142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
146};
147
148static void spa_sync_version(void *arg, dmu_tx_t *tx);
149static void spa_sync_props(void *arg, dmu_tx_t *tx);
150static boolean_t spa_has_active_shared_spare(spa_t *spa);
151static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
153 char **ereport);
154static void spa_vdev_resilver_done(spa_t *spa);
155
156uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
157#ifdef PSRSET_BIND
158id_t zio_taskq_psrset_bind = PS_NONE;
159#endif
160#ifdef SYSDC
161boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
162#endif
163uint_t zio_taskq_basedc = 80; /* base duty cycle */
164
165boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
166extern int zfs_sync_pass_deferred_free;
167
168#ifndef illumos
169extern void spa_deadman(void *arg);
170#endif
171
172/*
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
175 */
176#define TRYIMPORT_NAME "$import"
177
178/*
179 * ==========================================================================
180 * SPA properties routines
181 * ==========================================================================
182 */
183
184/*
185 * Add a (source=src, propname=propval) list to an nvlist.
186 */
187static void
188spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
189 uint64_t intval, zprop_source_t src)
190{
191 const char *propname = zpool_prop_to_name(prop);
192 nvlist_t *propval;
193
194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
196
197 if (strval != NULL)
198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
199 else
200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
201
202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
203 nvlist_free(propval);
204}
205
206/*
207 * Get property values from the spa configuration.
208 */
209static void
210spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
211{
212 vdev_t *rvd = spa->spa_root_vdev;
213 dsl_pool_t *pool = spa->spa_dsl_pool;
214 uint64_t size;
215 uint64_t alloc;
216 uint64_t space;
217 uint64_t cap, version;
218 zprop_source_t src = ZPROP_SRC_NONE;
219 spa_config_dirent_t *dp;
220
221 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
222
223 if (rvd != NULL) {
224 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
225 size = metaslab_class_get_space(spa_normal_class(spa));
226 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
230 size - alloc, src);
231
232 space = 0;
233 for (int c = 0; c < rvd->vdev_children; c++) {
234 vdev_t *tvd = rvd->vdev_child[c];
235 space += tvd->vdev_max_asize - tvd->vdev_asize;
236 }
237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
238 src);
239
240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
241 (spa_mode(spa) == FREAD), src);
242
243 cap = (size == 0) ? 0 : (alloc * 100 / size);
244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
245
246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
247 ddt_get_pool_dedup_ratio(spa), src);
248
249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
250 rvd->vdev_state, src);
251
252 version = spa_version(spa);
253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
254 src = ZPROP_SRC_DEFAULT;
255 else
256 src = ZPROP_SRC_LOCAL;
257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
258 }
259
260 if (pool != NULL) {
261 dsl_dir_t *freedir = pool->dp_free_dir;
262
263 /*
264 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
265 * when opening pools before this version freedir will be NULL.
266 */
267 if (freedir != NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
269 freedir->dd_phys->dd_used_bytes, src);
270 } else {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
272 NULL, 0, src);
273 }
274 }
275
276 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
277
278 if (spa->spa_comment != NULL) {
279 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
280 0, ZPROP_SRC_LOCAL);
281 }
282
283 if (spa->spa_root != NULL)
284 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
285 0, ZPROP_SRC_LOCAL);
286
287 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
288 if (dp->scd_path == NULL) {
289 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
290 "none", 0, ZPROP_SRC_LOCAL);
291 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
292 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
293 dp->scd_path, 0, ZPROP_SRC_LOCAL);
294 }
295 }
296}
297
298/*
299 * Get zpool property values.
300 */
301int
302spa_prop_get(spa_t *spa, nvlist_t **nvp)
303{
304 objset_t *mos = spa->spa_meta_objset;
305 zap_cursor_t zc;
306 zap_attribute_t za;
307 int err;
308
309 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
310
311 mutex_enter(&spa->spa_props_lock);
312
313 /*
314 * Get properties from the spa config.
315 */
316 spa_prop_get_config(spa, nvp);
317
318 /* If no pool property object, no more prop to get. */
319 if (mos == NULL || spa->spa_pool_props_object == 0) {
320 mutex_exit(&spa->spa_props_lock);
321 return (0);
322 }
323
324 /*
325 * Get properties from the MOS pool property object.
326 */
327 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
328 (err = zap_cursor_retrieve(&zc, &za)) == 0;
329 zap_cursor_advance(&zc)) {
330 uint64_t intval = 0;
331 char *strval = NULL;
332 zprop_source_t src = ZPROP_SRC_DEFAULT;
333 zpool_prop_t prop;
334
335 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
336 continue;
337
338 switch (za.za_integer_length) {
339 case 8:
340 /* integer property */
341 if (za.za_first_integer !=
342 zpool_prop_default_numeric(prop))
343 src = ZPROP_SRC_LOCAL;
344
345 if (prop == ZPOOL_PROP_BOOTFS) {
346 dsl_pool_t *dp;
347 dsl_dataset_t *ds = NULL;
348
349 dp = spa_get_dsl(spa);
350 dsl_pool_config_enter(dp, FTAG);
351 if (err = dsl_dataset_hold_obj(dp,
352 za.za_first_integer, FTAG, &ds)) {
353 dsl_pool_config_exit(dp, FTAG);
354 break;
355 }
356
357 strval = kmem_alloc(
358 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
359 KM_SLEEP);
360 dsl_dataset_name(ds, strval);
361 dsl_dataset_rele(ds, FTAG);
362 dsl_pool_config_exit(dp, FTAG);
363 } else {
364 strval = NULL;
365 intval = za.za_first_integer;
366 }
367
368 spa_prop_add_list(*nvp, prop, strval, intval, src);
369
370 if (strval != NULL)
371 kmem_free(strval,
372 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
373
374 break;
375
376 case 1:
377 /* string property */
378 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
379 err = zap_lookup(mos, spa->spa_pool_props_object,
380 za.za_name, 1, za.za_num_integers, strval);
381 if (err) {
382 kmem_free(strval, za.za_num_integers);
383 break;
384 }
385 spa_prop_add_list(*nvp, prop, strval, 0, src);
386 kmem_free(strval, za.za_num_integers);
387 break;
388
389 default:
390 break;
391 }
392 }
393 zap_cursor_fini(&zc);
394 mutex_exit(&spa->spa_props_lock);
395out:
396 if (err && err != ENOENT) {
397 nvlist_free(*nvp);
398 *nvp = NULL;
399 return (err);
400 }
401
402 return (0);
403}
404
405/*
406 * Validate the given pool properties nvlist and modify the list
407 * for the property values to be set.
408 */
409static int
410spa_prop_validate(spa_t *spa, nvlist_t *props)
411{
412 nvpair_t *elem;
413 int error = 0, reset_bootfs = 0;
414 uint64_t objnum = 0;
415 boolean_t has_feature = B_FALSE;
416
417 elem = NULL;
418 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
419 uint64_t intval;
420 char *strval, *slash, *check, *fname;
421 const char *propname = nvpair_name(elem);
422 zpool_prop_t prop = zpool_name_to_prop(propname);
423
424 switch (prop) {
425 case ZPROP_INVAL:
426 if (!zpool_prop_feature(propname)) {
427 error = SET_ERROR(EINVAL);
428 break;
429 }
430
431 /*
432 * Sanitize the input.
433 */
434 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
435 error = SET_ERROR(EINVAL);
436 break;
437 }
438
439 if (nvpair_value_uint64(elem, &intval) != 0) {
440 error = SET_ERROR(EINVAL);
441 break;
442 }
443
444 if (intval != 0) {
445 error = SET_ERROR(EINVAL);
446 break;
447 }
448
449 fname = strchr(propname, '@') + 1;
450 if (zfeature_lookup_name(fname, NULL) != 0) {
451 error = SET_ERROR(EINVAL);
452 break;
453 }
454
455 has_feature = B_TRUE;
456 break;
457
458 case ZPOOL_PROP_VERSION:
459 error = nvpair_value_uint64(elem, &intval);
460 if (!error &&
461 (intval < spa_version(spa) ||
462 intval > SPA_VERSION_BEFORE_FEATURES ||
463 has_feature))
464 error = SET_ERROR(EINVAL);
465 break;
466
467 case ZPOOL_PROP_DELEGATION:
468 case ZPOOL_PROP_AUTOREPLACE:
469 case ZPOOL_PROP_LISTSNAPS:
470 case ZPOOL_PROP_AUTOEXPAND:
471 error = nvpair_value_uint64(elem, &intval);
472 if (!error && intval > 1)
473 error = SET_ERROR(EINVAL);
474 break;
475
476 case ZPOOL_PROP_BOOTFS:
477 /*
478 * If the pool version is less than SPA_VERSION_BOOTFS,
479 * or the pool is still being created (version == 0),
480 * the bootfs property cannot be set.
481 */
482 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
483 error = SET_ERROR(ENOTSUP);
484 break;
485 }
486
487 /*
488 * Make sure the vdev config is bootable
489 */
490 if (!vdev_is_bootable(spa->spa_root_vdev)) {
491 error = SET_ERROR(ENOTSUP);
492 break;
493 }
494
495 reset_bootfs = 1;
496
497 error = nvpair_value_string(elem, &strval);
498
499 if (!error) {
500 objset_t *os;
501 uint64_t compress;
502
503 if (strval == NULL || strval[0] == '\0') {
504 objnum = zpool_prop_default_numeric(
505 ZPOOL_PROP_BOOTFS);
506 break;
507 }
508
509 if (error = dmu_objset_hold(strval, FTAG, &os))
510 break;
511
512 /* Must be ZPL and not gzip compressed. */
513
514 if (dmu_objset_type(os) != DMU_OST_ZFS) {
515 error = SET_ERROR(ENOTSUP);
516 } else if ((error =
517 dsl_prop_get_int_ds(dmu_objset_ds(os),
518 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
519 &compress)) == 0 &&
520 !BOOTFS_COMPRESS_VALID(compress)) {
521 error = SET_ERROR(ENOTSUP);
522 } else {
523 objnum = dmu_objset_id(os);
524 }
525 dmu_objset_rele(os, FTAG);
526 }
527 break;
528
529 case ZPOOL_PROP_FAILUREMODE:
530 error = nvpair_value_uint64(elem, &intval);
531 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
532 intval > ZIO_FAILURE_MODE_PANIC))
533 error = SET_ERROR(EINVAL);
534
535 /*
536 * This is a special case which only occurs when
537 * the pool has completely failed. This allows
538 * the user to change the in-core failmode property
539 * without syncing it out to disk (I/Os might
540 * currently be blocked). We do this by returning
541 * EIO to the caller (spa_prop_set) to trick it
542 * into thinking we encountered a property validation
543 * error.
544 */
545 if (!error && spa_suspended(spa)) {
546 spa->spa_failmode = intval;
547 error = SET_ERROR(EIO);
548 }
549 break;
550
551 case ZPOOL_PROP_CACHEFILE:
552 if ((error = nvpair_value_string(elem, &strval)) != 0)
553 break;
554
555 if (strval[0] == '\0')
556 break;
557
558 if (strcmp(strval, "none") == 0)
559 break;
560
561 if (strval[0] != '/') {
562 error = SET_ERROR(EINVAL);
563 break;
564 }
565
566 slash = strrchr(strval, '/');
567 ASSERT(slash != NULL);
568
569 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
570 strcmp(slash, "/..") == 0)
571 error = SET_ERROR(EINVAL);
572 break;
573
574 case ZPOOL_PROP_COMMENT:
575 if ((error = nvpair_value_string(elem, &strval)) != 0)
576 break;
577 for (check = strval; *check != '\0'; check++) {
578 /*
579 * The kernel doesn't have an easy isprint()
580 * check. For this kernel check, we merely
581 * check ASCII apart from DEL. Fix this if
582 * there is an easy-to-use kernel isprint().
583 */
584 if (*check >= 0x7f) {
585 error = SET_ERROR(EINVAL);
586 break;
587 }
588 check++;
589 }
590 if (strlen(strval) > ZPROP_MAX_COMMENT)
591 error = E2BIG;
592 break;
593
594 case ZPOOL_PROP_DEDUPDITTO:
595 if (spa_version(spa) < SPA_VERSION_DEDUP)
596 error = SET_ERROR(ENOTSUP);
597 else
598 error = nvpair_value_uint64(elem, &intval);
599 if (error == 0 &&
600 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
601 error = SET_ERROR(EINVAL);
602 break;
603 }
604
605 if (error)
606 break;
607 }
608
609 if (!error && reset_bootfs) {
610 error = nvlist_remove(props,
611 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
612
613 if (!error) {
614 error = nvlist_add_uint64(props,
615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
616 }
617 }
618
619 return (error);
620}
621
622void
623spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
624{
625 char *cachefile;
626 spa_config_dirent_t *dp;
627
628 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
629 &cachefile) != 0)
630 return;
631
632 dp = kmem_alloc(sizeof (spa_config_dirent_t),
633 KM_SLEEP);
634
635 if (cachefile[0] == '\0')
636 dp->scd_path = spa_strdup(spa_config_path);
637 else if (strcmp(cachefile, "none") == 0)
638 dp->scd_path = NULL;
639 else
640 dp->scd_path = spa_strdup(cachefile);
641
642 list_insert_head(&spa->spa_config_list, dp);
643 if (need_sync)
644 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
645}
646
647int
648spa_prop_set(spa_t *spa, nvlist_t *nvp)
649{
650 int error;
651 nvpair_t *elem = NULL;
652 boolean_t need_sync = B_FALSE;
653
654 if ((error = spa_prop_validate(spa, nvp)) != 0)
655 return (error);
656
657 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
658 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
659
660 if (prop == ZPOOL_PROP_CACHEFILE ||
661 prop == ZPOOL_PROP_ALTROOT ||
662 prop == ZPOOL_PROP_READONLY)
663 continue;
664
665 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
666 uint64_t ver;
667
668 if (prop == ZPOOL_PROP_VERSION) {
669 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
670 } else {
671 ASSERT(zpool_prop_feature(nvpair_name(elem)));
672 ver = SPA_VERSION_FEATURES;
673 need_sync = B_TRUE;
674 }
675
676 /* Save time if the version is already set. */
677 if (ver == spa_version(spa))
678 continue;
679
680 /*
681 * In addition to the pool directory object, we might
682 * create the pool properties object, the features for
683 * read object, the features for write object, or the
684 * feature descriptions object.
685 */
686 error = dsl_sync_task(spa->spa_name, NULL,
687 spa_sync_version, &ver, 6);
688 if (error)
689 return (error);
690 continue;
691 }
692
693 need_sync = B_TRUE;
694 break;
695 }
696
697 if (need_sync) {
698 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
699 nvp, 6));
700 }
701
702 return (0);
703}
704
705/*
706 * If the bootfs property value is dsobj, clear it.
707 */
708void
709spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
710{
711 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
712 VERIFY(zap_remove(spa->spa_meta_objset,
713 spa->spa_pool_props_object,
714 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
715 spa->spa_bootfs = 0;
716 }
717}
718
719/*ARGSUSED*/
720static int
721spa_change_guid_check(void *arg, dmu_tx_t *tx)
722{
723 uint64_t *newguid = arg;
724 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
725 vdev_t *rvd = spa->spa_root_vdev;
726 uint64_t vdev_state;
727
728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
729 vdev_state = rvd->vdev_state;
730 spa_config_exit(spa, SCL_STATE, FTAG);
731
732 if (vdev_state != VDEV_STATE_HEALTHY)
733 return (SET_ERROR(ENXIO));
734
735 ASSERT3U(spa_guid(spa), !=, *newguid);
736
737 return (0);
738}
739
740static void
741spa_change_guid_sync(void *arg, dmu_tx_t *tx)
742{
743 uint64_t *newguid = arg;
744 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
745 uint64_t oldguid;
746 vdev_t *rvd = spa->spa_root_vdev;
747
748 oldguid = spa_guid(spa);
749
750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
751 rvd->vdev_guid = *newguid;
752 rvd->vdev_guid_sum += (*newguid - oldguid);
753 vdev_config_dirty(rvd);
754 spa_config_exit(spa, SCL_STATE, FTAG);
755
756 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
757 oldguid, *newguid);
758}
759
760/*
761 * Change the GUID for the pool. This is done so that we can later
762 * re-import a pool built from a clone of our own vdevs. We will modify
763 * the root vdev's guid, our own pool guid, and then mark all of our
764 * vdevs dirty. Note that we must make sure that all our vdevs are
765 * online when we do this, or else any vdevs that weren't present
766 * would be orphaned from our pool. We are also going to issue a
767 * sysevent to update any watchers.
768 */
769int
770spa_change_guid(spa_t *spa)
771{
772 int error;
773 uint64_t guid;
774
775 mutex_enter(&spa->spa_vdev_top_lock);
776 mutex_enter(&spa_namespace_lock);
777 guid = spa_generate_guid(NULL);
778
779 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
780 spa_change_guid_sync, &guid, 5);
781
782 if (error == 0) {
783 spa_config_sync(spa, B_FALSE, B_TRUE);
784 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
785 }
786
787 mutex_exit(&spa_namespace_lock);
788 mutex_exit(&spa->spa_vdev_top_lock);
789
790 return (error);
791}
792
793/*
794 * ==========================================================================
795 * SPA state manipulation (open/create/destroy/import/export)
796 * ==========================================================================
797 */
798
799static int
800spa_error_entry_compare(const void *a, const void *b)
801{
802 spa_error_entry_t *sa = (spa_error_entry_t *)a;
803 spa_error_entry_t *sb = (spa_error_entry_t *)b;
804 int ret;
805
806 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
807 sizeof (zbookmark_t));
808
809 if (ret < 0)
810 return (-1);
811 else if (ret > 0)
812 return (1);
813 else
814 return (0);
815}
816
817/*
818 * Utility function which retrieves copies of the current logs and
819 * re-initializes them in the process.
820 */
821void
822spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
823{
824 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
825
826 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
827 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
828
829 avl_create(&spa->spa_errlist_scrub,
830 spa_error_entry_compare, sizeof (spa_error_entry_t),
831 offsetof(spa_error_entry_t, se_avl));
832 avl_create(&spa->spa_errlist_last,
833 spa_error_entry_compare, sizeof (spa_error_entry_t),
834 offsetof(spa_error_entry_t, se_avl));
835}
836
837static void
838spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
839{
840 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
841 enum zti_modes mode = ztip->zti_mode;
842 uint_t value = ztip->zti_value;
843 uint_t count = ztip->zti_count;
844 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
845 char name[32];
846 uint_t flags = 0;
847 boolean_t batch = B_FALSE;
848
849 if (mode == ZTI_MODE_NULL) {
850 tqs->stqs_count = 0;
851 tqs->stqs_taskq = NULL;
852 return;
853 }
854
855 ASSERT3U(count, >, 0);
856
857 tqs->stqs_count = count;
858 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
859
860 switch (mode) {
861 case ZTI_MODE_FIXED:
862 ASSERT3U(value, >=, 1);
863 value = MAX(value, 1);
864 break;
865
866 case ZTI_MODE_BATCH:
867 batch = B_TRUE;
868 flags |= TASKQ_THREADS_CPU_PCT;
869 value = zio_taskq_batch_pct;
870 break;
871
872 default:
873 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
874 "spa_activate()",
875 zio_type_name[t], zio_taskq_types[q], mode, value);
876 break;
877 }
878
879 for (uint_t i = 0; i < count; i++) {
880 taskq_t *tq;
881
882 if (count > 1) {
883 (void) snprintf(name, sizeof (name), "%s_%s_%u",
884 zio_type_name[t], zio_taskq_types[q], i);
885 } else {
886 (void) snprintf(name, sizeof (name), "%s_%s",
887 zio_type_name[t], zio_taskq_types[q]);
888 }
889
890#ifdef SYSDC
891 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
892 if (batch)
893 flags |= TASKQ_DC_BATCH;
894
895 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
896 spa->spa_proc, zio_taskq_basedc, flags);
897 } else {
898#endif
899 pri_t pri = maxclsyspri;
900 /*
901 * The write issue taskq can be extremely CPU
902 * intensive. Run it at slightly lower priority
903 * than the other taskqs.
904 */
905 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
906 pri--;
907
908 tq = taskq_create_proc(name, value, pri, 50,
909 INT_MAX, spa->spa_proc, flags);
910#ifdef SYSDC
911 }
912#endif
913
914 tqs->stqs_taskq[i] = tq;
915 }
916}
917
918static void
919spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
920{
921 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
922
923 if (tqs->stqs_taskq == NULL) {
924 ASSERT0(tqs->stqs_count);
925 return;
926 }
927
928 for (uint_t i = 0; i < tqs->stqs_count; i++) {
929 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
930 taskq_destroy(tqs->stqs_taskq[i]);
931 }
932
933 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
934 tqs->stqs_taskq = NULL;
935}
936
937/*
938 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
939 * Note that a type may have multiple discrete taskqs to avoid lock contention
940 * on the taskq itself. In that case we choose which taskq at random by using
941 * the low bits of gethrtime().
942 */
943void
944spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
945 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
946{
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
948 taskq_t *tq;
949
950 ASSERT3P(tqs->stqs_taskq, !=, NULL);
951 ASSERT3U(tqs->stqs_count, !=, 0);
952
953 if (tqs->stqs_count == 1) {
954 tq = tqs->stqs_taskq[0];
955 } else {
956 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
957 }
958
959 taskq_dispatch_ent(tq, func, arg, flags, ent);
960}
961
962static void
963spa_create_zio_taskqs(spa_t *spa)
964{
965 for (int t = 0; t < ZIO_TYPES; t++) {
966 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
967 spa_taskqs_init(spa, t, q);
968 }
969 }
970}
971
972#ifdef _KERNEL
973#ifdef SPA_PROCESS
974static void
975spa_thread(void *arg)
976{
977 callb_cpr_t cprinfo;
978
979 spa_t *spa = arg;
980 user_t *pu = PTOU(curproc);
981
982 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
983 spa->spa_name);
984
985 ASSERT(curproc != &p0);
986 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
987 "zpool-%s", spa->spa_name);
988 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
989
990#ifdef PSRSET_BIND
991 /* bind this thread to the requested psrset */
992 if (zio_taskq_psrset_bind != PS_NONE) {
993 pool_lock();
994 mutex_enter(&cpu_lock);
995 mutex_enter(&pidlock);
996 mutex_enter(&curproc->p_lock);
997
998 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
999 0, NULL, NULL) == 0) {
1000 curthread->t_bind_pset = zio_taskq_psrset_bind;
1001 } else {
1002 cmn_err(CE_WARN,
1003 "Couldn't bind process for zfs pool \"%s\" to "
1004 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1005 }
1006
1007 mutex_exit(&curproc->p_lock);
1008 mutex_exit(&pidlock);
1009 mutex_exit(&cpu_lock);
1010 pool_unlock();
1011 }
1012#endif
1013
1014#ifdef SYSDC
1015 if (zio_taskq_sysdc) {
1016 sysdc_thread_enter(curthread, 100, 0);
1017 }
1018#endif
1019
1020 spa->spa_proc = curproc;
1021 spa->spa_did = curthread->t_did;
1022
1023 spa_create_zio_taskqs(spa);
1024
1025 mutex_enter(&spa->spa_proc_lock);
1026 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1027
1028 spa->spa_proc_state = SPA_PROC_ACTIVE;
1029 cv_broadcast(&spa->spa_proc_cv);
1030
1031 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1032 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1033 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1034 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1035
1036 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1037 spa->spa_proc_state = SPA_PROC_GONE;
1038 spa->spa_proc = &p0;
1039 cv_broadcast(&spa->spa_proc_cv);
1040 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1041
1042 mutex_enter(&curproc->p_lock);
1043 lwp_exit();
1044}
1045#endif /* SPA_PROCESS */
1046#endif
1047
1048/*
1049 * Activate an uninitialized pool.
1050 */
1051static void
1052spa_activate(spa_t *spa, int mode)
1053{
1054 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1055
1056 spa->spa_state = POOL_STATE_ACTIVE;
1057 spa->spa_mode = mode;
1058
1059 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1060 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1061
1062 /* Try to create a covering process */
1063 mutex_enter(&spa->spa_proc_lock);
1064 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1065 ASSERT(spa->spa_proc == &p0);
1066 spa->spa_did = 0;
1067
1068#ifdef SPA_PROCESS
1069 /* Only create a process if we're going to be around a while. */
1070 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1071 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1072 NULL, 0) == 0) {
1073 spa->spa_proc_state = SPA_PROC_CREATED;
1074 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1075 cv_wait(&spa->spa_proc_cv,
1076 &spa->spa_proc_lock);
1077 }
1078 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1079 ASSERT(spa->spa_proc != &p0);
1080 ASSERT(spa->spa_did != 0);
1081 } else {
1082#ifdef _KERNEL
1083 cmn_err(CE_WARN,
1084 "Couldn't create process for zfs pool \"%s\"\n",
1085 spa->spa_name);
1086#endif
1087 }
1088 }
1089#endif /* SPA_PROCESS */
1090 mutex_exit(&spa->spa_proc_lock);
1091
1092 /* If we didn't create a process, we need to create our taskqs. */
1093 ASSERT(spa->spa_proc == &p0);
1094 if (spa->spa_proc == &p0) {
1095 spa_create_zio_taskqs(spa);
1096 }
1097
1098 /*
1099 * Start TRIM thread.
1100 */
1101 trim_thread_create(spa);
1102
1103 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1104 offsetof(vdev_t, vdev_config_dirty_node));
1105 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1106 offsetof(vdev_t, vdev_state_dirty_node));
1107
1108 txg_list_create(&spa->spa_vdev_txg_list,
1109 offsetof(struct vdev, vdev_txg_node));
1110
1111 avl_create(&spa->spa_errlist_scrub,
1112 spa_error_entry_compare, sizeof (spa_error_entry_t),
1113 offsetof(spa_error_entry_t, se_avl));
1114 avl_create(&spa->spa_errlist_last,
1115 spa_error_entry_compare, sizeof (spa_error_entry_t),
1116 offsetof(spa_error_entry_t, se_avl));
1117}
1118
1119/*
1120 * Opposite of spa_activate().
1121 */
1122static void
1123spa_deactivate(spa_t *spa)
1124{
1125 ASSERT(spa->spa_sync_on == B_FALSE);
1126 ASSERT(spa->spa_dsl_pool == NULL);
1127 ASSERT(spa->spa_root_vdev == NULL);
1128 ASSERT(spa->spa_async_zio_root == NULL);
1129 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1130
1131 /*
1132 * Stop TRIM thread in case spa_unload() wasn't called directly
1133 * before spa_deactivate().
1134 */
1135 trim_thread_destroy(spa);
1136
1137 txg_list_destroy(&spa->spa_vdev_txg_list);
1138
1139 list_destroy(&spa->spa_config_dirty_list);
1140 list_destroy(&spa->spa_state_dirty_list);
1141
1142 for (int t = 0; t < ZIO_TYPES; t++) {
1143 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1144 spa_taskqs_fini(spa, t, q);
1145 }
1146 }
1147
1148 metaslab_class_destroy(spa->spa_normal_class);
1149 spa->spa_normal_class = NULL;
1150
1151 metaslab_class_destroy(spa->spa_log_class);
1152 spa->spa_log_class = NULL;
1153
1154 /*
1155 * If this was part of an import or the open otherwise failed, we may
1156 * still have errors left in the queues. Empty them just in case.
1157 */
1158 spa_errlog_drain(spa);
1159
1160 avl_destroy(&spa->spa_errlist_scrub);
1161 avl_destroy(&spa->spa_errlist_last);
1162
1163 spa->spa_state = POOL_STATE_UNINITIALIZED;
1164
1165 mutex_enter(&spa->spa_proc_lock);
1166 if (spa->spa_proc_state != SPA_PROC_NONE) {
1167 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1168 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1169 cv_broadcast(&spa->spa_proc_cv);
1170 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1171 ASSERT(spa->spa_proc != &p0);
1172 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1173 }
1174 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1175 spa->spa_proc_state = SPA_PROC_NONE;
1176 }
1177 ASSERT(spa->spa_proc == &p0);
1178 mutex_exit(&spa->spa_proc_lock);
1179
1180#ifdef SPA_PROCESS
1181 /*
1182 * We want to make sure spa_thread() has actually exited the ZFS
1183 * module, so that the module can't be unloaded out from underneath
1184 * it.
1185 */
1186 if (spa->spa_did != 0) {
1187 thread_join(spa->spa_did);
1188 spa->spa_did = 0;
1189 }
1190#endif /* SPA_PROCESS */
1191}
1192
1193/*
1194 * Verify a pool configuration, and construct the vdev tree appropriately. This
1195 * will create all the necessary vdevs in the appropriate layout, with each vdev
1196 * in the CLOSED state. This will prep the pool before open/creation/import.
1197 * All vdev validation is done by the vdev_alloc() routine.
1198 */
1199static int
1200spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1201 uint_t id, int atype)
1202{
1203 nvlist_t **child;
1204 uint_t children;
1205 int error;
1206
1207 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1208 return (error);
1209
1210 if ((*vdp)->vdev_ops->vdev_op_leaf)
1211 return (0);
1212
1213 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1214 &child, &children);
1215
1216 if (error == ENOENT)
1217 return (0);
1218
1219 if (error) {
1220 vdev_free(*vdp);
1221 *vdp = NULL;
1222 return (SET_ERROR(EINVAL));
1223 }
1224
1225 for (int c = 0; c < children; c++) {
1226 vdev_t *vd;
1227 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1228 atype)) != 0) {
1229 vdev_free(*vdp);
1230 *vdp = NULL;
1231 return (error);
1232 }
1233 }
1234
1235 ASSERT(*vdp != NULL);
1236
1237 return (0);
1238}
1239
1240/*
1241 * Opposite of spa_load().
1242 */
1243static void
1244spa_unload(spa_t *spa)
1245{
1246 int i;
1247
1248 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1249
1250 /*
1251 * Stop TRIM thread.
1252 */
1253 trim_thread_destroy(spa);
1254
1255 /*
1256 * Stop async tasks.
1257 */
1258 spa_async_suspend(spa);
1259
1260 /*
1261 * Stop syncing.
1262 */
1263 if (spa->spa_sync_on) {
1264 txg_sync_stop(spa->spa_dsl_pool);
1265 spa->spa_sync_on = B_FALSE;
1266 }
1267
1268 /*
1269 * Wait for any outstanding async I/O to complete.
1270 */
1271 if (spa->spa_async_zio_root != NULL) {
1272 (void) zio_wait(spa->spa_async_zio_root);
1273 spa->spa_async_zio_root = NULL;
1274 }
1275
1276 bpobj_close(&spa->spa_deferred_bpobj);
1277
1278 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1279
1280 /*
1281 * Close all vdevs.
1282 */
1283 if (spa->spa_root_vdev)
1284 vdev_free(spa->spa_root_vdev);
1285 ASSERT(spa->spa_root_vdev == NULL);
1286
1287 /*
1288 * Close the dsl pool.
1289 */
1290 if (spa->spa_dsl_pool) {
1291 dsl_pool_close(spa->spa_dsl_pool);
1292 spa->spa_dsl_pool = NULL;
1293 spa->spa_meta_objset = NULL;
1294 }
1295
1296 ddt_unload(spa);
1297
1298
1299 /*
1300 * Drop and purge level 2 cache
1301 */
1302 spa_l2cache_drop(spa);
1303
1304 for (i = 0; i < spa->spa_spares.sav_count; i++)
1305 vdev_free(spa->spa_spares.sav_vdevs[i]);
1306 if (spa->spa_spares.sav_vdevs) {
1307 kmem_free(spa->spa_spares.sav_vdevs,
1308 spa->spa_spares.sav_count * sizeof (void *));
1309 spa->spa_spares.sav_vdevs = NULL;
1310 }
1311 if (spa->spa_spares.sav_config) {
1312 nvlist_free(spa->spa_spares.sav_config);
1313 spa->spa_spares.sav_config = NULL;
1314 }
1315 spa->spa_spares.sav_count = 0;
1316
1317 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1318 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1319 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1320 }
1321 if (spa->spa_l2cache.sav_vdevs) {
1322 kmem_free(spa->spa_l2cache.sav_vdevs,
1323 spa->spa_l2cache.sav_count * sizeof (void *));
1324 spa->spa_l2cache.sav_vdevs = NULL;
1325 }
1326 if (spa->spa_l2cache.sav_config) {
1327 nvlist_free(spa->spa_l2cache.sav_config);
1328 spa->spa_l2cache.sav_config = NULL;
1329 }
1330 spa->spa_l2cache.sav_count = 0;
1331
1332 spa->spa_async_suspended = 0;
1333
1334 if (spa->spa_comment != NULL) {
1335 spa_strfree(spa->spa_comment);
1336 spa->spa_comment = NULL;
1337 }
1338
1339 spa_config_exit(spa, SCL_ALL, FTAG);
1340}
1341
1342/*
1343 * Load (or re-load) the current list of vdevs describing the active spares for
1344 * this pool. When this is called, we have some form of basic information in
1345 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1346 * then re-generate a more complete list including status information.
1347 */
1348static void
1349spa_load_spares(spa_t *spa)
1350{
1351 nvlist_t **spares;
1352 uint_t nspares;
1353 int i;
1354 vdev_t *vd, *tvd;
1355
1356 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1357
1358 /*
1359 * First, close and free any existing spare vdevs.
1360 */
1361 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1362 vd = spa->spa_spares.sav_vdevs[i];
1363
1364 /* Undo the call to spa_activate() below */
1365 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1366 B_FALSE)) != NULL && tvd->vdev_isspare)
1367 spa_spare_remove(tvd);
1368 vdev_close(vd);
1369 vdev_free(vd);
1370 }
1371
1372 if (spa->spa_spares.sav_vdevs)
1373 kmem_free(spa->spa_spares.sav_vdevs,
1374 spa->spa_spares.sav_count * sizeof (void *));
1375
1376 if (spa->spa_spares.sav_config == NULL)
1377 nspares = 0;
1378 else
1379 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1380 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1381
1382 spa->spa_spares.sav_count = (int)nspares;
1383 spa->spa_spares.sav_vdevs = NULL;
1384
1385 if (nspares == 0)
1386 return;
1387
1388 /*
1389 * Construct the array of vdevs, opening them to get status in the
1390 * process. For each spare, there is potentially two different vdev_t
1391 * structures associated with it: one in the list of spares (used only
1392 * for basic validation purposes) and one in the active vdev
1393 * configuration (if it's spared in). During this phase we open and
1394 * validate each vdev on the spare list. If the vdev also exists in the
1395 * active configuration, then we also mark this vdev as an active spare.
1396 */
1397 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1398 KM_SLEEP);
1399 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1400 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1401 VDEV_ALLOC_SPARE) == 0);
1402 ASSERT(vd != NULL);
1403
1404 spa->spa_spares.sav_vdevs[i] = vd;
1405
1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1407 B_FALSE)) != NULL) {
1408 if (!tvd->vdev_isspare)
1409 spa_spare_add(tvd);
1410
1411 /*
1412 * We only mark the spare active if we were successfully
1413 * able to load the vdev. Otherwise, importing a pool
1414 * with a bad active spare would result in strange
1415 * behavior, because multiple pool would think the spare
1416 * is actively in use.
1417 *
1418 * There is a vulnerability here to an equally bizarre
1419 * circumstance, where a dead active spare is later
1420 * brought back to life (onlined or otherwise). Given
1421 * the rarity of this scenario, and the extra complexity
1422 * it adds, we ignore the possibility.
1423 */
1424 if (!vdev_is_dead(tvd))
1425 spa_spare_activate(tvd);
1426 }
1427
1428 vd->vdev_top = vd;
1429 vd->vdev_aux = &spa->spa_spares;
1430
1431 if (vdev_open(vd) != 0)
1432 continue;
1433
1434 if (vdev_validate_aux(vd) == 0)
1435 spa_spare_add(vd);
1436 }
1437
1438 /*
1439 * Recompute the stashed list of spares, with status information
1440 * this time.
1441 */
1442 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1443 DATA_TYPE_NVLIST_ARRAY) == 0);
1444
1445 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1446 KM_SLEEP);
1447 for (i = 0; i < spa->spa_spares.sav_count; i++)
1448 spares[i] = vdev_config_generate(spa,
1449 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1451 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1452 for (i = 0; i < spa->spa_spares.sav_count; i++)
1453 nvlist_free(spares[i]);
1454 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1455}
1456
1457/*
1458 * Load (or re-load) the current list of vdevs describing the active l2cache for
1459 * this pool. When this is called, we have some form of basic information in
1460 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1461 * then re-generate a more complete list including status information.
1462 * Devices which are already active have their details maintained, and are
1463 * not re-opened.
1464 */
1465static void
1466spa_load_l2cache(spa_t *spa)
1467{
1468 nvlist_t **l2cache;
1469 uint_t nl2cache;
1470 int i, j, oldnvdevs;
1471 uint64_t guid;
1472 vdev_t *vd, **oldvdevs, **newvdevs;
1473 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1474
1475 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1476
1477 if (sav->sav_config != NULL) {
1478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1479 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1480 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1481 } else {
1482 nl2cache = 0;
1483 newvdevs = NULL;
1484 }
1485
1486 oldvdevs = sav->sav_vdevs;
1487 oldnvdevs = sav->sav_count;
1488 sav->sav_vdevs = NULL;
1489 sav->sav_count = 0;
1490
1491 /*
1492 * Process new nvlist of vdevs.
1493 */
1494 for (i = 0; i < nl2cache; i++) {
1495 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1496 &guid) == 0);
1497
1498 newvdevs[i] = NULL;
1499 for (j = 0; j < oldnvdevs; j++) {
1500 vd = oldvdevs[j];
1501 if (vd != NULL && guid == vd->vdev_guid) {
1502 /*
1503 * Retain previous vdev for add/remove ops.
1504 */
1505 newvdevs[i] = vd;
1506 oldvdevs[j] = NULL;
1507 break;
1508 }
1509 }
1510
1511 if (newvdevs[i] == NULL) {
1512 /*
1513 * Create new vdev
1514 */
1515 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1516 VDEV_ALLOC_L2CACHE) == 0);
1517 ASSERT(vd != NULL);
1518 newvdevs[i] = vd;
1519
1520 /*
1521 * Commit this vdev as an l2cache device,
1522 * even if it fails to open.
1523 */
1524 spa_l2cache_add(vd);
1525
1526 vd->vdev_top = vd;
1527 vd->vdev_aux = sav;
1528
1529 spa_l2cache_activate(vd);
1530
1531 if (vdev_open(vd) != 0)
1532 continue;
1533
1534 (void) vdev_validate_aux(vd);
1535
1536 if (!vdev_is_dead(vd))
1537 l2arc_add_vdev(spa, vd);
1538 }
1539 }
1540
1541 /*
1542 * Purge vdevs that were dropped
1543 */
1544 for (i = 0; i < oldnvdevs; i++) {
1545 uint64_t pool;
1546
1547 vd = oldvdevs[i];
1548 if (vd != NULL) {
1549 ASSERT(vd->vdev_isl2cache);
1550
1551 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1552 pool != 0ULL && l2arc_vdev_present(vd))
1553 l2arc_remove_vdev(vd);
1554 vdev_clear_stats(vd);
1555 vdev_free(vd);
1556 }
1557 }
1558
1559 if (oldvdevs)
1560 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1561
1562 if (sav->sav_config == NULL)
1563 goto out;
1564
1565 sav->sav_vdevs = newvdevs;
1566 sav->sav_count = (int)nl2cache;
1567
1568 /*
1569 * Recompute the stashed list of l2cache devices, with status
1570 * information this time.
1571 */
1572 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1573 DATA_TYPE_NVLIST_ARRAY) == 0);
1574
1575 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1576 for (i = 0; i < sav->sav_count; i++)
1577 l2cache[i] = vdev_config_generate(spa,
1578 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1579 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1580 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1581out:
1582 for (i = 0; i < sav->sav_count; i++)
1583 nvlist_free(l2cache[i]);
1584 if (sav->sav_count)
1585 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1586}
1587
1588static int
1589load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1590{
1591 dmu_buf_t *db;
1592 char *packed = NULL;
1593 size_t nvsize = 0;
1594 int error;
1595 *value = NULL;
1596
1597 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1598 if (error != 0)
1599 return (error);
1600 nvsize = *(uint64_t *)db->db_data;
1601 dmu_buf_rele(db, FTAG);
1602
1603 packed = kmem_alloc(nvsize, KM_SLEEP);
1604 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1605 DMU_READ_PREFETCH);
1606 if (error == 0)
1607 error = nvlist_unpack(packed, nvsize, value, 0);
1608 kmem_free(packed, nvsize);
1609
1610 return (error);
1611}
1612
1613/*
1614 * Checks to see if the given vdev could not be opened, in which case we post a
1615 * sysevent to notify the autoreplace code that the device has been removed.
1616 */
1617static void
1618spa_check_removed(vdev_t *vd)
1619{
1620 for (int c = 0; c < vd->vdev_children; c++)
1621 spa_check_removed(vd->vdev_child[c]);
1622
1623 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1624 !vd->vdev_ishole) {
1625 zfs_post_autoreplace(vd->vdev_spa, vd);
1626 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1627 }
1628}
1629
1630/*
1631 * Validate the current config against the MOS config
1632 */
1633static boolean_t
1634spa_config_valid(spa_t *spa, nvlist_t *config)
1635{
1636 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1637 nvlist_t *nv;
1638
1639 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1640
1641 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1642 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1643
1644 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1645
1646 /*
1647 * If we're doing a normal import, then build up any additional
1648 * diagnostic information about missing devices in this config.
1649 * We'll pass this up to the user for further processing.
1650 */
1651 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1652 nvlist_t **child, *nv;
1653 uint64_t idx = 0;
1654
1655 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1656 KM_SLEEP);
1657 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1658
1659 for (int c = 0; c < rvd->vdev_children; c++) {
1660 vdev_t *tvd = rvd->vdev_child[c];
1661 vdev_t *mtvd = mrvd->vdev_child[c];
1662
1663 if (tvd->vdev_ops == &vdev_missing_ops &&
1664 mtvd->vdev_ops != &vdev_missing_ops &&
1665 mtvd->vdev_islog)
1666 child[idx++] = vdev_config_generate(spa, mtvd,
1667 B_FALSE, 0);
1668 }
1669
1670 if (idx) {
1671 VERIFY(nvlist_add_nvlist_array(nv,
1672 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1673 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1674 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1675
1676 for (int i = 0; i < idx; i++)
1677 nvlist_free(child[i]);
1678 }
1679 nvlist_free(nv);
1680 kmem_free(child, rvd->vdev_children * sizeof (char **));
1681 }
1682
1683 /*
1684 * Compare the root vdev tree with the information we have
1685 * from the MOS config (mrvd). Check each top-level vdev
1686 * with the corresponding MOS config top-level (mtvd).
1687 */
1688 for (int c = 0; c < rvd->vdev_children; c++) {
1689 vdev_t *tvd = rvd->vdev_child[c];
1690 vdev_t *mtvd = mrvd->vdev_child[c];
1691
1692 /*
1693 * Resolve any "missing" vdevs in the current configuration.
1694 * If we find that the MOS config has more accurate information
1695 * about the top-level vdev then use that vdev instead.
1696 */
1697 if (tvd->vdev_ops == &vdev_missing_ops &&
1698 mtvd->vdev_ops != &vdev_missing_ops) {
1699
1700 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1701 continue;
1702
1703 /*
1704 * Device specific actions.
1705 */
1706 if (mtvd->vdev_islog) {
1707 spa_set_log_state(spa, SPA_LOG_CLEAR);
1708 } else {
1709 /*
1710 * XXX - once we have 'readonly' pool
1711 * support we should be able to handle
1712 * missing data devices by transitioning
1713 * the pool to readonly.
1714 */
1715 continue;
1716 }
1717
1718 /*
1719 * Swap the missing vdev with the data we were
1720 * able to obtain from the MOS config.
1721 */
1722 vdev_remove_child(rvd, tvd);
1723 vdev_remove_child(mrvd, mtvd);
1724
1725 vdev_add_child(rvd, mtvd);
1726 vdev_add_child(mrvd, tvd);
1727
1728 spa_config_exit(spa, SCL_ALL, FTAG);
1729 vdev_load(mtvd);
1730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731
1732 vdev_reopen(rvd);
1733 } else if (mtvd->vdev_islog) {
1734 /*
1735 * Load the slog device's state from the MOS config
1736 * since it's possible that the label does not
1737 * contain the most up-to-date information.
1738 */
1739 vdev_load_log_state(tvd, mtvd);
1740 vdev_reopen(tvd);
1741 }
1742 }
1743 vdev_free(mrvd);
1744 spa_config_exit(spa, SCL_ALL, FTAG);
1745
1746 /*
1747 * Ensure we were able to validate the config.
1748 */
1749 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1750}
1751
1752/*
1753 * Check for missing log devices
1754 */
1755static boolean_t
1756spa_check_logs(spa_t *spa)
1757{
1758 boolean_t rv = B_FALSE;
1759
1760 switch (spa->spa_log_state) {
1761 case SPA_LOG_MISSING:
1762 /* need to recheck in case slog has been restored */
1763 case SPA_LOG_UNKNOWN:
1764 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1765 NULL, DS_FIND_CHILDREN) != 0);
1766 if (rv)
1767 spa_set_log_state(spa, SPA_LOG_MISSING);
1768 break;
1769 }
1770 return (rv);
1771}
1772
1773static boolean_t
1774spa_passivate_log(spa_t *spa)
1775{
1776 vdev_t *rvd = spa->spa_root_vdev;
1777 boolean_t slog_found = B_FALSE;
1778
1779 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1780
1781 if (!spa_has_slogs(spa))
1782 return (B_FALSE);
1783
1784 for (int c = 0; c < rvd->vdev_children; c++) {
1785 vdev_t *tvd = rvd->vdev_child[c];
1786 metaslab_group_t *mg = tvd->vdev_mg;
1787
1788 if (tvd->vdev_islog) {
1789 metaslab_group_passivate(mg);
1790 slog_found = B_TRUE;
1791 }
1792 }
1793
1794 return (slog_found);
1795}
1796
1797static void
1798spa_activate_log(spa_t *spa)
1799{
1800 vdev_t *rvd = spa->spa_root_vdev;
1801
1802 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1803
1804 for (int c = 0; c < rvd->vdev_children; c++) {
1805 vdev_t *tvd = rvd->vdev_child[c];
1806 metaslab_group_t *mg = tvd->vdev_mg;
1807
1808 if (tvd->vdev_islog)
1809 metaslab_group_activate(mg);
1810 }
1811}
1812
1813int
1814spa_offline_log(spa_t *spa)
1815{
1816 int error;
1817
1818 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1819 NULL, DS_FIND_CHILDREN);
1820 if (error == 0) {
1821 /*
1822 * We successfully offlined the log device, sync out the
1823 * current txg so that the "stubby" block can be removed
1824 * by zil_sync().
