4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
32 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfs_context.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/dsl_deleg.h>
43 #include <sys/dnode.h>
46 #include <sys/dmu_tx.h>
49 #include <sys/dmu_impl.h>
50 #include <sys/zfs_ioctl.h>
52 #include <sys/zfs_onexit.h>
53 #include <sys/dsl_destroy.h>
57 * Needed to close a window in dnode_move() that allows the objset to be freed
58 * before it can be safely accessed.
63 * Tunable to overwrite the maximum number of threads for the parallization
64 * of dmu_objset_find_dp, needed to speed up the import of pools with many
66 * Default is 4 times the number of leaf vdevs.
68 int dmu_find_threads = 0;
70 static void dmu_objset_find_dp_cb(void *arg);
75 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
85 dmu_objset_spa(objset_t *os)
91 dmu_objset_zil(objset_t *os)
97 dmu_objset_pool(objset_t *os)
101 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
102 return (ds->ds_dir->dd_pool);
104 return (spa_get_dsl(os->os_spa));
108 dmu_objset_ds(objset_t *os)
110 return (os->os_dsl_dataset);
114 dmu_objset_type(objset_t *os)
116 return (os->os_phys->os_type);
120 dmu_objset_name(objset_t *os, char *buf)
122 dsl_dataset_name(os->os_dsl_dataset, buf);
126 dmu_objset_id(objset_t *os)
128 dsl_dataset_t *ds = os->os_dsl_dataset;
130 return (ds ? ds->ds_object : 0);
134 dmu_objset_syncprop(objset_t *os)
136 return (os->os_sync);
140 dmu_objset_logbias(objset_t *os)
142 return (os->os_logbias);
146 checksum_changed_cb(void *arg, uint64_t newval)
151 * Inheritance should have been done by now.
153 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
155 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
159 compression_changed_cb(void *arg, uint64_t newval)
164 * Inheritance and range checking should have been done by now.
166 ASSERT(newval != ZIO_COMPRESS_INHERIT);
168 os->os_compress = zio_compress_select(os->os_spa, newval,
173 copies_changed_cb(void *arg, uint64_t newval)
178 * Inheritance and range checking should have been done by now.
181 ASSERT(newval <= spa_max_replication(os->os_spa));
183 os->os_copies = newval;
187 dedup_changed_cb(void *arg, uint64_t newval)
190 spa_t *spa = os->os_spa;
191 enum zio_checksum checksum;
194 * Inheritance should have been done by now.
196 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
198 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
200 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
201 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
205 primary_cache_changed_cb(void *arg, uint64_t newval)
210 * Inheritance and range checking should have been done by now.
212 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
213 newval == ZFS_CACHE_METADATA);
215 os->os_primary_cache = newval;
219 secondary_cache_changed_cb(void *arg, uint64_t newval)
224 * Inheritance and range checking should have been done by now.
226 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
227 newval == ZFS_CACHE_METADATA);
229 os->os_secondary_cache = newval;
233 sync_changed_cb(void *arg, uint64_t newval)
238 * Inheritance and range checking should have been done by now.
240 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
241 newval == ZFS_SYNC_DISABLED);
243 os->os_sync = newval;
245 zil_set_sync(os->os_zil, newval);
249 redundant_metadata_changed_cb(void *arg, uint64_t newval)
254 * Inheritance and range checking should have been done by now.
256 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
257 newval == ZFS_REDUNDANT_METADATA_MOST);
259 os->os_redundant_metadata = newval;
263 logbias_changed_cb(void *arg, uint64_t newval)
267 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
268 newval == ZFS_LOGBIAS_THROUGHPUT);
269 os->os_logbias = newval;
271 zil_set_logbias(os->os_zil, newval);
275 recordsize_changed_cb(void *arg, uint64_t newval)
279 os->os_recordsize = newval;
283 dmu_objset_byteswap(void *buf, size_t size)
285 objset_phys_t *osp = buf;
287 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
288 dnode_byteswap(&osp->os_meta_dnode);
289 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
290 osp->os_type = BSWAP_64(osp->os_type);
291 osp->os_flags = BSWAP_64(osp->os_flags);
292 if (size == sizeof (objset_phys_t)) {
293 dnode_byteswap(&osp->os_userused_dnode);
294 dnode_byteswap(&osp->os_groupused_dnode);
299 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
305 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
307 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
308 os->os_dsl_dataset = ds;
311 if (!BP_IS_HOLE(os->os_rootbp)) {
312 arc_flags_t aflags = ARC_FLAG_WAIT;
314 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
315 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
317 if (DMU_OS_IS_L2CACHEABLE(os))
318 aflags |= ARC_FLAG_L2CACHE;
319 if (DMU_OS_IS_L2COMPRESSIBLE(os))
320 aflags |= ARC_FLAG_L2COMPRESS;
322 dprintf_bp(os->os_rootbp, "reading %s", "");
323 err = arc_read(NULL, spa, os->os_rootbp,
324 arc_getbuf_func, &os->os_phys_buf,
325 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
327 kmem_free(os, sizeof (objset_t));
328 /* convert checksum errors into IO errors */
330 err = SET_ERROR(EIO);
334 /* Increase the blocksize if we are permitted. */
335 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
336 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
337 arc_buf_t *buf = arc_buf_alloc(spa,
338 sizeof (objset_phys_t), &os->os_phys_buf,
340 bzero(buf->b_data, sizeof (objset_phys_t));
341 bcopy(os->os_phys_buf->b_data, buf->b_data,
342 arc_buf_size(os->os_phys_buf));
343 (void) arc_buf_remove_ref(os->os_phys_buf,
345 os->os_phys_buf = buf;
348 os->os_phys = os->os_phys_buf->b_data;
349 os->os_flags = os->os_phys->os_flags;
351 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
352 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
353 os->os_phys_buf = arc_buf_alloc(spa, size,
354 &os->os_phys_buf, ARC_BUFC_METADATA);
355 os->os_phys = os->os_phys_buf->b_data;
356 bzero(os->os_phys, size);
360 * Note: the changed_cb will be called once before the register
361 * func returns, thus changing the checksum/compression from the
362 * default (fletcher2/off). Snapshots don't need to know about
363 * checksum/compression/copies.
