_free(spa->spa_config_syncing);
3758         spa->spa_config_syncing = config;
3759
3760         spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3761 }
3762
3763 /*
3764  * Set zpool properties.
3765  */
3766 static void
3767 spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3768 {
3769         spa_t *spa = arg1;
3770         objset_t *mos = spa->spa_meta_objset;
3771         nvlist_t *nvp = arg2;
3772         nvpair_t *elem;
3773         uint64_t intval;
3774         char *strval;
3775         zpool_prop_t prop;
3776         const char *propname;
3777         zprop_type_t proptype;
3778         spa_config_dirent_t *dp;
3779
3780         mutex_enter(&spa->spa_props_lock);
3781
3782         elem = NULL;
3783         while ((elem = nvlist_next_nvpair(nvp, elem))) {
3784                 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3785                 case ZPOOL_PROP_VERSION:
3786                         /*
3787                          * Only set version for non-zpool-creation cases
3788                          * (set/import). spa_create() needs special care
3789                          * for version setting.
3790                          */
3791                         if (tx->tx_txg != TXG_INITIAL) {
3792                                 VERIFY(nvpair_value_uint64(elem,
3793                                     &intval) == 0);
3794                                 ASSERT(intval <= SPA_VERSION);
3795                                 ASSERT(intval >= spa_version(spa));
3796                                 spa->spa_uberblock.ub_version = intval;
3797                                 vdev_config_dirty(spa->spa_root_vdev);
3798                         }
3799                         break;
3800
3801                 case ZPOOL_PROP_ALTROOT:
3802                         /*
3803                          * 'altroot' is a non-persistent property. It should
3804                          * have been set temporarily at creation or import time.
3805                          */
3806                         ASSERT(spa->spa_root != NULL);
3807                         break;
3808
3809                 case ZPOOL_PROP_CACHEFILE:
3810                         /*
3811                          * 'cachefile' is a non-persistent property, but note
3812                          * an async request that the config cache needs to be
3813                          * udpated.
3814                          */
3815                         VERIFY(nvpair_value_string(elem, &strval) == 0);
3816
3817                         dp = kmem_alloc(sizeof (spa_config_dirent_t), KM_SLEEP);
3818
3819                         if (strval[0] == '\0')
3820                                 dp->scd_path = spa_strdup(spa_config_path);
3821                         else if (strcmp(strval, "none") == 0)
3822                                 dp->scd_path = NULL;
3823                         else
3824                                 dp->scd_path = spa_strdup(strval);
3825
3826                         list_insert_head(&spa->spa_config_list, dp);
3827                         spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3828                         break;
3829                 default:
3830                         /*
3831                          * Set pool property values in the poolprops mos object.
3832                          */
3833                         if (spa->spa_pool_props_object == 0) {
3834                                 objset_t *mos = spa->spa_meta_objset;
3835
3836                                 VERIFY((spa->spa_pool_props_object =
3837                                     zap_create(mos, DMU_OT_POOL_PROPS,
3838                                     DMU_OT_NONE, 0, tx)) > 0);
3839
3840                                 VERIFY(zap_update(mos,
3841                                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
3842                                     8, 1, &spa->spa_pool_props_object, tx)
3843                                     == 0);
3844                         }
3845
3846                         /* normalize the property name */
3847                         propname = zpool_prop_to_name(prop);
3848                         proptype = zpool_prop_get_type(prop);
3849
3850                         if (nvpair_type(elem) == DATA_TYPE_STRING) {
3851                                 ASSERT(proptype == PROP_TYPE_STRING);
3852                                 VERIFY(nvpair_value_string(elem, &strval) == 0);
3853                                 VERIFY(zap_update(mos,
3854                                     spa->spa_pool_props_object, propname,
3855                                     1, strlen(strval) + 1, strval, tx) == 0);
3856
3857                         } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
3858                                 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
3859
3860                                 if (proptype == PROP_TYPE_INDEX) {
3861                                         const char *unused;
3862                                         VERIFY(zpool_prop_index_to_string(
3863                                             prop, intval, &unused) == 0);
3864                                 }
3865                                 VERIFY(zap_update(mos,
3866                                     spa->spa_pool_props_object, propname,
3867                                     8, 1, &intval, tx) == 0);
3868                         } else {
3869                                 ASSERT(0); /* not allowed */
3870                         }
3871
3872                         switch (prop) {
3873                         case ZPOOL_PROP_DELEGATION:
3874                                 spa->spa_delegation = intval;
3875                                 break;
3876                         case ZPOOL_PROP_BOOTFS:
3877                                 spa->spa_bootfs = intval;
3878                                 break;
3879                         case ZPOOL_PROP_FAILUREMODE:
3880                                 spa->spa_failmode = intval;
3881                                 break;
3882                         default:
3883                                 break;
3884                         }
3885                 }
3886
3887                 /* log internal history if this is not a zpool create */
3888                 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
3889                     tx->tx_txg != TXG_INITIAL) {
3890                         spa_history_internal_log(LOG_POOL_PROPSET,
3891                             spa, tx, cr, "%s %lld %s",
3892                             nvpair_name(elem), intval, spa_name(spa));
3893                 }
3894         }
3895
3896         mutex_exit(&spa->spa_props_lock);
3897 }
3898
3899 /*
3900  * Sync the specified transaction group.  New blocks may be dirtied as
3901  * part of the process, so we iterate until it converges.