1825 */
1826 txg_wait_synced(spa->spa_dsl_pool, 0);
1827 }
1828 return (error);
1829}
1830
1831static void
1832spa_aux_check_removed(spa_aux_vdev_t *sav)
1833{
1834 int i;
1835
1836 for (i = 0; i < sav->sav_count; i++)
1837 spa_check_removed(sav->sav_vdevs[i]);
1838}
1839
1840void
1841spa_claim_notify(zio_t *zio)
1842{
1843 spa_t *spa = zio->io_spa;
1844
1845 if (zio->io_error)
1846 return;
1847
1848 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1849 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1850 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1851 mutex_exit(&spa->spa_props_lock);
1852}
1853
1854typedef struct spa_load_error {
1855 uint64_t sle_meta_count;
1856 uint64_t sle_data_count;
1857} spa_load_error_t;
1858
1859static void
1860spa_load_verify_done(zio_t *zio)
1861{
1862 blkptr_t *bp = zio->io_bp;
1863 spa_load_error_t *sle = zio->io_private;
1864 dmu_object_type_t type = BP_GET_TYPE(bp);
1865 int error = zio->io_error;
1866
1867 if (error) {
1868 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1869 type != DMU_OT_INTENT_LOG)
1870 atomic_add_64(&sle->sle_meta_count, 1);
1871 else
1872 atomic_add_64(&sle->sle_data_count, 1);
1873 }
1874 zio_data_buf_free(zio->io_data, zio->io_size);
1875}
1876
1877/*ARGSUSED*/
1878static int
1879spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1880 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1881{
1882 if (!BP_IS_HOLE(bp)) {
1883 zio_t *rio = arg;
1884 size_t size = BP_GET_PSIZE(bp);
1885 void *data = zio_data_buf_alloc(size);
1886
1887 zio_nowait(zio_read(rio, spa, bp, data, size,
1888 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1889 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1890 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1891 }
1892 return (0);
1893}
1894
1895static int
1896spa_load_verify(spa_t *spa)
1897{
1898 zio_t *rio;
1899 spa_load_error_t sle = { 0 };
1900 zpool_rewind_policy_t policy;
1901 boolean_t verify_ok = B_FALSE;
1902 int error;
1903
1904 zpool_get_rewind_policy(spa->spa_config, &policy);
1905
1906 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1907 return (0);
1908
1909 rio = zio_root(spa, NULL, &sle,
1910 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1911
1912 error = traverse_pool(spa, spa->spa_verify_min_txg,
1913 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1914
1915 (void) zio_wait(rio);
1916
1917 spa->spa_load_meta_errors = sle.sle_meta_count;
1918 spa->spa_load_data_errors = sle.sle_data_count;
1919
1920 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1921 sle.sle_data_count <= policy.zrp_maxdata) {
1922 int64_t loss = 0;
1923
1924 verify_ok = B_TRUE;
1925 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1926 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1927
1928 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1929 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1930 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1931 VERIFY(nvlist_add_int64(spa->spa_load_info,
1932 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1933 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1934 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1935 } else {
1936 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1937 }
1938
1939 if (error) {
1940 if (error != ENXIO && error != EIO)
1941 error = SET_ERROR(EIO);
1942 return (error);
1943 }
1944
1945 return (verify_ok ? 0 : EIO);
1946}
1947
1948/*
1949 * Find a value in the pool props object.
1950 */
1951static void
1952spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1953{
1954 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1955 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1956}
1957
1958/*
1959 * Find a value in the pool directory object.
1960 */
1961static int
1962spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1963{
1964 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1965 name, sizeof (uint64_t), 1, val));
1966}
1967
1968static int
1969spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1970{
1971 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1972 return (err);
1973}
1974
1975/*
1976 * Fix up config after a partly-completed split. This is done with the
1977 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1978 * pool have that entry in their config, but only the splitting one contains
1979 * a list of all the guids of the vdevs that are being split off.
1980 *
1981 * This function determines what to do with that list: either rejoin
1982 * all the disks to the pool, or complete the splitting process. To attempt
1983 * the rejoin, each disk that is offlined is marked online again, and
1984 * we do a reopen() call. If the vdev label for every disk that was
1985 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1986 * then we call vdev_split() on each disk, and complete the split.
1987 *
1988 * Otherwise we leave the config alone, with all the vdevs in place in
1989 * the original pool.
1990 */
1991static void
1992spa_try_repair(spa_t *spa, nvlist_t *config)
1993{
1994 uint_t extracted;
1995 uint64_t *glist;
1996 uint_t i, gcount;
1997 nvlist_t *nvl;
1998 vdev_t **vd;
1999 boolean_t attempt_reopen;
2000
2001 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2002 return;
2003
2004 /* check that the config is complete */
2005 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2006 &glist, &gcount) != 0)
2007 return;
2008
2009 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2010
2011 /* attempt to online all the vdevs & validate */
2012 attempt_reopen = B_TRUE;
2013 for (i = 0; i < gcount; i++) {
2014 if (glist[i] == 0) /* vdev is hole */
2015 continue;
2016
2017 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2018 if (vd[i] == NULL) {
2019 /*
2020 * Don't bother attempting to reopen the disks;
2021 * just do the split.
2022 */
2023 attempt_reopen = B_FALSE;
2024 } else {
2025 /* attempt to re-online it */
2026 vd[i]->vdev_offline = B_FALSE;
2027 }
2028 }
2029
2030 if (attempt_reopen) {
2031 vdev_reopen(spa->spa_root_vdev);
2032
2033 /* check each device to see what state it's in */
2034 for (extracted = 0, i = 0; i < gcount; i++) {
2035 if (vd[i] != NULL &&
2036 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2037 break;
2038 ++extracted;
2039 }
2040 }
2041
2042 /*
2043 * If every disk has been moved to the new pool, or if we never
2044 * even attempted to look at them, then we split them off for
2045 * good.
2046 */
2047 if (!attempt_reopen || gcount == extracted) {
2048 for (i = 0; i < gcount; i++)
2049 if (vd[i] != NULL)
2050 vdev_split(vd[i]);
2051 vdev_reopen(spa->spa_root_vdev);
2052 }
2053
2054 kmem_free(vd, gcount * sizeof (vdev_t *));
2055}
2056
2057static int
2058spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2059 boolean_t mosconfig)
2060{
2061 nvlist_t *config = spa->spa_config;
2062 char *ereport = FM_EREPORT_ZFS_POOL;
2063 char *comment;
2064 int error;
2065 uint64_t pool_guid;
2066 nvlist_t *nvl;
2067
2068 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2069 return (SET_ERROR(EINVAL));
2070
2071 ASSERT(spa->spa_comment == NULL);
2072 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2073 spa->spa_comment = spa_strdup(comment);
2074
2075 /*
2076 * Versioning wasn't explicitly added to the label until later, so if
2077 * it's not present treat it as the initial version.
2078 */
2079 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2080 &spa->spa_ubsync.ub_version) != 0)
2081 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2082
2083 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2084 &spa->spa_config_txg);
2085
2086 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2087 spa_guid_exists(pool_guid, 0)) {
2088 error = SET_ERROR(EEXIST);
2089 } else {
2090 spa->spa_config_guid = pool_guid;
2091
2092 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2093 &nvl) == 0) {
2094 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2095 KM_SLEEP) == 0);
2096 }
2097
2098 nvlist_free(spa->spa_load_info);
2099 spa->spa_load_info = fnvlist_alloc();
2100
2101 gethrestime(&spa->spa_loaded_ts);
2102 error = spa_load_impl(spa, pool_guid, config, state, type,
2103 mosconfig, &ereport);
2104 }
2105
2106 spa->spa_minref = refcount_count(&spa->spa_refcount);
2107 if (error) {
2108 if (error != EEXIST) {
2109 spa->spa_loaded_ts.tv_sec = 0;
2110 spa->spa_loaded_ts.tv_nsec = 0;
2111 }
2112 if (error != EBADF) {
2113 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2114 }
2115 }
2116 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2117 spa->spa_ena = 0;
2118
2119 return (error);
2120}
2121
2122/*
2123 * Load an existing storage pool, using the pool's builtin spa_config as a
2124 * source of configuration information.
2125 */
2126static int
2127spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2129 char **ereport)
2130{
2131 int error = 0;
2132 nvlist_t *nvroot = NULL;
2133 nvlist_t *label;
2134 vdev_t *rvd;
2135 uberblock_t *ub = &spa->spa_uberblock;
2136 uint64_t children, config_cache_txg = spa->spa_config_txg;
2137 int orig_mode = spa->spa_mode;
2138 int parse;
2139 uint64_t obj;
2140 boolean_t missing_feat_write = B_FALSE;
2141
2142 /*
2143 * If this is an untrusted config, access the pool in read-only mode.
2144 * This prevents things like resilvering recently removed devices.
2145 */
2146 if (!mosconfig)
2147 spa->spa_mode = FREAD;
2148
2149 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2150
2151 spa->spa_load_state = state;
2152
2153 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2154 return (SET_ERROR(EINVAL));
2155
2156 parse = (type == SPA_IMPORT_EXISTING ?
2157 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2158
2159 /*
2160 * Create "The Godfather" zio to hold all async IOs
2161 */
2162 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2163 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2164
2165 /*
2166 * Parse the configuration into a vdev tree. We explicitly set the
2167 * value that will be returned by spa_version() since parsing the
2168 * configuration requires knowing the version number.
2169 */
2170 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2171 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2172 spa_config_exit(spa, SCL_ALL, FTAG);
2173
2174 if (error != 0)
2175 return (error);
2176
2177 ASSERT(spa->spa_root_vdev == rvd);
2178
2179 if (type != SPA_IMPORT_ASSEMBLE) {
2180 ASSERT(spa_guid(spa) == pool_guid);
2181 }
2182
2183 /*
2184 * Try to open all vdevs, loading each label in the process.
2185 */
2186 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2187 error = vdev_open(rvd);
2188 spa_config_exit(spa, SCL_ALL, FTAG);
2189 if (error != 0)
2190 return (error);
2191
2192 /*
2193 * We need to validate the vdev labels against the configuration that
2194 * we have in hand, which is dependent on the setting of mosconfig. If
2195 * mosconfig is true then we're validating the vdev labels based on
2196 * that config. Otherwise, we're validating against the cached config
2197 * (zpool.cache) that was read when we loaded the zfs module, and then
2198 * later we will recursively call spa_load() and validate against
2199 * the vdev config.
2200 *
2201 * If we're assembling a new pool that's been split off from an
2202 * existing pool, the labels haven't yet been updated so we skip
2203 * validation for now.
2204 */
2205 if (type != SPA_IMPORT_ASSEMBLE) {
2206 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2207 error = vdev_validate(rvd, mosconfig);
2208 spa_config_exit(spa, SCL_ALL, FTAG);
2209
2210 if (error != 0)
2211 return (error);
2212
2213 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2214 return (SET_ERROR(ENXIO));
2215 }
2216
2217 /*
2218 * Find the best uberblock.
2219 */
2220 vdev_uberblock_load(rvd, ub, &label);
2221
2222 /*
2223 * If we weren't able to find a single valid uberblock, return failure.
2224 */
2225 if (ub->ub_txg == 0) {
2226 nvlist_free(label);
2227 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2228 }
2229
2230 /*
2231 * If the pool has an unsupported version we can't open it.
2232 */
2233 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2234 nvlist_free(label);
2235 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2236 }
2237
2238 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2239 nvlist_t *features;
2240
2241 /*
2242 * If we weren't able to find what's necessary for reading the
2243 * MOS in the label, return failure.
2244 */
2245 if (label == NULL || nvlist_lookup_nvlist(label,
2246 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2247 nvlist_free(label);
2248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2249 ENXIO));
2250 }
2251
2252 /*
2253 * Update our in-core representation with the definitive values
2254 * from the label.
2255 */
2256 nvlist_free(spa->spa_label_features);
2257 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2258 }
2259
2260 nvlist_free(label);
2261
2262 /*
2263 * Look through entries in the label nvlist's features_for_read. If
2264 * there is a feature listed there which we don't understand then we
2265 * cannot open a pool.
2266 */
2267 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2268 nvlist_t *unsup_feat;
2269
2270 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2271 0);
2272
2273 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2274 NULL); nvp != NULL;
2275 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2276 if (!zfeature_is_supported(nvpair_name(nvp))) {
2277 VERIFY(nvlist_add_string(unsup_feat,
2278 nvpair_name(nvp), "") == 0);
2279 }
2280 }
2281
2282 if (!nvlist_empty(unsup_feat)) {
2283 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2284 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2285 nvlist_free(unsup_feat);
2286 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2287 ENOTSUP));
2288 }
2289
2290 nvlist_free(unsup_feat);
2291 }
2292
2293 /*
2294 * If the vdev guid sum doesn't match the uberblock, we have an
2295 * incomplete configuration. We first check to see if the pool
2296 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2297 * If it is, defer the vdev_guid_sum check till later so we
2298 * can handle missing vdevs.
2299 */
2300 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2301 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2302 rvd->vdev_guid_sum != ub->ub_guid_sum)
2303 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2304
2305 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2307 spa_try_repair(spa, config);
2308 spa_config_exit(spa, SCL_ALL, FTAG);
2309 nvlist_free(spa->spa_config_splitting);
2310 spa->spa_config_splitting = NULL;
2311 }
2312
2313 /*
2314 * Initialize internal SPA structures.
2315 */
2316 spa->spa_state = POOL_STATE_ACTIVE;
2317 spa->spa_ubsync = spa->spa_uberblock;
2318 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2319 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2320 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2321 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2322 spa->spa_claim_max_txg = spa->spa_first_txg;
2323 spa->spa_prev_software_version = ub->ub_software_version;
2324
2325 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2326 if (error)
2327 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2328 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2329
2330 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2331 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2332
2333 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2334 boolean_t missing_feat_read = B_FALSE;
2335 nvlist_t *unsup_feat, *enabled_feat;
2336
2337 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2338 &spa->spa_feat_for_read_obj) != 0) {
2339 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340 }
2341
2342 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2343 &spa->spa_feat_for_write_obj) != 0) {
2344 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2345 }
2346
2347 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2348 &spa->spa_feat_desc_obj) != 0) {
2349 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2350 }
2351
2352 enabled_feat = fnvlist_alloc();
2353 unsup_feat = fnvlist_alloc();
2354
2355 if (!spa_features_check(spa, B_FALSE,
2356 unsup_feat, enabled_feat))
2357 missing_feat_read = B_TRUE;
2358
2359 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2360 if (!spa_features_check(spa, B_TRUE,
2361 unsup_feat, enabled_feat)) {
2362 missing_feat_write = B_TRUE;
2363 }
2364 }
2365
2366 fnvlist_add_nvlist(spa->spa_load_info,
2367 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2368
2369 if (!nvlist_empty(unsup_feat)) {
2370 fnvlist_add_nvlist(spa->spa_load_info,
2371 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2372 }
2373
2374 fnvlist_free(enabled_feat);
2375 fnvlist_free(unsup_feat);
2376
2377 if (!missing_feat_read) {
2378 fnvlist_add_boolean(spa->spa_load_info,
2379 ZPOOL_CONFIG_CAN_RDONLY);
2380 }
2381
2382 /*
2383 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2384 * twofold: to determine whether the pool is available for
2385 * import in read-write mode and (if it is not) whether the
2386 * pool is available for import in read-only mode. If the pool
2387 * is available for import in read-write mode, it is displayed
2388 * as available in userland; if it is not available for import
2389 * in read-only mode, it is displayed as unavailable in
2390 * userland. If the pool is available for import in read-only
2391 * mode but not read-write mode, it is displayed as unavailable
2392 * in userland with a special note that the pool is actually
2393 * available for open in read-only mode.
2394 *
2395 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2396 * missing a feature for write, we must first determine whether
2397 * the pool can be opened read-only before returning to
2398 * userland in order to know whether to display the
2399 * abovementioned note.
2400 */
2401 if (missing_feat_read || (missing_feat_write &&
2402 spa_writeable(spa))) {
2403 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2404 ENOTSUP));
2405 }
2406
2407 /*
2408 * Load refcounts for ZFS features from disk into an in-memory
2409 * cache during SPA initialization.
2410 */
2411 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2412 uint64_t refcount;
2413
2414 error = feature_get_refcount_from_disk(spa,
2415 &spa_feature_table[i], &refcount);
2416 if (error == 0) {
2417 spa->spa_feat_refcount_cache[i] = refcount;
2418 } else if (error == ENOTSUP) {
2419 spa->spa_feat_refcount_cache[i] =
2420 SPA_FEATURE_DISABLED;
2421 } else {
2422 return (spa_vdev_err(rvd,
2423 VDEV_AUX_CORRUPT_DATA, EIO));
2424 }
2425 }
2426 }
2427
2428 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2429 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2430 &spa->spa_feat_enabled_txg_obj) != 0) {
2431 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432 }
2433 }
2434
2435 spa->spa_is_initializing = B_TRUE;
2436 error = dsl_pool_open(spa->spa_dsl_pool);
2437 spa->spa_is_initializing = B_FALSE;
2438 if (error != 0)
2439 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440
2441 if (!mosconfig) {
2442 uint64_t hostid;
2443 nvlist_t *policy = NULL, *nvconfig;
2444
2445 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447
2448 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2449 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2450 char *hostname;
2451 unsigned long myhostid = 0;
2452
2453 VERIFY(nvlist_lookup_string(nvconfig,
2454 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2455
2456#ifdef _KERNEL
2457 myhostid = zone_get_hostid(NULL);
2458#else /* _KERNEL */
2459 /*
2460 * We're emulating the system's hostid in userland, so
2461 * we can't use zone_get_hostid().
2462 */
2463 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2464#endif /* _KERNEL */
2465 if (check_hostid && hostid != 0 && myhostid != 0 &&
2466 hostid != myhostid) {
2467 nvlist_free(nvconfig);
2468 cmn_err(CE_WARN, "pool '%s' could not be "
2469 "loaded as it was last accessed by "
2470 "another system (host: %s hostid: 0x%lx). "
2471 "See: http://illumos.org/msg/ZFS-8000-EY",
2472 spa_name(spa), hostname,
2473 (unsigned long)hostid);
2474 return (SET_ERROR(EBADF));
2475 }
2476 }
2477 if (nvlist_lookup_nvlist(spa->spa_config,
2478 ZPOOL_REWIND_POLICY, &policy) == 0)
2479 VERIFY(nvlist_add_nvlist(nvconfig,
2480 ZPOOL_REWIND_POLICY, policy) == 0);
2481
2482 spa_config_set(spa, nvconfig);
2483 spa_unload(spa);
2484 spa_deactivate(spa);
2485 spa_activate(spa, orig_mode);
2486
2487 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2488 }
2489
2490 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2491 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2492 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2493 if (error != 0)
2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2495
2496 /*
2497 * Load the bit that tells us to use the new accounting function
2498 * (raid-z deflation). If we have an older pool, this will not
2499 * be present.
2500 */
2501 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2502 if (error != 0 && error != ENOENT)
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504
2505 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2506 &spa->spa_creation_version);
2507 if (error != 0 && error != ENOENT)
2508 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2509
2510 /*
2511 * Load the persistent error log. If we have an older pool, this will
2512 * not be present.
2513 */
2514 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2515 if (error != 0 && error != ENOENT)
2516 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2517
2518 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2519 &spa->spa_errlog_scrub);
2520 if (error != 0 && error != ENOENT)
2521 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2522
2523 /*
2524 * Load the history object. If we have an older pool, this
2525 * will not be present.
2526 */
2527 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2528 if (error != 0 && error != ENOENT)
2529 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2530
2531 /*
2532 * If we're assembling the pool from the split-off vdevs of
2533 * an existing pool, we don't want to attach the spares & cache
2534 * devices.
2535 */
2536
2537 /*
2538 * Load any hot spares for this pool.
2539 */
2540 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2541 if (error != 0 && error != ENOENT)
2542 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2543 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2544 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2545 if (load_nvlist(spa, spa->spa_spares.sav_object,
2546 &spa->spa_spares.sav_config) != 0)
2547 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2548
2549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2550 spa_load_spares(spa);
2551 spa_config_exit(spa, SCL_ALL, FTAG);
2552 } else if (error == 0) {
2553 spa->spa_spares.sav_sync = B_TRUE;
2554 }
2555
2556 /*
2557 * Load any level 2 ARC devices for this pool.
2558 */
2559 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2560 &spa->spa_l2cache.sav_object);
2561 if (error != 0 && error != ENOENT)
2562 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2563 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2564 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2565 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2566 &spa->spa_l2cache.sav_config) != 0)
2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568
2569 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2570 spa_load_l2cache(spa);
2571 spa_config_exit(spa, SCL_ALL, FTAG);
2572 } else if (error == 0) {
2573 spa->spa_l2cache.sav_sync = B_TRUE;
2574 }
2575
2576 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2577
2578 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2579 if (error && error != ENOENT)
2580 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2581
2582 if (error == 0) {
2583 uint64_t autoreplace;
2584
2585 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2586 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2587 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2588 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2589 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2590 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2591 &spa->spa_dedup_ditto);
2592
2593 spa->spa_autoreplace = (autoreplace != 0);
2594 }
2595
2596 /*
2597 * If the 'autoreplace' property is set, then post a resource notifying
2598 * the ZFS DE that it should not issue any faults for unopenable
2599 * devices. We also iterate over the vdevs, and post a sysevent for any
2600 * unopenable vdevs so that the normal autoreplace handler can take
2601 * over.
2602 */
2603 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2604 spa_check_removed(spa->spa_root_vdev);
2605 /*
2606 * For the import case, this is done in spa_import(), because
2607 * at this point we're using the spare definitions from
2608 * the MOS config, not necessarily from the userland config.
2609 */
2610 if (state != SPA_LOAD_IMPORT) {
2611 spa_aux_check_removed(&spa->spa_spares);
2612 spa_aux_check_removed(&spa->spa_l2cache);
2613 }
2614 }
2615
2616 /*
2617 * Load the vdev state for all toplevel vdevs.
2618 */
2619 vdev_load(rvd);
2620
2621 /*
2622 * Propagate the leaf DTLs we just loaded all the way up the tree.
2623 */
2624 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2625 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2626 spa_config_exit(spa, SCL_ALL, FTAG);
2627
2628 /*
2629 * Load the DDTs (dedup tables).
2630 */
2631 error = ddt_load(spa);
2632 if (error != 0)
2633 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2634
2635 spa_update_dspace(spa);
2636
2637 /*
2638 * Validate the config, using the MOS config to fill in any
2639 * information which might be missing. If we fail to validate
2640 * the config then declare the pool unfit for use. If we're
2641 * assembling a pool from a split, the log is not transferred
2642 * over.
2643 */
2644 if (type != SPA_IMPORT_ASSEMBLE) {
2645 nvlist_t *nvconfig;
2646
2647 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2649
2650 if (!spa_config_valid(spa, nvconfig)) {
2651 nvlist_free(nvconfig);
2652 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2653 ENXIO));
2654 }
2655 nvlist_free(nvconfig);
2656
2657 /*
2658 * Now that we've validated the config, check the state of the
2659 * root vdev. If it can't be opened, it indicates one or
2660 * more toplevel vdevs are faulted.
2661 */
2662 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2663 return (SET_ERROR(ENXIO));
2664
2665 if (spa_check_logs(spa)) {
2666 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2667 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2668 }
2669 }
2670
2671 if (missing_feat_write) {
2672 ASSERT(state == SPA_LOAD_TRYIMPORT);
2673
2674 /*
2675 * At this point, we know that we can open the pool in
2676 * read-only mode but not read-write mode. We now have enough
2677 * information and can return to userland.
2678 */
2679 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2680 }
2681
2682 /*
2683 * We've successfully opened the pool, verify that we're ready
2684 * to start pushing transactions.
2685 */
2686 if (state != SPA_LOAD_TRYIMPORT) {
2687 if (error = spa_load_verify(spa))
2688 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2689 error));
2690 }
2691
2692 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2693 spa->spa_load_max_txg == UINT64_MAX)) {
2694 dmu_tx_t *tx;
2695 int need_update = B_FALSE;
2696
2697 ASSERT(state != SPA_LOAD_TRYIMPORT);
2698
2699 /*
2700 * Claim log blocks that haven't been committed yet.
2701 * This must all happen in a single txg.
2702 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2703 * invoked from zil_claim_log_block()'s i/o done callback.
2704 * Price of rollback is that we abandon the log.
2705 */
2706 spa->spa_claiming = B_TRUE;
2707
2708 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2709 spa_first_txg(spa));
2710 (void) dmu_objset_find(spa_name(spa),
2711 zil_claim, tx, DS_FIND_CHILDREN);
2712 dmu_tx_commit(tx);
2713
2714 spa->spa_claiming = B_FALSE;
2715
2716 spa_set_log_state(spa, SPA_LOG_GOOD);
2717 spa->spa_sync_on = B_TRUE;
2718 txg_sync_start(spa->spa_dsl_pool);
2719
2720 /*
2721 * Wait for all claims to sync. We sync up to the highest
2722 * claimed log block birth time so that claimed log blocks
2723 * don't appear to be from the future. spa_claim_max_txg
2724 * will have been set for us by either zil_check_log_chain()
2725 * (invoked from spa_check_logs()) or zil_claim() above.
2726 */
2727 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2728
2729 /*
2730 * If the config cache is stale, or we have uninitialized
2731 * metaslabs (see spa_vdev_add()), then update the config.
2732 *
2733 * If this is a verbatim import, trust the current
2734 * in-core spa_config and update the disk labels.
2735 */
2736 if (config_cache_txg != spa->spa_config_txg ||
2737 state == SPA_LOAD_IMPORT ||
2738 state == SPA_LOAD_RECOVER ||
2739 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2740 need_update = B_TRUE;
2741
2742 for (int c = 0; c < rvd->vdev_children; c++)
2743 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2744 need_update = B_TRUE;
2745
2746 /*
2747 * Update the config cache asychronously in case we're the
2748 * root pool, in which case the config cache isn't writable yet.
2749 */
2750 if (need_update)
2751 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2752
2753 /*
2754 * Check all DTLs to see if anything needs resilvering.
2755 */
2756 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2757 vdev_resilver_needed(rvd, NULL, NULL))
2758 spa_async_request(spa, SPA_ASYNC_RESILVER);
2759
2760 /*
2761 * Log the fact that we booted up (so that we can detect if
2762 * we rebooted in the middle of an operation).
2763 */
2764 spa_history_log_version(spa, "open");
2765
2766 /*
2767 * Delete any inconsistent datasets.
2768 */
2769 (void) dmu_objset_find(spa_name(spa),
2770 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2771
2772 /*
2773 * Clean up any stale temporary dataset userrefs.
2774 */
2775 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2776 }
2777
2778 return (0);
2779}
2780
2781static int
2782spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2783{
2784 int mode = spa->spa_mode;
2785
2786 spa_unload(spa);
2787 spa_deactivate(spa);
2788
2789 spa->spa_load_max_txg--;
2790
2791 spa_activate(spa, mode);
2792 spa_async_suspend(spa);
2793
2794 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2795}
2796
2797/*
2798 * If spa_load() fails this function will try loading prior txg's. If
2799 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2800 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2801 * function will not rewind the pool and will return the same error as
2802 * spa_load().
2803 */
2804static int
2805spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2806 uint64_t max_request, int rewind_flags)
2807{
2808 nvlist_t *loadinfo = NULL;
2809 nvlist_t *config = NULL;
2810 int load_error, rewind_error;
2811 uint64_t safe_rewind_txg;
2812 uint64_t min_txg;
2813
2814 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2815 spa->spa_load_max_txg = spa->spa_load_txg;
2816 spa_set_log_state(spa, SPA_LOG_CLEAR);
2817 } else {
2818 spa->spa_load_max_txg = max_request;
2819 }
2820
2821 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2822 mosconfig);
2823 if (load_error == 0)
2824 return (0);
2825
2826 if (spa->spa_root_vdev != NULL)
2827 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2828
2829 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2830 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2831
2832 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2833 nvlist_free(config);
2834 return (load_error);
2835 }
2836
2837 if (state == SPA_LOAD_RECOVER) {
2838 /* Price of rolling back is discarding txgs, including log */
2839 spa_set_log_state(spa, SPA_LOG_CLEAR);
2840 } else {
2841 /*
2842 * If we aren't rolling back save the load info from our first
2843 * import attempt so that we can restore it after attempting
2844 * to rewind.
2845 */
2846 loadinfo = spa->spa_load_info;
2847 spa->spa_load_info = fnvlist_alloc();
2848 }
2849
2850 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2851 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2852 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2853 TXG_INITIAL : safe_rewind_txg;
2854
2855 /*
2856 * Continue as long as we're finding errors, we're still within
2857 * the acceptable rewind range, and we're still finding uberblocks
2858 */
2859 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2860 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2861 if (spa->spa_load_max_txg < safe_rewind_txg)
2862 spa->spa_extreme_rewind = B_TRUE;
2863 rewind_error = spa_load_retry(spa, state, mosconfig);
2864 }
2865
2866 spa->spa_extreme_rewind = B_FALSE;
2867 spa->spa_load_max_txg = UINT64_MAX;
2868
2869 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2870 spa_config_set(spa, config);
2871
2872 if (state == SPA_LOAD_RECOVER) {
2873 ASSERT3P(loadinfo, ==, NULL);
2874 return (rewind_error);
2875 } else {
2876 /* Store the rewind info as part of the initial load info */
2877 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2878 spa->spa_load_info);
2879
2880 /* Restore the initial load info */
2881 fnvlist_free(spa->spa_load_info);
2882 spa->spa_load_info = loadinfo;
2883
2884 return (load_error);
2885 }
2886}
2887
2888/*
2889 * Pool Open/Import
2890 *
2891 * The import case is identical to an open except that the configuration is sent
2892 * down from userland, instead of grabbed from the configuration cache. For the
2893 * case of an open, the pool configuration will exist in the
2894 * POOL_STATE_UNINITIALIZED state.
2895 *
2896 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2897 * the same time open the pool, without having to keep around the spa_t in some
2898 * ambiguous state.
2899 */
2900static int
2901spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2902 nvlist_t **config)
2903{
2904 spa_t *spa;
2905 spa_load_state_t state = SPA_LOAD_OPEN;
2906 int error;
2907 int locked = B_FALSE;
2908 int firstopen = B_FALSE;
2909
2910 *spapp = NULL;
2911
2912 /*
2913 * As disgusting as this is, we need to support recursive calls to this
2914 * function because dsl_dir_open() is called during spa_load(), and ends
2915 * up calling spa_open() again. The real fix is to figure out how to
2916 * avoid dsl_dir_open() calling this in the first place.
2917 */
2918 if (mutex_owner(&spa_namespace_lock) != curthread) {
2919 mutex_enter(&spa_namespace_lock);
2920 locked = B_TRUE;
2921 }
2922
2923 if ((spa = spa_lookup(pool)) == NULL) {
2924 if (locked)
2925 mutex_exit(&spa_namespace_lock);
2926 return (SET_ERROR(ENOENT));
2927 }
2928
2929 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2930 zpool_rewind_policy_t policy;
2931
2932 firstopen = B_TRUE;
2933
2934 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2935 &policy);
2936 if (policy.zrp_request & ZPOOL_DO_REWIND)
2937 state = SPA_LOAD_RECOVER;
2938
2939 spa_activate(spa, spa_mode_global);
2940
2941 if (state != SPA_LOAD_RECOVER)
2942 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2943
2944 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2945 policy.zrp_request);
2946
2947 if (error == EBADF) {
2948 /*
2949 * If vdev_validate() returns failure (indicated by
2950 * EBADF), it indicates that one of the vdevs indicates
2951 * that the pool has been exported or destroyed. If
2952 * this is the case, the config cache is out of sync and
2953 * we should remove the pool from the namespace.
2954 */
2955 spa_unload(spa);
2956 spa_deactivate(spa);
2957 spa_config_sync(spa, B_TRUE, B_TRUE);
2958 spa_remove(spa);
2959 if (locked)
2960 mutex_exit(&spa_namespace_lock);
2961 return (SET_ERROR(ENOENT));
2962 }
2963
2964 if (error) {
2965 /*
2966 * We can't open the pool, but we still have useful
2967 * information: the state of each vdev after the
2968 * attempted vdev_open(). Return this to the user.
2969 */
2970 if (config != NULL && spa->spa_config) {
2971 VERIFY(nvlist_dup(spa->spa_config, config,
2972 KM_SLEEP) == 0);
2973 VERIFY(nvlist_add_nvlist(*config,
2974 ZPOOL_CONFIG_LOAD_INFO,
2975 spa->spa_load_info) == 0);
2976 }
2977 spa_unload(spa);
2978 spa_deactivate(spa);
2979 spa->spa_last_open_failed = error;
2980 if (locked)
2981 mutex_exit(&spa_namespace_lock);
2982 *spapp = NULL;
2983 return (error);
2984 }
2985 }
2986
2987 spa_open_ref(spa, tag);
2988
2989 if (config != NULL)
2990 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2991
2992 /*
2993 * If we've recovered the pool, pass back any information we
2994 * gathered while doing the load.
2995 */
2996 if (state == SPA_LOAD_RECOVER) {
2997 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2998 spa->spa_load_info) == 0);
2999 }
3000
3001 if (locked) {
3002 spa->spa_last_open_failed = 0;
3003 spa->spa_last_ubsync_txg = 0;
3004 spa->spa_load_txg = 0;
3005 mutex_exit(&spa_namespace_lock);
3006#ifdef __FreeBSD__
3007#ifdef _KERNEL
3008 if (firstopen)
3009 zvol_create_minors(spa->spa_name);
3010#endif
3011#endif
3012 }
3013
3014 *spapp = spa;
3015
3016 return (0);
3017}
3018
3019int
3020spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3021 nvlist_t **config)
3022{
3023 return (spa_open_common(name, spapp, tag, policy, config));
3024}
3025
3026int
3027spa_open(const char *name, spa_t **spapp, void *tag)
3028{
3029 return (spa_open_common(name, spapp, tag, NULL, NULL));
3030}
3031
3032/*
3033 * Lookup the given spa_t, incrementing the inject count in the process,
3034 * preventing it from being exported or destroyed.
3035 */
3036spa_t *
3037spa_inject_addref(char *name)
3038{
3039 spa_t *spa;
3040
3041 mutex_enter(&spa_namespace_lock);
3042 if ((spa = spa_lookup(name)) == NULL) {
3043 mutex_exit(&spa_namespace_lock);
3044 return (NULL);
3045 }
3046 spa->spa_inject_ref++;
3047 mutex_exit(&spa_namespace_lock);
3048
3049 return (spa);
3050}
3051
3052void
3053spa_inject_delref(spa_t *spa)
3054{
3055 mutex_enter(&spa_namespace_lock);
3056 spa->spa_inject_ref--;
3057 mutex_exit(&spa_namespace_lock);
3058}
3059
3060/*
3061 * Add spares device information to the nvlist.
3062 */
3063static void
3064spa_add_spares(spa_t *spa, nvlist_t *config)
3065{
3066 nvlist_t **spares;
3067 uint_t i, nspares;
3068 nvlist_t *nvroot;
3069 uint64_t guid;
3070 vdev_stat_t *vs;
3071 uint_t vsc;
3072 uint64_t pool;
3073
3074 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3075
3076 if (spa->spa_spares.sav_count == 0)
3077 return;
3078
3079 VERIFY(nvlist_lookup_nvlist(config,
3080 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3081 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3082 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3083 if (nspares != 0) {
3084 VERIFY(nvlist_add_nvlist_array(nvroot,
3085 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3086 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3087 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3088
3089 /*
3090 * Go through and find any spares which have since been
3091 * repurposed as an active spare. If this is the case, update
3092 * their status appropriately.
3093 */
3094 for (i = 0; i < nspares; i++) {
3095 VERIFY(nvlist_lookup_uint64(spares[i],
3096 ZPOOL_CONFIG_GUID, &guid) == 0);
3097 if (spa_spare_exists(guid, &pool, NULL) &&
3098 pool != 0ULL) {
3099 VERIFY(nvlist_lookup_uint64_array(
3100 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3101 (uint64_t **)&vs, &vsc) == 0);
3102 vs->vs_state = VDEV_STATE_CANT_OPEN;
3103 vs->vs_aux = VDEV_AUX_SPARED;
3104 }
3105 }
3106 }
3107}
3108
3109/*
3110 * Add l2cache device information to the nvlist, including vdev stats.
3111 */
3112static void
3113spa_add_l2cache(spa_t *spa, nvlist_t *config)
3114{
3115 nvlist_t **l2cache;
3116 uint_t i, j, nl2cache;
3117 nvlist_t *nvroot;
3118 uint64_t guid;
3119 vdev_t *vd;
3120 vdev_stat_t *vs;
3121 uint_t vsc;
3122
3123 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3124
3125 if (spa->spa_l2cache.sav_count == 0)
3126 return;
3127
3128 VERIFY(nvlist_lookup_nvlist(config,
3129 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3130 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3131 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3132 if (nl2cache != 0) {
3133 VERIFY(nvlist_add_nvlist_array(nvroot,
3134 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3135 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3136 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3137
3138 /*
3139 * Update level 2 cache device stats.
3140 */
3141
3142 for (i = 0; i < nl2cache; i++) {
3143 VERIFY(nvlist_lookup_uint64(l2cache[i],
3144 ZPOOL_CONFIG_GUID, &guid) == 0);
3145
3146 vd = NULL;
3147 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3148 if (guid ==
3149 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3150 vd = spa->spa_l2cache.sav_vdevs[j];
3151 break;
3152 }
3153 }
3154 ASSERT(vd != NULL);
3155
3156 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3157 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3158 == 0);
3159 vdev_get_stats(vd, vs);
3160 }
3161 }
3162}
3163
3164static void
3165spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3166{
3167 nvlist_t *features;
3168 zap_cursor_t zc;
3169 zap_attribute_t za;
3170
3171 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3172 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3173
3174 /* We may be unable to read features if pool is suspended. */
3175 if (spa_suspended(spa))
3176 goto out;
3177
3178 if (spa->spa_feat_for_read_obj != 0) {
3179 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3180 spa->spa_feat_for_read_obj);
3181 zap_cursor_retrieve(&zc, &za) == 0;
3182 zap_cursor_advance(&zc)) {
3183 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3184 za.za_num_integers == 1);
3185 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3186 za.za_first_integer));
3187 }
3188 zap_cursor_fini(&zc);
3189 }
3190
3191 if (spa->spa_feat_for_write_obj != 0) {
3192 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3193 spa->spa_feat_for_write_obj);
3194 zap_cursor_retrieve(&zc, &za) == 0;
3195 zap_cursor_advance(&zc)) {
3196 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3197 za.za_num_integers == 1);
3198 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3199 za.za_first_integer));
3200 }
3201 zap_cursor_fini(&zc);
3202 }
3203
3204out:
3205 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3206 features) == 0);
3207 nvlist_free(features);
3208}
3209
3210int
3211spa_get_stats(const char *name, nvlist_t **config,
3212 char *altroot, size_t buflen)
3213{
3214 int error;
3215 spa_t *spa;
3216
3217 *config = NULL;
3218 error = spa_open_common(name, &spa, FTAG, NULL, config);
3219
3220 if (spa != NULL) {
3221 /*
3222 * This still leaves a window of inconsistency where the spares
3223 * or l2cache devices could change and the config would be
3224 * self-inconsistent.
3225 */
3226 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3227
3228 if (*config != NULL) {
3229 uint64_t loadtimes[2];
3230
3231 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3232 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3233 VERIFY(nvlist_add_uint64_array(*config,
3234 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3235
3236 VERIFY(nvlist_add_uint64(*config,
3237 ZPOOL_CONFIG_ERRCOUNT,
3238 spa_get_errlog_size(spa)) == 0);
3239
3240 if (spa_suspended(spa))
3241 VERIFY(nvlist_add_uint64(*config,
3242 ZPOOL_CONFIG_SUSPENDED,
3243 spa->spa_failmode) == 0);
3244
3245 spa_add_spares(spa, *config);
3246 spa_add_l2cache(spa, *config);
3247 spa_add_feature_stats(spa, *config);
3248 }
3249 }
3250
3251 /*
3252 * We want to get the alternate root even for faulted pools, so we cheat
3253 * and call spa_lookup() directly.
3254 */
3255 if (altroot) {
3256 if (spa == NULL) {
3257 mutex_enter(&spa_namespace_lock);
3258 spa = spa_lookup(name);
3259 if (spa)
3260 spa_altroot(spa, altroot, buflen);
3261 else
3262 altroot[0] = '\0';
3263 spa = NULL;
3264 mutex_exit(&spa_namespace_lock);
3265 } else {
3266 spa_altroot(spa, altroot, buflen);
3267 }
3268 }
3269
3270 if (spa != NULL) {
3271 spa_config_exit(spa, SCL_CONFIG, FTAG);
3272 spa_close(spa, FTAG);
3273 }
3274
3275 return (error);
3276}
3277
3278/*
3279 * Validate that the auxiliary device array is well formed. We must have an
3280 * array of nvlists, each which describes a valid leaf vdev. If this is an
3281 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3282 * specified, as long as they are well-formed.
3283 */
3284static int
3285spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3286 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3287 vdev_labeltype_t label)
3288{
3289 nvlist_t **dev;
3290 uint_t i, ndev;
3291 vdev_t *vd;
3292 int error;
3293
3294 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3295
3296 /*
3297 * It's acceptable to have no devs specified.
3298 */
3299 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3300 return (0);
3301
3302 if (ndev == 0)
3303 return (SET_ERROR(EINVAL));
3304
3305 /*
3306 * Make sure the pool is formatted with a version that supports this
3307 * device type.
3308 */
3309 if (spa_version(spa) < version)
3310 return (SET_ERROR(ENOTSUP));
3311
3312 /*
3313 * Set the pending device list so we correctly handle device in-use
3314 * checking.
3315 */
3316 sav->sav_pending = dev;
3317 sav->sav_npending = ndev;
3318
3319 for (i = 0; i < ndev; i++) {
3320 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3321 mode)) != 0)
3322 goto out;
3323
3324 if (!vd->vdev_ops->vdev_op_leaf) {
3325 vdev_free(vd);
3326 error = SET_ERROR(EINVAL);
3327 goto out;
3328 }
3329
3330 /*
3331 * The L2ARC currently only supports disk devices in
3332 * kernel context. For user-level testing, we allow it.
3333 */
3334#ifdef _KERNEL
3335 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3336 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3337 error = SET_ERROR(ENOTBLK);
3338 vdev_free(vd);
3339 goto out;
3340 }
3341#endif
3342 vd->vdev_top = vd;
3343
3344 if ((error = vdev_open(vd)) == 0 &&
3345 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3346 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3347 vd->vdev_guid) == 0);
3348 }
3349
3350 vdev_free(vd);
3351
3352 if (error &&
3353 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3354 goto out;
3355 else
3356 error = 0;
3357 }
3358
3359out:
3360 sav->sav_pending = NULL;
3361 sav->sav_npending = 0;
3362 return (error);
3363}
3364
3365static int
3366spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3367{
3368 int error;
3369
3370 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3371
3372 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3373 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3374 VDEV_LABEL_SPARE)) != 0) {
3375 return (error);
3376 }
3377
3378 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3379 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3380 VDEV_LABEL_L2CACHE));
3381}
3382
3383static void
3384spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3385 const char *config)
3386{
3387 int i;
3388
3389 if (sav->sav_config != NULL) {
3390 nvlist_t **olddevs;
3391 uint_t oldndevs;
3392 nvlist_t **newdevs;
3393
3394 /*
3395 * Generate new dev list by concatentating with the
3396 * current dev list.
3397 */
3398 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3399 &olddevs, &oldndevs) == 0);
3400
3401 newdevs = kmem_alloc(sizeof (void *) *
3402 (ndevs + oldndevs), KM_SLEEP);
3403 for (i = 0; i < oldndevs; i++)
3404 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3405 KM_SLEEP) == 0);
3406 for (i = 0; i < ndevs; i++)
3407 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3408 KM_SLEEP) == 0);
3409
3410 VERIFY(nvlist_remove(sav->sav_config, config,
3411 DATA_TYPE_NVLIST_ARRAY) == 0);
3412
3413 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3414 config, newdevs, ndevs + oldndevs) == 0);
3415 for (i = 0; i < oldndevs + ndevs; i++)
3416 nvlist_free(newdevs[i]);
3417 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3418 } else {
3419 /*
3420 * Generate a new dev list.
3421 */
3422 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3423 KM_SLEEP) == 0);
3424 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3425 devs, ndevs) == 0);
3426 }
3427}
3428
3429/*
3430 * Stop and drop level 2 ARC devices
3431 */
3432void
3433spa_l2cache_drop(spa_t *spa)
3434{
3435 vdev_t *vd;
3436 int i;
3437 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3438
3439 for (i = 0; i < sav->sav_count; i++) {
3440 uint64_t pool;
3441
3442 vd = sav->sav_vdevs[i];
3443 ASSERT(vd != NULL);
3444
3445 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3446 pool != 0ULL && l2arc_vdev_present(vd))
3447 l2arc_remove_vdev(vd);
3448 }
3449}
3450
3451/*
3452 * Pool Creation
3453 */
3454int
3455spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3456 nvlist_t *zplprops)
3457{
3458 spa_t *spa;
3459 char *altroot = NULL;
3460 vdev_t *rvd;
3461 dsl_pool_t *dp;
3462 dmu_tx_t *tx;
3463 int error = 0;
3464 uint64_t txg = TXG_INITIAL;
3465 nvlist_t **spares, **l2cache;
3466 uint_t nspares, nl2cache;
3467 uint64_t version, obj;
3468 boolean_t has_features;
3469
3470 /*
3471 * If this pool already exists, return failure.
3472 */
3473 mutex_enter(&spa_namespace_lock);
3474 if (spa_lookup(pool) != NULL) {
3475 mutex_exit(&spa_namespace_lock);
3476 return (SET_ERROR(EEXIST));
3477 }
3478
3479 /*
3480 * Allocate a new spa_t structure.
3481 */
3482 (void) nvlist_lookup_string(props,
3483 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3484 spa = spa_add(pool, NULL, altroot);
3485 spa_activate(spa, spa_mode_global);
3486
3487 if (props && (error = spa_prop_validate(spa, props))) {
3488 spa_deactivate(spa);
3489 spa_remove(spa);
3490 mutex_exit(&spa_namespace_lock);
3491 return (error);
3492 }
3493
3494 has_features = B_FALSE;
3495 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3496 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3497 if (zpool_prop_feature(nvpair_name(elem)))
3498 has_features = B_TRUE;
3499 }
3500
3501 if (has_features || nvlist_lookup_uint64(props,
3502 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3503 version = SPA_VERSION;
3504 }
3505 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3506
3507 spa->spa_first_txg = txg;
3508 spa->spa_uberblock.ub_txg = txg - 1;
3509 spa->spa_uberblock.ub_version = version;
3510 spa->spa_ubsync = spa->spa_uberblock;
3511
3512 /*
3513 * Create "The Godfather" zio to hold all async IOs
3514 */
3515 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3516 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3517
3518 /*
3519 * Create the root vdev.
3520 */
3521 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3522
3523 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3524
3525 ASSERT(error != 0 || rvd != NULL);
3526 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3527
3528 if (error == 0 && !zfs_allocatable_devs(nvroot))
3529 error = SET_ERROR(EINVAL);
3530
3531 if (error == 0 &&
3532 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3533 (error = spa_validate_aux(spa, nvroot, txg,
3534 VDEV_ALLOC_ADD)) == 0) {
3535 for (int c = 0; c < rvd->vdev_children; c++) {
3536 vdev_ashift_optimize(rvd->vdev_child[c]);
3537 vdev_metaslab_set_size(rvd->vdev_child[c]);
3538 vdev_expand(rvd->vdev_child[c], txg);
3539 }
3540 }
3541
3542 spa_config_exit(spa, SCL_ALL, FTAG);
3543
3544 if (error != 0) {
3545 spa_unload(spa);
3546 spa_deactivate(spa);
3547 spa_remove(spa);
3548 mutex_exit(&spa_namespace_lock);
3549 return (error);
3550 }
3551
3552 /*
3553 * Get the list of spares, if specified.
3554 */
3555 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3556 &spares, &nspares) == 0) {
3557 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3558 KM_SLEEP) == 0);
3559 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3560 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3561 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3562 spa_load_spares(spa);
3563 spa_config_exit(spa, SCL_ALL, FTAG);
3564 spa->spa_spares.sav_sync = B_TRUE;
3565 }
3566
3567 /*
3568 * Get the list of level 2 cache devices, if specified.
3569 */
3570 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3571 &l2cache, &nl2cache) == 0) {
3572 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3573 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3574 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3575 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3576 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3577 spa_load_l2cache(spa);
3578 spa_config_exit(spa, SCL_ALL, FTAG);
3579 spa->spa_l2cache.sav_sync = B_TRUE;
3580 }
3581
3582 spa->spa_is_initializing = B_TRUE;
3583 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3584 spa->spa_meta_objset = dp->dp_meta_objset;
3585 spa->spa_is_initializing = B_FALSE;
3586
3587 /*
3588 * Create DDTs (dedup tables).