366 boolean_t needlock = B_FALSE;
369 * Note: it's valid to open the objset if the dataset is
370 * long-held, in which case the pool_config lock will not
373 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
375 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
377 err = dsl_prop_register(ds,
378 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
379 primary_cache_changed_cb, os);
381 err = dsl_prop_register(ds,
382 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
383 secondary_cache_changed_cb, os);
385 if (!ds->ds_is_snapshot) {
387 err = dsl_prop_register(ds,
388 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
389 checksum_changed_cb, os);
392 err = dsl_prop_register(ds,
393 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
394 compression_changed_cb, os);
397 err = dsl_prop_register(ds,
398 zfs_prop_to_name(ZFS_PROP_COPIES),
399 copies_changed_cb, os);
402 err = dsl_prop_register(ds,
403 zfs_prop_to_name(ZFS_PROP_DEDUP),
404 dedup_changed_cb, os);
407 err = dsl_prop_register(ds,
408 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
409 logbias_changed_cb, os);
412 err = dsl_prop_register(ds,
413 zfs_prop_to_name(ZFS_PROP_SYNC),
414 sync_changed_cb, os);
417 err = dsl_prop_register(ds,
419 ZFS_PROP_REDUNDANT_METADATA),
420 redundant_metadata_changed_cb, os);
423 err = dsl_prop_register(ds,
424 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
425 recordsize_changed_cb, os);
429 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
431 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
433 kmem_free(os, sizeof (objset_t));
437 /* It's the meta-objset. */
438 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
439 os->os_compress = ZIO_COMPRESS_ON;
440 os->os_copies = spa_max_replication(spa);
441 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
442 os->os_dedup_verify = B_FALSE;
443 os->os_logbias = ZFS_LOGBIAS_LATENCY;
444 os->os_sync = ZFS_SYNC_STANDARD;
445 os->os_primary_cache = ZFS_CACHE_ALL;
446 os->os_secondary_cache = ZFS_CACHE_ALL;
449 if (ds == NULL || !ds->ds_is_snapshot)
450 os->os_zil_header = os->os_phys->os_zil_header;
451 os->os_zil = zil_alloc(os, &os->os_zil_header);
453 for (i = 0; i < TXG_SIZE; i++) {
454 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
455 offsetof(dnode_t, dn_dirty_link[i]));
456 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
457 offsetof(dnode_t, dn_dirty_link[i]));
459 list_create(&os->os_dnodes, sizeof (dnode_t),
460 offsetof(dnode_t, dn_link));
461 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
462 offsetof(dmu_buf_impl_t, db_link));
464 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
465 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
466 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
468 dnode_special_open(os, &os->os_phys->os_meta_dnode,
469 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
470 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
471 dnode_special_open(os, &os->os_phys->os_userused_dnode,
472 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
473 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
474 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
482 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
487 * We shouldn't be doing anything with dsl_dataset_t's unless the
488 * pool_config lock is held, or the dataset is long-held.
490 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
491 dsl_dataset_long_held(ds));
493 mutex_enter(&ds->ds_opening_lock);
494 if (ds->ds_objset == NULL) {
496 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
497 ds, dsl_dataset_get_blkptr(ds), &os);
500 mutex_enter(&ds->ds_lock);
501 ASSERT(ds->ds_objset == NULL);
503 mutex_exit(&ds->ds_lock);
506 *osp = ds->ds_objset;
507 mutex_exit(&ds->ds_opening_lock);
512 * Holds the pool while the objset is held. Therefore only one objset
513 * can be held at a time.
516 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
522 err = dsl_pool_hold(name, tag, &dp);
525 err = dsl_dataset_hold(dp, name, tag, &ds);
527 dsl_pool_rele(dp, tag);
531 err = dmu_objset_from_ds(ds, osp);
533 dsl_dataset_rele(ds, tag);
534 dsl_pool_rele(dp, tag);
541 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
542 boolean_t readonly, void *tag, objset_t **osp)
546 err = dmu_objset_from_ds(ds, osp);
548 dsl_dataset_disown(ds, tag);
549 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
550 dsl_dataset_disown(ds, tag);
551 return (SET_ERROR(EINVAL));
552 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
553 dsl_dataset_disown(ds, tag);
554 return (SET_ERROR(EROFS));
560 * dsl_pool must not be held when this is called.
561 * Upon successful return, there will be a longhold on the dataset,
562 * and the dsl_pool will not be held.