3902  */
3903 void
3904 spa_sync(spa_t *spa, uint64_t txg)
3905 {
3906         dsl_pool_t *dp = spa->spa_dsl_pool;
3907         objset_t *mos = spa->spa_meta_objset;
3908         bplist_t *bpl = &spa->spa_sync_bplist;
3909         vdev_t *rvd = spa->spa_root_vdev;
3910         vdev_t *vd;
3911         dmu_tx_t *tx;
3912         int dirty_vdevs;
3913         int error;
3914
3915         /*
3916          * Lock out configuration changes.
3917          */
3918         spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3919
3920         spa->spa_syncing_txg = txg;
3921         spa->spa_sync_pass = 0;
3922
3923         /*
3924          * If there are any pending vdev state changes, convert them
3925          * into config changes that go out with this transaction group.
3926          */
3927         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3928         while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
3929                 vdev_state_clean(vd);
3930                 vdev_config_dirty(vd);
3931         }
3932         spa_config_exit(spa, SCL_STATE, FTAG);
3933
3934         VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
3935
3936         tx = dmu_tx_create_assigned(dp, txg);
3937
3938         /*
3939          * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
3940          * set spa_deflate if we have no raid-z vdevs.
3941          */
3942         if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
3943             spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
3944                 int i;
3945
3946                 for (i = 0; i < rvd->vdev_children; i++) {
3947                         vd = rvd->vdev_child[i];
3948                         if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
3949                                 break;
3950                 }
3951                 if (i == rvd->vdev_children) {
3952                         spa->spa_deflate = TRUE;
3953                         VERIFY(0 == zap_add(spa->spa_meta_objset,
3954                             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3955                             sizeof (uint64_t), 1, &spa->spa_deflate, tx));
3956                 }
3957         }
3958
3959         if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
3960             spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
3961                 dsl_pool_create_origin(dp, tx);
3962
3963                 /* Keeping the origin open increases spa_minref */
3964                 spa->spa_minref += 3;
3965         }
3966
3967         if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
3968             spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
3969                 dsl_pool_upgrade_clones(dp, tx);
3970         }
3971
3972         /*
3973          * If anything has changed in this txg, push the deferred frees
3974          * from the previous txg.  If not, leave them alone so that we
3975          * don't generate work on an otherwise idle system.
3976          */
3977         if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
3978             !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
3979             !txg_list_empty(&dp->dp_sync_tasks, txg))
3980                 spa_sync_deferred_frees(spa, txg);
3981
3982         /*
3983          * Iterate to convergence.
3984          */
3985         do {
3986                 spa->spa_sync_pass++;
3987
3988                 spa_sync_config_object(spa, tx);
3989                 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
3990                     ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
3991                 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
3992                     ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
3993                 spa_errlog_sync(spa, txg);
3994                 dsl_pool_sync(dp, txg);
3995
3996                 dirty_vdevs = 0;
3997                 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
3998                         vdev_sync(vd, txg);
3999                         dirty_vdevs++;
4000                 }
4001
4002                 bplist_sync(bpl, tx);
4003         } while (dirty_vdevs);
4004
4005         bplist_close(bpl);
4006
4007         dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4008
4009         /*
4010          * Rewrite the vdev configuration (which includes the uberblock)
4011          * to commit the transaction group.
4012          *
4013          * If there are no dirty vdevs, we sync the uberblock to a few
4014          * random top-level vdevs that are known to be visible in the
4015          * config cache (see spa_vdev_add() for a complete description).