3589 */
3590 ddt_create(spa);
3591
3592 spa_update_dspace(spa);
3593
3594 tx = dmu_tx_create_assigned(dp, txg);
3595
3596 /*
3597 * Create the pool config object.
3598 */
3599 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3600 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3601 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3602
3603 if (zap_add(spa->spa_meta_objset,
3604 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3605 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3606 cmn_err(CE_PANIC, "failed to add pool config");
3607 }
3608
3609 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3610 spa_feature_create_zap_objects(spa, tx);
3611
3612 if (zap_add(spa->spa_meta_objset,
3613 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3614 sizeof (uint64_t), 1, &version, tx) != 0) {
3615 cmn_err(CE_PANIC, "failed to add pool version");
3616 }
3617
3618 /* Newly created pools with the right version are always deflated. */
3619 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3620 spa->spa_deflate = TRUE;
3621 if (zap_add(spa->spa_meta_objset,
3622 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3623 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3624 cmn_err(CE_PANIC, "failed to add deflate");
3625 }
3626 }
3627
3628 /*
3629 * Create the deferred-free bpobj. Turn off compression
3630 * because sync-to-convergence takes longer if the blocksize
3631 * keeps changing.
3632 */
3633 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3634 dmu_object_set_compress(spa->spa_meta_objset, obj,
3635 ZIO_COMPRESS_OFF, tx);
3636 if (zap_add(spa->spa_meta_objset,
3637 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3638 sizeof (uint64_t), 1, &obj, tx) != 0) {
3639 cmn_err(CE_PANIC, "failed to add bpobj");
3640 }
3641 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3642 spa->spa_meta_objset, obj));
3643
3644 /*
3645 * Create the pool's history object.
3646 */
3647 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3648 spa_history_create_obj(spa, tx);
3649
3650 /*
3651 * Set pool properties.
3652 */
3653 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3654 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3655 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3656 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3657
3658 if (props != NULL) {
3659 spa_configfile_set(spa, props, B_FALSE);
3660 spa_sync_props(props, tx);
3661 }
3662
3663 dmu_tx_commit(tx);
3664
3665 spa->spa_sync_on = B_TRUE;
3666 txg_sync_start(spa->spa_dsl_pool);
3667
3668 /*
3669 * We explicitly wait for the first transaction to complete so that our
3670 * bean counters are appropriately updated.
3671 */
3672 txg_wait_synced(spa->spa_dsl_pool, txg);
3673
3674 spa_config_sync(spa, B_FALSE, B_TRUE);
3675
3676 spa_history_log_version(spa, "create");
3677
3678 spa->spa_minref = refcount_count(&spa->spa_refcount);
3679
3680 mutex_exit(&spa_namespace_lock);
3681
3682 return (0);
3683}
3684
3685#ifdef _KERNEL
3686#if defined(sun)
3687/*
3688 * Get the root pool information from the root disk, then import the root pool
3689 * during the system boot up time.
3690 */
3691extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3692
3693static nvlist_t *
3694spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3695{
3696 nvlist_t *config;
3697 nvlist_t *nvtop, *nvroot;
3698 uint64_t pgid;
3699
3700 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3701 return (NULL);
3702
3703 /*
3704 * Add this top-level vdev to the child array.
3705 */
3706 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3707 &nvtop) == 0);
3708 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3709 &pgid) == 0);
3710 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3711
3712 /*
3713 * Put this pool's top-level vdevs into a root vdev.
3714 */
3715 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3716 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3717 VDEV_TYPE_ROOT) == 0);
3718 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3719 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3720 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3721 &nvtop, 1) == 0);
3722
3723 /*
3724 * Replace the existing vdev_tree with the new root vdev in
3725 * this pool's configuration (remove the old, add the new).
3726 */
3727 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3728 nvlist_free(nvroot);
3729 return (config);
3730}
3731
3732/*
3733 * Walk the vdev tree and see if we can find a device with "better"
3734 * configuration. A configuration is "better" if the label on that
3735 * device has a more recent txg.
3736 */
3737static void
3738spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3739{
3740 for (int c = 0; c < vd->vdev_children; c++)
3741 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3742
3743 if (vd->vdev_ops->vdev_op_leaf) {
3744 nvlist_t *label;
3745 uint64_t label_txg;
3746
3747 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3748 &label) != 0)
3749 return;
3750
3751 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3752 &label_txg) == 0);
3753
3754 /*
3755 * Do we have a better boot device?
3756 */
3757 if (label_txg > *txg) {
3758 *txg = label_txg;
3759 *avd = vd;
3760 }
3761 nvlist_free(label);
3762 }
3763}
3764
3765/*
3766 * Import a root pool.
3767 *
3768 * For x86. devpath_list will consist of devid and/or physpath name of
3769 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3770 * The GRUB "findroot" command will return the vdev we should boot.
3771 *
3772 * For Sparc, devpath_list consists the physpath name of the booting device
3773 * no matter the rootpool is a single device pool or a mirrored pool.
3774 * e.g.
3775 * "/pci@1f,0/ide@d/disk@0,0:a"
3776 */
3777int
3778spa_import_rootpool(char *devpath, char *devid)
3779{
3780 spa_t *spa;
3781 vdev_t *rvd, *bvd, *avd = NULL;
3782 nvlist_t *config, *nvtop;
3783 uint64_t guid, txg;
3784 char *pname;
3785 int error;
3786
3787 /*
3788 * Read the label from the boot device and generate a configuration.
3789 */
3790 config = spa_generate_rootconf(devpath, devid, &guid);
3791#if defined(_OBP) && defined(_KERNEL)
3792 if (config == NULL) {
3793 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3794 /* iscsi boot */
3795 get_iscsi_bootpath_phy(devpath);
3796 config = spa_generate_rootconf(devpath, devid, &guid);
3797 }
3798 }
3799#endif
3800 if (config == NULL) {
3801 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3802 devpath);
3803 return (SET_ERROR(EIO));
3804 }
3805
3806 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3807 &pname) == 0);
3808 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3809
3810 mutex_enter(&spa_namespace_lock);
3811 if ((spa = spa_lookup(pname)) != NULL) {
3812 /*
3813 * Remove the existing root pool from the namespace so that we
3814 * can replace it with the correct config we just read in.
3815 */
3816 spa_remove(spa);
3817 }
3818
3819 spa = spa_add(pname, config, NULL);
3820 spa->spa_is_root = B_TRUE;
3821 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3822
3823 /*
3824 * Build up a vdev tree based on the boot device's label config.
3825 */
3826 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3827 &nvtop) == 0);
3828 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3829 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3830 VDEV_ALLOC_ROOTPOOL);
3831 spa_config_exit(spa, SCL_ALL, FTAG);
3832 if (error) {
3833 mutex_exit(&spa_namespace_lock);
3834 nvlist_free(config);
3835 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3836 pname);
3837 return (error);
3838 }
3839
3840 /*
3841 * Get the boot vdev.
3842 */
3843 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3844 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3845 (u_longlong_t)guid);
3846 error = SET_ERROR(ENOENT);
3847 goto out;
3848 }
3849
3850 /*
3851 * Determine if there is a better boot device.
3852 */
3853 avd = bvd;
3854 spa_alt_rootvdev(rvd, &avd, &txg);
3855 if (avd != bvd) {
3856 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3857 "try booting from '%s'", avd->vdev_path);
3858 error = SET_ERROR(EINVAL);
3859 goto out;
3860 }
3861
3862 /*
3863 * If the boot device is part of a spare vdev then ensure that
3864 * we're booting off the active spare.
3865 */
3866 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3867 !bvd->vdev_isspare) {
3868 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3869 "try booting from '%s'",
3870 bvd->vdev_parent->
3871 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3872 error = SET_ERROR(EINVAL);
3873 goto out;
3874 }
3875
3876 error = 0;
3877out:
3878 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3879 vdev_free(rvd);
3880 spa_config_exit(spa, SCL_ALL, FTAG);
3881 mutex_exit(&spa_namespace_lock);
3882
3883 nvlist_free(config);
3884 return (error);
3885}
3886
3887#else
3888
3889extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3890 uint64_t *count);
3891
3892static nvlist_t *
3893spa_generate_rootconf(const char *name)
3894{
3895 nvlist_t **configs, **tops;
3896 nvlist_t *config;
3897 nvlist_t *best_cfg, *nvtop, *nvroot;
3898 uint64_t *holes;
3899 uint64_t best_txg;
3900 uint64_t nchildren;
3901 uint64_t pgid;
3902 uint64_t count;
3903 uint64_t i;
3904 uint_t nholes;
3905
3906 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3907 return (NULL);
3908
3909 ASSERT3U(count, !=, 0);
3910 best_txg = 0;
3911 for (i = 0; i < count; i++) {
3912 uint64_t txg;
3913
3914 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3915 &txg) == 0);
3916 if (txg > best_txg) {
3917 best_txg = txg;
3918 best_cfg = configs[i];
3919 }
3920 }
3921
3922 /*
3923 * Multi-vdev root pool configuration discovery is not supported yet.
3924 */
3925 nchildren = 1;
3926 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3927 holes = NULL;
3928 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3929 &holes, &nholes);
3930
3931 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3932 for (i = 0; i < nchildren; i++) {
3933 if (i >= count)
3934 break;
3935 if (configs[i] == NULL)
3936 continue;
3937 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3938 &nvtop) == 0);
3939 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3940 }
3941 for (i = 0; holes != NULL && i < nholes; i++) {
3942 if (i >= nchildren)
3943 continue;
3944 if (tops[holes[i]] != NULL)
3945 continue;
3946 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3947 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3948 VDEV_TYPE_HOLE) == 0);
3949 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3950 holes[i]) == 0);
3951 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3952 0) == 0);
3953 }
3954 for (i = 0; i < nchildren; i++) {
3955 if (tops[i] != NULL)
3956 continue;
3957 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3958 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3959 VDEV_TYPE_MISSING) == 0);
3960 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3961 i) == 0);
3962 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3963 0) == 0);
3964 }
3965
3966 /*
3967 * Create pool config based on the best vdev config.
3968 */
3969 nvlist_dup(best_cfg, &config, KM_SLEEP);
3970
3971 /*
3972 * Put this pool's top-level vdevs into a root vdev.
3973 */
3974 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3975 &pgid) == 0);
3976 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3977 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3978 VDEV_TYPE_ROOT) == 0);
3979 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3980 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3981 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3982 tops, nchildren) == 0);
3983
3984 /*
3985 * Replace the existing vdev_tree with the new root vdev in
3986 * this pool's configuration (remove the old, add the new).
3987 */
3988 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3989
3990 /*
3991 * Drop vdev config elements that should not be present at pool level.
3992 */
3993 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3994 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3995
3996 for (i = 0; i < count; i++)
3997 nvlist_free(configs[i]);
3998 kmem_free(configs, count * sizeof(void *));
3999 for (i = 0; i < nchildren; i++)
4000 nvlist_free(tops[i]);
4001 kmem_free(tops, nchildren * sizeof(void *));
4002 nvlist_free(nvroot);
4003 return (config);
4004}
4005
4006int
4007spa_import_rootpool(const char *name)
4008{
4009 spa_t *spa;
4010 vdev_t *rvd, *bvd, *avd = NULL;
4011 nvlist_t *config, *nvtop;
4012 uint64_t txg;
4013 char *pname;
4014 int error;
4015
4016 /*
4017 * Read the label from the boot device and generate a configuration.
4018 */
4019 config = spa_generate_rootconf(name);
4020
4021 mutex_enter(&spa_namespace_lock);
4022 if (config != NULL) {
4023 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4024 &pname) == 0 && strcmp(name, pname) == 0);
4025 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4026 == 0);
4027
4028 if ((spa = spa_lookup(pname)) != NULL) {
4029 /*
4030 * Remove the existing root pool from the namespace so
4031 * that we can replace it with the correct config
4032 * we just read in.
4033 */
4034 spa_remove(spa);
4035 }
4036 spa = spa_add(pname, config, NULL);
4037
4038 /*
4039 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4040 * via spa_version().
4041 */
4042 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4043 &spa->spa_ubsync.ub_version) != 0)
4044 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4045 } else if ((spa = spa_lookup(name)) == NULL) {
4046 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4047 name);
4048 return (EIO);
4049 } else {
4050 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4051 }
4052 spa->spa_is_root = B_TRUE;
4053 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4054
4055 /*
4056 * Build up a vdev tree based on the boot device's label config.
4057 */
4058 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4059 &nvtop) == 0);
4060 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4061 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4062 VDEV_ALLOC_ROOTPOOL);
4063 spa_config_exit(spa, SCL_ALL, FTAG);
4064 if (error) {
4065 mutex_exit(&spa_namespace_lock);
4066 nvlist_free(config);
4067 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4068 pname);
4069 return (error);
4070 }
4071
4072 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4073 vdev_free(rvd);
4074 spa_config_exit(spa, SCL_ALL, FTAG);
4075 mutex_exit(&spa_namespace_lock);
4076
4077 nvlist_free(config);
4078 return (0);
4079}
4080
4081#endif /* sun */
4082#endif
4083
4084/*
4085 * Import a non-root pool into the system.
4086 */
4087int
4088spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4089{
4090 spa_t *spa;
4091 char *altroot = NULL;
4092 spa_load_state_t state = SPA_LOAD_IMPORT;
4093 zpool_rewind_policy_t policy;
4094 uint64_t mode = spa_mode_global;
4095 uint64_t readonly = B_FALSE;
4096 int error;
4097 nvlist_t *nvroot;
4098 nvlist_t **spares, **l2cache;
4099 uint_t nspares, nl2cache;
4100
4101 /*
4102 * If a pool with this name exists, return failure.
4103 */
4104 mutex_enter(&spa_namespace_lock);
4105 if (spa_lookup(pool) != NULL) {
4106 mutex_exit(&spa_namespace_lock);
4107 return (SET_ERROR(EEXIST));
4108 }
4109
4110 /*
4111 * Create and initialize the spa structure.
4112 */
4113 (void) nvlist_lookup_string(props,
4114 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4115 (void) nvlist_lookup_uint64(props,
4116 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4117 if (readonly)
4118 mode = FREAD;
4119 spa = spa_add(pool, config, altroot);
4120 spa->spa_import_flags = flags;
4121
4122 /*
4123 * Verbatim import - Take a pool and insert it into the namespace
4124 * as if it had been loaded at boot.
4125 */
4126 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4127 if (props != NULL)
4128 spa_configfile_set(spa, props, B_FALSE);
4129
4130 spa_config_sync(spa, B_FALSE, B_TRUE);
4131
4132 mutex_exit(&spa_namespace_lock);
4133 return (0);
4134 }
4135
4136 spa_activate(spa, mode);
4137
4138 /*
4139 * Don't start async tasks until we know everything is healthy.
4140 */
4141 spa_async_suspend(spa);
4142
4143 zpool_get_rewind_policy(config, &policy);
4144 if (policy.zrp_request & ZPOOL_DO_REWIND)
4145 state = SPA_LOAD_RECOVER;
4146
4147 /*
4148 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4149 * because the user-supplied config is actually the one to trust when
4150 * doing an import.
4151 */
4152 if (state != SPA_LOAD_RECOVER)
4153 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4154
4155 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4156 policy.zrp_request);
4157
4158 /*
4159 * Propagate anything learned while loading the pool and pass it
4160 * back to caller (i.e. rewind info, missing devices, etc).
4161 */
4162 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4163 spa->spa_load_info) == 0);
4164
4165 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4166 /*
4167 * Toss any existing sparelist, as it doesn't have any validity
4168 * anymore, and conflicts with spa_has_spare().
4169 */
4170 if (spa->spa_spares.sav_config) {
4171 nvlist_free(spa->spa_spares.sav_config);
4172 spa->spa_spares.sav_config = NULL;
4173 spa_load_spares(spa);
4174 }
4175 if (spa->spa_l2cache.sav_config) {
4176 nvlist_free(spa->spa_l2cache.sav_config);
4177 spa->spa_l2cache.sav_config = NULL;
4178 spa_load_l2cache(spa);
4179 }
4180
4181 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4182 &nvroot) == 0);
4183 if (error == 0)
4184 error = spa_validate_aux(spa, nvroot, -1ULL,
4185 VDEV_ALLOC_SPARE);
4186 if (error == 0)
4187 error = spa_validate_aux(spa, nvroot, -1ULL,
4188 VDEV_ALLOC_L2CACHE);
4189 spa_config_exit(spa, SCL_ALL, FTAG);
4190
4191 if (props != NULL)
4192 spa_configfile_set(spa, props, B_FALSE);
4193
4194 if (error != 0 || (props && spa_writeable(spa) &&
4195 (error = spa_prop_set(spa, props)))) {
4196 spa_unload(spa);
4197 spa_deactivate(spa);
4198 spa_remove(spa);
4199 mutex_exit(&spa_namespace_lock);
4200 return (error);
4201 }
4202
4203 spa_async_resume(spa);
4204
4205 /*
4206 * Override any spares and level 2 cache devices as specified by
4207 * the user, as these may have correct device names/devids, etc.
4208 */
4209 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4210 &spares, &nspares) == 0) {
4211 if (spa->spa_spares.sav_config)
4212 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4213 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4214 else
4215 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4216 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4217 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4218 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4219 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4220 spa_load_spares(spa);
4221 spa_config_exit(spa, SCL_ALL, FTAG);
4222 spa->spa_spares.sav_sync = B_TRUE;
4223 }
4224 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4225 &l2cache, &nl2cache) == 0) {
4226 if (spa->spa_l2cache.sav_config)
4227 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4228 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4229 else
4230 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4231 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4232 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4233 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4234 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4235 spa_load_l2cache(spa);
4236 spa_config_exit(spa, SCL_ALL, FTAG);
4237 spa->spa_l2cache.sav_sync = B_TRUE;
4238 }
4239
4240 /*
4241 * Check for any removed devices.
4242 */
4243 if (spa->spa_autoreplace) {
4244 spa_aux_check_removed(&spa->spa_spares);
4245 spa_aux_check_removed(&spa->spa_l2cache);
4246 }
4247
4248 if (spa_writeable(spa)) {
4249 /*
4250 * Update the config cache to include the newly-imported pool.
4251 */
4252 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4253 }
4254
4255 /*
4256 * It's possible that the pool was expanded while it was exported.
4257 * We kick off an async task to handle this for us.
4258 */
4259 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4260
4261 mutex_exit(&spa_namespace_lock);
4262 spa_history_log_version(spa, "import");
4263
4264#ifdef __FreeBSD__
4265#ifdef _KERNEL
4266 zvol_create_minors(pool);
4267#endif
4268#endif
4269 return (0);
4270}
4271
4272nvlist_t *
4273spa_tryimport(nvlist_t *tryconfig)
4274{
4275 nvlist_t *config = NULL;
4276 char *poolname;
4277 spa_t *spa;
4278 uint64_t state;
4279 int error;
4280
4281 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4282 return (NULL);
4283
4284 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4285 return (NULL);
4286
4287 /*
4288 * Create and initialize the spa structure.
4289 */
4290 mutex_enter(&spa_namespace_lock);
4291 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4292 spa_activate(spa, FREAD);
4293
4294 /*
4295 * Pass off the heavy lifting to spa_load().
4296 * Pass TRUE for mosconfig because the user-supplied config
4297 * is actually the one to trust when doing an import.
4298 */
4299 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4300
4301 /*
4302 * If 'tryconfig' was at least parsable, return the current config.
4303 */
4304 if (spa->spa_root_vdev != NULL) {
4305 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4306 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4307 poolname) == 0);
4308 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4309 state) == 0);
4310 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4311 spa->spa_uberblock.ub_timestamp) == 0);
4312 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4313 spa->spa_load_info) == 0);
4314
4315 /*
4316 * If the bootfs property exists on this pool then we
4317 * copy it out so that external consumers can tell which
4318 * pools are bootable.
4319 */
4320 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4321 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4322
4323 /*
4324 * We have to play games with the name since the
4325 * pool was opened as TRYIMPORT_NAME.
4326 */
4327 if (dsl_dsobj_to_dsname(spa_name(spa),
4328 spa->spa_bootfs, tmpname) == 0) {
4329 char *cp;
4330 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4331
4332 cp = strchr(tmpname, '/');
4333 if (cp == NULL) {
4334 (void) strlcpy(dsname, tmpname,
4335 MAXPATHLEN);
4336 } else {
4337 (void) snprintf(dsname, MAXPATHLEN,
4338 "%s/%s", poolname, ++cp);
4339 }
4340 VERIFY(nvlist_add_string(config,
4341 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4342 kmem_free(dsname, MAXPATHLEN);
4343 }
4344 kmem_free(tmpname, MAXPATHLEN);
4345 }
4346
4347 /*
4348 * Add the list of hot spares and level 2 cache devices.
4349 */
4350 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4351 spa_add_spares(spa, config);
4352 spa_add_l2cache(spa, config);
4353 spa_config_exit(spa, SCL_CONFIG, FTAG);
4354 }
4355
4356 spa_unload(spa);
4357 spa_deactivate(spa);
4358 spa_remove(spa);
4359 mutex_exit(&spa_namespace_lock);
4360
4361 return (config);
4362}
4363
4364/*
4365 * Pool export/destroy
4366 *
4367 * The act of destroying or exporting a pool is very simple. We make sure there
4368 * is no more pending I/O and any references to the pool are gone. Then, we
4369 * update the pool state and sync all the labels to disk, removing the
4370 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4371 * we don't sync the labels or remove the configuration cache.
4372 */
4373static int
4374spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4375 boolean_t force, boolean_t hardforce)
4376{
4377 spa_t *spa;
4378
4379 if (oldconfig)
4380 *oldconfig = NULL;
4381
4382 if (!(spa_mode_global & FWRITE))
4383 return (SET_ERROR(EROFS));
4384
4385 mutex_enter(&spa_namespace_lock);
4386 if ((spa = spa_lookup(pool)) == NULL) {
4387 mutex_exit(&spa_namespace_lock);
4388 return (SET_ERROR(ENOENT));
4389 }
4390
4391 /*
4392 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4393 * reacquire the namespace lock, and see if we can export.
4394 */
4395 spa_open_ref(spa, FTAG);
4396 mutex_exit(&spa_namespace_lock);
4397 spa_async_suspend(spa);
4398 mutex_enter(&spa_namespace_lock);
4399 spa_close(spa, FTAG);
4400
4401 /*
4402 * The pool will be in core if it's openable,
4403 * in which case we can modify its state.
4404 */
4405 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4406 /*
4407 * Objsets may be open only because they're dirty, so we
4408 * have to force it to sync before checking spa_refcnt.
4409 */
4410 txg_wait_synced(spa->spa_dsl_pool, 0);
4411
4412 /*
4413 * A pool cannot be exported or destroyed if there are active
4414 * references. If we are resetting a pool, allow references by
4415 * fault injection handlers.
4416 */
4417 if (!spa_refcount_zero(spa) ||
4418 (spa->spa_inject_ref != 0 &&
4419 new_state != POOL_STATE_UNINITIALIZED)) {
4420 spa_async_resume(spa);
4421 mutex_exit(&spa_namespace_lock);
4422 return (SET_ERROR(EBUSY));
4423 }
4424
4425 /*
4426 * A pool cannot be exported if it has an active shared spare.
4427 * This is to prevent other pools stealing the active spare
4428 * from an exported pool. At user's own will, such pool can
4429 * be forcedly exported.
4430 */
4431 if (!force && new_state == POOL_STATE_EXPORTED &&
4432 spa_has_active_shared_spare(spa)) {
4433 spa_async_resume(spa);
4434 mutex_exit(&spa_namespace_lock);
4435 return (SET_ERROR(EXDEV));
4436 }
4437
4438 /*
4439 * We want this to be reflected on every label,
4440 * so mark them all dirty. spa_unload() will do the
4441 * final sync that pushes these changes out.
4442 */
4443 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4444 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4445 spa->spa_state = new_state;
4446 spa->spa_final_txg = spa_last_synced_txg(spa) +
4447 TXG_DEFER_SIZE + 1;
4448 vdev_config_dirty(spa->spa_root_vdev);
4449 spa_config_exit(spa, SCL_ALL, FTAG);
4450 }
4451 }
4452
4453 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4454
4455 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4456 spa_unload(spa);
4457 spa_deactivate(spa);
4458 }
4459
4460 if (oldconfig && spa->spa_config)
4461 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4462
4463 if (new_state != POOL_STATE_UNINITIALIZED) {
4464 if (!hardforce)
4465 spa_config_sync(spa, B_TRUE, B_TRUE);
4466 spa_remove(spa);
4467 }
4468 mutex_exit(&spa_namespace_lock);
4469
4470 return (0);
4471}
4472
4473/*
4474 * Destroy a storage pool.
4475 */
4476int
4477spa_destroy(char *pool)
4478{
4479 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4480 B_FALSE, B_FALSE));
4481}
4482
4483/*
4484 * Export a storage pool.
4485 */
4486int
4487spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4488 boolean_t hardforce)
4489{
4490 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4491 force, hardforce));
4492}
4493
4494/*
4495 * Similar to spa_export(), this unloads the spa_t without actually removing it
4496 * from the namespace in any way.
4497 */
4498int
4499spa_reset(char *pool)
4500{
4501 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4502 B_FALSE, B_FALSE));
4503}
4504
4505/*
4506 * ==========================================================================
4507 * Device manipulation
4508 * ==========================================================================
4509 */
4510
4511/*
4512 * Add a device to a storage pool.
4513 */
4514int
4515spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4516{
4517 uint64_t txg, id;
4518 int error;
4519 vdev_t *rvd = spa->spa_root_vdev;
4520 vdev_t *vd, *tvd;
4521 nvlist_t **spares, **l2cache;
4522 uint_t nspares, nl2cache;
4523
4524 ASSERT(spa_writeable(spa));
4525
4526 txg = spa_vdev_enter(spa);
4527
4528 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4529 VDEV_ALLOC_ADD)) != 0)
4530 return (spa_vdev_exit(spa, NULL, txg, error));
4531
4532 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4533
4534 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4535 &nspares) != 0)
4536 nspares = 0;
4537
4538 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4539 &nl2cache) != 0)
4540 nl2cache = 0;
4541
4542 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4543 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4544
4545 if (vd->vdev_children != 0 &&
4546 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4547 return (spa_vdev_exit(spa, vd, txg, error));
4548
4549 /*
4550 * We must validate the spares and l2cache devices after checking the
4551 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4552 */
4553 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4554 return (spa_vdev_exit(spa, vd, txg, error));
4555
4556 /*
4557 * Transfer each new top-level vdev from vd to rvd.
4558 */
4559 for (int c = 0; c < vd->vdev_children; c++) {
4560
4561 /*
4562 * Set the vdev id to the first hole, if one exists.
4563 */
4564 for (id = 0; id < rvd->vdev_children; id++) {
4565 if (rvd->vdev_child[id]->vdev_ishole) {
4566 vdev_free(rvd->vdev_child[id]);
4567 break;
4568 }
4569 }
4570 tvd = vd->vdev_child[c];
4571 vdev_remove_child(vd, tvd);
4572 tvd->vdev_id = id;
4573 vdev_add_child(rvd, tvd);
4574 vdev_config_dirty(tvd);
4575 }
4576
4577 if (nspares != 0) {
4578 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4579 ZPOOL_CONFIG_SPARES);
4580 spa_load_spares(spa);
4581 spa->spa_spares.sav_sync = B_TRUE;
4582 }
4583
4584 if (nl2cache != 0) {
4585 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4586 ZPOOL_CONFIG_L2CACHE);
4587 spa_load_l2cache(spa);
4588 spa->spa_l2cache.sav_sync = B_TRUE;
4589 }
4590
4591 /*
4592 * We have to be careful when adding new vdevs to an existing pool.
4593 * If other threads start allocating from these vdevs before we
4594 * sync the config cache, and we lose power, then upon reboot we may
4595 * fail to open the pool because there are DVAs that the config cache
4596 * can't translate. Therefore, we first add the vdevs without
4597 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4598 * and then let spa_config_update() initialize the new metaslabs.
4599 *
4600 * spa_load() checks for added-but-not-initialized vdevs, so that
4601 * if we lose power at any point in this sequence, the remaining
4602 * steps will be completed the next time we load the pool.
4603 */
4604 (void) spa_vdev_exit(spa, vd, txg, 0);
4605
4606 mutex_enter(&spa_namespace_lock);
4607 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4608 mutex_exit(&spa_namespace_lock);
4609
4610 return (0);
4611}
4612
4613/*
4614 * Attach a device to a mirror. The arguments are the path to any device
4615 * in the mirror, and the nvroot for the new device. If the path specifies
4616 * a device that is not mirrored, we automatically insert the mirror vdev.
4617 *
4618 * If 'replacing' is specified, the new device is intended to replace the
4619 * existing device; in this case the two devices are made into their own
4620 * mirror using the 'replacing' vdev, which is functionally identical to
4621 * the mirror vdev (it actually reuses all the same ops) but has a few
4622 * extra rules: you can't attach to it after it's been created, and upon
4623 * completion of resilvering, the first disk (the one being replaced)
4624 * is automatically detached.
4625 */
4626int
4627spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4628{
4629 uint64_t txg, dtl_max_txg;
4630 vdev_t *rvd = spa->spa_root_vdev;
4631 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4632 vdev_ops_t *pvops;
4633 char *oldvdpath, *newvdpath;
4634 int newvd_isspare;
4635 int error;
4636
4637 ASSERT(spa_writeable(spa));
4638
4639 txg = spa_vdev_enter(spa);
4640
4641 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4642
4643 if (oldvd == NULL)
4644 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4645
4646 if (!oldvd->vdev_ops->vdev_op_leaf)
4647 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4648
4649 pvd = oldvd->vdev_parent;
4650
4651 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4652 VDEV_ALLOC_ATTACH)) != 0)
4653 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4654
4655 if (newrootvd->vdev_children != 1)
4656 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4657
4658 newvd = newrootvd->vdev_child[0];
4659
4660 if (!newvd->vdev_ops->vdev_op_leaf)
4661 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4662
4663 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4664 return (spa_vdev_exit(spa, newrootvd, txg, error));
4665
4666 /*
4667 * Spares can't replace logs
4668 */
4669 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4670 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4671
4672 if (!replacing) {
4673 /*
4674 * For attach, the only allowable parent is a mirror or the root
4675 * vdev.
4676 */
4677 if (pvd->vdev_ops != &vdev_mirror_ops &&
4678 pvd->vdev_ops != &vdev_root_ops)
4679 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4680
4681 pvops = &vdev_mirror_ops;
4682 } else {
4683 /*
4684 * Active hot spares can only be replaced by inactive hot
4685 * spares.
4686 */
4687 if (pvd->vdev_ops == &vdev_spare_ops &&
4688 oldvd->vdev_isspare &&
4689 !spa_has_spare(spa, newvd->vdev_guid))
4690 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4691
4692 /*
4693 * If the source is a hot spare, and the parent isn't already a
4694 * spare, then we want to create a new hot spare. Otherwise, we
4695 * want to create a replacing vdev. The user is not allowed to
4696 * attach to a spared vdev child unless the 'isspare' state is
4697 * the same (spare replaces spare, non-spare replaces
4698 * non-spare).
4699 */
4700 if (pvd->vdev_ops == &vdev_replacing_ops &&
4701 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4702 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4703 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4704 newvd->vdev_isspare != oldvd->vdev_isspare) {
4705 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4706 }
4707
4708 if (newvd->vdev_isspare)
4709 pvops = &vdev_spare_ops;
4710 else
4711 pvops = &vdev_replacing_ops;
4712 }
4713
4714 /*
4715 * Make sure the new device is big enough.
4716 */
4717 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4718 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4719
4720 /*
4721 * The new device cannot have a higher alignment requirement
4722 * than the top-level vdev.
4723 */
4724 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4725 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4726
4727 /*
4728 * If this is an in-place replacement, update oldvd's path and devid
4729 * to make it distinguishable from newvd, and unopenable from now on.
4730 */
4731 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4732 spa_strfree(oldvd->vdev_path);
4733 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4734 KM_SLEEP);
4735 (void) sprintf(oldvd->vdev_path, "%s/%s",
4736 newvd->vdev_path, "old");
4737 if (oldvd->vdev_devid != NULL) {
4738 spa_strfree(oldvd->vdev_devid);
4739 oldvd->vdev_devid = NULL;
4740 }
4741 }
4742
4743 /* mark the device being resilvered */
4744 newvd->vdev_resilver_txg = txg;
4745
4746 /*
4747 * If the parent is not a mirror, or if we're replacing, insert the new
4748 * mirror/replacing/spare vdev above oldvd.
4749 */
4750 if (pvd->vdev_ops != pvops)
4751 pvd = vdev_add_parent(oldvd, pvops);
4752
4753 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4754 ASSERT(pvd->vdev_ops == pvops);
4755 ASSERT(oldvd->vdev_parent == pvd);
4756
4757 /*
4758 * Extract the new device from its root and add it to pvd.
4759 */
4760 vdev_remove_child(newrootvd, newvd);
4761 newvd->vdev_id = pvd->vdev_children;
4762 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4763 vdev_add_child(pvd, newvd);
4764
4765 tvd = newvd->vdev_top;
4766 ASSERT(pvd->vdev_top == tvd);
4767 ASSERT(tvd->vdev_parent == rvd);
4768
4769 vdev_config_dirty(tvd);
4770
4771 /*
4772 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4773 * for any dmu_sync-ed blocks. It will propagate upward when
4774 * spa_vdev_exit() calls vdev_dtl_reassess().
4775 */
4776 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4777
4778 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4779 dtl_max_txg - TXG_INITIAL);
4780
4781 if (newvd->vdev_isspare) {
4782 spa_spare_activate(newvd);
4783 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4784 }
4785
4786 oldvdpath = spa_strdup(oldvd->vdev_path);
4787 newvdpath = spa_strdup(newvd->vdev_path);
4788 newvd_isspare = newvd->vdev_isspare;
4789
4790 /*
4791 * Mark newvd's DTL dirty in this txg.
4792 */
4793 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4794
4795 /*
4796 * Schedule the resilver to restart in the future. We do this to
4797 * ensure that dmu_sync-ed blocks have been stitched into the
4798 * respective datasets.
4799 */
4800 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4801
4802 /*
4803 * Commit the config
4804 */
4805 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4806
4807 spa_history_log_internal(spa, "vdev attach", NULL,
4808 "%s vdev=%s %s vdev=%s",
4809 replacing && newvd_isspare ? "spare in" :
4810 replacing ? "replace" : "attach", newvdpath,
4811 replacing ? "for" : "to", oldvdpath);
4812
4813 spa_strfree(oldvdpath);
4814 spa_strfree(newvdpath);
4815
4816 if (spa->spa_bootfs)
4817 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4818
4819 return (0);
4820}
4821
4822/*
4823 * Detach a device from a mirror or replacing vdev.
4824 *
4825 * If 'replace_done' is specified, only detach if the parent
4826 * is a replacing vdev.
4827 */
4828int
4829spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4830{
4831 uint64_t txg;
4832 int error;
4833 vdev_t *rvd = spa->spa_root_vdev;
4834 vdev_t *vd, *pvd, *cvd, *tvd;
4835 boolean_t unspare = B_FALSE;
4836 uint64_t unspare_guid = 0;
4837 char *vdpath;
4838
4839 ASSERT(spa_writeable(spa));
4840
4841 txg = spa_vdev_enter(spa);
4842
4843 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4844
4845 if (vd == NULL)
4846 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4847
4848 if (!vd->vdev_ops->vdev_op_leaf)
4849 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4850
4851 pvd = vd->vdev_parent;
4852
4853 /*
4854 * If the parent/child relationship is not as expected, don't do it.
4855 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4856 * vdev that's replacing B with C. The user's intent in replacing
4857 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4858 * the replace by detaching C, the expected behavior is to end up
4859 * M(A,B). But suppose that right after deciding to detach C,
4860 * the replacement of B completes. We would have M(A,C), and then
4861 * ask to detach C, which would leave us with just A -- not what
4862 * the user wanted. To prevent this, we make sure that the
4863 * parent/child relationship hasn't changed -- in this example,
4864 * that C's parent is still the replacing vdev R.
4865 */
4866 if (pvd->vdev_guid != pguid && pguid != 0)
4867 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4868
4869 /*
4870 * Only 'replacing' or 'spare' vdevs can be replaced.
4871 */
4872 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4873 pvd->vdev_ops != &vdev_spare_ops)
4874 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4875
4876 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4877 spa_version(spa) >= SPA_VERSION_SPARES);
4878
4879 /*
4880 * Only mirror, replacing, and spare vdevs support detach.
4881 */
4882 if (pvd->vdev_ops != &vdev_replacing_ops &&
4883 pvd->vdev_ops != &vdev_mirror_ops &&
4884 pvd->vdev_ops != &vdev_spare_ops)
4885 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4886
4887 /*
4888 * If this device has the only valid copy of some data,
4889 * we cannot safely detach it.
4890 */
4891 if (vdev_dtl_required(vd))
4892 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4893
4894 ASSERT(pvd->vdev_children >= 2);
4895
4896 /*
4897 * If we are detaching the second disk from a replacing vdev, then
4898 * check to see if we changed the original vdev's path to have "/old"
4899 * at the end in spa_vdev_attach(). If so, undo that change now.
4900 */
4901 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4902 vd->vdev_path != NULL) {
4903 size_t len = strlen(vd->vdev_path);
4904
4905 for (int c = 0; c < pvd->vdev_children; c++) {
4906 cvd = pvd->vdev_child[c];
4907
4908 if (cvd == vd || cvd->vdev_path == NULL)
4909 continue;
4910
4911 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4912 strcmp(cvd->vdev_path + len, "/old") == 0) {
4913 spa_strfree(cvd->vdev_path);
4914 cvd->vdev_path = spa_strdup(vd->vdev_path);
4915 break;
4916 }
4917 }
4918 }
4919
4920 /*
4921 * If we are detaching the original disk from a spare, then it implies
4922 * that the spare should become a real disk, and be removed from the
4923 * active spare list for the pool.
4924 */
4925 if (pvd->vdev_ops == &vdev_spare_ops &&
4926 vd->vdev_id == 0 &&
4927 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4928 unspare = B_TRUE;
4929
4930 /*
4931 * Erase the disk labels so the disk can be used for other things.
4932 * This must be done after all other error cases are handled,
4933 * but before we disembowel vd (so we can still do I/O to it).
4934 * But if we can't do it, don't treat the error as fatal --
4935 * it may be that the unwritability of the disk is the reason
4936 * it's being detached!
4937 */
4938 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4939
4940 /*
4941 * Remove vd from its parent and compact the parent's children.
4942 */
4943 vdev_remove_child(pvd, vd);
4944 vdev_compact_children(pvd);
4945
4946 /*
4947 * Remember one of the remaining children so we can get tvd below.
4948 */
4949 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4950
4951 /*
4952 * If we need to remove the remaining child from the list of hot spares,
4953 * do it now, marking the vdev as no longer a spare in the process.
4954 * We must do this before vdev_remove_parent(), because that can
4955 * change the GUID if it creates a new toplevel GUID. For a similar
4956 * reason, we must remove the spare now, in the same txg as the detach;
4957 * otherwise someone could attach a new sibling, change the GUID, and
4958 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4959 */
4960 if (unspare) {
4961 ASSERT(cvd->vdev_isspare);
4962 spa_spare_remove(cvd);
4963 unspare_guid = cvd->vdev_guid;
4964 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4965 cvd->vdev_unspare = B_TRUE;
4966 }
4967
4968 /*
4969 * If the parent mirror/replacing vdev only has one child,
4970 * the parent is no longer needed. Remove it from the tree.
4971 */
4972 if (pvd->vdev_children == 1) {
4973 if (pvd->vdev_ops == &vdev_spare_ops)
4974 cvd->vdev_unspare = B_FALSE;
4975 vdev_remove_parent(cvd);
4976 }
4977
4978
4979 /*
4980 * We don't set tvd until now because the parent we just removed
4981 * may have been the previous top-level vdev.
4982 */
4983 tvd = cvd->vdev_top;
4984 ASSERT(tvd->vdev_parent == rvd);
4985
4986 /*
4987 * Reevaluate the parent vdev state.
4988 */
4989 vdev_propagate_state(cvd);
4990
4991 /*
4992 * If the 'autoexpand' property is set on the pool then automatically
4993 * try to expand the size of the pool. For example if the device we
4994 * just detached was smaller than the others, it may be possible to
4995 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4996 * first so that we can obtain the updated sizes of the leaf vdevs.
4997 */
4998 if (spa->spa_autoexpand) {
4999 vdev_reopen(tvd);
5000 vdev_expand(tvd, txg);
5001 }
5002
5003 vdev_config_dirty(tvd);
5004
5005 /*
5006 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5007 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5008 * But first make sure we're not on any *other* txg's DTL list, to
5009 * prevent vd from being accessed after it's freed.
5010 */
5011 vdpath = spa_strdup(vd->vdev_path);
5012 for (int t = 0; t < TXG_SIZE; t++)
5013 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5014 vd->vdev_detached = B_TRUE;
5015 vdev_dirty(tvd, VDD_DTL, vd, txg);
5016
5017 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5018
5019 /* hang on to the spa before we release the lock */
5020 spa_open_ref(spa, FTAG);
5021
5022 error = spa_vdev_exit(spa, vd, txg, 0);
5023
5024 spa_history_log_internal(spa, "detach", NULL,
5025 "vdev=%s", vdpath);
5026 spa_strfree(vdpath);
5027
5028 /*
5029 * If this was the removal of the original device in a hot spare vdev,
5030 * then we want to go through and remove the device from the hot spare
5031 * list of every other pool.
5032 */
5033 if (unspare) {
5034 spa_t *altspa = NULL;
5035
5036 mutex_enter(&spa_namespace_lock);
5037 while ((altspa = spa_next(altspa)) != NULL) {
5038 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5039 altspa == spa)
5040 continue;
5041
5042 spa_open_ref(altspa, FTAG);
5043 mutex_exit(&spa_namespace_lock);
5044 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5045 mutex_enter(&spa_namespace_lock);
5046 spa_close(altspa, FTAG);
5047 }
5048 mutex_exit(&spa_namespace_lock);
5049
5050 /* search the rest of the vdevs for spares to remove */
5051 spa_vdev_resilver_done(spa);
5052 }
5053
5054 /* all done with the spa; OK to release */
5055 mutex_enter(&spa_namespace_lock);
5056 spa_close(spa, FTAG);
5057 mutex_exit(&spa_namespace_lock);
5058
5059 return (error);
5060}
5061
5062/*
5063 * Split a set of devices from their mirrors, and create a new pool from them.
5064 */
5065int
5066spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5067 nvlist_t *props, boolean_t exp)
5068{
5069 int error = 0;
5070 uint64_t txg, *glist;
5071 spa_t *newspa;
5072 uint_t c, children, lastlog;
5073 nvlist_t **child, *nvl, *tmp;
5074 dmu_tx_t *tx;
5075 char *altroot = NULL;
5076 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5077 boolean_t activate_slog;
5078
5079 ASSERT(spa_writeable(spa));
5080
5081 txg = spa_vdev_enter(spa);
5082
5083 /* clear the log and flush everything up to now */
5084 activate_slog = spa_passivate_log(spa);
5085 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5086 error = spa_offline_log(spa);
5087 txg = spa_vdev_config_enter(spa);
5088
5089 if (activate_slog)
5090 spa_activate_log(spa);
5091
5092 if (error != 0)
5093 return (spa_vdev_exit(spa, NULL, txg, error));
5094
5095 /* check new spa name before going any further */
5096 if (spa_lookup(newname) != NULL)
5097 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5098
5099 /*
5100 * scan through all the children to ensure they're all mirrors
5101 */
5102 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5103 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5104 &children) != 0)
5105 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5106
5107 /* first, check to ensure we've got the right child count */
5108 rvd = spa->spa_root_vdev;
5109 lastlog = 0;
5110 for (c = 0; c < rvd->vdev_children; c++) {
5111 vdev_t *vd = rvd->vdev_child[c];
5112
5113 /* don't count the holes & logs as children */
5114 if (vd->vdev_islog || vd->vdev_ishole) {
5115 if (lastlog == 0)
5116 lastlog = c;
5117 continue;
5118 }
5119
5120 lastlog = 0;
5121 }
5122 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5123 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5124
5125 /* next, ensure no spare or cache devices are part of the split */
5126 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5127 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5128 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5129
5130 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5131 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5132
5133 /* then, loop over each vdev and validate it */
5134 for (c = 0; c < children; c++) {
5135 uint64_t is_hole = 0;
5136
5137 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5138 &is_hole);
5139
5140 if (is_hole != 0) {
5141 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5142 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5143 continue;
5144 } else {
5145 error = SET_ERROR(EINVAL);
5146 break;
5147 }
5148 }
5149
5150 /* which disk is going to be split? */
5151 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5152 &glist[c]) != 0) {
5153 error = SET_ERROR(EINVAL);
5154 break;
5155 }
5156
5157 /* look it up in the spa */
5158 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5159 if (vml[c] == NULL) {
5160 error = SET_ERROR(ENODEV);
5161 break;
5162 }
5163
5164 /* make sure there's nothing stopping the split */
5165 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5166 vml[c]->vdev_islog ||
5167 vml[c]->vdev_ishole ||
5168 vml[c]->vdev_isspare ||
5169 vml[c]->vdev_isl2cache ||
5170 !vdev_writeable(vml[c]) ||
5171 vml[c]->vdev_children != 0 ||
5172 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5173 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5174 error = SET_ERROR(EINVAL);
5175 break;
5176 }
5177
5178 if (vdev_dtl_required(vml[c])) {
5179 error = SET_ERROR(EBUSY);
5180 break;
5181 }
5182
5183 /* we need certain info from the top level */
5184 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5185 vml[c]->vdev_top->vdev_ms_array) == 0);
5186 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5187 vml[c]->vdev_top->vdev_ms_shift) == 0);
5188 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5189 vml[c]->vdev_top->vdev_asize) == 0);
5190 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5191 vml[c]->vdev_top->vdev_ashift) == 0);
5192 }
5193
5194 if (error != 0) {
5195 kmem_free(vml, children * sizeof (vdev_t *));
5196 kmem_free(glist, children * sizeof (uint64_t));
5197 return (spa_vdev_exit(spa, NULL, txg, error));
5198 }
5199
5200 /* stop writers from using the disks */
5201 for (c = 0; c < children; c++) {
5202 if (vml[c] != NULL)
5203 vml[c]->vdev_offline = B_TRUE;
5204 }
5205 vdev_reopen(spa->spa_root_vdev);
5206
5207 /*
5208 * Temporarily record the splitting vdevs in the spa config. This
5209 * will disappear once the config is regenerated.