565 dmu_objset_own(const char *name, dmu_objset_type_t type,
566 boolean_t readonly, void *tag, objset_t **osp)
572 err = dsl_pool_hold(name, FTAG, &dp);
575 err = dsl_dataset_own(dp, name, tag, &ds);
577 dsl_pool_rele(dp, FTAG);
580 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
581 dsl_pool_rele(dp, FTAG);
587 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
588 boolean_t readonly, void *tag, objset_t **osp)
593 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
597 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
601 dmu_objset_rele(objset_t *os, void *tag)
603 dsl_pool_t *dp = dmu_objset_pool(os);
604 dsl_dataset_rele(os->os_dsl_dataset, tag);
605 dsl_pool_rele(dp, tag);
609 * When we are called, os MUST refer to an objset associated with a dataset
610 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
611 * == tag. We will then release and reacquire ownership of the dataset while
612 * holding the pool config_rwlock to avoid intervening namespace or ownership
615 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
616 * release the hold on its dataset and acquire a new one on the dataset of the
617 * same name so that it can be partially torn down and reconstructed.
620 dmu_objset_refresh_ownership(objset_t *os, void *tag)
623 dsl_dataset_t *ds, *newds;
624 char name[ZFS_MAX_DATASET_NAME_LEN];
626 ds = os->os_dsl_dataset;
627 VERIFY3P(ds, !=, NULL);
628 VERIFY3P(ds->ds_owner, ==, tag);
629 VERIFY(dsl_dataset_long_held(ds));
631 dsl_dataset_name(ds, name);
632 dp = dmu_objset_pool(os);
633 dsl_pool_config_enter(dp, FTAG);
634 dmu_objset_disown(os, tag);
635 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
636 VERIFY3P(newds, ==, os->os_dsl_dataset);
637 dsl_pool_config_exit(dp, FTAG);
641 dmu_objset_disown(objset_t *os, void *tag)
643 dsl_dataset_disown(os->os_dsl_dataset, tag);
647 dmu_objset_evict_dbufs(objset_t *os)
652 mutex_enter(&os->os_lock);
653 dn = list_head(&os->os_dnodes);
656 * Skip dnodes without holds. We have to do this dance
657 * because dnode_add_ref() only works if there is already a
658 * hold. If the dnode has no holds, then it has no dbufs.
660 if (dnode_add_ref(dn, FTAG)) {
661 list_insert_after(&os->os_dnodes, dn, &dn_marker);
662 mutex_exit(&os->os_lock);
664 dnode_evict_dbufs(dn);
665 dnode_rele(dn, FTAG);
667 mutex_enter(&os->os_lock);
668 dn = list_next(&os->os_dnodes, &dn_marker);
669 list_remove(&os->os_dnodes, &dn_marker);
671 dn = list_next(&os->os_dnodes, dn);
674 mutex_exit(&os->os_lock);
676 if (DMU_USERUSED_DNODE(os) != NULL) {
677 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
678 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
680 dnode_evict_dbufs(DMU_META_DNODE(os));
684 * Objset eviction processing is split into into two pieces.
685 * The first marks the objset as evicting, evicts any dbufs that
686 * have a refcount of zero, and then queues up the objset for the
687 * second phase of eviction. Once os->os_dnodes has been cleared by
688 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
689 * The second phase closes the special dnodes, dequeues the objset from
690 * the list of those undergoing eviction, and finally frees the objset.
692 * NOTE: Due to asynchronous eviction processing (invocation of
693 * dnode_buf_pageout()), it is possible for the meta dnode for the
694 * objset to have no holds even though os->os_dnodes is not empty.
697 dmu_objset_evict(objset_t *os)
699 dsl_dataset_t *ds = os->os_dsl_dataset;
701 for (int t = 0; t < TXG_SIZE; t++)
702 ASSERT(!dmu_objset_is_dirty(os, t));
705 dsl_prop_unregister_all(ds, os);
710 dmu_objset_evict_dbufs(os);
712 mutex_enter(&os->os_lock);
713 spa_evicting_os_register(os->os_spa, os);
714 if (list_is_empty(&os->os_dnodes)) {
715 mutex_exit(&os->os_lock);
716 dmu_objset_evict_done(os);
718 mutex_exit(&os->os_lock);
723 dmu_objset_evict_done(objset_t *os)
725 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
727 dnode_special_close(&os->os_meta_dnode);
728 if (DMU_USERUSED_DNODE(os)) {
729 dnode_special_close(&os->os_userused_dnode);
730 dnode_special_close(&os->os_groupused_dnode);
732 zil_free(os->os_zil);
734 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
737 * This is a barrier to prevent the objset from going away in
738 * dnode_move() until we can safely ensure that the objset is still in
739 * use. We consider the objset valid before the barrier and invalid
742 rw_enter(&os_lock, RW_READER);
745 mutex_destroy(&os->os_lock);
746 mutex_destroy(&os->os_obj_lock);
747 mutex_destroy(&os->os_user_ptr_lock);
748 spa_evicting_os_deregister(os->os_spa, os);
749 kmem_free(os, sizeof (objset_t));
753 dmu_objset_snap_cmtime(objset_t *os)
755 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
758 /* called from dsl for meta-objset */
760 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
761 dmu_objset_type_t type, dmu_tx_t *tx)
766 ASSERT(dmu_tx_is_syncing(tx));
769 VERIFY0(dmu_objset_from_ds(ds, &os));
771 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
773 mdn = DMU_META_DNODE(os);
775 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
776 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
779 * We don't want to have to increase the meta-dnode's nlevels
780 * later, because then we could do it in quescing context while
781 * we are also accessing it in open context.