4016          * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4017          */
4018         for (;;) {
4019                 /*
4020                  * We hold SCL_STATE to prevent vdev open/close/etc.
4021                  * while we're attempting to write the vdev labels.
4022                  */
4023                 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4024
4025                 if (list_is_empty(&spa->spa_config_dirty_list)) {
4026                         vdev_t *svd[SPA_DVAS_PER_BP];
4027                         int svdcount = 0;
4028                         int children = rvd->vdev_children;
4029                         int c0 = spa_get_random(children);
4030                         int c;
4031
4032                         for (c = 0; c < children; c++) {
4033                                 vd = rvd->vdev_child[(c0 + c) % children];
4034                                 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4035                                         continue;
4036                                 svd[svdcount++] = vd;
4037                                 if (svdcount == SPA_DVAS_PER_BP)
4038                                         break;
4039                         }
4040                         error = vdev_config_sync(svd, svdcount, txg);
4041                 } else {
4042                         error = vdev_config_sync(rvd->vdev_child,
4043                             rvd->vdev_children, txg);
4044                 }
4045
4046                 spa_config_exit(spa, SCL_STATE, FTAG);
4047
4048                 if (error == 0)
4049                         break;
4050                 zio_suspend(spa, NULL);
4051                 zio_resume_wait(spa);
4052         }
4053         dmu_tx_commit(tx);
4054
4055         /*
4056          * Clear the dirty config list.
4057          */
4058         while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4059                 vdev_config_clean(vd);
4060
4061         /*
4062          * Now that the new config has synced transactionally,
4063          * let it become visible to the config cache.
4064          */
4065         if (spa->spa_config_syncing != NULL) {
4066                 spa_config_set(spa, spa->spa_config_syncing);
4067                 spa->spa_config_txg = txg;
4068                 spa->spa_config_syncing = NULL;
4069         }
4070
4071         spa->spa_ubsync = spa->spa_uberblock;
4072
4073         /*
4074          * Clean up the ZIL records for the synced txg.
4075          */
4076         dsl_pool_zil_clean(dp);
4077
4078         /*
4079          * Update usable space statistics.
4080          */
4081         while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4082                 vdev_sync_done(vd, txg);
4083
4084         /*
4085          * It had better be the case that we didn't dirty anything
4086          * since vdev_config_sync().
4087          */
4088         ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4089         ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4090         ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4091         ASSERT(bpl->bpl_queue == NULL);
4092
4093         spa_config_exit(spa, SCL_CONFIG, FTAG);
4094
4095         /*
4096          * If any async tasks have been requested, kick them off.
4097          */
4098         spa_async_dispatch(spa);
4099 }
4100
4101 /*
4102  * Sync all pools.  We don't want to hold the namespace lock across these
4103  * operations, so we take a reference on the spa_t and drop the lock during the
4104  * sync.
4105  */
4106 void
4107 spa_sync_allpools(void)
4108 {
4109         spa_t *spa = NULL;
4110         mutex_enter(&spa_namespace_lock);
4111         while ((spa = spa_next(spa)) != NULL) {
4112                 if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4113                         continue;
4114                 spa_open_ref(spa, FTAG);
4115                 mutex_exit(&spa_namespace_lock);
4116                 txg_wait_synced(spa_get_dsl(spa), 0);
4117                 mutex_enter(&spa_namespace_lock);
4118                 spa_close(spa, FTAG);
4119         }
4120         mutex_exit(&spa_namespace_lock);
4121 }
4122
4123 /*
4124  * ==========================================================================
4125  * Miscellaneous routines
4126  * ==========================================================================
4127  */
4128
4129 /*
4130  * Remove all pools in the system.
4131  */
4132 void
4133 spa_evict_all(void)
4134 {
4135         spa_t *spa;
4136
4137         /*
4138          * Remove all cached state.  All pools should be closed now,
4139          * so every spa in the AVL tree should be unreferenced.
4140          */
4141         mutex_enter(&spa_namespace_lock);
4142         while ((spa = spa_next(NULL)) != NULL) {
4143                 /*
4144                  * Stop async tasks.  The async thread may need to detach
4145                  * a device that's been replaced, which requires grabbing
4146                  * spa_namespace_lock, so we must drop it here.