5210 */
5211 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5212 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5213 glist, children) == 0);
5214 kmem_free(glist, children * sizeof (uint64_t));
5215
5216 mutex_enter(&spa->spa_props_lock);
5217 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5218 nvl) == 0);
5219 mutex_exit(&spa->spa_props_lock);
5220 spa->spa_config_splitting = nvl;
5221 vdev_config_dirty(spa->spa_root_vdev);
5222
5223 /* configure and create the new pool */
5224 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5225 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5226 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5227 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5228 spa_version(spa)) == 0);
5229 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5230 spa->spa_config_txg) == 0);
5231 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5232 spa_generate_guid(NULL)) == 0);
5233 (void) nvlist_lookup_string(props,
5234 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5235
5236 /* add the new pool to the namespace */
5237 newspa = spa_add(newname, config, altroot);
5238 newspa->spa_config_txg = spa->spa_config_txg;
5239 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5240
5241 /* release the spa config lock, retaining the namespace lock */
5242 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5243
5244 if (zio_injection_enabled)
5245 zio_handle_panic_injection(spa, FTAG, 1);
5246
5247 spa_activate(newspa, spa_mode_global);
5248 spa_async_suspend(newspa);
5249
5250#ifndef sun
5251 /* mark that we are creating new spa by splitting */
5252 newspa->spa_splitting_newspa = B_TRUE;
5253#endif
5254 /* create the new pool from the disks of the original pool */
5255 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5256#ifndef sun
5257 newspa->spa_splitting_newspa = B_FALSE;
5258#endif
5259 if (error)
5260 goto out;
5261
5262 /* if that worked, generate a real config for the new pool */
5263 if (newspa->spa_root_vdev != NULL) {
5264 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5265 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5266 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5267 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5268 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5269 B_TRUE));
5270 }
5271
5272 /* set the props */
5273 if (props != NULL) {
5274 spa_configfile_set(newspa, props, B_FALSE);
5275 error = spa_prop_set(newspa, props);
5276 if (error)
5277 goto out;
5278 }
5279
5280 /* flush everything */
5281 txg = spa_vdev_config_enter(newspa);
5282 vdev_config_dirty(newspa->spa_root_vdev);
5283 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5284
5285 if (zio_injection_enabled)
5286 zio_handle_panic_injection(spa, FTAG, 2);
5287
5288 spa_async_resume(newspa);
5289
5290 /* finally, update the original pool's config */
5291 txg = spa_vdev_config_enter(spa);
5292 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5293 error = dmu_tx_assign(tx, TXG_WAIT);
5294 if (error != 0)
5295 dmu_tx_abort(tx);
5296 for (c = 0; c < children; c++) {
5297 if (vml[c] != NULL) {
5298 vdev_split(vml[c]);
5299 if (error == 0)
5300 spa_history_log_internal(spa, "detach", tx,
5301 "vdev=%s", vml[c]->vdev_path);
5302 vdev_free(vml[c]);
5303 }
5304 }
5305 vdev_config_dirty(spa->spa_root_vdev);
5306 spa->spa_config_splitting = NULL;
5307 nvlist_free(nvl);
5308 if (error == 0)
5309 dmu_tx_commit(tx);
5310 (void) spa_vdev_exit(spa, NULL, txg, 0);
5311
5312 if (zio_injection_enabled)
5313 zio_handle_panic_injection(spa, FTAG, 3);
5314
5315 /* split is complete; log a history record */
5316 spa_history_log_internal(newspa, "split", NULL,
5317 "from pool %s", spa_name(spa));
5318
5319 kmem_free(vml, children * sizeof (vdev_t *));
5320
5321 /* if we're not going to mount the filesystems in userland, export */
5322 if (exp)
5323 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5324 B_FALSE, B_FALSE);
5325
5326 return (error);
5327
5328out:
5329 spa_unload(newspa);
5330 spa_deactivate(newspa);
5331 spa_remove(newspa);
5332
5333 txg = spa_vdev_config_enter(spa);
5334
5335 /* re-online all offlined disks */
5336 for (c = 0; c < children; c++) {
5337 if (vml[c] != NULL)
5338 vml[c]->vdev_offline = B_FALSE;
5339 }
5340 vdev_reopen(spa->spa_root_vdev);
5341
5342 nvlist_free(spa->spa_config_splitting);
5343 spa->spa_config_splitting = NULL;
5344 (void) spa_vdev_exit(spa, NULL, txg, error);
5345
5346 kmem_free(vml, children * sizeof (vdev_t *));
5347 return (error);
5348}
5349
5350static nvlist_t *
5351spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5352{
5353 for (int i = 0; i < count; i++) {
5354 uint64_t guid;
5355
5356 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5357 &guid) == 0);
5358
5359 if (guid == target_guid)
5360 return (nvpp[i]);
5361 }
5362
5363 return (NULL);
5364}
5365
5366static void
5367spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5368 nvlist_t *dev_to_remove)
5369{
5370 nvlist_t **newdev = NULL;
5371
5372 if (count > 1)
5373 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5374
5375 for (int i = 0, j = 0; i < count; i++) {
5376 if (dev[i] == dev_to_remove)
5377 continue;
5378 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5379 }
5380
5381 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5382 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5383
5384 for (int i = 0; i < count - 1; i++)
5385 nvlist_free(newdev[i]);
5386
5387 if (count > 1)
5388 kmem_free(newdev, (count - 1) * sizeof (void *));
5389}
5390
5391/*
5392 * Evacuate the device.
5393 */
5394static int
5395spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5396{
5397 uint64_t txg;
5398 int error = 0;
5399
5400 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5401 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5402 ASSERT(vd == vd->vdev_top);
5403
5404 /*
5405 * Evacuate the device. We don't hold the config lock as writer
5406 * since we need to do I/O but we do keep the
5407 * spa_namespace_lock held. Once this completes the device
5408 * should no longer have any blocks allocated on it.
5409 */
5410 if (vd->vdev_islog) {
5411 if (vd->vdev_stat.vs_alloc != 0)
5412 error = spa_offline_log(spa);
5413 } else {
5414 error = SET_ERROR(ENOTSUP);
5415 }
5416
5417 if (error)
5418 return (error);
5419
5420 /*
5421 * The evacuation succeeded. Remove any remaining MOS metadata
5422 * associated with this vdev, and wait for these changes to sync.
5423 */
5424 ASSERT0(vd->vdev_stat.vs_alloc);
5425 txg = spa_vdev_config_enter(spa);
5426 vd->vdev_removing = B_TRUE;
5427 vdev_dirty_leaves(vd, VDD_DTL, txg);
5428 vdev_config_dirty(vd);
5429 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5430
5431 return (0);
5432}
5433
5434/*
5435 * Complete the removal by cleaning up the namespace.
5436 */
5437static void
5438spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5439{
5440 vdev_t *rvd = spa->spa_root_vdev;
5441 uint64_t id = vd->vdev_id;
5442 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5443
5444 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5445 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5446 ASSERT(vd == vd->vdev_top);
5447
5448 /*
5449 * Only remove any devices which are empty.
5450 */
5451 if (vd->vdev_stat.vs_alloc != 0)
5452 return;
5453
5454 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5455
5456 if (list_link_active(&vd->vdev_state_dirty_node))
5457 vdev_state_clean(vd);
5458 if (list_link_active(&vd->vdev_config_dirty_node))
5459 vdev_config_clean(vd);
5460
5461 vdev_free(vd);
5462
5463 if (last_vdev) {
5464 vdev_compact_children(rvd);
5465 } else {
5466 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5467 vdev_add_child(rvd, vd);
5468 }
5469 vdev_config_dirty(rvd);
5470
5471 /*
5472 * Reassess the health of our root vdev.
5473 */
5474 vdev_reopen(rvd);
5475}
5476
5477/*
5478 * Remove a device from the pool -
5479 *
5480 * Removing a device from the vdev namespace requires several steps
5481 * and can take a significant amount of time. As a result we use
5482 * the spa_vdev_config_[enter/exit] functions which allow us to
5483 * grab and release the spa_config_lock while still holding the namespace
5484 * lock. During each step the configuration is synced out.
5485 *
5486 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5487 * devices.
5488 */
5489int
5490spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5491{
5492 vdev_t *vd;
5493 metaslab_group_t *mg;
5494 nvlist_t **spares, **l2cache, *nv;
5495 uint64_t txg = 0;
5496 uint_t nspares, nl2cache;
5497 int error = 0;
5498 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5499
5500 ASSERT(spa_writeable(spa));
5501
5502 if (!locked)
5503 txg = spa_vdev_enter(spa);
5504
5505 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5506
5507 if (spa->spa_spares.sav_vdevs != NULL &&
5508 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5509 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5510 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5511 /*
5512 * Only remove the hot spare if it's not currently in use
5513 * in this pool.
5514 */
5515 if (vd == NULL || unspare) {
5516 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5517 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5518 spa_load_spares(spa);
5519 spa->spa_spares.sav_sync = B_TRUE;
5520 } else {
5521 error = SET_ERROR(EBUSY);
5522 }
5523 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5524 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5525 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5526 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5527 /*
5528 * Cache devices can always be removed.
5529 */
5530 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5531 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5532 spa_load_l2cache(spa);
5533 spa->spa_l2cache.sav_sync = B_TRUE;
5534 } else if (vd != NULL && vd->vdev_islog) {
5535 ASSERT(!locked);
5536 ASSERT(vd == vd->vdev_top);
5537
5538 /*
5539 * XXX - Once we have bp-rewrite this should
5540 * become the common case.
5541 */
5542
5543 mg = vd->vdev_mg;
5544
5545 /*
5546 * Stop allocating from this vdev.
5547 */
5548 metaslab_group_passivate(mg);
5549
5550 /*
5551 * Wait for the youngest allocations and frees to sync,
5552 * and then wait for the deferral of those frees to finish.
5553 */
5554 spa_vdev_config_exit(spa, NULL,
5555 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5556
5557 /*
5558 * Attempt to evacuate the vdev.
5559 */
5560 error = spa_vdev_remove_evacuate(spa, vd);
5561
5562 txg = spa_vdev_config_enter(spa);
5563
5564 /*
5565 * If we couldn't evacuate the vdev, unwind.
5566 */
5567 if (error) {
5568 metaslab_group_activate(mg);
5569 return (spa_vdev_exit(spa, NULL, txg, error));
5570 }
5571
5572 /*
5573 * Clean up the vdev namespace.
5574 */
5575 spa_vdev_remove_from_namespace(spa, vd);
5576
5577 } else if (vd != NULL) {
5578 /*
5579 * Normal vdevs cannot be removed (yet).
5580 */
5581 error = SET_ERROR(ENOTSUP);
5582 } else {
5583 /*
5584 * There is no vdev of any kind with the specified guid.
5585 */
5586 error = SET_ERROR(ENOENT);
5587 }
5588
5589 if (!locked)
5590 return (spa_vdev_exit(spa, NULL, txg, error));
5591
5592 return (error);
5593}
5594
5595/*
5596 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5597 * currently spared, so we can detach it.
5598 */
5599static vdev_t *
5600spa_vdev_resilver_done_hunt(vdev_t *vd)
5601{
5602 vdev_t *newvd, *oldvd;
5603
5604 for (int c = 0; c < vd->vdev_children; c++) {
5605 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5606 if (oldvd != NULL)
5607 return (oldvd);
5608 }
5609
5610 /*
5611 * Check for a completed replacement. We always consider the first
5612 * vdev in the list to be the oldest vdev, and the last one to be
5613 * the newest (see spa_vdev_attach() for how that works). In
5614 * the case where the newest vdev is faulted, we will not automatically
5615 * remove it after a resilver completes. This is OK as it will require
5616 * user intervention to determine which disk the admin wishes to keep.
5617 */
5618 if (vd->vdev_ops == &vdev_replacing_ops) {
5619 ASSERT(vd->vdev_children > 1);
5620
5621 newvd = vd->vdev_child[vd->vdev_children - 1];
5622 oldvd = vd->vdev_child[0];
5623
5624 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5625 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5626 !vdev_dtl_required(oldvd))
5627 return (oldvd);
5628 }
5629
5630 /*
5631 * Check for a completed resilver with the 'unspare' flag set.
5632 */
5633 if (vd->vdev_ops == &vdev_spare_ops) {
5634 vdev_t *first = vd->vdev_child[0];
5635 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5636
5637 if (last->vdev_unspare) {
5638 oldvd = first;
5639 newvd = last;
5640 } else if (first->vdev_unspare) {
5641 oldvd = last;
5642 newvd = first;
5643 } else {
5644 oldvd = NULL;
5645 }
5646
5647 if (oldvd != NULL &&
5648 vdev_dtl_empty(newvd, DTL_MISSING) &&
5649 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5650 !vdev_dtl_required(oldvd))
5651 return (oldvd);
5652
5653 /*
5654 * If there are more than two spares attached to a disk,
5655 * and those spares are not required, then we want to
5656 * attempt to free them up now so that they can be used
5657 * by other pools. Once we're back down to a single
5658 * disk+spare, we stop removing them.
5659 */
5660 if (vd->vdev_children > 2) {
5661 newvd = vd->vdev_child[1];
5662
5663 if (newvd->vdev_isspare && last->vdev_isspare &&
5664 vdev_dtl_empty(last, DTL_MISSING) &&
5665 vdev_dtl_empty(last, DTL_OUTAGE) &&
5666 !vdev_dtl_required(newvd))
5667 return (newvd);
5668 }
5669 }
5670
5671 return (NULL);
5672}
5673
5674static void
5675spa_vdev_resilver_done(spa_t *spa)
5676{
5677 vdev_t *vd, *pvd, *ppvd;
5678 uint64_t guid, sguid, pguid, ppguid;
5679
5680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5681
5682 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5683 pvd = vd->vdev_parent;
5684 ppvd = pvd->vdev_parent;
5685 guid = vd->vdev_guid;
5686 pguid = pvd->vdev_guid;
5687 ppguid = ppvd->vdev_guid;
5688 sguid = 0;
5689 /*
5690 * If we have just finished replacing a hot spared device, then
5691 * we need to detach the parent's first child (the original hot
5692 * spare) as well.
5693 */
5694 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5695 ppvd->vdev_children == 2) {
5696 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5697 sguid = ppvd->vdev_child[1]->vdev_guid;
5698 }
5699 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5700
5701 spa_config_exit(spa, SCL_ALL, FTAG);
5702 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5703 return;
5704 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5705 return;
5706 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5707 }
5708
5709 spa_config_exit(spa, SCL_ALL, FTAG);
5710}
5711
5712/*
5713 * Update the stored path or FRU for this vdev.
5714 */
5715int
5716spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5717 boolean_t ispath)
5718{
5719 vdev_t *vd;
5720 boolean_t sync = B_FALSE;
5721
5722 ASSERT(spa_writeable(spa));
5723
5724 spa_vdev_state_enter(spa, SCL_ALL);
5725
5726 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5727 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5728
5729 if (!vd->vdev_ops->vdev_op_leaf)
5730 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5731
5732 if (ispath) {
5733 if (strcmp(value, vd->vdev_path) != 0) {
5734 spa_strfree(vd->vdev_path);
5735 vd->vdev_path = spa_strdup(value);
5736 sync = B_TRUE;
5737 }
5738 } else {
5739 if (vd->vdev_fru == NULL) {
5740 vd->vdev_fru = spa_strdup(value);
5741 sync = B_TRUE;
5742 } else if (strcmp(value, vd->vdev_fru) != 0) {
5743 spa_strfree(vd->vdev_fru);
5744 vd->vdev_fru = spa_strdup(value);
5745 sync = B_TRUE;
5746 }
5747 }
5748
5749 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5750}
5751
5752int
5753spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5754{
5755 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5756}
5757
5758int
5759spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5760{
5761 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5762}
5763
5764/*
5765 * ==========================================================================
5766 * SPA Scanning
5767 * ==========================================================================
5768 */
5769
5770int
5771spa_scan_stop(spa_t *spa)
5772{
5773 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5774 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5775 return (SET_ERROR(EBUSY));
5776 return (dsl_scan_cancel(spa->spa_dsl_pool));
5777}
5778
5779int
5780spa_scan(spa_t *spa, pool_scan_func_t func)
5781{
5782 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5783
5784 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5785 return (SET_ERROR(ENOTSUP));
5786
5787 /*
5788 * If a resilver was requested, but there is no DTL on a
5789 * writeable leaf device, we have nothing to do.
5790 */
5791 if (func == POOL_SCAN_RESILVER &&
5792 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5793 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5794 return (0);
5795 }
5796
5797 return (dsl_scan(spa->spa_dsl_pool, func));
5798}
5799
5800/*
5801 * ==========================================================================
5802 * SPA async task processing
5803 * ==========================================================================
5804 */
5805
5806static void
5807spa_async_remove(spa_t *spa, vdev_t *vd)
5808{
5809 if (vd->vdev_remove_wanted) {
5810 vd->vdev_remove_wanted = B_FALSE;
5811 vd->vdev_delayed_close = B_FALSE;
5812 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5813
5814 /*
5815 * We want to clear the stats, but we don't want to do a full
5816 * vdev_clear() as that will cause us to throw away
5817 * degraded/faulted state as well as attempt to reopen the
5818 * device, all of which is a waste.
5819 */
5820 vd->vdev_stat.vs_read_errors = 0;
5821 vd->vdev_stat.vs_write_errors = 0;
5822 vd->vdev_stat.vs_checksum_errors = 0;
5823
5824 vdev_state_dirty(vd->vdev_top);
5825 }
5826
5827 for (int c = 0; c < vd->vdev_children; c++)
5828 spa_async_remove(spa, vd->vdev_child[c]);
5829}
5830
5831static void
5832spa_async_probe(spa_t *spa, vdev_t *vd)
5833{
5834 if (vd->vdev_probe_wanted) {
5835 vd->vdev_probe_wanted = B_FALSE;
5836 vdev_reopen(vd); /* vdev_open() does the actual probe */
5837 }
5838
5839 for (int c = 0; c < vd->vdev_children; c++)
5840 spa_async_probe(spa, vd->vdev_child[c]);
5841}
5842
5843static void
5844spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5845{
5846 sysevent_id_t eid;
5847 nvlist_t *attr;
5848 char *physpath;
5849
5850 if (!spa->spa_autoexpand)
5851 return;
5852
5853 for (int c = 0; c < vd->vdev_children; c++) {
5854 vdev_t *cvd = vd->vdev_child[c];
5855 spa_async_autoexpand(spa, cvd);
5856 }
5857
5858 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5859 return;
5860
5861 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5862 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5863
5864 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5865 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5866
5867 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5868 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5869
5870 nvlist_free(attr);
5871 kmem_free(physpath, MAXPATHLEN);
5872}
5873
5874static void
5875spa_async_thread(void *arg)
5876{
5877 spa_t *spa = arg;
5878 int tasks;
5879
5880 ASSERT(spa->spa_sync_on);
5881
5882 mutex_enter(&spa->spa_async_lock);
5883 tasks = spa->spa_async_tasks;
5884 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5885 mutex_exit(&spa->spa_async_lock);
5886
5887 /*
5888 * See if the config needs to be updated.
5889 */
5890 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5891 uint64_t old_space, new_space;
5892
5893 mutex_enter(&spa_namespace_lock);
5894 old_space = metaslab_class_get_space(spa_normal_class(spa));
5895 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5896 new_space = metaslab_class_get_space(spa_normal_class(spa));
5897 mutex_exit(&spa_namespace_lock);
5898
5899 /*
5900 * If the pool grew as a result of the config update,
5901 * then log an internal history event.
5902 */
5903 if (new_space != old_space) {
5904 spa_history_log_internal(spa, "vdev online", NULL,
5905 "pool '%s' size: %llu(+%llu)",
5906 spa_name(spa), new_space, new_space - old_space);
5907 }
5908 }
5909
5910 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5911 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5912 spa_async_autoexpand(spa, spa->spa_root_vdev);
5913 spa_config_exit(spa, SCL_CONFIG, FTAG);
5914 }
5915
5916 /*
5917 * See if any devices need to be probed.
5918 */
5919 if (tasks & SPA_ASYNC_PROBE) {
5920 spa_vdev_state_enter(spa, SCL_NONE);
5921 spa_async_probe(spa, spa->spa_root_vdev);
5922 (void) spa_vdev_state_exit(spa, NULL, 0);
5923 }
5924
5925 /*
5926 * If any devices are done replacing, detach them.
5927 */
5928 if (tasks & SPA_ASYNC_RESILVER_DONE)
5929 spa_vdev_resilver_done(spa);
5930
5931 /*
5932 * Kick off a resilver.
5933 */
5934 if (tasks & SPA_ASYNC_RESILVER)
5935 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5936
5937 /*
5938 * Let the world know that we're done.
5939 */
5940 mutex_enter(&spa->spa_async_lock);
5941 spa->spa_async_thread = NULL;
5942 cv_broadcast(&spa->spa_async_cv);
5943 mutex_exit(&spa->spa_async_lock);
5944 thread_exit();
5945}
5946
5947static void
5948spa_async_thread_vd(void *arg)
5949{
5950 spa_t *spa = arg;
5951 int tasks;
5952
5953 ASSERT(spa->spa_sync_on);
5954
5955 mutex_enter(&spa->spa_async_lock);
5956 tasks = spa->spa_async_tasks;
5957retry:
5958 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5959 mutex_exit(&spa->spa_async_lock);
5960
5961 /*
5962 * See if any devices need to be marked REMOVED.
5963 */
5964 if (tasks & SPA_ASYNC_REMOVE) {
5965 spa_vdev_state_enter(spa, SCL_NONE);
5966 spa_async_remove(spa, spa->spa_root_vdev);
5967 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5968 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5969 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5970 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5971 (void) spa_vdev_state_exit(spa, NULL, 0);
5972 }
5973
5974 /*
5975 * Let the world know that we're done.
5976 */
5977 mutex_enter(&spa->spa_async_lock);
5978 tasks = spa->spa_async_tasks;
5979 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5980 goto retry;
5981 spa->spa_async_thread_vd = NULL;
5982 cv_broadcast(&spa->spa_async_cv);
5983 mutex_exit(&spa->spa_async_lock);
5984 thread_exit();
5985}
5986
5987void
5988spa_async_suspend(spa_t *spa)
5989{
5990 mutex_enter(&spa->spa_async_lock);
5991 spa->spa_async_suspended++;
5992 while (spa->spa_async_thread != NULL &&
5993 spa->spa_async_thread_vd != NULL)
5994 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5995 mutex_exit(&spa->spa_async_lock);
5996}
5997
5998void
5999spa_async_resume(spa_t *spa)
6000{
6001 mutex_enter(&spa->spa_async_lock);
6002 ASSERT(spa->spa_async_suspended != 0);
6003 spa->spa_async_suspended--;
6004 mutex_exit(&spa->spa_async_lock);
6005}
6006
6007static boolean_t
6008spa_async_tasks_pending(spa_t *spa)
6009{
6010 uint_t non_config_tasks;
6011 uint_t config_task;
6012 boolean_t config_task_suspended;
6013
6014 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6015 SPA_ASYNC_REMOVE);
6016 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6017 if (spa->spa_ccw_fail_time == 0) {
6018 config_task_suspended = B_FALSE;
6019 } else {
6020 config_task_suspended =
6021 (gethrtime() - spa->spa_ccw_fail_time) <
6022 (zfs_ccw_retry_interval * NANOSEC);
6023 }
6024
6025 return (non_config_tasks || (config_task && !config_task_suspended));
6026}
6027
6028static void
6029spa_async_dispatch(spa_t *spa)
6030{
6031 mutex_enter(&spa->spa_async_lock);
6032 if (spa_async_tasks_pending(spa) &&
6033 !spa->spa_async_suspended &&
6034 spa->spa_async_thread == NULL &&
6035 rootdir != NULL)
6036 spa->spa_async_thread = thread_create(NULL, 0,
6037 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6038 mutex_exit(&spa->spa_async_lock);
6039}
6040
6041static void
6042spa_async_dispatch_vd(spa_t *spa)
6043{
6044 mutex_enter(&spa->spa_async_lock);
6045 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6046 !spa->spa_async_suspended &&
6047 spa->spa_async_thread_vd == NULL &&
6048 rootdir != NULL)
6049 spa->spa_async_thread_vd = thread_create(NULL, 0,
6050 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6051 mutex_exit(&spa->spa_async_lock);
6052}
6053
6054void
6055spa_async_request(spa_t *spa, int task)
6056{
6057 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6058 mutex_enter(&spa->spa_async_lock);
6059 spa->spa_async_tasks |= task;
6060 mutex_exit(&spa->spa_async_lock);
6061 spa_async_dispatch_vd(spa);
6062}
6063
6064/*
6065 * ==========================================================================
6066 * SPA syncing routines
6067 * ==========================================================================
6068 */
6069
6070static int
6071bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6072{
6073 bpobj_t *bpo = arg;
6074 bpobj_enqueue(bpo, bp, tx);
6075 return (0);
6076}
6077
6078static int
6079spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6080{
6081 zio_t *zio = arg;
6082
6083 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6084 BP_GET_PSIZE(bp), zio->io_flags));
6085 return (0);
6086}
6087
6088/*
6089 * Note: this simple function is not inlined to make it easier to dtrace the
6090 * amount of time spent syncing frees.
6091 */
6092static void
6093spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6094{
6095 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6096 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6097 VERIFY(zio_wait(zio) == 0);
6098}
6099
6100/*
6101 * Note: this simple function is not inlined to make it easier to dtrace the
6102 * amount of time spent syncing deferred frees.
6103 */
6104static void
6105spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6106{
6107 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6108 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6109 spa_free_sync_cb, zio, tx), ==, 0);
6110 VERIFY0(zio_wait(zio));
6111}
6112
6113
6114static void
6115spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6116{
6117 char *packed = NULL;
6118 size_t bufsize;
6119 size_t nvsize = 0;
6120 dmu_buf_t *db;
6121
6122 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6123
6124 /*
6125 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6126 * information. This avoids the dmu_buf_will_dirty() path and
6127 * saves us a pre-read to get data we don't actually care about.
6128 */
6129 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6130 packed = kmem_alloc(bufsize, KM_SLEEP);
6131
6132 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6133 KM_SLEEP) == 0);
6134 bzero(packed + nvsize, bufsize - nvsize);
6135
6136 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6137
6138 kmem_free(packed, bufsize);
6139
6140 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6141 dmu_buf_will_dirty(db, tx);
6142 *(uint64_t *)db->db_data = nvsize;
6143 dmu_buf_rele(db, FTAG);
6144}
6145
6146static void
6147spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6148 const char *config, const char *entry)
6149{
6150 nvlist_t *nvroot;
6151 nvlist_t **list;
6152 int i;
6153
6154 if (!sav->sav_sync)
6155 return;
6156
6157 /*
6158 * Update the MOS nvlist describing the list of available devices.
6159 * spa_validate_aux() will have already made sure this nvlist is
6160 * valid and the vdevs are labeled appropriately.
6161 */
6162 if (sav->sav_object == 0) {
6163 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6164 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6165 sizeof (uint64_t), tx);
6166 VERIFY(zap_update(spa->spa_meta_objset,
6167 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6168 &sav->sav_object, tx) == 0);
6169 }
6170
6171 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6172 if (sav->sav_count == 0) {
6173 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6174 } else {
6175 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6176 for (i = 0; i < sav->sav_count; i++)
6177 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6178 B_FALSE, VDEV_CONFIG_L2CACHE);
6179 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6180 sav->sav_count) == 0);
6181 for (i = 0; i < sav->sav_count; i++)
6182 nvlist_free(list[i]);
6183 kmem_free(list, sav->sav_count * sizeof (void *));
6184 }
6185
6186 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6187 nvlist_free(nvroot);
6188
6189 sav->sav_sync = B_FALSE;
6190}
6191
6192static void
6193spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6194{
6195 nvlist_t *config;
6196
6197 if (list_is_empty(&spa->spa_config_dirty_list))
6198 return;
6199
6200 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6201
6202 config = spa_config_generate(spa, spa->spa_root_vdev,
6203 dmu_tx_get_txg(tx), B_FALSE);
6204
6205 /*
6206 * If we're upgrading the spa version then make sure that
6207 * the config object gets updated with the correct version.
6208 */
6209 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6210 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6211 spa->spa_uberblock.ub_version);
6212
6213 spa_config_exit(spa, SCL_STATE, FTAG);
6214
6215 if (spa->spa_config_syncing)
6216 nvlist_free(spa->spa_config_syncing);
6217 spa->spa_config_syncing = config;
6218
6219 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6220}
6221
6222static void
6223spa_sync_version(void *arg, dmu_tx_t *tx)
6224{
6225 uint64_t *versionp = arg;
6226 uint64_t version = *versionp;
6227 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6228
6229 /*
6230 * Setting the version is special cased when first creating the pool.
6231 */
6232 ASSERT(tx->tx_txg != TXG_INITIAL);
6233
6234 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6235 ASSERT(version >= spa_version(spa));
6236
6237 spa->spa_uberblock.ub_version = version;
6238 vdev_config_dirty(spa->spa_root_vdev);
6239 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6240}
6241
6242/*
6243 * Set zpool properties.
6244 */
6245static void
6246spa_sync_props(void *arg, dmu_tx_t *tx)
6247{
6248 nvlist_t *nvp = arg;
6249 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6250 objset_t *mos = spa->spa_meta_objset;
6251 nvpair_t *elem = NULL;
6252
6253 mutex_enter(&spa->spa_props_lock);
6254
6255 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6256 uint64_t intval;
6257 char *strval, *fname;
6258 zpool_prop_t prop;
6259 const char *propname;
6260 zprop_type_t proptype;
6261 spa_feature_t fid;
6262
6263 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6264 case ZPROP_INVAL:
6265 /*
6266 * We checked this earlier in spa_prop_validate().
6267 */
6268 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6269
6270 fname = strchr(nvpair_name(elem), '@') + 1;
6271 VERIFY0(zfeature_lookup_name(fname, &fid));
6272
6273 spa_feature_enable(spa, fid, tx);
6274 spa_history_log_internal(spa, "set", tx,
6275 "%s=enabled", nvpair_name(elem));
6276 break;
6277
6278 case ZPOOL_PROP_VERSION:
6279 intval = fnvpair_value_uint64(elem);
6280 /*
6281 * The version is synced seperatly before other
6282 * properties and should be correct by now.
6283 */
6284 ASSERT3U(spa_version(spa), >=, intval);
6285 break;
6286
6287 case ZPOOL_PROP_ALTROOT:
6288 /*
6289 * 'altroot' is a non-persistent property. It should
6290 * have been set temporarily at creation or import time.
6291 */
6292 ASSERT(spa->spa_root != NULL);
6293 break;
6294
6295 case ZPOOL_PROP_READONLY:
6296 case ZPOOL_PROP_CACHEFILE:
6297 /*
6298 * 'readonly' and 'cachefile' are also non-persisitent
6299 * properties.
6300 */
6301 break;
6302 case ZPOOL_PROP_COMMENT:
6303 strval = fnvpair_value_string(elem);
6304 if (spa->spa_comment != NULL)
6305 spa_strfree(spa->spa_comment);
6306 spa->spa_comment = spa_strdup(strval);
6307 /*
6308 * We need to dirty the configuration on all the vdevs
6309 * so that their labels get updated. It's unnecessary
6310 * to do this for pool creation since the vdev's
6311 * configuratoin has already been dirtied.
6312 */
6313 if (tx->tx_txg != TXG_INITIAL)
6314 vdev_config_dirty(spa->spa_root_vdev);
6315 spa_history_log_internal(spa, "set", tx,
6316 "%s=%s", nvpair_name(elem), strval);
6317 break;
6318 default:
6319 /*
6320 * Set pool property values in the poolprops mos object.
6321 */
6322 if (spa->spa_pool_props_object == 0) {
6323 spa->spa_pool_props_object =
6324 zap_create_link(mos, DMU_OT_POOL_PROPS,
6325 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6326 tx);
6327 }
6328
6329 /* normalize the property name */
6330 propname = zpool_prop_to_name(prop);
6331 proptype = zpool_prop_get_type(prop);
6332
6333 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6334 ASSERT(proptype == PROP_TYPE_STRING);
6335 strval = fnvpair_value_string(elem);
6336 VERIFY0(zap_update(mos,
6337 spa->spa_pool_props_object, propname,
6338 1, strlen(strval) + 1, strval, tx));
6339 spa_history_log_internal(spa, "set", tx,
6340 "%s=%s", nvpair_name(elem), strval);
6341 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6342 intval = fnvpair_value_uint64(elem);
6343
6344 if (proptype == PROP_TYPE_INDEX) {
6345 const char *unused;
6346 VERIFY0(zpool_prop_index_to_string(
6347 prop, intval, &unused));
6348 }
6349 VERIFY0(zap_update(mos,
6350 spa->spa_pool_props_object, propname,
6351 8, 1, &intval, tx));
6352 spa_history_log_internal(spa, "set", tx,
6353 "%s=%lld", nvpair_name(elem), intval);
6354 } else {
6355 ASSERT(0); /* not allowed */
6356 }
6357
6358 switch (prop) {
6359 case ZPOOL_PROP_DELEGATION:
6360 spa->spa_delegation = intval;
6361 break;
6362 case ZPOOL_PROP_BOOTFS:
6363 spa->spa_bootfs = intval;
6364 break;
6365 case ZPOOL_PROP_FAILUREMODE:
6366 spa->spa_failmode = intval;
6367 break;
6368 case ZPOOL_PROP_AUTOEXPAND:
6369 spa->spa_autoexpand = intval;
6370 if (tx->tx_txg != TXG_INITIAL)
6371 spa_async_request(spa,
6372 SPA_ASYNC_AUTOEXPAND);
6373 break;
6374 case ZPOOL_PROP_DEDUPDITTO:
6375 spa->spa_dedup_ditto = intval;
6376 break;
6377 default:
6378 break;
6379 }
6380 }
6381
6382 }
6383
6384 mutex_exit(&spa->spa_props_lock);
6385}
6386
6387/*
6388 * Perform one-time upgrade on-disk changes. spa_version() does not
6389 * reflect the new version this txg, so there must be no changes this
6390 * txg to anything that the upgrade code depends on after it executes.
6391 * Therefore this must be called after dsl_pool_sync() does the sync
6392 * tasks.
6393 */
6394static void
6395spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6396{
6397 dsl_pool_t *dp = spa->spa_dsl_pool;
6398
6399 ASSERT(spa->spa_sync_pass == 1);
6400
6401 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6402
6403 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6404 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6405 dsl_pool_create_origin(dp, tx);
6406
6407 /* Keeping the origin open increases spa_minref */
6408 spa->spa_minref += 3;
6409 }
6410
6411 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6412 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6413 dsl_pool_upgrade_clones(dp, tx);
6414 }
6415
6416 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6417 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6418 dsl_pool_upgrade_dir_clones(dp, tx);
6419
6420 /* Keeping the freedir open increases spa_minref */
6421 spa->spa_minref += 3;
6422 }
6423
6424 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6425 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6426 spa_feature_create_zap_objects(spa, tx);
6427 }
6428 rrw_exit(&dp->dp_config_rwlock, FTAG);
6429}
6430
6431/*
6432 * Sync the specified transaction group. New blocks may be dirtied as
6433 * part of the process, so we iterate until it converges.
6434 */
6435void
6436spa_sync(spa_t *spa, uint64_t txg)
6437{
6438 dsl_pool_t *dp = spa->spa_dsl_pool;
6439 objset_t *mos = spa->spa_meta_objset;
6440 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6441 vdev_t *rvd = spa->spa_root_vdev;
6442 vdev_t *vd;
6443 dmu_tx_t *tx;
6444 int error;
6445
6446 VERIFY(spa_writeable(spa));
6447
6448 /*
6449 * Lock out configuration changes.
6450 */
6451 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6452
6453 spa->spa_syncing_txg = txg;
6454 spa->spa_sync_pass = 0;
6455
6456 /*
6457 * If there are any pending vdev state changes, convert them
6458 * into config changes that go out with this transaction group.
6459 */
6460 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6461 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6462 /*
6463 * We need the write lock here because, for aux vdevs,
6464 * calling vdev_config_dirty() modifies sav_config.
6465 * This is ugly and will become unnecessary when we
6466 * eliminate the aux vdev wart by integrating all vdevs
6467 * into the root vdev tree.
6468 */
6469 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6470 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6471 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6472 vdev_state_clean(vd);
6473 vdev_config_dirty(vd);
6474 }
6475 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6476 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6477 }
6478 spa_config_exit(spa, SCL_STATE, FTAG);
6479
6480 tx = dmu_tx_create_assigned(dp, txg);
6481
6482 spa->spa_sync_starttime = gethrtime();
6483#ifdef illumos
6484 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6485 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6486#else /* FreeBSD */
6487#ifdef _KERNEL
6488 callout_reset(&spa->spa_deadman_cycid,
6489 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6490#endif
6491#endif
6492
6493 /*
6494 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6495 * set spa_deflate if we have no raid-z vdevs.
6496 */
6497 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6498 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6499 int i;
6500
6501 for (i = 0; i < rvd->vdev_children; i++) {
6502 vd = rvd->vdev_child[i];
6503 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6504 break;
6505 }
6506 if (i == rvd->vdev_children) {
6507 spa->spa_deflate = TRUE;
6508 VERIFY(0 == zap_add(spa->spa_meta_objset,
6509 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6510 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6511 }
6512 }
6513
6514 /*
6515 * If anything has changed in this txg, or if someone is waiting
6516 * for this txg to sync (eg, spa_vdev_remove()), push the
6517 * deferred frees from the previous txg. If not, leave them
6518 * alone so that we don't generate work on an otherwise idle
6519 * system.
6520 */
6521 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6522 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6523 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6524 ((dsl_scan_active(dp->dp_scan) ||
6525 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6526 spa_sync_deferred_frees(spa, tx);
6527 }
6528
6529 /*
6530 * Iterate to convergence.
6531 */
6532 do {
6533 int pass = ++spa->spa_sync_pass;
6534
6535 spa_sync_config_object(spa, tx);
6536 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6537 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6538 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6539 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6540 spa_errlog_sync(spa, txg);
6541 dsl_pool_sync(dp, txg);
6542
6543 if (pass < zfs_sync_pass_deferred_free) {
6544 spa_sync_frees(spa, free_bpl, tx);
6545 } else {
6546 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6547 &spa->spa_deferred_bpobj, tx);
6548 }
6549
6550 ddt_sync(spa, txg);
6551 dsl_scan_sync(dp, tx);
6552
6553 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6554 vdev_sync(vd, txg);
6555
6556 if (pass == 1)
6557 spa_sync_upgrades(spa, tx);
6558
6559 } while (dmu_objset_is_dirty(mos, txg));
6560
6561 /*
6562 * Rewrite the vdev configuration (which includes the uberblock)
6563 * to commit the transaction group.
6564 *
6565 * If there are no dirty vdevs, we sync the uberblock to a few
6566 * random top-level vdevs that are known to be visible in the
6567 * config cache (see spa_vdev_add() for a complete description).
6568 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6569 */
6570 for (;;) {
6571 /*
6572 * We hold SCL_STATE to prevent vdev open/close/etc.
6573 * while we're attempting to write the vdev labels.
6574 */
6575 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6576
6577 if (list_is_empty(&spa->spa_config_dirty_list)) {
6578 vdev_t *svd[SPA_DVAS_PER_BP];
6579 int svdcount = 0;
6580 int children = rvd->vdev_children;
6581 int c0 = spa_get_random(children);
6582
6583 for (int c = 0; c < children; c++) {
6584 vd = rvd->vdev_child[(c0 + c) % children];
6585 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6586 continue;
6587 svd[svdcount++] = vd;
6588 if (svdcount == SPA_DVAS_PER_BP)
6589 break;
6590 }
6591 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6592 if (error != 0)
6593 error = vdev_config_sync(svd, svdcount, txg,
6594 B_TRUE);
6595 } else {
6596 error = vdev_config_sync(rvd->vdev_child,
6597 rvd->vdev_children, txg, B_FALSE);
6598 if (error != 0)
6599 error = vdev_config_sync(rvd->vdev_child,
6600 rvd->vdev_children, txg, B_TRUE);
6601 }
6602
6603 if (error == 0)
6604 spa->spa_last_synced_guid = rvd->vdev_guid;
6605
6606 spa_config_exit(spa, SCL_STATE, FTAG);
6607
6608 if (error == 0)
6609 break;
6610 zio_suspend(spa, NULL);
6611 zio_resume_wait(spa);
6612 }
6613 dmu_tx_commit(tx);
6614
6615#ifdef illumos
6616 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6617#else /* FreeBSD */
6618#ifdef _KERNEL
6619 callout_drain(&spa->spa_deadman_cycid);
6620#endif
6621#endif
6622
6623 /*
6624 * Clear the dirty config list.
6625 */
6626 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6627 vdev_config_clean(vd);
6628
6629 /*
6630 * Now that the new config has synced transactionally,
6631 * let it become visible to the config cache.
6632 */
6633 if (spa->spa_config_syncing != NULL) {
6634 spa_config_set(spa, spa->spa_config_syncing);
6635 spa->spa_config_txg = txg;
6636 spa->spa_config_syncing = NULL;
6637 }
6638
6639 spa->spa_ubsync = spa->spa_uberblock;
6640
6641 dsl_pool_sync_done(dp, txg);
6642
6643 /*
6644 * Update usable space statistics.
6645 */
6646 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6647 vdev_sync_done(vd, txg);
6648
6649 spa_update_dspace(spa);
6650
6651 /*
6652 * It had better be the case that we didn't dirty anything
6653 * since vdev_config_sync().
6654 */
6655 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6656 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6657 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6658
6659 spa->spa_sync_pass = 0;
6660
6661 spa_config_exit(spa, SCL_CONFIG, FTAG);
6662
6663 spa_handle_ignored_writes(spa);
6664
6665 /*
6666 * If any async tasks have been requested, kick them off.
6667 */
6668 spa_async_dispatch(spa);
6669 spa_async_dispatch_vd(spa);
6670}
6671
6672/*
6673 * Sync all pools. We don't want to hold the namespace lock across these
6674 * operations, so we take a reference on the spa_t and drop the lock during the
6675 * sync.
6676 */
6677void
6678spa_sync_allpools(void)
6679{
6680 spa_t *spa = NULL;
6681 mutex_enter(&spa_namespace_lock);
6682 while ((spa = spa_next(spa)) != NULL) {
6683 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6684 !spa_writeable(spa) || spa_suspended(spa))
6685 continue;
6686 spa_open_ref(spa, FTAG);
6687 mutex_exit(&spa_namespace_lock);
6688 txg_wait_synced(spa_get_dsl(spa), 0);
6689 mutex_enter(&spa_namespace_lock);
6690 spa_close(spa, FTAG);
6691 }
6692 mutex_exit(&spa_namespace_lock);
6693}
6694
6695/*
6696 * ==========================================================================
6697 * Miscellaneous routines
6698 * ==========================================================================
6699 */
6700
6701/*
6702 * Remove all pools in the system.
6703 */
6704void
6705spa_evict_all(void)
6706{
6707 spa_t *spa;
6708
6709 /*
6710 * Remove all cached state. All pools should be closed now,
6711 * so every spa in the AVL tree should be unreferenced.
6712 */
6713 mutex_enter(&spa_namespace_lock);
6714 while ((spa = spa_next(NULL)) != NULL) {
6715 /*
6716 * Stop async tasks. The async thread may need to detach
6717 * a device that's been replaced, which requires grabbing
6718 * spa_namespace_lock, so we must drop it here.
6719 */
6720 spa_open_ref(spa, FTAG);
6721 mutex_exit(&spa_namespace_lock);
6722 spa_async_suspend(spa);
6723 mutex_enter(&spa_namespace_lock);
6724 spa_close(spa, FTAG);
6725
6726 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6727 spa_unload(spa);
6728 spa_deactivate(spa);
6729 }
6730 spa_remove(spa);
6731 }
6732 mutex_exit(&spa_namespace_lock);
6733}
6734
6735vdev_t *
6736spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6737{
6738 vdev_t *vd;
6739 int i;
6740
6741 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6742 return (vd);
6743
6744 if (aux) {
6745 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6746 vd = spa->spa_l2cache.sav_vdevs[i];
6747 if (vd->vdev_guid == guid)
6748 return (vd);
6749 }
6750
6751 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6752 vd = spa->spa_spares.sav_vdevs[i];
6753 if (vd->vdev_guid == guid)
6754 return (vd);
6755 }
6756 }
6757
6758 return (NULL);
6759}
6760
6761void
6762spa_upgrade(spa_t *spa, uint64_t version)
6763{
6764 ASSERT(spa_writeable(spa));
6765
6766 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6767
6768 /*
6769 * This should only be called for a non-faulted pool, and since a
6770 * future version would result in an unopenable pool, this shouldn't be
6771 * possible.
6772 */
6773 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6774 ASSERT(version >= spa->spa_uberblock.ub_version);
6775
6776 spa->spa_uberblock.ub_version = version;
6777 vdev_config_dirty(spa->spa_root_vdev);
6778
6779 spa_config_exit(spa, SCL_ALL, FTAG);
6780
6781 txg_wait_synced(spa_get_dsl(spa), 0);
6782}
6783
6784boolean_t
6785spa_has_spare(spa_t *spa, uint64_t guid)
6786{
6787 int i;
6788 uint64_t spareguid;
6789 spa_aux_vdev_t *sav = &spa->spa_spares;
6790
6791 for (i = 0; i < sav->sav_count; i++)
6792 if (sav->sav_vdevs[i]->vdev_guid == guid)
6793 return (B_TRUE);
6794
6795 for (i = 0; i < sav->sav_npending; i++) {
6796 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6797 &spareguid) == 0 && spareguid == guid)
6798 return (B_TRUE);
6799 }
6800
6801 return (B_FALSE);
6802}
6803
6804/*
6805 * Check if a pool has an active shared spare device.
6806 * Note: reference count of an active spare is 2, as a spare and as a replace
6807 */
6808static boolean_t
6809spa_has_active_shared_spare(spa_t *spa)
6810{
6811 int i, refcnt;
6812 uint64_t pool;
6813 spa_aux_vdev_t *sav = &spa->spa_spares;
6814
6815 for (i = 0; i < sav->sav_count; i++) {
6816 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6817 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6818 refcnt > 2)
6819 return (B_TRUE);
6820 }
6821
6822 return (B_FALSE);
6823}
6824
6825/*
6826 * Post a sysevent corresponding to the given event. The 'name' must be one of
6827 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6828 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6829 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6830 * or zdb as real changes.
6831 */
6832void
6833spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6834{
6835#ifdef _KERNEL
6836 sysevent_t *ev;
6837 sysevent_attr_list_t *attr = NULL;
6838 sysevent_value_t value;
6839 sysevent_id_t eid;
6840
6841 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6842 SE_SLEEP);
6843
6844 value.value_type = SE_DATA_TYPE_STRING;
6845 value.value.sv_string = spa_name(spa);
6846 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6847 goto done;
6848
6849 value.value_type = SE_DATA_TYPE_UINT64;
6850 value.value.sv_uint64 = spa_guid(spa);
6851 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6852 goto done;
6853
6854 if (vd) {
6855 value.value_type = SE_DATA_TYPE_UINT64;
6856 value.value.sv_uint64 = vd->vdev_guid;
6857 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6858 SE_SLEEP) != 0)
6859 goto done;
6860
6861 if (vd->vdev_path) {
6862 value.value_type = SE_DATA_TYPE_STRING;
6863 value.value.sv_string = vd->vdev_path;
6864 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6865 &value, SE_SLEEP) != 0)
6866 goto done;
6867 }
6868 }
6869
6870 if (sysevent_attach_attributes(ev, attr) != 0)
6871 goto done;
6872 attr = NULL;
6873
6874 (void) log_sysevent(ev, SE_SLEEP, &eid);
6875
6876done:
6877 if (attr)
6878 sysevent_free_attr(attr);
6879 sysevent_free(ev);
6880#endif
6881}
| 142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
146};
147
148static void spa_sync_version(void *arg, dmu_tx_t *tx);
149static void spa_sync_props(void *arg, dmu_tx_t *tx);
150static boolean_t spa_has_active_shared_spare(spa_t *spa);
151static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
153 char **ereport);
154static void spa_vdev_resilver_done(spa_t *spa);
155
156uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
157#ifdef PSRSET_BIND
158id_t zio_taskq_psrset_bind = PS_NONE;
159#endif
160#ifdef SYSDC
161boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
162#endif
163uint_t zio_taskq_basedc = 80; /* base duty cycle */
164
165boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
166extern int zfs_sync_pass_deferred_free;
167
168#ifndef illumos
169extern void spa_deadman(void *arg);
170#endif
171
172/*
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
175 */
176#define TRYIMPORT_NAME "$import"
177
178/*
179 * ==========================================================================
180 * SPA properties routines
181 * ==========================================================================
182 */
183
184/*
185 * Add a (source=src, propname=propval) list to an nvlist.