783 * This precaution is not necessary for the MOS (ds == NULL),
784 * because the MOS is only updated in syncing context.
785 * This is most fortunate: the MOS is the only objset that
786 * needs to be synced multiple times as spa_sync() iterates
787 * to convergence, so minimizing its dn_nlevels matters.
793 * Determine the number of levels necessary for the meta-dnode
794 * to contain DN_MAX_OBJECT dnodes.
796 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
797 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
798 DN_MAX_OBJECT * sizeof (dnode_phys_t))
801 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
802 mdn->dn_nlevels = levels;
805 ASSERT(type != DMU_OST_NONE);
806 ASSERT(type != DMU_OST_ANY);
807 ASSERT(type < DMU_OST_NUMTYPES);
808 os->os_phys->os_type = type;
809 if (dmu_objset_userused_enabled(os)) {
810 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
811 os->os_flags = os->os_phys->os_flags;
814 dsl_dataset_dirty(ds, tx);
819 typedef struct dmu_objset_create_arg {
820 const char *doca_name;
822 void (*doca_userfunc)(objset_t *os, void *arg,
823 cred_t *cr, dmu_tx_t *tx);
825 dmu_objset_type_t doca_type;
827 } dmu_objset_create_arg_t;
831 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
833 dmu_objset_create_arg_t *doca = arg;
834 dsl_pool_t *dp = dmu_tx_pool(tx);
839 if (strchr(doca->doca_name, '@') != NULL)
840 return (SET_ERROR(EINVAL));
842 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
843 return (SET_ERROR(ENAMETOOLONG));
845 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
849 dsl_dir_rele(pdd, FTAG);
850 return (SET_ERROR(EEXIST));
852 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
854 dsl_dir_rele(pdd, FTAG);
860 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
862 dmu_objset_create_arg_t *doca = arg;
863 dsl_pool_t *dp = dmu_tx_pool(tx);
871 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
873 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
874 doca->doca_cred, tx);
876 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
877 bp = dsl_dataset_get_blkptr(ds);
878 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
879 ds, bp, doca->doca_type, tx);
881 if (doca->doca_userfunc != NULL) {
882 doca->doca_userfunc(os, doca->doca_userarg,
883 doca->doca_cred, tx);
886 spa_history_log_internal_ds(ds, "create", tx, "");
887 dsl_dataset_rele(ds, FTAG);
888 dsl_dir_rele(pdd, FTAG);
892 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
893 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
895 dmu_objset_create_arg_t doca;
897 doca.doca_name = name;
898 doca.doca_cred = CRED();
899 doca.doca_flags = flags;
900 doca.doca_userfunc = func;
901 doca.doca_userarg = arg;
902 doca.doca_type = type;
904 return (dsl_sync_task(name,
905 dmu_objset_create_check, dmu_objset_create_sync, &doca,
906 5, ZFS_SPACE_CHECK_NORMAL));
909 typedef struct dmu_objset_clone_arg {
910 const char *doca_clone;
911 const char *doca_origin;
913 } dmu_objset_clone_arg_t;
917 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
919 dmu_objset_clone_arg_t *doca = arg;
923 dsl_dataset_t *origin;
924 dsl_pool_t *dp = dmu_tx_pool(tx);
926 if (strchr(doca->doca_clone, '@') != NULL)
927 return (SET_ERROR(EINVAL));
929 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
930 return (SET_ERROR(ENAMETOOLONG));
932 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
936 dsl_dir_rele(pdd, FTAG);
937 return (SET_ERROR(EEXIST));
940 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
943 dsl_dir_rele(pdd, FTAG);
944 return (SET_ERROR(EDQUOT));
946 dsl_dir_rele(pdd, FTAG);
948 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
952 /* You can only clone snapshots, not the head datasets. */
953 if (!origin->ds_is_snapshot) {
954 dsl_dataset_rele(origin, FTAG);
955 return (SET_ERROR(EINVAL));
957 dsl_dataset_rele(origin, FTAG);
963 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
965 dmu_objset_clone_arg_t *doca = arg;
966 dsl_pool_t *dp = dmu_tx_pool(tx);
969 dsl_dataset_t *origin, *ds;
971 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
973 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
974 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
976 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
977 doca->doca_cred, tx);
979 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
980 dsl_dataset_name(origin, namebuf);
981 spa_history_log_internal_ds(ds, "clone", tx,
982 "origin=%s (%llu)", namebuf, origin->ds_object);
983 dsl_dataset_rele(ds, FTAG);
984 dsl_dataset_rele(origin, FTAG);
985 dsl_dir_rele(pdd, FTAG);
989 dmu_objset_clone(const char *clone, const char *origin)
991 dmu_objset_clone_arg_t doca;
993 doca.doca_clone = clone;
994 doca.doca_origin = origin;
995 doca.doca_cred = CRED();
997 return (dsl_sync_task(clone,
998 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
999 5, ZFS_SPACE_CHECK_NORMAL));
1003 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1006 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1007 nvlist_t *snaps = fnvlist_alloc();
1009 fnvlist_add_boolean(snaps, longsnap);
1011 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1012 fnvlist_free(snaps);
1017 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1021 while (dn = list_head(list)) {
1022 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1023 ASSERT(dn->dn_dbuf->db_data_pending);
1025 * Initialize dn_zio outside dnode_sync() because the
1026 * meta-dnode needs to set it ouside dnode_sync().