4147                  */
4148                 spa_open_ref(spa, FTAG);
4149                 mutex_exit(&spa_namespace_lock);
4150                 spa_async_suspend(spa);
4151                 mutex_enter(&spa_namespace_lock);
4152                 spa_close(spa, FTAG);
4153
4154                 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4155                         spa_unload(spa);
4156                         spa_deactivate(spa);
4157                 }
4158                 spa_remove(spa);
4159         }
4160         mutex_exit(&spa_namespace_lock);
4161 }
4162
4163 vdev_t *
4164 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t l2cache)
4165 {
4166         vdev_t *vd;
4167         int i;
4168
4169         if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4170                 return (vd);
4171
4172         if (l2cache) {
4173                 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4174                         vd = spa->spa_l2cache.sav_vdevs[i];
4175                         if (vd->vdev_guid == guid)
4176                                 return (vd);
4177                 }
4178         }
4179
4180         return (NULL);
4181 }
4182
4183 void
4184 spa_upgrade(spa_t *spa, uint64_t version)
4185 {
4186         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4187
4188         /*
4189          * This should only be called for a non-faulted pool, and since a
4190          * future version would result in an unopenable pool, this shouldn't be
4191          * possible.
4192          */
4193         ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4194         ASSERT(version >= spa->spa_uberblock.ub_version);
4195
4196         spa->spa_uberblock.ub_version = version;
4197         vdev_config_dirty(spa->spa_root_vdev);
4198
4199         spa_config_exit(spa, SCL_ALL, FTAG);
4200
4201         txg_wait_synced(spa_get_dsl(spa), 0);
4202 }
4203
4204 boolean_t
4205 spa_has_spare(spa_t *spa, uint64_t guid)
4206 {
4207         int i;
4208         uint64_t spareguid;
4209         spa_aux_vdev_t *sav = &spa->spa_spares;
4210
4211         for (i = 0; i < sav->sav_count; i++)
4212                 if (sav->sav_vdevs[i]->vdev_guid == guid)
4213                         return (B_TRUE);
4214
4215         for (i = 0; i < sav->sav_npending; i++) {
4216                 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4217                     &spareguid) == 0 && spareguid == guid)
4218                         return (B_TRUE);
4219         }
4220
4221         return (B_FALSE);
4222 }
4223
4224 /*
4225  * Check if a pool has an active shared spare device.
4226  * Note: reference count of an active spare is 2, as a spare and as a replace
4227  */
4228 static boolean_t
4229 spa_has_active_shared_spare(spa_t *spa)
4230 {
4231         int i, refcnt;
4232         uint64_t pool;
4233         spa_aux_vdev_t *sav = &spa->spa_spares;
4234
4235         for (i = 0; i < sav->sav_count; i++) {
4236                 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4237                     &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4238                     refcnt > 2)
4239                         return (B_TRUE);
4240         }
4241
4242         return (B_FALSE);
4243 }
4244
4245 /*
4246  * Post a sysevent corresponding to the given event.  The 'name' must be one of
4247  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4248  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4249  * in the userland libzpool, as we don't want consumers to misinterpret ztest
4250  * or zdb as real changes.
4251  */
4252 void
4253 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4254 {
4255 #ifndef __NetBSD__
4256 #ifdef _KERNEL
4257         sysevent_t              *ev;
4258         sysevent_attr_list_t    *attr = NULL;
4259         sysevent_value_t        value;
4260         sysevent_id_t           eid;
4261
4262         ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4263             SE_SLEEP);
4264
4265         value.value_type = SE_DATA_TYPE_STRING;
4266         value.value.sv_string = spa_name(spa);
4267         if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4268                 goto done;
4269
4270         value.value_type = SE_DATA_TYPE_UINT64;
4271         value.value.sv_uint64 = spa_guid(spa);
4272         if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4273                 goto done;
4274
4275         if (vd) {
4276                 value.value_type = SE_DATA_TYPE_UINT64;
4277                 value.value.sv_uint64 = vd->vdev_guid;
4278                 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4279                     SE_SLEEP) != 0)
4280                         goto done;
4281
4282                 if (vd->vdev_path) {
4283                         value.value_type = SE_DATA_TYPE_STRING;
4284                         value.value.sv_string = vd->vdev_path;
4285                         if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4286                             &value, SE_SLEEP) != 0)
4287                                 goto done;
4288                 }
4289         }
4290
4291         if (sysevent_attach_attributes(ev, attr) != 0)
4292                 goto done;
4293         attr = NULL;
4294
4295         (void) log_sysevent(ev, SE_SLEEP, &eid);
4296
4297 done:
4298         if (attr)
4299                 sysevent_free_attr(attr);
4300         sysevent_free(ev);
4301 #endif
4302 #endif /* __NetBSD__ */
4303 }