186 */
187static void
188spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
189 uint64_t intval, zprop_source_t src)
190{
191 const char *propname = zpool_prop_to_name(prop);
192 nvlist_t *propval;
193
194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
196
197 if (strval != NULL)
198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
199 else
200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
201
202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
203 nvlist_free(propval);
204}
205
206/*
207 * Get property values from the spa configuration.
208 */
209static void
210spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
211{
212 vdev_t *rvd = spa->spa_root_vdev;
213 dsl_pool_t *pool = spa->spa_dsl_pool;
214 uint64_t size;
215 uint64_t alloc;
216 uint64_t space;
217 uint64_t cap, version;
218 zprop_source_t src = ZPROP_SRC_NONE;
219 spa_config_dirent_t *dp;
220
221 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
222
223 if (rvd != NULL) {
224 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
225 size = metaslab_class_get_space(spa_normal_class(spa));
226 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
227 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
228 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
229 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
230 size - alloc, src);
231
232 space = 0;
233 for (int c = 0; c < rvd->vdev_children; c++) {
234 vdev_t *tvd = rvd->vdev_child[c];
235 space += tvd->vdev_max_asize - tvd->vdev_asize;
236 }
237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
238 src);
239
240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
241 (spa_mode(spa) == FREAD), src);
242
243 cap = (size == 0) ? 0 : (alloc * 100 / size);
244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
245
246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
247 ddt_get_pool_dedup_ratio(spa), src);
248
249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
250 rvd->vdev_state, src);
251
252 version = spa_version(spa);
253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
254 src = ZPROP_SRC_DEFAULT;
255 else
256 src = ZPROP_SRC_LOCAL;
257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
258 }
259
260 if (pool != NULL) {
261 dsl_dir_t *freedir = pool->dp_free_dir;
262
263 /*
264 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
265 * when opening pools before this version freedir will be NULL.
266 */
267 if (freedir != NULL) {
268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
269 freedir->dd_phys->dd_used_bytes, src);
270 } else {
271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
272 NULL, 0, src);
273 }
274 }
275
276 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
277
278 if (spa->spa_comment != NULL) {
279 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
280 0, ZPROP_SRC_LOCAL);
281 }
282
283 if (spa->spa_root != NULL)
284 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
285 0, ZPROP_SRC_LOCAL);
286
287 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
288 if (dp->scd_path == NULL) {
289 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
290 "none", 0, ZPROP_SRC_LOCAL);
291 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
292 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
293 dp->scd_path, 0, ZPROP_SRC_LOCAL);
294 }
295 }
296}
297
298/*
299 * Get zpool property values.
300 */
301int
302spa_prop_get(spa_t *spa, nvlist_t **nvp)
303{
304 objset_t *mos = spa->spa_meta_objset;
305 zap_cursor_t zc;
306 zap_attribute_t za;
307 int err;
308
309 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
310
311 mutex_enter(&spa->spa_props_lock);
312
313 /*
314 * Get properties from the spa config.
315 */
316 spa_prop_get_config(spa, nvp);
317
318 /* If no pool property object, no more prop to get. */
319 if (mos == NULL || spa->spa_pool_props_object == 0) {
320 mutex_exit(&spa->spa_props_lock);
321 return (0);
322 }
323
324 /*
325 * Get properties from the MOS pool property object.
326 */
327 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
328 (err = zap_cursor_retrieve(&zc, &za)) == 0;
329 zap_cursor_advance(&zc)) {
330 uint64_t intval = 0;
331 char *strval = NULL;
332 zprop_source_t src = ZPROP_SRC_DEFAULT;
333 zpool_prop_t prop;
334
335 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
336 continue;
337
338 switch (za.za_integer_length) {
339 case 8:
340 /* integer property */
341 if (za.za_first_integer !=
342 zpool_prop_default_numeric(prop))
343 src = ZPROP_SRC_LOCAL;
344
345 if (prop == ZPOOL_PROP_BOOTFS) {
346 dsl_pool_t *dp;
347 dsl_dataset_t *ds = NULL;
348
349 dp = spa_get_dsl(spa);
350 dsl_pool_config_enter(dp, FTAG);
351 if (err = dsl_dataset_hold_obj(dp,
352 za.za_first_integer, FTAG, &ds)) {
353 dsl_pool_config_exit(dp, FTAG);
354 break;
355 }
356
357 strval = kmem_alloc(
358 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
359 KM_SLEEP);
360 dsl_dataset_name(ds, strval);
361 dsl_dataset_rele(ds, FTAG);
362 dsl_pool_config_exit(dp, FTAG);
363 } else {
364 strval = NULL;
365 intval = za.za_first_integer;
366 }
367
368 spa_prop_add_list(*nvp, prop, strval, intval, src);
369
370 if (strval != NULL)
371 kmem_free(strval,
372 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
373
374 break;
375
376 case 1:
377 /* string property */
378 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
379 err = zap_lookup(mos, spa->spa_pool_props_object,
380 za.za_name, 1, za.za_num_integers, strval);
381 if (err) {
382 kmem_free(strval, za.za_num_integers);
383 break;
384 }
385 spa_prop_add_list(*nvp, prop, strval, 0, src);
386 kmem_free(strval, za.za_num_integers);
387 break;
388
389 default:
390 break;
391 }
392 }
393 zap_cursor_fini(&zc);
394 mutex_exit(&spa->spa_props_lock);
395out:
396 if (err && err != ENOENT) {
397 nvlist_free(*nvp);
398 *nvp = NULL;
399 return (err);
400 }
401
402 return (0);
403}
404
405/*
406 * Validate the given pool properties nvlist and modify the list
407 * for the property values to be set.
408 */
409static int
410spa_prop_validate(spa_t *spa, nvlist_t *props)
411{
412 nvpair_t *elem;
413 int error = 0, reset_bootfs = 0;
414 uint64_t objnum = 0;
415 boolean_t has_feature = B_FALSE;
416
417 elem = NULL;
418 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
419 uint64_t intval;
420 char *strval, *slash, *check, *fname;
421 const char *propname = nvpair_name(elem);
422 zpool_prop_t prop = zpool_name_to_prop(propname);
423
424 switch (prop) {
425 case ZPROP_INVAL:
426 if (!zpool_prop_feature(propname)) {
427 error = SET_ERROR(EINVAL);
428 break;
429 }
430
431 /*
432 * Sanitize the input.
433 */
434 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
435 error = SET_ERROR(EINVAL);
436 break;
437 }
438
439 if (nvpair_value_uint64(elem, &intval) != 0) {
440 error = SET_ERROR(EINVAL);
441 break;
442 }
443
444 if (intval != 0) {
445 error = SET_ERROR(EINVAL);
446 break;
447 }
448
449 fname = strchr(propname, '@') + 1;
450 if (zfeature_lookup_name(fname, NULL) != 0) {
451 error = SET_ERROR(EINVAL);
452 break;
453 }
454
455 has_feature = B_TRUE;
456 break;
457
458 case ZPOOL_PROP_VERSION:
459 error = nvpair_value_uint64(elem, &intval);
460 if (!error &&
461 (intval < spa_version(spa) ||
462 intval > SPA_VERSION_BEFORE_FEATURES ||
463 has_feature))
464 error = SET_ERROR(EINVAL);
465 break;
466
467 case ZPOOL_PROP_DELEGATION:
468 case ZPOOL_PROP_AUTOREPLACE:
469 case ZPOOL_PROP_LISTSNAPS:
470 case ZPOOL_PROP_AUTOEXPAND:
471 error = nvpair_value_uint64(elem, &intval);
472 if (!error && intval > 1)
473 error = SET_ERROR(EINVAL);
474 break;
475
476 case ZPOOL_PROP_BOOTFS:
477 /*
478 * If the pool version is less than SPA_VERSION_BOOTFS,
479 * or the pool is still being created (version == 0),
480 * the bootfs property cannot be set.
481 */
482 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
483 error = SET_ERROR(ENOTSUP);
484 break;
485 }
486
487 /*
488 * Make sure the vdev config is bootable
489 */
490 if (!vdev_is_bootable(spa->spa_root_vdev)) {
491 error = SET_ERROR(ENOTSUP);
492 break;
493 }
494
495 reset_bootfs = 1;
496
497 error = nvpair_value_string(elem, &strval);
498
499 if (!error) {
500 objset_t *os;
501 uint64_t compress;
502
503 if (strval == NULL || strval[0] == '\0') {
504 objnum = zpool_prop_default_numeric(
505 ZPOOL_PROP_BOOTFS);
506 break;
507 }
508
509 if (error = dmu_objset_hold(strval, FTAG, &os))
510 break;
511
512 /* Must be ZPL and not gzip compressed. */
513
514 if (dmu_objset_type(os) != DMU_OST_ZFS) {
515 error = SET_ERROR(ENOTSUP);
516 } else if ((error =
517 dsl_prop_get_int_ds(dmu_objset_ds(os),
518 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
519 &compress)) == 0 &&
520 !BOOTFS_COMPRESS_VALID(compress)) {
521 error = SET_ERROR(ENOTSUP);
522 } else {
523 objnum = dmu_objset_id(os);
524 }
525 dmu_objset_rele(os, FTAG);
526 }
527 break;
528
529 case ZPOOL_PROP_FAILUREMODE:
530 error = nvpair_value_uint64(elem, &intval);
531 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
532 intval > ZIO_FAILURE_MODE_PANIC))
533 error = SET_ERROR(EINVAL);
534
535 /*
536 * This is a special case which only occurs when
537 * the pool has completely failed. This allows
538 * the user to change the in-core failmode property
539 * without syncing it out to disk (I/Os might
540 * currently be blocked). We do this by returning
541 * EIO to the caller (spa_prop_set) to trick it
542 * into thinking we encountered a property validation
543 * error.
544 */
545 if (!error && spa_suspended(spa)) {
546 spa->spa_failmode = intval;
547 error = SET_ERROR(EIO);
548 }
549 break;
550
551 case ZPOOL_PROP_CACHEFILE:
552 if ((error = nvpair_value_string(elem, &strval)) != 0)
553 break;
554
555 if (strval[0] == '\0')
556 break;
557
558 if (strcmp(strval, "none") == 0)
559 break;
560
561 if (strval[0] != '/') {
562 error = SET_ERROR(EINVAL);
563 break;
564 }
565
566 slash = strrchr(strval, '/');
567 ASSERT(slash != NULL);
568
569 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
570 strcmp(slash, "/..") == 0)
571 error = SET_ERROR(EINVAL);
572 break;
573
574 case ZPOOL_PROP_COMMENT:
575 if ((error = nvpair_value_string(elem, &strval)) != 0)
576 break;
577 for (check = strval; *check != '\0'; check++) {
578 /*
579 * The kernel doesn't have an easy isprint()
580 * check. For this kernel check, we merely
581 * check ASCII apart from DEL. Fix this if
582 * there is an easy-to-use kernel isprint().
583 */
584 if (*check >= 0x7f) {
585 error = SET_ERROR(EINVAL);
586 break;
587 }
588 check++;
589 }
590 if (strlen(strval) > ZPROP_MAX_COMMENT)
591 error = E2BIG;
592 break;
593
594 case ZPOOL_PROP_DEDUPDITTO:
595 if (spa_version(spa) < SPA_VERSION_DEDUP)
596 error = SET_ERROR(ENOTSUP);
597 else
598 error = nvpair_value_uint64(elem, &intval);
599 if (error == 0 &&
600 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
601 error = SET_ERROR(EINVAL);
602 break;
603 }
604
605 if (error)
606 break;
607 }
608
609 if (!error && reset_bootfs) {
610 error = nvlist_remove(props,
611 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
612
613 if (!error) {
614 error = nvlist_add_uint64(props,
615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
616 }
617 }
618
619 return (error);
620}
621
622void
623spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
624{
625 char *cachefile;
626 spa_config_dirent_t *dp;
627
628 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
629 &cachefile) != 0)
630 return;
631
632 dp = kmem_alloc(sizeof (spa_config_dirent_t),
633 KM_SLEEP);
634
635 if (cachefile[0] == '\0')
636 dp->scd_path = spa_strdup(spa_config_path);
637 else if (strcmp(cachefile, "none") == 0)
638 dp->scd_path = NULL;
639 else
640 dp->scd_path = spa_strdup(cachefile);
641
642 list_insert_head(&spa->spa_config_list, dp);
643 if (need_sync)
644 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
645}
646
647int
648spa_prop_set(spa_t *spa, nvlist_t *nvp)
649{
650 int error;
651 nvpair_t *elem = NULL;
652 boolean_t need_sync = B_FALSE;
653
654 if ((error = spa_prop_validate(spa, nvp)) != 0)
655 return (error);
656
657 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
658 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
659
660 if (prop == ZPOOL_PROP_CACHEFILE ||
661 prop == ZPOOL_PROP_ALTROOT ||
662 prop == ZPOOL_PROP_READONLY)
663 continue;
664
665 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
666 uint64_t ver;
667
668 if (prop == ZPOOL_PROP_VERSION) {
669 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
670 } else {
671 ASSERT(zpool_prop_feature(nvpair_name(elem)));
672 ver = SPA_VERSION_FEATURES;
673 need_sync = B_TRUE;
674 }
675
676 /* Save time if the version is already set. */
677 if (ver == spa_version(spa))
678 continue;
679
680 /*
681 * In addition to the pool directory object, we might
682 * create the pool properties object, the features for
683 * read object, the features for write object, or the
684 * feature descriptions object.
685 */
686 error = dsl_sync_task(spa->spa_name, NULL,
687 spa_sync_version, &ver, 6);
688 if (error)
689 return (error);
690 continue;
691 }
692
693 need_sync = B_TRUE;
694 break;
695 }
696
697 if (need_sync) {
698 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
699 nvp, 6));
700 }
701
702 return (0);
703}
704
705/*
706 * If the bootfs property value is dsobj, clear it.
707 */
708void
709spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
710{
711 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
712 VERIFY(zap_remove(spa->spa_meta_objset,
713 spa->spa_pool_props_object,
714 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
715 spa->spa_bootfs = 0;
716 }
717}
718
719/*ARGSUSED*/
720static int
721spa_change_guid_check(void *arg, dmu_tx_t *tx)
722{
723 uint64_t *newguid = arg;
724 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
725 vdev_t *rvd = spa->spa_root_vdev;
726 uint64_t vdev_state;
727
728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
729 vdev_state = rvd->vdev_state;
730 spa_config_exit(spa, SCL_STATE, FTAG);
731
732 if (vdev_state != VDEV_STATE_HEALTHY)
733 return (SET_ERROR(ENXIO));
734
735 ASSERT3U(spa_guid(spa), !=, *newguid);
736
737 return (0);
738}
739
740static void
741spa_change_guid_sync(void *arg, dmu_tx_t *tx)
742{
743 uint64_t *newguid = arg;
744 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
745 uint64_t oldguid;
746 vdev_t *rvd = spa->spa_root_vdev;
747
748 oldguid = spa_guid(spa);
749
750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
751 rvd->vdev_guid = *newguid;
752 rvd->vdev_guid_sum += (*newguid - oldguid);
753 vdev_config_dirty(rvd);
754 spa_config_exit(spa, SCL_STATE, FTAG);
755
756 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
757 oldguid, *newguid);
758}
759
760/*
761 * Change the GUID for the pool. This is done so that we can later
762 * re-import a pool built from a clone of our own vdevs. We will modify
763 * the root vdev's guid, our own pool guid, and then mark all of our
764 * vdevs dirty. Note that we must make sure that all our vdevs are
765 * online when we do this, or else any vdevs that weren't present
766 * would be orphaned from our pool. We are also going to issue a
767 * sysevent to update any watchers.
768 */
769int
770spa_change_guid(spa_t *spa)
771{
772 int error;
773 uint64_t guid;
774
775 mutex_enter(&spa->spa_vdev_top_lock);
776 mutex_enter(&spa_namespace_lock);
777 guid = spa_generate_guid(NULL);
778
779 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
780 spa_change_guid_sync, &guid, 5);
781
782 if (error == 0) {
783 spa_config_sync(spa, B_FALSE, B_TRUE);
784 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
785 }
786
787 mutex_exit(&spa_namespace_lock);
788 mutex_exit(&spa->spa_vdev_top_lock);
789
790 return (error);
791}
792
793/*
794 * ==========================================================================
795 * SPA state manipulation (open/create/destroy/import/export)
796 * ==========================================================================
797 */
798
799static int
800spa_error_entry_compare(const void *a, const void *b)
801{
802 spa_error_entry_t *sa = (spa_error_entry_t *)a;
803 spa_error_entry_t *sb = (spa_error_entry_t *)b;
804 int ret;
805
806 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
807 sizeof (zbookmark_t));
808
809 if (ret < 0)
810 return (-1);
811 else if (ret > 0)
812 return (1);
813 else
814 return (0);
815}
816
817/*
818 * Utility function which retrieves copies of the current logs and
819 * re-initializes them in the process.
820 */
821void
822spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
823{
824 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
825
826 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
827 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
828
829 avl_create(&spa->spa_errlist_scrub,
830 spa_error_entry_compare, sizeof (spa_error_entry_t),
831 offsetof(spa_error_entry_t, se_avl));
832 avl_create(&spa->spa_errlist_last,
833 spa_error_entry_compare, sizeof (spa_error_entry_t),
834 offsetof(spa_error_entry_t, se_avl));
835}
836
837static void
838spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
839{
840 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
841 enum zti_modes mode = ztip->zti_mode;
842 uint_t value = ztip->zti_value;
843 uint_t count = ztip->zti_count;
844 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
845 char name[32];
846 uint_t flags = 0;
847 boolean_t batch = B_FALSE;
848
849 if (mode == ZTI_MODE_NULL) {
850 tqs->stqs_count = 0;
851 tqs->stqs_taskq = NULL;
852 return;
853 }
854
855 ASSERT3U(count, >, 0);
856
857 tqs->stqs_count = count;
858 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
859
860 switch (mode) {
861 case ZTI_MODE_FIXED:
862 ASSERT3U(value, >=, 1);
863 value = MAX(value, 1);
864 break;
865
866 case ZTI_MODE_BATCH:
867 batch = B_TRUE;
868 flags |= TASKQ_THREADS_CPU_PCT;
869 value = zio_taskq_batch_pct;
870 break;
871
872 default:
873 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
874 "spa_activate()",
875 zio_type_name[t], zio_taskq_types[q], mode, value);
876 break;
877 }
878
879 for (uint_t i = 0; i < count; i++) {
880 taskq_t *tq;
881
882 if (count > 1) {
883 (void) snprintf(name, sizeof (name), "%s_%s_%u",
884 zio_type_name[t], zio_taskq_types[q], i);
885 } else {
886 (void) snprintf(name, sizeof (name), "%s_%s",
887 zio_type_name[t], zio_taskq_types[q]);
888 }
889
890#ifdef SYSDC
891 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
892 if (batch)
893 flags |= TASKQ_DC_BATCH;
894
895 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
896 spa->spa_proc, zio_taskq_basedc, flags);
897 } else {
898#endif
899 pri_t pri = maxclsyspri;
900 /*
901 * The write issue taskq can be extremely CPU
902 * intensive. Run it at slightly lower priority
903 * than the other taskqs.
904 */
905 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
906 pri--;
907
908 tq = taskq_create_proc(name, value, pri, 50,
909 INT_MAX, spa->spa_proc, flags);
910#ifdef SYSDC
911 }
912#endif
913
914 tqs->stqs_taskq[i] = tq;
915 }
916}
917
918static void
919spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
920{
921 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
922
923 if (tqs->stqs_taskq == NULL) {
924 ASSERT0(tqs->stqs_count);
925 return;
926 }
927
928 for (uint_t i = 0; i < tqs->stqs_count; i++) {
929 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
930 taskq_destroy(tqs->stqs_taskq[i]);
931 }
932
933 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
934 tqs->stqs_taskq = NULL;
935}
936
937/*
938 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
939 * Note that a type may have multiple discrete taskqs to avoid lock contention
940 * on the taskq itself. In that case we choose which taskq at random by using
941 * the low bits of gethrtime().
942 */
943void
944spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
945 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
946{
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
948 taskq_t *tq;
949
950 ASSERT3P(tqs->stqs_taskq, !=, NULL);
951 ASSERT3U(tqs->stqs_count, !=, 0);
952
953 if (tqs->stqs_count == 1) {
954 tq = tqs->stqs_taskq[0];
955 } else {
956 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
957 }
958
959 taskq_dispatch_ent(tq, func, arg, flags, ent);
960}
961
962static void
963spa_create_zio_taskqs(spa_t *spa)
964{
965 for (int t = 0; t < ZIO_TYPES; t++) {
966 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
967 spa_taskqs_init(spa, t, q);
968 }
969 }
970}
971
972#ifdef _KERNEL
973#ifdef SPA_PROCESS
974static void
975spa_thread(void *arg)
976{
977 callb_cpr_t cprinfo;
978
979 spa_t *spa = arg;
980 user_t *pu = PTOU(curproc);
981
982 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
983 spa->spa_name);
984
985 ASSERT(curproc != &p0);
986 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
987 "zpool-%s", spa->spa_name);
988 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
989
990#ifdef PSRSET_BIND
991 /* bind this thread to the requested psrset */
992 if (zio_taskq_psrset_bind != PS_NONE) {
993 pool_lock();
994 mutex_enter(&cpu_lock);
995 mutex_enter(&pidlock);
996 mutex_enter(&curproc->p_lock);
997
998 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
999 0, NULL, NULL) == 0) {
1000 curthread->t_bind_pset = zio_taskq_psrset_bind;
1001 } else {
1002 cmn_err(CE_WARN,
1003 "Couldn't bind process for zfs pool \"%s\" to "
1004 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1005 }
1006
1007 mutex_exit(&curproc->p_lock);
1008 mutex_exit(&pidlock);
1009 mutex_exit(&cpu_lock);
1010 pool_unlock();
1011 }
1012#endif
1013
1014#ifdef SYSDC
1015 if (zio_taskq_sysdc) {
1016 sysdc_thread_enter(curthread, 100, 0);
1017 }
1018#endif
1019
1020 spa->spa_proc = curproc;
1021 spa->spa_did = curthread->t_did;
1022
1023 spa_create_zio_taskqs(spa);
1024
1025 mutex_enter(&spa->spa_proc_lock);
1026 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1027
1028 spa->spa_proc_state = SPA_PROC_ACTIVE;
1029 cv_broadcast(&spa->spa_proc_cv);
1030
1031 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1032 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1033 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1034 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1035
1036 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1037 spa->spa_proc_state = SPA_PROC_GONE;
1038 spa->spa_proc = &p0;
1039 cv_broadcast(&spa->spa_proc_cv);
1040 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1041
1042 mutex_enter(&curproc->p_lock);
1043 lwp_exit();
1044}
1045#endif /* SPA_PROCESS */
1046#endif
1047
1048/*
1049 * Activate an uninitialized pool.
1050 */
1051static void
1052spa_activate(spa_t *spa, int mode)
1053{
1054 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1055
1056 spa->spa_state = POOL_STATE_ACTIVE;
1057 spa->spa_mode = mode;
1058
1059 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1060 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1061
1062 /* Try to create a covering process */
1063 mutex_enter(&spa->spa_proc_lock);
1064 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1065 ASSERT(spa->spa_proc == &p0);
1066 spa->spa_did = 0;
1067
1068#ifdef SPA_PROCESS
1069 /* Only create a process if we're going to be around a while. */
1070 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1071 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1072 NULL, 0) == 0) {
1073 spa->spa_proc_state = SPA_PROC_CREATED;
1074 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1075 cv_wait(&spa->spa_proc_cv,
1076 &spa->spa_proc_lock);
1077 }
1078 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1079 ASSERT(spa->spa_proc != &p0);
1080 ASSERT(spa->spa_did != 0);
1081 } else {
1082#ifdef _KERNEL
1083 cmn_err(CE_WARN,
1084 "Couldn't create process for zfs pool \"%s\"\n",
1085 spa->spa_name);
1086#endif
1087 }
1088 }
1089#endif /* SPA_PROCESS */
1090 mutex_exit(&spa->spa_proc_lock);
1091
1092 /* If we didn't create a process, we need to create our taskqs. */
1093 ASSERT(spa->spa_proc == &p0);
1094 if (spa->spa_proc == &p0) {
1095 spa_create_zio_taskqs(spa);
1096 }
1097
1098 /*
1099 * Start TRIM thread.
1100 */
1101 trim_thread_create(spa);
1102
1103 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1104 offsetof(vdev_t, vdev_config_dirty_node));
1105 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1106 offsetof(vdev_t, vdev_state_dirty_node));
1107
1108 txg_list_create(&spa->spa_vdev_txg_list,
1109 offsetof(struct vdev, vdev_txg_node));
1110
1111 avl_create(&spa->spa_errlist_scrub,
1112 spa_error_entry_compare, sizeof (spa_error_entry_t),
1113 offsetof(spa_error_entry_t, se_avl));
1114 avl_create(&spa->spa_errlist_last,
1115 spa_error_entry_compare, sizeof (spa_error_entry_t),
1116 offsetof(spa_error_entry_t, se_avl));
1117}
1118
1119/*
1120 * Opposite of spa_activate().
1121 */
1122static void
1123spa_deactivate(spa_t *spa)
1124{
1125 ASSERT(spa->spa_sync_on == B_FALSE);
1126 ASSERT(spa->spa_dsl_pool == NULL);
1127 ASSERT(spa->spa_root_vdev == NULL);
1128 ASSERT(spa->spa_async_zio_root == NULL);
1129 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1130
1131 /*
1132 * Stop TRIM thread in case spa_unload() wasn't called directly
1133 * before spa_deactivate().
1134 */
1135 trim_thread_destroy(spa);
1136
1137 txg_list_destroy(&spa->spa_vdev_txg_list);
1138
1139 list_destroy(&spa->spa_config_dirty_list);
1140 list_destroy(&spa->spa_state_dirty_list);
1141
1142 for (int t = 0; t < ZIO_TYPES; t++) {
1143 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1144 spa_taskqs_fini(spa, t, q);
1145 }
1146 }
1147
1148 metaslab_class_destroy(spa->spa_normal_class);
1149 spa->spa_normal_class = NULL;
1150
1151 metaslab_class_destroy(spa->spa_log_class);
1152 spa->spa_log_class = NULL;
1153
1154 /*
1155 * If this was part of an import or the open otherwise failed, we may
1156 * still have errors left in the queues. Empty them just in case.
1157 */
1158 spa_errlog_drain(spa);
1159
1160 avl_destroy(&spa->spa_errlist_scrub);
1161 avl_destroy(&spa->spa_errlist_last);
1162
1163 spa->spa_state = POOL_STATE_UNINITIALIZED;
1164
1165 mutex_enter(&spa->spa_proc_lock);
1166 if (spa->spa_proc_state != SPA_PROC_NONE) {
1167 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1168 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1169 cv_broadcast(&spa->spa_proc_cv);
1170 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1171 ASSERT(spa->spa_proc != &p0);
1172 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1173 }
1174 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1175 spa->spa_proc_state = SPA_PROC_NONE;
1176 }
1177 ASSERT(spa->spa_proc == &p0);
1178 mutex_exit(&spa->spa_proc_lock);
1179
1180#ifdef SPA_PROCESS
1181 /*
1182 * We want to make sure spa_thread() has actually exited the ZFS
1183 * module, so that the module can't be unloaded out from underneath
1184 * it.
1185 */
1186 if (spa->spa_did != 0) {
1187 thread_join(spa->spa_did);
1188 spa->spa_did = 0;
1189 }
1190#endif /* SPA_PROCESS */
1191}
1192
1193/*
1194 * Verify a pool configuration, and construct the vdev tree appropriately. This
1195 * will create all the necessary vdevs in the appropriate layout, with each vdev
1196 * in the CLOSED state. This will prep the pool before open/creation/import.
1197 * All vdev validation is done by the vdev_alloc() routine.
1198 */
1199static int
1200spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1201 uint_t id, int atype)
1202{
1203 nvlist_t **child;
1204 uint_t children;
1205 int error;
1206
1207 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1208 return (error);
1209
1210 if ((*vdp)->vdev_ops->vdev_op_leaf)
1211 return (0);
1212
1213 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1214 &child, &children);
1215
1216 if (error == ENOENT)
1217 return (0);
1218
1219 if (error) {
1220 vdev_free(*vdp);
1221 *vdp = NULL;
1222 return (SET_ERROR(EINVAL));
1223 }
1224
1225 for (int c = 0; c < children; c++) {
1226 vdev_t *vd;
1227 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1228 atype)) != 0) {
1229 vdev_free(*vdp);
1230 *vdp = NULL;
1231 return (error);
1232 }
1233 }
1234
1235 ASSERT(*vdp != NULL);
1236
1237 return (0);
1238}
1239
1240/*
1241 * Opposite of spa_load().
1242 */
1243static void
1244spa_unload(spa_t *spa)
1245{
1246 int i;
1247
1248 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1249
1250 /*
1251 * Stop TRIM thread.
1252 */
1253 trim_thread_destroy(spa);
1254
1255 /*
1256 * Stop async tasks.
1257 */
1258 spa_async_suspend(spa);
1259
1260 /*
1261 * Stop syncing.
1262 */
1263 if (spa->spa_sync_on) {
1264 txg_sync_stop(spa->spa_dsl_pool);
1265 spa->spa_sync_on = B_FALSE;
1266 }
1267
1268 /*
1269 * Wait for any outstanding async I/O to complete.
1270 */
1271 if (spa->spa_async_zio_root != NULL) {
1272 (void) zio_wait(spa->spa_async_zio_root);
1273 spa->spa_async_zio_root = NULL;
1274 }
1275
1276 bpobj_close(&spa->spa_deferred_bpobj);
1277
1278 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1279
1280 /*
1281 * Close all vdevs.
1282 */
1283 if (spa->spa_root_vdev)
1284 vdev_free(spa->spa_root_vdev);
1285 ASSERT(spa->spa_root_vdev == NULL);
1286
1287 /*
1288 * Close the dsl pool.
1289 */
1290 if (spa->spa_dsl_pool) {
1291 dsl_pool_close(spa->spa_dsl_pool);
1292 spa->spa_dsl_pool = NULL;
1293 spa->spa_meta_objset = NULL;
1294 }
1295
1296 ddt_unload(spa);
1297
1298
1299 /*
1300 * Drop and purge level 2 cache
1301 */
1302 spa_l2cache_drop(spa);
1303
1304 for (i = 0; i < spa->spa_spares.sav_count; i++)
1305 vdev_free(spa->spa_spares.sav_vdevs[i]);
1306 if (spa->spa_spares.sav_vdevs) {
1307 kmem_free(spa->spa_spares.sav_vdevs,
1308 spa->spa_spares.sav_count * sizeof (void *));
1309 spa->spa_spares.sav_vdevs = NULL;
1310 }
1311 if (spa->spa_spares.sav_config) {
1312 nvlist_free(spa->spa_spares.sav_config);
1313 spa->spa_spares.sav_config = NULL;
1314 }
1315 spa->spa_spares.sav_count = 0;
1316
1317 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1318 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1319 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1320 }
1321 if (spa->spa_l2cache.sav_vdevs) {
1322 kmem_free(spa->spa_l2cache.sav_vdevs,
1323 spa->spa_l2cache.sav_count * sizeof (void *));
1324 spa->spa_l2cache.sav_vdevs = NULL;
1325 }
1326 if (spa->spa_l2cache.sav_config) {
1327 nvlist_free(spa->spa_l2cache.sav_config);
1328 spa->spa_l2cache.sav_config = NULL;
1329 }
1330 spa->spa_l2cache.sav_count = 0;
1331
1332 spa->spa_async_suspended = 0;
1333
1334 if (spa->spa_comment != NULL) {
1335 spa_strfree(spa->spa_comment);
1336 spa->spa_comment = NULL;
1337 }
1338
1339 spa_config_exit(spa, SCL_ALL, FTAG);
1340}
1341
1342/*
1343 * Load (or re-load) the current list of vdevs describing the active spares for
1344 * this pool. When this is called, we have some form of basic information in
1345 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1346 * then re-generate a more complete list including status information.
1347 */
1348static void
1349spa_load_spares(spa_t *spa)
1350{
1351 nvlist_t **spares;
1352 uint_t nspares;
1353 int i;
1354 vdev_t *vd, *tvd;
1355
1356 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1357
1358 /*
1359 * First, close and free any existing spare vdevs.
1360 */
1361 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1362 vd = spa->spa_spares.sav_vdevs[i];
1363
1364 /* Undo the call to spa_activate() below */
1365 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1366 B_FALSE)) != NULL && tvd->vdev_isspare)
1367 spa_spare_remove(tvd);
1368 vdev_close(vd);
1369 vdev_free(vd);
1370 }
1371
1372 if (spa->spa_spares.sav_vdevs)
1373 kmem_free(spa->spa_spares.sav_vdevs,
1374 spa->spa_spares.sav_count * sizeof (void *));
1375
1376 if (spa->spa_spares.sav_config == NULL)
1377 nspares = 0;
1378 else
1379 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1380 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1381
1382 spa->spa_spares.sav_count = (int)nspares;
1383 spa->spa_spares.sav_vdevs = NULL;
1384
1385 if (nspares == 0)
1386 return;
1387
1388 /*
1389 * Construct the array of vdevs, opening them to get status in the
1390 * process. For each spare, there is potentially two different vdev_t
1391 * structures associated with it: one in the list of spares (used only
1392 * for basic validation purposes) and one in the active vdev
1393 * configuration (if it's spared in). During this phase we open and
1394 * validate each vdev on the spare list. If the vdev also exists in the
1395 * active configuration, then we also mark this vdev as an active spare.
1396 */
1397 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1398 KM_SLEEP);
1399 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1400 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1401 VDEV_ALLOC_SPARE) == 0);
1402 ASSERT(vd != NULL);
1403
1404 spa->spa_spares.sav_vdevs[i] = vd;
1405
1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1407 B_FALSE)) != NULL) {
1408 if (!tvd->vdev_isspare)
1409 spa_spare_add(tvd);
1410
1411 /*
1412 * We only mark the spare active if we were successfully
1413 * able to load the vdev. Otherwise, importing a pool
1414 * with a bad active spare would result in strange
1415 * behavior, because multiple pool would think the spare
1416 * is actively in use.
1417 *
1418 * There is a vulnerability here to an equally bizarre
1419 * circumstance, where a dead active spare is later
1420 * brought back to life (onlined or otherwise). Given
1421 * the rarity of this scenario, and the extra complexity
1422 * it adds, we ignore the possibility.
1423 */
1424 if (!vdev_is_dead(tvd))
1425 spa_spare_activate(tvd);
1426 }
1427
1428 vd->vdev_top = vd;
1429 vd->vdev_aux = &spa->spa_spares;
1430
1431 if (vdev_open(vd) != 0)
1432 continue;
1433
1434 if (vdev_validate_aux(vd) == 0)
1435 spa_spare_add(vd);
1436 }
1437
1438 /*
1439 * Recompute the stashed list of spares, with status information
1440 * this time.
1441 */
1442 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1443 DATA_TYPE_NVLIST_ARRAY) == 0);
1444
1445 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1446 KM_SLEEP);
1447 for (i = 0; i < spa->spa_spares.sav_count; i++)
1448 spares[i] = vdev_config_generate(spa,
1449 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1451 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1452 for (i = 0; i < spa->spa_spares.sav_count; i++)
1453 nvlist_free(spares[i]);
1454 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1455}
1456
1457/*
1458 * Load (or re-load) the current list of vdevs describing the active l2cache for
1459 * this pool. When this is called, we have some form of basic information in
1460 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1461 * then re-generate a more complete list including status information.
1462 * Devices which are already active have their details maintained, and are
1463 * not re-opened.
1464 */
1465static void
1466spa_load_l2cache(spa_t *spa)
1467{
1468 nvlist_t **l2cache;
1469 uint_t nl2cache;
1470 int i, j, oldnvdevs;
1471 uint64_t guid;
1472 vdev_t *vd, **oldvdevs, **newvdevs;
1473 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1474
1475 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1476
1477 if (sav->sav_config != NULL) {
1478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1479 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1480 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1481 } else {
1482 nl2cache = 0;
1483 newvdevs = NULL;
1484 }
1485
1486 oldvdevs = sav->sav_vdevs;
1487 oldnvdevs = sav->sav_count;
1488 sav->sav_vdevs = NULL;
1489 sav->sav_count = 0;
1490
1491 /*
1492 * Process new nvlist of vdevs.
1493 */
1494 for (i = 0; i < nl2cache; i++) {
1495 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1496 &guid) == 0);
1497
1498 newvdevs[i] = NULL;
1499 for (j = 0; j < oldnvdevs; j++) {
1500 vd = oldvdevs[j];
1501 if (vd != NULL && guid == vd->vdev_guid) {
1502 /*
1503 * Retain previous vdev for add/remove ops.
1504 */
1505 newvdevs[i] = vd;
1506 oldvdevs[j] = NULL;
1507 break;
1508 }
1509 }
1510
1511 if (newvdevs[i] == NULL) {
1512 /*
1513 * Create new vdev
1514 */
1515 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1516 VDEV_ALLOC_L2CACHE) == 0);
1517 ASSERT(vd != NULL);
1518 newvdevs[i] = vd;
1519
1520 /*
1521 * Commit this vdev as an l2cache device,
1522 * even if it fails to open.
1523 */
1524 spa_l2cache_add(vd);
1525
1526 vd->vdev_top = vd;
1527 vd->vdev_aux = sav;
1528
1529 spa_l2cache_activate(vd);
1530
1531 if (vdev_open(vd) != 0)
1532 continue;
1533
1534 (void) vdev_validate_aux(vd);
1535
1536 if (!vdev_is_dead(vd))
1537 l2arc_add_vdev(spa, vd);
1538 }
1539 }
1540
1541 /*
1542 * Purge vdevs that were dropped
1543 */
1544 for (i = 0; i < oldnvdevs; i++) {
1545 uint64_t pool;
1546
1547 vd = oldvdevs[i];
1548 if (vd != NULL) {
1549 ASSERT(vd->vdev_isl2cache);
1550
1551 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1552 pool != 0ULL && l2arc_vdev_present(vd))
1553 l2arc_remove_vdev(vd);
1554 vdev_clear_stats(vd);
1555 vdev_free(vd);
1556 }
1557 }
1558
1559 if (oldvdevs)
1560 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1561
1562 if (sav->sav_config == NULL)
1563 goto out;
1564
1565 sav->sav_vdevs = newvdevs;
1566 sav->sav_count = (int)nl2cache;
1567
1568 /*
1569 * Recompute the stashed list of l2cache devices, with status
1570 * information this time.
1571 */
1572 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1573 DATA_TYPE_NVLIST_ARRAY) == 0);
1574
1575 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1576 for (i = 0; i < sav->sav_count; i++)
1577 l2cache[i] = vdev_config_generate(spa,
1578 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1579 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1580 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1581out:
1582 for (i = 0; i < sav->sav_count; i++)
1583 nvlist_free(l2cache[i]);
1584 if (sav->sav_count)
1585 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1586}
1587
1588static int
1589load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1590{
1591 dmu_buf_t *db;
1592 char *packed = NULL;
1593 size_t nvsize = 0;
1594 int error;
1595 *value = NULL;
1596
1597 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1598 if (error != 0)
1599 return (error);
1600 nvsize = *(uint64_t *)db->db_data;
1601 dmu_buf_rele(db, FTAG);
1602
1603 packed = kmem_alloc(nvsize, KM_SLEEP);
1604 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1605 DMU_READ_PREFETCH);
1606 if (error == 0)
1607 error = nvlist_unpack(packed, nvsize, value, 0);
1608 kmem_free(packed, nvsize);
1609
1610 return (error);
1611}
1612
1613/*
1614 * Checks to see if the given vdev could not be opened, in which case we post a
1615 * sysevent to notify the autoreplace code that the device has been removed.
1616 */
1617static void
1618spa_check_removed(vdev_t *vd)
1619{
1620 for (int c = 0; c < vd->vdev_children; c++)
1621 spa_check_removed(vd->vdev_child[c]);
1622
1623 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1624 !vd->vdev_ishole) {
1625 zfs_post_autoreplace(vd->vdev_spa, vd);
1626 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1627 }
1628}
1629
1630/*
1631 * Validate the current config against the MOS config
1632 */
1633static boolean_t
1634spa_config_valid(spa_t *spa, nvlist_t *config)
1635{
1636 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1637 nvlist_t *nv;
1638
1639 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1640
1641 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1642 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1643
1644 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1645
1646 /*
1647 * If we're doing a normal import, then build up any additional
1648 * diagnostic information about missing devices in this config.
1649 * We'll pass this up to the user for further processing.
1650 */
1651 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1652 nvlist_t **child, *nv;
1653 uint64_t idx = 0;
1654
1655 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1656 KM_SLEEP);
1657 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1658
1659 for (int c = 0; c < rvd->vdev_children; c++) {
1660 vdev_t *tvd = rvd->vdev_child[c];
1661 vdev_t *mtvd = mrvd->vdev_child[c];
1662
1663 if (tvd->vdev_ops == &vdev_missing_ops &&
1664 mtvd->vdev_ops != &vdev_missing_ops &&
1665 mtvd->vdev_islog)
1666 child[idx++] = vdev_config_generate(spa, mtvd,
1667 B_FALSE, 0);
1668 }
1669
1670 if (idx) {
1671 VERIFY(nvlist_add_nvlist_array(nv,
1672 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1673 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1674 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1675
1676 for (int i = 0; i < idx; i++)
1677 nvlist_free(child[i]);
1678 }
1679 nvlist_free(nv);
1680 kmem_free(child, rvd->vdev_children * sizeof (char **));
1681 }
1682
1683 /*
1684 * Compare the root vdev tree with the information we have
1685 * from the MOS config (mrvd). Check each top-level vdev
1686 * with the corresponding MOS config top-level (mtvd).
1687 */
1688 for (int c = 0; c < rvd->vdev_children; c++) {
1689 vdev_t *tvd = rvd->vdev_child[c];
1690 vdev_t *mtvd = mrvd->vdev_child[c];
1691
1692 /*
1693 * Resolve any "missing" vdevs in the current configuration.
1694 * If we find that the MOS config has more accurate information
1695 * about the top-level vdev then use that vdev instead.
1696 */
1697 if (tvd->vdev_ops == &vdev_missing_ops &&
1698 mtvd->vdev_ops != &vdev_missing_ops) {
1699
1700 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1701 continue;
1702
1703 /*
1704 * Device specific actions.
1705 */
1706 if (mtvd->vdev_islog) {
1707 spa_set_log_state(spa, SPA_LOG_CLEAR);
1708 } else {
1709 /*
1710 * XXX - once we have 'readonly' pool
1711 * support we should be able to handle
1712 * missing data devices by transitioning
1713 * the pool to readonly.
1714 */
1715 continue;
1716 }
1717
1718 /*
1719 * Swap the missing vdev with the data we were
1720 * able to obtain from the MOS config.
1721 */
1722 vdev_remove_child(rvd, tvd);
1723 vdev_remove_child(mrvd, mtvd);
1724
1725 vdev_add_child(rvd, mtvd);
1726 vdev_add_child(mrvd, tvd);
1727
1728 spa_config_exit(spa, SCL_ALL, FTAG);
1729 vdev_load(mtvd);
1730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731
1732 vdev_reopen(rvd);
1733 } else if (mtvd->vdev_islog) {
1734 /*
1735 * Load the slog device's state from the MOS config
1736 * since it's possible that the label does not
1737 * contain the most up-to-date information.
1738 */
1739 vdev_load_log_state(tvd, mtvd);
1740 vdev_reopen(tvd);
1741 }
1742 }
1743 vdev_free(mrvd);
1744 spa_config_exit(spa, SCL_ALL, FTAG);
1745
1746 /*
1747 * Ensure we were able to validate the config.
1748 */
1749 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1750}
1751
1752/*
1753 * Check for missing log devices
1754 */
1755static boolean_t
1756spa_check_logs(spa_t *spa)
1757{
1758 boolean_t rv = B_FALSE;
1759
1760 switch (spa->spa_log_state) {
1761 case SPA_LOG_MISSING:
1762 /* need to recheck in case slog has been restored */
1763 case SPA_LOG_UNKNOWN:
1764 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1765 NULL, DS_FIND_CHILDREN) != 0);
1766 if (rv)
1767 spa_set_log_state(spa, SPA_LOG_MISSING);
1768 break;
1769 }
1770 return (rv);
1771}
1772
1773static boolean_t
1774spa_passivate_log(spa_t *spa)
1775{
1776 vdev_t *rvd = spa->spa_root_vdev;
1777 boolean_t slog_found = B_FALSE;
1778
1779 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1780
1781 if (!spa_has_slogs(spa))
1782 return (B_FALSE);
1783
1784 for (int c = 0; c < rvd->vdev_children; c++) {
1785 vdev_t *tvd = rvd->vdev_child[c];
1786 metaslab_group_t *mg = tvd->vdev_mg;
1787
1788 if (tvd->vdev_islog) {
1789 metaslab_group_passivate(mg);
1790 slog_found = B_TRUE;
1791 }
1792 }
1793
1794 return (slog_found);
1795}
1796
1797static void
1798spa_activate_log(spa_t *spa)
1799{
1800 vdev_t *rvd = spa->spa_root_vdev;
1801
1802 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1803
1804 for (int c = 0; c < rvd->vdev_children; c++) {
1805 vdev_t *tvd = rvd->vdev_child[c];
1806 metaslab_group_t *mg = tvd->vdev_mg;
1807
1808 if (tvd->vdev_islog)
1809 metaslab_group_activate(mg);
1810 }
1811}
1812
1813int
1814spa_offline_log(spa_t *spa)
1815{
1816 int error;
1817
1818 error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1819 NULL, DS_FIND_CHILDREN);
1820 if (error == 0) {
1821 /*
1822 * We successfully offlined the log device, sync out the
1823 * current txg so that the "stubby" block can be removed
1824 * by zil_sync().
1825 */
1826 txg_wait_synced(spa->spa_dsl_pool, 0);
1827 }
1828 return (error);
1829}
1830
1831static void
1832spa_aux_check_removed(spa_aux_vdev_t *sav)
1833{
1834 int i;
1835
1836 for (i = 0; i < sav->sav_count; i++)
1837 spa_check_removed(sav->sav_vdevs[i]);
1838}
1839
1840void
1841spa_claim_notify(zio_t *zio)
1842{
1843 spa_t *spa = zio->io_spa;
1844
1845 if (zio->io_error)
1846 return;
1847
1848 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1849 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1850 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1851 mutex_exit(&spa->spa_props_lock);
1852}
1853
1854typedef struct spa_load_error {
1855 uint64_t sle_meta_count;
1856 uint64_t sle_data_count;
1857} spa_load_error_t;
1858
1859static void
1860spa_load_verify_done(zio_t *zio)
1861{
1862 blkptr_t *bp = zio->io_bp;
1863 spa_load_error_t *sle = zio->io_private;
1864 dmu_object_type_t type = BP_GET_TYPE(bp);
1865 int error = zio->io_error;
1866
1867 if (error) {
1868 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1869 type != DMU_OT_INTENT_LOG)
1870 atomic_add_64(&sle->sle_meta_count, 1);
1871 else
1872 atomic_add_64(&sle->sle_data_count, 1);
1873 }
1874 zio_data_buf_free(zio->io_data, zio->io_size);
1875}
1876
1877/*ARGSUSED*/
1878static int
1879spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1880 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1881{
1882 if (!BP_IS_HOLE(bp)) {
1883 zio_t *rio = arg;
1884 size_t size = BP_GET_PSIZE(bp);
1885 void *data = zio_data_buf_alloc(size);
1886
1887 zio_nowait(zio_read(rio, spa, bp, data, size,
1888 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1889 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1890 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1891 }
1892 return (0);
1893}
1894
1895static int
1896spa_load_verify(spa_t *spa)
1897{
1898 zio_t *rio;
1899 spa_load_error_t sle = { 0 };
1900 zpool_rewind_policy_t policy;
1901 boolean_t verify_ok = B_FALSE;
1902 int error;
1903
1904 zpool_get_rewind_policy(spa->spa_config, &policy);
1905
1906 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1907 return (0);
1908
1909 rio = zio_root(spa, NULL, &sle,
1910 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1911
1912 error = traverse_pool(spa, spa->spa_verify_min_txg,
1913 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1914
1915 (void) zio_wait(rio);
1916
1917 spa->spa_load_meta_errors = sle.sle_meta_count;
1918 spa->spa_load_data_errors = sle.sle_data_count;
1919
1920 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1921 sle.sle_data_count <= policy.zrp_maxdata) {
1922 int64_t loss = 0;
1923
1924 verify_ok = B_TRUE;
1925 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1926 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1927
1928 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1929 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1930 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1931 VERIFY(nvlist_add_int64(spa->spa_load_info,
1932 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1933 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1934 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1935 } else {
1936 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1937 }
1938
1939 if (error) {
1940 if (error != ENXIO && error != EIO)
1941 error = SET_ERROR(EIO);
1942 return (error);
1943 }
1944
1945 return (verify_ok ? 0 : EIO);
1946}
1947
1948/*
1949 * Find a value in the pool props object.