1028 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1031 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1032 list_remove(list, dn);
1035 (void) dnode_add_ref(dn, newlist);
1036 list_insert_tail(newlist, dn);
1045 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1047 blkptr_t *bp = zio->io_bp;
1049 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1051 ASSERT(!BP_IS_EMBEDDED(bp));
1052 ASSERT3P(bp, ==, os->os_rootbp);
1053 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1054 ASSERT0(BP_GET_LEVEL(bp));
1057 * Update rootbp fill count: it should be the number of objects
1058 * allocated in the object set (not counting the "special"
1059 * objects that are stored in the objset_phys_t -- the meta
1060 * dnode and user/group accounting objects).
1063 for (int i = 0; i < dnp->dn_nblkptr; i++)
1064 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1069 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1071 blkptr_t *bp = zio->io_bp;
1072 blkptr_t *bp_orig = &zio->io_bp_orig;
1075 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1076 ASSERT(BP_EQUAL(bp, bp_orig));
1078 dsl_dataset_t *ds = os->os_dsl_dataset;
1079 dmu_tx_t *tx = os->os_synctx;
1081 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1082 dsl_dataset_block_born(ds, bp, tx);
1086 /* called from dsl */
1088 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1091 zbookmark_phys_t zb;
1095 list_t *newlist = NULL;
1096 dbuf_dirty_record_t *dr;
1098 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1100 ASSERT(dmu_tx_is_syncing(tx));
1101 /* XXX the write_done callback should really give us the tx... */
1104 if (os->os_dsl_dataset == NULL) {
1106 * This is the MOS. If we have upgraded,
1107 * spa_max_replication() could change, so reset
1110 os->os_copies = spa_max_replication(os->os_spa);
1114 * Create the root block IO
1116 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1117 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1118 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1119 arc_release(os->os_phys_buf, &os->os_phys_buf);
1121 dmu_write_policy(os, NULL, 0, 0, &zp);
1123 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1124 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1125 DMU_OS_IS_L2COMPRESSIBLE(os),
1126 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1127 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1130 * Sync special dnodes - the parent IO for the sync is the root block
1132 DMU_META_DNODE(os)->dn_zio = zio;
1133 dnode_sync(DMU_META_DNODE(os), tx);
1135 os->os_phys->os_flags = os->os_flags;
1137 if (DMU_USERUSED_DNODE(os) &&
1138 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1139 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1140 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1141 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1142 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1145 txgoff = tx->tx_txg & TXG_MASK;
1147 if (dmu_objset_userused_enabled(os)) {
1148 newlist = &os->os_synced_dnodes;
1150 * We must create the list here because it uses the
1151 * dn_dirty_link[] of this txg.
1153 list_create(newlist, sizeof (dnode_t),
1154 offsetof(dnode_t, dn_dirty_link[txgoff]));
1157 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1158 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1160 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1161 while (dr = list_head(list)) {
1162 ASSERT0(dr->dr_dbuf->db_level);
1163 list_remove(list, dr);
1165 zio_nowait(dr->dr_zio);
1168 * Free intent log blocks up to this tx.
1170 zil_sync(os->os_zil, tx);
1171 os->os_phys->os_zil_header = os->os_zil_header;
1176 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1178 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1179 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1182 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1185 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1191 dmu_objset_userused_enabled(objset_t *os)
1193 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1194 used_cbs[os->os_phys->os_type] != NULL &&
1195 DMU_USERUSED_DNODE(os) != NULL);
1199 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1200 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1202 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1203 int64_t delta = DNODE_SIZE + used;
1206 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1208 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1214 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1217 list_t *list = &os->os_synced_dnodes;
1219 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1221 while (dn = list_head(list)) {
1223 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1224 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1225 dn->dn_phys->dn_flags &
1226 DNODE_FLAG_USERUSED_ACCOUNTED);
1228 /* Allocate the user/groupused objects if necessary. */
1229 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1230 VERIFY(0 == zap_create_claim(os,
1231 DMU_USERUSED_OBJECT,
1232 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1233 VERIFY(0 == zap_create_claim(os,
1234 DMU_GROUPUSED_OBJECT,
1235 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1239 * We intentionally modify the zap object even if the
1240 * net delta is zero. Otherwise
1241 * the block of the zap obj could be shared between
1242 * datasets but need to be different between them after
1246 flags = dn->dn_id_flags;
1248 if (flags & DN_ID_OLD_EXIST) {
1249 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1250 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1252 if (flags & DN_ID_NEW_EXIST) {
1253 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1254 dn->dn_phys->dn_flags, dn->dn_newuid,
1255 dn->dn_newgid, B_FALSE, tx);
1258 mutex_enter(&dn->dn_mtx);
1260 dn->dn_oldflags = 0;
1261 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1262 dn->dn_olduid = dn->dn_newuid;
1263 dn->dn_oldgid = dn->dn_newgid;
1264 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1265 if (dn->dn_bonuslen == 0)
1266 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1268 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1270 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1271 mutex_exit(&dn->dn_mtx);
1273 list_remove(list, dn);
1274 dnode_rele(dn, list);
1279 * Returns a pointer to data to find uid/gid from
1281 * If a dirty record for transaction group that is syncing can't
1282 * be found then NULL is returned. In the NULL case it is assumed
1283 * the uid/gid aren't changing.
1286 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1288 dbuf_dirty_record_t *dr, **drp;
1291 if (db->db_dirtycnt == 0)
1292 return (db->db.db_data); /* Nothing is changing */
1294 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1295 if (dr->dr_txg == tx->tx_txg)
1303 DB_DNODE_ENTER(dr->dr_dbuf);
1304 dn = DB_DNODE(dr->dr_dbuf);
1306 if (dn->dn_bonuslen == 0 &&
1307 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1308 data = dr->dt.dl.dr_data->b_data;
1310 data = dr->dt.dl.dr_data;
1312 DB_DNODE_EXIT(dr->dr_dbuf);
1319 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1321 objset_t *os = dn->dn_objset;
1323 dmu_buf_impl_t *db = NULL;
1324 uint64_t *user = NULL;
1325 uint64_t *group = NULL;
1326 int flags = dn->dn_id_flags;
1328 boolean_t have_spill = B_FALSE;
1330 if (!dmu_objset_userused_enabled(dn->dn_objset))
1333 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1334 DN_ID_CHKED_SPILL)))
1337 if (before && dn->dn_bonuslen != 0)
1338 data = DN_BONUS(dn->dn_phys);
1339 else if (!before && dn->dn_bonuslen != 0) {
1342 mutex_enter(&db->db_mtx);
1343 data = dmu_objset_userquota_find_data(db, tx);
1345 data = DN_BONUS(dn->dn_phys);
1347 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1350 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1351 rf |= DB_RF_HAVESTRUCT;
1352 error = dmu_spill_hold_by_dnode(dn,
1353 rf | DB_RF_MUST_SUCCEED,
1354 FTAG, (dmu_buf_t **)&db);
1356 mutex_enter(&db->db_mtx);
1357 data = (before) ? db->db.db_data :
1358 dmu_objset_userquota_find_data(db, tx);
1359 have_spill = B_TRUE;
1361 mutex_enter(&dn->dn_mtx);
1362 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1363 mutex_exit(&dn->dn_mtx);
1369 user = &dn->dn_olduid;
1370 group = &dn->dn_oldgid;
1372 user = &dn->dn_newuid;
1373 group = &dn->dn_newgid;
1377 * Must always call the callback in case the object
1378 * type has changed and that type isn't an object type to track
1380 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1384 * Preserve existing uid/gid when the callback can't determine
1385 * what the new uid/gid are and the callback returned EEXIST.
1386 * The EEXIST error tells us to just use the existing uid/gid.
1387 * If we don't know what the old values are then just assign
1388 * them to 0, since that is a new file being created.
1390 if (!before && data == NULL && error == EEXIST) {
1391 if (flags & DN_ID_OLD_EXIST) {
1392 dn->dn_newuid = dn->dn_olduid;
1393 dn->dn_newgid = dn->dn_oldgid;
1402 mutex_exit(&db->db_mtx);
1404 mutex_enter(&dn->dn_mtx);
1405 if (error == 0 && before)
1406 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1407 if (error == 0 && !before)
1408 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1411 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1413 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1415 mutex_exit(&dn->dn_mtx);
1417 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1421 dmu_objset_userspace_present(objset_t *os)
1423 return (os->os_phys->os_flags &
1424 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1428 dmu_objset_userspace_upgrade(objset_t *os)
1433 if (dmu_objset_userspace_present(os))
1435 if (!dmu_objset_userused_enabled(os))
1436 return (SET_ERROR(ENOTSUP));
1437 if (dmu_objset_is_snapshot(os))
1438 return (SET_ERROR(EINVAL));
1441 * We simply need to mark every object dirty, so that it will be
1442 * synced out and now accounted. If this is called
1443 * concurrently, or if we already did some work before crashing,
1444 * that's fine, since we track each object's accounted state
1448 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1453 if (issig(JUSTLOOKING) && issig(FORREAL))
1454 return (SET_ERROR(EINTR));
1456 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1459 tx = dmu_tx_create(os);
1460 dmu_tx_hold_bonus(tx, obj);
1461 objerr = dmu_tx_assign(tx, TXG_WAIT);
1466 dmu_buf_will_dirty(db, tx);
1467 dmu_buf_rele(db, FTAG);
1471 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1472 txg_wait_synced(dmu_objset_pool(os), 0);
1477 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1478 uint64_t *usedobjsp, uint64_t *availobjsp)
1480 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1481 usedobjsp, availobjsp);
1485 dmu_objset_fsid_guid(objset_t *os)
1487 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1491 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1493 stat->dds_type = os->os_phys->os_type;
1494 if (os->os_dsl_dataset)
1495 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1499 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1501 ASSERT(os->os_dsl_dataset ||
1502 os->os_phys->os_type == DMU_OST_META);
1504 if (os->os_dsl_dataset != NULL)
1505 dsl_dataset_stats(os->os_dsl_dataset, nv);
1507 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1508 os->os_phys->os_type);
1509 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1510 dmu_objset_userspace_present(os));
1514 dmu_objset_is_snapshot(objset_t *os)
1516 if (os->os_dsl_dataset != NULL)
1517 return (os->os_dsl_dataset->ds_is_snapshot);
1523 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1524 boolean_t *conflict)
1526 dsl_dataset_t *ds = os->os_dsl_dataset;