1950 */
1951static void
1952spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1953{
1954 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1955 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1956}
1957
1958/*
1959 * Find a value in the pool directory object.
1960 */
1961static int
1962spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1963{
1964 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1965 name, sizeof (uint64_t), 1, val));
1966}
1967
1968static int
1969spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1970{
1971 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1972 return (err);
1973}
1974
1975/*
1976 * Fix up config after a partly-completed split. This is done with the
1977 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1978 * pool have that entry in their config, but only the splitting one contains
1979 * a list of all the guids of the vdevs that are being split off.
1980 *
1981 * This function determines what to do with that list: either rejoin
1982 * all the disks to the pool, or complete the splitting process. To attempt
1983 * the rejoin, each disk that is offlined is marked online again, and
1984 * we do a reopen() call. If the vdev label for every disk that was
1985 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1986 * then we call vdev_split() on each disk, and complete the split.
1987 *
1988 * Otherwise we leave the config alone, with all the vdevs in place in
1989 * the original pool.
1990 */
1991static void
1992spa_try_repair(spa_t *spa, nvlist_t *config)
1993{
1994 uint_t extracted;
1995 uint64_t *glist;
1996 uint_t i, gcount;
1997 nvlist_t *nvl;
1998 vdev_t **vd;
1999 boolean_t attempt_reopen;
2000
2001 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2002 return;
2003
2004 /* check that the config is complete */
2005 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2006 &glist, &gcount) != 0)
2007 return;
2008
2009 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2010
2011 /* attempt to online all the vdevs & validate */
2012 attempt_reopen = B_TRUE;
2013 for (i = 0; i < gcount; i++) {
2014 if (glist[i] == 0) /* vdev is hole */
2015 continue;
2016
2017 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2018 if (vd[i] == NULL) {
2019 /*
2020 * Don't bother attempting to reopen the disks;
2021 * just do the split.
2022 */
2023 attempt_reopen = B_FALSE;
2024 } else {
2025 /* attempt to re-online it */
2026 vd[i]->vdev_offline = B_FALSE;
2027 }
2028 }
2029
2030 if (attempt_reopen) {
2031 vdev_reopen(spa->spa_root_vdev);
2032
2033 /* check each device to see what state it's in */
2034 for (extracted = 0, i = 0; i < gcount; i++) {
2035 if (vd[i] != NULL &&
2036 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2037 break;
2038 ++extracted;
2039 }
2040 }
2041
2042 /*
2043 * If every disk has been moved to the new pool, or if we never
2044 * even attempted to look at them, then we split them off for
2045 * good.
2046 */
2047 if (!attempt_reopen || gcount == extracted) {
2048 for (i = 0; i < gcount; i++)
2049 if (vd[i] != NULL)
2050 vdev_split(vd[i]);
2051 vdev_reopen(spa->spa_root_vdev);
2052 }
2053
2054 kmem_free(vd, gcount * sizeof (vdev_t *));
2055}
2056
2057static int
2058spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2059 boolean_t mosconfig)
2060{
2061 nvlist_t *config = spa->spa_config;
2062 char *ereport = FM_EREPORT_ZFS_POOL;
2063 char *comment;
2064 int error;
2065 uint64_t pool_guid;
2066 nvlist_t *nvl;
2067
2068 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2069 return (SET_ERROR(EINVAL));
2070
2071 ASSERT(spa->spa_comment == NULL);
2072 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2073 spa->spa_comment = spa_strdup(comment);
2074
2075 /*
2076 * Versioning wasn't explicitly added to the label until later, so if
2077 * it's not present treat it as the initial version.
2078 */
2079 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2080 &spa->spa_ubsync.ub_version) != 0)
2081 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2082
2083 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2084 &spa->spa_config_txg);
2085
2086 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2087 spa_guid_exists(pool_guid, 0)) {
2088 error = SET_ERROR(EEXIST);
2089 } else {
2090 spa->spa_config_guid = pool_guid;
2091
2092 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2093 &nvl) == 0) {
2094 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2095 KM_SLEEP) == 0);
2096 }
2097
2098 nvlist_free(spa->spa_load_info);
2099 spa->spa_load_info = fnvlist_alloc();
2100
2101 gethrestime(&spa->spa_loaded_ts);
2102 error = spa_load_impl(spa, pool_guid, config, state, type,
2103 mosconfig, &ereport);
2104 }
2105
2106 spa->spa_minref = refcount_count(&spa->spa_refcount);
2107 if (error) {
2108 if (error != EEXIST) {
2109 spa->spa_loaded_ts.tv_sec = 0;
2110 spa->spa_loaded_ts.tv_nsec = 0;
2111 }
2112 if (error != EBADF) {
2113 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2114 }
2115 }
2116 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2117 spa->spa_ena = 0;
2118
2119 return (error);
2120}
2121
2122/*
2123 * Load an existing storage pool, using the pool's builtin spa_config as a
2124 * source of configuration information.
2125 */
2126static int
2127spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2129 char **ereport)
2130{
2131 int error = 0;
2132 nvlist_t *nvroot = NULL;
2133 nvlist_t *label;
2134 vdev_t *rvd;
2135 uberblock_t *ub = &spa->spa_uberblock;
2136 uint64_t children, config_cache_txg = spa->spa_config_txg;
2137 int orig_mode = spa->spa_mode;
2138 int parse;
2139 uint64_t obj;
2140 boolean_t missing_feat_write = B_FALSE;
2141
2142 /*
2143 * If this is an untrusted config, access the pool in read-only mode.
2144 * This prevents things like resilvering recently removed devices.
2145 */
2146 if (!mosconfig)
2147 spa->spa_mode = FREAD;
2148
2149 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2150
2151 spa->spa_load_state = state;
2152
2153 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2154 return (SET_ERROR(EINVAL));
2155
2156 parse = (type == SPA_IMPORT_EXISTING ?
2157 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2158
2159 /*
2160 * Create "The Godfather" zio to hold all async IOs
2161 */
2162 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2163 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2164
2165 /*
2166 * Parse the configuration into a vdev tree. We explicitly set the
2167 * value that will be returned by spa_version() since parsing the
2168 * configuration requires knowing the version number.
2169 */
2170 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2171 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2172 spa_config_exit(spa, SCL_ALL, FTAG);
2173
2174 if (error != 0)
2175 return (error);
2176
2177 ASSERT(spa->spa_root_vdev == rvd);
2178
2179 if (type != SPA_IMPORT_ASSEMBLE) {
2180 ASSERT(spa_guid(spa) == pool_guid);
2181 }
2182
2183 /*
2184 * Try to open all vdevs, loading each label in the process.
2185 */
2186 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2187 error = vdev_open(rvd);
2188 spa_config_exit(spa, SCL_ALL, FTAG);
2189 if (error != 0)
2190 return (error);
2191
2192 /*
2193 * We need to validate the vdev labels against the configuration that
2194 * we have in hand, which is dependent on the setting of mosconfig. If
2195 * mosconfig is true then we're validating the vdev labels based on
2196 * that config. Otherwise, we're validating against the cached config
2197 * (zpool.cache) that was read when we loaded the zfs module, and then
2198 * later we will recursively call spa_load() and validate against
2199 * the vdev config.
2200 *
2201 * If we're assembling a new pool that's been split off from an
2202 * existing pool, the labels haven't yet been updated so we skip
2203 * validation for now.
2204 */
2205 if (type != SPA_IMPORT_ASSEMBLE) {
2206 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2207 error = vdev_validate(rvd, mosconfig);
2208 spa_config_exit(spa, SCL_ALL, FTAG);
2209
2210 if (error != 0)
2211 return (error);
2212
2213 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2214 return (SET_ERROR(ENXIO));
2215 }
2216
2217 /*
2218 * Find the best uberblock.
2219 */
2220 vdev_uberblock_load(rvd, ub, &label);
2221
2222 /*
2223 * If we weren't able to find a single valid uberblock, return failure.
2224 */
2225 if (ub->ub_txg == 0) {
2226 nvlist_free(label);
2227 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2228 }
2229
2230 /*
2231 * If the pool has an unsupported version we can't open it.
2232 */
2233 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2234 nvlist_free(label);
2235 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2236 }
2237
2238 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2239 nvlist_t *features;
2240
2241 /*
2242 * If we weren't able to find what's necessary for reading the
2243 * MOS in the label, return failure.
2244 */
2245 if (label == NULL || nvlist_lookup_nvlist(label,
2246 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2247 nvlist_free(label);
2248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2249 ENXIO));
2250 }
2251
2252 /*
2253 * Update our in-core representation with the definitive values
2254 * from the label.
2255 */
2256 nvlist_free(spa->spa_label_features);
2257 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2258 }
2259
2260 nvlist_free(label);
2261
2262 /*
2263 * Look through entries in the label nvlist's features_for_read. If
2264 * there is a feature listed there which we don't understand then we
2265 * cannot open a pool.
2266 */
2267 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2268 nvlist_t *unsup_feat;
2269
2270 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2271 0);
2272
2273 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2274 NULL); nvp != NULL;
2275 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2276 if (!zfeature_is_supported(nvpair_name(nvp))) {
2277 VERIFY(nvlist_add_string(unsup_feat,
2278 nvpair_name(nvp), "") == 0);
2279 }
2280 }
2281
2282 if (!nvlist_empty(unsup_feat)) {
2283 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2284 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2285 nvlist_free(unsup_feat);
2286 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2287 ENOTSUP));
2288 }
2289
2290 nvlist_free(unsup_feat);
2291 }
2292
2293 /*
2294 * If the vdev guid sum doesn't match the uberblock, we have an
2295 * incomplete configuration. We first check to see if the pool
2296 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2297 * If it is, defer the vdev_guid_sum check till later so we
2298 * can handle missing vdevs.
2299 */
2300 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2301 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2302 rvd->vdev_guid_sum != ub->ub_guid_sum)
2303 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2304
2305 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2307 spa_try_repair(spa, config);
2308 spa_config_exit(spa, SCL_ALL, FTAG);
2309 nvlist_free(spa->spa_config_splitting);
2310 spa->spa_config_splitting = NULL;
2311 }
2312
2313 /*
2314 * Initialize internal SPA structures.
2315 */
2316 spa->spa_state = POOL_STATE_ACTIVE;
2317 spa->spa_ubsync = spa->spa_uberblock;
2318 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2319 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2320 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2321 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2322 spa->spa_claim_max_txg = spa->spa_first_txg;
2323 spa->spa_prev_software_version = ub->ub_software_version;
2324
2325 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2326 if (error)
2327 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2328 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2329
2330 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2331 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2332
2333 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2334 boolean_t missing_feat_read = B_FALSE;
2335 nvlist_t *unsup_feat, *enabled_feat;
2336
2337 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2338 &spa->spa_feat_for_read_obj) != 0) {
2339 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340 }
2341
2342 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2343 &spa->spa_feat_for_write_obj) != 0) {
2344 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2345 }
2346
2347 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2348 &spa->spa_feat_desc_obj) != 0) {
2349 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2350 }
2351
2352 enabled_feat = fnvlist_alloc();
2353 unsup_feat = fnvlist_alloc();
2354
2355 if (!spa_features_check(spa, B_FALSE,
2356 unsup_feat, enabled_feat))
2357 missing_feat_read = B_TRUE;
2358
2359 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2360 if (!spa_features_check(spa, B_TRUE,
2361 unsup_feat, enabled_feat)) {
2362 missing_feat_write = B_TRUE;
2363 }
2364 }
2365
2366 fnvlist_add_nvlist(spa->spa_load_info,
2367 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2368
2369 if (!nvlist_empty(unsup_feat)) {
2370 fnvlist_add_nvlist(spa->spa_load_info,
2371 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2372 }
2373
2374 fnvlist_free(enabled_feat);
2375 fnvlist_free(unsup_feat);
2376
2377 if (!missing_feat_read) {
2378 fnvlist_add_boolean(spa->spa_load_info,
2379 ZPOOL_CONFIG_CAN_RDONLY);
2380 }
2381
2382 /*
2383 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2384 * twofold: to determine whether the pool is available for
2385 * import in read-write mode and (if it is not) whether the
2386 * pool is available for import in read-only mode. If the pool
2387 * is available for import in read-write mode, it is displayed
2388 * as available in userland; if it is not available for import
2389 * in read-only mode, it is displayed as unavailable in
2390 * userland. If the pool is available for import in read-only
2391 * mode but not read-write mode, it is displayed as unavailable
2392 * in userland with a special note that the pool is actually
2393 * available for open in read-only mode.
2394 *
2395 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2396 * missing a feature for write, we must first determine whether
2397 * the pool can be opened read-only before returning to
2398 * userland in order to know whether to display the
2399 * abovementioned note.
2400 */
2401 if (missing_feat_read || (missing_feat_write &&
2402 spa_writeable(spa))) {
2403 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2404 ENOTSUP));
2405 }
2406
2407 /*
2408 * Load refcounts for ZFS features from disk into an in-memory
2409 * cache during SPA initialization.
2410 */
2411 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2412 uint64_t refcount;
2413
2414 error = feature_get_refcount_from_disk(spa,
2415 &spa_feature_table[i], &refcount);
2416 if (error == 0) {
2417 spa->spa_feat_refcount_cache[i] = refcount;
2418 } else if (error == ENOTSUP) {
2419 spa->spa_feat_refcount_cache[i] =
2420 SPA_FEATURE_DISABLED;
2421 } else {
2422 return (spa_vdev_err(rvd,
2423 VDEV_AUX_CORRUPT_DATA, EIO));
2424 }
2425 }
2426 }
2427
2428 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2429 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2430 &spa->spa_feat_enabled_txg_obj) != 0) {
2431 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432 }
2433 }
2434
2435 spa->spa_is_initializing = B_TRUE;
2436 error = dsl_pool_open(spa->spa_dsl_pool);
2437 spa->spa_is_initializing = B_FALSE;
2438 if (error != 0)
2439 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440
2441 if (!mosconfig) {
2442 uint64_t hostid;
2443 nvlist_t *policy = NULL, *nvconfig;
2444
2445 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447
2448 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2449 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2450 char *hostname;
2451 unsigned long myhostid = 0;
2452
2453 VERIFY(nvlist_lookup_string(nvconfig,
2454 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2455
2456#ifdef _KERNEL
2457 myhostid = zone_get_hostid(NULL);
2458#else /* _KERNEL */
2459 /*
2460 * We're emulating the system's hostid in userland, so
2461 * we can't use zone_get_hostid().
2462 */
2463 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2464#endif /* _KERNEL */
2465 if (check_hostid && hostid != 0 && myhostid != 0 &&
2466 hostid != myhostid) {
2467 nvlist_free(nvconfig);
2468 cmn_err(CE_WARN, "pool '%s' could not be "
2469 "loaded as it was last accessed by "
2470 "another system (host: %s hostid: 0x%lx). "
2471 "See: http://illumos.org/msg/ZFS-8000-EY",
2472 spa_name(spa), hostname,
2473 (unsigned long)hostid);
2474 return (SET_ERROR(EBADF));
2475 }
2476 }
2477 if (nvlist_lookup_nvlist(spa->spa_config,
2478 ZPOOL_REWIND_POLICY, &policy) == 0)
2479 VERIFY(nvlist_add_nvlist(nvconfig,
2480 ZPOOL_REWIND_POLICY, policy) == 0);
2481
2482 spa_config_set(spa, nvconfig);
2483 spa_unload(spa);
2484 spa_deactivate(spa);
2485 spa_activate(spa, orig_mode);
2486
2487 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2488 }
2489
2490 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2491 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2492 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2493 if (error != 0)
2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2495
2496 /*
2497 * Load the bit that tells us to use the new accounting function
2498 * (raid-z deflation). If we have an older pool, this will not
2499 * be present.
2500 */
2501 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2502 if (error != 0 && error != ENOENT)
2503 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504
2505 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2506 &spa->spa_creation_version);
2507 if (error != 0 && error != ENOENT)
2508 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2509
2510 /*
2511 * Load the persistent error log. If we have an older pool, this will
2512 * not be present.
2513 */
2514 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2515 if (error != 0 && error != ENOENT)
2516 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2517
2518 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2519 &spa->spa_errlog_scrub);
2520 if (error != 0 && error != ENOENT)
2521 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2522
2523 /*
2524 * Load the history object. If we have an older pool, this
2525 * will not be present.
2526 */
2527 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2528 if (error != 0 && error != ENOENT)
2529 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2530
2531 /*
2532 * If we're assembling the pool from the split-off vdevs of
2533 * an existing pool, we don't want to attach the spares & cache
2534 * devices.
2535 */
2536
2537 /*
2538 * Load any hot spares for this pool.
2539 */
2540 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2541 if (error != 0 && error != ENOENT)
2542 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2543 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2544 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2545 if (load_nvlist(spa, spa->spa_spares.sav_object,
2546 &spa->spa_spares.sav_config) != 0)
2547 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2548
2549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2550 spa_load_spares(spa);
2551 spa_config_exit(spa, SCL_ALL, FTAG);
2552 } else if (error == 0) {
2553 spa->spa_spares.sav_sync = B_TRUE;
2554 }
2555
2556 /*
2557 * Load any level 2 ARC devices for this pool.
2558 */
2559 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2560 &spa->spa_l2cache.sav_object);
2561 if (error != 0 && error != ENOENT)
2562 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2563 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2564 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2565 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2566 &spa->spa_l2cache.sav_config) != 0)
2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568
2569 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2570 spa_load_l2cache(spa);
2571 spa_config_exit(spa, SCL_ALL, FTAG);
2572 } else if (error == 0) {
2573 spa->spa_l2cache.sav_sync = B_TRUE;
2574 }
2575
2576 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2577
2578 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2579 if (error && error != ENOENT)
2580 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2581
2582 if (error == 0) {
2583 uint64_t autoreplace;
2584
2585 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2586 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2587 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2588 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2589 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2590 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2591 &spa->spa_dedup_ditto);
2592
2593 spa->spa_autoreplace = (autoreplace != 0);
2594 }
2595
2596 /*
2597 * If the 'autoreplace' property is set, then post a resource notifying
2598 * the ZFS DE that it should not issue any faults for unopenable
2599 * devices. We also iterate over the vdevs, and post a sysevent for any
2600 * unopenable vdevs so that the normal autoreplace handler can take
2601 * over.
2602 */
2603 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2604 spa_check_removed(spa->spa_root_vdev);
2605 /*
2606 * For the import case, this is done in spa_import(), because
2607 * at this point we're using the spare definitions from
2608 * the MOS config, not necessarily from the userland config.
2609 */
2610 if (state != SPA_LOAD_IMPORT) {
2611 spa_aux_check_removed(&spa->spa_spares);
2612 spa_aux_check_removed(&spa->spa_l2cache);
2613 }
2614 }
2615
2616 /*
2617 * Load the vdev state for all toplevel vdevs.
2618 */
2619 vdev_load(rvd);
2620
2621 /*
2622 * Propagate the leaf DTLs we just loaded all the way up the tree.
2623 */
2624 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2625 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2626 spa_config_exit(spa, SCL_ALL, FTAG);
2627
2628 /*
2629 * Load the DDTs (dedup tables).
2630 */
2631 error = ddt_load(spa);
2632 if (error != 0)
2633 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2634
2635 spa_update_dspace(spa);
2636
2637 /*
2638 * Validate the config, using the MOS config to fill in any
2639 * information which might be missing. If we fail to validate
2640 * the config then declare the pool unfit for use. If we're
2641 * assembling a pool from a split, the log is not transferred
2642 * over.
2643 */
2644 if (type != SPA_IMPORT_ASSEMBLE) {
2645 nvlist_t *nvconfig;
2646
2647 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2648 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2649
2650 if (!spa_config_valid(spa, nvconfig)) {
2651 nvlist_free(nvconfig);
2652 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2653 ENXIO));
2654 }
2655 nvlist_free(nvconfig);
2656
2657 /*
2658 * Now that we've validated the config, check the state of the
2659 * root vdev. If it can't be opened, it indicates one or
2660 * more toplevel vdevs are faulted.
2661 */
2662 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2663 return (SET_ERROR(ENXIO));
2664
2665 if (spa_check_logs(spa)) {
2666 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2667 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2668 }
2669 }
2670
2671 if (missing_feat_write) {
2672 ASSERT(state == SPA_LOAD_TRYIMPORT);
2673
2674 /*
2675 * At this point, we know that we can open the pool in
2676 * read-only mode but not read-write mode. We now have enough
2677 * information and can return to userland.
2678 */
2679 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2680 }
2681
2682 /*
2683 * We've successfully opened the pool, verify that we're ready
2684 * to start pushing transactions.
2685 */
2686 if (state != SPA_LOAD_TRYIMPORT) {
2687 if (error = spa_load_verify(spa))
2688 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2689 error));
2690 }
2691
2692 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2693 spa->spa_load_max_txg == UINT64_MAX)) {
2694 dmu_tx_t *tx;
2695 int need_update = B_FALSE;
2696
2697 ASSERT(state != SPA_LOAD_TRYIMPORT);
2698
2699 /*
2700 * Claim log blocks that haven't been committed yet.
2701 * This must all happen in a single txg.
2702 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2703 * invoked from zil_claim_log_block()'s i/o done callback.
2704 * Price of rollback is that we abandon the log.
2705 */
2706 spa->spa_claiming = B_TRUE;
2707
2708 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2709 spa_first_txg(spa));
2710 (void) dmu_objset_find(spa_name(spa),
2711 zil_claim, tx, DS_FIND_CHILDREN);
2712 dmu_tx_commit(tx);
2713
2714 spa->spa_claiming = B_FALSE;
2715
2716 spa_set_log_state(spa, SPA_LOG_GOOD);
2717 spa->spa_sync_on = B_TRUE;
2718 txg_sync_start(spa->spa_dsl_pool);
2719
2720 /*
2721 * Wait for all claims to sync. We sync up to the highest
2722 * claimed log block birth time so that claimed log blocks
2723 * don't appear to be from the future. spa_claim_max_txg
2724 * will have been set for us by either zil_check_log_chain()
2725 * (invoked from spa_check_logs()) or zil_claim() above.
2726 */
2727 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2728
2729 /*
2730 * If the config cache is stale, or we have uninitialized
2731 * metaslabs (see spa_vdev_add()), then update the config.
2732 *
2733 * If this is a verbatim import, trust the current
2734 * in-core spa_config and update the disk labels.
2735 */
2736 if (config_cache_txg != spa->spa_config_txg ||
2737 state == SPA_LOAD_IMPORT ||
2738 state == SPA_LOAD_RECOVER ||
2739 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2740 need_update = B_TRUE;
2741
2742 for (int c = 0; c < rvd->vdev_children; c++)
2743 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2744 need_update = B_TRUE;
2745
2746 /*
2747 * Update the config cache asychronously in case we're the
2748 * root pool, in which case the config cache isn't writable yet.
2749 */
2750 if (need_update)
2751 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2752
2753 /*
2754 * Check all DTLs to see if anything needs resilvering.
2755 */
2756 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2757 vdev_resilver_needed(rvd, NULL, NULL))
2758 spa_async_request(spa, SPA_ASYNC_RESILVER);
2759
2760 /*
2761 * Log the fact that we booted up (so that we can detect if
2762 * we rebooted in the middle of an operation).
2763 */
2764 spa_history_log_version(spa, "open");
2765
2766 /*
2767 * Delete any inconsistent datasets.
2768 */
2769 (void) dmu_objset_find(spa_name(spa),
2770 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2771
2772 /*
2773 * Clean up any stale temporary dataset userrefs.
2774 */
2775 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2776 }
2777
2778 return (0);
2779}
2780
2781static int
2782spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2783{
2784 int mode = spa->spa_mode;
2785
2786 spa_unload(spa);
2787 spa_deactivate(spa);
2788
2789 spa->spa_load_max_txg--;
2790
2791 spa_activate(spa, mode);
2792 spa_async_suspend(spa);
2793
2794 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2795}
2796
2797/*
2798 * If spa_load() fails this function will try loading prior txg's. If
2799 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2800 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2801 * function will not rewind the pool and will return the same error as
2802 * spa_load().
2803 */
2804static int
2805spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2806 uint64_t max_request, int rewind_flags)
2807{
2808 nvlist_t *loadinfo = NULL;
2809 nvlist_t *config = NULL;
2810 int load_error, rewind_error;
2811 uint64_t safe_rewind_txg;
2812 uint64_t min_txg;
2813
2814 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2815 spa->spa_load_max_txg = spa->spa_load_txg;
2816 spa_set_log_state(spa, SPA_LOG_CLEAR);
2817 } else {
2818 spa->spa_load_max_txg = max_request;
2819 }
2820
2821 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2822 mosconfig);
2823 if (load_error == 0)
2824 return (0);
2825
2826 if (spa->spa_root_vdev != NULL)
2827 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2828
2829 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2830 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2831
2832 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2833 nvlist_free(config);
2834 return (load_error);
2835 }
2836
2837 if (state == SPA_LOAD_RECOVER) {
2838 /* Price of rolling back is discarding txgs, including log */
2839 spa_set_log_state(spa, SPA_LOG_CLEAR);
2840 } else {
2841 /*
2842 * If we aren't rolling back save the load info from our first
2843 * import attempt so that we can restore it after attempting
2844 * to rewind.
2845 */
2846 loadinfo = spa->spa_load_info;
2847 spa->spa_load_info = fnvlist_alloc();
2848 }
2849
2850 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2851 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2852 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2853 TXG_INITIAL : safe_rewind_txg;
2854
2855 /*
2856 * Continue as long as we're finding errors, we're still within
2857 * the acceptable rewind range, and we're still finding uberblocks
2858 */
2859 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2860 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2861 if (spa->spa_load_max_txg < safe_rewind_txg)
2862 spa->spa_extreme_rewind = B_TRUE;
2863 rewind_error = spa_load_retry(spa, state, mosconfig);
2864 }
2865
2866 spa->spa_extreme_rewind = B_FALSE;
2867 spa->spa_load_max_txg = UINT64_MAX;
2868
2869 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2870 spa_config_set(spa, config);
2871
2872 if (state == SPA_LOAD_RECOVER) {
2873 ASSERT3P(loadinfo, ==, NULL);
2874 return (rewind_error);
2875 } else {
2876 /* Store the rewind info as part of the initial load info */
2877 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2878 spa->spa_load_info);
2879
2880 /* Restore the initial load info */
2881 fnvlist_free(spa->spa_load_info);
2882 spa->spa_load_info = loadinfo;
2883
2884 return (load_error);
2885 }
2886}
2887
2888/*
2889 * Pool Open/Import
2890 *
2891 * The import case is identical to an open except that the configuration is sent
2892 * down from userland, instead of grabbed from the configuration cache. For the
2893 * case of an open, the pool configuration will exist in the
2894 * POOL_STATE_UNINITIALIZED state.
2895 *
2896 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2897 * the same time open the pool, without having to keep around the spa_t in some
2898 * ambiguous state.
2899 */
2900static int
2901spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2902 nvlist_t **config)
2903{
2904 spa_t *spa;
2905 spa_load_state_t state = SPA_LOAD_OPEN;
2906 int error;
2907 int locked = B_FALSE;
2908 int firstopen = B_FALSE;
2909
2910 *spapp = NULL;
2911
2912 /*
2913 * As disgusting as this is, we need to support recursive calls to this
2914 * function because dsl_dir_open() is called during spa_load(), and ends
2915 * up calling spa_open() again. The real fix is to figure out how to
2916 * avoid dsl_dir_open() calling this in the first place.
2917 */
2918 if (mutex_owner(&spa_namespace_lock) != curthread) {
2919 mutex_enter(&spa_namespace_lock);
2920 locked = B_TRUE;
2921 }
2922
2923 if ((spa = spa_lookup(pool)) == NULL) {
2924 if (locked)
2925 mutex_exit(&spa_namespace_lock);
2926 return (SET_ERROR(ENOENT));
2927 }
2928
2929 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2930 zpool_rewind_policy_t policy;
2931
2932 firstopen = B_TRUE;
2933
2934 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2935 &policy);
2936 if (policy.zrp_request & ZPOOL_DO_REWIND)
2937 state = SPA_LOAD_RECOVER;
2938
2939 spa_activate(spa, spa_mode_global);
2940
2941 if (state != SPA_LOAD_RECOVER)
2942 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2943
2944 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2945 policy.zrp_request);
2946
2947 if (error == EBADF) {
2948 /*
2949 * If vdev_validate() returns failure (indicated by
2950 * EBADF), it indicates that one of the vdevs indicates
2951 * that the pool has been exported or destroyed. If
2952 * this is the case, the config cache is out of sync and
2953 * we should remove the pool from the namespace.
2954 */
2955 spa_unload(spa);
2956 spa_deactivate(spa);
2957 spa_config_sync(spa, B_TRUE, B_TRUE);
2958 spa_remove(spa);
2959 if (locked)
2960 mutex_exit(&spa_namespace_lock);
2961 return (SET_ERROR(ENOENT));
2962 }
2963
2964 if (error) {
2965 /*
2966 * We can't open the pool, but we still have useful
2967 * information: the state of each vdev after the
2968 * attempted vdev_open(). Return this to the user.
2969 */
2970 if (config != NULL && spa->spa_config) {
2971 VERIFY(nvlist_dup(spa->spa_config, config,
2972 KM_SLEEP) == 0);
2973 VERIFY(nvlist_add_nvlist(*config,
2974 ZPOOL_CONFIG_LOAD_INFO,
2975 spa->spa_load_info) == 0);
2976 }
2977 spa_unload(spa);
2978 spa_deactivate(spa);
2979 spa->spa_last_open_failed = error;
2980 if (locked)
2981 mutex_exit(&spa_namespace_lock);
2982 *spapp = NULL;
2983 return (error);
2984 }
2985 }
2986
2987 spa_open_ref(spa, tag);
2988
2989 if (config != NULL)
2990 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2991
2992 /*
2993 * If we've recovered the pool, pass back any information we
2994 * gathered while doing the load.
2995 */
2996 if (state == SPA_LOAD_RECOVER) {
2997 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2998 spa->spa_load_info) == 0);
2999 }
3000
3001 if (locked) {
3002 spa->spa_last_open_failed = 0;
3003 spa->spa_last_ubsync_txg = 0;
3004 spa->spa_load_txg = 0;
3005 mutex_exit(&spa_namespace_lock);
3006#ifdef __FreeBSD__
3007#ifdef _KERNEL
3008 if (firstopen)
3009 zvol_create_minors(spa->spa_name);
3010#endif
3011#endif
3012 }
3013
3014 *spapp = spa;
3015
3016 return (0);
3017}
3018
3019int
3020spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3021 nvlist_t **config)
3022{
3023 return (spa_open_common(name, spapp, tag, policy, config));
3024}
3025
3026int
3027spa_open(const char *name, spa_t **spapp, void *tag)
3028{
3029 return (spa_open_common(name, spapp, tag, NULL, NULL));
3030}
3031
3032/*
3033 * Lookup the given spa_t, incrementing the inject count in the process,
3034 * preventing it from being exported or destroyed.
3035 */
3036spa_t *
3037spa_inject_addref(char *name)
3038{
3039 spa_t *spa;
3040
3041 mutex_enter(&spa_namespace_lock);
3042 if ((spa = spa_lookup(name)) == NULL) {
3043 mutex_exit(&spa_namespace_lock);
3044 return (NULL);
3045 }
3046 spa->spa_inject_ref++;
3047 mutex_exit(&spa_namespace_lock);
3048
3049 return (spa);
3050}
3051
3052void
3053spa_inject_delref(spa_t *spa)
3054{
3055 mutex_enter(&spa_namespace_lock);
3056 spa->spa_inject_ref--;
3057 mutex_exit(&spa_namespace_lock);
3058}
3059
3060/*
3061 * Add spares device information to the nvlist.
3062 */
3063static void
3064spa_add_spares(spa_t *spa, nvlist_t *config)
3065{
3066 nvlist_t **spares;
3067 uint_t i, nspares;
3068 nvlist_t *nvroot;
3069 uint64_t guid;
3070 vdev_stat_t *vs;
3071 uint_t vsc;
3072 uint64_t pool;
3073
3074 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3075
3076 if (spa->spa_spares.sav_count == 0)
3077 return;
3078
3079 VERIFY(nvlist_lookup_nvlist(config,
3080 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3081 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3082 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3083 if (nspares != 0) {
3084 VERIFY(nvlist_add_nvlist_array(nvroot,
3085 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3086 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3087 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3088
3089 /*
3090 * Go through and find any spares which have since been
3091 * repurposed as an active spare. If this is the case, update
3092 * their status appropriately.
3093 */
3094 for (i = 0; i < nspares; i++) {
3095 VERIFY(nvlist_lookup_uint64(spares[i],
3096 ZPOOL_CONFIG_GUID, &guid) == 0);
3097 if (spa_spare_exists(guid, &pool, NULL) &&
3098 pool != 0ULL) {
3099 VERIFY(nvlist_lookup_uint64_array(
3100 spares[i], ZPOOL_CONFIG_VDEV_STATS,
3101 (uint64_t **)&vs, &vsc) == 0);
3102 vs->vs_state = VDEV_STATE_CANT_OPEN;
3103 vs->vs_aux = VDEV_AUX_SPARED;
3104 }
3105 }
3106 }
3107}
3108
3109/*
3110 * Add l2cache device information to the nvlist, including vdev stats.
3111 */
3112static void
3113spa_add_l2cache(spa_t *spa, nvlist_t *config)
3114{
3115 nvlist_t **l2cache;
3116 uint_t i, j, nl2cache;
3117 nvlist_t *nvroot;
3118 uint64_t guid;
3119 vdev_t *vd;
3120 vdev_stat_t *vs;
3121 uint_t vsc;
3122
3123 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3124
3125 if (spa->spa_l2cache.sav_count == 0)
3126 return;
3127
3128 VERIFY(nvlist_lookup_nvlist(config,
3129 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3130 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3131 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3132 if (nl2cache != 0) {
3133 VERIFY(nvlist_add_nvlist_array(nvroot,
3134 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3135 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3136 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3137
3138 /*
3139 * Update level 2 cache device stats.
3140 */
3141
3142 for (i = 0; i < nl2cache; i++) {
3143 VERIFY(nvlist_lookup_uint64(l2cache[i],
3144 ZPOOL_CONFIG_GUID, &guid) == 0);
3145
3146 vd = NULL;
3147 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3148 if (guid ==
3149 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3150 vd = spa->spa_l2cache.sav_vdevs[j];
3151 break;
3152 }
3153 }
3154 ASSERT(vd != NULL);
3155
3156 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3157 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3158 == 0);
3159 vdev_get_stats(vd, vs);
3160 }
3161 }
3162}
3163
3164static void
3165spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3166{
3167 nvlist_t *features;
3168 zap_cursor_t zc;
3169 zap_attribute_t za;
3170
3171 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3172 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3173
3174 /* We may be unable to read features if pool is suspended. */
3175 if (spa_suspended(spa))
3176 goto out;
3177
3178 if (spa->spa_feat_for_read_obj != 0) {
3179 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3180 spa->spa_feat_for_read_obj);
3181 zap_cursor_retrieve(&zc, &za) == 0;
3182 zap_cursor_advance(&zc)) {
3183 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3184 za.za_num_integers == 1);
3185 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3186 za.za_first_integer));
3187 }
3188 zap_cursor_fini(&zc);
3189 }
3190
3191 if (spa->spa_feat_for_write_obj != 0) {
3192 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3193 spa->spa_feat_for_write_obj);
3194 zap_cursor_retrieve(&zc, &za) == 0;
3195 zap_cursor_advance(&zc)) {
3196 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3197 za.za_num_integers == 1);
3198 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3199 za.za_first_integer));
3200 }
3201 zap_cursor_fini(&zc);
3202 }
3203
3204out:
3205 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3206 features) == 0);
3207 nvlist_free(features);
3208}
3209
3210int
3211spa_get_stats(const char *name, nvlist_t **config,
3212 char *altroot, size_t buflen)
3213{
3214 int error;
3215 spa_t *spa;
3216
3217 *config = NULL;
3218 error = spa_open_common(name, &spa, FTAG, NULL, config);
3219
3220 if (spa != NULL) {
3221 /*
3222 * This still leaves a window of inconsistency where the spares
3223 * or l2cache devices could change and the config would be
3224 * self-inconsistent.
3225 */
3226 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3227
3228 if (*config != NULL) {
3229 uint64_t loadtimes[2];
3230
3231 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3232 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3233 VERIFY(nvlist_add_uint64_array(*config,
3234 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3235
3236 VERIFY(nvlist_add_uint64(*config,
3237 ZPOOL_CONFIG_ERRCOUNT,
3238 spa_get_errlog_size(spa)) == 0);
3239
3240 if (spa_suspended(spa))
3241 VERIFY(nvlist_add_uint64(*config,
3242 ZPOOL_CONFIG_SUSPENDED,
3243 spa->spa_failmode) == 0);
3244
3245 spa_add_spares(spa, *config);
3246 spa_add_l2cache(spa, *config);
3247 spa_add_feature_stats(spa, *config);
3248 }
3249 }
3250
3251 /*
3252 * We want to get the alternate root even for faulted pools, so we cheat
3253 * and call spa_lookup() directly.
3254 */
3255 if (altroot) {
3256 if (spa == NULL) {
3257 mutex_enter(&spa_namespace_lock);
3258 spa = spa_lookup(name);
3259 if (spa)
3260 spa_altroot(spa, altroot, buflen);
3261 else
3262 altroot[0] = '\0';
3263 spa = NULL;
3264 mutex_exit(&spa_namespace_lock);
3265 } else {
3266 spa_altroot(spa, altroot, buflen);
3267 }
3268 }
3269
3270 if (spa != NULL) {
3271 spa_config_exit(spa, SCL_CONFIG, FTAG);
3272 spa_close(spa, FTAG);
3273 }
3274
3275 return (error);
3276}
3277
3278/*
3279 * Validate that the auxiliary device array is well formed. We must have an
3280 * array of nvlists, each which describes a valid leaf vdev. If this is an
3281 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3282 * specified, as long as they are well-formed.
3283 */
3284static int
3285spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3286 spa_aux_vdev_t *sav, const char *config, uint64_t version,
3287 vdev_labeltype_t label)
3288{
3289 nvlist_t **dev;
3290 uint_t i, ndev;
3291 vdev_t *vd;
3292 int error;
3293
3294 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3295
3296 /*
3297 * It's acceptable to have no devs specified.
3298 */
3299 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3300 return (0);
3301
3302 if (ndev == 0)
3303 return (SET_ERROR(EINVAL));
3304
3305 /*
3306 * Make sure the pool is formatted with a version that supports this
3307 * device type.
3308 */
3309 if (spa_version(spa) < version)
3310 return (SET_ERROR(ENOTSUP));
3311
3312 /*
3313 * Set the pending device list so we correctly handle device in-use
3314 * checking.
3315 */
3316 sav->sav_pending = dev;
3317 sav->sav_npending = ndev;
3318
3319 for (i = 0; i < ndev; i++) {
3320 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3321 mode)) != 0)
3322 goto out;
3323
3324 if (!vd->vdev_ops->vdev_op_leaf) {
3325 vdev_free(vd);
3326 error = SET_ERROR(EINVAL);
3327 goto out;
3328 }
3329
3330 /*
3331 * The L2ARC currently only supports disk devices in
3332 * kernel context. For user-level testing, we allow it.
3333 */
3334#ifdef _KERNEL
3335 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3336 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3337 error = SET_ERROR(ENOTBLK);
3338 vdev_free(vd);
3339 goto out;
3340 }
3341#endif
3342 vd->vdev_top = vd;
3343
3344 if ((error = vdev_open(vd)) == 0 &&
3345 (error = vdev_label_init(vd, crtxg, label)) == 0) {
3346 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3347 vd->vdev_guid) == 0);
3348 }
3349
3350 vdev_free(vd);
3351
3352 if (error &&
3353 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3354 goto out;
3355 else
3356 error = 0;
3357 }
3358
3359out:
3360 sav->sav_pending = NULL;
3361 sav->sav_npending = 0;
3362 return (error);
3363}
3364
3365static int
3366spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3367{
3368 int error;
3369
3370 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3371
3372 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3373 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3374 VDEV_LABEL_SPARE)) != 0) {
3375 return (error);
3376 }
3377
3378 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3379 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3380 VDEV_LABEL_L2CACHE));
3381}
3382
3383static void
3384spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3385 const char *config)
3386{
3387 int i;
3388
3389 if (sav->sav_config != NULL) {
3390 nvlist_t **olddevs;
3391 uint_t oldndevs;
3392 nvlist_t **newdevs;
3393
3394 /*
3395 * Generate new dev list by concatentating with the
3396 * current dev list.
3397 */
3398 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3399 &olddevs, &oldndevs) == 0);
3400
3401 newdevs = kmem_alloc(sizeof (void *) *
3402 (ndevs + oldndevs), KM_SLEEP);
3403 for (i = 0; i < oldndevs; i++)
3404 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3405 KM_SLEEP) == 0);
3406 for (i = 0; i < ndevs; i++)
3407 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3408 KM_SLEEP) == 0);
3409
3410 VERIFY(nvlist_remove(sav->sav_config, config,
3411 DATA_TYPE_NVLIST_ARRAY) == 0);
3412
3413 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3414 config, newdevs, ndevs + oldndevs) == 0);
3415 for (i = 0; i < oldndevs + ndevs; i++)
3416 nvlist_free(newdevs[i]);
3417 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3418 } else {
3419 /*
3420 * Generate a new dev list.
3421 */
3422 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3423 KM_SLEEP) == 0);
3424 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3425 devs, ndevs) == 0);
3426 }
3427}
3428
3429/*
3430 * Stop and drop level 2 ARC devices
3431 */
3432void
3433spa_l2cache_drop(spa_t *spa)
3434{
3435 vdev_t *vd;
3436 int i;
3437 spa_aux_vdev_t *sav = &spa->spa_l2cache;
3438
3439 for (i = 0; i < sav->sav_count; i++) {
3440 uint64_t pool;
3441
3442 vd = sav->sav_vdevs[i];
3443 ASSERT(vd != NULL);
3444
3445 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3446 pool != 0ULL && l2arc_vdev_present(vd))
3447 l2arc_remove_vdev(vd);
3448 }
3449}
3450
3451/*
3452 * Pool Creation
3453 */
3454int
3455spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3456 nvlist_t *zplprops)
3457{
3458 spa_t *spa;
3459 char *altroot = NULL;
3460 vdev_t *rvd;
3461 dsl_pool_t *dp;
3462 dmu_tx_t *tx;
3463 int error = 0;
3464 uint64_t txg = TXG_INITIAL;
3465 nvlist_t **spares, **l2cache;
3466 uint_t nspares, nl2cache;
3467 uint64_t version, obj;
3468 boolean_t has_features;
3469
3470 /*
3471 * If this pool already exists, return failure.
3472 */
3473 mutex_enter(&spa_namespace_lock);
3474 if (spa_lookup(pool) != NULL) {
3475 mutex_exit(&spa_namespace_lock);
3476 return (SET_ERROR(EEXIST));
3477 }
3478
3479 /*
3480 * Allocate a new spa_t structure.
3481 */
3482 (void) nvlist_lookup_string(props,
3483 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3484 spa = spa_add(pool, NULL, altroot);
3485 spa_activate(spa, spa_mode_global);
3486
3487 if (props && (error = spa_prop_validate(spa, props))) {
3488 spa_deactivate(spa);
3489 spa_remove(spa);
3490 mutex_exit(&spa_namespace_lock);
3491 return (error);
3492 }
3493
3494 has_features = B_FALSE;
3495 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3496 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3497 if (zpool_prop_feature(nvpair_name(elem)))
3498 has_features = B_TRUE;
3499 }
3500
3501 if (has_features || nvlist_lookup_uint64(props,
3502 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3503 version = SPA_VERSION;
3504 }
3505 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3506
3507 spa->spa_first_txg = txg;
3508 spa->spa_uberblock.ub_txg = txg - 1;
3509 spa->spa_uberblock.ub_version = version;
3510 spa->spa_ubsync = spa->spa_uberblock;
3511
3512 /*
3513 * Create "The Godfather" zio to hold all async IOs
3514 */
3515 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3516 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3517
3518 /*
3519 * Create the root vdev.
3520 */
3521 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3522
3523 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3524
3525 ASSERT(error != 0 || rvd != NULL);
3526 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3527
3528 if (error == 0 && !zfs_allocatable_devs(nvroot))
3529 error = SET_ERROR(EINVAL);
3530
3531 if (error == 0 &&
3532 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3533 (error = spa_validate_aux(spa, nvroot, txg,
3534 VDEV_ALLOC_ADD)) == 0) {
3535 for (int c = 0; c < rvd->vdev_children; c++) {
3536 vdev_ashift_optimize(rvd->vdev_child[c]);
3537 vdev_metaslab_set_size(rvd->vdev_child[c]);
3538 vdev_expand(rvd->vdev_child[c], txg);
3539 }
3540 }
3541
3542 spa_config_exit(spa, SCL_ALL, FTAG);
3543
3544 if (error != 0) {
3545 spa_unload(spa);
3546 spa_deactivate(spa);
3547 spa_remove(spa);
3548 mutex_exit(&spa_namespace_lock);
3549 return (error);
3550 }
3551
3552 /*
3553 * Get the list of spares, if specified.
3554 */
3555 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3556 &spares, &nspares) == 0) {
3557 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3558 KM_SLEEP) == 0);
3559 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3560 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3561 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3562 spa_load_spares(spa);
3563 spa_config_exit(spa, SCL_ALL, FTAG);
3564 spa->spa_spares.sav_sync = B_TRUE;
3565 }
3566
3567 /*
3568 * Get the list of level 2 cache devices, if specified.
3569 */
3570 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3571 &l2cache, &nl2cache) == 0) {
3572 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3573 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3574 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3575 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3576 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3577 spa_load_l2cache(spa);
3578 spa_config_exit(spa, SCL_ALL, FTAG);
3579 spa->spa_l2cache.sav_sync = B_TRUE;
3580 }
3581
3582 spa->spa_is_initializing = B_TRUE;
3583 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3584 spa->spa_meta_objset = dp->dp_meta_objset;
3585 spa->spa_is_initializing = B_FALSE;
3586
3587 /*
3588 * Create DDTs (dedup tables).
3589 */
3590 ddt_create(spa);
3591
3592 spa_update_dspace(spa);
3593
3594 tx = dmu_tx_create_assigned(dp, txg);
3595
3596 /*
3597 * Create the pool config object.