1529 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1530 return (SET_ERROR(ENOENT));
1532 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1533 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1534 MT_FIRST, real, maxlen, conflict));
1538 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1539 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1541 dsl_dataset_t *ds = os->os_dsl_dataset;
1542 zap_cursor_t cursor;
1543 zap_attribute_t attr;
1545 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1547 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1548 return (SET_ERROR(ENOENT));
1550 zap_cursor_init_serialized(&cursor,
1551 ds->ds_dir->dd_pool->dp_meta_objset,
1552 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1554 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1555 zap_cursor_fini(&cursor);
1556 return (SET_ERROR(ENOENT));
1559 if (strlen(attr.za_name) + 1 > namelen) {
1560 zap_cursor_fini(&cursor);
1561 return (SET_ERROR(ENAMETOOLONG));
1564 (void) strcpy(name, attr.za_name);
1566 *idp = attr.za_first_integer;
1568 *case_conflict = attr.za_normalization_conflict;
1569 zap_cursor_advance(&cursor);
1570 *offp = zap_cursor_serialize(&cursor);
1571 zap_cursor_fini(&cursor);
1577 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1578 uint64_t *idp, uint64_t *offp)
1580 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1581 zap_cursor_t cursor;
1582 zap_attribute_t attr;
1584 /* there is no next dir on a snapshot! */
1585 if (os->os_dsl_dataset->ds_object !=
1586 dsl_dir_phys(dd)->dd_head_dataset_obj)
1587 return (SET_ERROR(ENOENT));
1589 zap_cursor_init_serialized(&cursor,
1590 dd->dd_pool->dp_meta_objset,
1591 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1593 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1594 zap_cursor_fini(&cursor);
1595 return (SET_ERROR(ENOENT));
1598 if (strlen(attr.za_name) + 1 > namelen) {
1599 zap_cursor_fini(&cursor);
1600 return (SET_ERROR(ENAMETOOLONG));
1603 (void) strcpy(name, attr.za_name);
1605 *idp = attr.za_first_integer;
1606 zap_cursor_advance(&cursor);
1607 *offp = zap_cursor_serialize(&cursor);
1608 zap_cursor_fini(&cursor);
1613 typedef struct dmu_objset_find_ctx {
1617 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1620 kmutex_t *dc_error_lock;
1622 } dmu_objset_find_ctx_t;
1625 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1627 dsl_pool_t *dp = dcp->dc_dp;
1628 dmu_objset_find_ctx_t *child_dcp;
1632 zap_attribute_t *attr;
1636 /* don't process if there already was an error */
1637 if (*dcp->dc_error != 0)
1640 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd);
1644 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1645 if (dd->dd_myname[0] == '$') {
1646 dsl_dir_rele(dd, FTAG);
1650 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1651 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1654 * Iterate over all children.
1656 if (dcp->dc_flags & DS_FIND_CHILDREN) {
1657 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1658 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1659 zap_cursor_retrieve(&zc, attr) == 0;
1660 (void) zap_cursor_advance(&zc)) {
1661 ASSERT3U(attr->za_integer_length, ==,
1663 ASSERT3U(attr->za_num_integers, ==, 1);
1665 child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
1667 child_dcp->dc_ddobj = attr->za_first_integer;
1668 if (dcp->dc_tq != NULL)
1669 (void) taskq_dispatch(dcp->dc_tq,
1670 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1672 dmu_objset_find_dp_impl(child_dcp);
1674 zap_cursor_fini(&zc);
1678 * Iterate over all snapshots.
1680 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
1682 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1687 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1688 dsl_dataset_rele(ds, FTAG);
1690 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1691 zap_cursor_retrieve(&zc, attr) == 0;
1692 (void) zap_cursor_advance(&zc)) {
1693 ASSERT3U(attr->za_integer_length, ==,
1695 ASSERT3U(attr->za_num_integers, ==, 1);
1697 err = dsl_dataset_hold_obj(dp,
1698 attr->za_first_integer, FTAG, &ds);
1701 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1702 dsl_dataset_rele(ds, FTAG);
1706 zap_cursor_fini(&zc);
1710 dsl_dir_rele(dd, FTAG);
1711 kmem_free(attr, sizeof (zap_attribute_t));
1719 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1722 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1723 dsl_dataset_rele(ds, FTAG);
1727 mutex_enter(dcp->dc_error_lock);
1728 /* only keep first error */
1729 if (*dcp->dc_error == 0)
1730 *dcp->dc_error = err;
1731 mutex_exit(dcp->dc_error_lock);
1734 kmem_free(dcp, sizeof (*dcp));
1738 dmu_objset_find_dp_cb(void *arg)
1740 dmu_objset_find_ctx_t *dcp = arg;
1741 dsl_pool_t *dp = dcp->dc_dp;
1744 * We need to get a pool_config_lock here, as there are several
1745 * asssert(pool_config_held) down the stack. Getting a lock via
1746 * dsl_pool_config_enter is risky, as it might be stalled by a
1747 * pending writer. This would deadlock, as the write lock can
1748 * only be granted when our parent thread gives up the lock.
1749 * The _prio interface gives us priority over a pending writer.