3598 */
3599 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3600 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3601 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3602
3603 if (zap_add(spa->spa_meta_objset,
3604 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3605 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3606 cmn_err(CE_PANIC, "failed to add pool config");
3607 }
3608
3609 if (spa_version(spa) >= SPA_VERSION_FEATURES)
3610 spa_feature_create_zap_objects(spa, tx);
3611
3612 if (zap_add(spa->spa_meta_objset,
3613 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3614 sizeof (uint64_t), 1, &version, tx) != 0) {
3615 cmn_err(CE_PANIC, "failed to add pool version");
3616 }
3617
3618 /* Newly created pools with the right version are always deflated. */
3619 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3620 spa->spa_deflate = TRUE;
3621 if (zap_add(spa->spa_meta_objset,
3622 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3623 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3624 cmn_err(CE_PANIC, "failed to add deflate");
3625 }
3626 }
3627
3628 /*
3629 * Create the deferred-free bpobj. Turn off compression
3630 * because sync-to-convergence takes longer if the blocksize
3631 * keeps changing.
3632 */
3633 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3634 dmu_object_set_compress(spa->spa_meta_objset, obj,
3635 ZIO_COMPRESS_OFF, tx);
3636 if (zap_add(spa->spa_meta_objset,
3637 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3638 sizeof (uint64_t), 1, &obj, tx) != 0) {
3639 cmn_err(CE_PANIC, "failed to add bpobj");
3640 }
3641 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3642 spa->spa_meta_objset, obj));
3643
3644 /*
3645 * Create the pool's history object.
3646 */
3647 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3648 spa_history_create_obj(spa, tx);
3649
3650 /*
3651 * Set pool properties.
3652 */
3653 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3654 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3655 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3656 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3657
3658 if (props != NULL) {
3659 spa_configfile_set(spa, props, B_FALSE);
3660 spa_sync_props(props, tx);
3661 }
3662
3663 dmu_tx_commit(tx);
3664
3665 spa->spa_sync_on = B_TRUE;
3666 txg_sync_start(spa->spa_dsl_pool);
3667
3668 /*
3669 * We explicitly wait for the first transaction to complete so that our
3670 * bean counters are appropriately updated.
3671 */
3672 txg_wait_synced(spa->spa_dsl_pool, txg);
3673
3674 spa_config_sync(spa, B_FALSE, B_TRUE);
3675
3676 spa_history_log_version(spa, "create");
3677
3678 spa->spa_minref = refcount_count(&spa->spa_refcount);
3679
3680 mutex_exit(&spa_namespace_lock);
3681
3682 return (0);
3683}
3684
3685#ifdef _KERNEL
3686#if defined(sun)
3687/*
3688 * Get the root pool information from the root disk, then import the root pool
3689 * during the system boot up time.
3690 */
3691extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3692
3693static nvlist_t *
3694spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3695{
3696 nvlist_t *config;
3697 nvlist_t *nvtop, *nvroot;
3698 uint64_t pgid;
3699
3700 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3701 return (NULL);
3702
3703 /*
3704 * Add this top-level vdev to the child array.
3705 */
3706 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3707 &nvtop) == 0);
3708 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3709 &pgid) == 0);
3710 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3711
3712 /*
3713 * Put this pool's top-level vdevs into a root vdev.
3714 */
3715 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3716 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3717 VDEV_TYPE_ROOT) == 0);
3718 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3719 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3720 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3721 &nvtop, 1) == 0);
3722
3723 /*
3724 * Replace the existing vdev_tree with the new root vdev in
3725 * this pool's configuration (remove the old, add the new).
3726 */
3727 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3728 nvlist_free(nvroot);
3729 return (config);
3730}
3731
3732/*
3733 * Walk the vdev tree and see if we can find a device with "better"
3734 * configuration. A configuration is "better" if the label on that
3735 * device has a more recent txg.
3736 */
3737static void
3738spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3739{
3740 for (int c = 0; c < vd->vdev_children; c++)
3741 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3742
3743 if (vd->vdev_ops->vdev_op_leaf) {
3744 nvlist_t *label;
3745 uint64_t label_txg;
3746
3747 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3748 &label) != 0)
3749 return;
3750
3751 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3752 &label_txg) == 0);
3753
3754 /*
3755 * Do we have a better boot device?
3756 */
3757 if (label_txg > *txg) {
3758 *txg = label_txg;
3759 *avd = vd;
3760 }
3761 nvlist_free(label);
3762 }
3763}
3764
3765/*
3766 * Import a root pool.
3767 *
3768 * For x86. devpath_list will consist of devid and/or physpath name of
3769 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3770 * The GRUB "findroot" command will return the vdev we should boot.
3771 *
3772 * For Sparc, devpath_list consists the physpath name of the booting device
3773 * no matter the rootpool is a single device pool or a mirrored pool.
3774 * e.g.
3775 * "/pci@1f,0/ide@d/disk@0,0:a"
3776 */
3777int
3778spa_import_rootpool(char *devpath, char *devid)
3779{
3780 spa_t *spa;
3781 vdev_t *rvd, *bvd, *avd = NULL;
3782 nvlist_t *config, *nvtop;
3783 uint64_t guid, txg;
3784 char *pname;
3785 int error;
3786
3787 /*
3788 * Read the label from the boot device and generate a configuration.
3789 */
3790 config = spa_generate_rootconf(devpath, devid, &guid);
3791#if defined(_OBP) && defined(_KERNEL)
3792 if (config == NULL) {
3793 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3794 /* iscsi boot */
3795 get_iscsi_bootpath_phy(devpath);
3796 config = spa_generate_rootconf(devpath, devid, &guid);
3797 }
3798 }
3799#endif
3800 if (config == NULL) {
3801 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3802 devpath);
3803 return (SET_ERROR(EIO));
3804 }
3805
3806 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3807 &pname) == 0);
3808 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3809
3810 mutex_enter(&spa_namespace_lock);
3811 if ((spa = spa_lookup(pname)) != NULL) {
3812 /*
3813 * Remove the existing root pool from the namespace so that we
3814 * can replace it with the correct config we just read in.
3815 */
3816 spa_remove(spa);
3817 }
3818
3819 spa = spa_add(pname, config, NULL);
3820 spa->spa_is_root = B_TRUE;
3821 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3822
3823 /*
3824 * Build up a vdev tree based on the boot device's label config.
3825 */
3826 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3827 &nvtop) == 0);
3828 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3829 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3830 VDEV_ALLOC_ROOTPOOL);
3831 spa_config_exit(spa, SCL_ALL, FTAG);
3832 if (error) {
3833 mutex_exit(&spa_namespace_lock);
3834 nvlist_free(config);
3835 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3836 pname);
3837 return (error);
3838 }
3839
3840 /*
3841 * Get the boot vdev.
3842 */
3843 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3844 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3845 (u_longlong_t)guid);
3846 error = SET_ERROR(ENOENT);
3847 goto out;
3848 }
3849
3850 /*
3851 * Determine if there is a better boot device.
3852 */
3853 avd = bvd;
3854 spa_alt_rootvdev(rvd, &avd, &txg);
3855 if (avd != bvd) {
3856 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3857 "try booting from '%s'", avd->vdev_path);
3858 error = SET_ERROR(EINVAL);
3859 goto out;
3860 }
3861
3862 /*
3863 * If the boot device is part of a spare vdev then ensure that
3864 * we're booting off the active spare.
3865 */
3866 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3867 !bvd->vdev_isspare) {
3868 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3869 "try booting from '%s'",
3870 bvd->vdev_parent->
3871 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3872 error = SET_ERROR(EINVAL);
3873 goto out;
3874 }
3875
3876 error = 0;
3877out:
3878 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3879 vdev_free(rvd);
3880 spa_config_exit(spa, SCL_ALL, FTAG);
3881 mutex_exit(&spa_namespace_lock);
3882
3883 nvlist_free(config);
3884 return (error);
3885}
3886
3887#else
3888
3889extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
3890 uint64_t *count);
3891
3892static nvlist_t *
3893spa_generate_rootconf(const char *name)
3894{
3895 nvlist_t **configs, **tops;
3896 nvlist_t *config;
3897 nvlist_t *best_cfg, *nvtop, *nvroot;
3898 uint64_t *holes;
3899 uint64_t best_txg;
3900 uint64_t nchildren;
3901 uint64_t pgid;
3902 uint64_t count;
3903 uint64_t i;
3904 uint_t nholes;
3905
3906 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
3907 return (NULL);
3908
3909 ASSERT3U(count, !=, 0);
3910 best_txg = 0;
3911 for (i = 0; i < count; i++) {
3912 uint64_t txg;
3913
3914 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
3915 &txg) == 0);
3916 if (txg > best_txg) {
3917 best_txg = txg;
3918 best_cfg = configs[i];
3919 }
3920 }
3921
3922 /*
3923 * Multi-vdev root pool configuration discovery is not supported yet.
3924 */
3925 nchildren = 1;
3926 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
3927 holes = NULL;
3928 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
3929 &holes, &nholes);
3930
3931 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
3932 for (i = 0; i < nchildren; i++) {
3933 if (i >= count)
3934 break;
3935 if (configs[i] == NULL)
3936 continue;
3937 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
3938 &nvtop) == 0);
3939 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
3940 }
3941 for (i = 0; holes != NULL && i < nholes; i++) {
3942 if (i >= nchildren)
3943 continue;
3944 if (tops[holes[i]] != NULL)
3945 continue;
3946 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
3947 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
3948 VDEV_TYPE_HOLE) == 0);
3949 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
3950 holes[i]) == 0);
3951 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
3952 0) == 0);
3953 }
3954 for (i = 0; i < nchildren; i++) {
3955 if (tops[i] != NULL)
3956 continue;
3957 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
3958 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
3959 VDEV_TYPE_MISSING) == 0);
3960 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
3961 i) == 0);
3962 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
3963 0) == 0);
3964 }
3965
3966 /*
3967 * Create pool config based on the best vdev config.
3968 */
3969 nvlist_dup(best_cfg, &config, KM_SLEEP);
3970
3971 /*
3972 * Put this pool's top-level vdevs into a root vdev.
3973 */
3974 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3975 &pgid) == 0);
3976 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3977 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3978 VDEV_TYPE_ROOT) == 0);
3979 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3980 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3981 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3982 tops, nchildren) == 0);
3983
3984 /*
3985 * Replace the existing vdev_tree with the new root vdev in
3986 * this pool's configuration (remove the old, add the new).
3987 */
3988 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3989
3990 /*
3991 * Drop vdev config elements that should not be present at pool level.
3992 */
3993 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
3994 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
3995
3996 for (i = 0; i < count; i++)
3997 nvlist_free(configs[i]);
3998 kmem_free(configs, count * sizeof(void *));
3999 for (i = 0; i < nchildren; i++)
4000 nvlist_free(tops[i]);
4001 kmem_free(tops, nchildren * sizeof(void *));
4002 nvlist_free(nvroot);
4003 return (config);
4004}
4005
4006int
4007spa_import_rootpool(const char *name)
4008{
4009 spa_t *spa;
4010 vdev_t *rvd, *bvd, *avd = NULL;
4011 nvlist_t *config, *nvtop;
4012 uint64_t txg;
4013 char *pname;
4014 int error;
4015
4016 /*
4017 * Read the label from the boot device and generate a configuration.
4018 */
4019 config = spa_generate_rootconf(name);
4020
4021 mutex_enter(&spa_namespace_lock);
4022 if (config != NULL) {
4023 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4024 &pname) == 0 && strcmp(name, pname) == 0);
4025 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
4026 == 0);
4027
4028 if ((spa = spa_lookup(pname)) != NULL) {
4029 /*
4030 * Remove the existing root pool from the namespace so
4031 * that we can replace it with the correct config
4032 * we just read in.
4033 */
4034 spa_remove(spa);
4035 }
4036 spa = spa_add(pname, config, NULL);
4037
4038 /*
4039 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
4040 * via spa_version().
4041 */
4042 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4043 &spa->spa_ubsync.ub_version) != 0)
4044 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4045 } else if ((spa = spa_lookup(name)) == NULL) {
4046 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
4047 name);
4048 return (EIO);
4049 } else {
4050 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
4051 }
4052 spa->spa_is_root = B_TRUE;
4053 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4054
4055 /*
4056 * Build up a vdev tree based on the boot device's label config.
4057 */
4058 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4059 &nvtop) == 0);
4060 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4061 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4062 VDEV_ALLOC_ROOTPOOL);
4063 spa_config_exit(spa, SCL_ALL, FTAG);
4064 if (error) {
4065 mutex_exit(&spa_namespace_lock);
4066 nvlist_free(config);
4067 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
4068 pname);
4069 return (error);
4070 }
4071
4072 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4073 vdev_free(rvd);
4074 spa_config_exit(spa, SCL_ALL, FTAG);
4075 mutex_exit(&spa_namespace_lock);
4076
4077 nvlist_free(config);
4078 return (0);
4079}
4080
4081#endif /* sun */
4082#endif
4083
4084/*
4085 * Import a non-root pool into the system.
4086 */
4087int
4088spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4089{
4090 spa_t *spa;
4091 char *altroot = NULL;
4092 spa_load_state_t state = SPA_LOAD_IMPORT;
4093 zpool_rewind_policy_t policy;
4094 uint64_t mode = spa_mode_global;
4095 uint64_t readonly = B_FALSE;
4096 int error;
4097 nvlist_t *nvroot;
4098 nvlist_t **spares, **l2cache;
4099 uint_t nspares, nl2cache;
4100
4101 /*
4102 * If a pool with this name exists, return failure.
4103 */
4104 mutex_enter(&spa_namespace_lock);
4105 if (spa_lookup(pool) != NULL) {
4106 mutex_exit(&spa_namespace_lock);
4107 return (SET_ERROR(EEXIST));
4108 }
4109
4110 /*
4111 * Create and initialize the spa structure.
4112 */
4113 (void) nvlist_lookup_string(props,
4114 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4115 (void) nvlist_lookup_uint64(props,
4116 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4117 if (readonly)
4118 mode = FREAD;
4119 spa = spa_add(pool, config, altroot);
4120 spa->spa_import_flags = flags;
4121
4122 /*
4123 * Verbatim import - Take a pool and insert it into the namespace
4124 * as if it had been loaded at boot.
4125 */
4126 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4127 if (props != NULL)
4128 spa_configfile_set(spa, props, B_FALSE);
4129
4130 spa_config_sync(spa, B_FALSE, B_TRUE);
4131
4132 mutex_exit(&spa_namespace_lock);
4133 return (0);
4134 }
4135
4136 spa_activate(spa, mode);
4137
4138 /*
4139 * Don't start async tasks until we know everything is healthy.
4140 */
4141 spa_async_suspend(spa);
4142
4143 zpool_get_rewind_policy(config, &policy);
4144 if (policy.zrp_request & ZPOOL_DO_REWIND)
4145 state = SPA_LOAD_RECOVER;
4146
4147 /*
4148 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
4149 * because the user-supplied config is actually the one to trust when
4150 * doing an import.
4151 */
4152 if (state != SPA_LOAD_RECOVER)
4153 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4154
4155 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4156 policy.zrp_request);
4157
4158 /*
4159 * Propagate anything learned while loading the pool and pass it
4160 * back to caller (i.e. rewind info, missing devices, etc).
4161 */
4162 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4163 spa->spa_load_info) == 0);
4164
4165 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4166 /*
4167 * Toss any existing sparelist, as it doesn't have any validity
4168 * anymore, and conflicts with spa_has_spare().
4169 */
4170 if (spa->spa_spares.sav_config) {
4171 nvlist_free(spa->spa_spares.sav_config);
4172 spa->spa_spares.sav_config = NULL;
4173 spa_load_spares(spa);
4174 }
4175 if (spa->spa_l2cache.sav_config) {
4176 nvlist_free(spa->spa_l2cache.sav_config);
4177 spa->spa_l2cache.sav_config = NULL;
4178 spa_load_l2cache(spa);
4179 }
4180
4181 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4182 &nvroot) == 0);
4183 if (error == 0)
4184 error = spa_validate_aux(spa, nvroot, -1ULL,
4185 VDEV_ALLOC_SPARE);
4186 if (error == 0)
4187 error = spa_validate_aux(spa, nvroot, -1ULL,
4188 VDEV_ALLOC_L2CACHE);
4189 spa_config_exit(spa, SCL_ALL, FTAG);
4190
4191 if (props != NULL)
4192 spa_configfile_set(spa, props, B_FALSE);
4193
4194 if (error != 0 || (props && spa_writeable(spa) &&
4195 (error = spa_prop_set(spa, props)))) {
4196 spa_unload(spa);
4197 spa_deactivate(spa);
4198 spa_remove(spa);
4199 mutex_exit(&spa_namespace_lock);
4200 return (error);
4201 }
4202
4203 spa_async_resume(spa);
4204
4205 /*
4206 * Override any spares and level 2 cache devices as specified by
4207 * the user, as these may have correct device names/devids, etc.
4208 */
4209 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4210 &spares, &nspares) == 0) {
4211 if (spa->spa_spares.sav_config)
4212 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4213 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4214 else
4215 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4216 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4217 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4218 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4219 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4220 spa_load_spares(spa);
4221 spa_config_exit(spa, SCL_ALL, FTAG);
4222 spa->spa_spares.sav_sync = B_TRUE;
4223 }
4224 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4225 &l2cache, &nl2cache) == 0) {
4226 if (spa->spa_l2cache.sav_config)
4227 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4228 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4229 else
4230 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4231 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4232 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4233 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4234 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4235 spa_load_l2cache(spa);
4236 spa_config_exit(spa, SCL_ALL, FTAG);
4237 spa->spa_l2cache.sav_sync = B_TRUE;
4238 }
4239
4240 /*
4241 * Check for any removed devices.
4242 */
4243 if (spa->spa_autoreplace) {
4244 spa_aux_check_removed(&spa->spa_spares);
4245 spa_aux_check_removed(&spa->spa_l2cache);
4246 }
4247
4248 if (spa_writeable(spa)) {
4249 /*
4250 * Update the config cache to include the newly-imported pool.
4251 */
4252 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4253 }
4254
4255 /*
4256 * It's possible that the pool was expanded while it was exported.
4257 * We kick off an async task to handle this for us.
4258 */
4259 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4260
4261 mutex_exit(&spa_namespace_lock);
4262 spa_history_log_version(spa, "import");
4263
4264#ifdef __FreeBSD__
4265#ifdef _KERNEL
4266 zvol_create_minors(pool);
4267#endif
4268#endif
4269 return (0);
4270}
4271
4272nvlist_t *
4273spa_tryimport(nvlist_t *tryconfig)
4274{
4275 nvlist_t *config = NULL;
4276 char *poolname;
4277 spa_t *spa;
4278 uint64_t state;
4279 int error;
4280
4281 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4282 return (NULL);
4283
4284 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4285 return (NULL);
4286
4287 /*
4288 * Create and initialize the spa structure.
4289 */
4290 mutex_enter(&spa_namespace_lock);
4291 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4292 spa_activate(spa, FREAD);
4293
4294 /*
4295 * Pass off the heavy lifting to spa_load().
4296 * Pass TRUE for mosconfig because the user-supplied config
4297 * is actually the one to trust when doing an import.
4298 */
4299 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4300
4301 /*
4302 * If 'tryconfig' was at least parsable, return the current config.
4303 */
4304 if (spa->spa_root_vdev != NULL) {
4305 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4306 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4307 poolname) == 0);
4308 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4309 state) == 0);
4310 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4311 spa->spa_uberblock.ub_timestamp) == 0);
4312 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4313 spa->spa_load_info) == 0);
4314
4315 /*
4316 * If the bootfs property exists on this pool then we
4317 * copy it out so that external consumers can tell which
4318 * pools are bootable.
4319 */
4320 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4321 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4322
4323 /*
4324 * We have to play games with the name since the
4325 * pool was opened as TRYIMPORT_NAME.
4326 */
4327 if (dsl_dsobj_to_dsname(spa_name(spa),
4328 spa->spa_bootfs, tmpname) == 0) {
4329 char *cp;
4330 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4331
4332 cp = strchr(tmpname, '/');
4333 if (cp == NULL) {
4334 (void) strlcpy(dsname, tmpname,
4335 MAXPATHLEN);
4336 } else {
4337 (void) snprintf(dsname, MAXPATHLEN,
4338 "%s/%s", poolname, ++cp);
4339 }
4340 VERIFY(nvlist_add_string(config,
4341 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4342 kmem_free(dsname, MAXPATHLEN);
4343 }
4344 kmem_free(tmpname, MAXPATHLEN);
4345 }
4346
4347 /*
4348 * Add the list of hot spares and level 2 cache devices.
4349 */
4350 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4351 spa_add_spares(spa, config);
4352 spa_add_l2cache(spa, config);
4353 spa_config_exit(spa, SCL_CONFIG, FTAG);
4354 }
4355
4356 spa_unload(spa);
4357 spa_deactivate(spa);
4358 spa_remove(spa);
4359 mutex_exit(&spa_namespace_lock);
4360
4361 return (config);
4362}
4363
4364/*
4365 * Pool export/destroy
4366 *
4367 * The act of destroying or exporting a pool is very simple. We make sure there
4368 * is no more pending I/O and any references to the pool are gone. Then, we
4369 * update the pool state and sync all the labels to disk, removing the
4370 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4371 * we don't sync the labels or remove the configuration cache.
4372 */
4373static int
4374spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4375 boolean_t force, boolean_t hardforce)
4376{
4377 spa_t *spa;
4378
4379 if (oldconfig)
4380 *oldconfig = NULL;
4381
4382 if (!(spa_mode_global & FWRITE))
4383 return (SET_ERROR(EROFS));
4384
4385 mutex_enter(&spa_namespace_lock);
4386 if ((spa = spa_lookup(pool)) == NULL) {
4387 mutex_exit(&spa_namespace_lock);
4388 return (SET_ERROR(ENOENT));
4389 }
4390
4391 /*
4392 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4393 * reacquire the namespace lock, and see if we can export.
4394 */
4395 spa_open_ref(spa, FTAG);
4396 mutex_exit(&spa_namespace_lock);
4397 spa_async_suspend(spa);
4398 mutex_enter(&spa_namespace_lock);
4399 spa_close(spa, FTAG);
4400
4401 /*
4402 * The pool will be in core if it's openable,
4403 * in which case we can modify its state.
4404 */
4405 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4406 /*
4407 * Objsets may be open only because they're dirty, so we
4408 * have to force it to sync before checking spa_refcnt.
4409 */
4410 txg_wait_synced(spa->spa_dsl_pool, 0);
4411
4412 /*
4413 * A pool cannot be exported or destroyed if there are active
4414 * references. If we are resetting a pool, allow references by
4415 * fault injection handlers.
4416 */
4417 if (!spa_refcount_zero(spa) ||
4418 (spa->spa_inject_ref != 0 &&
4419 new_state != POOL_STATE_UNINITIALIZED)) {
4420 spa_async_resume(spa);
4421 mutex_exit(&spa_namespace_lock);
4422 return (SET_ERROR(EBUSY));
4423 }
4424
4425 /*
4426 * A pool cannot be exported if it has an active shared spare.
4427 * This is to prevent other pools stealing the active spare
4428 * from an exported pool. At user's own will, such pool can
4429 * be forcedly exported.
4430 */
4431 if (!force && new_state == POOL_STATE_EXPORTED &&
4432 spa_has_active_shared_spare(spa)) {
4433 spa_async_resume(spa);
4434 mutex_exit(&spa_namespace_lock);
4435 return (SET_ERROR(EXDEV));
4436 }
4437
4438 /*
4439 * We want this to be reflected on every label,
4440 * so mark them all dirty. spa_unload() will do the
4441 * final sync that pushes these changes out.
4442 */
4443 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4444 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4445 spa->spa_state = new_state;
4446 spa->spa_final_txg = spa_last_synced_txg(spa) +
4447 TXG_DEFER_SIZE + 1;
4448 vdev_config_dirty(spa->spa_root_vdev);
4449 spa_config_exit(spa, SCL_ALL, FTAG);
4450 }
4451 }
4452
4453 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4454
4455 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4456 spa_unload(spa);
4457 spa_deactivate(spa);
4458 }
4459
4460 if (oldconfig && spa->spa_config)
4461 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4462
4463 if (new_state != POOL_STATE_UNINITIALIZED) {
4464 if (!hardforce)
4465 spa_config_sync(spa, B_TRUE, B_TRUE);
4466 spa_remove(spa);
4467 }
4468 mutex_exit(&spa_namespace_lock);
4469
4470 return (0);
4471}
4472
4473/*
4474 * Destroy a storage pool.
4475 */
4476int
4477spa_destroy(char *pool)
4478{
4479 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4480 B_FALSE, B_FALSE));
4481}
4482
4483/*
4484 * Export a storage pool.
4485 */
4486int
4487spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4488 boolean_t hardforce)
4489{
4490 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4491 force, hardforce));
4492}
4493
4494/*
4495 * Similar to spa_export(), this unloads the spa_t without actually removing it
4496 * from the namespace in any way.
4497 */
4498int
4499spa_reset(char *pool)
4500{
4501 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4502 B_FALSE, B_FALSE));
4503}
4504
4505/*
4506 * ==========================================================================
4507 * Device manipulation
4508 * ==========================================================================
4509 */
4510
4511/*
4512 * Add a device to a storage pool.
4513 */
4514int
4515spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4516{
4517 uint64_t txg, id;
4518 int error;
4519 vdev_t *rvd = spa->spa_root_vdev;
4520 vdev_t *vd, *tvd;
4521 nvlist_t **spares, **l2cache;
4522 uint_t nspares, nl2cache;
4523
4524 ASSERT(spa_writeable(spa));
4525
4526 txg = spa_vdev_enter(spa);
4527
4528 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4529 VDEV_ALLOC_ADD)) != 0)
4530 return (spa_vdev_exit(spa, NULL, txg, error));
4531
4532 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
4533
4534 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4535 &nspares) != 0)
4536 nspares = 0;
4537
4538 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4539 &nl2cache) != 0)
4540 nl2cache = 0;
4541
4542 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4543 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4544
4545 if (vd->vdev_children != 0 &&
4546 (error = vdev_create(vd, txg, B_FALSE)) != 0)
4547 return (spa_vdev_exit(spa, vd, txg, error));
4548
4549 /*
4550 * We must validate the spares and l2cache devices after checking the
4551 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
4552 */
4553 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4554 return (spa_vdev_exit(spa, vd, txg, error));
4555
4556 /*
4557 * Transfer each new top-level vdev from vd to rvd.
4558 */
4559 for (int c = 0; c < vd->vdev_children; c++) {
4560
4561 /*
4562 * Set the vdev id to the first hole, if one exists.
4563 */
4564 for (id = 0; id < rvd->vdev_children; id++) {
4565 if (rvd->vdev_child[id]->vdev_ishole) {
4566 vdev_free(rvd->vdev_child[id]);
4567 break;
4568 }
4569 }
4570 tvd = vd->vdev_child[c];
4571 vdev_remove_child(vd, tvd);
4572 tvd->vdev_id = id;
4573 vdev_add_child(rvd, tvd);
4574 vdev_config_dirty(tvd);
4575 }
4576
4577 if (nspares != 0) {
4578 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4579 ZPOOL_CONFIG_SPARES);
4580 spa_load_spares(spa);
4581 spa->spa_spares.sav_sync = B_TRUE;
4582 }
4583
4584 if (nl2cache != 0) {
4585 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4586 ZPOOL_CONFIG_L2CACHE);
4587 spa_load_l2cache(spa);
4588 spa->spa_l2cache.sav_sync = B_TRUE;
4589 }
4590
4591 /*
4592 * We have to be careful when adding new vdevs to an existing pool.
4593 * If other threads start allocating from these vdevs before we
4594 * sync the config cache, and we lose power, then upon reboot we may
4595 * fail to open the pool because there are DVAs that the config cache
4596 * can't translate. Therefore, we first add the vdevs without
4597 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4598 * and then let spa_config_update() initialize the new metaslabs.
4599 *
4600 * spa_load() checks for added-but-not-initialized vdevs, so that
4601 * if we lose power at any point in this sequence, the remaining
4602 * steps will be completed the next time we load the pool.
4603 */
4604 (void) spa_vdev_exit(spa, vd, txg, 0);
4605
4606 mutex_enter(&spa_namespace_lock);
4607 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4608 mutex_exit(&spa_namespace_lock);
4609
4610 return (0);
4611}
4612
4613/*
4614 * Attach a device to a mirror. The arguments are the path to any device
4615 * in the mirror, and the nvroot for the new device. If the path specifies
4616 * a device that is not mirrored, we automatically insert the mirror vdev.
4617 *
4618 * If 'replacing' is specified, the new device is intended to replace the
4619 * existing device; in this case the two devices are made into their own
4620 * mirror using the 'replacing' vdev, which is functionally identical to
4621 * the mirror vdev (it actually reuses all the same ops) but has a few
4622 * extra rules: you can't attach to it after it's been created, and upon
4623 * completion of resilvering, the first disk (the one being replaced)
4624 * is automatically detached.
4625 */
4626int
4627spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4628{
4629 uint64_t txg, dtl_max_txg;
4630 vdev_t *rvd = spa->spa_root_vdev;
4631 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4632 vdev_ops_t *pvops;
4633 char *oldvdpath, *newvdpath;
4634 int newvd_isspare;
4635 int error;
4636
4637 ASSERT(spa_writeable(spa));
4638
4639 txg = spa_vdev_enter(spa);
4640
4641 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4642
4643 if (oldvd == NULL)
4644 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4645
4646 if (!oldvd->vdev_ops->vdev_op_leaf)
4647 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4648
4649 pvd = oldvd->vdev_parent;
4650
4651 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4652 VDEV_ALLOC_ATTACH)) != 0)
4653 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4654
4655 if (newrootvd->vdev_children != 1)
4656 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4657
4658 newvd = newrootvd->vdev_child[0];
4659
4660 if (!newvd->vdev_ops->vdev_op_leaf)
4661 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4662
4663 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4664 return (spa_vdev_exit(spa, newrootvd, txg, error));
4665
4666 /*
4667 * Spares can't replace logs
4668 */
4669 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4670 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4671
4672 if (!replacing) {
4673 /*
4674 * For attach, the only allowable parent is a mirror or the root
4675 * vdev.
4676 */
4677 if (pvd->vdev_ops != &vdev_mirror_ops &&
4678 pvd->vdev_ops != &vdev_root_ops)
4679 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4680
4681 pvops = &vdev_mirror_ops;
4682 } else {
4683 /*
4684 * Active hot spares can only be replaced by inactive hot
4685 * spares.
4686 */
4687 if (pvd->vdev_ops == &vdev_spare_ops &&
4688 oldvd->vdev_isspare &&
4689 !spa_has_spare(spa, newvd->vdev_guid))
4690 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4691
4692 /*
4693 * If the source is a hot spare, and the parent isn't already a
4694 * spare, then we want to create a new hot spare. Otherwise, we
4695 * want to create a replacing vdev. The user is not allowed to
4696 * attach to a spared vdev child unless the 'isspare' state is
4697 * the same (spare replaces spare, non-spare replaces
4698 * non-spare).
4699 */
4700 if (pvd->vdev_ops == &vdev_replacing_ops &&
4701 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4702 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4703 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4704 newvd->vdev_isspare != oldvd->vdev_isspare) {
4705 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4706 }
4707
4708 if (newvd->vdev_isspare)
4709 pvops = &vdev_spare_ops;
4710 else
4711 pvops = &vdev_replacing_ops;
4712 }
4713
4714 /*
4715 * Make sure the new device is big enough.
4716 */
4717 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4718 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4719
4720 /*
4721 * The new device cannot have a higher alignment requirement
4722 * than the top-level vdev.
4723 */
4724 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4725 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4726
4727 /*
4728 * If this is an in-place replacement, update oldvd's path and devid
4729 * to make it distinguishable from newvd, and unopenable from now on.
4730 */
4731 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4732 spa_strfree(oldvd->vdev_path);
4733 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4734 KM_SLEEP);
4735 (void) sprintf(oldvd->vdev_path, "%s/%s",
4736 newvd->vdev_path, "old");
4737 if (oldvd->vdev_devid != NULL) {
4738 spa_strfree(oldvd->vdev_devid);
4739 oldvd->vdev_devid = NULL;
4740 }
4741 }
4742
4743 /* mark the device being resilvered */
4744 newvd->vdev_resilver_txg = txg;
4745
4746 /*
4747 * If the parent is not a mirror, or if we're replacing, insert the new
4748 * mirror/replacing/spare vdev above oldvd.
4749 */
4750 if (pvd->vdev_ops != pvops)
4751 pvd = vdev_add_parent(oldvd, pvops);
4752
4753 ASSERT(pvd->vdev_top->vdev_parent == rvd);
4754 ASSERT(pvd->vdev_ops == pvops);
4755 ASSERT(oldvd->vdev_parent == pvd);
4756
4757 /*
4758 * Extract the new device from its root and add it to pvd.
4759 */
4760 vdev_remove_child(newrootvd, newvd);
4761 newvd->vdev_id = pvd->vdev_children;
4762 newvd->vdev_crtxg = oldvd->vdev_crtxg;
4763 vdev_add_child(pvd, newvd);
4764
4765 tvd = newvd->vdev_top;
4766 ASSERT(pvd->vdev_top == tvd);
4767 ASSERT(tvd->vdev_parent == rvd);
4768
4769 vdev_config_dirty(tvd);
4770
4771 /*
4772 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4773 * for any dmu_sync-ed blocks. It will propagate upward when
4774 * spa_vdev_exit() calls vdev_dtl_reassess().
4775 */
4776 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4777
4778 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4779 dtl_max_txg - TXG_INITIAL);
4780
4781 if (newvd->vdev_isspare) {
4782 spa_spare_activate(newvd);
4783 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4784 }
4785
4786 oldvdpath = spa_strdup(oldvd->vdev_path);
4787 newvdpath = spa_strdup(newvd->vdev_path);
4788 newvd_isspare = newvd->vdev_isspare;
4789
4790 /*
4791 * Mark newvd's DTL dirty in this txg.
4792 */
4793 vdev_dirty(tvd, VDD_DTL, newvd, txg);
4794
4795 /*
4796 * Schedule the resilver to restart in the future. We do this to
4797 * ensure that dmu_sync-ed blocks have been stitched into the
4798 * respective datasets.
4799 */
4800 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4801
4802 /*
4803 * Commit the config
4804 */
4805 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4806
4807 spa_history_log_internal(spa, "vdev attach", NULL,
4808 "%s vdev=%s %s vdev=%s",
4809 replacing && newvd_isspare ? "spare in" :
4810 replacing ? "replace" : "attach", newvdpath,
4811 replacing ? "for" : "to", oldvdpath);
4812
4813 spa_strfree(oldvdpath);
4814 spa_strfree(newvdpath);
4815
4816 if (spa->spa_bootfs)
4817 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4818
4819 return (0);
4820}
4821
4822/*
4823 * Detach a device from a mirror or replacing vdev.
4824 *
4825 * If 'replace_done' is specified, only detach if the parent
4826 * is a replacing vdev.
4827 */
4828int
4829spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4830{
4831 uint64_t txg;
4832 int error;
4833 vdev_t *rvd = spa->spa_root_vdev;
4834 vdev_t *vd, *pvd, *cvd, *tvd;
4835 boolean_t unspare = B_FALSE;
4836 uint64_t unspare_guid = 0;
4837 char *vdpath;
4838
4839 ASSERT(spa_writeable(spa));
4840
4841 txg = spa_vdev_enter(spa);
4842
4843 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4844
4845 if (vd == NULL)
4846 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4847
4848 if (!vd->vdev_ops->vdev_op_leaf)
4849 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4850
4851 pvd = vd->vdev_parent;
4852
4853 /*
4854 * If the parent/child relationship is not as expected, don't do it.
4855 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4856 * vdev that's replacing B with C. The user's intent in replacing
4857 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4858 * the replace by detaching C, the expected behavior is to end up
4859 * M(A,B). But suppose that right after deciding to detach C,
4860 * the replacement of B completes. We would have M(A,C), and then
4861 * ask to detach C, which would leave us with just A -- not what
4862 * the user wanted. To prevent this, we make sure that the
4863 * parent/child relationship hasn't changed -- in this example,
4864 * that C's parent is still the replacing vdev R.
4865 */
4866 if (pvd->vdev_guid != pguid && pguid != 0)
4867 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4868
4869 /*
4870 * Only 'replacing' or 'spare' vdevs can be replaced.
4871 */
4872 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4873 pvd->vdev_ops != &vdev_spare_ops)
4874 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4875
4876 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4877 spa_version(spa) >= SPA_VERSION_SPARES);
4878
4879 /*
4880 * Only mirror, replacing, and spare vdevs support detach.
4881 */
4882 if (pvd->vdev_ops != &vdev_replacing_ops &&
4883 pvd->vdev_ops != &vdev_mirror_ops &&
4884 pvd->vdev_ops != &vdev_spare_ops)
4885 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4886
4887 /*
4888 * If this device has the only valid copy of some data,
4889 * we cannot safely detach it.
4890 */
4891 if (vdev_dtl_required(vd))
4892 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4893
4894 ASSERT(pvd->vdev_children >= 2);
4895
4896 /*
4897 * If we are detaching the second disk from a replacing vdev, then
4898 * check to see if we changed the original vdev's path to have "/old"
4899 * at the end in spa_vdev_attach(). If so, undo that change now.
4900 */
4901 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4902 vd->vdev_path != NULL) {
4903 size_t len = strlen(vd->vdev_path);
4904
4905 for (int c = 0; c < pvd->vdev_children; c++) {
4906 cvd = pvd->vdev_child[c];
4907
4908 if (cvd == vd || cvd->vdev_path == NULL)
4909 continue;
4910
4911 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4912 strcmp(cvd->vdev_path + len, "/old") == 0) {
4913 spa_strfree(cvd->vdev_path);
4914 cvd->vdev_path = spa_strdup(vd->vdev_path);
4915 break;
4916 }
4917 }
4918 }
4919
4920 /*
4921 * If we are detaching the original disk from a spare, then it implies
4922 * that the spare should become a real disk, and be removed from the
4923 * active spare list for the pool.
4924 */
4925 if (pvd->vdev_ops == &vdev_spare_ops &&
4926 vd->vdev_id == 0 &&
4927 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4928 unspare = B_TRUE;
4929
4930 /*
4931 * Erase the disk labels so the disk can be used for other things.
4932 * This must be done after all other error cases are handled,
4933 * but before we disembowel vd (so we can still do I/O to it).
4934 * But if we can't do it, don't treat the error as fatal --
4935 * it may be that the unwritability of the disk is the reason
4936 * it's being detached!
4937 */
4938 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4939
4940 /*
4941 * Remove vd from its parent and compact the parent's children.
4942 */
4943 vdev_remove_child(pvd, vd);
4944 vdev_compact_children(pvd);
4945
4946 /*
4947 * Remember one of the remaining children so we can get tvd below.
4948 */
4949 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4950
4951 /*
4952 * If we need to remove the remaining child from the list of hot spares,
4953 * do it now, marking the vdev as no longer a spare in the process.
4954 * We must do this before vdev_remove_parent(), because that can
4955 * change the GUID if it creates a new toplevel GUID. For a similar
4956 * reason, we must remove the spare now, in the same txg as the detach;
4957 * otherwise someone could attach a new sibling, change the GUID, and
4958 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4959 */
4960 if (unspare) {
4961 ASSERT(cvd->vdev_isspare);
4962 spa_spare_remove(cvd);
4963 unspare_guid = cvd->vdev_guid;
4964 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4965 cvd->vdev_unspare = B_TRUE;
4966 }
4967
4968 /*
4969 * If the parent mirror/replacing vdev only has one child,
4970 * the parent is no longer needed. Remove it from the tree.
4971 */
4972 if (pvd->vdev_children == 1) {
4973 if (pvd->vdev_ops == &vdev_spare_ops)
4974 cvd->vdev_unspare = B_FALSE;
4975 vdev_remove_parent(cvd);
4976 }
4977
4978
4979 /*
4980 * We don't set tvd until now because the parent we just removed
4981 * may have been the previous top-level vdev.
4982 */
4983 tvd = cvd->vdev_top;
4984 ASSERT(tvd->vdev_parent == rvd);
4985
4986 /*
4987 * Reevaluate the parent vdev state.
4988 */
4989 vdev_propagate_state(cvd);
4990
4991 /*
4992 * If the 'autoexpand' property is set on the pool then automatically
4993 * try to expand the size of the pool. For example if the device we
4994 * just detached was smaller than the others, it may be possible to
4995 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4996 * first so that we can obtain the updated sizes of the leaf vdevs.
4997 */
4998 if (spa->spa_autoexpand) {
4999 vdev_reopen(tvd);
5000 vdev_expand(tvd, txg);
5001 }
5002
5003 vdev_config_dirty(tvd);
5004
5005 /*
5006 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
5007 * vd->vdev_detached is set and free vd's DTL object in syncing context.
5008 * But first make sure we're not on any *other* txg's DTL list, to
5009 * prevent vd from being accessed after it's freed.
5010 */
5011 vdpath = spa_strdup(vd->vdev_path);
5012 for (int t = 0; t < TXG_SIZE; t++)
5013 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
5014 vd->vdev_detached = B_TRUE;
5015 vdev_dirty(tvd, VDD_DTL, vd, txg);
5016
5017 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
5018
5019 /* hang on to the spa before we release the lock */
5020 spa_open_ref(spa, FTAG);
5021
5022 error = spa_vdev_exit(spa, vd, txg, 0);
5023
5024 spa_history_log_internal(spa, "detach", NULL,
5025 "vdev=%s", vdpath);
5026 spa_strfree(vdpath);
5027
5028 /*
5029 * If this was the removal of the original device in a hot spare vdev,
5030 * then we want to go through and remove the device from the hot spare
5031 * list of every other pool.
5032 */
5033 if (unspare) {
5034 spa_t *altspa = NULL;
5035
5036 mutex_enter(&spa_namespace_lock);
5037 while ((altspa = spa_next(altspa)) != NULL) {
5038 if (altspa->spa_state != POOL_STATE_ACTIVE ||
5039 altspa == spa)
5040 continue;
5041
5042 spa_open_ref(altspa, FTAG);
5043 mutex_exit(&spa_namespace_lock);
5044 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5045 mutex_enter(&spa_namespace_lock);
5046 spa_close(altspa, FTAG);
5047 }
5048 mutex_exit(&spa_namespace_lock);
5049
5050 /* search the rest of the vdevs for spares to remove */
5051 spa_vdev_resilver_done(spa);
5052 }
5053
5054 /* all done with the spa; OK to release */
5055 mutex_enter(&spa_namespace_lock);
5056 spa_close(spa, FTAG);
5057 mutex_exit(&spa_namespace_lock);
5058
5059 return (error);
5060}
5061
5062/*
5063 * Split a set of devices from their mirrors, and create a new pool from them.
5064 */
5065int
5066spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5067 nvlist_t *props, boolean_t exp)
5068{
5069 int error = 0;
5070 uint64_t txg, *glist;
5071 spa_t *newspa;
5072 uint_t c, children, lastlog;
5073 nvlist_t **child, *nvl, *tmp;
5074 dmu_tx_t *tx;
5075 char *altroot = NULL;
5076 vdev_t *rvd, **vml = NULL; /* vdev modify list */
5077 boolean_t activate_slog;
5078
5079 ASSERT(spa_writeable(spa));
5080
5081 txg = spa_vdev_enter(spa);
5082
5083 /* clear the log and flush everything up to now */
5084 activate_slog = spa_passivate_log(spa);
5085 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5086 error = spa_offline_log(spa);
5087 txg = spa_vdev_config_enter(spa);
5088
5089 if (activate_slog)
5090 spa_activate_log(spa);
5091
5092 if (error != 0)
5093 return (spa_vdev_exit(spa, NULL, txg, error));
5094
5095 /* check new spa name before going any further */
5096 if (spa_lookup(newname) != NULL)
5097 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5098
5099 /*
5100 * scan through all the children to ensure they're all mirrors
5101 */
5102 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5103 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5104 &children) != 0)
5105 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5106
5107 /* first, check to ensure we've got the right child count */
5108 rvd = spa->spa_root_vdev;
5109 lastlog = 0;
5110 for (c = 0; c < rvd->vdev_children; c++) {
5111 vdev_t *vd = rvd->vdev_child[c];
5112
5113 /* don't count the holes & logs as children */
5114 if (vd->vdev_islog || vd->vdev_ishole) {
5115 if (lastlog == 0)
5116 lastlog = c;
5117 continue;
5118 }
5119
5120 lastlog = 0;
5121 }
5122 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5123 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5124
5125 /* next, ensure no spare or cache devices are part of the split */
5126 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5127 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5128 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5129
5130 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5131 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5132
5133 /* then, loop over each vdev and validate it */
5134 for (c = 0; c < children; c++) {
5135 uint64_t is_hole = 0;
5136
5137 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5138 &is_hole);
5139
5140 if (is_hole != 0) {
5141 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5142 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5143 continue;
5144 } else {
5145 error = SET_ERROR(EINVAL);
5146 break;
5147 }
5148 }
5149
5150 /* which disk is going to be split? */
5151 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5152 &glist[c]) != 0) {
5153 error = SET_ERROR(EINVAL);
5154 break;
5155 }
5156
5157 /* look it up in the spa */
5158 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5159 if (vml[c] == NULL) {
5160 error = SET_ERROR(ENODEV);
5161 break;
5162 }
5163
5164 /* make sure there's nothing stopping the split */
5165 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5166 vml[c]->vdev_islog ||
5167 vml[c]->vdev_ishole ||
5168 vml[c]->vdev_isspare ||
5169 vml[c]->vdev_isl2cache ||
5170 !vdev_writeable(vml[c]) ||
5171 vml[c]->vdev_children != 0 ||
5172 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5173 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5174 error = SET_ERROR(EINVAL);
5175 break;
5176 }
5177
5178 if (vdev_dtl_required(vml[c])) {
5179 error = SET_ERROR(EBUSY);
5180 break;
5181 }
5182
5183 /* we need certain info from the top level */
5184 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5185 vml[c]->vdev_top->vdev_ms_array) == 0);
5186 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5187 vml[c]->vdev_top->vdev_ms_shift) == 0);
5188 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5189 vml[c]->vdev_top->vdev_asize) == 0);
5190 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5191 vml[c]->vdev_top->vdev_ashift) == 0);
5192 }
5193
5194 if (error != 0) {
5195 kmem_free(vml, children * sizeof (vdev_t *));
5196 kmem_free(glist, children * sizeof (uint64_t));
5197 return (spa_vdev_exit(spa, NULL, txg, error));
5198 }
5199
5200 /* stop writers from using the disks */
5201 for (c = 0; c < children; c++) {
5202 if (vml[c] != NULL)
5203 vml[c]->vdev_offline = B_TRUE;
5204 }
5205 vdev_reopen(spa->spa_root_vdev);
5206
5207 /*
5208 * Temporarily record the splitting vdevs in the spa config. This
5209 * will disappear once the config is regenerated.