1751 dsl_pool_config_enter_prio(dp, FTAG);
1753 dmu_objset_find_dp_impl(dcp);
1755 dsl_pool_config_exit(dp, FTAG);
1759 * Find objsets under and including ddobj, call func(ds) on each.
1760 * The order for the enumeration is completely undefined.
1761 * func is called with dsl_pool_config held.
1764 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1765 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1770 dmu_objset_find_ctx_t *dcp;
1773 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
1774 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
1777 dcp->dc_ddobj = ddobj;
1778 dcp->dc_func = func;
1780 dcp->dc_flags = flags;
1781 dcp->dc_error_lock = &err_lock;
1782 dcp->dc_error = &error;
1784 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
1786 * In case a write lock is held we can't make use of
1787 * parallelism, as down the stack of the worker threads
1788 * the lock is asserted via dsl_pool_config_held.
1789 * In case of a read lock this is solved by getting a read
1790 * lock in each worker thread, which isn't possible in case
1791 * of a writer lock. So we fall back to the synchronous path
1793 * In the future it might be possible to get some magic into
1794 * dsl_pool_config_held in a way that it returns true for
1795 * the worker threads so that a single lock held from this
1796 * thread suffices. For now, stay single threaded.
1798 dmu_objset_find_dp_impl(dcp);
1799 mutex_destroy(&err_lock);
1804 ntasks = dmu_find_threads;
1806 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
1807 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
1810 kmem_free(dcp, sizeof (*dcp));
1811 mutex_destroy(&err_lock);
1813 return (SET_ERROR(ENOMEM));
1817 /* dcp will be freed by task */
1818 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
1821 * PORTING: this code relies on the property of taskq_wait to wait
1822 * until no more tasks are queued and no more tasks are active. As
1823 * we always queue new tasks from within other tasks, task_wait
1824 * reliably waits for the full recursion to finish, even though we
1825 * enqueue new tasks after taskq_wait has been called.
1826 * On platforms other than illumos, taskq_wait may not have this
1831 mutex_destroy(&err_lock);
1837 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1838 * The dp_config_rwlock must not be held when this is called, and it
1839 * will not be held when the callback is called.
1840 * Therefore this function should only be used when the pool is not changing
1841 * (e.g. in syncing context), or the callback can deal with the possible races.
1844 dmu_objset_find_impl(spa_t *spa, const char *name,
1845 int func(const char *, void *), void *arg, int flags)
1848 dsl_pool_t *dp = spa_get_dsl(spa);
1851 zap_attribute_t *attr;
1856 dsl_pool_config_enter(dp, FTAG);
1858 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1860 dsl_pool_config_exit(dp, FTAG);
1864 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1865 if (dd->dd_myname[0] == '$') {
1866 dsl_dir_rele(dd, FTAG);
1867 dsl_pool_config_exit(dp, FTAG);
1871 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1872 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1875 * Iterate over all children.
1877 if (flags & DS_FIND_CHILDREN) {
1878 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1879 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1880 zap_cursor_retrieve(&zc, attr) == 0;
1881 (void) zap_cursor_advance(&zc)) {
1882 ASSERT3U(attr->za_integer_length, ==,
1884 ASSERT3U(attr->za_num_integers, ==, 1);
1886 child = kmem_asprintf("%s/%s", name, attr->za_name);
1887 dsl_pool_config_exit(dp, FTAG);
1888 err = dmu_objset_find_impl(spa, child,
1890 dsl_pool_config_enter(dp, FTAG);
1895 zap_cursor_fini(&zc);
1898 dsl_dir_rele(dd, FTAG);
1899 dsl_pool_config_exit(dp, FTAG);
1900 kmem_free(attr, sizeof (zap_attribute_t));
1906 * Iterate over all snapshots.
1908 if (flags & DS_FIND_SNAPSHOTS) {
1909 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1914 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1915 dsl_dataset_rele(ds, FTAG);
1917 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1918 zap_cursor_retrieve(&zc, attr) == 0;
1919 (void) zap_cursor_advance(&zc)) {
1920 ASSERT3U(attr->za_integer_length, ==,
1922 ASSERT3U(attr->za_num_integers, ==, 1);
1924 child = kmem_asprintf("%s@%s",
1925 name, attr->za_name);
1926 dsl_pool_config_exit(dp, FTAG);
1927 err = func(child, arg);
1928 dsl_pool_config_enter(dp, FTAG);
1933 zap_cursor_fini(&zc);
1937 dsl_dir_rele(dd, FTAG);
1938 kmem_free(attr, sizeof (zap_attribute_t));
1939 dsl_pool_config_exit(dp, FTAG);
1944 /* Apply to self. */
1945 return (func(name, arg));
1949 * See comment above dmu_objset_find_impl().
1952 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1958 error = spa_open(name, &spa, FTAG);
1961 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1962 spa_close(spa, FTAG);
1967 dmu_objset_set_user(objset_t *os, void *user_ptr)
1969 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1970 os->os_user_ptr = user_ptr;
1974 dmu_objset_get_user(objset_t *os)
1976 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1977 return (os->os_user_ptr);
1981 * Determine name of filesystem, given name of snapshot.
1982 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
1985 dmu_fsname(const char *snapname, char *buf)
1987 char *atp = strchr(snapname, '@');
1989 return (SET_ERROR(EINVAL));
1990 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
1991 return (SET_ERROR(ENAMETOOLONG));
1992 (void) strlcpy(buf, snapname, atp - snapname + 1);