5210 */
5211 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5212 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5213 glist, children) == 0);
5214 kmem_free(glist, children * sizeof (uint64_t));
5215
5216 mutex_enter(&spa->spa_props_lock);
5217 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5218 nvl) == 0);
5219 mutex_exit(&spa->spa_props_lock);
5220 spa->spa_config_splitting = nvl;
5221 vdev_config_dirty(spa->spa_root_vdev);
5222
5223 /* configure and create the new pool */
5224 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5225 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5226 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5227 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5228 spa_version(spa)) == 0);
5229 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5230 spa->spa_config_txg) == 0);
5231 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5232 spa_generate_guid(NULL)) == 0);
5233 (void) nvlist_lookup_string(props,
5234 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5235
5236 /* add the new pool to the namespace */
5237 newspa = spa_add(newname, config, altroot);
5238 newspa->spa_config_txg = spa->spa_config_txg;
5239 spa_set_log_state(newspa, SPA_LOG_CLEAR);
5240
5241 /* release the spa config lock, retaining the namespace lock */
5242 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5243
5244 if (zio_injection_enabled)
5245 zio_handle_panic_injection(spa, FTAG, 1);
5246
5247 spa_activate(newspa, spa_mode_global);
5248 spa_async_suspend(newspa);
5249
5250#ifndef sun
5251 /* mark that we are creating new spa by splitting */
5252 newspa->spa_splitting_newspa = B_TRUE;
5253#endif
5254 /* create the new pool from the disks of the original pool */
5255 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5256#ifndef sun
5257 newspa->spa_splitting_newspa = B_FALSE;
5258#endif
5259 if (error)
5260 goto out;
5261
5262 /* if that worked, generate a real config for the new pool */
5263 if (newspa->spa_root_vdev != NULL) {
5264 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5265 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5266 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5267 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5268 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5269 B_TRUE));
5270 }
5271
5272 /* set the props */
5273 if (props != NULL) {
5274 spa_configfile_set(newspa, props, B_FALSE);
5275 error = spa_prop_set(newspa, props);
5276 if (error)
5277 goto out;
5278 }
5279
5280 /* flush everything */
5281 txg = spa_vdev_config_enter(newspa);
5282 vdev_config_dirty(newspa->spa_root_vdev);
5283 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5284
5285 if (zio_injection_enabled)
5286 zio_handle_panic_injection(spa, FTAG, 2);
5287
5288 spa_async_resume(newspa);
5289
5290 /* finally, update the original pool's config */
5291 txg = spa_vdev_config_enter(spa);
5292 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5293 error = dmu_tx_assign(tx, TXG_WAIT);
5294 if (error != 0)
5295 dmu_tx_abort(tx);
5296 for (c = 0; c < children; c++) {
5297 if (vml[c] != NULL) {
5298 vdev_split(vml[c]);
5299 if (error == 0)
5300 spa_history_log_internal(spa, "detach", tx,
5301 "vdev=%s", vml[c]->vdev_path);
5302 vdev_free(vml[c]);
5303 }
5304 }
5305 vdev_config_dirty(spa->spa_root_vdev);
5306 spa->spa_config_splitting = NULL;
5307 nvlist_free(nvl);
5308 if (error == 0)
5309 dmu_tx_commit(tx);
5310 (void) spa_vdev_exit(spa, NULL, txg, 0);
5311
5312 if (zio_injection_enabled)
5313 zio_handle_panic_injection(spa, FTAG, 3);
5314
5315 /* split is complete; log a history record */
5316 spa_history_log_internal(newspa, "split", NULL,
5317 "from pool %s", spa_name(spa));
5318
5319 kmem_free(vml, children * sizeof (vdev_t *));
5320
5321 /* if we're not going to mount the filesystems in userland, export */
5322 if (exp)
5323 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5324 B_FALSE, B_FALSE);
5325
5326 return (error);
5327
5328out:
5329 spa_unload(newspa);
5330 spa_deactivate(newspa);
5331 spa_remove(newspa);
5332
5333 txg = spa_vdev_config_enter(spa);
5334
5335 /* re-online all offlined disks */
5336 for (c = 0; c < children; c++) {
5337 if (vml[c] != NULL)
5338 vml[c]->vdev_offline = B_FALSE;
5339 }
5340 vdev_reopen(spa->spa_root_vdev);
5341
5342 nvlist_free(spa->spa_config_splitting);
5343 spa->spa_config_splitting = NULL;
5344 (void) spa_vdev_exit(spa, NULL, txg, error);
5345
5346 kmem_free(vml, children * sizeof (vdev_t *));
5347 return (error);
5348}
5349
5350static nvlist_t *
5351spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5352{
5353 for (int i = 0; i < count; i++) {
5354 uint64_t guid;
5355
5356 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5357 &guid) == 0);
5358
5359 if (guid == target_guid)
5360 return (nvpp[i]);
5361 }
5362
5363 return (NULL);
5364}
5365
5366static void
5367spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5368 nvlist_t *dev_to_remove)
5369{
5370 nvlist_t **newdev = NULL;
5371
5372 if (count > 1)
5373 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5374
5375 for (int i = 0, j = 0; i < count; i++) {
5376 if (dev[i] == dev_to_remove)
5377 continue;
5378 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5379 }
5380
5381 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5382 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5383
5384 for (int i = 0; i < count - 1; i++)
5385 nvlist_free(newdev[i]);
5386
5387 if (count > 1)
5388 kmem_free(newdev, (count - 1) * sizeof (void *));
5389}
5390
5391/*
5392 * Evacuate the device.
5393 */
5394static int
5395spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5396{
5397 uint64_t txg;
5398 int error = 0;
5399
5400 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5401 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5402 ASSERT(vd == vd->vdev_top);
5403
5404 /*
5405 * Evacuate the device. We don't hold the config lock as writer
5406 * since we need to do I/O but we do keep the
5407 * spa_namespace_lock held. Once this completes the device
5408 * should no longer have any blocks allocated on it.
5409 */
5410 if (vd->vdev_islog) {
5411 if (vd->vdev_stat.vs_alloc != 0)
5412 error = spa_offline_log(spa);
5413 } else {
5414 error = SET_ERROR(ENOTSUP);
5415 }
5416
5417 if (error)
5418 return (error);
5419
5420 /*
5421 * The evacuation succeeded. Remove any remaining MOS metadata
5422 * associated with this vdev, and wait for these changes to sync.
5423 */
5424 ASSERT0(vd->vdev_stat.vs_alloc);
5425 txg = spa_vdev_config_enter(spa);
5426 vd->vdev_removing = B_TRUE;
5427 vdev_dirty_leaves(vd, VDD_DTL, txg);
5428 vdev_config_dirty(vd);
5429 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5430
5431 return (0);
5432}
5433
5434/*
5435 * Complete the removal by cleaning up the namespace.
5436 */
5437static void
5438spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5439{
5440 vdev_t *rvd = spa->spa_root_vdev;
5441 uint64_t id = vd->vdev_id;
5442 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5443
5444 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5445 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5446 ASSERT(vd == vd->vdev_top);
5447
5448 /*
5449 * Only remove any devices which are empty.
5450 */
5451 if (vd->vdev_stat.vs_alloc != 0)
5452 return;
5453
5454 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5455
5456 if (list_link_active(&vd->vdev_state_dirty_node))
5457 vdev_state_clean(vd);
5458 if (list_link_active(&vd->vdev_config_dirty_node))
5459 vdev_config_clean(vd);
5460
5461 vdev_free(vd);
5462
5463 if (last_vdev) {
5464 vdev_compact_children(rvd);
5465 } else {
5466 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5467 vdev_add_child(rvd, vd);
5468 }
5469 vdev_config_dirty(rvd);
5470
5471 /*
5472 * Reassess the health of our root vdev.
5473 */
5474 vdev_reopen(rvd);
5475}
5476
5477/*
5478 * Remove a device from the pool -
5479 *
5480 * Removing a device from the vdev namespace requires several steps
5481 * and can take a significant amount of time. As a result we use
5482 * the spa_vdev_config_[enter/exit] functions which allow us to
5483 * grab and release the spa_config_lock while still holding the namespace
5484 * lock. During each step the configuration is synced out.
5485 *
5486 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5487 * devices.
5488 */
5489int
5490spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5491{
5492 vdev_t *vd;
5493 metaslab_group_t *mg;
5494 nvlist_t **spares, **l2cache, *nv;
5495 uint64_t txg = 0;
5496 uint_t nspares, nl2cache;
5497 int error = 0;
5498 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5499
5500 ASSERT(spa_writeable(spa));
5501
5502 if (!locked)
5503 txg = spa_vdev_enter(spa);
5504
5505 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5506
5507 if (spa->spa_spares.sav_vdevs != NULL &&
5508 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5509 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5510 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5511 /*
5512 * Only remove the hot spare if it's not currently in use
5513 * in this pool.
5514 */
5515 if (vd == NULL || unspare) {
5516 spa_vdev_remove_aux(spa->spa_spares.sav_config,
5517 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5518 spa_load_spares(spa);
5519 spa->spa_spares.sav_sync = B_TRUE;
5520 } else {
5521 error = SET_ERROR(EBUSY);
5522 }
5523 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5524 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5525 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5526 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5527 /*
5528 * Cache devices can always be removed.
5529 */
5530 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5531 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5532 spa_load_l2cache(spa);
5533 spa->spa_l2cache.sav_sync = B_TRUE;
5534 } else if (vd != NULL && vd->vdev_islog) {
5535 ASSERT(!locked);
5536 ASSERT(vd == vd->vdev_top);
5537
5538 /*
5539 * XXX - Once we have bp-rewrite this should
5540 * become the common case.
5541 */
5542
5543 mg = vd->vdev_mg;
5544
5545 /*
5546 * Stop allocating from this vdev.
5547 */
5548 metaslab_group_passivate(mg);
5549
5550 /*
5551 * Wait for the youngest allocations and frees to sync,
5552 * and then wait for the deferral of those frees to finish.
5553 */
5554 spa_vdev_config_exit(spa, NULL,
5555 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5556
5557 /*
5558 * Attempt to evacuate the vdev.
5559 */
5560 error = spa_vdev_remove_evacuate(spa, vd);
5561
5562 txg = spa_vdev_config_enter(spa);
5563
5564 /*
5565 * If we couldn't evacuate the vdev, unwind.
5566 */
5567 if (error) {
5568 metaslab_group_activate(mg);
5569 return (spa_vdev_exit(spa, NULL, txg, error));
5570 }
5571
5572 /*
5573 * Clean up the vdev namespace.
5574 */
5575 spa_vdev_remove_from_namespace(spa, vd);
5576
5577 } else if (vd != NULL) {
5578 /*
5579 * Normal vdevs cannot be removed (yet).
5580 */
5581 error = SET_ERROR(ENOTSUP);
5582 } else {
5583 /*
5584 * There is no vdev of any kind with the specified guid.
5585 */
5586 error = SET_ERROR(ENOENT);
5587 }
5588
5589 if (!locked)
5590 return (spa_vdev_exit(spa, NULL, txg, error));
5591
5592 return (error);
5593}
5594
5595/*
5596 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5597 * currently spared, so we can detach it.
5598 */
5599static vdev_t *
5600spa_vdev_resilver_done_hunt(vdev_t *vd)
5601{
5602 vdev_t *newvd, *oldvd;
5603
5604 for (int c = 0; c < vd->vdev_children; c++) {
5605 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5606 if (oldvd != NULL)
5607 return (oldvd);
5608 }
5609
5610 /*
5611 * Check for a completed replacement. We always consider the first
5612 * vdev in the list to be the oldest vdev, and the last one to be
5613 * the newest (see spa_vdev_attach() for how that works). In
5614 * the case where the newest vdev is faulted, we will not automatically
5615 * remove it after a resilver completes. This is OK as it will require
5616 * user intervention to determine which disk the admin wishes to keep.
5617 */
5618 if (vd->vdev_ops == &vdev_replacing_ops) {
5619 ASSERT(vd->vdev_children > 1);
5620
5621 newvd = vd->vdev_child[vd->vdev_children - 1];
5622 oldvd = vd->vdev_child[0];
5623
5624 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5625 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5626 !vdev_dtl_required(oldvd))
5627 return (oldvd);
5628 }
5629
5630 /*
5631 * Check for a completed resilver with the 'unspare' flag set.
5632 */
5633 if (vd->vdev_ops == &vdev_spare_ops) {
5634 vdev_t *first = vd->vdev_child[0];
5635 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5636
5637 if (last->vdev_unspare) {
5638 oldvd = first;
5639 newvd = last;
5640 } else if (first->vdev_unspare) {
5641 oldvd = last;
5642 newvd = first;
5643 } else {
5644 oldvd = NULL;
5645 }
5646
5647 if (oldvd != NULL &&
5648 vdev_dtl_empty(newvd, DTL_MISSING) &&
5649 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5650 !vdev_dtl_required(oldvd))
5651 return (oldvd);
5652
5653 /*
5654 * If there are more than two spares attached to a disk,
5655 * and those spares are not required, then we want to
5656 * attempt to free them up now so that they can be used
5657 * by other pools. Once we're back down to a single
5658 * disk+spare, we stop removing them.
5659 */
5660 if (vd->vdev_children > 2) {
5661 newvd = vd->vdev_child[1];
5662
5663 if (newvd->vdev_isspare && last->vdev_isspare &&
5664 vdev_dtl_empty(last, DTL_MISSING) &&
5665 vdev_dtl_empty(last, DTL_OUTAGE) &&
5666 !vdev_dtl_required(newvd))
5667 return (newvd);
5668 }
5669 }
5670
5671 return (NULL);
5672}
5673
5674static void
5675spa_vdev_resilver_done(spa_t *spa)
5676{
5677 vdev_t *vd, *pvd, *ppvd;
5678 uint64_t guid, sguid, pguid, ppguid;
5679
5680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5681
5682 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5683 pvd = vd->vdev_parent;
5684 ppvd = pvd->vdev_parent;
5685 guid = vd->vdev_guid;
5686 pguid = pvd->vdev_guid;
5687 ppguid = ppvd->vdev_guid;
5688 sguid = 0;
5689 /*
5690 * If we have just finished replacing a hot spared device, then
5691 * we need to detach the parent's first child (the original hot
5692 * spare) as well.
5693 */
5694 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5695 ppvd->vdev_children == 2) {
5696 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5697 sguid = ppvd->vdev_child[1]->vdev_guid;
5698 }
5699 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5700
5701 spa_config_exit(spa, SCL_ALL, FTAG);
5702 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5703 return;
5704 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5705 return;
5706 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5707 }
5708
5709 spa_config_exit(spa, SCL_ALL, FTAG);
5710}
5711
5712/*
5713 * Update the stored path or FRU for this vdev.
5714 */
5715int
5716spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5717 boolean_t ispath)
5718{
5719 vdev_t *vd;
5720 boolean_t sync = B_FALSE;
5721
5722 ASSERT(spa_writeable(spa));
5723
5724 spa_vdev_state_enter(spa, SCL_ALL);
5725
5726 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5727 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5728
5729 if (!vd->vdev_ops->vdev_op_leaf)
5730 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5731
5732 if (ispath) {
5733 if (strcmp(value, vd->vdev_path) != 0) {
5734 spa_strfree(vd->vdev_path);
5735 vd->vdev_path = spa_strdup(value);
5736 sync = B_TRUE;
5737 }
5738 } else {
5739 if (vd->vdev_fru == NULL) {
5740 vd->vdev_fru = spa_strdup(value);
5741 sync = B_TRUE;
5742 } else if (strcmp(value, vd->vdev_fru) != 0) {
5743 spa_strfree(vd->vdev_fru);
5744 vd->vdev_fru = spa_strdup(value);
5745 sync = B_TRUE;
5746 }
5747 }
5748
5749 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5750}
5751
5752int
5753spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5754{
5755 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5756}
5757
5758int
5759spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5760{
5761 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5762}
5763
5764/*
5765 * ==========================================================================
5766 * SPA Scanning
5767 * ==========================================================================
5768 */
5769
5770int
5771spa_scan_stop(spa_t *spa)
5772{
5773 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5774 if (dsl_scan_resilvering(spa->spa_dsl_pool))
5775 return (SET_ERROR(EBUSY));
5776 return (dsl_scan_cancel(spa->spa_dsl_pool));
5777}
5778
5779int
5780spa_scan(spa_t *spa, pool_scan_func_t func)
5781{
5782 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5783
5784 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5785 return (SET_ERROR(ENOTSUP));
5786
5787 /*
5788 * If a resilver was requested, but there is no DTL on a
5789 * writeable leaf device, we have nothing to do.
5790 */
5791 if (func == POOL_SCAN_RESILVER &&
5792 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5793 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5794 return (0);
5795 }
5796
5797 return (dsl_scan(spa->spa_dsl_pool, func));
5798}
5799
5800/*
5801 * ==========================================================================
5802 * SPA async task processing
5803 * ==========================================================================
5804 */
5805
5806static void
5807spa_async_remove(spa_t *spa, vdev_t *vd)
5808{
5809 if (vd->vdev_remove_wanted) {
5810 vd->vdev_remove_wanted = B_FALSE;
5811 vd->vdev_delayed_close = B_FALSE;
5812 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5813
5814 /*
5815 * We want to clear the stats, but we don't want to do a full
5816 * vdev_clear() as that will cause us to throw away
5817 * degraded/faulted state as well as attempt to reopen the
5818 * device, all of which is a waste.
5819 */
5820 vd->vdev_stat.vs_read_errors = 0;
5821 vd->vdev_stat.vs_write_errors = 0;
5822 vd->vdev_stat.vs_checksum_errors = 0;
5823
5824 vdev_state_dirty(vd->vdev_top);
5825 }
5826
5827 for (int c = 0; c < vd->vdev_children; c++)
5828 spa_async_remove(spa, vd->vdev_child[c]);
5829}
5830
5831static void
5832spa_async_probe(spa_t *spa, vdev_t *vd)
5833{
5834 if (vd->vdev_probe_wanted) {
5835 vd->vdev_probe_wanted = B_FALSE;
5836 vdev_reopen(vd); /* vdev_open() does the actual probe */
5837 }
5838
5839 for (int c = 0; c < vd->vdev_children; c++)
5840 spa_async_probe(spa, vd->vdev_child[c]);
5841}
5842
5843static void
5844spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5845{
5846 sysevent_id_t eid;
5847 nvlist_t *attr;
5848 char *physpath;
5849
5850 if (!spa->spa_autoexpand)
5851 return;
5852
5853 for (int c = 0; c < vd->vdev_children; c++) {
5854 vdev_t *cvd = vd->vdev_child[c];
5855 spa_async_autoexpand(spa, cvd);
5856 }
5857
5858 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5859 return;
5860
5861 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5862 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5863
5864 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5865 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5866
5867 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5868 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
5869
5870 nvlist_free(attr);
5871 kmem_free(physpath, MAXPATHLEN);
5872}
5873
5874static void
5875spa_async_thread(void *arg)
5876{
5877 spa_t *spa = arg;
5878 int tasks;
5879
5880 ASSERT(spa->spa_sync_on);
5881
5882 mutex_enter(&spa->spa_async_lock);
5883 tasks = spa->spa_async_tasks;
5884 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
5885 mutex_exit(&spa->spa_async_lock);
5886
5887 /*
5888 * See if the config needs to be updated.
5889 */
5890 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5891 uint64_t old_space, new_space;
5892
5893 mutex_enter(&spa_namespace_lock);
5894 old_space = metaslab_class_get_space(spa_normal_class(spa));
5895 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5896 new_space = metaslab_class_get_space(spa_normal_class(spa));
5897 mutex_exit(&spa_namespace_lock);
5898
5899 /*
5900 * If the pool grew as a result of the config update,
5901 * then log an internal history event.
5902 */
5903 if (new_space != old_space) {
5904 spa_history_log_internal(spa, "vdev online", NULL,
5905 "pool '%s' size: %llu(+%llu)",
5906 spa_name(spa), new_space, new_space - old_space);
5907 }
5908 }
5909
5910 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5911 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5912 spa_async_autoexpand(spa, spa->spa_root_vdev);
5913 spa_config_exit(spa, SCL_CONFIG, FTAG);
5914 }
5915
5916 /*
5917 * See if any devices need to be probed.
5918 */
5919 if (tasks & SPA_ASYNC_PROBE) {
5920 spa_vdev_state_enter(spa, SCL_NONE);
5921 spa_async_probe(spa, spa->spa_root_vdev);
5922 (void) spa_vdev_state_exit(spa, NULL, 0);
5923 }
5924
5925 /*
5926 * If any devices are done replacing, detach them.
5927 */
5928 if (tasks & SPA_ASYNC_RESILVER_DONE)
5929 spa_vdev_resilver_done(spa);
5930
5931 /*
5932 * Kick off a resilver.
5933 */
5934 if (tasks & SPA_ASYNC_RESILVER)
5935 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5936
5937 /*
5938 * Let the world know that we're done.
5939 */
5940 mutex_enter(&spa->spa_async_lock);
5941 spa->spa_async_thread = NULL;
5942 cv_broadcast(&spa->spa_async_cv);
5943 mutex_exit(&spa->spa_async_lock);
5944 thread_exit();
5945}
5946
5947static void
5948spa_async_thread_vd(void *arg)
5949{
5950 spa_t *spa = arg;
5951 int tasks;
5952
5953 ASSERT(spa->spa_sync_on);
5954
5955 mutex_enter(&spa->spa_async_lock);
5956 tasks = spa->spa_async_tasks;
5957retry:
5958 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
5959 mutex_exit(&spa->spa_async_lock);
5960
5961 /*
5962 * See if any devices need to be marked REMOVED.
5963 */
5964 if (tasks & SPA_ASYNC_REMOVE) {
5965 spa_vdev_state_enter(spa, SCL_NONE);
5966 spa_async_remove(spa, spa->spa_root_vdev);
5967 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5968 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5969 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5970 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5971 (void) spa_vdev_state_exit(spa, NULL, 0);
5972 }
5973
5974 /*
5975 * Let the world know that we're done.
5976 */
5977 mutex_enter(&spa->spa_async_lock);
5978 tasks = spa->spa_async_tasks;
5979 if ((tasks & SPA_ASYNC_REMOVE) != 0)
5980 goto retry;
5981 spa->spa_async_thread_vd = NULL;
5982 cv_broadcast(&spa->spa_async_cv);
5983 mutex_exit(&spa->spa_async_lock);
5984 thread_exit();
5985}
5986
5987void
5988spa_async_suspend(spa_t *spa)
5989{
5990 mutex_enter(&spa->spa_async_lock);
5991 spa->spa_async_suspended++;
5992 while (spa->spa_async_thread != NULL &&
5993 spa->spa_async_thread_vd != NULL)
5994 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5995 mutex_exit(&spa->spa_async_lock);
5996}
5997
5998void
5999spa_async_resume(spa_t *spa)
6000{
6001 mutex_enter(&spa->spa_async_lock);
6002 ASSERT(spa->spa_async_suspended != 0);
6003 spa->spa_async_suspended--;
6004 mutex_exit(&spa->spa_async_lock);
6005}
6006
6007static boolean_t
6008spa_async_tasks_pending(spa_t *spa)
6009{
6010 uint_t non_config_tasks;
6011 uint_t config_task;
6012 boolean_t config_task_suspended;
6013
6014 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
6015 SPA_ASYNC_REMOVE);
6016 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
6017 if (spa->spa_ccw_fail_time == 0) {
6018 config_task_suspended = B_FALSE;
6019 } else {
6020 config_task_suspended =
6021 (gethrtime() - spa->spa_ccw_fail_time) <
6022 (zfs_ccw_retry_interval * NANOSEC);
6023 }
6024
6025 return (non_config_tasks || (config_task && !config_task_suspended));
6026}
6027
6028static void
6029spa_async_dispatch(spa_t *spa)
6030{
6031 mutex_enter(&spa->spa_async_lock);
6032 if (spa_async_tasks_pending(spa) &&
6033 !spa->spa_async_suspended &&
6034 spa->spa_async_thread == NULL &&
6035 rootdir != NULL)
6036 spa->spa_async_thread = thread_create(NULL, 0,
6037 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
6038 mutex_exit(&spa->spa_async_lock);
6039}
6040
6041static void
6042spa_async_dispatch_vd(spa_t *spa)
6043{
6044 mutex_enter(&spa->spa_async_lock);
6045 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
6046 !spa->spa_async_suspended &&
6047 spa->spa_async_thread_vd == NULL &&
6048 rootdir != NULL)
6049 spa->spa_async_thread_vd = thread_create(NULL, 0,
6050 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
6051 mutex_exit(&spa->spa_async_lock);
6052}
6053
6054void
6055spa_async_request(spa_t *spa, int task)
6056{
6057 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
6058 mutex_enter(&spa->spa_async_lock);
6059 spa->spa_async_tasks |= task;
6060 mutex_exit(&spa->spa_async_lock);
6061 spa_async_dispatch_vd(spa);
6062}
6063
6064/*
6065 * ==========================================================================
6066 * SPA syncing routines
6067 * ==========================================================================
6068 */
6069
6070static int
6071bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6072{
6073 bpobj_t *bpo = arg;
6074 bpobj_enqueue(bpo, bp, tx);
6075 return (0);
6076}
6077
6078static int
6079spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6080{
6081 zio_t *zio = arg;
6082
6083 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6084 BP_GET_PSIZE(bp), zio->io_flags));
6085 return (0);
6086}
6087
6088/*
6089 * Note: this simple function is not inlined to make it easier to dtrace the
6090 * amount of time spent syncing frees.
6091 */
6092static void
6093spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6094{
6095 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6096 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6097 VERIFY(zio_wait(zio) == 0);
6098}
6099
6100/*
6101 * Note: this simple function is not inlined to make it easier to dtrace the
6102 * amount of time spent syncing deferred frees.
6103 */
6104static void
6105spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6106{
6107 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6108 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6109 spa_free_sync_cb, zio, tx), ==, 0);
6110 VERIFY0(zio_wait(zio));
6111}
6112
6113
6114static void
6115spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6116{
6117 char *packed = NULL;
6118 size_t bufsize;
6119 size_t nvsize = 0;
6120 dmu_buf_t *db;
6121
6122 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6123
6124 /*
6125 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6126 * information. This avoids the dmu_buf_will_dirty() path and
6127 * saves us a pre-read to get data we don't actually care about.
6128 */
6129 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6130 packed = kmem_alloc(bufsize, KM_SLEEP);
6131
6132 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6133 KM_SLEEP) == 0);
6134 bzero(packed + nvsize, bufsize - nvsize);
6135
6136 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6137
6138 kmem_free(packed, bufsize);
6139
6140 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6141 dmu_buf_will_dirty(db, tx);
6142 *(uint64_t *)db->db_data = nvsize;
6143 dmu_buf_rele(db, FTAG);
6144}
6145
6146static void
6147spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6148 const char *config, const char *entry)
6149{
6150 nvlist_t *nvroot;
6151 nvlist_t **list;
6152 int i;
6153
6154 if (!sav->sav_sync)
6155 return;
6156
6157 /*
6158 * Update the MOS nvlist describing the list of available devices.
6159 * spa_validate_aux() will have already made sure this nvlist is
6160 * valid and the vdevs are labeled appropriately.
6161 */
6162 if (sav->sav_object == 0) {
6163 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6164 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6165 sizeof (uint64_t), tx);
6166 VERIFY(zap_update(spa->spa_meta_objset,
6167 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6168 &sav->sav_object, tx) == 0);
6169 }
6170
6171 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6172 if (sav->sav_count == 0) {
6173 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6174 } else {
6175 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6176 for (i = 0; i < sav->sav_count; i++)
6177 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6178 B_FALSE, VDEV_CONFIG_L2CACHE);
6179 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6180 sav->sav_count) == 0);
6181 for (i = 0; i < sav->sav_count; i++)
6182 nvlist_free(list[i]);
6183 kmem_free(list, sav->sav_count * sizeof (void *));
6184 }
6185
6186 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6187 nvlist_free(nvroot);
6188
6189 sav->sav_sync = B_FALSE;
6190}
6191
6192static void
6193spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6194{
6195 nvlist_t *config;
6196
6197 if (list_is_empty(&spa->spa_config_dirty_list))
6198 return;
6199
6200 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6201
6202 config = spa_config_generate(spa, spa->spa_root_vdev,
6203 dmu_tx_get_txg(tx), B_FALSE);
6204
6205 /*
6206 * If we're upgrading the spa version then make sure that
6207 * the config object gets updated with the correct version.
6208 */
6209 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6210 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6211 spa->spa_uberblock.ub_version);
6212
6213 spa_config_exit(spa, SCL_STATE, FTAG);
6214
6215 if (spa->spa_config_syncing)
6216 nvlist_free(spa->spa_config_syncing);
6217 spa->spa_config_syncing = config;
6218
6219 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6220}
6221
6222static void
6223spa_sync_version(void *arg, dmu_tx_t *tx)
6224{
6225 uint64_t *versionp = arg;
6226 uint64_t version = *versionp;
6227 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6228
6229 /*
6230 * Setting the version is special cased when first creating the pool.
6231 */
6232 ASSERT(tx->tx_txg != TXG_INITIAL);
6233
6234 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6235 ASSERT(version >= spa_version(spa));
6236
6237 spa->spa_uberblock.ub_version = version;
6238 vdev_config_dirty(spa->spa_root_vdev);
6239 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6240}
6241
6242/*
6243 * Set zpool properties.
6244 */
6245static void
6246spa_sync_props(void *arg, dmu_tx_t *tx)
6247{
6248 nvlist_t *nvp = arg;
6249 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6250 objset_t *mos = spa->spa_meta_objset;
6251 nvpair_t *elem = NULL;
6252
6253 mutex_enter(&spa->spa_props_lock);
6254
6255 while ((elem = nvlist_next_nvpair(nvp, elem))) {
6256 uint64_t intval;
6257 char *strval, *fname;
6258 zpool_prop_t prop;
6259 const char *propname;
6260 zprop_type_t proptype;
6261 spa_feature_t fid;
6262
6263 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6264 case ZPROP_INVAL:
6265 /*
6266 * We checked this earlier in spa_prop_validate().
6267 */
6268 ASSERT(zpool_prop_feature(nvpair_name(elem)));
6269
6270 fname = strchr(nvpair_name(elem), '@') + 1;
6271 VERIFY0(zfeature_lookup_name(fname, &fid));
6272
6273 spa_feature_enable(spa, fid, tx);
6274 spa_history_log_internal(spa, "set", tx,
6275 "%s=enabled", nvpair_name(elem));
6276 break;
6277
6278 case ZPOOL_PROP_VERSION:
6279 intval = fnvpair_value_uint64(elem);
6280 /*
6281 * The version is synced seperatly before other
6282 * properties and should be correct by now.
6283 */
6284 ASSERT3U(spa_version(spa), >=, intval);
6285 break;
6286
6287 case ZPOOL_PROP_ALTROOT:
6288 /*
6289 * 'altroot' is a non-persistent property. It should
6290 * have been set temporarily at creation or import time.
6291 */
6292 ASSERT(spa->spa_root != NULL);
6293 break;
6294
6295 case ZPOOL_PROP_READONLY:
6296 case ZPOOL_PROP_CACHEFILE:
6297 /*
6298 * 'readonly' and 'cachefile' are also non-persisitent
6299 * properties.
6300 */
6301 break;
6302 case ZPOOL_PROP_COMMENT:
6303 strval = fnvpair_value_string(elem);
6304 if (spa->spa_comment != NULL)
6305 spa_strfree(spa->spa_comment);
6306 spa->spa_comment = spa_strdup(strval);
6307 /*
6308 * We need to dirty the configuration on all the vdevs
6309 * so that their labels get updated. It's unnecessary
6310 * to do this for pool creation since the vdev's
6311 * configuratoin has already been dirtied.
6312 */
6313 if (tx->tx_txg != TXG_INITIAL)
6314 vdev_config_dirty(spa->spa_root_vdev);
6315 spa_history_log_internal(spa, "set", tx,
6316 "%s=%s", nvpair_name(elem), strval);
6317 break;
6318 default:
6319 /*
6320 * Set pool property values in the poolprops mos object.
6321 */
6322 if (spa->spa_pool_props_object == 0) {
6323 spa->spa_pool_props_object =
6324 zap_create_link(mos, DMU_OT_POOL_PROPS,
6325 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6326 tx);
6327 }
6328
6329 /* normalize the property name */
6330 propname = zpool_prop_to_name(prop);
6331 proptype = zpool_prop_get_type(prop);
6332
6333 if (nvpair_type(elem) == DATA_TYPE_STRING) {
6334 ASSERT(proptype == PROP_TYPE_STRING);
6335 strval = fnvpair_value_string(elem);
6336 VERIFY0(zap_update(mos,
6337 spa->spa_pool_props_object, propname,
6338 1, strlen(strval) + 1, strval, tx));
6339 spa_history_log_internal(spa, "set", tx,
6340 "%s=%s", nvpair_name(elem), strval);
6341 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6342 intval = fnvpair_value_uint64(elem);
6343
6344 if (proptype == PROP_TYPE_INDEX) {
6345 const char *unused;
6346 VERIFY0(zpool_prop_index_to_string(
6347 prop, intval, &unused));
6348 }
6349 VERIFY0(zap_update(mos,
6350 spa->spa_pool_props_object, propname,
6351 8, 1, &intval, tx));
6352 spa_history_log_internal(spa, "set", tx,
6353 "%s=%lld", nvpair_name(elem), intval);
6354 } else {
6355 ASSERT(0); /* not allowed */
6356 }
6357
6358 switch (prop) {
6359 case ZPOOL_PROP_DELEGATION:
6360 spa->spa_delegation = intval;
6361 break;
6362 case ZPOOL_PROP_BOOTFS:
6363 spa->spa_bootfs = intval;
6364 break;
6365 case ZPOOL_PROP_FAILUREMODE:
6366 spa->spa_failmode = intval;
6367 break;
6368 case ZPOOL_PROP_AUTOEXPAND:
6369 spa->spa_autoexpand = intval;
6370 if (tx->tx_txg != TXG_INITIAL)
6371 spa_async_request(spa,
6372 SPA_ASYNC_AUTOEXPAND);
6373 break;
6374 case ZPOOL_PROP_DEDUPDITTO:
6375 spa->spa_dedup_ditto = intval;
6376 break;
6377 default:
6378 break;
6379 }
6380 }
6381
6382 }
6383
6384 mutex_exit(&spa->spa_props_lock);
6385}
6386
6387/*
6388 * Perform one-time upgrade on-disk changes. spa_version() does not
6389 * reflect the new version this txg, so there must be no changes this
6390 * txg to anything that the upgrade code depends on after it executes.
6391 * Therefore this must be called after dsl_pool_sync() does the sync
6392 * tasks.
6393 */
6394static void
6395spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6396{
6397 dsl_pool_t *dp = spa->spa_dsl_pool;
6398
6399 ASSERT(spa->spa_sync_pass == 1);
6400
6401 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6402
6403 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6404 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6405 dsl_pool_create_origin(dp, tx);
6406
6407 /* Keeping the origin open increases spa_minref */
6408 spa->spa_minref += 3;
6409 }
6410
6411 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6412 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6413 dsl_pool_upgrade_clones(dp, tx);
6414 }
6415
6416 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6417 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6418 dsl_pool_upgrade_dir_clones(dp, tx);
6419
6420 /* Keeping the freedir open increases spa_minref */
6421 spa->spa_minref += 3;
6422 }
6423
6424 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6425 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6426 spa_feature_create_zap_objects(spa, tx);
6427 }
6428 rrw_exit(&dp->dp_config_rwlock, FTAG);
6429}
6430
6431/*
6432 * Sync the specified transaction group. New blocks may be dirtied as
6433 * part of the process, so we iterate until it converges.
6434 */
6435void
6436spa_sync(spa_t *spa, uint64_t txg)
6437{
6438 dsl_pool_t *dp = spa->spa_dsl_pool;
6439 objset_t *mos = spa->spa_meta_objset;
6440 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6441 vdev_t *rvd = spa->spa_root_vdev;
6442 vdev_t *vd;
6443 dmu_tx_t *tx;
6444 int error;
6445
6446 VERIFY(spa_writeable(spa));
6447
6448 /*
6449 * Lock out configuration changes.
6450 */
6451 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6452
6453 spa->spa_syncing_txg = txg;
6454 spa->spa_sync_pass = 0;
6455
6456 /*
6457 * If there are any pending vdev state changes, convert them
6458 * into config changes that go out with this transaction group.
6459 */
6460 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6461 while (list_head(&spa->spa_state_dirty_list) != NULL) {
6462 /*
6463 * We need the write lock here because, for aux vdevs,
6464 * calling vdev_config_dirty() modifies sav_config.
6465 * This is ugly and will become unnecessary when we
6466 * eliminate the aux vdev wart by integrating all vdevs
6467 * into the root vdev tree.
6468 */
6469 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6470 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6471 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6472 vdev_state_clean(vd);
6473 vdev_config_dirty(vd);
6474 }
6475 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6476 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6477 }
6478 spa_config_exit(spa, SCL_STATE, FTAG);
6479
6480 tx = dmu_tx_create_assigned(dp, txg);
6481
6482 spa->spa_sync_starttime = gethrtime();
6483#ifdef illumos
6484 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6485 spa->spa_sync_starttime + spa->spa_deadman_synctime));
6486#else /* FreeBSD */
6487#ifdef _KERNEL
6488 callout_reset(&spa->spa_deadman_cycid,
6489 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa);
6490#endif
6491#endif
6492
6493 /*
6494 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6495 * set spa_deflate if we have no raid-z vdevs.
6496 */
6497 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6498 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6499 int i;
6500
6501 for (i = 0; i < rvd->vdev_children; i++) {
6502 vd = rvd->vdev_child[i];
6503 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6504 break;
6505 }
6506 if (i == rvd->vdev_children) {
6507 spa->spa_deflate = TRUE;
6508 VERIFY(0 == zap_add(spa->spa_meta_objset,
6509 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6510 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6511 }
6512 }
6513
6514 /*
6515 * If anything has changed in this txg, or if someone is waiting
6516 * for this txg to sync (eg, spa_vdev_remove()), push the
6517 * deferred frees from the previous txg. If not, leave them
6518 * alone so that we don't generate work on an otherwise idle
6519 * system.
6520 */
6521 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6522 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6523 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6524 ((dsl_scan_active(dp->dp_scan) ||
6525 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6526 spa_sync_deferred_frees(spa, tx);
6527 }
6528
6529 /*
6530 * Iterate to convergence.
6531 */
6532 do {
6533 int pass = ++spa->spa_sync_pass;
6534
6535 spa_sync_config_object(spa, tx);
6536 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6537 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6538 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6539 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6540 spa_errlog_sync(spa, txg);
6541 dsl_pool_sync(dp, txg);
6542
6543 if (pass < zfs_sync_pass_deferred_free) {
6544 spa_sync_frees(spa, free_bpl, tx);
6545 } else {
6546 bplist_iterate(free_bpl, bpobj_enqueue_cb,
6547 &spa->spa_deferred_bpobj, tx);
6548 }
6549
6550 ddt_sync(spa, txg);
6551 dsl_scan_sync(dp, tx);
6552
6553 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6554 vdev_sync(vd, txg);
6555
6556 if (pass == 1)
6557 spa_sync_upgrades(spa, tx);
6558
6559 } while (dmu_objset_is_dirty(mos, txg));
6560
6561 /*
6562 * Rewrite the vdev configuration (which includes the uberblock)
6563 * to commit the transaction group.
6564 *
6565 * If there are no dirty vdevs, we sync the uberblock to a few
6566 * random top-level vdevs that are known to be visible in the
6567 * config cache (see spa_vdev_add() for a complete description).
6568 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6569 */
6570 for (;;) {
6571 /*
6572 * We hold SCL_STATE to prevent vdev open/close/etc.
6573 * while we're attempting to write the vdev labels.
6574 */
6575 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6576
6577 if (list_is_empty(&spa->spa_config_dirty_list)) {
6578 vdev_t *svd[SPA_DVAS_PER_BP];
6579 int svdcount = 0;
6580 int children = rvd->vdev_children;
6581 int c0 = spa_get_random(children);
6582
6583 for (int c = 0; c < children; c++) {
6584 vd = rvd->vdev_child[(c0 + c) % children];
6585 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6586 continue;
6587 svd[svdcount++] = vd;
6588 if (svdcount == SPA_DVAS_PER_BP)
6589 break;
6590 }
6591 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6592 if (error != 0)
6593 error = vdev_config_sync(svd, svdcount, txg,
6594 B_TRUE);
6595 } else {
6596 error = vdev_config_sync(rvd->vdev_child,
6597 rvd->vdev_children, txg, B_FALSE);
6598 if (error != 0)
6599 error = vdev_config_sync(rvd->vdev_child,
6600 rvd->vdev_children, txg, B_TRUE);
6601 }
6602
6603 if (error == 0)
6604 spa->spa_last_synced_guid = rvd->vdev_guid;
6605
6606 spa_config_exit(spa, SCL_STATE, FTAG);
6607
6608 if (error == 0)
6609 break;
6610 zio_suspend(spa, NULL);
6611 zio_resume_wait(spa);
6612 }
6613 dmu_tx_commit(tx);
6614
6615#ifdef illumos
6616 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6617#else /* FreeBSD */
6618#ifdef _KERNEL
6619 callout_drain(&spa->spa_deadman_cycid);
6620#endif
6621#endif
6622
6623 /*
6624 * Clear the dirty config list.
6625 */
6626 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6627 vdev_config_clean(vd);
6628
6629 /*
6630 * Now that the new config has synced transactionally,
6631 * let it become visible to the config cache.
6632 */
6633 if (spa->spa_config_syncing != NULL) {
6634 spa_config_set(spa, spa->spa_config_syncing);
6635 spa->spa_config_txg = txg;
6636 spa->spa_config_syncing = NULL;
6637 }
6638
6639 spa->spa_ubsync = spa->spa_uberblock;
6640
6641 dsl_pool_sync_done(dp, txg);
6642
6643 /*
6644 * Update usable space statistics.
6645 */
6646 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6647 vdev_sync_done(vd, txg);
6648
6649 spa_update_dspace(spa);
6650
6651 /*
6652 * It had better be the case that we didn't dirty anything
6653 * since vdev_config_sync().
6654 */
6655 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6656 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6657 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6658
6659 spa->spa_sync_pass = 0;
6660
6661 spa_config_exit(spa, SCL_CONFIG, FTAG);
6662
6663 spa_handle_ignored_writes(spa);
6664
6665 /*
6666 * If any async tasks have been requested, kick them off.
6667 */
6668 spa_async_dispatch(spa);
6669 spa_async_dispatch_vd(spa);
6670}
6671
6672/*
6673 * Sync all pools. We don't want to hold the namespace lock across these
6674 * operations, so we take a reference on the spa_t and drop the lock during the
6675 * sync.
6676 */
6677void
6678spa_sync_allpools(void)
6679{
6680 spa_t *spa = NULL;
6681 mutex_enter(&spa_namespace_lock);
6682 while ((spa = spa_next(spa)) != NULL) {
6683 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6684 !spa_writeable(spa) || spa_suspended(spa))
6685 continue;
6686 spa_open_ref(spa, FTAG);
6687 mutex_exit(&spa_namespace_lock);
6688 txg_wait_synced(spa_get_dsl(spa), 0);
6689 mutex_enter(&spa_namespace_lock);
6690 spa_close(spa, FTAG);
6691 }
6692 mutex_exit(&spa_namespace_lock);
6693}
6694
6695/*
6696 * ==========================================================================
6697 * Miscellaneous routines
6698 * ==========================================================================
6699 */
6700
6701/*
6702 * Remove all pools in the system.
6703 */
6704void
6705spa_evict_all(void)
6706{
6707 spa_t *spa;
6708
6709 /*
6710 * Remove all cached state. All pools should be closed now,
6711 * so every spa in the AVL tree should be unreferenced.
6712 */
6713 mutex_enter(&spa_namespace_lock);
6714 while ((spa = spa_next(NULL)) != NULL) {
6715 /*
6716 * Stop async tasks. The async thread may need to detach
6717 * a device that's been replaced, which requires grabbing
6718 * spa_namespace_lock, so we must drop it here.
6719 */
6720 spa_open_ref(spa, FTAG);
6721 mutex_exit(&spa_namespace_lock);
6722 spa_async_suspend(spa);
6723 mutex_enter(&spa_namespace_lock);
6724 spa_close(spa, FTAG);
6725
6726 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6727 spa_unload(spa);
6728 spa_deactivate(spa);
6729 }
6730 spa_remove(spa);
6731 }
6732 mutex_exit(&spa_namespace_lock);
6733}
6734
6735vdev_t *
6736spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6737{
6738 vdev_t *vd;
6739 int i;
6740
6741 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6742 return (vd);
6743
6744 if (aux) {
6745 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6746 vd = spa->spa_l2cache.sav_vdevs[i];
6747 if (vd->vdev_guid == guid)
6748 return (vd);
6749 }
6750
6751 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6752 vd = spa->spa_spares.sav_vdevs[i];
6753 if (vd->vdev_guid == guid)
6754 return (vd);
6755 }
6756 }
6757
6758 return (NULL);
6759}
6760
6761void
6762spa_upgrade(spa_t *spa, uint64_t version)
6763{
6764 ASSERT(spa_writeable(spa));
6765
6766 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6767
6768 /*
6769 * This should only be called for a non-faulted pool, and since a
6770 * future version would result in an unopenable pool, this shouldn't be
6771 * possible.
6772 */
6773 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6774 ASSERT(version >= spa->spa_uberblock.ub_version);
6775
6776 spa->spa_uberblock.ub_version = version;
6777 vdev_config_dirty(spa->spa_root_vdev);
6778
6779 spa_config_exit(spa, SCL_ALL, FTAG);
6780
6781 txg_wait_synced(spa_get_dsl(spa), 0);
6782}
6783
6784boolean_t
6785spa_has_spare(spa_t *spa, uint64_t guid)
6786{
6787 int i;
6788 uint64_t spareguid;
6789 spa_aux_vdev_t *sav = &spa->spa_spares;
6790
6791 for (i = 0; i < sav->sav_count; i++)
6792 if (sav->sav_vdevs[i]->vdev_guid == guid)
6793 return (B_TRUE);
6794
6795 for (i = 0; i < sav->sav_npending; i++) {
6796 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6797 &spareguid) == 0 && spareguid == guid)
6798 return (B_TRUE);
6799 }
6800
6801 return (B_FALSE);
6802}
6803
6804/*
6805 * Check if a pool has an active shared spare device.
6806 * Note: reference count of an active spare is 2, as a spare and as a replace
6807 */
6808static boolean_t
6809spa_has_active_shared_spare(spa_t *spa)
6810{
6811 int i, refcnt;
6812 uint64_t pool;
6813 spa_aux_vdev_t *sav = &spa->spa_spares;
6814
6815 for (i = 0; i < sav->sav_count; i++) {
6816 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6817 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6818 refcnt > 2)
6819 return (B_TRUE);
6820 }
6821
6822 return (B_FALSE);
6823}
6824
6825/*
6826 * Post a sysevent corresponding to the given event. The 'name' must be one of
6827 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
6828 * filled in from the spa and (optionally) the vdev. This doesn't do anything
6829 * in the userland libzpool, as we don't want consumers to misinterpret ztest
6830 * or zdb as real changes.
6831 */
6832void
6833spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6834{
6835#ifdef _KERNEL
6836 sysevent_t *ev;
6837 sysevent_attr_list_t *attr = NULL;
6838 sysevent_value_t value;
6839 sysevent_id_t eid;
6840
6841 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6842 SE_SLEEP);
6843
6844 value.value_type = SE_DATA_TYPE_STRING;
6845 value.value.sv_string = spa_name(spa);
6846 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6847 goto done;
6848
6849 value.value_type = SE_DATA_TYPE_UINT64;
6850 value.value.sv_uint64 = spa_guid(spa);
6851 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6852 goto done;
6853
6854 if (vd) {
6855 value.value_type = SE_DATA_TYPE_UINT64;
6856 value.value.sv_uint64 = vd->vdev_guid;
6857 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6858 SE_SLEEP) != 0)
6859 goto done;
6860
6861 if (vd->vdev_path) {
6862 value.value_type = SE_DATA_TYPE_STRING;
6863 value.value.sv_string = vd->vdev_path;
6864 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6865 &value, SE_SLEEP) != 0)
6866 goto done;
6867 }
6868 }
6869
6870 if (sysevent_attach_attributes(ev, attr) != 0)
6871 goto done;
6872 attr = NULL;
6873
6874 (void) log_sysevent(ev, SE_SLEEP, &eid);
6875
6876done:
6877 if (attr)
6878 sysevent_free_attr(attr);
6879 sysevent_free(ev);
6880#endif
6881}
|